kohvakka-medium access control and hardware prototype designs for low-energy wireless sensor...

Tampereen teknillinen yliopisto. Julkaisu 808 Tampere University of Technology. Publication 808 Mikko Kohvakka Medium Access Control and Hardware Prototype Designs for Low-Energy Wireless Sensor Networks Thesis for the degree of Doctor of Technology to be presented with due permission for public examination and criticism in Tietotalo Building, Auditorium TB109, at Tampere University of Technology, on the 29th of May 2009, at 12 noon. Tampereen teknillinen yliopisto - Tampere University of Technology Tampere 2009

ISBN 978-952-15-2153-9 ISSN 1459-2045

ABSTRACT

A Wireless Sensor Network (WSN) is an emerging technology consisting of small,cheap, and ultra-low energy sensor nodes, which cooperatively monitor physicalquantities, actuate, and perform data processing tasks. A deployment may comprisethousands of randomly distributed autonomous nodes, which must self-configure andcreate a multi-hop network topology.

This Thesis focuses on low-energy WSNs targeting to long network lifetime. Themain research problem is the combination of adaptive and scalable multi-hop net-working with constrained energy budget, processing power, and communication band-width. The research problem is approached by energy-efficient protocols and low-power sensor node platforms.

The main contribution of this Thesis is an energy-efficient Medium Access Control(MAC) design for TUTWSN (Tampere University of Technology Wireless SensorNetwork). The design comprises channel access and networking mechanisms, whichspecify data exchange, link synchronization, network self-configuration, and neigh-bor discovery operations. The second outcome are several low-power sensor nodeplatforms, which have been designed and implemented to evaluate the performanceof the MAC design and hardware components in real deployments. The third out-come are the performance models and analysis of several MAC designs includingTUTWSN, IEEE 802.15.4, and the most essential research proposals.

The results and conclusion of this Thesis indicate that it is possible to implementmulti-hop WSNs in harsh and dynamic operation conditions with years of lifetimeusing current low-cost components and batteries. Energy analysis results indicatethat the lowest energy consumption is achieved by using simple and high data-ratetransceivers. It is also critical to minimize sleep mode power consumption of allcomponents and to use accurate wake-up timers. However, the selection of com-ponents constitutes only a minor part of the solution, and an energy-efficient MAClayer design being able to minimize radio duty cycle is required. A theoretically idealMAC eliminates idle listening, overhearing, collisions, and control traffic overhead.The performance analysis shows that TUTWSN MAC achieves the highest energy-

ii Abstract

efficiency in both router and leaf nodes compared to existing proposals and standards.Compared to the ideal MAC, the energy consumption of TUTWSN MAC is only2.85% - 27.1% higher, depending on traffic load, radio, and node type. IEEE 802.15.4performs the second best resulting in 2.92% to 229% energy overhead. Analysis andmeasurements indicate that TUTWSN can maintain high energy-efficiency also indynamic networks. The MAC and platform designs are measured and validated inlong-term deployments using full-scale WSN implementations.

The results of this Thesis can be used in the WSN research, development, and im-plementation in general. The designed mechanisms in the MAC layer are presentedand analyzed separately on each other. Presented performance models can be eas-ily adapted to other protocols. In addition, the developed sensor node platforms areapplicable for experimenting other applications and protocol.

PREFACE

The research work for this Thesis was carried out in the Department of ComputerSystems at Tampere University of Technology during the years 2002 - 2008.

I would like to express my sincere gratitude to my supervisor Professor Marko Hän-nikäinen for his guidance, support, and motivation during the research. I am alsograteful to Professor Timo D. Hämäläinen for his farsighted guidance and for the op-portunity to carry out the research in DACI research group. I would also like to thankthe reviewers of my Thesis, Professor Jari Porras and Assistant Professor EvgenyOsipov for their valuable comments, and Professor Petri Mähönen for agreeing toserve as an opponent in the defense.

Many thanks to the members of the DACI research group for the inspiring and cre-ative atmosphere. Special thanks to Dr. Mauri Kuorilehto, Mr. Jukka Suhonen,M.Sc., Mr. Ville Kaseva, M.Sc., Mr. Tero Arpinen, M.Sc., and all the members ofTUTWSN team for their valuable work that have made this Thesis possible. Also,thanks to Dr. Panu Hämäläinen, Dr. Timo Vanhatupa, and Dr. Jari Heikkinen forvaluable comments and discussions.

My Thesis was financially supported by Finnish Funding Agency for Technology andInnovation (TEKES), Academy of Finland, Nokia Foundation, Tekniikan edistämis-säätiö (TES), and Ulla Tuomisen Säätiö.

Finally, I would like to express my gratitude to my family and friends for their supportand encouragement. Especially, I am deeply grateful to my wife Salla for her loveand understanding through these years.

Tampere, April 2009

Mikko Kohvakka

iv Preface

TABLE OF CONTENTS

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii

Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

List of Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix

List of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 WSN Technology Overview . . . . . . . . . . . . . . . . . . . . . 2

1.2 Key WSN Design Requirements . . . . . . . . . . . . . . . . . . . 5

1.3 Scope, Objectives and Methods of the Research . . . . . . . . . . . 7

1.4 Research Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.5 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2. Applications and Standards . . . . . . . . . . . . . . . . . . . . . . . . 11

2.1 WSN Application Space . . . . . . . . . . . . . . . . . . . . . . . 11

2.2 Deployments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.3 Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.3.1 Wireless Communication Standards . . . . . . . . . . . . . 15

2.3.2 Standards Related to WSNs . . . . . . . . . . . . . . . . . 17

3. Sensor Node Platforms . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.1 Platform Components . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.1.1 Communication subsystem . . . . . . . . . . . . . . . . . . 24

3.1.2 Computing subsystem . . . . . . . . . . . . . . . . . . . . 25

vi Table of Contents

3.1.3 Sensing subsystem . . . . . . . . . . . . . . . . . . . . . . 27

3.1.4 Power subsystem . . . . . . . . . . . . . . . . . . . . . . . 27

3.2 Existing Platforms . . . . . . . . . . . . . . . . . . . . . . . . . . 29

4. Medium Access Control for WSNs . . . . . . . . . . . . . . . . . . . . 35

4.1 Low-Energy MAC Design . . . . . . . . . . . . . . . . . . . . . . 35

4.2 Existing MAC Protocols . . . . . . . . . . . . . . . . . . . . . . . 36

4.2.1 Unsynchronized Low Duty-Cycle MAC Protocols . . . . . 37

4.2.2 Synchronized Low Duty-Cycle MAC Protocols . . . . . . . 39

5. TUTWSN MAC Design . . . . . . . . . . . . . . . . . . . . . . . . . . 45

5.1 TUTWSN Networking Technology . . . . . . . . . . . . . . . . . . 45

5.2 TUTWSN Channel Access Mechanism . . . . . . . . . . . . . . . 47

5.3 TUTWSN Networking Mechanisms . . . . . . . . . . . . . . . . . 52

5.3.1 Self-Configuration . . . . . . . . . . . . . . . . . . . . . . 52

5.3.2 Neighbor Discovery . . . . . . . . . . . . . . . . . . . . . 58

6. Performance Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

6.1 Utilized Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 64

6.2 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

6.2.1 Ideal-MAC . . . . . . . . . . . . . . . . . . . . . . . . . . 66

6.2.2 B-MAC . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

6.2.3 SCP-MAC . . . . . . . . . . . . . . . . . . . . . . . . . . 68

6.2.4 X-MAC . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

6.2.5 T-MAC . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

6.2.6 IEEE 802.15.4 . . . . . . . . . . . . . . . . . . . . . . . . 73

6.2.7 TUTWSN MAC . . . . . . . . . . . . . . . . . . . . . . . 74

6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

Table of Contents vii

7. TUTWSN Sensor Node Platforms and Deployments . . . . . . . . . . . . 79

7.1 TUTWSN Node Designs . . . . . . . . . . . . . . . . . . . . . . . 79

7.1.1 Node Designs in 2004 . . . . . . . . . . . . . . . . . . . . 80

7.1.2 Node Designs in 2005 . . . . . . . . . . . . . . . . . . . . 83

7.1.3 Node Designs in 2006 . . . . . . . . . . . . . . . . . . . . 88

7.1.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

7.2 TUTWSN Deployments . . . . . . . . . . . . . . . . . . . . . . . 93

7.2.1 Indoor Temperature Sensing Deployment . . . . . . . . . . 93

7.2.2 Environmental Monitoring Deployment . . . . . . . . . . . 97

7.2.3 Building Monitoring Deployment . . . . . . . . . . . . . . 99

8. Summary of Publications . . . . . . . . . . . . . . . . . . . . . . . . . 103

9. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

LIST OF PUBLICATIONS

This Thesis consists of an introductory part and the following publications. Thepublications are divided into three categories: protocol designs, performance analy-sis, and hardware platforms. The publications presenting new protocol designs arereferred as [P1] - [P4]. The publications covering the performance analysis and opti-mizations of existing protocols are referred as [P5], [P6]. The publications discussingthe implementation of hardware prototypes and real application deployments are re-ferred as [P7] - [P10].

[P1] M. Kohvakka, "TUTWSN MAC Protocol," book chapter in Ultra-low energywireless sensor networks in practice: theory, realization and deployment, M.Kuorilehto, M. Kohvakka, J. Suhonen, P. Hämäläinen, M. Hännikäinen, andT.D. Hämäläinen, Chichester: John Wiley, 2007, pp. 145-182.

[P2] M. Kohvakka, J. Suhonen, M. Hännikäinen, and T. D. Hämäläinen, “Trans-mission Power Based Path Loss Metering for Wireless Sensor Networks,”in Proceedings of the 17th Annual IEEE International Symposium on Per-sonal, Indoor and Mobile Radio Communications (PIMRC 2006), Helsinki,Finland, Sep. 11–14, 2006, pp. 1–5.

[P3] M. Kohvakka, M. Kuorilehto, M. Hännikäinen, and T. D. Hämäläinen, “Net-work Signaling Channel for Improving ZigBee Performance in DynamicCluster-Tree Networks,” EURASIP Journal on Wireless Communications andNetworking, vol. 8, no. 3, pp. 1–15, January 2008.

[P4] M. Kohvakka, J. Suhonen, M. Kuorilehto, V. Kaseva, M. Hännikäinen, and T.D. Hämäläinen, “Energy-Efficient Neighbor Discovery Protocol for MobileWireless Sensor Networks,” Ad Hoc Networks, vol. 7, no. 1, pp. 24–41,January 2009.

[P5] M. Kohvakka, M. Hännikäinen, and T. D. Hämäläinen, “Energy OptimizedBeacon Transmission Rate in a Wireless Sensor Network,” in Proceedings

x List of Publications

of the 16th International Symposium on Personal Indoor and Mobile RadioCommunications (PIMRC 2005), Berlin, Germany, Sep. 11–14, 2005, pp.1269–1273.

[P6] M. Kohvakka, M. Kuorilehto, M. Hännikäinen, and T. D. Hämäläinen, “Per-formance Analysis of IEEE 802.15.4 and ZigBee for Large-Scale WirelessSensor Network Applications,” in Proceedings of the 3rd ACM InternationalWorkshop on Performance Evaluation of Wireless Ad Hoc, Sensor, and Ubiq-uitous Networks (PE-WASUN 2006), Torremolinos, Spain, Oct. 6, 2006, pp.48–57.

[P7] M. Kohvakka, M. Hännikäinen, and T. D. Hämäläinen, “Wireless SensorPrototype Platform,” in Proceedings of the 29th Annual Conference of theIEEE Industrial Electronics Society (IECON 2003), Virginia, USA, Nov. 2–6, 2003, pp. 1499–1504.

[P8] M. Kohvakka, M. Hännikäinen, and T. D. Hämäläinen, “Ultra Low EnergyWireless Temperature Sensor Network Implementation,” in Proceedings ofthe 16th International Symposium on Personal Indoor and Mobile RadioCommunications (PIMRC 2005), Berlin, Germany, Sep. 11–14, 2005, pp.801–805.

[P9] M. Kohvakka, M. Hännikäinen, and T. D. Hämäläinen, “Wireless SensorNetwork Implementation for Industrial Linear Position Metering,” in Pro-ceedings of the 8th Euromicro Conference on Digital System Design – Archi-tectures, Methods, and Tools (DSD 2005), Porto, Portugal, Aug. 30–Sept. 3,2005, pp. 267–273.

[P10] M. Kohvakka, T. Arpinen, M. Hännikäinen, and T. D. Hämäläinen, “High-Performance Multi-Radio WSN Platform,” in Proceedings of the 2nd Inter-national Workshop on Multi-hop Ad Hoc Networks: from theory to reality(REALMAN 2006), Florence, Italy, May 26, 2006, Italy, pp. 95–97.

LIST OF ABBREVIATIONS

ACK Acknowledgment

ADC Analog-to-Digital Converter

AES Advanced Encryption Standard

API Application Programming Interface

ASIC Application Specific Integrated Circuit

BER Bit Error Rate

B-MAC Berkeley Media Access Control

CAN Controller Area Network

CAP Contention Access Period

CCA Clear Channel Assessment

CDMA Code Division Multiple Access

CFP Contention-Free Period

COTS Commercial Off-The-Shelf

CSMA Carrier Sense Multiple Access

CSMA/CA Carrier Sense Multiple Access with Collision Avoidance

DSL Digital Subscriber Line

DSSS Direct Sequence Spread Spectrum

ENDP Energy-efficient Neighbor Discovery Protocol

EEPROM Electrically Erasable Programmable Read-Only Memory

xii List of Abbreviations

FDMA Frequency Division Multiple Access

FET Field-Effect Transistor

FHSS Frequency Hopping Spread Spectrum

FPGA Field Programmable Gate-Array

GDI Great Duck Island

GPS Global Positioning System

GPRS General Packet Radio Service

GSM Global System for Mobile Communications

HR High data Rate

HVAC Heating, Ventilation & Air Conditioning

IC Integrated Circuit

IEEE Institute of Electrical and Electronics Engineers

IR Infra-Red

ISM Industrial, Scientific, Medicine

LAN Local Area Network

LEACH Low-Energy Adaptive Clustering Hierarchy

LED Light Emitting Diode

LOS Line-of-Sight

LPL Low Power Listening

LR-WPAN Low-Rate Wireless Personal Area Network

LR Low data Rate

MAC Medium Access Control

MACA Multiple Access with Collision Avoidance

MCU Micro-Controller Unit

xiii

MIPS Million Instructions Per Second

NCAP Network Capable Application Processor

NCB Network Channel Beaconing

NFC Near Field Communication

NiMH Nickel-Metal Hydride

PACT Power Aware Clustered TDMA

PAMAS Power Aware Multi-Access protocol with Signaling

PCB Printed Circuit Board

PDA Personal Digital Assistant

PHY Physical

PIR Passive Infra-Red

QoS Quality of Service

RF Radio Frequency

RFID Radio Frequency Identification

RSSI Received Signal Strength Indicator

SCP-MAC Scheduled Channel Polling Medium Access Control

SDU Synchronization Data Unit

SIG Special Interest Group

S-MAC Sensor-MAC

SMACS Self-Organizing Medium Access Control for Sensor Networks

SoC System-on-Chip

SpeckMAC-B Speck Medium Access Control Backoff

SpeckMAC-D Speck Medium Access Control Data

SRAM Static Random Access Memory

xiv List of Abbreviations

SRSA Self-Organizing Slot Allocation

STEM Sparse Topology and Energy Management

SYNC Synchronization

TDMA Time Division Multiple Access

TEDS Transducer Electronic Data Sheet

TII Transducer Independent Interface

TIM Transducer Interface Module

T-MAC Timeout-MAC

TRAMA Traffic-Adaptive Medium Access

TUTWSN Tampere University of Technology Wireless Sensor Network

TX Transmit

UI User Interface

UMTS Universal Mobile Telecommunications System

USB Universal Serial Bus

WiMAX Wordwide Interoperability for Microwave Access

WiseMAC Wireless Sensor MAC

WLAN Wireless Local Area Network

WMAN Wireless Metropolitan Area Network

WPAN Wireless Personal Area Network

WSN Wireless Sensor Network

WWAN Wireless Wide Area Network

Z-MAC Zebra MAC

1. INTRODUCTION

The most profound technologies are those that disappear.They weave themselves into the fabric of everyday life

until they are indistinguishable from it.– – Mark Weiser

Wireless network technologies have been under rapid development during the recentyears [111]. Most people feel the strong impact of wireless technology mainly dueto the astonishing growth of cell-phone markets [117]. In the future, the highest po-tential for growth will be in other types of networks [111]. One of the most potentialtechnologies is wireless sensor networks [32].

Sensor networks gather information on entities of interest by multiple distributedsensor elements [32, 202]. Early sensor networks are found in the national powergrid with its many sensors [43], in the radar networks used in air traffic control [32],and in factory floor, where fieldbusses interconnect various sensors, actuators, fieldcontrollers and man-machine interfaces [196,214]. These sensor networks consist oflarge and expensive sensor elements interconnected by a wired network [32].

The rapid development and miniaturization of computing and communication cir-cuits has created the vision of a Wireless Sensor Network (WSN), where thousandsof tiny and cheap sensing nodes operate fully autonomously in interaction with theirenvironment [4,5,40,124,207]. Nodes perform wireless communication by simple ra-dios, while small processors provide sophisticated functionality [25,28,132]. By co-operation and in-network data fusion, nodes refine the measurement accuracy, detectand classify occurred events, and control actuators according to the events [31, 112].Ultra-low energy consumption enables the network lifetime of years with small bat-teries, or supply energy scavenging solely from operation environment [150]. WSNshave a vast number of foreseen application fields including military, environmen-tal and condition monitoring, building automation, object tracking, and interactivegames [5, 32, 88, 124, 211]. They have also been seen as an enabling technology forubiquitous networks, where computing power is embedded invisibly around us, and

2 1. Introduction

0

500

1000

1500

2000

2500

3000

3500

1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007

Year

Pu

bli

sh

ed

Art

icle

s p

er

Year

Fig. 1. Annually published IEEE research articles containing a phrase "wireless sensor net-work".

accessed through intelligent interfaces [65, 98, 206, 209].

WSNs have gained extensively growing academic and commercial interest duringthe recent years [207, 209, 210]. Fig. 1 illustrates the number of annually pub-lished articles in the conferences and journals of Institute of Electrical and ElectronicsEngineers (IEEE) containing a phrase "wireless sensor network" [78]. The first arti-cle [25] was published in 1996, and less than 50 articles were published until 2001.However, during a single year 2007, over 3300 WSN related articles were published.Business Week [208] predicted in September 1999 that WSNs will be one of the mostimportant technologies of the 21st century.

1.1 WSN Technology Overview

So far, WSNs have been implemented mainly for research purposes [14, 16, 39,69, 101, 123, 145, 155, 158, 168, 172, 179, 185, 199, 200, 219, 225], while only fewcommercial WSNs exist. Most of the commercial WSNs are development kits orradio modules necessitating engineering work in application development. Exam-ples of development kit providers are Dust Networks, Inc. [48], Nanotron Technolo-gies GmbH [116], Dynastream Innovations, Inc. [50], and Crossbow Technology,Inc. [36]. Sensicast [166] also provides some end-to-end solutions with sensor inte-grations and interfaces for external networks.

Since the application fields and their requirements for the network are diverse, WSNimplementations are application specific. The energy consumption, communicationbandwidth, and networking performance of nodes are optimized to execute a givenapplication task. There are a number of limitations in the current WSN products, both

1.1. WSN Technology Overview 3

Terminal

with UI

Database

Sinks with

gateways

Sensor elements

Inspected phenomenon

Hardware architecture

Wireless links

Ext

erna

lne

twor

k

CommunicationComputingSensing

Radio

Power

Battery Voltage regulator

Application

server

Sink embedded in a

portable computer

Sensor

SensorMCU

AD

CFig. 2. Example WSN application scenario.

in software and hardware including power consumption, networking performanceand physical size [224]. Due to the limitations, existing WSN realizations can imple-ment only a subset of the envisioned features, and a trade-off has been made betweenenergy consumption and data transfer performance.

For clarifying the characteristics of the WSN pursued in this Thesis, an examplescenario is depicted in Fig. 2. An arbitrary number of nodes are randomly deployedin the area of an inspected phenomenon, where they self-configure network topology.In the figure, all nodes are similar (homogenous) in their hardware. It is possiblethat nodes are diverse (heterogenous) in their sensing, processing, and networkingcapabilities, when nodes can be specialized in different tasks. The network containsone or more sink nodes, which request other nodes to perform measurements, andthen collect measured values for further use. Data is routed in the network by a chainof short and low-energy hops (multi-hop routing).

Sinks may operate in various locations in the network. Typically, a sink is integratedwith a gateway connecting WSN to an external network. An external computer net-work may contain an application server that delivers applications for user terminals,performs data processing, and handles access to a database. The external network canalso be another WSN operating e.g. at a different frequency band. A sink may also be

4 1. Introduction

Table 1. Component power consumptions and allowed activity levels.Component Power consumption Allowed activity

Radio (reception) 50 - 100 mW 0.1 - 1%Radio (transmission) 25 - 50 mW 0.1 - 1%MCU 1 - 5 mW 1 - 5%Sensors 0.5 - 10 mW 0.01 - 1%All components (sleep) 5 - 100 µW 100%

embedded in a portable computer providing a User Interface (UI). Moreover, nodesmaking actuating decisions according to received information from other nodes aresinks.

The focus of this Thesis is on low-energy WSNs, where long network lifetime is moreimportant than throughput, latency, and data processing performance. Network life-time can be increased by using low-power hardware components and high-capacitybatteries. For clarifying the capability of current Commercial Off-The-Shelf (COTS)components, an example node assembly targeting to long lifetime and small size ispresented [53, 152] . The node comprises a Micro-Controller Unit (MCU) havingaround 1 Million Instructions Per Second (MIPS) processing speed, tens of kilobytesprogram memory, few kilobytes data memory, and an Analog-to-Digital Converter(ADC). In addition, the node contains a short-range radio, and a set of sensors. Sup-ply power is obtained from AA-size batteries.

Assuming a target lifetime of one year using AA-size batteries, the available powerbudget is around 1 mW. For reference, Table 1 presents the typical power consump-tions of hardware components. For reaching the budget, allowed activity for trans-missions, receptions, sensing and data processing is in the order of one percent, whilethe rest of time must be spent in a low power sleep mode. In a sleep mode, a nodeturns off the MCU and the radio such that only a low power wake up timer is activeto be able to wake up according to a schedule. Thus, a low-energy WSN implemen-tation necessitates the combination of low-power hardware components with energy-efficient protocols, which can eliminate the unnecessary activity of the hardware.

Wireless communication between nodes is enabled by network protocols. A protocolis a set of rules, which define what (semantics), how (syntax), and when (timing) tocommunicate for exchanging data between two entities [177]. Network protocols aretypically divided into layers according to their responsibilities, and together they forma protocol stack. Each layer has precisely defined interfaces, which permits flexibleupdates and changes in the software and hardware implementations in a modularmanner.

1.2. Key WSN Design Requirements 5

Multi-hop routing

Medium access control

Radio transceiver

Transport

Application

programming interface

Ap

pli

cati

on

Ap

pli

cati

on

Ap

pli

cati

on

Ap

pli

cati

on

Ap

pli

cati

on

Fig. 3. A WSN protocol stack used in this Thesis.

An WSN protocol stack implementation used in this Thesis is presented in Fig. 3. Aradio transceiver (radio) transmits and receives messages one bit or symbol at a timeby making a conversion between digital data and analog symbols of the medium.

A Medium Access Control (MAC) protocol determines how and when to utilize ra-dio functions for discovering network neighborhood, establishing wireless links, andexchanging different frame types. A routing protocol creates multi-hop routing pathsbetween source and destination nodes, while the transport protocol implements end-to-end flow control, if necessary. An Application Programming Interface (API) ab-stracts the underlying communication and hardware for applications. Several appli-cations may be executed in parallel, for example sensing, actuating, data fusion, andnode diagnostics.

1.2 Key WSN Design Requirements

The key design requirements of low-energy WSNs differ in many ways from tradi-tional wireless computer networks. As presented in Table 2, high throughput and lowlatency are the most critical requirements for wireless computer networks, e.g. IEEE802.11 Wireless Local Area Network (WLAN) [79].

The most important design requirement for the low-energy WSN is resource con-strained implementation. In this Thesis, WSN should be implementable in practiceusing current COTS components. Due to the size, cost, and lifetime requirementsof applications, WSN nodes have scarce communication, computation, and energyresources.

Another important requirement is networking performance. WSN should be adaptivefor ensuring network robustness in uncertain operation conditions, where network

6 1. Introduction

Table 2. Relation of requirements for wireless computer networks and low-energy WSNs.Requirement Criticality for wireless Criticality for

computer networks low-energy WSNs

Resource constrained Low Very highimplementationAdaptivity Low HighScalability Moderate HighFairness Moderate ModerateLatency High LowThroughput Very high Low

size, node locations, and Radio Frequency (RF) propagation conditions vary dynam-ically. In order to enable mobile applications, network should tolerate node mobilityat a pedestrian walking speed. In addition, scalability to support various networksizes up to tens of thousands of nodes, and node densities up to hundreds of activenodes in a radio range is required.

Due to the characteristics of WSN applications, data transfer performance has lowercriticality than the above mentioned requirements. Fairness is desirable such thatall nodes can transmit data to sinks equally, and that all sinks can receive data fromnodes equally. A tolerable latency from nodes to a sink in a large network is evenminutes, while a sufficient throughput is in the order of one kbits/s.

Existing standards for wireless communications, such as cellular telephone networks,WiMAX [215], IEEE 802.11 WLAN [79], and Bluetooth [20] are unsatisfactoryfor WSNs due to their limited energy-efficiency, scalability and adaptivity. IEEE802.15.4 Low-Rate Wireless Personal Area Network (LR-WPAN) can be configuredto a low-power operation mode and thus, it is a promising standard for enablingWSNs [64]. Still, it has problems of combining the energy-efficiency with adaptivityand scalability required by many potential applications [118].

Although many protocols have been proposed for traditional wireless ad hoc net-works [126], these protocols aim to provide good throughput and delay characteris-tics. However, energy consumption takes up secondary importance, since batteriescan be easily replaced or charged when needed. Hence, these protocols cannot pro-vide adequate energy-efficiency for WSNs [5].

There are two wireless technologies, which can be categorized as low-end WSNs:Radio Frequency Identification (RFID) and Near Field Communication (NFC). RFIDtechnology is typically seen as an intelligent version of the bar codes [75, 86]. In

1.3. Scope, Objectives and Methods of the Research 7

general, RFID tags can communicate with a high-power reader device only [27].NFC technology can be seen as a contactless smart card reader and writer, which isembedded in portable electronic devices, such as mobile phones. NFC devices canform a simple peer-to-peer network and exchange data in a range of about 10 cm [29].The suitability of these technologies for WSNs is severely limited due to their shortrange, small network size, and limited networking capability.

1.3 Scope, Objectives and Methods of the Research

The scope of this research is on the MAC layer protocols and component-based sen-sor node hardware prototypes (platforms) for low-energy WSNs. Applications andupper protocol layers are covered only briefly for evaluating the requirements for theMAC and platform designs. Sensing, simulations, software design flow, and hard-ware design of Integrated Circuits (ICs) are outside the scope of this Thesis.

The main objective of this Thesis is to solve the problems of resource constrainedWSN implementation and networking performance by a MAC layer design. An ob-jective is to design feasible solutions for real deployments in harsh operation condi-tions. For obtaining realistic and feasible results, the MAC layer is designed accord-ing to the characteristics of current state-off-the-art low-power COTS components.

The methods for conducting the research results of this Thesis are presented in Fig.4. In the first phase, a literature review is carried out and target applications areidentified. At the same time, the capabilities and behavior of current state-of-the-art hardware components are analyzed and the requirements for MAC design areevaluated. According to this information, the initial version of a MAC design isformulated.

Next, the performance of the MAC design and other MAC approaches are evaluatedby modeling them analytically. The performance of the approaches are analyzed andcompared, while varying protocol, network, and application parameters. Accordingto analysis results, the MAC design is refined.

As a potential MAC design is found, its feasibility is studied in real applications byexperimental measurements. An objective is to identify the non-ideal characteristicsof hardware components and radio environment. This information is used in the MACdesign for improving and optimizing its performance. Thus, real sensor node plat-forms are designed and implemented using state-of-the-art commercial components.The power consumptions and operation mode switching times of the platforms are

8 1. Introduction

Results of

this Thesis

Results of

this Thesis

Platform designPlatform design

Prototype

deployments and

measurements

Prototype

deployments and

measurements

ModelsModels

Analysis of

hardware

components

Analysis of

hardware

componentsLiterature reviewLiterature review

Measured

network

performance

Measured

platform

performance

Analyzed

protocol

performance

Evaluation

of MAC

requirements

Evaluation

of MAC

requirements

Performance

analysis

Performance

analysis

MAC designMAC design

Analysis of

results

Analysis of

results

Fig. 4. Methods for obtaining the results of this Thesis

back-annotated to the performance analysis phase for improving analysis accuracyand for refining the MAC design.

Next, a prototype network is deployed and its performance is measured. The mea-sured performance is used for improving and optimizing MAC and sensor node plat-form designs.

As results, this Thesis presents MAC protocol designs and sensor node platforms,which fulfill the presented requirements of low energy WSNs.

1.4 Research Outcomes

The results of this Thesis are summarized in Fig. 5. The main contribution is a MACdesign for TUTWSN (Tampere University of Technology Wireless Sensor Network)

1.4. Research Outcomes 9

ResultsResults

MAC designs

[P1]-[P4]

MAC designs

[P1]-[P4]

Sensor node

platforms

[P7]-[P10]

Sensor node

platforms

[P7]-[P10]

Performance

models

[P5]-[P6]

Performance

models

[P5]-[P6]

Models for

networking

mechanisms

Models for

networking

mechanisms

Models for

channel access

mechanisms

Models for

channel access

mechanisms

Networking

mechanisms

Networking

mechanismsChannel access

mechanisms

Channel access

mechanisms

Fig. 5. Results of this Thesis.

([P1] - [P4]). The design comprises channel access and networking mechanisms.While channel access mechanism defines mainly frame exchanges and link synchro-nization, the networking mechanisms specify network self-configuration and neigh-bor discovery mechanisms.

The second research outcome are sensor node platforms ([P7] - [P10]). Several plat-forms have been designed and implemented to evaluate the performance of the MACdesign and hardware components in real deployments.

The third main outcome of the research are performance models for various channelaccess and networking mechanisms ([P5], [P6]). MAC layer models are defined forTUTWSN, IEEE 802.15.4, and the most essential research proposals. Besides indi-cating the performance of the TUTWSN MAC design, the models are used for show-ing the energy-efficiency of the designed networking mechanisms for IEEE 802.15.4[P3].

As a summary, the main contributions of the Thesis are:

- A survey of existing energy-efficient MAC protocols, standards, and low-powersensor node platforms for WSNs.

- Energy-efficient and scalable channel access mechanism for a resource con-strained WSN called TUTWSN MAC.

- Networking mechanisms for improving network adaptivity in dynamic envi-ronments.

- Several low-power sensor node platforms for WSNs. The platforms enable realWSN deployments and experimental performance evaluation.

10 1. Introduction

- Performance models and analysis of several MAC designs including TUTWSN,IEEE 802.15.4, and the most essential research proposals.

- Deployment cases that validate the analysis results and the feasibility of de-signed mechanisms and platforms.

1.5 Thesis Outline

The Thesis consists of an introductory part and 10 publications [P1-P10]. The intro-ductory part motivates the work, presents technical background, and summarizes andanalyzes the results. The main results are presented in the publications. Publication[P1] presents the overview of MAC layer, while the results in details are presented inpublications [P2] - [P10]. The rest of the introductory part is organized as follows:

Chapter 2 presents the WSN application space and performed deployments. Thechapter includes also an overview of existing standards related to WSNs.

Chapter 3 discusses the design principles of sensor node platforms, and presents ex-isting platforms. The chapter presents the characteristics of low-power componentsand provides a basis for a MAC layer design.

Chapter 4 discusses the most essential MAC protocols for WSNs. The chapter pro-vides a research background for TUTWSN MAC design.

Chapter 5 composes the research results of channel access and networking mecha-nism.

Chapter 6 presents the research results of performance models considering the mostessential MAC layer designs for WSNs.

Chapter 7 presents the research results of sensor node platform designs and deploy-ments cases.

Chapter 8 summarizes the publications included in the Thesis

Chapter 9 concludes the Thesis.

2. APPLICATIONS AND STANDARDS

This chapter presents the current state of WSN research in application development.First, the design space of WSN applications is presented including application tasksand usage classes. Then, the performed WSN deployments are discussed. As a refer-ence, this chapter presents also existing standards for wireless communication. Thestate of the research in sensor node platforms and MAC protocols will be presentedin the following chapters.

2.1 WSN Application Space

WSN application space is continuously emerging and extending together with thedevelopment of low-power circuits and communication protocols. Potential appli-cations have been found in home automation, environmental and industrial monitor-ing, military, personal security, asset management, and traffic control [5, 32, 40, 88,144, 146, 160]. WSN applications can execute one or more application tasks, whichare build upon the sensing, actuating, communication, and computing capabilities ofWSN nodes. The application tasks can be divided into five categories [5, 66, 93]:

Data logging: Determine periodically the value of a physical quantity in agiven location, for example temperature, humidity and gas concentration.

Event detection: Detect the occurrence of an event of interest, for examplemotion or the exceeding of a predetermined physical quantity.

Object classification: Identify an object or an event according to measure-ments using different sensors.

Object tracking: Trace the movement, direction and speed of an object ac-cording to measurements at different locations.

Control: Control an actuator according to commands or measured sensor val-ues within the network, for example a valve and an electrical switch.

12 2. Applications and Standards

Wireless data networks can be classified into six classes according to message la-tency and reliability requirements, as presented in Table 3 [83]. The classification isreferred for clarifying the application space of WSNs.

The classes 5 - 4 are categorized as monitoring networks. The class 5 has the lightestmessage latency requirements, and it is suitable for data logging networks. Ratherlong delays are tolerated and some messages can be missed. The class 4 has slightlyhigher latency requirements and no messages should be missed. The class 4 is suit-able for flagging networks performing event detection tasks.

The classes 3 - 1 are categorized as control networks. The class 3 contains open-loop control networks, where a human is in the loop. The class 2 contains closedloop supervisory control for non-critical control tasks. The class 1 networks performclosed loop regulatory control, where the criticality for latency and reliability is veryhigh.

Class 0 contains safety networks having the highest requirements for message latencyand reliability. These networks are used for emergency action.

The low-energy WSNs discussed in this Thesis are suitable for the classes 4 and 5,and the class 3 will be reached in the near future. The highest energy-efficiency andscalability are achieved at the class 5. As the requirements for latency and reliabil-ity raise, the need for a centralized network management also increases limiting thescalability and energy-efficiency.

Table 3. Usage classes of wireless data networks.Class Description Criticality of Suitability of

latency and current low-reliability energy WSNs

0 Emergency action Always critical Very low1 Closed loop regulatory control Often critical Very low2 Closed loop supervisory control Often non-critical Low3 Open loop control Non-critical Moderate4 Flagging Non-critical High5 Data logging Non-critical Very high

2.2. Deployments 13

Table 4. Examples of experimental WSN deployments.Deployment Year Scale Duration Data Node MAC Power

(months) period platform protocol (mW)GDI [185] 2003 98 3.8 20 min Mica2Dot B-MAC 1.6Vineyard [16] 2003 65 6 5 min Mica2 CSMA 58Macroscope [199] 2004 33 1.5 5 min Mica2Dot CSMA 1.5-6.3ZebraNet [225] 2004 7 12 8 min custom Z-MAC 30-70Heathland [200] 2005 24 0.5 1 hour ESB CSMA 30SensorScope [13] 2007 17 1.5 2 min Tiny node S-MAC N/A

2.2 Deployments

To illustrate potential application types in practice, few interesting WSN deploymentsare presented. The presented deployments have been selected according to the fol-lowing requirements: networks should utilize multi-hop networking, and the deploy-ment duration should be at least one week. The deployments are summarized in Table4. The columns of the table give an overview and characteristics of the deployments.Focus is on the network size, duration of the deployment, utilized sensor node plat-form and MAC protocol, and average power consumption of nodes. The data periodpresents the activity of the network giving the frequency of data communication.

A deployment on Great Duck Island (GDI) [185] monitors the occupancy of small,underground nesting burrows and the role of micro-climatic factors in their habitatselection on an offshore breading colony. A 115 days long deployment in 2003 com-prised 98 Mica2Dot [38] motes, which executed Berkeley Media Access Control (B-MAC) [129] protocol. Data was collected at 20 minutes intervals, and forwardedto a sink by multi-hop routing. The deployment indicated that most links are shortlived: the median link is used to deliver only 13 packets. The packet loss was verysignificant, since nearly half of packets were lost during routing. Node lifetimes var-ied from 1 to 110 days causing connectivity problems for the network. The averagepower consumption was around 1.6 mW.

A temperature monitoring network deployed in a vineyard for 6 a months period in2003 is presented in [16]. The deployment employed 65 Mica2 [37] motes, which ex-ecuted TinyOS [58] protocol stack including a Carrier Sense Multiple Access (CSMA)type MAC layer. Their data was collected at 5 minutes intervals and multi-hop routedto a base station. Nodes utilized large 42 Ah battery packs. The reported lifetimes ofrouting nodes were about 3 months. According to the battery capacity, the calculatedpower consumption of routing nodes was 58 mW. The deployment demonstrated that


a WSN brings value to an agricultural setting and that the total cost of a WSN islower than a wired data logger type network.

Macroscope [199] deployment recorded temperature, humidity, and solar radiationon the surface of a 70-meter tall redwood tree during 44 days period in the summerof 2004. The deployment consisted 33 Mica2Dot [38] motes, which collected dataat 5 minutes intervals and multi-hop routed the data to a sink node. Nodes utilizedTASK [26] protocol stack including a CSMA type MAC layer. The lifetime of thenodes using 560 mAh batteries were 11 to 44 days. Thus, the calculated powerconsumptions were between 1.5 mW and 6.4 mW. The experiment indicated that thenetwork suffered from high packet loss, especially at nodes far from the sink. Thepacket loss was most probably caused by collisions.

ZebraNet [225] experiment tracked animal movements in Kenya on an area of 36km2. Seven nodes were attached on Zebras, and their locations were tracked during12 months in 2004. Global Positioning System (GPS) was activated at 8 minutes pe-riods at a time, and the data was multi-hop routed to a base station. Nodes compriseda low power MCU, a 900 MHz radio and a GPS receiver, and they executed ZebraMAC (Z-MAC) [143] protocol. The power consumptions of nodes were between 30mW and 70 mW. Each node was powered by a rechargeable battery and a 200 gramssolar panel. The experiment indicated that harsh operation conditions and the lack ofa large ground plane reduced radio range significantly.

Heathland experiment [200] in Northern Germany analyzed wireless communicationin an outdoor environment in March 2005. The deployment comprised 24 ESB [157]nodes, which formed a multi-hop network. The MAC protocol was based on a sim-ple CSMA [91]. Data collection interval was 1 hour. As nodes were powered bythree AA batteries, node lifetimes near the sink were 16 days. Thus, their powerconsumptions were around 30 mW. The experiment demonstrated that the quality ofradio links varies considerably in the long run. The required adaptive protocols canbe evaluated only in field trials.

SensorScope [13] measured the hydrological model of the Grand Saint Bernard passduring a 1.5 months period in 2007. The pass is 2400 m high and located in the Alpsbetween Switzerland and Italy. In the deployment, 17 Tiny nodes [47] were locatedon a 900 m long line, and they measured humidity values at 120 s intervals. Nodes ex-ecuted Sensor-MAC (S-MAC) [220] type channel access and multi-hop routed datato a sink having General Packet Radio Service (GPRS) connectivity. The networksuffered from significant packet losses, which were mostly caused by hardware mal-functions.

2.3. Standards 15

Most of the performed deployments have been targeted for environmental monitor-ing, where data gathering interval has been several minutes. The scale of long-term WSN deployments has been tens of nodes. The average power consumptionof nodes has ranged between 1.5 mW and 58 mW, and adequate network lifetimehas been achieved by using large battery packs or solar panels [16, 225]. Commonproblems in the deployments have been unequal power consumption of nodes andweak tolerance against harsh operation conditions, network dynamics, and conges-tion [199, 200, 225].

Commercial networks are still rare and detailed information about their performanceis not available. Most of them are development kits or radio modules necessitatingengineering work in application development, for example Dust Networks, Inc. [48],Nanotron Technologies GmbH [116], Dynastream Innovations, Inc. [50], and Cross-bow Technology, Inc. [36]. Sensicast [166] provides also various sensors and inter-faces for external networks.

2.3 Standards

There exist numerous standards for wireless communication technologies for en-abling inter-operability between products from different manufacturers. For end usersthe utilization of standards provides many benefits in technology support, productavailability, and expandability. In this Thesis, the standards set a starting point andreference for the research. Next, the current status of standards is discussed in theWSN point of view, and the need for a proprietary solution is reasoned.

2.3.1 Wireless Communication Standards

The classification of standards originated by IEEE categorizes wireless communica-tion technologies according to their range, data rate, and power consumption [75,76,170]. The categories of existing wireless technologies are presented in Table 5. Thepresented value ranges are not absolute but merely indicative [27, 64, 75, 111, 214].Although WSN is not included in the conventional IEEE originated classification, itis presented for comparison.

Wireless Wide Area Networks (WWANs) provide the widest geographical cover-age. The most well-known WWANs are digital cellular telephone networks, suchas Global System for Mobile Communications (GSM), and their extensions for data


services, e.g. GPRS, and Universal Mobile Telecommunications System (UMTS).Also communication satellites belong to this category.

Wireless Metropolitan Area Networks (WMANs) are emerging technologies devel-oped to specify broadband wireless access as an alternative to cable networks andDigital Subscriber Lines (DSLs). Thus, the provided data rates are much higherthan in WWANs. WMAN is often called Wordwide Interoperability for MicrowaveAccess (WiMAX) by an industry group called the WiMAX forum. Examples ofWMANs are IEEE 802.16 and its mobile extensions.

WLANs were originally developed for extending or replacing wired computer LocalArea Networks (LANs). Currently, WLAN is widely employed for providing net-work access with location freedom in homes, public buildings and enterprises, andfor municipal public network implementations. The dominating WLAN technologyis IEEE 802.11 [79] with its numerous extensions for higher communication speeds,Quality of Service (QoS) support, security, and mesh networking.

Wireless Personal Area Networks (WPANs) are generally targeted at data commu-nications between personal devices, including Personal Digital Assistants (PDAs),mobile phones, headsets, and laptops. The most well-known and mature WPAN tech-nology is Bluetooth [20], which is also known as IEEE 802.15.1 [80]. The WPANcategory also includes two emerging standards: an IEEE 802.15.3 [81] standard forhigher data rate multimedia content delivery, and an IEEE 802.15.4 [82] standard forlow-rate and low-power communications. WPANs include very wide range of appli-cations, and their characteristics are not clearly distinct from WLANs, sharing thesame operational environments and application domains. The differences are in thenon-functional requirements, such as cost, power, and networking range.

Table 5. Categorization of wireless communication technologies.

Range Data rate Power Exampleconsumption technologies

WWAN > 10 km < 10 Mbits/s medium/high GSM, GPRS, UMTS, satelliteWMAN < 10 km < 100 Mbits/s high IEEE 802.16, HIPERMANWLAN < 100 m < 100 Mbits/s medium IEEE 802.11, HIPERLAN/2WPAN < 10 m < 10 Mbits/s low Bluetooth, IEEE 802.15.4WSN < 1 km < 100 kbits/s ultra-low Proprietary, IEEE 802.15.4

2.3. Standards 17

2.3.2 Standards Related to WSNs

In contrast to WLANs and WPANs, WSNs are envisioned to have much more net-work devices, which are application-oriented rather than measured by the coverageof a single radio cell or the nominal capacity of a link [5, 88]. WSNs nodes arestand-alone stations without the need or even possibility for human intervention. Thenetwork performance is measured as its capability to serve the implemented appli-cations. Depending on the application, the data rate of a single node varies fromfew bits/s to hundreds of kilobits/s, and network coverage ranges from centimeters toseveral kilometers [94]. There is no pre-existing physical infrastructure that restrictsthe topology. Messages should not be sent to individual nodes but to geographicallocations or regions defined by data content [178]. As resources are constrained, thefeasibility of WSN lies on the joint effort of the nodes [178]. The most potentialstandardized technologies for realizing WSNs are ZigBee [64, 229], and emergingISA-SP100.11a [83].

ZigBee AND IEEE 802.15.4

ZigBee [229] is an open specification for low data-rate wireless control and moni-toring networks, where low-power consumption is a key requirement. The candidateapplications are wireless sensors, lighting controls, and surveillance. The first ver-sion of the ZigBee specification was announced in December 2004. A refinement ofthe specification has been launched in December 2006 [55, 131].

ZigBee builds upon the MAC and Physical (PHY) layers defined by IEEE 802.15.4LR-WPAN, as presented in Fig. 6 [70]. IEEE 802.15.4 is responsible for the channelaccess mechanism, acknowledged frame delivery, network association, and disasso-ciation.

A Network (NWK) layer provides network self-organization and multi-hop routingcapability. NWK performs route discovery and maintenance, and message relayingfunctions. NWK can initiate a new network and assign network addresses to newnodes associating with the network for the first time. A Security Service Provider(SSP) offers security functions including Advanced Encryption Standard (AES) en-cryption, key generation, key distribution, authentication and access control lists[227]. Overall node management is performed by a ZigBee Device Object (ZDO).Application endpoints may call ZDO in order to discover other ZigBee nodes on thenetwork and services they offer, and to define security and network settings. An Ap-plication Support (APS) sub-layer connects NWK, SSP and endpoints, and routes


messages to different endpoints. An application layer at the top of the stack de-termines node relationships, and supervises network initiation and association func-tions. The application layer contains application profiles, which define the format ofexchanged messages for a given applications. ZigBee defines a set of public profilesfor common application scenarios. In addition, vendors may add additional featureswith private profiles [211].

IEEE 802.15.4 defines two Direct Sequence Spread Spectrum (DSSS) radio typesoperating in Industrial, Scientific, Medicine (ISM) frequency bands. A low-bandPHY operates in the 868 MHz or 915 MHz frequency band and has a data-rate of20 kbps or 40 kbps, respectively. A high-band PHY operating in the 2.4 GHz bandspecifies a data-rate of 250 kbps.

IEEE 802.15.4 defines three types of logical devices, a Personal Area Network (PAN)coordinator, a coordinator, and a device. PAN coordinator is the primary controllerof PAN, which initiates the network and operates often as a gateway to other net-works. Each PAN must have exactly one PAN coordinator. Coordinators collaboratewith each other for executing data routing and network self-organization operations.Devices do not have data routing capability and they can communicate only withcoordinators.

Besides star and peer-to-peer network topologies, ZigBee supports a cluster-treetopology. The network consists of clusters, each having a coordinator as a clusterhead and multiple devices as leaf nodes. A PAN coordinator initiates the networkand serves as the sink. The network is formed by parent-child relationships, wherenew nodes associate as children with the existing coordinators. This well-definedstructure simplifies multi-hop routing and allows energy saving on the MAC layer.The applicability of ZigBee on WSN applications is limited by scalability, the energyconsumption of coordinators, and the support for one sink only.

Application

NWK

MAC

PHY

ZigBee Platform

IEEE 802.15.4

ZDOApp Support (APS)

SSP

Application objects

Endpoints

ZigBee Applications

Fig. 6. ZigBee protocol stack.

2.3. Standards 19

MiWi

MiWi [57] developed by Microchip is a simpler version of ZigBee operating abovea IEEE 802.15.4 compliant radio. MiWi is suitable for smaller networks having atmost 1024 nodes. Supported network topologies are star and mesh. Simplificationsfor ZigBee stack reduces the cost of MCU even 40% - 60%. A disadvantage isthat MiWi does not support the low duty-cycle mode of ZigBee. Thus, the energyconsumption of MiWi is too high for most of WSN applications.

Z-Wave

Z-Wave [151] is another simpler version of ZigBee operating at 868 MHz and 915MHz frequency bands. The maximum number of nodes in a network is 232. Sup-ported network topologies are star and mesh. The suitability of Z-Wave for WSN ap-plications is limited by scalability, and energy-efficiency, since the cluster-tree topol-ogy with power saving features is not supported.

Bluetooth

Bluetooth [20] is a wireless technology specified by the Bluetooth Special InterestGroup (SIG). IEEE has standardized the MAC and PHY layers of Bluetooth as IEEE802.15.1 [80]. Originally, it was only intended as a simple serial cable replacementfor electronic devices. Presently, the technology supports various more advancedfunctionalities, such as ad hoc networking and access point operation for Internetconnections.

Bluetooth operates in the 2.4 GHz unlicensed ISM band and utilizes the FrequencyHopping Spread Spectrum (FHSS) technique in the radio interface [20]. Currentversions have up to 3 Mbps data rate, while the link range is from 10 cm to 100 m.A network composed of Bluetooth devices is called a piconet, which consists of amaster and up to seven slave devices. Piconets can be linked together to form a largernetwork, known as scatternet. The applicability of Bluetooth in WSN applications islimited by scalability and energy consumption.

Ultra-Low-Power Bluetooth

Ultra-Low-Power Bluetooth (ULPB) [19,212] is a light-weight version of Bluetooth.It has been announced by Nokia Corp as Wibree at October 2006. ULPB is operating


at 2.4 GHz frequency band and it supports a star network topology with one masterand seven slave nodes. For reducing the power consumption and expenses, ULPButilizes lower transmission power and lower symbol rate. It is expected frequencyhopping is not utilized, either. It is expected that the ULPB can reduce the powerconsumption of Bluetooth to one tenth.

ULPB may have a common RF part with Bluetooth making its integration into cel-lular phones and laptop computers cheaper. Yet, small devices, such as watches andsport sensors may utilize just a ULPB radio. Hence, ULPB can connect togethertwo market segments: devices having Bluetooth and simple devices for which Blue-tooth is too powerful and energy consuming. Yet, the applicability of ULPB in WSNapplications is limited by scalability and network coverage.

ANT

ANT [49] developed by Dynastream Innovations is a simple low data-rate and low-latency technology specifying PHY, MAC and NWK layers. ANT operates at 2.4GHz frequency band and has 1 Mbps radio data rate. ANT network is based on astar-topology, but more complex topologies can be achieved by using several chan-nels: each node can be simultaneously a master and a slave on different channels.Master nodes always receive, while slaves transmit when new data is provided. Apractical limit for network size is few thousands nodes. The disadvantages of ANTare high power consumption in master nodes and low scalability due to random accesstransmissions. These limit the usability of ANT in large multi-hop WSN applications.

WirelessHART

WirelessHART [67] is an open wireless communication standard ratified in 2007by the HART Communication Foundation [67]. WirelessHART is specifically de-signed for process measurement and control applications having stringent require-ments for end-to-end communication delay, reliability, and security [176]. Wire-lessHART utilizes a time-synchronized TDMA MAC on top of the IEEE 802.15.4physical layer. MAC employs network wide time synchronization, channel hopping,channel blacklisting, and AES encryption. WirelessHART utilizes a self-organizingmulti-hop mesh network with centralized control. For guaranteeing network per-formance, a central network manager is responsible for route updates and commu-nication scheduling for entire network. The suitability of WirelessHART on WSNapplications is limited. The centralized control of TDMA schedule and routes limits

2.3. Standards 21

network size and the tolerance against network dynamics. The centralized controlcauses a significant control frame overhead increasing network energy consumption.

ISA-SP100.11a

ISA-SP100.11a [83] is an emerging standard for non-critical process industry appli-cations, where tolerable delays are in the order of 100 ms. Target applications includesensors, valves and actuators having low data rate requirements [197]. The standardis expected to be published during 2008 [213]. Later versions of the standard willalso address factory automation and building automation [83]. Similarly with Wire-lessHART, ISA100.11a utilizes time synchronized TDMA and frequency hoppingwith channel blacklisting on top of the IEEE 802.15.4 physical layer. ISA-SP100.11asupports multi-hop routing in a mesh topology and battery powered routers. In con-trast to WirelessHART, ISA-SP100.11a specifies utilization of multiple gateways andflexible TDMA slot length for improving data routing performance [22]. It is ex-pected that ISA-SP100.11a will be a suitable technology for some WSN applications,too. The usability in WSN applications is possible limited by scalability, energy con-sumption, and tolerance against network dynamics.

IEEE 1451 Standard Family

IEEE 1451 is a suite of smart transducer interface standards, which describes com-munication interfaces for connecting sensors and actuators (transducers) to micro-processors, instrumentation systems, and networks.

IEEE 1451 defines two terms: Transducer Interface Module (TIM) and NetworkCapable Application Processor (NCAP). TIM is a device, which contains a set oftransducers, signal conditioning and data conversion circuitry, and software modules.They consist of IEEE 1451 standard modules and a standardized wireless or wirednetwork technology. NCAP is any kind of network-connected computing device,which receives data from a set of TIMs.

IEEE 1451.0 standard defines the functional specification of TIM, the discovery andmanagement of TIMs, and a set of sensor API calls with message exchange protocolsand commands required for interfacing with transducers. In addition, IEEE 1451.0defines a Transducer Electronic Data Sheet (TEDS), which is used to describe theentire TIM including transducer, signal conditioner and data converter. Hence, TEDSeliminates error prone, manual entering of data and system configuration and allowstransducers to be installed, upgraded, replaced or moved by plug-and-play principle.


IEEE 1451 supports numerous standardized technologies to establish wired and wire-less connections between NCAP and TIMs by IEEE 1451.2 through IEEE 1451.6.IEEE 1451.2 defines wired point-to-point communication through UART or a Trans-ducer Independent Interface (TII). IEEE 1451.3 defines distributed multi-drop sys-tem, where a large number of TIMs may be connected along a wired multi-drop bus.IEEE 1451.4 specifies mixed-mode communication protocols, which carry analogsensor values with digital TEDS data. IEEE 1451.6 defines a high-speed ControllerArea Network (CAN) bus. IEEE 1451.5 standard [77] defines wireless sensors andthus, it is most closely related with WSNs. Supported communication technologiesare IEEE 802.11a/b/g, IEEE 801.15.1 and IEEE 802.15.4.

Conclusion of Standards

Current standards for wireless communications are diverse, but none of these coverthe entire WSN application space [93]. Currently, ZigBee and its variations are themost potential standardized technologies for WSNs. Their feasibility is mostly lim-ited by the energy consumption of routers, scalability, and inadequate performancein dynamic networks. Yet, standards are improving and new standards are emergingcontinuously. One of the most interesting upcoming standards is the ISA-SP100.11a,which will most probably cover most of the application areas of ZigBee.

One of the most difficult challenges in WSN is to develop energy-efficient MAC layerprotocols for very large and dynamically changing networks. This area is not coveredby the current standards and the upcoming standards released in the near future. Thisarea has been selected as the main focus in this Thesis.

3. SENSOR NODE PLATFORMS

Sensor node platforms implement the physical layer (hardware) of the protocol stack.In conventional wireless networks, such as WLANs, hardware affects very signif-icantly on the achieved performance and the energy consumption of network ter-minals. In WSNs, the energy-efficiency is implemented mostly by the MAC layer,which at best can reduce the activity of a hardware to below 1% in low data-ratemonitoring applications. Clearly, it is important to minimize hardware power con-sumption in active operation modes. Even more important is to minimize the powerconsumption in idle and sleep modes, which may dominate the power consumptionand limit network lifetime in very low data-rate applications. Next, suitable low-power hardware components and existing sensor node platforms are discussed.

3.1 Platform Components

WSN applications typically necessitate small and cheap hardware realization havingthe battery lifetime in the order of years. The given requirements for are fulfilledbest by Application Specific Integrated Circuits (ASICs), which perform computationpowerfully and energy-efficiently by an application specific hardware [133]. Smallphysical size is achieved, since a single System-on-Chip (SoC) circuit contains allessential digital circuits. Due to high design and initial costs, and fixed hardware,ASIC suits best for implementing a mature and standardized technology having veryhigh production volumes. For WSNs, COTS components are often the most feasibleoption. From now on, the focus will be on COTS based hardware components.

A general hardware architecture of a sensor node platform is presented in Fig. 7. Thearchitecture can be divided into four subsystems:

Communication subsystem enabling wireless communication,

Computing subsystem allowing data processing and the management of nodefunctionality,

24 3. Sensor Node Platforms

Communication

subsystem

Computing

subsystem

Sensing

subsystem

Radio

transceiver

Radio

transceiver

Power subsystem

Energy storageEnergy storage Voltage regulatorVoltage regulator

MCUMCU

CoreCore

FlashFlash

SRAMSRAM

EEPROMEEPROMADCADC

TimerTimer

I/O portsI/O ports

Energy scavenging unitEnergy scavenging unit

ActuatorActuator

Digital

sensor

Digital

sensor

Analog

sensor

Analog

sensor

Power Data

Fig. 7. Sensor node hardware architecture.

Sensing subsystem connecting the wireless sensor node to the outside world,

Power subsystem providing the system supply voltage.

The central component of the platform is MCU that forms the computing subsystem.In addition, the protocols of communication and sensing subsystems are executed onMCU.

3.1.1 Communication subsystem

The communication subsystem consists of a wireless transceiver and an antenna. Awireless transceiver can be based on acoustic, optical or RF waves. Acoustic commu-nication is typically used for communication under water [92] or to measure distancesbased on time-of-flight measurements [15]. The disadvantages of acoustic commu-nication are long and variable propagation delay, high path loss, noise, and very lowdata rate, which limit the achieved energy-efficiency. Also, a large external antenna isneeded. In mobile networks, Doppler spread is significant reducing the data rate [92].

Optical communication [46, 216] has low energy consumption especially in recep-tion mode, and it can utilize very small antenna. A transmitter can be implementedby a Light Emitting Diode (LED) or a laser, and a receiver by a photo diode. How-ever, radiation is highly directional and a Line-of-Sight (LOS) condition is required.

3.1. Platform Components 25

Hence, the alignment of a transmitter to a receiver is difficult or even impossible inlarge-scale WSN applications [216].

RF communication combines the benefits of high data rate, long range and nearlyomni-directional radiation, which make it the most suitable for WSNs. Disadvan-tages are large antenna size, and higher energy consumption compared to opticaltechnology [46].

In general, an RF transceiver (radio) has four operation modes: transmit, receive, idle,and sleep. Radio is active in transmit and receive modes, when power consumptionis also the highest. In idle mode, most of circuitry is shut down, but main oscillatorsremain active. Thus, the transition to the active mode is fast. The lowest powerconsumption is achieved in sleep mode, when all circuitry is switched off. Also, thetransient time to the active mode is relatively long [P7].

Most short-range radios utilized with WSNs operate in the 433, 868, 915, and 2400MHz license-free ISM frequency bands. The operating frequency affects RF wavepropagation characteristics. In industry, the 2400 MHz band is beneficial, since theinterferences emitted from the machinery are significant below few hundred mega-hertz and drops rapidly above 1 GHz frequency [136]. The 2400 MHz band is alsothe widest providing more channels and typically higher data rate than the lower ISMbands. Depending on the frequency band and antenna type, operating range with 1mW Transmit (TX) power is from few meters to hundreds meters [P7]. The charac-teristics of the most potential commercial low power radios are summarized in Table6 [P7]. Microchip, Nordic Semiconductor, and Texas Instruments utilize on-chipbuffers for the adaptation of a high-speed radio with a low-speed MCU [P7]. Currentconsumptions are specified at the lowest band and 0 dBm transmission power. Thetable indicates that data rate and frequency band has only a low effect on current con-sumption. The last two paragraphs present the energy consumption of a received andtransmitted bit of data, which are determined using 3.0 V supply voltage. The com-parison indicates that radios operating at the 2.4 GHz frequency band are the mostenergy-efficient, which is mostly caused by their high data rates.

3.1.2 Computing subsystem

Computing subsystem consists of MCU, which integrates a processor core with pro-gram and data memories, timers, configurable I/O ports, ADC and other peripher-als. Program memory is typically Flash, while data memory consists of Static Ran-dom Access Memory (SRAM) and Electrically Erasable Programmable Read-Only


Table 6. Radio features, current consumptions, and energy efficiencies.Radio Data rate Band Buffer Sleep RX TX RX TX

(kbps) (MHz) (B) (µA) (mA) (mA) (nJ/b) (nJ/b)MC MRF24J40 [107] 250 2400 128 2 18 22 264 216NS nRF2401A [120] 1000 2400 32 0.9 19.0 13.0 39 57NS nRF24L01 [121] 2000 2400 32 0.9 12.3 11.3 17 18NS nRF905 [122] 50 433-915 32 2.5 14.0 12.5 750 840RFM TR1001 [141] 115.2 868 no 0.7 3.8 12 313 99RFM TR3100 [142] 576 433 no 0.7 7.0 10 52 36SE XE1201A [163] 64 433 no 0.2 6.0 11.0 516 281SE XE1203F [165] 152.3 433-915 no 0.2 14.0 33.0 650 276TI CC2400 [193] 1000 2400 32 1.5 24.0 19.0 57 72TI CC2420 [194] 250 2400 128 1 18.8 17.4 209 226TI CC2500 [189] 500 2400 64 0.4 17.0 21.2 127 102TI CC1000 [192] 76.8 433-915 no 0.2 9.3 10.4 406 363TI CC1020 [187] 153.6 433-915 no 0.2 19.9 16.2 316 389TI CC1100 [188] 500 433-915 64 0.4 16.5 15.5 93 99

Abbreviations: Microchip (MC), Nordic Semiconductor (NS), RF Monolithics (RFM), Semtech (SE),Texas Instruments (TI)

Memory (EEPROM). The characteristics of potential MCU from different manufac-turers are compared in Table 7. The energy-efficiencies of MCUs can be comparedaccording to their current consumption at one MIPS processing speed. Since the in-struction set affects the performance, only orders of magnitude are important [128].The comparison indicates that Semtech XE8802 and TI MSP430F1611 MCUs haveare the most energy-efficient [P7].

Table 7. The comparison of the features of low power MCUs.MCU FLASH SRAM EEPROM Sleep 1 MIPS

(kB) (kB) (B) (µA) (mA)Atmel AT89C51RE2 (8051) [11] 128 8 0 75 7.4Atmel ATmega103L (AVR) [9] 128 4 4096 1 1.38Atmel AT91FR40162S (ARM) [10] 2048 256 0 400 0.96Cypress CY8C29666 [41] 32 2 0 5 10Freescale M68HC08 [59] 61 2 0 22 3.75Microchip PIC18LF8722 [106] 128 3.9 1024 2.32 1.0Microchip PIC24FJ128 [108] 128 8 0 21 1.6Semtech XE8802 (CoolRisc) [164] 22 1 0 1.9 0.3TI MSP430F1611 [190] 48 10 0 1.3 0.33

3.1. Platform Components 27

3.1.3 Sensing subsystem

Sensing subsystem consists of a set of sensors and actuators. The sensors observephenomena such as thermal, optic, or acoustic event. They are equipped with ananalog or digital interfacing for reading sensor values [3].

There exists a large variety of low power sensors suitable for WSNs. For example,sensors are available for acceleration, air pressure, humidity, illumination, infra-red,magnetic field, geographic position, and temperature. Important requirements forsensors are low power consumption and short sensing time, which determine the en-ergy consumption of a single sensing. In addition, adequate accuracy is requiredwithin the entire temperature range. The features of some example sensors are pre-sented in Table 8. Most of the sensors fulfill the requirements well. The suitability ofa GPS position sensor for WSNs is questionable due to high energy consumption.

A WSN node can also operate as a decision unit, which takes sensor readings fromthe WSN as input and generates action commands as output. These action commandsare then transformed into actions by actuators [3]. Besides an electric switch and aservo drive, an actuator can be a mobile robot. In order to improve the reliability ofactions, the robot can be a WSN node and act based on its own sensor readings andthe data of the other WSN nodes in the network [3, 45].

3.1.4 Power subsystem

The power subsystem stores supply energy and converts it to an appropriate supplyvoltage level. The subsystem consists of an energy storage, a voltage regulator, andpossible an energy scavenging unit.

Table 8. Features of typical sensors.Physical Example sensor Accuracy Active Sensing Energyquantity current time cons.Acceleration VTI SCA3000 [204] 1% 120 µA 10 ms 3.6 µ JAir pressure VTI SCP1000 [205] 150 Pa 25 µA 110 ms 8.3 µ JHumidity Sensorion SHT15 [167] 2% 300 µA 210 ms 190 µ JIllumination Avago APDS-9002 [12] 50% 2.0 mA 1.0 ms 6.0 µ JInfra-red Fuji MS-320 [60] - 35 µA cont. -Magnetic field Hitachi HM55B [74] 5% 9.0 mA 30 ms 810 µ JPosition Fastrax iTRAX03 [56] 1.0 m 32 mA 4.0 s 380 mJTemperature Dallas DS620U [42] 0.5C 800 µA 200 ms 480 µ J


The energy storage can be a non-rechargeable (primary) battery, a rechargeable (sec-ondary) battery, or a supercapacitor. Primary batteries are cheap and have the highestenergy density. They are the most common power source for WSNs. Secondary bat-teries have lower energy density and are more expensive, but they can be recharged500 - 1000 times. Secondary batteries are suitable, if batteries can be rechargedoccasionally. Generally, supercapacitors are capacitors having very high capaci-tance. Compared to secondary batteries, supercapacitors have lower energy densityand they are more expensive. However, their lifetime is in the order of a millioncharging/discharging cycles. Supercapacitors are suitable to be used with an energygenerator, since energy is typically generated in peaks during short periods of time,and the amount of stored energy can be relatively low.

The most potential sources for energy scavenging are photovoltaics and vibrations[149]. Solar cells are mature technology and they can provide up to 15 mW/cm3

power at outdoor conditions. In indoor conditions, achieved power reduces to 10µW/cm3. A promising method for converting vibration to source power is a piezo-electric conversion. Commonly occurring vibrations can provide up to 200 µW / cm3

power [150]. Other possible energy sources are temperature gradients (40 µW/cm3 at5 C temperature differential) [127, 180] and air flow (380 µW/cm3 at 5 m/s) [149].

A voltage regulator stabilizes system supply voltage to an appropriate and constantlevel. In order to minimizing static current consumptions, it is reasonable to use theminimum supply voltage specified by system components. The scarce energy budgetsets strict requirements also for the voltage regulator. A typical current consumptionwaveform measured from two Tampere University of Technology Wireless SensorNetwork (TUTWSN) nodes is presented in Fig. 8. The nodes are typically over 99%of time in a sleep mode. The regulator must be able to supply short peaks of relativelyhigh, tens of milliamperes currents, while the quiescent current consumption must bevery low.

Voltage regulators can be divided into linear and switched-mode regulators. Linearregulators control the output voltage by adjusting the voltage drop across a seriespower transistor, which is located between the unregulated input and the regulatedoutput voltage. This transistor conducts continuously and the energy of voltage dropis converted to heat. Thus, linear regulators are simple, cheap, small and have verylow electromagnetic interferences and quiescent current [115].

Switched-mode step-down type regulators convert and input voltage to a lower out-put voltage by a switching transistor that is opened and closed periodically. Then, theswitching current is fed to a simple coil-capacitor (L-C) filter and a diode, which aver-

3.2. Existing Platforms 29

Subnode

Headnode

10 μA

20 mA

Fig. 8. Measured current waveform from TUTWSN subnode and headnode.

age the output voltage. Thus, the energy loss of the regulator is theoretically zero. Inpractice, some energy is consumed in the voltage drops, resistances and leakage cur-rents of the switching transistor, diode, coil and capacitor, and for the supply power ofa controller circuitry. Compared to linear regulators, switching mode regulators aremore expensive and larger, and have significantly higher electromagnetic emissionsand quiescent current [115].

Switched-mode regulators provide higher energy-efficiency, when the supply currentand dropout voltage are high. Yet, in WSN applications linear regulators are moreenergy-efficient due to the low average supply current levels. For comparing theperformance of linear (L) and switched-mode (S) regulators, Table 9 presents thefeatures of some of the lowest power regulators available on the market.

3.2 Existing Platforms

Sensor node platforms have been improved significantly during the last decade alongwith the advances in low power processing and communication technology. Due tothe strict energy constrains, and the visions of complex networking and data fusion, itis very challenging to fulfill all the requirements with the current level of technology.Thus, the platform research can be divided into two branches: high performanceplatforms, and low power platforms [72].


Table 9. Features of low-power voltage regulators.Regulator Type Quiescent Max. load Max. input Dropout

current (µA) (mA) (V ) (mV )Maxim MAX1725 [102] L 2.0 20 12.0 300Maxim MAX8880 [103] L 3.5 200 12.0 350Microchip MCP1702 [110] L 2.0 250 13.2 330Minilogic ML62 [113] L 2.8 150 10 800SII S-1206 [162] L 1.0 250 6.5 350TI TPS71501 [195] L 3.2 50 24 415Analogic Tech AAT1112 [1] S 42 1500 5.5 200Microchip MCP1603 [109] S 45 500 5.5 250TI TPS62000 [191] S 50 600 5.5 230

The high performance platforms have been developed for researching complex dataprocessing and fusion in sensor nodes. The design target has been the reduction oftransmitted data by efficient data processing. These platforms utilize high perfor-mance processors having at least tens of MIPSs processing performance and hun-dreds of kbytes program and data memories. For long-lived battery operation, theirenergy consumption is not adequate. However, these high power platforms can beused as a part of a WSN for data processing and data routing. Examples of the highperformance platforms are Piconode [139], µAMPS [112], and Stargate [35].

Low power platforms are aiming to maximize the lifetime and minimizing the physi-cal size of nodes. These are obtained by minimizing hardware complexity and energyconsumption. These platforms are capable for performing low data rate communica-tion and data processing required for networking and simple applications. Next, themost essential sensor node platforms are discussed.

The most well-known work among low power sensor node platforms has been con-ducted in the University of California, Berkeley. They have built a number of plat-forms called motes by a SmartDust project. The SmartDust [207] project was aimingto fabricate a cubic-millimeter sized WSN node containing sensing, computing andcommunication systems. An important goal of the project was to explore the lim-itations of microfabrication technology. Before that, Mote platforms were built toapproximate the capabilities of an envisioned SmartDust node with COTS compo-nents.

In 2001, the first mote called Mica [73] was released. Mica is a general purposeplatform for WSN research having Atmel ATmega103L MCU and RFM TR1000 ra-dio that operates at 915 MHz frequency band and has up to 115.2 kbps data rate.The MCU contains 128 kbytes program memory and 4 kbytes data memory, and


has 4 MIPS processing speed. In addition, the platform contains a 4-Mbit externalnon-volatile memory and Atmel AT90LS2343 co-processor for handling wireless re-programming of the main processor. The most critical shortcoming of the platformis a switched-mode voltage regulator, which uses 200 - 300 µA quiescent currentreducing significantly achieved node lifetime [130].

An improved version of Mica called Mica2 [37] was released in 2002. Sleep modecurrent consumption was reduced to about 17 µA by discarding the switched-moderegulator and utilizing Atmel ATmega128L MCU. The radio was changed to ChipconCC1000 enabling tunable frequency channel from 300 MHz to 900 MHz. A smallerversion of Mica2 called Mica2Dot [38] was also released in the same year. Thesetwo platforms became very popular in WSN research for many years.

MicaZ [34] replaces the CC1000 radio with an IEEE 802.15.4 compatible ChipconCC2420 radio. This modification increased radio data rate from 76.8 kbps to 250kbps. In addition, a wide-band DSSS modulation scheme provided better toleranceagainst noise and interferences.

Telos [130] is a new type of mote released in 2004. Telos combines the CC2420radio with a very low power TI MSP430 MCU. A special feature is an UniversalSerial Bus (USB) connector, which is used for programming the platform. Sleepmode current consumption is only 5.1 µA. An improved version of Telos is calledTmote Sky [114].

Also other research projects have been developing sensor node platforms intensively.Medusa MK-2 [156] is a versatile platform released in 2002 by University of Califor-nia, Los Angeles. The platform combines ATmega128L MCU with TR1000 radio.Hence, it is quite similar to Mica and Mica2 motes. A special feature of MedusaMK-2 platform is a higher performance 16/32-bit Atmel AT91FR4081 ARM7TDMIco-processor, which is responsible for the processing of sensor data and localizationalgorithms. The platform is equipped with light, temperature, and acceleration sen-sors. The sleep mode current consumption is 27 µA, which is adequate for long-livedWSN applications.

ETH Zurich has developed a sensor node platform called BTnode in a Smart-Itsproject [174]. A BTnode combines Bluetooth radio with the ATmega128L MCU.A BTnode rev3 [18], released in 2004, is a dual-radio platform, which can employsimultaneously a Zeevo ZV4002 Bluetooth radio and the CC1000 radio. Hence, theBTnode rev3 platform can form a backbone network with Bluetooth technology, andexchange data with low power Mica2 and Mica2Dot platforms by the CC1000 ra-dio [72]. A disadvantage is that typical sleep mode current consumption is 3 mA [23],


which limits the achievable node lifetime very significantly.

A ScatterWeb Embedded Sensor Board (ESB) [157] is a simple and low power plat-form developed by Freie Universität Berlin, Germany in 2003. The platform consistsof the MSP430 MCU, RFM TR1001 radio and sensors for measuring light, acousticnoise, vibration, and movement by a Passive Infra-Red (PIR) sensor. In addition, amicrophone, a speaker, and an Infra-Red (IR) transceiver are included. Long-livedWSN applications are possible, since the sleep mode current consumption is only8 µA. Communication with external network is enabled by various gateway nodeshaving interfaces, for example, to Ethernet, GSM/GPRS, USB and RS-485. A verysimilar hardware architecture is employed also in nodes [140] developed by an Euro-pean EYES research project.

A TinyNode [47] developed by Shockfish SA, Switzerland, is a very small and lowpower platform developed for research and industrial applications. The platform con-tains only the core components, while additional functionality and sensors are pro-vided by extension boards. In addition, the platform contains a 512 kbytes externalnon-volatile memory for storing several firmware images and for logging data. In-cluded main components are MSP430 MCU, Semtech XE1205 radio and a temper-ature sensor. The radio can operate in 433 MHz, 868 MHz and 915 MHz frequencybands having up to 152 kbps data rate. Long lifetime is possible, since the sleep modecurrent consumption is only 5.1 µA .

A ProSpeckz [8] is a very small and simple platform, which is equipped with aCY8C29666 Programmable System-on-Chip (PSoC) from Cypress Microsystems.PSoC combines a simple 8-bit MCU core having 16 kbytes of program memory and256 bytes of data memory with versatile software reconfigurable analogue circuits toexternal interfaces and components. Employed radio is the TI CC2420. Although,the PSoC platform provides versatile processing capabilities for analog sensor values,high 330 µA current consumption in sleep mode reduces significantly the achievablelifetime.

Besides the COTS platforms presented above, a lot of research work has been con-ducted for developing SoC platforms targeting to even smaller size and higher energy-efficiency. For example, a WiseNET SoC sensor node [54] developed in Swiss Cen-ter for Electronics and Microtechnology integrates a low-power radio with CoolRISCMCU core, low-leakage memories, two ADCs and power management blocks. Thereception mode current consumption is only 2 mA, which is nearly one order of mag-nitude less than in typical low power radios. Yet, a data rate is only 25 kbps. Thetransmission mode current consumption at 10 dBm output power is 24 mA. The sleep


mode current consumption of the radio block is 3.5 µA.

The characteristics of the presented sensor node platforms are summarized in Table10. At best, low power platforms can perform various sensing tasks and they enablethe extending of network lifetime to even years. This necessitates an energy-efficientMAC protocol, which maximizes the time node spends in the sleep mode. This willbe discussed in details in the next Chapters.

In the future, data processing and sensing tasks within the network will be increasedand they become more complex. In the component side, this necessitates the devel-opment of processors having high performance, energy-efficiency, and large memoryresources. Also, the development of diverse low power sensor elements is importantfor enabling the application development.

This Thesis will present a series of sensor node platforms, which further improvethe achieved network lifetime by focusing on the minimization of radio energy con-sumption. Also, various sensor integrations, interfaces to external networks, andapplication experiments will be presented.


Table10.C

omparison

ofexistinglow

-power

sensornode

platforms.

PlatformM

CU

SensorsR

adioD

atarate

(kbps)R

Fband

(MH

z)Sleep

(µA)

Size(m

m2)

Mica

[73]A

Tm

ega103Lno

TR

1000115.2

915200-300

1856M

ica2[37]

AT

mega128L

noC

C1000

76.8433-915

171856

Mica2dot[38]

AT

mega128L

noC

C1000

76.8433-915

17492

MicaZ

[34]A

Tm

ega128Lno

CC

2420250

240030

1856B

Tnode

ver3[23]

Atm

ega128Lno

ZV

40021000

24003000

1890+

CC

1000+

76.8+

433-915M

edusaM

K-2

[156]A

Tm

ega128LT,L

,AT

R1000

115.2915

27ca.4500

+A

RM

7E

YE

Snode

[140]M

SP430no

TR

1001115.2

8685.1

ca.2600ScatterW

ebE

SB[157]

MSP430

L,A

C,A

,PIRT

R1001

115.2868

8ca.3000

Tinynode

[47]M

SP430no

XE

1205152

433-9155.1

1200T

mote

sky[114]

MSP430

H,T,L

CC

2420250

24005.1

2621ProSpeckz

[8]C

Y8C

29666no

CC

2420250

2400330

704

Abbreviations:A

cceleration(A

),Acoustic

(AC

),Hum

idity(H

),Light(L

),PassiveInfra-R

ed(PIR

),Temperature

(T)

4. MEDIUM ACCESS CONTROL FOR WSNS

A MAC layer operates on top of the physical layer and manages radio transmissionsand receptions on a shared wireless medium. Hence, it has a very high effect onthe performance and energy consumption of a network. This chapter discusses MACdesign requirements and existing MAC protocols for low-energy WSNs.

4.1 Low-Energy MAC Design

As WSNs are designed to operate in large geographic areas, the energy consumptionof data transmissions is reduced by forwarding data in the network by several lowenergy hops (multi-hop routing). According to a distance-dependent path loss model[161], the mean path loss (PL) increases exponentially with distance (d)

PL(d) ∝(

dd0

)n

, (1)

where n is the path loss exponent, d0 is the reference distance, and d is the ac-tual distance between the transmitter and the receiver. The value of the path lossexponent is typically between 2 and 3.5 depending on the operation environment[68, 137, 138, 161]. Thus, the reduction of a hop length to a half reduces the requiredtransmission power to between 1/4 and 1/12. Yet, this model considers only the radi-ated transmission power and ignores the static power consumption of the transmitterand receiver circuitry, which dominate the power consumption at short hop distances.Thus, the advantage of multi-hop routing is limited [44, 226].

To be able to reach adequate energy-efficiency, a MAC protocol should be able tominimize all unnecessary activity of a radio. Most importantly, a MAC protocolshould minimize [17, 88, 134, 207, 223]:

Idle listening: Idle listening occurs when a node is actively receiving a chan-nel, but there is no meaningful activity on the channel resulting wasted energy.

36 4. Medium Access Control for WSNs

Collisions: When two nodes transmit simultaneously at the same frequencychannel their transmissions collide in the overlapping area of their transmis-sion ranges. In this area the received data is most probably corrupted causinguseless receive cost at the destination node and useless transmit cost at thesource node.

Overhearing: An unicast transmission on a shared wireless broadcast mediummay cause other nodes than the intended destination to receive a data packet,which is most probably useless to them and consumes unnecessarily energy.

Protocol overhead: The headers and trailers of data packets, and the controlpacket exchange may cause significant reception and transmission cost for allnodes consuming unnecessarily energy.

Start-up transients: The transient from sleep to active mode, and switchingbetween active modes dissipate a lot of energy causing wasted energy.

As a principle, the radio should be activated only when transmitting or receivinga packet that is vital for the node operation. Moreover, a source and a destinationnode should be activated and tuned on a correct RF channel simultaneously, whileother nodes remain in the sleep mode. This is very difficult in large networks, whereoperation conditions are uncertain and computing resources are constrained. Next,the operation principles of the most essential MAC protocols are discussed.

4.2 Existing MAC Protocols

MAC protocols can be categorized into contention and contention-free protocols. Incontention protocols, nodes compete for a shared channel, while trying to avoid framecollisions.

ALOHA [147] is the simplest contention protocol, where nodes transmit data with-out coordination. Slotted ALOHA reduces collisions by dividing time into slots andtransmitting data on the slot boundaries only. CSMA [91] further reduces collisionsand improves achievable throughput by checking channel activity prior to transmis-sions and avoiding transmission during busy channel situations. Carrier Sense Mul-tiple Access with Collision Avoidance (CSMA/CA) [33] is a modification of CSMA,which reduces the congestion on a channel by deferring a transmission for a randominterval, if the channel is sensed to be busy (backoff). Still, collisions may occurdue to a hidden node problem [198]: nodes separated by two hops may not detect

4.2. Existing MAC Protocols 37

each other, and their transmissions may collide on the receivers of their intermedi-ate nodes. Multiple Access with Collision Avoidance (MACA) [89] reduces hid-den node collisions by performing a Request-To-Send (RTS) - Clear-To-Send (CTS)handshaking prior to a data transmission. This handshaking is also defined as anoption in CSMA/CA. While contention-based protocols work well under low traf-fic loads, their performance and reliability degrades drastically under higher loadsbecause of collisions and retransmissions.

In contention-free protocols, nodes get unique time slots or frequency channels fortransmissions. Ideally, collisions are eliminated. Time Division Multiple Access(TDMA) [228] divides time into numerous slots, where only one node is allowedto transmit on each slot. Other alternatives are Frequency Division Multiple Access(FDMA) [2] and Code Division Multiple Access (CDMA) [85], which provide contention-free operation by separate frequency channels and spreading codes, respectively.

Contention-free protocols achieve high performance and reliability regardless of thetraffic load. Yet, the bandwidth should be reserved in advance, which increases con-trol traffic overhead [30, 84, 100, 186, 228].

Due to the power consumption of current low-power radios especially in the receptionmode, the energy-efficiency of the conventional MAC approaches is not adequate forthe low energy WSN as such. Further energy saving is achieved by duty cycling: timeis divided into a short active period and a long sleep period, which are repeated con-secutively. These low duty-cycle protocols can also be divided into two categories:unsynchronized and synchronized protocols, according to the synchronization of dataexchanges.

4.2.1 Unsynchronized Low Duty-Cycle MAC Protocols

Unsynchronized low duty-cycle MAC protocols are based on a Low Power Listening(LPL) mechanism, where nodes poll channel asynchronously to test for possible traf-fic, as presented in Fig. 9. Transmissions are preceded with a preamble (Tpreample)that is longer than the channel-polling interval (Tinterval). Hence, the preamble partacts like a wake up signal. If a busy channel is detected, nodes begin to listen to thechannel until a data packet is received or a time-out occurs.

B-MAC [129] is a simple LPL protocol, which utilizes CSMA for collision avoid-ance. The energy-efficiency of B-MAC is significantly limited by the transmissionand reception energy costs due to the long preamble. In addition, energy-efficiency


Source

DestinationRx

Tx

Rx

Tx

Tpreamplec

Preample

Channel sampling Data frame Carrier sensingc

time

time

Wait for data

Tinterval

Tinterval

Fig. 9. Operation of unsynchronized low duty-cycle protocols.

is reduced by overhearing of frames intended to other nodes and idle listening causedby the frequent channel sampling.

Z-MAC [143] operates above B-MAC, but utilizes TDMA for managing congestion.As a principle, each node owns a slot during which a smaller contention window isused compared to other nodes. Thus, the slot owner always has the best possibility toaccess the channel. Under low contention, Z-MAC behaves like CSMA, and underhigh contention more like TDMA. The utilization of slots improves the fairness andthroughput of B-MAC. Yet, the improvement on energy-efficiency is only limited.

There are numerous variations of B-MAC targeting at the reduction of the pream-ple energy. Speck Medium Access Control Backoff (SpeckMAC-B) [217] replacesthe long preamble with numerous short wake up packets containing a destinationaddress and an exact time to the actual data transmission. Thus, nodes may returnto sleep mode after receiving one wake up packet. Speck Medium Access ControlData (SpeckMAC-D) [217] replaces the long preamble with consecutive data packetsreducing the required channel reception time. In X-MAC [24], a sender transmitsmultiple short preambles with the address of the intended receiver. Each preample isfollowed by a short reception period. Upon receiving a preamble, the desired desti-nation node sends an Acknowledgment (ACK) between the preambles. Other nodescan enter early a sleep mode for reducing overhearing. After receiving the ACK, thesource node begins the transmission of a data frame. Disadvantages of these pro-tocols are the transmission cost of a preample and idle listening caused by CSMAmechanism, channel polling, overhearing and radio start-up transients.

There are two variations of B-MAC, which reduce preample energy by utilizing syn-chronization. Thus, they can also be categorized into the group of synchronized pro-tocols. They are presented here, since their operation principle is very similar with B-MAC and other unsynchronized protocols. Wireless Sensor MAC (WiseMAC) [51]


utilizes ALOHA for transmissions. A network consists of an access point and numer-ous sensor nodes in a star-topology. The access point learns the sampling schedules ofeach sensor node and starts preamble transmission just prior to the channel samplingmoment of a desired destination node. Major disadvantages of the protocol are verylimited coverage and connectivity of the network due to the star-topology. ScheduledChannel Polling Medium Access Control (SCP-MAC) [221] is a synchronized vari-ation of B-MAC, which operates in a mesh network by making the channel pollingschedules of all nodes identical. Hence, only a short preamble is required to reach allneighbors. Synchronization is performed by transmitting periodically synchroniza-tion packets containing the schedule information, or piggybacking the information isdata packets. SCP-MAC is currently the most energy-efficient unsynchronized lowduty-cycle protocol. Still, the idle listening during a backoff mechanism and chan-nel polling, collisions, frequent radio start-up transients, and overhearing reduces itsenergy-efficiency.

The preample can be eliminated by utilizing a wake-up radio [63]. The wake-up radiomechanism is based on the assumption that the listen mode of the wake-up radio is ul-tra low power and it can be active constantly. At the same time, the normal data radiois in the sleep mode as long as there is no packet transmission or reception required.Power Aware Multi-Access protocol with Signaling (PAMAS) [173] protocol enablecollision avoidance mechanism by utilizing RTS-CTS handshaking in the wake-upradio channel. Sparse Topology and Energy Management (STEM) [159] protocol re-duces the energy consumption of a wake-up radio by the LPL scheme. The wake-upradio protocols are successful in avoiding overhearing and idle listening in the dataradio. Their major problems are the energy consumption and cost of the wake-up ra-dio. In addition, the difference in the transmission ranges between data and wake-upradio may pose significant problems.

Unsynchronized protocols are relatively simple and robust, and require a small amountof memory compared to synchronized protocols. A general drawback is rather highoverhearing, since each node must receive at least the beginning of each frame trans-mitted within radio range. Thus, they suit best for relatively simple WSNs utilizingvery low data rates. Unsynchronized protocols tolerate dynamics in networks, buttheir energy-efficiency is limited by the channel sampling mechanism [223].

4.2.2 Synchronized Low Duty-Cycle MAC Protocols

Synchronized low duty-cycle MAC protocols utilize scheduling to ensure that listen-ers and transmitters have a regular, short active period (Tactive) in which to exchange


Source

DestinationRx

Tx

Rx

Tx

Tactive

c c

wakeup_period

Tsleep

Active period Beacon Data frame Carrier sensingc

time

time

T

Fig. 10. Operation of synchronized low duty-cycle protocols.

one or more frames with neighbors. The active period is followed by a sleep pe-riod (Tsleep) for enabling energy saving. The active and sleep periods are repeatedat Twakeup_period intervals, as illustrated in Fig. 10. The active period is denoted asa superframe, when a node typically first broadcasts a control frame called beaconfor signaling its schedule and status information, and then listen to the channel forpossible incoming data until the end of the active period. Due to the synchronizedoperation, nodes know the exact moments of active periods in advance eliminatingthe need of long preambles. As a global synchronization is very difficult in large net-works [99], synchronization is maintained locally by receiving periodically beaconsfrom one or more neighboring nodes.

For establishing the synchronized operation, neighboring nodes are typically discov-ered by a network scan. The network scan means a long-term reception of frequencychannels for receiving beacons from neighbors, since their schedules and frequencychannels are unknown. Clearly, this is energy-hungry. However, the synchronizedoperation after the network scan is very energy-efficient [221, 223].

S-MAC [220] is one of the first synchronized low duty-cycle MAC protocols. Theprotocol utilizes a fixed active period length and an adjustable, network specific wakeup period. Neighboring nodes may coordinate their active periods to occur simulta-neously to form virtual clusters. At the beginning of an active period nodes wakeup and exchange Synchronization (SYNC) frames for synchronizing their operation.Then, nodes having data to be send transmit RTS messages using CSMA/CA mecha-nism. According to received RTS messages, a destination node select a desired sourcenode and sends a CTS message. Then, the source and destination nodes continue byexchanging data and ACK frames at most until the end of the active period. Theenergy-efficiency of S-MAC is reduced by long SYNC and RTS phases, and fixedactive period length causing idle listening. In addition, the fixed duty cycle causes


poor adaptation to changing traffic conditions.

Timeout-MAC (T-MAC) [201] protocol is a variation of the S-MAC, which utilizes ashort listening window after the CTS phase and each frame exchange. If no activityoccurs during the listening window, node returns to sleep mode. Thus, the length ofthe active period is adjusted according to traffic. Still, the energy-efficiency is limitedby the idle listening in SYNC and RTS phases.

DMAC [61] is a variation of S-MAC, which staggers the active/sleep schedules of thenodes in a tree topology. This allows continuous low latency packet forwarding flowfrom nodes to a sink. DMAC has mechanisms for notifying nodes of data deliveryprogress and duty-cycle adjustments.

Several variations of TDMA are also proposed for low-energy WSN. Low-EnergyAdaptive Clustering Hierarchy (LEACH) [71] protocol uses TDMA with clusterednetwork topology. LEACH utilizes a single base station, with which all cluster headsemploy direct communications. Inter-cluster interferences are managed by CDMA.The problems of direct communication are limited network coverage and low energy-efficiency caused by high transmission power levels. However, cluster members op-erate quite energy-efficiently. For increasing network lifetime, LEACH proposes tocompress data in cluster heads and to rotate the role of a cluster head among clustermembers.

Self-Organizing Slot Allocation (SRSA) [218] protocol is a variation of LEACH,which manages inter-cluster interferences by a distributed time division algorithm.Time slot allocations for communication are adjusted according to observed colli-sions within a cluster. Reliance only on the local information ensures high scalability.Still, the problems with direct communication are not solved.

Power Aware Clustered TDMA (PACT) [125] protocol is a variation of LEACH,which performs multi-hop data routing between clusters by inter-cluster gatewaynodes. Disadvantages of PACT are high control traffic overhead and idle listeningin larger networks. Relatively complex data slot scheduling algorithm performs wellin static networks, but lacks support for dynamic network.

Self-Organizing Medium Access Control for Sensor Networks (SMACS) [175] pro-tocol assigns a locally unique contention-free slot for each link. Neighbor discoveryis performed at semi-regular intervals by broadcasting invitation messages on a com-mon signaling channel. Then, the channel is received for possible responses and otherinvitation messages. According to invitation messages, each pair of nodes mutuallyagrees a periodic time and frequency slot for data exchanges. A major disadvantage


is the energy consumption of a neighbor discovery requiring a long-term radio re-ception. This severely limits energy-efficiency and adaptivity in dynamic networks,where link lifetimes are short.

Traffic-Adaptive Medium Access (TRAMA) [135] is a scalable TDMA protocol de-signed for multi-hop networks. Time is divided into slots. A distributed algorithmselects only one sender for each slot per a two-hop neighborhood allowing collision-free operation. A sender can command a set of neighbors to receive a given data frameproviding efficient unicast, multicast and broadcast transmissions. Nodes that are notselected as senders or receivers at a particular time slot go to a sleep mode. Neighborinformation is updated during periodic random access periods. TRAMA can providecollision-free medium access in a static network. Energy-efficiency is reduced bysignaling traffic overhead and the random access periods requiring a long-term radioreception. Hence, the energy-efficiency and performance decrease significantly indynamic networks.

IEEE 802.15.4 [82] standard defines a MAC layer that can operate on beacon-enabledand non-beacon modes. In the non-beacon mode, a protocol is a simple CSMA/CA.Energy-efficient synchronized low duty-cycle operation is provided by the beacon-enabled mode, where all communications are performed in a superframe structure.The superframe is divided into three parts: the beacon, Contention Access Period(CAP) and Contention-Free Period (CFP). CAP is a mandatory part of a superframeduring which channel is accessed using a slotted CSMA/CA scheme. CFP is anoptional feature of IEEE 802.15.4 MAC, in which a channel access is performed indedicated time slots. CFP can be utilized only for a direct communication with aPAN coordinator. Thus, its applicability and benefits are very limited in multi-hopnetworks.

IEEE 802.15.4 supports a BatteryLifeExtension option, which effectively reducesthe coordinator energy consumption by minimizing the lengths of contention win-dow and CAP. Due to the shortened CAP, this option significantly reduces availablethroughput. In addition, a collision probability is high due to shorter backoff time.Thus, this option is suitable only for small and very low data rate networks [P6].

The cluster-tree type IEEE 802.15.4 network can provide comparably good energy-efficiency in static and sparse networks. A major disadvantage is that coordinatorsmust be active entire CAP causing significant idle listening. In addition, the hiddennode problem reduces performance in dense networks, since any handshaking priorto transmissions are not used.

In current synchronized low duty-cycle protocols, the major advantage is that a sender


knows a receiver’s wake up time in advance and thus can transmit efficiently. Indynamic networks, synchronized links are short-lived and new neighbors need tobe searched frequently, which increases energy consumption rapidly. In contentionprotocols, a major disadvantage is the energy cost of receiving an entire active period[129]. Contention-free protocols have better energy-efficiency is stationary networks,but their performance reduces rapidly as network dynamics increases.

The features and major contributions of the presented MAC protocols are summa-rized in Table 11. The main problem of existing MAC protocols is the energy con-sumption of idle listening, which is mostly caused by channel polling, preamplereception, contention mechanism, and neighbor discovery. Idle listening is furtherincreased in WSN applications, where link quality varies and nodes are mobile. Inaddition, overhearing and frequent start-up transients limit the energy-efficiency ofunsynchronized protocols. Hence, the energy-inefficiency of existing MAC protocolslimits significantly network lifetime and disables many potential application areas.This Thesis presents a novel MAC protocol, which achieves higher energy-efficiencythan the existing protocols. In addition, networking protocols for maintaining energy-efficient operation in dynamic networks are presented.


Table11.C

omparison

oflowpow

erW

SNM

ACprotocols.

ProtocolN

etwork

#ofchannels

Channelaccess

Idlelistening

SynchronizationM

ajorcontributionstopology

avoidanceU

nsynchronizedB

-MA

CM

esh1

CSM

AL

PLN

oC

ombination

ofCSM

Aw

ithpream

plesam

plingPA

MA

SM

esh2

CSM

AW

ake-upradio

No

Wake-up

radiow

ithR

TS-C

TS

SCP-M

AC

Mesh

1C

SMA

SynchronizedL

PLY

esL

PLw

itha

synchronizedsleep

scheduleSpeckM

AC

Mesh

1C

SMA

LPL

No

Reduction

offrame

receptiontim

eST

EM

Mesh

2C

SMA

Wake-up

radioN

oW

ake-upradio

with

LPL

WiseM

AC

Star1

Random

accessL

PLN

oL

earningofL

PLschedules

X-M

AC

Mesh

1C

SMA

LPL

No

Multiple

shortpreamples

Z-M

AC

Mesh

1T

DM

A/C

SMA

LPL

Yes

Hybrid

TD

MA

/CSM

Aabove

B-M

AC

SynchronizedD

MA

CM

esh1

CSM

APeriodic

sleepY

esStaggered

active/sleepschedules

LE

AC

HC

lustered1

TD

MA

TD

MA

Yes

Low

energyclustering,

rotationofclusterheads

PAC

TC

lustered1

TD

MA

TD

MA

with

Yes

LE

AC

Hw

ithadaptive

dutycycle

multihop

routingS-M

AC

Mesh

1C

SMA

Periodicsleep

Yes

Active/sleep

schedulesSM

AC

SM

eshM

anyT

DM

A/FD

MA

TD

MA

Yes

Hybrid

TD

MA

/FDM

ASR

SAC

lustered1

TD

MA

TD

MA

InsideA

daptiveslotassignm

entsclusters

forreducinginterferences

T-MA

CM

esh1

CSM

APeriodic

sleepY

esA

daptiveduty

cycleT

RA

MA

Mesh

1T

DM

AT

DM

Aw

ithY

esD

istributedalgorithm

foradaptive

dutycycle

sender/receiverselection

5. TUTWSN MAC DESIGN

The results of this Thesis are summarized from now on. This chapter presents theMAC layer design for TUTWSN. The next chapter presents performance modelsfor WSN MAC protocols. Then, TUTWSN sensor node platform designs and ex-perimental deployments are presented. The details of the results are published in[P1]-[P10]. The performance of the designs is verified by performance analysis andexperimental measurements. First, TUTWSN networking technology and the designrequirements for the MAC layer are presented.

5.1 TUTWSN Networking Technology

TUTWSN networking technology has been developed for low data-rate monitoringapplications, where a large number of nodes self-configure themselves, perform sens-ing and data processing tasks, and communicate with one or more sinks with very lowenergy consumption. Target applications fields are environmental monitoring, build-ing monitoring, condition monitoring, object and human tracking, and access control.Target usage classes for message timeliness are 4 and 5 meaning non-critical flaggingand data logging tasks [83].

The applications have strict requirements for network lifetime, scalability, and adap-tivity for network dynamics. An important requirement has been that the designedprotocols are feasible in practice in harsh operation environment using low-powerhardware components. For these requirements, current networking technologies donot provide adequate energy-efficiency and performance. Thus, a completely newnetworking technology has been developed.

In this Thesis, the main focus is on an energy-efficient MAC layer, which consists ofchannel access and networking mechanisms. The main targets for the channel accessmechanism have been maximized energy-efficiency and reliability. The networkingmechanisms perform network self-configuration and neighbor discovery operations.The main objectives for them have been high energy-efficiency and adaptivity for

46 5. TUTWSN MAC Design

APIAPI

Multi-hop routingMulti-hop routing

MACMAC

Sensor node platformSensor node platform

TUTWSN stack TUTWSN MAC architecture

Frame processingFrame processing

Queue controlQueue control

Frame assemblyFrame assembly

Error controlError control

Channel accessChannel access

Networkmanagement

Networkmanagement

Neighbor discoveryNeighbor discovery

Self-configurationSelf-configuration

Bandwidth

management

Bandwidth

management

ApplicationsApplications

Fig. 11. TUTWSN protocol stack and MAC layer architecture.

tolerating unreliable radio links and node mobility. An important objective for theentire MAC layer has been compatibility with a simple and low-power hardware.This Thesis discusses also low-power and cost-effective sensor node platforms, whichhave been developed for various sensing applications. Thus, the requirement for longbattery lifetime is solved by both MAC layer and sensor node platform designs.

The TUTWSN protocol stack is presented in Fig. 11. The focus of this Thesis is pre-sented with boxes having a dark gray background. The MAC layer operates betweenthe radio on a sensor node platform, and a multi-hop routing layer [184]. Abovethe routing layer are a transport layer and API [87]. API implements an abstractionlayer between applications and network protocols. Application layer contain varioussensing [181], data processing, and diagnostics [182] applications.

The architecture of the designed MAC layer is presented on the right side of the Fig.11. Functions associated to frame processing are grouped to queue control, frame as-sembly, error control, and channel access. In operation, frames received from routinglayer are queued, and then forwarded to a frame assembly according to frame priori-ties. Frame assembly builds headers, manages addressing, and performs piggyback-ing of control signaling in MAC frames. The error control performs error detection,acknowledgements, and retransmissions. The channel access controls frame trans-missions and receptions on RF channels using contention based and contention-freemechanisms.

Network management contains mechanisms for bandwidth management, neighbordiscovery, and network self-configuration. Bandwidth management controls slot allo-cations for the contention-free channel access mechanism. Neighbor discovery con-tains reactive and proactive mechanisms called Network Channel Beaconing (NCB)and Energy-efficient Neighbor Discovery Protocol (ENDP), which have been devel-oped for maximizing network performance and energy-efficiency in dynamic radioenvironments. The self-configuration manages the formation of network topology,

5.2. TUTWSN Channel Access Mechanism 47

TUTWSN MAC

Design

TUTWSN MAC

Design

Channel

access

mechanism

Channel

access

mechanism

Networking

mechanisms

Networking

mechanisms

Neighbor

discovery

Neighbor

discoverySelf-

configuration

Self-

configurationContention-free

mechanism

Contention-free

mechanismContention

mechanism

Contention

mechanism

Network

topology

formation

Network

topology

formation

Superframe

interlacing

Superframe

interlacingPath-loss

estimation

Path-loss

estimation

Network

channel

beaconing

Network

channel

beaconing

Energy-efficient

neighbor

discovery protocol

Energy-efficient

neighbor

discovery protocol

Fig. 12. Designed MAC layer mechanisms.

the interlacing of superframes for minimizing interferences, and the estimation of ra-dio wave path-loss to neighboring nodes. In this Thesis, the self-configuration andneighbor discovery mechanisms are called networking mechanisms. The designedMAC layer mechanisms are summarized in Fig. 12.

5.2 TUTWSN Channel Access Mechanism

TUTWSN channel access mechanism pursues to minimize the duty cycle of commu-nication. Until now, related proposals for channel access mechanism have reducedenergy consumption by focusing on the minimization of long-term idle listening,overhearing, and the active period length, as discussed in the Section 4. Only a smallresearch effort has been paid on the minimization of the energy overhead of a channelaccess as a whole.

For finding out the most essential causes of energy overhead, a simple energy analysisof a CSMA channel access is presented. CSMA can be considered a typical channelaccess mechanism in WSNs and it is used for example in IEEE 802.15.4, S-MAC, T-MAC, SCP-MAC, and X-MAC. The results of the analysis will clarify the guidelinesfor the designed TUTWSN channel access mechanism. To be able to focus purely onthe data exchange between two nodes, an analyzed network contains only a sourceand a destination node. The actions of these nodes are presented in Fig. 13. At thebeginning of a channel access period, a destination node activates its receiver lasting


Carrier sensing

tBOT

Source

DestinationRX

TX

RX

TX

Radio start-up transient

C

Data frame

D

A

tAW

C

D

A Ack frame

tST

Fig. 13. Analyzed frame exchange mechanism.

tST and begins receiving the channel for possible incoming frames. Prior to a dataframe transmission, the source node waits a random backoff time (tBOT ), activates itsreceiver, and performs a carrier sensing (tCCA). For improving energy-efficiency, ablind backoff is assumed, where a source spends tBOT in sleep mode. If the channelis idle, the source turns the receiver off, activates its transmitter, and transmits a dataframe (tDATA). The energy of inactivating the radio is negligible and it can be ignored.After receiving the data frame, the destination node turns off the receiver, checks thecorrectness of the data (tAW ), activates a transmitter and transmits ACK. Since thewait time (tAW ) prior to the reception of ACK is not pre-determined, and depends onthe frame content and data processing performance, the source node needs to be inreception mode entire tAW .

The consumed energy is divided into an effective energy comprising data and ACKexchange energies, and overhead energy consisting of radio start-up, backoff, andACK wait energies. Next, models for these energies are determined.

The presented frame exchange procedure consists of two transmitter start-up andthree receiver start-up transients (tST ) during which power consumption equals to atransmitting power (PT X ) and a receiving power (PRX ), respectively. Hence, the totalstart-up energy (EST ) of a frame exchange is

EST = tST (2PT X +3PRX) . (2)

Although the source node may sleep during the backoff delay, the destination node


needs to be in the reception mode. An average idle listening time consists of a halfof a contention window length (tCW ) and a carrier sensing time (tCCA). Hence, thebackoff energy consumption (EBO) is

EBO =( tCW

2+ tCCA

)PRX . (3)

The ACK wait energy consumption (EAW ) caused by an average ACK wait delay(tAW ) is

EAW = tAW PRX . (4)

The data exchange energy (EDATA) consists purely of the transmission and receptionenergies of a data frame (LDATA). As radio data rate is R, EDATA is

EDATA =LDATA

R(PT X +PRX) . (5)

Similarly, the ACK exchange energy is

EACK =LACK

R(PT X +PRX) . (6)

Next, the energy consumptions are determined for a low power sensor node plat-form utilizing Microchip PIC18LF4620 MCU. Since the energy characteristics oflow power radios are diverse, the energy consumptions are determined for two differ-ent types of generally used commercial off-the-shelf radios: a High data Rate (HR)Nordic Semiconductor nRF2401A [120] radio having 1 Mbps data rate, and a Lowdata Rate (LR) Chipcon CC1000 [192] radio having 76.8 kbps data rate. The utilizedparameter values are presented in Table 12 in subsection 6.3., p. 76. The analysisfocuses on short (<128 Bytes) frame lengths, since they results the highest energy-efficiency at high (> 1x10−4) Bit Error Rate (BER) conditions [7,154]. High BER istypical for WSNs due to difficult operation environment and narrow band radios [7].

The resulted energies as the function of data frame size are presented in Fig. 14. Gen-erally, the HR radio has nearly one order of magnitude lower effective energy con-sumption compared to the LR radio. The energy overhead is nearly equal for bothradio types. The energy overhead is caused mostly by the backoff mechanism andcarrier sensing causing idle listening. The mechanism also necessitates frequent op-eration mode changes causing significant start-up transient energy consumption. The


LR (CC1000)

1

10

100

1000

0 16 32 48 64 80 96 112 128

Data frame size (Bytes)

En

erg

y (

µJ

)

EffectiveBack-offStart-upACK wait

HR (nRF2401A)

1

10

100

1000

0 16 32 48 64 80 96 112 128

Data frame size (Bytes)

En

erg

y (

µJ

)

EffectiveBack-offStart-upACK wait

Fig. 14. The effective and overhead energies of HR (nRF2401A) and LR (CC1000) platforms.

results clearly indicate that energy overhead is dominating the energy consumptionof the HR radio. For the LR radio, the energy overhead is also significant. In practice,busy channel situations and collisions make the energy overhead even higher [21].

The designed TUTWSN channel access mechanism pursues to maximize energy-efficiency by minimizing idle listening, unnecessary start-up transients, overhearing,control frame overhead, and collisions. These are minimized by two ways:

Predetermined frame exchange moments: Nodes maintain accurate localsynchronization and exchange frames exactly at predetermined moments.

Reservation based channel access: Nodes avoid collisions and the energyoverhead of contention mechanism by reserving their transmission moments inadvance.

The designed channel access mechanism is illustrated in Fig. 15. For enabling frameexchanges, some or all nodes maintain a superframe structure (superframe manager).Superframes are repeated at regular intervals called an access cycle. A node canreceive data and control frames from its neighbors in its own superframe, whilst frametransmissions to a neighbor is performed in the superframe of the desired neighbor.The rest of time, nodes can sleep and conserve energy. For eliminating collisions,superframes have locally unique schedules such that they do not overlap with eachother.

At the beginning of each superframe, a superframe manager transmits a beacon. Thebeacon contains crucial information for the channel access, networking, and routing.For the channel access, two fields are the most essential: time to the next superframe,


TX

RX

TX

RXNeighbor A

time

time

Beacon Contention Contention-free

slots slots

tA

Uplink boundary Downlink boundary

Super-

frameIdle time Idle timeSuper-

frame

Access cycle

TX

RXNeighbor B

time

tS

time

TX

RX

TX

RXNeighbor A

time

time

Beacon Contention Contention-free

slots slots

tA

Uplink boundary Downlink boundary

Super-

frameIdle time Idle timeSuper-

frame

Access cycle

TX

RXNeighbor B

time

tS

time

Fig. 15. TUTWSN access cycle and superframe.

which is used for maintaining synchronization, and reserved slot allocation table,which is used for granting transmission rights for associated neighbors.

The beacon is followed by time slots for data exchanges. Each superframe containstwo types of time slots: contention slots and contention-free slots. Data frames utilizereservation based contention-free slots, while contention slots are used for controlframes allowing network association and slot reservations. While the amount of theslots can vary depending on the application, the current TUTWSN implementationutilizes four contention slots and up to eight contention-free slots in each superframe.

Time slots are further divided into uplink and downlink subslots. Uplink subslots areused for transmitting data or control frames to the superframe manager. Downlinksubslots are used by the superframe manager for transmitting ACKs. Each subslotcontains an active part (tA), which is used for transmitting a single frame, and a sleeppart (tS), which is used for data processing and preparing a next frame during whichradio is in idle or sleep mode. Since the beacon at the beginning of each superframeperforms synchronization, the clock drift is negligible at the slot boundaries. In thecurrent implementation, tA is slightly longer than the transmission time of a singleframe, and length of each subslot is 10 ms.

To minimize the energy overhead, uplink frames are transmitted without any collisionavoidance mechanism. Occasional collisions in contention slots are managed by re-transmissions. All downlink subslots are contention-free, since only the superframemanager can transmit in these slots. Collisions between superframes are avoidedby the superframe interlacing mechanism presented in the next section. As carrier


sensing functionality is not needed, energy overhead is low and the designed channelaccess mechanism can be implemented with a simple and low cost hardware.

Since the superframe manager cannot predict which contention slots will be used,unnecessary reception of slots is unavoidable causing idle listening. This is commonfor all contention based mechanisms. The reduction of the number of contentionslots reduces the idle listening of superframe managers, but increases the probabilityof collisions. This may reduce the energy-efficiency and performance of an entirenetwork [153, 222].

In the designed contention mechanism, the energy consumption is minimized by threeways. First, the reception is always terminated as soon as an unused contention slot isdetected, or at last when tA has expired. Second, the utilization of contention slots isminimized by piggybacking bandwidth adjustment signaling in data frames. Third,the number of contention slots is dynamically adjusted according to their loading[P1].

The designed contention-free mechanism is inherently energy-efficient, since the uti-lized reserved slots in each superframe are determined in advance using bandwidthadjustment signaling and the slot allocation table. The idle listening is nearly elimi-nated, since only utilized subslots are received. A minor idle listening is caused bythe inaccuracy of time synchronization and occasional link failures causing receptionfailures in the contention-free slots.

5.3 TUTWSN Networking Mechanisms

In this Thesis, networking mechanisms are included in the MAC layer. Networkingmechanisms perform the creation and management of a scalable, reliable, and adap-tive network topology for data routing. The networking mechanisms are divided intoself-configuration and neighbor discovery mechanisms. The mechanisms are pre-sented independently of each other to improve clarity and to facilitate the adaptationof individual mechanisms with other types of MAC layers.

5.3.1 Self-Configuration

The designed self-configuration mechanism performs three main functions: networktopology formation, superframe interlacing, and path-loss estimation. The networktopology formation determines the operation mode of each node and maintained ra-dio links between nodes. The superframe interlacing ensures collision-free operation

5.3. TUTWSN Networking Mechanisms 53

for the channel access mechanism in large and dense networks by eliminating over-lapping superframes. The path-loss estimation measures the attenuation of signalstrength between nodes allowing the utilization of simple, low power and low costradios without Received Signal Strength Indicator (RSSI) functionality. The path-loss estimation enables a dynamic transmission power control and the maintenanceof robust routing paths. Self-configuration operations are decentralized, and eachnode makes its configuration according to observed information from neighborhood.

Network Topology Formation

To reduce the energy consumption of frame transmissions in large networks, multi-hop data routing between nodes is utilized [62, 96]. Frames are routed from a sourceto a destination along a chain of low-energy hops. Each node along the chain receivesdata from a neighbor (child) one hop closer to the source, maintains synchronizationwith a next-hop node (parent) by periodically receiving its beacons, and transmitsdata according to time slot assignments.

The selection of network topology between flat and clustered affects significantly onnetwork energy consumption and bandwidth utilization [P1], [99, 203]. In the flattopology, all nodes participate in data routing and consume nearly equally power andnetwork bandwidth. In the clustered topology, a network is formed as interconnectedstar networks. The master of each star is a cluster head, while other nodes are leafnodes. Cluster heads utilize a majority of energy and bandwidth by managing super-frames and exchanging data with other clusters. Leaf nodes synchronize themselveswith a superframe schedule and transmit data on demand without the need of theirown superframes, which reduces the bandwidth utilization of a network.

The designed network topology is based on the clustered topology. Each cluster con-sists of a cluster head (headnode), leaf nodes (subnodes), and associated headnodes(child headnodes) from neighboring clusters. The operation of a child headnode in anext hop cluster is similar with a subnode, which receives beacons and transmits dataaccording to time slot assignments.

The utilization of a clustered topology is rationalized by a simple analysis, which con-siders the energy consumptions of clustered and flat topologies using the TUTWSNchannel access mechanism. The analysis assumes that the energy consumptions offrame transmissions (ET X ) and receptions (ERX ) are equal for all frame types, whichis realistic for the TUTWSN channel access utilizing a low power radio. Moreover,the density of cluster heads in the clustered topology is assumed to be adequate for


maintaining optimal hop lengths. Therefore, the number of required data and ACKframe exchanges for flat and clustered topologies is equal, as the entire network isconsidered.

A router node is defined as any node in the flat topology, and a cluster head in theclustered topology. The energy overhead of the router node (EOR) consists of thetransmissions and receptions of beacons, and the reception of SA contention slots.For one access cycle, the energy overhead is

EOR = ET X +(SA +1)ERX . (7)

The energy overhead of a leaf node (EOL) using TUTWSN channel access is causedby the reception of beacons. Hence, for one access cycle, EOL equals to ERX .

In the flat topology, all nodes have equal energy overhead, which equals to EOR. In theclustered topology, where each cluster consists of nS leaf nodes, the average energyoverhead per a node (EOC) is

EOC =ET X +(SA +1+nS)ERX

nS +1. (8)

If nS = 0, then EOC = EOR. This is clear, since all nodes are cluster heads and thenetwork is similar with the flat topology. As nS increases, EOC decreases, and whennS → ∞, then EOC → ERX , which equals to EOL. Hence, the clustered topology hasalways lower energy overhead than the flat topology, assuming that the network has atleast one leaf node. The energy-efficiency of clustering is even more obvious, whencluster heads aggregate received data reducing the amount of forwarded data [203].

Network connectivity between clusters can be formed as a cluster-mesh or a cluster-tree topology. In the cluster-mesh topology, each cluster head maintains connectivitywith all neighboring cluster heads resulting robust network, but higher energy con-sumption. In the cluster-tree topology, each cluster head maintains connectivity withone cluster head only, which is one hop closer to a sink locating at the root of thetree. This improves energy-efficiency, but reduces the tolerance against link failuresdue to low connectivity.

This Thesis presents a multi-cluster-tree topology [P1], which combines the strengthsof cluster-tree and cluster-mesh topologies. The multi-cluster-tree topology consistsof multiple super-positioned cluster-tree networks. An example multi-cluster-treetopology is illustrated in Fig. 16, where arrows indicate the directions of uplink


Subnode

Headnode

Sink

Cluster-tree 1

Cluster-tree 2

Fig. 16. Multi-cluster-tree network topology (k = 2).

routing paths. Each subnode and headnode maintains synchronization with k neigh-bors by receiving their beacons. This allows the adjustment of connectivity for bothsubnodes and headnodes allowing a trade-off between network robustness and en-ergy consumption. Compared to the cluster-tree topology, which supports only oneroute to a single sink, the multi-cluster-tree allows the utilization of multiple sinks,multiple routes, and load balancing between headnodes. The value of k is uniform forentire network and it is selected before a deployment according to expected networkdynamics. According to measurements with TUTWSN nodes, an optimal value for kis between 2 and 4 [P1].

Superframe interlacing

For guaranteeing contention-free channel access in a multi-hop network, the overlap-ping of superframes in two-hop neighborhood (interference range) is eliminated byinterlacing. Typically, interlacing is implemented by time division, for example inIEEE 802.15.4 [82], DMAC [61], SRSA [218], and TRAMA. The time division lim-its network density especially when the superframe length is relatively long comparedto the access cycle length. In the designed superframe interlacing mechanism, scala-bility is improved by time and frequency division. For reasoning this, a short analysisof the maximum scalability is presented. The maximum number (α) of nodes in aninterference area can be determined by the access cycle length (TAC), the superframelength, the average number of subnodes in each cluster, and the number of utilizednon-interfering frequency channels (nCH) as

α =TACnCH(1+nS)

2(1+SA +SR)(tA + tS)+ tguard, (9)


where SA and SR are the maximum number of contention and contention-free slots,tguard is a short guard time between consecutive superframes. α is maximized bymaximizing TAC, nCH , and nS, and by minimizing tA and tS. It can be clearly seen inthe equation that by utilizing a high data-rate radio operating at a wide frequency bandprovides the highest scalability. In the current 2.4 GHz TUTWSN implementation,TAC = 4 s, nCH = 20, nS = 8, SA = 4, SR = 8, tA + tS = 10 ms, and tguard = 100 ms.Thus, α equals to 2000 nodes per an interference area. If only one channel is used(nCH = 1), α would be reduced to 10 nodes per an interference area. In TUTWSNsensor node platforms, the interference range in indoor conditions is around 100 mequaling to the area of 31400 m2.

In the designed superframe interlacing mechanism, each headnode selects semi-randomlya time slot and a frequency channel (superslot) for its superframe among the freeslots detected by a network scan. The simple randomization minimizes the energyoverhead of signaling traffic. The superslot is selected at a node start-up and if inter-ferences are detected by increased link error rate [P1].

Path-Loss Estimation Scheme

The measurement of path-loss to neighboring nodes is crucial for efficient networkself-configuration; nodes should form and maintain robust routing paths with suffi-cient radio link quality and energy-efficient hop lengths [171]. Moreover, measuredpath-loss enables dynamic adjustment of transmission power levels for minimizingthe energy consumption of the transmitter. Conventionally, the path-loss is mea-sured by a RSSI functionality of a radio. A disadvantage of the RSSI functionalityis increased radio hardware complexity. In general, the cheapest and lowest energyradios lack RSSI functionality, for example Nordic Semiconductor nRF2401A [120],and nRF24L01 [121].

The designed path-loss estimation scheme [P2] measures path-loss to a neighboringnode according to the successfully and unsuccessfully received beacons, which theneighbor transmits at different transmission power levels. The operation principle ofpath loss metering is illustrated in Fig. 17, where Node 3 determines path losses toNode 1 and Node 2 by receiving their beacons. The transmission ranges of beaconstransmitted at four power levels are illustrated by circles. As Node 1 is transmitting,Node 3 is within the range of the second lowest transmission power level resulting inlow path loss. Only the highest transmission power reaches node 2 resulting in highpath loss.


Range with the highest powerbeacon

Range with the lowest powerbeacon

Node 1 Node 2

Node 3

Node 1 (TX)

Node 2 (TX)

Node 3 (RX)

Lowest power beacon

Highest power beacon

Failed receptions

Successful reception

Successfulreception

Radio powerdown

Fig. 17. The operational principle of a path loss metering.

In TUTWSN MAC implementation, all beacons are replaced by the train of beacons.Power consumption is further optimized by transmitting only two beacons in eachtrain. Each train consists of the highest power beacon and an altering power beacon.The transmission power of the altering power beacon is alternately selected amongall possible transmission power levels except the highest power, and it is signaled inthe beacon header. Thus, the path loss estimation is gradually improved as multipletrains are received [P2].

Experimental performance measurements [P2] in indoor and outdoor environmentsindicate sufficient accuracy for managing robust network operation. Measurementspresented in [90] show that the path loss metering scheme has adequate accuracy fornode localization [169]. A performance analysis [P2] in a large scale ZigBee networkindicate that the path loss metering consume less than 0.5% of the entire node powerconsumption. By utilizing a simple (Nordic semiconductor nRF24L01) radio withthe path loss metering scheme results in over 51% power saving compared to a lowpower RSSI equipped radio (TI CC2400) [P2].


5.3.2 Neighbor Discovery

Low transmission power levels, harsh and dynamically changing operation environ-ment, and mobile applications cause unreliability for wireless links. After a linkfailure, a new synchronized link is established by a neighbor discovery operation.Typically, the neighbor discovery is performed by a network scan by listening a fre-quency channel until adequate number of beacons is received. To be able to detectall neighbors operating at different frequency channels, the network scan duration ineach utilized channel must be at least one beacon transmission interval. This methodis utilized, for example, in IEEE 802.15.4 [82].

An energy analysis [P5] with a typical low power radio indicates that each second ofa network scan consumes the energy equaling to the transmission of nearly 2800 bea-con frames. Clearly, the scanning increases significantly the energy consumption ofa node, especially as network dynamics is high and the number of utilized frequencychannels is high.

In this Thesis, two solutions for energy-efficient neighbor discovery are presented:a Network Channel Beaconing (NCB) scheme [P1], [P3], [P5], which can be com-plemented with an Energy-efficient Neighbor Discovery Protocol (ENDP) [P4] forfurther improving adaptivity. The objective of the both solutions is to maximizeachieved energy efficiency and to minimize data routing breaks in dynamic networks.

Network channel beaconing

The designed NCB scheme dedicates one globally known channel for short networkbeacons, which are broadcast by all headnodes more frequently than ordinary bea-cons (cluster beacons). A network beacons contain the frequency channel and exacttime to the next superframe. In addition, required information for selecting an ap-propriate headnode as a parent is included. The network beacons are received duringa neighbor discovery only. As illustrated in Fig. 18, the neighbor discovery is per-formed by receiving the network channel for one network beacon interval [P3]. Forclarity, the additional beacons of the path-loss estimation scheme are not presentedin the figure. In normal operation, beacons are received only in the cluster channel.

A potential shortcoming of the network channel beaconing mechanism is unequaltransmission range in different frequency channels due to antenna characteristics anddifferent interference levels. To avoid this, the neighbor discovery is completed withthe reception of cluster beacons from the most suitable parents in their cluster chan-


time

time

Networkchannel

Headnode D (channel 4)

Superframestime

Headnode C (channel 3)

time

Headnode B (channel 2)

time

Headnode A (channel 1)

Network channel reception

for neighbor discovery

Network beacons of:

Headnode A: Headnode B: Headnode C: Headnode D:

Network beacon interval

Access cycle

Fig. 18. Neighbor discovery with network beacons.

nels. This validates the link quality measurements and the set of new parents withwhich synchronized links can be established.

The performance of the network channel beaconing scheme has been analyzed in ageneral case with a network using the synchronized low duty-cycle channel accessscheme [P5], and in a ZigBee network [P3].

To show the achieved energy saving, Fig. 19 presents the results of an energy analy-sis [P3] comparing network maintenance power consumptions with and without theNCB scheme. The network maintenance power is the sum of network scanning andbeacon transmission powers. Cluster beacon interval and average network scan in-terval are set to 3.96 s and 2000 s, respectively. The analysis is performed using thehardware parameters of TI CC2420 [194] radio, and PIC18LF8722 [106] MCU. Asthe standard specified passive scans are used, the network maintenance power con-sumptions for networks using one and four channels are 132 µW and 468 µW. WhenNCB is utilized, the minimum of network maintenance power is 42 µW regardless ofthe number of channels. This power is achieved at 2.6 Hz network beacon rate. Thus,NCB achieves up to 96% energy saving compared to the standard ZigBee [P3].

An analysis in [P3] indicates that an energy optimal network beacon transmission ratef ∗b is a function network scanning interval (INS), beacon frame transmission energy(ET X ), radio power consumption in the reception mode (PRX ), and the average numberof leaf nodes in each cluster (nS), as


0

50

100

150

200

250

300

350

400

450

500

0.1 1 10 100

Network Beacon Rate (Hz)

Ma

inte

na

nc

e P

ow

er

(µW

)

Network maintenance (ZigBee, 4 ch)

Network maintenance (ZigBee, 1 ch)

Network maintenance (NCB)

Beacon exchange (NCB)

Network scanning (NCB)

Fig. 19. Beacon exchange, network scanning and network maintenance power consumptionsand comparison with standard ZigBee neighbor discovery.

f ∗b =

√PRX(nS +1)

ET X INS. (10)

The function of f ∗b is similar for ZigBee, TUTWSN and other networks utilizing thesynchronized low duty-cycle MAC scheme with regular beacon transmissions andnetwork scans.

Energy-efficient neighbor discovery protocol

The designed ENDP reduces the need of network scans by a proactive distributionof synchronization information about potential parents in a two-hop neighborhood.Nodes transmit the synchronization information about their direct parents in beacons.A Synchronization Data Unit (SDU) referring to a node consists of two parts: thechannel on which the node is operating, and the time base difference (offset) betweenthe node and the sender of SDU. As NCB is used, SDUs are transmitted in networkbeacons [P4].

The operation of ENDP is illustrated in Fig. 20. In the figure, the number of main-tained synchronized links (k) is two. Node M receives SDUs from its parents I and L.Node I composes SDUs from its parents A and B. Similarly, Node L composes SDUsfrom nodes H and G yielding that node M gathers synchronization information aboutnodes A, B, H, and G [P4].


M

L

F

E

K

GH

J

DC

I

A

B

Ch: 12

Ch: 35

Ch: 56

Ch: 55

Ch: 16

Offset:+220 ms

Offset:+150 ms

Offset:+100 ms

Offset:+390 ms Offset:+450 ms

Offset:+120 ms

Ch: 21

Node I, ch55, neighbors

[ch35,+150ms; ch12,+100ms]

Node L, ch21, neighbors[ch56,+450ms; ch16,+390ms]

Ch:07

Node M, ch07, neighbors

[ch55,+220ms; ch21,+120ms]

Fig. 20. Connectivity and distributed neighbor information. Arrows indicate synchronizationbetween nodes.

If a link failure occurs or the link quality to a neighbor degrades significantly, anode first tries to receive beacons and form a new synchronized link according tothe received SDUs. The energy consumption of these beacon receptions is negligiblecompared to the energy consumption of a network scan. A network scan is performedonly if a suitable parent is not discovered by SDUs. The performance of ENDP canbe adjusted according to the parameter k of the multi-cluster-tree topology.

The energy-efficiency of ENDP is analyzed in [P4]. The analysis utilizes the hard-ware parameters of a TUTWSN temperature sensing prototype consisting of Mi-crochip PIC18LF8722 [106] MCU and Nordic Semiconductor nRF2401A [120] ra-dio having 1 Mbps data rate. In the analysis, radio range is set to 10 m. Fig. 21compares the network maintenance power consumptions as the function of networkbeacon rate at 0.1 m/s and 10 m/s node velocities. Curves are plotted using NCB withand without ENDP. When ENDP is used, k is varied from 1 to 4. When ENDP is notused, k is fixed to one resulting in the lowest power consumption. For comparison,the network maintenance power consumption using conventional network scans inone channel without NCB is plotted [P4].

The network maintenance power consumptions at 0.1 m/s and 10 m/s node veloc-ities using conventional network scans are 1.23 mW and 60.17 mW, respectively.By adding the NCB scheme, network maintenance power consumption decreases atoptimal network beacon rate to 95 µW and 742 µW. With ENDP, the minimum net-


0.01

0.1

1

10

100

0.01 0.1 1 10 100


Ma

inte

na

nc

e P

ow

er

(mW

)

Conventional scan Without ENDP ENDP, k=1

ENDP, k=2 ENDP, k=3 ENDP, k=4

v = 10 m/s

0.01

0.1

1

10

100

0.01 0.1 1 10 100


Ma

inte

na

nc

e P

ow

er

(mW

)

v = 0.1 m/s

Fig. 21. Network maintenance powers with and without ENDP and network beacons.

work maintenance power further reduces to 87 µW (k = 1) and 397 µW (k = 2). Theresults show that ENDP reduces the network maintenance power consumption, asnode mobility is higher than 1 m/s and the achieved energy saving increases withnode mobility [P4]. Experimental measurements of ENDP are presented in Chapter7 validating the results in practice.

6. PERFORMANCE MODELS

This chapter presents performance models for analyzing the power consumptionsof the most essential low power channel access mechanisms and comparing themagainst the designed TUTWSN MAC. The focus is on data and ACK frame ex-changes and on the maintenance of link synchronization by a beacon or SYNC frameexchange.

Performance models are determined for the following MAC protocols: T-MAC [201]and B-MAC [129], which are well known synchronized and unsynchronized lowduty-cycle protocols, X-MAC [24] and SCP-MAC [221], which are two interestingproposals for unsynchronized protocols, and IEEE 802.15.4 [82], which is the mostpotential standardized technology for WSNs. For comparison, an ideal MAC proto-col is defined and modeled.

The following performance models are based on the analysis of S. Yoon presentedin [223]. For this Thesis, the set of models has been extended by IEEE 802.15.4 andTUTWSN MAC protocols. In addition, the effects of start-up transitions, contentionwindows and crystal tolerance have been modeled more accurately. In addition, themodels and their presentation has been simplified and clarified.

The performance models are derived using the following assumptions:

- Each sensor node measures one sensor sample and forwards it to a next-hopnode during one data generation interval;

- Each data frame is followed by an ACK for fair comparison;

- There are no transmission errors nor collisions;

- There is no contention, and carrier sense attempts produce an idle result;

- The power consumption of idle listening equals to the reception mode power;

- The active time of MCU equals to the active time of radio.

64 6. Performance Models

Therefore, the performance models can focus on the power consumption of the chan-nel access mechanisms, while the effects of data processing, contention, and controlframe exchanges are eliminated. For contention based protocols, obtained resultsare slightly better than in practice with contention. As TUTWSN MAC utilizescontention-free mechanism for data and ACK exchanges, the obtained results forTUTWSN are realistic.

6.1 Utilized Parameters

For determining the channel access models, all essential parameters describing thecharacteristics of a sensor node platform, application, and network topology are iden-tified. The sensor node platform is defined by the following parameters:

ε Crystal tolerance of a wake up timerPY The power consumed for Y , where Y ∈ RX (receive), S (sleep), T X (transmit)R The data rate of a radiotST Radio start-up transient duration (crystal running)

Application and network topology are defined by the following parameters:

LX The length of frame X , where X ∈ ACK, B (beacon or SYNC), CT S, DATA,P (preample), RT S

n The number of direct neighbors for a given nodenDL The number of descendent nodes of a given node in the routing tree, i.e. the

number of data frames the node needs to forward during one data generationinterval

TDATA Data generation interval in each node

In addition, there are protocol implementation specific parameters. Generally utilizedparameters of that kind are:

tCCA The time for a clear channel assessment or carrier sensingtCW Contention window length in CSMAtsleep Sleep period lengthTAC Access cycle length or channel polling intervalTSY NC SYNC or beacon frame transmission interval

6.2. Models 65

6.2 Models

Energy consumptions are analyzed for a router node (A), and a leaf node (B) pre-sented in Fig. 22. Both nodes have eight neighbors (n), which may cause overhearingand interferences for the channel access. Data generation interval (TDATA) is equal foreach node, and it varies from 1 s to 1000 s. Arrows in the figure indicate data routingdirections. The traffic load is accumulated in routers, since they transmit their owndata and the multi-hop routed data from nDL nodes. For example, the router nodeA routes data from three nodes (nDL = B, D, E), while the router node C routesdata from four nodes (nDL = A, B, D, E). This increases the power consumptionof these routers, but also the overhearing and interferences among other nodes in theirtransmission range.

Average power consumptions (P) for each protocol are calculated by normalizedtransmission (tT X ) and reception (tRX ) activities and their power consumptions as

P = tT X PT X + tRX PRX +(1− tT X − tRX)PS. (11)

The normalized activity is determined by dividing the duration of an activity by theinterval of the activity resulting in a percentage value of the activity. Data exchangesare normalized by TDATA during which all nodes in the network generate exactlyone data frame. Similarly, the transmission and reception activity for maintainingsynchronization is normalized by TSY NC.

A BC

D

E

Fig. 22. Network topology for channel access comparison.


6.2.1 Ideal-MAC

First, an ideal MAC (Ideal-MAC) protocol [223] is defined. All nodes can exchangedata and ACK frames without the need of any synchronization or contention mecha-nism. Nodes can sleep all the time between frame exchanges. Hence, the Ideal-MACdoes not cause any idle listening or control frame overhead.

The required activity for exchanging one data frame is presented in Fig. 23. AlthoughMAC is ideal and practically impossible to implement, a sensor node platform isrealistic. Thus, each data transmission and reception is preceded by a radio start-uptransient (tST ). Thus, tT X and tRX for the leaf node are

tT X =(

tST +LDATA

R

)1

TDATA, (12)

tRX =(

tST +LACK

R

)1

TDATA. (13)

The router node receives data frames from nDL leaf nodes, transmit them ACKs,forwards the received and own data frames to a parent and receives ACKs. Thus, tT X

and tRX for the router are

tT X =(

tST +LDATA

R

)nDL +1TDATA

+(

tST +LACK

R

)nDL

TDATA. (14)

tRX =(

tST +LDATA

R

)nDL

TDATA+

(tST +

LACK

R

)nDL +1TDATA

. (15)

6.2.2 B-MAC

B-MAC [129] uses the LPL scheme, where nodes sleep (tsleep), wake up (tST ) andpoll channel (tCCA) periodically at TAC intervals. The frame exchanges of B-MAC are

DATA

ACK

s

time

time

TX

RX

TX

RXSource

Destination

s

s

s

tST

Fig. 23. The activity of radio in Ideal-MAC.

6.2. Models 67

presented in Fig. 24.

The normalized channel polling time (tPOLL) is

tPOLL =tST + tCCA

TAC. (16)

The leaf node transmits data frames at TDATA intervals. Each transmission is precededby a carrier sensing (tCCA) and a preamble transmission lasting TAC. Thus, tT X for theleaf node is

tT X =(

tST +TAC +LDATA

R

)1

TDATA. (17)

In B-MAC, all data in a radio range is received. As the leaf node has n neighbors,and the router in a range forwards nDL data frames from leaf nodes, totally n + nDL

data frames are received during TDATA. Since channel is polled randomly, averagepreamble reception time is a half of TAC. Thus, tRX for the leaf node is

tRX = tPOLL +(

TAC

2− tCCA +

LDATA

R

)n+nDL

TDATA+

(tST +

LACK

R

)1

TDATA. (18)

The operation of the router node is similar to the leaf node, except the amount ofexchanged data. The router node forwards data frames from nDL nodes and transmitsACKs. The normalized transmission and reception times for the B-MAC router are

tT X =(

tST +TAC +LDATA

R

)nDL +1TDATA

+(

tST +LACK

R

)nDL

TDATA. (19)

DATA

time

time

TX

RX

TX

RXSource

DestinationACK

tCCAtCCA tsleepPreample

TAC

TAC

TAC /2

s

s

ss

ss

s

tST

Fig. 24. The activity of radio in B-MAC.


tRX = tPOLL +(

TAC

2− tCCA +

LDATA

R

)n+nDL +1

TDATA+

(tST +

LACK

R

)nDL +1TDATA

. (20)

The power consumption reaches its unique minimum at an optimal polling interval(T ∗AC) obtained by setting ∂(tT X PT X + tRX PRX)/∂TAC = 0. Performance results arecalculated using the optimal polling interval of the router, which is

T ∗AC =

√TDATA(tST + tCCA)

(nDL +1)PT X/PRX +(n+nDL +1)/2. (21)

6.2.3 SCP-MAC

SCP-MAC [221] replaces the long preample with a short wake-up tone by wakingup the senders and the receiver at the same time. The best-case situation is consid-ered, where all synchronization signaling is piggybacked with data frames. Thus, thesynchronization does not cause control frame exchanges. The frame exchanges ofSCP-MAC are presented in Fig. 25.

SCP-MAC utilizes similar channel polling than B-MAC, and tPOLL equals to (16).The duration of the wake-up tone (tTONE) is determined according to the clock drift(ε), the rate of frame receptions containing synchronization information, and the min-imum tone duration (tCCA) to detect a transmission. Thus, tTONE is

tTONE =4TDATAεn+nDL

+ tCCA. (22)

DATA

time

time

TX

RX

TX

RXSource

Destination

tCW /4

tCCA tsleep

TAC

Tone

ACK

SB

tCCA tCCA

s

s

s

s

s s

s

sstST

tTONE

tCW /4

CW1 CW2

Fig. 25. The activity of radio in SCP-MAC.

6.2. Models 69

A data frame transmission consists of the wake-up tone, and the frame transmis-sion. SCP-MAC utilizes two contention windows (CW1 and CW2) being tCW /2 long.Thus, an average backoff time in each contention window is tCW /4 during which thesource node is asleep. After a start-up transient, the destination node receives onaverage halves of the wake-up tone and the second contention window. Data framesare piggybacked with synchronization data (SB). Thus, tT X and tRX of the SCP-MACleaf node are

tT X =(

2tST + tTONE +LSB +LDATA

R

)1

TDATA. (23)

tRX = tPOLL +(

3tST +2tCCA +LACK

R

)1

TDATA

+(

3tST +tTONE

2+

tCW

4+ tCCA +

LSB +LDATA

R

)n+nDL

TDATA. (24)

The SCP-MAC router transmits nDL + 1 data frames to a next hop node, and ACKsto the leaf nodes. Thus, tT X and tRX of the SCP-MAC router node are

tT X =(

2tST + tTONE +LSB +LDATA

R

)nDL +1TDATA

+(

tST +LACK

R

)nDL

TDATA. (25)

tRX = tPOLL +(

3tST +2tCCA +LACK

R

)nDL +1TDATA

+(

3tST +tTONE

2+

tCW

4+ tCCA +

LSB +LDATA

R

)n+nDL +1

TDATA. (26)

For achieving the best energy-efficiency, a node should only poll the channel whenthere is a transmission from a neighbor. Performance results are calculated using theoptimal polling interval of the router, which is

T ∗AC =TDATA

nDL +1. (27)


6.2.4 X-MAC

In X-MAC [24], each data frame transmission is preceded by the strobed preamble,as presented in Fig. 26. The minimum channel polling time to receive at least oneentire preamble (P) equals to the lengths of two preamples (tp) and one ACK (tal).The normalized channel-polling time in X-MAC is

tPOLL =2tp + tal

TAC,where (28)

tp = tST +LP

R, (29)

tal = tST +LACK

R. (30)

Assuming uniform distribution of wake-up moments, the average length of the strobedpreamble is a half of the wake up period (TAC), during which TAC/(2(tp + tal)) pream-ples are transmitted. Overhearing is limited to the channel polling time. Hence, tT X

and tRX the leaf node are

tT X =(

TAC

2(tp + tal)tp + tST +

LDATA

R

)1

TDATA, (31)

tRX = tPOLL +(

TAC

2(tp + tal)+1

)tal

TDATA. (32)

DATA

time

time

TX

RX

TX

RXSource

Destination

tsleep

TAC / 2

P

tp

2tp + talP

s

s

s

s

Ps

s

s

s

s

s

s

s

s

AC

K

AC

K

tal

s

TAC

tST

Fig. 26. The activity of radio in X-MAC.

6.2. Models 71

As the router node receives and forwards data frames from nDL nodes, the normalizedtransmission and reception times of the X-MAC router node are

tT X =(

TAC

2(tp + tal)tp +

LDATA

R

)nDL +1TDATA

+2talnDL

TDATA, (33)

tRX = tPOLL +(

TAC

2(tp + tal)+1

)tal (nDL +1)

TDATA+

(tST +

LDATA

R

)nDL

TDATA. (34)

The power consumption reaches its unique minimum at an optimal polling interval(T ∗AC) obtained by setting ∂P/∂TAC = 0. Performance results are calculated using theoptimal polling interval of the router, which is

T ∗AC =

√2TDATA(tp + tal)(2tp + tal)

(tpPT X/PRX + tal)(nDL +1). (35)

6.2.5 T-MAC

Next, the synchronized low duty-cycle protocols are modeled. For comparability, TAC

is determined such that each active period in the router node contains nF data frames

TAC =nFTDATA

nDL +1. (36)

In T-MAC [201], each node polls channel for RTS messages at TAC intervals, aspresented in Fig. 27. If no traffic exists, radio is turned off after a period of TA.Hence, the normalized channel polling time without traffic is

tPOLL =tST +TA

TAC,where (37)

TA = tST + tCW + tRT S. (38)

A data frame transmission consists of a random delay within a Contention Window(CW) being followed by an RTS - CTS - DATA - ACK frame exchange. LB bytes longSYNC frames are transmitted at TSY NC intervals using a random delay within CW.For maximum energy-efficiency, adaptive listening and virtual clusters are assumed.


According to received RTS frames, nodes are in sleep mode during the transmissionsintended to other nodes. Thus, tT X and tRX for the leaf node are

tT X =(

2tST +LRT S +LDATA

R

)1

TDATA+

(tST +

LB

R

)1

TSY NC, (39)

tRX = tPOLL +(

2tST +tCW

2+

LRT S

R

)n+nDL

TDATA

+(

2tST +LCT S +LACK

R

)1

TDATA+

(tST + tCW − LB

R

)1

TSY NC. (40)

The router node receives data frames from nDL nodes, transmits nDL + 1 frames to anext-hop node, and transmits and receives SYNC frames. Thus, tT X and tRX for theT-MAC router node are

tT X =(

2tST +LRT S +LDATA

R

)nDL +1TDATA

+(

2tST +LCT S +LACK

R

)nDL

TDATA+

(tST +

LB

R

)1

TSY NC, (41)

DATA

AC

K

time

time

TX

RX

TX

RXSource

Destination

B

CT

S

RT

S

tCW

/2

tCW

ss

ss

s

s

s

s

ss

ss

s

B

times

TA

s s

SY

NC

SY

NC

TSYNC

TA

TAC

s

s

1. data exchange 2. – n. data

exchanges

…

s

ACTIVE

PERIOD

TAC

s

tST

Fig. 27. The activity of radio in T-MAC.

6.2. Models 73

DATA

AC

K

time

time

TX

RX

TX

RXSource

DestinationB

tCW

/2

s

s

ss

ss

ss

tCCA

DATAA

CK

tCW

/2

ss

ss

ss

tCCA

…

2TAC

time

tCAP

TAC

B

1. data exchange 2. data exchange

CAP B CAP

s

3. – n. data

exchanges

tST

Fig. 28. The activity of radio in IEEE 802.15.4.

tRX = tPOLL +(

3tST +tCW

2+

LRT S +LDATA

R

)nDL

TDATA

+(

tST + tCW − LB

R

)1

TSY NC+

(tST +

tCW

2+

LRT S

R

)n+nDL +1

TDATA

+(

2tST +LCT S +LACK

R

)nDL +1TDATA

. (42)

6.2.6 IEEE 802.15.4

For obtaining the best energy-efficiency, IEEE 802.15.4 [82] is analyzed in the beacon-enabled mode, with inactive time, and employing a cluster-tree network topology.Nodes maintain synchronization by receiving beacon frames from a parent at thebeginning of active periods. Beacons are transmitted by cluster heads only. Therequired activity of radio is presented in Fig. 28.

Beacons are transmitted at TAC intervals. As a crystal tolerance is ε, the normalizedbeacon reception (polling) time is

tPOLL =(

tST +2TACε+LB

R

)1

TAC. (43)

A data frame transmission is preceded by a random backoff delay and two ClearChannel Assessment (CCA) operations [82]. IEEE 802.15.4 utilizes blind backoffs,where nodes spend the backoff delay in the sleep mode. A data frame is followed by


an ACK, which is transmitted within a maximum waiting time of 864 µs. Assumingthe best case situation, where the hardware processes the received frame instantly,ACK is transmitted without a delay. Thus, tT X and tRX for the IEEE 802.15.4 leafnode are

tT X =(

tST +LDATA

R

)1

TDATA, (44)

tRX = tPOLL +(

3tST +2tCCA +LACK

R

)1

TDATA. (45)

The IEEE 802.15.4 router node (coordinator) transmit beacons, receives entire CAPexcept ACK transmissions, and forward data to a next hop node. Thus, tT X and tRX

for the IEEE 802.15.4 router node are

tT X =(

tST +LB

R

)1

TAC+

(tST +

LDATA

R

)nDL +1TDATA

+(

tST +LACK

R

)nDL

TDATA, (46)

tRX = tPOLL +tCAP

TAC−

(tST +

LACK

R

)nDL

TDATA+

(3tST +2tCCA +

LACK

R

)nDL +1TDATA

. (47)

For achieving the maximum energy-efficiency, analysis results are determined usingthe minimum CAP length, where the required frame exchanges can be performed.The minimum CAP length is

tCAP =(

4tST +tCW

2+2tCCA +

LDATA +LACK

R

)nF . (48)

6.2.7 TUTWSN MAC

For comparison, the energy consumption of TUTWSN is modeled. Due to the sta-tionary network, the following analysis considers the basic channel access mech-anism without the networking mechanisms. Each node maintains synchronizationwith one parent. Superframe contains SA contention slots and the required number ofcontention-free slots. The activity of radio in TUTWSN is presented in Fig. 29.

6.3. Results 75

In TUTWSN, the normalized channel polling time tPOLL is similar with IEEE 802.15.4.The leaf node (subnode) receives the beacons and transmits data using a contention-free slot. Thus, tT X is similar with IEEE 802.15.4 leaf node, and tRX for the leaf nodeis

tRX = tPOLL +(

tST +LACK

R

)1

TDATA. (49)

The TUTWSN router node (headnode) receives and transmits beacons, receives SA/TAC

contention slots and nDL/TDATA contention-free slots, and transmits received andown generated data to a parent in contention-free slots. Thus, tT X and tRX for theTUTWSN router are

tT X =(

tST +LB

R

)1

TAC+

(tST +

LACK

R

)nDL

TDATA+

(tST +

LDATA

R

)nDL +1TDATA

, (50)

tRX = tPOLL +(

tST +LDATA

R

)(SA

TAC+

nDL

TDATA

)+

(tST +

LACK

R

)nDL +1TDATA

. (51)

6.3 Results

By utilizing parameter values presented in Table 12, average power consumptions aredetermined for HR and LR platforms.

DATA

AC

K

time

time

TX

RX

TX

RXSource

DestinationB

s

s

s

ss

s

2TAC

time

TAC

Superframe

s s

DATA

AC

K

s

ss

s

…

SA

contention

slots

1.

contention-

free slot

2.

contention-

free slot

SuperframeSleep

…

3. – n.

contention-

free slots

tST

Fig. 29. The activity of radio in TUTWSN MAC.


The results for HR platform are presented in Fig. 30. According to the results,synchronized protocols outperform clearly unsynchronized proposals in the analyzednetwork. For both node types, the order of power consumptions is the same. B-MACresults in the highest power, while TUTWSN performs closest to the Ideal-MAC.As data generation interval increases from 1 s to 1000 s, the power consumptionsof the Ideal-MAC leaf and router nodes are 68 µW to 37 µW, and 270 µW to 37µW, respectively. The power consumption of TUTWSN leaf and router nodes are23.4% to 6.54%, and 18.8% to 6.60% higher than the Ideal-MAC and the same datageneration intervals. IEEE 802.15.4 performs the second best and results in 80.4% to6.64 % and 229% to 8.14% higher power consumption than the Ideal-MAC.

The utilization of LR platform increases power consumptions, as presented in Fig.31. However, TUTWSN MAC maintains its energy-efficiency resulting in the low-est power consumption. The power consumptions of the Ideal-MAC leaf and routernodes are 171 µW to 37 µW, and 945 µW to 38 µW, respectively. The power consump-tion of TUTWSN leaf and router nodes are 27.1% to 2.85%, and 20.2% to 3.18%higher than the Ideal-MAC and the same data generation intervals. IEEE 802.15.4performs the second best and results in 42.1% to 2.92% and 66.3% to 4.33% higherpower consumption than the Ideal-MAC.

The results indicate that the designed channel access mechanism achieves higher

Table 12. Utilized parameter values.Parameter Value (HR) Value (LR)ε 20 ppm 20 ppmLACK , LCT S, LRT S, LP 8B 8BLB, LDATA 32B 32BLSB (SCP-MAC) 2B 2Bn 8 8nF 8 8nDL 3 3PRX 60.2 mW 25.4 mWPS 37 µW 37 µWPT X 34.7 mW 29.9 mWR 1 Mbps 76.8 kbpsSA (TUTWSN) 2 2TSY NC (T-MAC) 90 s 90 stCCA 128 µs 256 µstCW 2 ms 4 mstST 195 µs 250 µs

6.3. Results 77

0.01

0.1

1

10

1 10 100 1000

Data Generation Interval (s)

Le

af

No

de

Po

we

r (m

W)

B-MAC X-MAC SCP-MAC T-MAC802.15.4 TUTWSN Ideal-MAC

0.01

0.1

1

10

1 10 100 1000

Data Generation Interval (s)R

ou

ter

No

de

Po

we

r (m

W)

Fig. 30. Power consumption comparison using HR (nRF2401A) radio.

energy-efficiency than the most essential existing low-power MAC protocols. De-pending on the traffic loading, radio type, and data routing capability, the energyoverhead of TUTWSN nodes is only 2.85% to 27.1%. The high energy-efficiency isachieved in both leaf and router nodes. It should be noted that the TUTWSN channelaccess can operate without the designed networking mechanisms (NCB and ENDP).Then, the neighbor discovery operations are performed similarly with IEEE 802.15.4passive scans.

0.01

0.1

1

10

1 10 100 1000


Le

af

No

de

Po

we

r (m

W)

B-MAC X-MAC SCP-MAC T-MAC802.15.4 TUTWSN Ideal-MAC

0.01

0.1

1

10

1 10 100 1000


Ro

ute

r N

od

e P

ow

er

(mW

)

Fig. 31. Power consumption comparison using LR (CC1000) radio.

7. TUTWSN SENSOR NODE PLATFORMS AND DEPLOYMENTS

This chapter summarizes the research results of TUTWSN sensor node platformsand deployments. The sensor node platforms have been designed for three reasons.First, for measuring and understanding the behavior of low power components inreal operation environment, which is critical for the design of a feasible MAC layer.Second, for measuring the performance of the TUTWSN MAC layer design in largescale networks, which is critical for validating and optimizing the design. Third,for developing and demonstrating new WSN applications and long-lived WSN nodesin practice. Important information obtained from the sensor node platforms havebeen the constraints and non-ideal characteristics of low-power hardware componentsand radio wave propagation, and their effect on protocol implementation, networkperformance, and power consumption.

The TUTWSN sensor node platforms have been designed in parallel with the MAClayer. The platforms have been tailored according to the memory, processing powerand timing accuracy requirements of the TUTWSN protocol stack and the MAClayer. Existing platforms have not been used, since they have not been commerciallyavailable, or their energy consumption has been too high.

7.1 TUTWSN Node Designs

This Thesis presents twelve TUTWSN sensor node platforms, which have been de-signed in the years 2004 to 2006. Along the time, TUTWSN protocol stack has beendeveloped and its memory requirements have been increased. One of the most criti-cal design decisions has been the selection of a radio and MCU. Since the radio is asignificant (in most nodes the largest) source of power consumption [P7], TUTWSNplatforms have utilized simple and low energy radios. MCU has been selected ac-cording to sleep mode power consumption, the energy-efficiency in active mode, andmemory resources for software. In addition, an accurate wake up timer has beenrequired for ensuring link synchronization accuracy.

80 7. TUTWSN Sensor Node Platforms and Deployments

The platforms are presented chronologically and summarized at the end of this Sec-tion. Performance results obtained by the platforms are presented in the next Section.

7.1.1 Node Designs in 2004

The first embedded TUTWSN sensor node platforms were designed in 2004. At thattime, TUTWSN protocol stack contained only the basic MAC functionality and verysimple upper protocol layers.

According to an energy analysis [P7], Nordic Semiconductor nRF2401A [120] radioand Xemics (Semtech) XE88LC02 [163] MCU were selected as the basis of the de-signed platforms. The radio operates at 2.4 GHz frequency band and has 1 Mbps datarate resulting in high energy-efficiency and enabling the utilization of a very smallceramic antenna. The MCU utilizes an energy-efficient CoolRISC core and 16-bitADC. Available 22kB program memory and 1 kB data memory were sufficient forthe protocol stack of that time.

TUTWSN Node

First, a platform called TUTWSN node [P8] was build. The platform was targetedat indoor temperature sensing applications. It was equipped with a simple analogtemperature sensor and a linear voltage regulator. External EEPROM was used fornon-volatile data. The implemented TUTWSN node is presented in Fig. 32.

Antenna

EEPROM

Temperature

sensor

Programming

connector

Test pins

Radio

MCURegulator

0 10 20 mm

Power switch

Fig. 32. TUTWSN node.

7.1. TUTWSN Node Designs 81

Although the platform was energy-efficient [P8], it had some problems in experimen-tal networks. A small ceramic antenna limited the radio range to around 15 meters. Inaddition, tolerances in a low-cost temperature sensor and its voltage reference causedinaccuracy to measurements.

TUTWSN Position Sensor Node

For demonstrating the feasibility of TUTWSN in industrial applications, a platformcalled TUTWSN position sensor node [P9] was implemented. The platform wasequipped with a linear position sensor, a small 150 mAh (3.6 V) rechargeable Nickel-Metal Hydride (NiMH) battery, and it was packed in a robust aluminum casing. Thebattery capacity was adequate, since the required battery lifetime was one month [P9].The Printed Circuit Board (PCB) and the aluminum casing of the node is presentedin Fig. 33.

According to measurements, the TUTWSN position sensor node obtained better radio

AntennaEEPROM

Sensor

connector

Radio

MCUVoltage

regulator

Battery

Programming

connectorDiagnostics

LEDs

Linear position

sensor

WSN node

prototypeMoving axis for

position measuring

Fig. 33. TUTWSN position sensor node.


range than the TUTWSN node. The usability of the platform was also improved bythe robust mechanical design and the on-board battery. In addition, the connectorfor the linear position sensor could be used for charging the battery. Measurementswith the TUTWSN position sensor nodes indicated that TUTWSN technology hasadequate performance for measuring low-speed movements by the sensor element.Observed problems of the platform were low battery capacity, and expensive antenna.In addition, strong fluctuation of a radio link quality caused frequent link failures andnetwork scans, which increased power consumption at long hop lengths.

TUTWSN Gateway

For implementing a sink and a gateway to PC, a platform called TUTWSN gatewaywas designed [P9]. The platform was equipped with an RS-232 serial interface. Forimproving usability, the platform obtained its supply energy directly from the RS-232port using a rectifier and large electrolytic filter capacitors. The TUTWSN gatewayis presented in Fig. 34.

TUTWSN SoC Node

A platform called TUTWSN SoC node was implemented [P8] for experimenting andmeasuring the performance of commercial SoC components integrating radio andMCU. The platform was targeted at a subnode operation.

The platform utilized Nordic Semiconductor nRF24E1 [119] chip, which contains anRF2401A radio, ADC and a 8051 MCU core having 4 kB program memory and256 B data memory. External EEPROM was required for storing program mem-ory. Because of a common radio interface, the platform was compatible with other

Antenna LEDs MCU RS-232 transceiver RS-232 connector Voltage regulator Radio

Filter capacitors Voltage rectifiers Programming connector

Fig. 34. TUTWSN gateway.


TUTWSN platforms. The platform was equipped with a simple temperature sensorand a linear voltage regulator. The TUTWSN SoC node on a 5 cent coin is presentedin Fig. 35. The size of the platform was only 23 mm x 13 mm x 5 mm.

The utilization of the SoC component reduced node size and cost. Yet, an inaccuratewake up timer caused serious problems for link synchronization. Link synchroniza-tion was possible by either extending the frame reception periods, or keeping MCUalways active. According to energy measurements, the latter resulted better energy-efficiency. Still, the measured average power consumption was over two orders ofmagnitude higher than the power consumption of other TUTWSN platforms.


As TUTWSN protocol stack was improved and extended, the memory resources ofthe Xemics MCU exceeded. In 2005, new family of sensor node platforms was de-signed and implemented. Microchip PIC18LF4620 [105] MCU was selected accord-ing to an analysis of the energy-efficiency, sleep mode power consumption, wake uptimers, and memory resources of available MCUs. It should be noted that the energy-efficiency of PIC is around four times worse than in Xemics, which slightly increasesthe power consumption of these platforms.

Antenna nRF24E1

Crystal

EEPROM

Programming

connector

Temperature

sensor

Voltage converter Power switch

Fig. 35. TUTWSN SoC node on a 5-euro-cent coin.


TUTWSN Temperature Sensor Node

First, a platform called TUTWSN temperature sensor node was designed [P2]. Theplatform was designed to tolerate arctic conditions, and all utilized components werespecified with an extended temperature range beginning from -40 C. The range ofthe nRF2401A [120] radio was improved by a loop antenna, which was implementedas a PCB trace and experimentally tuned for the best performance. The accuracy oftemperature measurements was improved by utilizing a Dallas DS620U [42] digitalthermometer. The platform was powered by a lithium battery having high perfor-mance in low temperature conditions. Support for a solar panel as a power sourcewas improved by an under voltage protection circuit in the power subsystem. Thecircuitry was required to eliminate the under voltage lockout condition of MCU andradio, since the supply voltage depends on the illumination.

The implemented TUTWSN temperature sensing node is presented in Fig. 36. Theplatform contained two attachable boards: a MCU board and an extension board,which was equipped with a battery holder. The utilization of detachable extensionboard allowed the change of a power source and easy implementation of applicationspecific sensors and network interfaces.

According to experiments, the platform had good mechanical design for outdoor fieldtests. Operation in temperatures as low as -37.5 C were experimentally verified.The digital thermometer improved highly the accuracy of temperature measurements.Measured transmission range of the loop antenna was up to 500 meters, which drasti-cally reduced the required number of hops, and transmission power levels. Yet, highdirectivity of the antenna reduced the accuracy of node localization.

Antenna

Transceiver

Battery

MCU and

temperature sensor

(bottom side)

Voltage regulator

Push button

and LED

Fig. 36. Implemented TUTWSN temperature sensor node.


TUTWSN Long-Range Node

For improving the applicability of TUTWSN in environmental monitoring applica-tions, a TUTWSN long-range node was implemented. The platform was targetingat over 1 km hop lengths enabling the coverage of large geographic areas by us-ing a relatively small number of nodes. The platform utilized similar design withthe TUTWSN temperature sensor node, except the radio. Due to the increased re-quirements for communication range, the platform utilized a Nordic SemiconductornRF905 [122] radio operating at the 433 MHz frequency band. The radio has 50kbps data rate and 10 mW maximum transmission power. Antenna was selectedand fine-tuned according to experimental measurements. According to the results, ahalf-wave loop was the best tradeoff between size and performance, and had nearlyomni-directional radiation pattern. The TUTWSN long-range node is presented inFig. 37.

Experiments with the platform indicated up to 3 km radio range in a deployed net-work. Thus, very large and sparse networks were possible. Observed problems withthe platform were lower energy-efficiency and scalability than with the 2.4 GHz ra-dio. These were caused by the low data rate, and only few usable channels due tosignificant inter-channel interferences and the narrow 433 MHz frequency band.

Antenna

Transceiver

Battery

MCU and temperature

sensor (bottom side)

Voltage

regulator

Push button

and LED

Fig. 37. TUTWSN long-range node.


TUTWSN Wrist Node

For indoor localization applications, a platform called TUTWSN wrist node was de-signed. Since the platform was intended for human localization and access con-trol applications, the design targeted at significantly smaller size than the previousPIC-based platforms. Due to the size constrains, a small ceramic antenna was se-lected. Since only subnode operation was required, MCU was changed to a smallerPIC18LF2520 [104] model. The node obtained supply power from a CR2450 lithiumcoin cell battery. Since the coin cell battery can supply only around 100 µA current, aNiMH battery was used as a temporary energy storage supplying high current peaks.

The implemented TUTWSN wrist node is presented in Fig. 38. The PCB is circularshaped having a diameter of 30 mm. The PCB is enclosed in a plastic enclosurepresented in the figure. The top side of the board contains the antenna, radio, sensor,push-button and LED. The bottom side contains MCU and a voltage regulator.

Observed problems with the platform were expensive power subsystem and shorttransmission range.

Transceiver

NiMH Battery

Temperature sensor

Antenna

Microcontroller

Voltage regulators

Push button

LED

Connectors for

CR2450 battery

Fig. 38. Implemented TUTWSN wrist node.


TUTWSN Badge Node

For reducing the cost of the power subsystem and improving the battery capacity,a variation the TUTWSN watch node called TUTWSN badge node was designed.Larger physical size allowed the use of AAA-type batteries, accelerometer, and alarger PCB antenna having longer range and a nearly omni-directional radiation pat-tern improving the performance of node localization. The movement of the nodecould be measured by both signal strength and acceleration measurements. The im-plemented TUTWSN badge node with a plastic container is presented in Fig. 39.

Multi-Radio Platform

For minimizing routing latency in real-time control applications, a multi-radio plat-form was developed [P10]. The platform was targeted at a mains-powered backbonenode operation, where absolute minimum latency was needed.

For compatibility with other TUTWSN nodes, the multi-radio platform was equippedwith nRF2401A [120] radios. Efficient routing of multiple frames simultaneouslywas enabled by utilizing four parallel radios controlled by an Altera Cyclone EP1C20[6] Field Programmable Gate-Array (FPGA) circuit. The FPGA has enough gates forfour NIOS processor software cores and MAC layer frame forwarding functionalityas a hardware accelerator implementation. The complete multi-radio platform is pre-sented in Fig. 40. The platform size is 125 mm × 112 mm × 30 mm [P10].

The platform enabled the minimization of routing latency by receiving and transmit-ting frames simultaneously using different radios. The platform determined the nexthop node as soon as the frame header was received after which the frame forwarding

Batteries Temperature sensor MCU Radio LEDs Push buttons

Fig. 39. TUTWSN badge node for human positioning.


Radio boards

I/O boardMother board

Cyclone

device

Programming

connector

DC input

Battery terminals

(bottom side)

Power

switch

7-segment

display

Analog

and digital

I/O

LEDs and

push-buttons

Fig. 40. Multi-radio platform.

began instantly. Theoretical minimum routing delay was as low as 106 µs per a hop[P10].


In 2006, the TUTWSN protocol stack was extended by several new protocols and sen-sor applications. The utilization of many new features was limited by the constrainedmemory resources. In addition, new monitoring applications necessitated the integra-tion of new types of sensors in platforms. Thus, new versions of the TUTWSN sensornode platforms were designed. The design targets were cost-effectiveness, improvedinterfaces for external networks, improved mechanical design, and versatile sensors.

TUTWSN Universal Node

First, a TUTWSN universal node was developed. The previous MCU was replacedwith a larger PIC18LF8722 [106] having 128 kB program memory and 4 kB datamemory. For allowing larger data buffers and the development of SensorOS oper-ating system on nodes, data memory was extended by an external 128 kB SRAM.Previously used nRF2401A radio was changed to nRF24L01 [121] having up to 2Mbps data rate with lower active mode power consumption. Antenna was modifiedfor achieving more omni-directional radiation. The node was equipped with acceler-ation, temperature and illumination sensors. User interface was implemented by pushbutton and two LEDs. In addition, Lantronix XPort [97] was optionally assembled


User interface

and illumination

sensor

Temperature

sensor and MCU

(bottom side)

AntennaRadio

SRAM

Voltage

converters

Place for

Ethernet module

Accelerometer

Battery holder Enclosure

Fig. 41. TUTWSN universal node and the plastic enclosure.

for providing Ethernet connectivity. Hence, the node could also operate as a stand-alone sink having Ethernet gateway functionality. In normal operation, the node waspowered by two AA-batteries. The utilization of Ethernet necessitates the use of amains power adapter.

Power subsystem was improved by a two-step dynamic voltage scaling. A highervoltage setting allowed the utilization of a higher clock frequency, while lower volt-age setting minimized the sleep mode power consumption. Moreover, the sleep modepower consumption was reduced by a Field-Effect Transistor (FET) circuitry, whichswitches off the supply voltage of sensors, when they are not used. The appearanceand mechanical strength of the node was improved by a designed splash-proof plas-tic enclosure. The PCB and the plastic enclosure of the TUTWSN universal node arepresented in Fig. 41.

TUTWSN Long-Range Universal Node

For improving the applicability of the TUTWSN universal node in environmentalmonitoring applications, a TUTWSN long-range universal node was designed. Thenode had similar hardware with the TUTWSN universal node, but the radio wasreplaced to nRF905 [122]. The radio is similar with the TUTWSN long-range nodeproviding up to multiple kilometers range. The node is presented in Fig. 42.


External SRAM

Antenna

Accelerometer

User interface and

illumination sensor

Transceiver

MCU (bottom side)

Voltage

regulator

Connection area

Programming

connector

Enclosure

Fig. 42. TUTWSN long-range universal node.

Multi-Sensor Node

For extending the sensing capability of nodes, a TUTWSN multi-sensor node wasdeveloped. Computing, communication, and power subsystems were similar to theTUTWSN universal node. The sensing subsystem contained sensors for acceleration[204], temperature [42], humidity [167], illumination [12], acoustic (microphone),orientation and magnetic field (compass) [74], image (VGA camera) [148], motion(PIR) [60], and location (GPS) [56]. The TUTWSN multi-sensor node is presentedin Fig. 43.

The designed TUTWSN platforms are summarized in Table 13.

7.1.4 Results

This Thesis has presented twelve TUTWSN sensor node platforms, which have beendesigned according to the requirements of TUTWSN protocols and applications.Along with the evolution of TUTWSN sensor node platforms, their sensing and dataprocessing capabilities have improved, and lifetime increased.

Low power consumption necessitates high energy-efficiency in radio and MCU. Mostof the designed TUTWSN platforms utilize low power and high data rate radios re-


External SRAM (bottom side)

Antenna

Batteries

Illumination sensor

Transceiver (bottom side)

MCU

Voltage regulation

Connection area

Push button and LED

Latches for external SRAM

GPS

GPS antenna

Humidity sensor

Microphone

Temperature sensor

Connector for

VGA camera

Connector for PIR sensor

Compass

Fig. 43. TUTWSN multi-sensor node.

sulting the lowest energy per a transmitted and received bit of data [P7]. Experimentsindicate that the energy efficiency of high data rate radios can be exploited only withan accurate (better than 20 ppm) wake-up timer enabling precise link synchroniza-tion. In addition, high data rate radios necessitate transmission and reception databuffers in the radio, due to the speed limitations of low power MCUs. The data bufferlength determines the maximum frame size. TUTWSN sensor node platforms utilizeradios from Nordic Semiconductor having 32 bytes data buffers, which are relativelyshort for WSN applications. In practice, the buffer size limits the data payload sizeand the number of reserved slots signaled in beacons. Thus, if a radio having largerdata buffers with the same power consumption exists, even higher energy efficiencyand data forwarding performance would be achieved.

Experiments also indicate that high scalability necessitates the utilization of radioshaving a large number of non-interfering channels. Thus, the platforms utilizingnarrow band radios in the wide 2.4 GHz frequency band enable the highest scalabil-ity. Platforms operating in the narrow 433 MHz frequency band suffer from limitedscalability and inter-channel interferences. These platforms are the most suitablefor sparse environmental monitoring networks enabling even kilometers long hoplengths.

The platforms prove that it is feasible to perform versatile and low power sensingin small nodes. To maximize energy-efficiency, it is essential to minimize the duty


Table 13. Summary of TUTWSN sensor node platforms.Platform MCU Radio Sensors and InterfacesTUTWSN node XE88LC02 nRF2401A TTUTWSN position XE88LC02 nRF2401A Lsensor nodeTUTWSN gateway XE88LC02 nRF2401A T, RS-232TUTWSN SoC node 8051 nRF2401A TTUTWSN temperature PIC18LF4620 nRF2401A T, RS-232sensor nodeTUTWSN long-range node PIC18LF4620 nRF905 T, RS-232TUTWSN wrist node PIC18LF2520 nRF2401A TTUTWSN badge node PIC18LF2520 nRF2401A A, TMulti-radio platform 4*NIOS 4*nRF2401A RS-232TUTWSN universal node PIC18LF8722 nRF24L01 A, I, T, Eth, RS-232TUTWSN long-range PIC18LF8722 nRF905 A, I, T, Eth, RS-232universal nodeTUTWSN multi-sensor PIC18LF8722 nRF24L01 A, C, G, M, P, T,node H, I, IM, RS-232

Abbreviations: acceleration (A), compass (C), GPS (G), humidity (H), illumination (I), image (IM),linear position (L), microphone (M), Passive infra-red (P), temperature (T), Ethernet (Eth)

cycle of sensing. Quantities changing slowly are easy to monitor, for example tem-perature and humidity. These sensors can be sampled infrequently at intervals of tensof seconds. A reliable and energy-efficient monitoring of rapidly changing quantitiesis more difficult. Generally, a high frequency sampling of a sensor increases sig-nificantly energy consumption. Higher energy-efficiency is achieved by utilizing alow power comparator circuit, which generates an interrupt to MCU, if a predeter-mined threshold is exceeded. These kind of sensors utilized with TUTWSN are theaccelerometer [204] and PIR motion detection sensor [60].

In the newest platform designs, node lifetime has been increased by the two leveldynamic voltage scaling. The sleep mode current consumption is minimized by us-ing the lowest possible supply voltage, which is 2.0 V in TUTWSN nodes. Higher,2.7 V supply voltage is used only on-demand, when the MCU clock frequency istemporarily increased, or when some sensor requires higher voltage, for example theaccelerometer [204]. According to measurements, dynamic voltage scaling reducesthe sleep mode power consumption around 30% compared to the utilization of a con-stant 2.7 V voltage. When battery voltage drops below 2.7 V, only the lower supply

7.2. TUTWSN Deployments 93

voltage is used, and node functionalities requiring 2.7 V are disabled. Thus, batteriescan be discharged to around 2.1 V voltage extending node lifetime at least tens ofpercents [52].

The measured performance of TUTWSN sensor node platforms and protocol stackin real deployments is presented in the next section.

7.2 TUTWSN Deployments

In this section, three TUTWSN deployments are presented. The main objective ofthe deployments is to prove and demonstrate that the MAC and platform designs,and WSN applications are feasible in practice in real operation conditions. Powerconsumption measurements during the deployments provide information about theenergy consumptions of data processing, sensing, and random synchronization errors,which are difficult to estimate analytically.

The first deployment is a small indoor temperature sensing network, where nodesutilize a simple protocol stack and the basic MAC functionality. The deploymentproves that the TUTWSN channel access mechanism and sensor node platforms arefeasible, and they can achieve high energy-efficiency in practice.

The second deployment discusses a long-term outdoor environmental monitoring net-work. The deployment verifies the operation of the designed MAC mechanisms andthe TUTWSN long-range nodes in harsh outdoor conditions.

The third deployment is an indoor building monitoring network, where nodes uti-lize the full TUTWSN protocol stack containing all designed channel access and net-working mechanisms. The deployment proves that high energy-efficiency is achievedwith the full MAC implementation, with mobile nodes, and with a larger and densernetwork size than the previous deployments.

The deployments are presented as follows. First, the application is presented shortly.Then, the power analysis of utilized platforms is presented, which is followed by theexperimental results of the deployment.

7.2.1 Indoor Temperature Sensing Deployment

Temperature sensing is one of the most straightforward applications for WSNs, andit can be used for example for Heating, Ventilation & Air Conditioning (HVAC)control in building. Compared to wired sensors, the utilization of WSN technology


will reduce cabling and installation costs, and it makes the system easily extendable.The data rate and latency requirements are low, but the required node lifetime shouldbe years [P8].

The indoor temperature sensing deployment [P8] was carried out in 2004. The targetof the deployment was to prove that the TUTWSN channel access mechanism withaccurate local time synchronization is feasible, and that the high energy-efficiencycan be achieved in practice using the designed sensor node platforms. Due to prac-tical reasons, the measured network is rather small containing five TUTWSN nodes,one TUTWSN SoC node, and a TUTWSN gateway. Nodes utilized a simple protocolstack, which consists of the channel access mechanism, path-loss estimation scheme,network channel beaconing mechanism, and simple upper protocol layers. The nodesacquired temperature values at 5 s intervals, and routed data to the TUTWSN gate-way. Utilized network beacon interval was fixed to 500 ms [P8].

Power Analysis

The power analysis measures the power consumptions of the main components indifferent operation modes. In the all presented deployments, power analysis is per-formed as follows. MCU power consumption is measured in active and sleep modes.The sleep mode, only a wake up timer is active. The ADC and sensor power con-sumptions are measured in active and shut down (off) modes. The radio power con-sumption is measured in sleep, reception, and transmission modes with a minimumand maximum transmission power levels. Results are measured using 3 V batteryvoltage. The results are used in the performance analysis for specifying the powerconsumption characteristics of the hardware.

The power analysis results of the TUTWSN SoC node and the TUTWSN node arepresented in Table 14 [P8]. The sleep mode power consumptions for the platformsare 15 µW and 24 µW. The maximum power consumptions are 66.03 mW and 59.55mW, which are achieved when all components are in the active mode and the radio isin the receive mode. The radio clearly dominates the momentary power consumptionin both platforms, especially in the reception mode. The node power consumptionover doubles during reception compared to transmission using the lowest -20 dBmtransmission power. The analysis also shows that the 8051 core of the TUTWSN SoCnode has 6 times higher power consumption than the CoolRISC core of TUTWSNnode [P8].


Table 14. The power analysis of the TUTWSN node and TUTWSN SoC node.MCU ADC Sensor Radio TUTWSN TUTWSN SoCmode mode mode mode node (mW ) node (mW )active active active RX 59.55 66.03active active active TX (0dBm) 37.70 44.10active active active TX (−20dBm) 26.60 32.90active active active sleep 3.04 11.64active active off sleep 3.01 11.60active off off sleep 1.69 10.56sleep off off sleep 0.024 0.015

Experimental Results

First, the functionality and timing accuracy of the MAC design is verified by os-cilloscope measurements. The current consumption waveforms of a subnode and aheadnode during a superframe are presented in Fig. 44. Transmissions and receptionsproduce narrow and high current peaks indicating precise time synchronization andvery low idle listening. MCU activity is shown after transmissions and receptions aswider and lower current peaks. In the figure, subnode receives network and clusterbeacons, transmits a slot reservation request in a contention slot and a data framein a contention-free slot. After each transmission, the subnode receives ACK. Theheadnode exchanges data with other nodes resulting activity in the fourth contentionslot and second contention-free slot [P8].

Contention slots Contention-free slotsNetwork

beacons

Cluster

beacons

RX

TX

RXRX RX

TX

RX

TX

RX

TX

RXRXRXRX

TX

TXTX

TX

Headnode

Subnode

TX

Contention slots Contention-free slotsNetwork

beacons

Cluster

beacons

RX

TX

RXRX RX

TX

RX

TX

RX

TX

RXRXRXRX

TX

TXTX

TX

Headnode

Subnode

TX

Fig. 44. Measured current consumption waveforms of a subnode and a headnode during asuperframe.


Table 15. Measured subnode power consumptions and estimated lifetimes.Platform Average Lifetime with Lifetime with

power CR2450 2 x AATUTWSN node 88.2 µW 879 days 10.1 yearsTUTWSN SoC node 10.6 mW 6.6 days 30.5 daysTUTWSN SoC node with 80.7 µW 960 days 11.0 yearspower saving (estimated)

Then, the power consumptions of the platforms operating as subnodes are measured.The power results with 5 s MAC access cycle length are presented in Table 15 [P8].The resulted power consumptions of TUTWSN node and TUTWSN SoC node are88.2 µW and 10.6 mW. The significantly higher power consumption in the SoC nodeis caused by an inaccurate wake-up timer having inadequate accuracy for timing thechannel access mechanism. The clock frequency of the timer varied between 1 kHzand 5 kHz depending on temperature and process variables. However, the calibrationof the timer was possible, but a small program memory size prevented its implemen-tation in practice. Thus, sleep mode was not utilized during the measurements. Theestimated power consumption with the full power saving would be 80.7 µW [P8].

According to the power consumption results, node lifetimes are estimated with twobattery types: a CR2450 lithium coin battery and two serially connected AA alkalinebatteries. With the full power saving, the lifetimes of both platforms with the CR2450are about 900 days, and with the two AA batteries over 10 years. This proves theenergy-efficiency of TUTWSN subnodes in practice. The lifetime of the TUTWSNSoC node with the limited power saving is below a month. This clearly shows theimportance of sleep mode power consumption and timing accuracy for the achievedpower consumption [P8].

Next, the power consumption of a headnode is measured using the TUTWSN nodeplatform. The MAC access cycle length is fixed to 5 s, and the number of associatedsubnodes is varied from 0 to 4. The results indicate that headnode power consumptionincreases from 390 µW to 560 µW, as presented in the left side of Fig. 45 [P8]. Asthe headnode operation is rotated among nodes, the average power consumption ofthe TUTWSN nodes reduce from 390 µW to 183 µW depending on the number ofsubnodes [P8].

The right side of Fig. 45 presents the resulted lifetimes using two AA batteries.The calculated lifetime of the headnode is between 2.3 and 1.6 years, depending onthe number of subnodes. As the rotation of headnode operation between nodes isapplied, the lifetime of nodes is up to 4.9 years, as the number of subnodes is four


0

100

200

300

400

500

600

0 1 2 3 4

Subnodes per headnode

Po

wer

co

nsu

mp

tio

n (

µW

)

Headnode

Network average

0

1

2

3

4

5

0 1 2 3 4

Subnodes per headnode

Lif

eti

me (

years

)

Headnode

Network average

Fig. 45. Measured power consumptions of TUTWSN nodes.

times higher than the headnodes. This proves the energy efficiency of TUTWSNheadnodes in practice.

The deployment proves that the designed channel access mechanism is functional andfeasible in practical implementations. As the drift of a wake-up timer clock source isin the order of 20 parts-per-million (ppm) or less, the energy-efficiency of subnodesand headnodes is very high enabling long-lived WSN applications. Similar powerconsumption results are also obtained from a linear position sensing deployment per-formed in 2004 [P9].

7.2.2 Environmental Monitoring Deployment

Environmental monitoring is one of the most potential WSN applications. Such net-works can be utilized in hostile environments and in large geographical areas to pro-vide accurate and localized data. By conventional wireless technology, the imple-mentation of such functionality is very expensive and difficult requiring large batterypacks and manual network configuration.

The described deployment has been the first large-scale and long-term field test ofTUTWSN network. The deployment has been active from autumn 2005 and is stillrunning. A four months time period nearly at the beginning of the deployment hasbeen published in [181]. The objective of the deployment has been to prove thatthe designed TUTWSN long-range nodes can survive and maintain high energy-efficiency in arctic operation conditions having high temperature variations and mois-ture levels. In addition, an objective was to prove that the designed path-loss esti-mation scheme is feasible in harsh outdoor conditions, and that nodes can maintain


adequate link reliability with long hop lengths.

The deployment utilized 20 TUTWSN long-range nodes, which formed a multi-hopnetwork in an area of few square kilometers. One of the nodes was equipped with aRS-232 interface and operated as a sink and a gateway to a computer. The resultednetwork was very sparse and the distances between nodes were up to 3 km. Nodesutilized a MAC layer utilizing the designed channel access, NCB and path-loss me-tering mechanisms, and the multi-cluster-tree topology. ENDP was not used. MACutilized 2 s access cycle length, 500 ms network beacon interval, 4 contention slotsand at maximum 8 contention-free slots per a superframe. Upper protocol layers con-tained a simple multi-hop routing protocol, API, and applications for data gatheringand node diagnostics [181].

Power Analysis

The results of a power analysis are presented in Table 16. The sleep mode power con-sumption is 31 µW. The maximum power consumption is 95.9 mW, which is achievedas all components are in active mode and the nRF905 radio is in transmission modeat 10 dBm transmission power. The power consumption is significantly higher thanin nodes having the nRF2401A radio. Also, radio data rate is only 50 kbps, whichfurther increases the energy consumption of exchanged frames. However, expectedtransmission range is nearly ten times longer than with the nRF2401A radio, whichcompensates well the higher energy consumption.

Table 16. The static power consumption of the TUTWSN long-range node.MCU mode ADC mode Sensor mode Radio mode Power (mW )

active active active TX (10dBm) 95.9active active active TX (6dBm) 65.9active active active TX (−2dBm) 47.9active active active TX (−10dBm) 32.9active active active RX 43.3active active active sleep 5.91active active off sleep 3.68active off off sleep 3.17sleep off off sleep 0.031


Experimental Results

The deployment provided experimental results about the network operation in forested,low temperature outdoor environment. According to the experiments, a forest atten-uates radio wave propagation very significantly. Achieved radio range in a forest wasbelow 100 meters, while the longest measured hop lengths were near to 3 km. Also,radio wave propagation was notably affected by snow fall, rain, humidity, temper-ature, and the frost and snow in trees, ground, and around the nodes. Hence, thequality of radio links altered continuously although the nodes were stationary [181].

The outdoor temperatures during the test period from November 2005 to March 2006ranged from -31.5 C to 12.0 C. High temperature variation reduced significantlythe accuracy of crystals and thus, the accuracy of time synchronization. Some nodeswere inside buildings and other outdoors resulting in nearly 50 C difference in theoperation temperatures. According to the experiments, the implemented sensor nodeplatforms performed well during the whole test period [181].

Network lifetime with 1600 mAh CR123 batteries was 6 - 8 months. As a com-parison, the calculated lifetime using two AA-type batteries per node is around oneyear. It should be noted that the network traffic consisted not only of temperaturemeasurements but also diagnostics information, which increased load. Lifetime waslower than in the deployments utilizing nRF2401A radios due to the higher powerconsumption and lower data rate of the radio [181].

7.2.3 Building Monitoring Deployment

The building monitoring deployment has been operating in indoor conditions at Tam-pere University of Technology premises from the year 2005. The network has beenused for demonstrating the TUTWSN technology, for verifying the operation of de-signed protocols, and for measuring the network performance. The network has beenmeasuring various quantities, for example temperature, humidity, CO2, motion, ac-celeration, and illumination [95, 182, 184], [P4].

The presented deployment [95] was carried out in 2006. One of the main objectives ofthe deployment was to prove that the ENDP scheme and multi-cluster-tree topologyare energy-efficient and feasible in practice. The deployment was also used for mea-suring the energy consumption of link failures and neighbor discovery operations,and to evaluate the effect of node mobility on them.

The deployment utilized up to 28 TUTWSN temperature sensor nodes, which mea-


sured temperature values at 30 s intervals. Ten of the nodes were equipped with a PIRmotion detection sensor. The platforms are executing the TUTWSN protocol stackcontaining the full MAC layer design, an interest-based routing protocol, API, andapplications for sensing and self-diagnostics [182]. Utilized access cycle length andnetwork beacon interval were 2 s and 500 ms [95].

Power Analysis

The results of a power analysis are presented in Table 17. The sleep mode powerconsumption is 37 µW. The power consumption at transmission mode is between29.57 mW and 42.17 mW, as the transmission power level ranges from -20 dBm to 0dBm. The maximum power of 60.17 mW is consumed as MCU is active and radio isin the reception mode.

Experimental Measurements

The deployment provided two main results: the power consumption of a stationarynetwork, and the performance of the ENDP scheme in a mobile network.

The power consumptions of various subnodes and headnodes were measured in a sta-tionary network, where headnodes routed data to a sink using several parallel routes.The routes were changing dynamically for adapting to varying RF propagation con-ditions and for balancing energy consumptions between nodes. ENDP was not usedand the number of maintained synchronized links (k) was fixed to 2. The averagepower consumptions of subnodes and headnodes were 257 µW and 693 µW [95].

The results are 134 µW and 300 µW higher than in the MAC layer performance anal-ysis. The difference is mostly caused by the energy consumptions of data processing,

Table 17. The static power consumption of the TUTWSN temperature sensor node.MCU mode Radio mode Power (mW )

active RX 60.17active TX (0dBm) 42.17active TX (−6dBm) 34.67active TX (−12dBm) 31.37active TX (−20dBm) 29.57active idle 3.29active sleep 3.17sleep sleep 0.037


sensor sampling, and the frame exchanges caused by the routing layer, which areignored in the analysis. In addition, the implementation is not fully optimized. Uti-lized time margins for reception are longer than the optimal margins utilized in theanalysis. By noticing the non-ideal implementation, the obtained experimental mea-surements are analogous with the analytical results, and the energy-efficiency of theMAC layer design is verified.

The performance of ENDP was measured in a network containing eight stationaryheadnodes and one mobile subnode [P4]. During the measurements, k was variedfrom 2 to 4. The mobile node transmitted a 32 byte diagnostics data frame on acontention-free time slot at 8 s intervals. In addition, control frames for associa-tion and slot reservation were transmitted after communication link changes. Threemobility levels were utilized: 0 m/s, 1 m/s, and 3 m/s [P4].

The measured power consumption are presented in Fig. 46. Without ENDP, thepower consumptions at 0 m/s, 1 m/s and 3 m/s mobility levels were 326 µW, 2.18mW and 4.76 mW. As ENDP was utilized, the power consumptions with an optimalk value were 627 µW (k = 2), 933 µW (k = 3), and 968 µW (k = 3). Thus, the achievedenergy saving is 57% - 80% compared to the situation without ENDP. At higher val-ues of k, the energy consumption of maintained synchronized links increases, whichreduces the achieved energy-efficiency. At lower values of k, the probability of find-ing a suitable neighbor using SDUs decreases increasing the energy consumption ofnetwork scans [P4].

The measured power consumption is higher than the analysis results presented in theChapter 5. This is caused by the energy consumptions of data processing and thesynchronization inaccuracy of a protocol implementation. In addition, the operation

0

1

2

3

4

5

Mobility: no Mobility: 1 m/s Mobility: 3 m/s

Measu

red

po

wer

(mW

) without ENDPENDP, k = 2ENDP, k = 3ENDP, k = 4

Fig. 46. Measured power consumption with and without ENDP at three levels of mobility.


environment had significant fluctuation in the radio link quality depending on nodeposition due to shadowing and multi-path propagation. Thus, the actual link failurerate was much higher than in the analysis. However, the results prove that ENDPimproves significantly the energy-efficiency of mobile nodes. ENDP can be usedwith IEEE 802.15.4, TUTWSN MAC and other synchronized low duty-cycle MACprotocols having regular beacon transmissions [P4].

8. SUMMARY OF PUBLICATIONS

The publications of this Thesis are based on the work of the author during yearsbetween 2003 and 2007. This chapter summarizes the contents of the publicationsand clarifies the contribution of the author. The co-authors have seen and agree withthe descriptions. None of the publications have previously been used as a part of adoctoral thesis.

The publications can be divided into three main groups. The publications [P1] -[P4] propose protocols and energy-efficient mechanisms for WSNs. The publications[P5], [P6] cover the performance analysis and optimizations of existing protocols.The publications [P7] - [P10] discuss the implementation of hardware prototypes andreal application deployments.

Publication [P1] is a book chapter presenting a summary of TUTWSN MAC layerdesign. Sections 13.1 - 13.3. present the channel access mechanism, network topol-ogy formation and superframe interlacing, which are previously unpublished mate-rial. Sections 13.4 - 13.8. present the energy-efficient neighbor discovery protocol,the path-loss estimation scheme and the measured power consumption of TUTWSNMAC, which have been reproduced from the publications [P2], [P4], [P5], [P9],and [183].

The author has been the main architect of the TUTWSM MAC layer, and he hasspecialized in energy-efficiency. The publication was mainly written by the author.Jukka Suhonen reproduced the section 13.5 from publication [183]. Mauri Kuori-lehto gave important ideas for the MAC design and revised the draft version of thepublication. Prof. Marko Hännikäinen and Prof. Timo D. Hämäläinen have super-vised the work, outlined the functional requirements, and revised the draft version ofthe publication.

Publication [P2] presents the path-loss estimation scheme, which allows signal strengthmeasurements using simple radios without RSSI functionality. The performance ofthe scheme is experimentally verified by indoor and outdoor measurements. In addi-tion, performance analysis indicates very low energy-overhead of the scheme, and a

104 8. Summary of Publications

significant energy saving in ZigBee by replacing an IEEE 802.15.4 compliant radioby a simpler one.

The author designed the path-loss estimation scheme, as well as designed and im-plemented hardware platforms. In addition, the author developed the performancemodels, performed the performance analysis, and experimental performance mea-surements. The publication was written by the author. Jukka Suhonen implementedby scheme in software and gave ideas for optimizing the performance. Prof. MarkoHännikäinen and Prof. Timo D. Hämäläinen gave ideas for the mechanism and re-vised the draft version of the publication.

Publication [P3] proposes the use of the network channel beaconing scheme in Zig-Bee. The publication presents an introduction on ZigBee protocol stack and neighbordiscovery operation, and implementation guidelines for the network channel beacon-ing scheme. The performance of the scheme is proven by a performance analysis andsimulations. In addition, an optimal network beacon rate has been determined as thefunction of node mobility and cluster size.

The author designed the network channel beaconing scheme, as well as performed theperformance analysis and optimizations. The publication was written by the author,except the Section 7. Jukka Suhonen performed the simulations by NS-2 and wrotethe Section 7 presenting the simulation results. In addition, Mauri Kuorilehto, Prof.Marko Hännikäinen and Prof. Timo D. Hämäläinen gave ideas for the mechanismand revised the draft version of the publication.

Publication [P4] presents the energy-efficient neighbor discovery protocol for mo-bile WSNs. The publication presents the mechanism for proactive distribution ofneighbor information and a neighbor discovery algorithm for using the distributedinformation for neighbor discovery. The performance of the protocol is proven byperformance analysis and experimental measurements.

The author has been the main architect of the energy-efficient neighbor discoveryprotocol. He designed the mechanisms for distributing and utilizing neighbor in-formation in network, performed performance analysis and optimizations, designedand implemented hardware prototypes, and performed experimental measurements.Jukka Suhonen improved the utilization of the neighbor information by designing aneighbor discovery algorithm, and wrote the software with the help of Ville Kaseva.The main parts of publication were written by the author. Jukka Suhonen wrote theSubsection 3.2 presenting the pseudocode of the algorithm. Mauri Kuorilehto, Prof.Marko Hännikäinen and Prof. Timo D. Hämäläinen gave ideas for the protocol andrevised the draft version of the publication.

105

Publication [P5] presents the optimization of beacon transmission rate for reducingenergy consumption in synchronized low-duty cycle MAC protocols. The publicationdivides the energy consumption of a WSN node in a node start-up, data exchange andnetwork maintenance energies, and then focuses on minimizing the network mainte-nance energy. It has been shown that the network maintenance power is a sum ofbeacon exchange and network scanning powers, and an optimal network beacon rateis a function of average link failure interval, and radio power consumption.

The author determined the performance models and analysis for beacon rate opti-mization, as well as designed and implemented the reference hardware platform. Thepublication was written by the author. Prof. Marko Hännikäinen and Prof. Timo D.Hämäläinen gave ideas for the protocol and revised the draft version of the publica-tion.

Publication [P6] presents a performance analysis of ZigBee in a large-scale wirelesssensor network application. The analysis focuses on a cluster-tree network topol-ogy and beacon-enabled mode. Analysis includes models for a hardware platform,channel access mechanism, and contention. As results, average power consumptions,goodputs and energy-efficiencies are obtained with various MAC parameter values.Analysis results are validated by simulations.

The author determined the performance models and performed the performance anal-ysis. The publication was written by the author, except Section 7. Mauri Kuorilehtoperformed simulations and wrote the Section 7 presenting the simulation results.Prof. Marko Hännikäinen and Prof. Timo D. Hämäläinen gave ideas for the anal-ysis and revised the draft version of the publication.

Publication [P7] presents a study and analysis of energy-efficient components forWSN node platforms. In addition, photovoltaics and magnetic coupling based en-ergy scavenging mechanisms are experimentally measured. According to the study,a WSN node prototype is implemented using partly evaluation boards of componentmanufacturers. In addition, a random access communication protocol and a simplesensing application has been implemented for demonstrating the performance of theprototype.

The author performed the study, analysis, and measurements. In addition, the authordesigned and implemented the WSN node prototype and the software. The publi-cation was written by the author. Prof. Marko Hännikäinen and Prof. Timo D.Hämäläinen gave ideas for the publication and revised the draft version of the publi-cation.

Publication [P8] presents the design and implementation of an ultra low energy WSN

106 8. Summary of Publications

for in-building temperature sensing. The publication contains the design, imple-mentation and performance measurements of two prototypes: TUTWSN node, andTUTWSN SoC prototype. The presented WSN executes a simple TUTWSN protocolstack containing the MAC layer, a multi-hop routing with a fixed stationary routingtree, and a simple sensing application. Results indicate nearly 5 years network life-time for TUTWSN nodes, when each node is powered by two AA-type batteries.Moreover, TUTWSN SoC prototype is around the size of a 5 cents coin, but inaccu-rate wake-up timer limits severely its energy-efficiency.

The author performed the design and implementation of the prototypes. In addition,the author designed the simple protocol stack utilized in the prototypes. Moreover,the author performed the performance measurements and analysis. The publicationwas written by the author. Prof. Marko Hännikäinen and Prof. Timo D. Hämäläinengave ideas for the prototypes and revised the draft version of the publication.

Publication [P9] presents the design and implementation of a WSN for industrial lin-ear position sensing application. The design and implementation of position sensingprototype and TUTWSN gateway having RS-232 interface are presented. The im-plemented network executes the same simple TUTWSN protocol stack than in [P8].Results indicate very high energy-efficiency, but the small capacity of used recharge-able battery limits network lifetime to around 2 months.

The author designed and implemented the prototypes. In addition, the author de-signed the protocol stack utilized in the prototypes, and performed the performancemeasurements and analysis. The publication was written by the author. Prof. MarkoHännikäinen and Prof. Timo D. Hämäläinen gave ideas for the prototypes and revisedthe draft version of the publication.

Publication [P10] presents the design and implementation of a high-performancemulti-radio platform. The publication focuses on hardware design covering electron-ics design and a multi-processor design in FPGA. For demonstrating the applicabilityof the platform, two reference applications are presented, which are targeting to min-imize routing latency, and to maximize routing throughput.

The author designed and implemented the electronics of the prototype. In addition,the author designed and analyzed the reference applications. Tero Arpinen designedand implemented the multi-processor design in the FPGA. The main parts of thepublication was written by the author. Tero Arpinen wrote the Section 3 present-ing the multi-processor architecture. Prof. Marko Hännikäinen and Prof. Timo D.Hämäläinen gave ideas for the prototype design and revised the draft version of thepublication.

9. CONCLUSIONS

The emerging of low-energy WSNs has been possible by the recent advances in low-power communication and computing circuits. Still, the power consumption of hard-ware is too high for long-lived WSN applications without energy-efficient communi-cation protocols and applications. The main research problem is the combination ofa scalable and adaptive multi-hop networking with constrained hardware resourcesand scarce energy budget.

This Thesis has presented a survey of existing energy-efficient MAC protocols andstandards. It was shown that none of existing MAC protocols fulfilled adequately theenergy-efficiency, scalability, and adaptivity requirements of low-energy WSNs. Thismotivated the design of a new MAC layer called TUTWSN MAC concentrating onchannel access and networking mechanisms. The main focus has been on ultra-lowduty cycle frame exchanges, scalable network self-configuration, and energy-efficientneighbor discovery mechanisms improving adaptivity against network dynamics.

The second outcome of this Thesis has been sensor node platforms. The developmentof low-power platforms has been vital for measuring, validating, and demonstratingthe designed communication protocols in real operation conditions. A survey of low-power hardware components and sensor node platforms revealed the potential to de-sign lower energy platforms than existing ones. This motivated the design of twelveTUTWSN sensor node platforms, which were emerged together with the TUTWSNprotocol stack and applications. The performance measurements of the platformswere used for developing and optimizing MAC layer mechanisms.

The third outcome of this Thesis has been performance models and analysis. Per-formance models have been presented for the state-of-the-art channel access mecha-nisms and their energy-efficiencies have been compared against TUTWSN MAC.

The results of this Thesis enable the implementation of multi-hop WSNs in harsh anddynamic operation conditions with years of lifetime using current low-cost compo-nents. In the MAC layer, the results have been achieved by minimizing idle listening,overhearing, collisions, and control traffic overhead. In sensor node platforms, the

108 9. Conclusions

results have been achieved by using transceivers having high data-rate and severalnon-interfering channels, and hardware components having low sleep mode powerconsumption and accurate wake-up timers. The performance analysis and compar-ison to existing proposals and standards have shown that TUTWSN MAC achievesthe highest energy-efficiency in both router and leaf nodes. Compared to the theoreti-cally ideal MAC, the energy consumption of TUTWSN MAC is only 2.85% - 27.1%higher, depending on traffic load, radio, and node type. IEEE 802.15.4 performsthe second best resulting in 2.92% to 229% energy overhead compared to the idealMAC. It has been shown that TUTWSN can maintain high energy-efficiency also indynamic networks. The long-term deployments with the platforms have validated thefeasibility of the MAC layer and platform designs in real operation environment.

The presented MAC layer mechanisms are presented and analyzed clearly and sep-arately on each other, and they can be utilized with other MAC layer designs, andsensor node platforms. The developed sensor node platforms present a good low-power hardware design and they are applicable for experimenting other applicationsand protocols. The performance models can determine the energy consumption ofchannel access mechanisms simply, clearly, and accurately, and they can be easilyadapted to other hardware, protocols, and applications.

Although the absolute energy consumption of low-power components constantly re-duces, it can be assumed that the operation principles will not change dramatically.Thus, the principles and results of this Thesis are valid also in future, while it is ex-pected that the power consumption, physical size, and cost of implementations willreduce.

Bibliography 109

[1] Advanced Analogic Technologies, Inc., “Product datasheet AAT1112 -1.5A, 1.4MHz step-down converter,” Available: http://www.analogictech.com/products/digitalfiles/AAT1112.pdf, 2007.

[2] S. Agarwal, R. Katz, S. Krishnamurthy, and S. Dao, “Distributed power con-trol in ad hoc wireless networks,” in Proc. 12th IEEE International Symposiumon Personal, Indoor and Mobile Radio Communications (PIMRC’01), vol. 2,San Diego, CA, USA, September 30 - October 3 2001, pp. 59–66.

[3] I. F. Akyildiz and I. H. Kasimoglu, “Wireless sensor and actor networks: Re-search challenges,” Elsevier Ad Hoc Networks, vol. 2, no. 4, pp. 351–367,October 2004.

[4] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci, “A survey onsensor networks,” IEEE Communications Magazine, vol. 40, no. 8, pp. 102–114, August 2002.

[5] ——, “Wireless sensor networks: a survey,” Elsevier Computer Networks,vol. 38, no. 4, pp. 393–422, March 2002.

[6] Altera, Corp., “Section I. cyclone FPGA family data sheet,” Available: http://www.altera.com/literature/hb/cyc/cyc_c5v1_01.pdf, 2007.

[7] J. Ammer and J. Rabaey, “Low power synchronization for wireless sensornetwork modems,” in Proc. IEEE Wireless Communications and NetworkingConference (WCNC’05), vol. 2, New Orleans, LA, USA, March 13-17 2005,pp. 670–675.

[8] D. K. Arvind and K. J. Wong, “Speckled computing: Disruptive technology fornetworked information appliances,” in Proc. IEEE Int’l Symposium on Con-sumer Electronics (ISCE’04), Reading, UK, September 2004, pp. 334–338.

[9] Atmel, Corp., “8-bit AVR microcontroller with 128K bytes in-systemprogrammable flash,” Available: http://www.atmel.com/dyn/resources/prod_documents/0945s.pdf, 2001.

[10] ——, “AT91 ARM Thumb microcontrollers AT91FR40162S,” Available:http://www.atmel.com/dyn/resources/prod_documents/doc6174.pdf, 2005.

[11] ——, “8-bit flash microcontroller AT89C51RE2,” Available: http://www.atmel.com/dyn/resources/prod_documents/doc7663.pdf, 2007.

110 Bibliography

[12] Avago Technologies, Ltd., “Data sheet - APDS-9002 miniature surface mountambient light photo diode,” Available: http://www.avagotech.com/, 2006.

[13] G. Barrenetxea, F. Ingelrest, G. Schaefer, and M. Vetterli, “Wireless sen-sor networks for environmental monitoring: the SensorScope experience,”in Proc. IEEE International Zurich Seminar on Communications, Zurich,Switzerland, March 12-14 2007, pp. 98–101.

[14] M. A. Batalin, M. Rahimi, Y. Yu, D. Liu, A. Kansal, G. S. Sukhatme, W. J.Kaiser, M. Hansen, G. J. Pottie, M. Srivastava, and D. Estrin, “Call and re-sponse: experiments in sampling the environment,” in Proc. 2nd Int’l Conf.on Embedded Networked Sensor Systems (SenSys’04), Baltimore, MD, USA,November 3–5 2004, pp. 25–38.

[15] M. Baunach, R. Kolla, and C. Mühlberger, “Beyond theory: Development ofa real world localization application as low power wsn,” in Proc. 32nd IEEEConference on Local Computer Networks (LCN’07), Dublin, Ireland, October15–18 2007, pp. 872–884.

[16] R. Beckwith, D. Teibel, and P. Bowen, “Report from the field: Results from anagricultural wireless sensor network,” in Proc. 29th Annual IEEE Int’l Conf.on Local Computer Networks (LCN’04), Tampa, FL, USA, November 16–182004, pp. 471–478.

[17] L. Benini, A. Bogliolo, and G. D. Micheli, “A survey of design techniques forsystem-level dynamic power management,” IEEE Transactions on Very LargeScale Integration Systems, vol. 8, no. 3, pp. 299–316, June 2004.

[18] J. Beutel, M. Dyer, M. Hinz, L. Meier, and M. Ringwald, “Next-generationprototyping of sensor networks,” in Proc. 2nd Internation Conf. on EmbeddedNetworked Sensor Systems (SenSys’04). Baltimore, MD, USA: ACM Press,November 2004, pp. 291–292.

[19] I. Bluetooth SIG, “Bluetooth low energy technology,” Available: http://www.bluetooth.com/Bluetooth/Products/Low_Energy.htm, 2008.

[20] Specification of the Bluetooth System, Bluetooth Special Interest Group (SIG)Std. 2.0 + EDR, 2004.

[21] B. Bougard, F. Catthoor, D. Daly, A. Chandrakasan, and W. Dehaene, “Energyefficiency of the IEEE 802.15.4 standard in dense wireless microsensor net-works: modeling and improvement perspectives,” in Proc. Design, Automation

Bibliography 111

and Test in Europe (DATE’05), vol. 1, Munich, Germany, March 7-11 2005,pp. 196–201.

[22] T. Bourke, “ISA100.11a completely obviates the need for WirelessHART,”Measurement and Testing, vol. 1, no. 1, pp. 18–19, October 2007.

[23] BTnode project @ ETH Zurich, “BTnode rev3 hardware ref-erence,” Available: http://www.btnode.ethz.ch/Documentation/BTnodeRev3HardwareReference/, 2007.

[24] M. Buettner, G. Yee, E. Anderson, and R. Han, “X-MAC: A short preambleMAC protocol for duty-cycled wireless sensor networks,” in Proc. 4th ACMConf. Embedded Networked Sensor Systems (SenSys’06), Boulder, Colorado,USA, October 31 – November 3 2006, pp. 307–320.

[25] K. Bult, A. Burstein, D. Chang, M. Dong, M. Fielding, E. Kruglick, J. Ho,F. Lin, T. Lin, W. Kaiser, H. Marcy, R. Mukai, P. Nelson, F. Newburg, K. Pister,G. Pottie, H. Sanchez, K. Sohrabi, O. S. amd K.B. Tan, G. Yung, S. Xue, andJ. Yao, “Low power systems for wireless microsensors,” in Proc. InternationalSymposium on Low Power Electronics and Design (ISLPED’96), Monterey,CA, USA, August 12–14 1996, pp. 17–21.

[26] P. Buonadonna, D. Gay, J. Hellerstein, W. Hong, and S. Madden, “TASK:Sensor network in a box,” in Proc. 2nd IEEE European Workshop on WirelessSensor Networks (EWSN’05), Istanbul, Turkey, January 31 - February 2 2005,pp. 133–144.

[27] X. Carcelle, T. Dang, and C. Devic, “Industrial wireless technologies: appli-cations for the electrical utilities,” in Proc. IEEE International Conference onIndustrial Informatics (INDIN’06), Singapore, August 16-18 2007, pp. 108–113.

[28] A. Chandrakasan, R. Min, M. Bhardwaj, S.-H. Cho, and A. Wang, “Power-aware wireless microsensor systems,” in Proc. 28th European Solid-State Cir-cuits Conference (ESSCIRC’02), Firenze, Italy, September 24-26 2002, pp.47–54.

[29] G. Chavira, S. Nava, R. Hervas, J. Bravojose, and C. Sanchez, “Combiningrfid and nfc technologies in an ami conference scenario,” in Proc. 8th MexicanInternational Conference on Current Trends in Computer Science (ENC’07),Morelia, Mexico, September 24-28 2007, pp. 165–172.

112 Bibliography

[30] I. Chlamtac, A. Farago, A. Myers, V. Syrotiuk, and G. Zaruba, “Adapt: A dy-namically self-adjusting media access control protocol for ad hoc networks,”in Proc. IEEE Global Telecommunications Conference (GLOBECOM’99),vol. 1A, Rio de Janeiro, Brazil, December 5-9 1999, pp. 11–15.

[31] C. Y. Chong, S. Mori, and K. C. Chang, “Distributed multitarget multisen-sor tracking,” in Multitarget Multisensor Tracking: Advanced Applications,Y. Bar-Shalom, Ed. Norwood, MA: Artech House, 1990, vol. 1, pp. 247–295.

[32] C.-Y. Chong and S. P. Kumar, “Sensor networks: Evolution, opportunities, andchallenges,” Proceedings of the IEEE, vol. 91, no. 8, pp. 1247–1256, August2003.

[33] A. Colvin, “CSMA with collision avoidance,” Computer Communications,vol. 6, no. 5, pp. 227–235, October 1983.

[34] Crossbow Technology, Inc., “MICAz wireless measurement system,” Avail-able: xbow.com/Products/Product_pdf_files/Wireless_pdf/6020-0060-01_A_MICAz.pdf, 2004.

[35] ——, “Stargate X-Scale processor platform,” Available: http://www.xbow.com/Products/Product_pdf_files/Wireless_pdf/6020-0049-01_B_STARGATE.pdf, 2004.

[36] ——, “2007 wireless sensor networks product reference guide,” Available:http://www.xbow.com/Products/Product_pdf_files/Wireless_pdf/Wireless_Catalog_2006.zip, 2007.

[37] ——, “MICA2 wireless measurement system,” Available: http://www.xbow.com/Products/Product_pdf_files/Wireless_pdf/MICA2_Datasheet.pdf, 2007.

[38] ——, “MICA2DOT wireless microsensor mote,” Available: www.xbow.com/Products/Product_pdf_files/Wireless_pdf/MICA2DOT_Datasheet.pdf, 2007.

[39] K. Crowley, J. Frisby, S. Murphy, M. Roantree, and D. Diamond, “Web-basedreal-time temperature monitoring of shellfish catches using a wireless sensornetwork,” Sensors and Actuators A: Physical, vol. 122, no. 2, pp. 222–230,August 2005.

[40] D. Culler, D. Estrin, and M. Srivastava, “Guest editors’ introduction:Overview of sensor networks,” Computer, vol. 37, no. 8, pp. 41–49, August2004.

Bibliography 113

[41] C. Cypress Semiconductor, “PSoC mixed-signal array final data sheet,” Avail-able: http://download.cypress.com.edgesuite.net/design_resources/datasheets/contents/cy8c29666_8.pdf, 2007.

[42] Dallas Semiconductor, Inc., “DS620 low-voltage ±0.5 c accuracy digitalthermometer and thermostat,” Available: http://datasheets.maxim-ic.com/en/ds/DS620.pdf, 2008.

[43] E. J. Davison and N. K. Tripathi, “The optimal decentralized control of a largepower system: Load and frequency control,” IEEE Trans. Automatic Control,vol. 23, no. 2, pp. 312–325, April 1978.

[44] J. Deng, Y. S. Han, P.-N. Chen, and P. K. Varshney, “Optimal transmissionrange for wireless ad hoc networks based on energy efficiency,” IEEE Trans-actions on Communications, vol. 55, no. 9, pp. 1772–1782, September 2007.

[45] A. Deshpande, C. Guestrin, and S. Madden, “Resource-aware wireless sensor-actuator networks,” Data Engineering Bulletin, vol. 28, no. 1, pp. 40–47,March 2005.

[46] L. Doherty, B. Warneke, B. Boser, and K. Pister, “Energy and performanceconsiderations for smart dust,” Int’l Journal of Parallel and Distributed Sys-tems and Networks, vol. 4, no. 3, pp. 121–133, December 2001.

[47] H. Dubois-Ferriere, R. Meier, L. Fabre, and P. Metrailler, “Tinynode: Acomprehensive platform for wireless sensor network applications,” in Proc.5th International Symposium on Information Processing in Sensor Networks(IPSN’06), Nashville, TN, USA, April 2006, pp. 358–365.

[48] Dust Networks, Inc., “SmartMesh-XR evaulation kit,” Available: http://www.dustnetworks.com/docs/Evaluation_Kit.pdf, 2005.

[49] Dynastream Innovations, Inc., “ANT message protocol and usage,” Avail-able: http://www.nordicsemi.com/files/Product/data_sheet/User_guide_ANT.pdf, 2007.

[50] ——, “User manual development kit ANTDKT3,” Available: http://www.thisisant.com/, 2008.

[51] A. El-Hoiydi, J.-D. Decotignie, and J. Hernandez, “Low power MAC protocolsfor infrastructure wireless sensor networks,” in Proc. 5th European WirelessConf. (EW’04), Barcelona, Spain, February 2004.

114 Bibliography

[52] Energizer Holdings, Inc., “Energizer e91 product data sheet,” Available: http://data.energizer.com/PDFs/E91.pdf, 2008.

[53] C. Enz, N. Scolari, and U. Yodprasit, “Ultra low-power radio design for wire-less sensor networks,” in Proc. IEEE Int. Workshop on Radio-Frequency Inte-gration Technology (RFIT’05), Singapore, November 30 - December 02 2005,pp. 1–17.

[54] C. C. Enz, A. El-Hoiydi, J.-D. Decotignie, and V. Peiris, “WiseNET: Anultralow-power wireless sensor network solution,” Computer, vol. 37, no. 8,pp. 62–70, August 2004.

[55] C. Evans-Pughe, “Bzzzz zzz [zigbee wireless standard],” IEE Review, vol. 49,no. 3, pp. 28–31, March 2003.

[56] Fastrax Oy, “iTrax MP family - OEM GPS receiver modules,” Available: http://www.fastraxgps.com/, 2007.

[57] D. Flowers and Y. Yang, “AN1066 - MiWi wireless networking proto-col stack,” Available: http://ww1.microchip.com/downloads/en/AppNotes/01066a.pdf, 2007.

[58] T. C. Forum, “An open-source OS for the networked sensor regime,” Available:http://www.tinyos.net, 2004.

[59] Freescale Semiconductor, Inc., “Data sheet M68HC08 microcontrollers,”Available: http://www.freescale.com/files/microcontrollers/doc/data_sheet/MC68HC908AP64.pdf, 2005.

[60] Fuji & CO. (Piezo Science), “MS300-series - PIR module with lens,” Avail-able: http://www.fuji-piezo.com/ms300.htm, 2006.

[61] Gang Lu, B. Krishnamachari, and C. S. Raghavendra, “An adaptive energy-efficient and low-latency MAC for data gathering in wireless sensor networks,”in Proc. Parallel and Distributed Processing Symposium (IPDPS’04), April2004, pp. 224–231.

[62] Q. Gaoa, K. Blowa, D. Holdinga, I. Marshallb, and X. Penga, “Radio rangeadjustment for energy efficient wireless sensor networks,” Ad Hoc Networks,vol. 4, no. 1, pp. 75–82, January 2006.

Bibliography 115

[63] C. Guo, L. Zhong, and J. Rabaey, “Low power distributed MAC for ad hoc sen-sor radio networks,” in Global Telecommunications Conf. (GLOBECOM’01),vol. 5, San Antonio, TX, USA, November 2001, pp. 2944–2948.

[64] J. A. Gutierrez, E. H. C. Jr, and R. L. Jr, Low-Rate Wireless Personal AreaNetworks: Enabling Wireless Sensors with IEEE 802.15.4. IEEE Press, 2004,vol. 1.

[65] M. Haenggi, Handbook of Sensor Networks: Compact Wireless and WiredSensing Systems, 1st ed., M. Ilyas and I. Mahgoub, Eds. CRC Press, Florida,2004.

[66] J. K. Hart and K. Martinez, “Environmental sensor networks: A revolutionin the earth system science?” Earth-Science Reviews, vol. 78, no. 3-4, pp.177–191, October 2006.

[67] HART Communication Foundation, “Main page,” Available: http://www.hartcomm2.org/index.html, 2008.

[68] D. Hawbacker and T. Rappaport, “Indoor wideband radiowave propagationmeasurements at 1.3 GHz and 4.0 GHz,” Electronics Letters, vol. 26, no. 21,pp. 1800–1802, October 1990.

[69] T. He, S. Krishnamurthy, L. Luo, T. Yan, L. Gu, R. Stoleru, G. Zhou, Q. Cao,P. Vicaire, J. A. Stankovic, T. F. Abdelzaher, J. Hui, and B. Krogh, “VigilNet:An integrated sensor network system for energy-efficient surveillance,” ACMTransactions on Sensor Networks, vol. 2, no. 1, pp. 1–38, February 2006.

[70] B. Heile, “Wireless sensors and control networks: Enabling new opportunitieswith ZigBee,” Available: http://www.zigbee.org, 2006.

[71] W. B. Heinzelman, A. P. Chandrakasan, and H. Balakrishnan, “An application-specific protocol architecture for wireless microsensor networks,” IEEE Wire-less Communications, vol. 1, no. 4, pp. 660–670, October 2002.

[72] J. Hill, M. Horton, R. Kling, and L. Krishnamurthy, “Wireless sensor net-works: The platforms enabling wireless sensor networks,” Communications ofthe ACM, vol. 6, no. 47, pp. 41–46, June 2004.

[73] J. Hill and D. Culler, “Mica: a wireless platform for deeply embedded net-works,” IEEE Micro, vol. 22, no. 6, pp. 12–24, November/December 2002.

116 Bibliography

[74] Hitachi Metals, Ltd., “Magnetic compass sensor HM55B data sheet,” Avail-able: http://www.hitachimetals.com/product/sensors/mems/files/h_dst_e.pdf,2008.

[75] P. Hämäläinen, “Cryptographic security designs and hardware architecturesfor wireless local area networks,” Tampere University of Technology, PhDThesis, Tampere, Finland, December 2006.

[76] M. Hännikäinen, “Design of quality of service support for wireless local areanetworks,” Tampere University of Technology, PhD Thesis, Tampere, Finland,November 2002.

[77] IEEE, “IEEE p1451.5 project,” Available: http://grouper.ieee.org/groups/1451/5/, 2006.

[78] ——, “IEEE xplore release 2.4,” Available: http://ieeexplore.ieee.org/, 2007.

[79] IEEE Std 802.11-1997 Information Technology- telecommunications And In-formation exchange Between Systems-Local And Metropolitan Area Networks-specific Requirements-part 11: Wireless Lan Medium Access Control (MAC)And Physical Layer (PHY) Specifications, IEEE 802.11, 1997.

[80] IEEE Standard for Information Technology—Telecommunications and Infor-mation Exchange Between Systems—Local and Metropolitan Area Networks—Specific Requirements—Part 15.1: Wireless Medium Access Control (MAC)and Physical Layer (PHY) Specifications for Wireless Personal Area Networks(WPANs), IEEE 802.15.1, 2002.

[81] IEEE Standard for Information Technology—Telecommunications and Infor-mation Exchange Between Systems—Local and Metropolitan Area Networks—Specific Requirements—Part 15.3: Wireless Medium Access Control (MAC)and Physical Layer (PHY) Specifications for High Rate Wireless Personal AreaNetworks (WPANs), IEEE 802.15.3, 2003.

[82] IEEE Standard for Information Technology—Telecommunications and Infor-mation Exchange Between Systems—Local and Metropolitan Area Networks—Specific Requirements—Part 15.4: Wireless Medium Access Control (MAC)and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal AreaNetworks (LR-WPAN), IEEE 802.15.4, 2003.

[83] ISA, “ISA100.11a release 1,” Available: http://www.isa.org/source/ISA100.11a_Release1_Status.ppt, 2007.

Bibliography 117

[84] S. Jiang, J. Rao, D. He, X. Ling, and C. C. Ko, “A simple distributed PRMAfor MANETs,” IEEE Transactions on Vehicular Technology, vol. 51, no. 2, pp.293–305, March 2002.

[85] K. Jin and D. Cho, “Multi-code MAC for multi-hop wireless ad hoc networks,”in Proc. 56th Vehicular Technology Conference (VTC’02 (fall)), vol. 2, Van-couver, British Columbia, Canada, September 24-29 2002, pp. 1100–1104.

[86] A. Juels, “RFID security and privacy: a research survey,” IEEE Journal onSelected Areas in Communications, vol. 24, no. 2, pp. 381 – 394, February2006.

[87] J. K. Juntunen, M. Kuorilehto, M. Kohvakka, V. A. Kaseva, M. Hännikäinen,and T. D. Hämäläinen, “WSN API: Application programming interface forwireless sensor networks,” in Proc. The 17th Annual IEEE Int’l Symposium onPersonal, Indoor and Mobile Radio Communications (PIMRC’06), Helsinki,Finland, September 11–14 2006.

[88] H. Karl and A. Willig, Protocols and Architectures for Wireless Sensor Net-works. John Wiley & Sons Ltd, 2005.

[89] P. Karn, “MACA - a new channel access method for packet radio,” in Proc.ARRL/CRRL Amateur Radio 9th Computer Networking Conf., 1990, pp. 134–140.

[90] V. Kaseva, M. Kohvakka, M. Kuorilehto, M. Hännikäinen, and T. D.Hämäläinen, “A wireless sensor network for RF-based indoor localiza-tion,” EURASIP Journal on Advances in Signal Processing, vol. 2008, no.doi:10.1155/2008/731835, pp. 1–27, 2008.

[91] L. Kleinrock and F. Tobagi, “Packet switching in radio channels: part i-thecarrier sense multiple access modes and their throughput-delay characteris-tics,” IEEE Transactions on Communications, vol. COM-23, no. 12, pp. 1400– 1416, December 1975.

[92] J. Kong, J.-H. Cui, D. Wu, and M. Gerla, “Building underwater ad-hoc net-works and sensor networks for large scale real-time aquatic applications,” inProc. IEEE Military Communications Conference (MILCOM’05), vol. 3, At-lantic City, NJ, USA, October 17–20 2005, pp. 1535–1541.

[93] M. Kuorilehto, “System level design issues in low-power wireless sensor net-works,” Tampere University of Technology, PhD Thesis, Tampere, Finland,June 2008.

118 Bibliography

[94] M. Kuorilehto, M. Hännikäinen, and T. D. Hämäläinen, “A survey of applica-tion distribution in wireless sensor networks,” EURASIP Journal on WirelessCommunications and Networking, Special Issue on Ad Hoc Networks: Cross-Layer Issues, vol. 2005, no. 5, pp. 774–788, December 2005.

[95] M. Kuorilehto, J. Suhonen, M. Hännikäinen, and T. D. Hämäläinen, “Tool-aided design and implementation of indoor surveillance wireless sensor net-work,” in Lecture Notes in Computer Science, S. Vassiliadis, S. Wong, andT. D. Hämäläinen, Eds. Springer, 2007, vol. 4599, pp. 396–407.

[96] J. Kuruvila, A. Nayak, and I. Stojmenovic, “Hop count optimal position-basedpacket routing algorithms for ad hoc wireless networks with a realistic physicallayer,” IEEE Journal of Selected Areas in Communications, vol. 23, no. 6, pp.1267–1275, June 2005.

[97] Lantronix, Inc., “XPort data sheet,” Available: http://www.lantronix.com/pdf/XPort_DS.pdf, 2005.

[98] D. Ley, “Emerging technologies for learning: Ubiquitous computing,” Avail-able: http://www.becta.org.uk/research, British Educational Communicationsand Technology Agency Becta, Tech. Rep. volume 2 (2007), 2007.

[99] C. Lin and M. Gerla, “Adaptive clustering for mobile wireless networks,” IEEEJournal of Selected Areas in Communications, vol. 15, no. 7, pp. 1265–1275,September 1997.

[100] M. J. Markowski and A. Sethi, “Fully distributed wireless MAC transmissionof real-time data,” in Proc. 4th IEEE Real-Time Technology and ApplicationsSymposium (RTAS’98), Denver, Colorado, USA, June 3-5 1998.

[101] K. Martinez, P. Padhy, A. Riddoch, R. Ong, and J. Hart, “Glacial environmentmonitoring using sensor networks,” in Proc. Workshop on Real-World WirelessSensor Networks (REALWSN’05), Stockholm, Sweden, June 20–21 2005.

[102] Maxim Integrated Products, “MAX1725/MAX1726 12V, Ultra-low-IQ, low-dropout linear regulators,” Available: http://datasheets.maxim-ic.com/en/ds/MAX1725-MAX1726.pdf, 2003.

[103] ——, “MAX8880/MAX8881 12V, Ultra-low-IQ, low-dropout linearregulators with pok,” Available: http://datasheets.maxim-ic.com/en/ds/MAX8880-MAX8881.pdf, 2007.

Bibliography 119

[104] Microchip Technology, Inc., “PIC18F2420/2520/4420/4520 data sheet,”Available: http://ww1.microchip.com/downloads/en/DeviceDoc/39631a.pdf,2004.

[105] ——, “PIC18F2525/2620/4525/4620 data sheet,” Available: http://ww1.microchip.com/downloads/en/DeviceDoc/39626b.pdf, 2004.

[106] ——, “PIC18F8722 family data sheet,” Available: http://ww1.microchip.com/downloads/en/DeviceDoc/39646b.pdf, 2004.

[107] ——, “MRF24J40 data sheet,” Available: http://ww1.microchip.com/downloads/en/DeviceDoc/39776a.pdf, 2006.

[108] ——, “PIC24FJ128GA family data sheet,” Available: http://ww1.microchip.com/downloads/en/DeviceDoc/39747C.pdf, 2006.

[109] ——, “MCP1603 2.0 MHz, 500 mA synchronous buck regulator,” Available:http://ww1.microchip.com/downloads/en/DeviceDoc/22042a.pdf, 2007.

[110] ——, “MCP1702 250 mA low quiescent current LDO regulator,” Available:http://ww1.microchip.com/downloads/en/DeviceDoc/22008b.pdf, 2007.

[111] A. Mihovska, F. Platbrood, G. Karetsos, S. Kyriazakos, R. Muijen,R. Guarneri, and J. Pereira, “Towards the wireless 2010 vision: A technologyroadmap,” Wireless Personal Communications, vol. 42, no. 3, pp. 303–336,August 2007.

[112] R. Min, M. Bhardwaj, S.-H. Cho, N. Ickes, E. Shih, A. Sinha, A. Wang, andA. Chandrakasan, “Energy-centric enabling technologies for wireless sensornetworks,” IEEE Wireless Communications, vol. 9, no. 4, pp. 28–39, August2002.

[113] Minilogic Device, Corp., “ML62 series positive voltage regulator,” Available:http://www.minilogic.com.hk/specification/ML62.pdf, 2005.

[114] Moteiv, Corp., “Tmote sky ultra low power IEEE 802.15.4 compliant wirelesssensor module,” Available: http://www.eecs.harvard.edu/~konrad/projects/shimmer/references/tmote-sky-datasheet.pdf, 2006.

[115] S. Mozar, “An evaluation of low cost power supply alternatives for high vol-ume consumer products,” in Proc. 10th Int. Symposium on Consumer Elec-tronics (ISCE’06), St. Petersburg, Russia, June 28- July 01 2006, pp. 1–4.

120 Bibliography

[116] Nanotron Technologies GmbH, “nanoNET development kit,” Available: http://www.nanotron.com/EN/pdf/Factsheet_nanoNET-Dev-Kit.pdf, 2008.

[117] Y. Neuvo, “Future wireless technologies,” in Proc. IEEE 6th CAS Symposiumon Emerging Technologies: Frontiers of Mobile and Wireless Communication,vol. I, Shanghai, China, May 31 - June 2 2004, pp. 1–3.

[118] D. Niculescu, “Communication paradigms for sensor networks,” IEEE Com-munications Magazine, vol. 43, no. 3, pp. 116–122, March 2005.

[119] Nordic Semiconductor ASA, “nRF24E1 product specification,” Available:http://www.nordicsemi.com/files/Product/data_sheet/Product_Specification_nRF24E1_1_3.pdf, 2006.

[120] ——, “Single chip 2.4 GHz transceiver nRF2401A,” Available:http://www.nordicsemi.no/files/Product/data_sheet/Product_Specification_nRF2401A_1_1.pdf, 2006.

[121] ——, “Single chip 2.4 GHz Transceiver nRF24L01,” Available:http://www.nordicsemi.no/files/Product/data_sheet/Product_Specification_nRF24L01_1.0.pdf, 2006.

[122] ——, “Single chip 433/868/915 MHz transceiver nRF905,” Available: http://www.nordicsemi.no/files/Product/data_sheet/nRF905_rev1_4.pdf, 2006.

[123] S. Oh, P. Chen, M. Manzo, and S. Sastry, “Instrumenting wireless sensor net-works for real-time surveillance,” in Proc. 2006 IEEE Int’l Conf. on Roboticsand Automation (ICRA’06), Orlando, FL, USA, May 15–19 2006, pp. 3128–3133.

[124] N. Ota and P. Wright, “Trends in wireless sensor networks for manufactur-ing,” International Journal of Manufacturing Research, vol. 1, no. 1, pp. 3–17,2006.

[125] G. Pei and C. Chien, “Low power TDMA in large wireless sensor networks,”in Proc. IEEE Military Communications Conf. (MILCOM’01), October 2001,pp. 347–351.

[126] C. Perkins, Ad Hoc Networks. Addison-Wesley, Reading, MA, 2000, vol. 1.

[127] D. Pescovitz, “The power of small tech,” Smalltimes, vol. 2, no. 1, January2002.

Bibliography 121

[128] C. Piguet, J.-M. Masgonty, C. Arm, S. Durand, T. Schneider, F. Rampogna,C. Scarnera, C. Iseli, J.-P. Bardyn, R. Pache, and E. Dijkstra, “Low-power de-sign of 8-b embedded coolrisc microcontroller cores,” IEEE Journal of Solid-State Circuits, vol. 32, no. 7, pp. 1067–1078, July 1997.

[129] J. Polastre, J. Hill, and D. Culler, “Versatile low power media access for wire-less sensor networks,” in Proc. 2nd Internation Conf. on Embedded NetworkedSensor Systems (Sensys’04), Baltimore, MD, USA, November 2004, pp. 95–107.

[130] J. Polastre, R. Szewczyk, and D. Culler, “Telos: enabling ultra-low powerwireless research,” in Proc. 4th International Symposium on Information Pro-cessing in Sensor Networks (IPSN’05), Los Angeles, CA, USA, April 2005,pp. 364–369.

[131] I. Poole, “What exactly is...zigbee,” IEE Review, vol. 2, no. 4, pp. 44–45,August-September 2004.

[132] G. Pottie and W. Kaiser, “Wireless integrated network sensors,” Communica-tions of the ACM, vol. 43, no. 5, pp. 51–58, May 2000.

[133] J. Rabaey, J. Ammer, J. da Silva Jr., D. Patel, and S. Roundy, “Picoradio sup-ports ad hoc ultra-low power wireless networking,” IEEE Computer Magazine,vol. 33, no. 7, pp. 42–48, July 2000.

[134] V. Raghunathan, C. Schurgers, S. Park, and M. Srivastava, “Energy-awarewireless microsensor networks,” IEEE Signal Processing Magazine, vol. 19,no. 2, pp. 40–50, March 2002.

[135] V. Rajendran, K. Obraczka, and J. Garcia-Luna-Aceves, “Energy-efficient,collision-free medium access control for wireless sensor networks,” in Proc.1st Int’l Conf. on Embedded Networked Sensor Systems (SenSys’03), Los An-geles, CA, USA, November 2003, pp. 181–192.

[136] T. Rappaport, “Indoor radio communications for factories of the future,” IEEECommunications Magazine, vol. 27, no. 5, pp. 15–24, May 1989.

[137] ——, Wireless Communications - Principles and Practice, 2nd ed. PrenticeHall, 1996, ch. 1.

[138] T. Rappaport and C. McGillem, “Uhf fading in factories,” IEEE Journal onSelected Areas in Communication, vol. 7, no. 1, pp. 40–48, January 1989.

122 Bibliography

[139] J. M. Reason and J. M. Rabaey, “A study of energy consumption and reliabilityin a multi-hop sensor network,” ACM SIGMOBILE Mobile Computing andCommunications Review, vol. 8, no. 1, pp. 84–97, January 2004.

[140] N. Reijers, G. Halkes, and K. Langendoen, “Link layer measurements in sen-sor networks,” in Proc. 1st IEEE Int’l Conf. on Mobile Ad-hoc and SensorSystems (MASS’04), FL, USA, October 2004, pp. 224–234.

[141] RF Monolithics, Inc., “TR1001 868.35 MHz hybrid transceiver,” Available:http://www.rfm.com/products/data/tr1001.pdf, 1999.

[142] ——, “TR3100 433.92 MHz hybrid transceiver,” Available: http://www.rfm.com/products/data/tr3100.pdf, 1999.

[143] I. Rhee, A. Warrier, M. Aia, and J. Min, “Z-MAC: A hybrid MAC for wirelesssensor networks,” in Proc. 3rd ACM Conf. on Embedded Networked SensorSystems (Sensys ’05), New York, NY, USA, November 2005, pp. 90–101.

[144] O. Riva and C. Borcea, “The urbanet revolution: Sensor power to the people!”IEEE Pervasive Computing, vol. 6, no. 2, pp. 41–49, April-June 2007.

[145] K. Römer, “Tracking real-world phenomena with smart dust,” in Proc. 1stEuropean Workshop on Wireless Sensor Networks (EWSN’04), ser. LNCS, no.2920. Berlin, Germany: Springer-Verlag, January 19–21 2004, pp. 28–43.

[146] K. Römer and F. Mattern, “The design space of wireless sensor networks,”IEEE Wireless Communications, vol. 11, no. 6, pp. 54–61, December 2004.

[147] L. Roberts, “ALOHA packet system with and without slots and capture,” ACMSIGCOMM Computer Communication Review, vol. 5, no. 2, pp. 28–42, April1975.

[148] Round Solutions GmbH & Co KG, “CAM-VGA100 user manual,” Avail-able: http://www.roundsolutions.com/techdocs/docs/CAM-VGA100-User\%20Manual-v2-2.pdf, 2006.

[149] S. Roundy, D. Steingart, L. Frechette, P. Wright, and J. Rabaey, “Powersources for wireless sensor networks,” in Proc. 1st Eur. Workshop on Wire-less Sensor Network (EWSN’04), Berlin, Germany, January 2004, pp. 1–17.

[150] S. Roundy, P. K. Wright, and J. Rabaey, “A study of low level vibrations as apower source for wireless sensor nodes,” Computer Communications, vol. 26,no. 11, pp. 1131–1144, July 2003.

Bibliography 123

[151] R. Sachs, “Z-Wave, the standard in wireless home control,” Available: http://www.zen-sys.com/, 2006.

[152] B. M. Sadler, “Fundamentals of energy-constrained sensor network systems,”IEEE A&E Systems Magazine, vol. 20, no. 1, pp. 17–35, August 2005.

[153] A. Sakata, T. Yamazato, H. Okada, and M. Katayamat, “Throughput compar-ison of CSMA and CDMA slotted ALOHA in inter-vehicle communication,”in Proc. 7th International Conference on ITS Telecommunications (ITST’07),Sophia Antipolis, France, June 6–8 2007, pp. 1–6.

[154] Y. Sankarasubramaniam, I. Akyildiz, and S. McLaughlin, “Energy efficiencybased packet size optimization in wireless sensor networks,” in Proc. 1st IEEEInt. Workshop on Sensor Network Platforms and Applications (SNPA’03), An-chorage, Alaska, May 11 2003, pp. 1–8.

[155] N. Sastry and D. Wagner, “Security considerations for IEEE 802.15.4 net-works,” in Proc. 2004 ACM Workshop on Wireless Security (WiSE’04),Philadelphia, PA, USA, October 1 2004, pp. 32–42.

[156] A. Savvides and M. B. Srivastava, “A distributed computation platform forwireless embedded sensing,” in Proc. IEEE Int’l Conf. on Computer Design:VLSI in Computers and Processors (ICCD’02), Freiburg, Germany, September16–18 2002, pp. 220–225.

[157] J. Schiller, A. Liers, and H. Ritter, “ScatterWeb: A wireless sensornet plat-form for research and teaching,” Elsevier Computer Communications, vol. 28,no. 13, pp. 1545–1551, August 2005.

[158] T. Schmid, H. Dubois-Ferriere, and M. Vetterli, “Sensorscope: Experienceswith a wireless building monitoring sensor network,” in Proc. Workshop onReal-World Wireless Sensor Networks (REALWSN’05), Stockholm, Sweden,June 20–21 2005.

[159] C. Schurgers, V. Tsiatsis, S. Ganeriwal, and M. Srivastava, “Optimizing sensornetworks in the energy-latency-density design space,” IEEE Transactions onMobile Computing, vol. 1, no. 1, pp. 70–80, January-March 2002.

[160] L. Schwiebert, S. Gupta, and J. Weinmann, “Research challenges in wire-less networks of biomedical sensors,” in Proc. 7th annual ACM/IEEE Inter-national Conference on Mobile Computing and Networking (MobiCom’01),Rome, Italy, July 16–21 2001, pp. 151–165.

124 Bibliography

[161] S. Seidel and T. Rappaport, “914 MHz path loss prediction models for indoorwireless communications in multifloored buildings,” IEEE Trans. Antennasand Propagation, vol. 40, no. 2, pp. 207–217, February 1992.

[162] Seiko Instruments, Inc., “S-1206 series ultra low current consumption and lowdropout CMOS voltage regulator,” Available: http://dl.sii-ic.com.edgesuite.net/spd_dtst/dt_sht_e/vol_regu/S1206_E.pdf, 2008.

[163] Semtech, Corp., “XE1201A 300-500 MHz low-power UHF transceiver,”Available: http://www.semtech.com/, 2005.

[164] ——, “XE8802 sensing machine,” Available: http://www.semtech.com/,2006.

[165] ——, “XE1203F 433 MHz/868 MHz/915 MHz low-power, integrated UHFtransceiver,” Available: http://www.semtech.com/, 2007.

[166] Sensicast, Inc., “SensiNet - the wireless sensor network from sensicast,” Avail-able: http://www.sensicast.com/, 2008.

[167] Sensorion AG, “SHT1x/SHT7x humidity & temperature sensor,” Available:http://www.sensirion.com/images/getFile?id=25, 2004.

[168] C. Sharp, S. Schaffert, A. Woo, N. Sastry, C. Karlof, S. Sastry, and D. Culler,“Design and implementation of a sensor network system for vehicle trackingand autonomous interception,” in Proc. 2nd European Workshop on WirelessSensor Networks (EWSN’05), Istanbul, Turkey, January 31–February 2 2005,pp. 93–107.

[169] M. Sichitiu, V. Ramadurai, and P. Peddabachagari, “Simple algorithm for out-door localization of wireless sensor networks with in-accurate range measure-ments,” in Proc. Int’l Conf. on Wireless Networks (ICWN’03), Las Vegas,Nevada, USA, June 23-26 2003, pp. 300–305.

[170] A. Sikora, “Design challenges for short-range wireless networks,” IEE Proc.Communications, vol. 151, no. 5, pp. 473–479, October 2002.

[171] M. Sikora, J. Laneman, M. Haenggi, J. D.J. Costello, and T. Fuja, “On the op-timum number of hops in linear wireless networks,” in Proc. IEEE InformationTheory Workshop (ITW’04), Texas, USA, October 24-29 2004, pp. 165–169.

[172] G. Simon, , M. Maróti, Ákos Lédeczi, G. Balogh, B. Kusy, A. Nádas, G. Pap,J. Sallai, and K. Frampton, “Sensor network-based countersniper system,” in

Bibliography 125

Proc. 2nd Int’l Conf. on Embedded Networked Sensor Systems (SenSys’04),Baltimore, MD, USA, November 3–5 2004, pp. 1–12.

[173] S. Singh and C. S. Raghavendra, “PAMAS: Power-aware multi-access pro-tocol with signaling for ad hoc networks,” ACM Computer CommunicationReview, vol. 28, no. 3, pp. 5–26, July 1998.

[174] Smart-its, “Smart-its home page,” Available: http://www.smart-its.org/, 2007.

[175] K. Sohrabi, J. Gao, V. Ailawadhi, and G. J. Pottie, “Protocols for self-organization of a wireless sensor network,” IEEE Personal Communications,vol. 7, no. 5, pp. 16–27, October 2000.

[176] J. Song, S. Han, A. Mok, D. Chen, M. Lucas, and M. Nixon, “WirelessHART:Applying wireless technology in real-time industrial process control,” in Proc.14th IEEE Real-Time and Embedded Technology and Applications Symposium(RTAS’08), St. Louis, MO, USA, April 22-24 2008, pp. 377–386.

[177] W. Stallings, Data and Computer Communications, 7th ed. Prentice-Hall,2004.

[178] J. A. Stankovic, T. F. Abdelzaher, C. Lu, L. Sha, and J. C. Hou, “Real-timecommunication and coordination in embedded sensor networks,” Proceedingsof the IEEE, vol. 91, no. 7, pp. 1002–1022, July 2003.

[179] D. C. Steere, A. Baptista, D. McNamee, C. Pu, and J. Walpole, “Researchchallenges in environmental observation and forecasting systems,” in Proc. 6thAnnual ACM/IEEE Int’l Conf. on Mobile Computing and Networking (Mobi-Com’00), Boston, MA, USA, August 6–11 2000, pp. 292–299.

[180] M. Stordeur and I. Stark, “Low power thermoelectric generator - self-sufficientenergy supply for micro systems,” in Proc. 16th Int. Conf. Thermoelectrics(ICT’97), Dresden, Germany, August 26–29 1997, pp. 575–577.

[181] J. Suhonen, M. Kohvakka, M. Hännikäinen, and T. Hämäläinen, “Design, im-plementation, and outdoor deployment for wireless sensor network for envi-ronmental monitoring,” in Proc. Embedded Computer Systems: Architectures,Modeling, and Simulation (SAMOS VI), Samos, Greece, July 2006, pp. 109–121.

[182] ——, “Embedded software architecture for diagnosing network and node fail-ures in wireless sensor networks,” in Proc. Embedded Computer Systems: Ar-

126 Bibliography

chitectures, Modeling, and Simulation (SAMOS VIII), Samos, Greece, July2008, pp. 1–10.

[183] J. Suhonen, M. Kohvakka, M. Kuorilehto, M. Hännikäinen, and T. D. Hä-mäläinen, “Cost-aware capacity optimization in dynamic multi-hop wsns,” inProc. Design, Automation and Test in Europe (DATE’07), Nice, France, April2007, pp. 666–671.

[184] J. Suhonen, M. Kuorilehto, M. Hännikäinen, and T. D. Hämäläinen, “Cost-aware dynamic routing protocol for wireless sensor networks - design and pro-totype experiments,” in Proc. 17th Annual IEEE Int’l Symposium on Personal,Indoor and Mobile Radio Communications (PIMRC’06), Helsinki, Finland,September 2006, pp. 1–5.

[185] R. Szewczyk, A. Mainwaring, J. Polastre, J. Anderson, and D. Culler, “Ananalysis of a large scale habitat monitoring application,” in Proc. 2nd Int’lConf. on Embedded Networked Sensor Systems (SenSys’04), Baltimore, MD,USA, November 3–5 2004, pp. 214–226.

[186] Z. Tang and J. Garcia-Luna-Aceves, “A protocol for topology-dependent trans-mission scheduling in wireless networks,” in Proc. IEEE Wireless Communi-cations and Networking Conference (WCNC’99), vol. 3, New Orleans, LA,USA, September 1999, pp. 1333–1337.

[187] Texas Instruments, Inc., “CC1020 single chip low power RF transceiver fornarrowband systems,” Available: http://www.ti.com/lit/gpn/cc1020, 2006.

[188] ——, “CC1100 single chip low cost low power RF transceiver,” Available:http://www.ti.com/lit/gpn/cc1100, 2006.

[189] ——, “CC2500 single chip low cost low power RF transceiver,” Available:http://www.ti.com/lit/gpn/cc2500, 2006.

[190] ——, “MSP430x15x, MSP430x16x, MSP430x161x mixed signal microcon-troller,” Available: http://focus.ti.com/lit/ds/symlink/msp430f169.pdf, 2006.

[191] ——, “TPS62000 high-efficiency step-down low power DC-DC converter,”Available: http://focus.ti.com/lit/ds/symlink/tps62000.pdf, 2006.

[192] ——, “CC1000 single chip very low power RF transceiver,” Available: http://www.ti.com/lit/gpn/cc1000, 2007.

Bibliography 127

[193] ——, “CC2400 2.4 GHz low-power RF transceiver,” Available: http://www.ti.com/lit/gpn/cc2400, 2007.

[194] ——, “CC2420 2.4 GHz IEEE 802.15.4 / ZigBee-ready RF transceiver,”Available: http://www.ti.com/lit/gpn/cc2420, 2007.

[195] ——, “TPS715x 50 mA, 24V, 3.2 µA supply current low-dropout linearregulator in SC70 package,” Available: http://focus.ti.com/lit/ds/symlink/tps71530.pdf, 2007.

[196] J.-P. Thomesse, “Fieldbus technology in industrial automation,” Proceedingsof the IEEE, vol. 93, no. 6, pp. 1073–1101, June 2005.

[197] P. Thubert, “6LoWPAN backbone router, internet-draft,” Available: http://tools.ietf.org/html/draft-thubert-6lowpan-backbone-router-01, 2008.

[198] F. A. Tobagi and L. Kleinrock, “Packet switching in radio channels: Part II- the hidden terminal problem in carrier sense multiple-access modes and thebusy-tone solution,” IEEE Trans. Communications, vol. 23, no. 12, pp. 1417–1433, December 1975.

[199] G. Tolle, J. Polastre, R. Szewczyk, D. Culler, N. Turner, K. Tu, S. Burgess,T. Dawson, P. Buonadonna, D. Gay, and W. Hong, “A macroscope in the red-woods,” in Proc. 3rd Int’l Conf. on Embedded Networked Sensor Systems (Sen-Sys’05), San Diego, CA, USA, November 2–4 2005, pp. 51–63.

[200] V. Turau, M. Witt, and C. Weyer, “Analysis of a real multi-hop sensor networkdeployment: The heathland experiment,” in Proc. 3rd Int’l Conf. on NetworkedSensing Systems (INSS’06), Chicago, IL, USA, May 31–June 2 2006.

[201] T. van Dam and K. Langendoen, “An adaptive energy-efficient MAC protocolfor wireless sensor networks,” in Proc. 1st Int’l Conf. on Embedded NetworkedSensor Systems (Sensys’03), Los Angeles, CA, USA, November 2003, pp.171–180.

[202] E. Verriest, B. Friedlander, and M. Morf, “Distributed processing in estima-tion and detection,” in Proc. 18th IEEE Conference on Decision and Controlincluding the Symposium on Adaptive Processes (CDC’79), Ft. Lauderdale,FL, USA, December 12–14 1979, pp. part 1: 153–158.

[203] N. Vlajic and D. Xia, “Wireless sensor networks: to cluster or not to cluster?”in Proc. International Symposium on World of Wireless, Mobile and Multi-

128 Bibliography

media Networks (WoWMoM ’06), Buffalo, NY, USA, June 26–29 2006, pp.11–9.

[204] VTI Technologies Oy, “SCA3000-E01 3-axis ultra low power accelerometerwith digital spi interface,” Available: http://www.vti.fi, 2002.

[205] ——, “SCP1000 series (120 kPa) absolute pressure sensor,” Available: http://www.vti.fi, 2007.

[206] N. Wang, N. Zhang, and M. Wang, “Wireless sensors in agriculture and foodindustry - recent development and future perspective,” Computers and Elec-tronics in Agriculture, vol. 50, no. 1, pp. 1–14, January 2006.

[207] B. Warneke, M. Last, B. Leibowitz, , and K. S. J. Pister, “Smart dust: Com-municating with a cubic-millimeter computer,” Computer, vol. 34, no. 1, pp.43–51, January 2001.

[208] B. Week, “21 ideas for the 21st century,” Business Week, August 30th, 1999,pp. 78 - 167, 1999.

[209] M. Weiser, “Hot topics: ubiquitous computing,” IEEE Computer, vol. 26,no. 10, pp. 71–72, October 1993.

[210] ——, “The computer for the 21st century,” ACM SIGMOBILE Mobile Com-puting and Communications Review, vol. 3, no. 3, pp. 3–11, July 1999.

[211] A. Wheeler, “Commercial applications of wireless sensor networks using zig-bee,” IEEE Communications Magazine, vol. 45, no. 4, pp. 70–77, April 2007.

[212] Wibree, “Ultra-low power radio technology for small devices,” Available: http://www.wibree.com/technology/Wibree_2Pager.pdf, 2007.

[213] A. Willig, “Recent and emerging topics in wireless industrial communications:A selection,” IEEE Transactions on Industrial Informatics, vol. 4, no. 2, pp.102–124, May 2008.

[214] A. Willig, K. Matheus, and a. Wolisz, “Wireless technology in industrial net-works,” Proceedings of the IEEE, vol. 93, no. 6, pp. 1130–1151, June 2005.

[215] WiMAX, “WiMAX forum website,” Available: http://www.wimaxforum.org,2007.

Bibliography 129

[216] M. Wolf and D. Kress, “Short-range wireless infrared transmission: the linkbudget compared to RF,” IEEE Wireless Communications Magazine, vol. 10,no. 2, pp. 8–14, April 2003.

[217] K.-J. Wong and D. Arvind, “SpeckMAC: Low-power decentralized MAC pro-tocols for low data rate transmissions in specknets,” in Proc. 2nd Int’l Work-shop on Multi-hop Ad Hoc Networks: from Theory to Reality (REALMAN’06),Florenze, Italy, May 2006, pp. 71–78.

[218] T. Wu and S. Biswas, “A self-reorganizing slot allocation protocol for multi-cluster sensor networks,” in Proc. 4th International Symposium on Informa-tion Processing in Sensor Networks (IPSN’05), Los Angeles, CA, USA, April2005, pp. 309–316.

[219] N. Xu, S. Rangwala, K. K. Chintalapudi, D. Ganesan, A. Broad, R. Govin-dan, and D. Estrin, “A wireless sensor network for structural monitoring,” inProc. 2nd Int’l Conf. on Embedded Networked Sensor Systems (SenSys’04),Baltimore, MD, USA, November 3–5 2004, pp. 13–24.

[220] W. Ye, J. Heidemann, and D. Estrin, “An energy-efficient MAC protocol forwireless sensor networks,” in Proc. 21st Annual Joint Conf. of the IEEE Com-puter and Communications Societies (INFOCOM’02), vol. 3, New York, NY,USA, June 23–27 2002, pp. 1567–1576.

[221] W. Ye, F. Silva, and J. Heideman, “Ultra-low duty cycle MAC with scheduledchannel polling,” in Proc. 4th ACM Conf. Embedded networked sensor systems(SenSys’06), Boulder, Colorado, USA, October 31 – November 3 2003, pp.329–334.

[222] J. Yinghui and L. Qingchong, “Throughput analysis of spread slotted alohasystems using multiuser receivers,” in Proc. Military Communications Confer-ence (MILCOM’02), Anaheim, CA, USA, October 7–10 2002, pp. 237–242.

[223] S. Yoon, “Power management in wireless sensor networks,” North CarolinaState University, PhD Thesis, Raleigh, NC, USA, January 2007.

[224] M. Yousef and N. El-Sheimy, “Wireless sensor network: research vs. realitydesign and deployment issues,” in Proc. 5th IEEE Annual Conference on Com-munication Networks and Services Research (CNSR’07), New Brunswick,Canada, May 14–17 2007, pp. 8–9.

130 Bibliography

[225] P. Zhang, C. M. Sadler, S. A. Lyon, and M. Martonosi, “Hardware designexperiences in zebranet,” in Proc. 2nd Int’l Conf. on Embedded networkedsensor systems (SenSys’04), Baltimore, MD, USA, November 3–5 2004, pp.227–238.

[226] R. Zhang and J.-M. Gorce, “Optimal transmission range for minimum energyconsumption in wireless sensor networks,” in Proc. IEEE Wireless Communi-cations and Networking Conference (WCNC’08), Budapest, Hungary, March31 – April 3 2008, pp. 757–762.

[227] J. Zheng, M. Lee, and M. Anshel, “Toward secure low rate wireless personalarea networks,” IEEE Trans. Mobile Computing, vol. 5, no. 10, pp. 1361–1373,October 2006.

[228] C. Zhu and M. Corson, “A five-phase reservation protocol (fprp) for mobile adhoc networks,” in Proc. 17th Annual Joint Conference of the IEEE Computerand Communications Societies (INFOCOM’98), vol. 1, San Fransisco, CA,USA, March 29 - April 2 1998, pp. 322–331.

[229] ZigBee Specification Version 1.0, ZigBee Alliance, 2004.

PUBLICATIONS

132 Publications

PUBLICATION 1

M. Kohvakka, "TUTWSN MAC Protocol," book chapter in Ultra-low energy wire-less sensor networks in practice: theory, realization and deployment, M. Kuorilehto,M. Kohvakka, J. Suhonen, P. Hämäläinen, M. Hännikäinen, and T.D. Hämäläinen,Chichester: John Wiley, 2007, pp. 145–182.

© 2007 Copyright John Wiley & Sons Limited. Reproduced with permission.

The publication is intentionally excluded from the electronic version due to copyrightreasons.

PUBLICATION 2

M. Kohvakka, J. Suhonen, M. Hännikäinen, and T. D. Hämäläinen, “TransmissionPower Based Path Loss Metering for Wireless Sensor Networks,” in Proceedingsof the 17th Annual IEEE International Symposium on Personal, Indoor and MobileRadio Communications (PIMRC 2006), Helsinki, Finland, Sep. 11–14, 2006, pp.1–5.

© 2006 IEEE. Reprinted, with permission, from the proceedings of the 17th AnnualIEEE International Symposium on Personal, Indoor and Mobile Radio Communica-tions 2006.

This material is posted here with permission of the IEEE. Such permission of theIEEE does not in any way imply IEEE endorsement of any of the Tampere Universityof Technology’s products or services. Internal or personal use of this material ispermitted. However, permission to reprint/republish this material for advertising orpromotional purposes or for creating new collective works for resale or redistributionmust be obtained from the IEEE by writing to [email protected].

The 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC’06)

1-4244-0330-8/06/$20.002006 IEEE

TRANSMISSION POWER BASED PATH LOSS METERING FOR WIRELESS SENSOR NETWORKS

Mikko Kohvakka, Jukka Suhonen, Marko Hännikäinen and Timo D. Hämäläinen Tampere University of Technology, Institute of Digital and Computer Systems

Tampere, Finland

ABSTRACT

The metering of path loss to neighbouring nodes is vital in multi-hop Wireless Sensor Networks (WSN) for managing network self-configuration, robust data routing, and node lo-calization. Conventionally, path loss is measured by a Re-ceived Signal Strength Indicator (RSSI) mechanism of a transceiver. Yet, the lowest hardware complexity, energy con-sumption and cost are reached by transceivers without RSSI. We present a new simple transmission power based path loss metering method for WSNs without RSSI mechanism. The method determines path loss from frames transmitted at dif-ferent power levels. Performance measurements with physical WSN prototypes indicate sufficient accuracy for network management. According to performance analysis, the power consumption of the proposed method is below 10 µW. An energy analysis in a large IEEE 802.15.4 network indicates 51% to 69% energy saving compared to RSSI-equipped transceivers by using the path loss metering method together with a simple and low power transceiver.

I. INTRODUCTION

Self-configuring Wireless Sensor Networks (WSN) are a rela-tively new family of wireless ad hoc networks developed for low data rate monitoring and actuating in home, office, indus-trial and outdoor environments [1]. WSN consist of even thousands of battery-powered WSN nodes, which combine sensing, processing and wireless communication with ultra-low energy consumption, size and cost. As the number of nodes is large, battery or node replacements are difficult, and nodes should operate for months to years without mainte-nance of any kind. The energy consumption in WSN nodes is typically dominated by transceiver circuitry [2]. To minimize transmission energy consumption, data is routed in the net-work by multiple low-energy hops instead of a single high-energy transmission. Communication links are formed dy-namically with any node in a range creating a mesh network topology.

An energy efficient and robust WSN operation requires that nodes are capable to measure radio wave attenuation (path loss) to neighbouring nodes. This information is crucial for efficient network self-configuration and routing; nodes can form and maintain robust routing paths with sufficient radio link quality and energy efficient hop lengths [3]. A weakened radio link can be recognized and route changed before an actual link failure. In addition, transmission power levels are dynamically minimized for reducing network energy con-sumption and interferences with other nodes. As radio wave attenuates with the transmission distance [4], measured path loss can be used for estimating distances between nodes (ranging), and further for node localization [5]. Location in-formation improves the applicability of gathered data in a

dynamic network, and allows even random node deploy-ments.

In practice, a path loss is estimated by measured signal strengths of received frames and their known transmission power levels [5]. Conventionally, signal strength is measured by a Received Signal Strength Indicator (RSSI) mechanism implemented on a transceiver PHY layer. A transceiver calcu-lates the RSSI based on the gain in the receiver chain in con-junction with the energy of the signal after channel filtering, and stores the value in a register. Yet, the simplest and lowest power commercial off-the-shelf radio transceivers lack the quite complex RSSI mechanism. To be able to achieve the lowest energy and cost WSN realizations, a transceiver with-out RSSI may be the only option, and the implementation of a path loss metering otherwise is required.

In this paper we present a new type of a path loss metering mechanism for minimum power WSNs. The method utilizes receiver sensitivity as a simple signal strength meter, and es-timates path loss according to succeed and failed receptions of beacon frames transmitted at different transmission power levels. This is possible since almost every transceiver has an adjustable transmission power. The accuracy of the path loss metering method is measured with low power WSN proto-types in indoor and outdoor environments. Achieved energy efficiency is demonstrated in an IEEE 802.15.4 Low-Rate Wireless Personal Area Network (LR-WPAN) [6] by analyz-ing coordinator power consumption with different low power transceivers.

The rest of this paper is organized as follows. A detailed de-scription of the method is presented in Section 2. Section 3 presents performance measurements and power consumption analysis. Section 4 concludes the paper.

II. PATH LOSS METERING DESCRIPTION

The presented path loss metering method is designed for low duty cycle Medium Access Control (MAC) protocols with a periodic wakeup scheme. In a periodic wakeup scheme, data transmissions and reception are performed in short data ex-change periods, while the rest of time nodes are in inactive mode conserving energy, as presented in Figure 1. Beacon transmissions at the beginning of data exchange periods are used for synchronizing data exchanges and signalling network

Beacon Data exchange period

Inactive periodNode 1

Node 2

Beacon interval

Figure 1: Periodic wakeup scheme.


management information. Most well-known low duty cycle MAC protocols are S-MAC [7] and its variants, and a beacon-enabled mode of LR-WPAN [6].

In this paper, a path loss A means the signal attenuation be-tween a transmitter output and receiver input ports including antenna gain, and the losses in impedance matching network. Thus, A is determined as the ratio of transmission power level PT to received signal strength PR that is measured according to succeed and failed frame receptions and receiver sensitivity Smin. Receiver sensitivity is defined as the minimum mean power received at the input port at which the Bit Error Rate (BER) does not exceed a specified value (typically 0.1%). This yields that a frame of l bits can be received successfully exactly at the probability of (1 - BER)l, when PR = Smin. For simplification we approximate that PR ≥ Smin, when a frame is received successfully. The upper and lower limits for the path loss are estimated by retransmitting a frame with different transmission power levels, and detecting the lowest transmis-sion power resulting a successful reception PT(succeed) and the highest power resulting a failed reception PT(failed) as

( ) ( )( ) ( )max minT failed T succeed

R R

P PA

P P< ≤ . (1)

As an example, calculated path loss values for a transceiver having Smin = -85 dBm and beacon transmission powers from -20 dBm to 0 dBm are presented in Table 1. Since path loss is typically quite equal for both directions [8], the same path loss is assumed to apply for transmissions to neighbours.

The path loss metering is implemented on MAC protocol beacons due to the following reasons. First, the interval of beacon transmissions of low duty cycle MAC protocols is very suitable for path loss measurements. In addition, beacon transmissions are synchronized, which allows the minimiza-tion of reception time. Beacons are also quite short, which provides energy efficient implementation without the need for a new frame type. Moreover, path loss information is typi-cally needed together with the node status information trans-mitted in beacons for network management.

The operational principle of the path loss metering method is presented in Figure 2, where Node 3 measures the path losses between Node 3 and two other nodes (Node 1 and Node 2) by receiving their periodic beacons. Achieved ranges of beacon transmissions with four power levels are marked by circles. As Node 1 is quite close to Node 3, a beacon transmit-ted at the second lowest transmission power level can be re-ceived resulting low path loss. As a distance from Node 2 is much longer, only the highest power beacon is successfully received. This indicates a high path loss.

For reducing energy consumption, beacons are transmitted in the increased order of transmission power. This enables that beacon frames are received until one reception succeeds, after which the receiver may be switched to a power down mode. As the contents of the beacons are similar, the recep-tion of higher power beacons does not bring any additional information, and they may be ignored.

III. PERFORMANCE ANALYSIS

The performance analysis of the path loss metering method consists of accuracy measurements in indoor and outdoor environment, and power consumption analysis in a large LR-WPAN network. A. Accuracy Measurements The accuracy of the path loss metering method is measured experimentally by two TUTWSN prototype platforms, pre-sented in Figure 3. The platform utilizes 2.4 GHz low power Nordic Semiconductor nRF2401A transceiver, and a Micro-chip PIC18LF4620 microcontroller. The transceiver has ad-justable transmit power level from -20 dBm to 0 dBm with four steps. Receiver sensitivity is -85 dBm at 1 Mbps data rate. A full-wave loop antenna provides a dipole-type radia-tion pattern, and around 4 dBi directivity (simulated). Since the transceiver lacks the RSSI mechanism, the platform is suitable for testing the performance of the path loss metering.

The transmission ranges of beacons with four different transmission power levels (-20 dBm, -10 dBm, -5 dBm and 0 dBm) are measured outdoors in a park in a fairly open space, and in a roadside mostly line-of-sight conditions. One plat-

Table 1: Path loss according to succeed (1) and failed (0) receptions.

transmission power level Resulted path loss -20 dBm -10 dBm -5 dBm 0 dBm

≤ 65 dB 1 1 1 1 65 – 75 dB 0 1 1 1 75 – 80 dB 0 0 1 1 80 – 85 dB 0 0 0 1

Node 1 (TX)

Node 2 (TX)

Node 3 (RX)

Range with the highest power beacon

Range with the lowest power beacon

Lowest power beacon

Highest power beacon

Failed receptionsSuccessful reception

Node 1 Node 2

Node 3

Successful reception

Radio power down

Figure 2: Path loss metering operational principle.

AntennaTransceiver

Battery

MCU

Temperature sensor

Voltage regulation

Push button and LED

Figure 3: TUTWSN prototype platform.


form is placed 1.5 m above a snowy ground and configured to periodically transmitting beacons. Another node is moved away from the transmitter around 2 m above the ground, and is receiving beacons. The beams of loop antennas are directed towards each other. The minimum transmission power levels of succeed beacon receptions (min(PT(succeed))) are recorded as a function of the distance between the transmitting and re-ceiving node, as presented in Figure 4. In an open space, a path loss increases quite proportionally to the distance. Bea-cons transmitted at -20 dBm, -10 dBm, -5 dBm, and 0 dBm power levels are received until around 25m, 190 m, 350 m and 370 m distances, respectively. In a roadside, a significant radio wave fading and gaining is caused by reflections from ground and surrounding snowy trees and rocks. Beacons transmitted at -20 dBm, -10 dBm, -5 dBm, and 0 dBm power levels are received until around 60 m, 130 m, 340 m and 440 m distances, respectively. The results are reasonable and well comparable to RSSI measurements, such as presented in [8].

The accuracy of the path loss metering method should be sufficient for simple outdoor localization algorithms, such as presented in [5]. According to the measurements, antenna directivity and radio wave reflections have higher effect on the ranging accuracy than the path loss measurements mecha-nism itself. Localization accuracy would be highly improved by a more omni-directional antenna, such as a monopole.

Indoor measurements are conducted in office environment, presented in Figure 5. A beacon transmitting node is marked with TX and placed on a corridor around 2 m above a floor such that antenna beam is directed along the long corridor. The measuring points are marked with the numbers 1 to 4, according to the minimum power level the successfully re-ceived beacons are transmitted with. The measurement results are very reasonable. With the smallest -20 dBm transmission power around 40 m distance is achieved. A glass door and a wall attenuate a radio wave 5 to 10 dB.

The accuracy of the path loss metering is sufficient for managing robust network operation. The accuracy for indoor node localization depends on antenna directivity and radiation power, and node density. Since walls and doors cause very significant signal attenuation and reflections, high node den-sity, omni-directional antennas and very low radiation power provide the highest localization accuracy. B. Power Consumption Analysis First, the average power consumed by the beacon transmis-

sions and receptions (beacon exchange) caused by the path loss metering mechanism is analyzed. Only the transceiver power consumption is analyzed ignoring all other hardware components. For the analysis, a Nordic Semiconductor nRF24L01 transceiver is selected. The transceiver operates at 2.4 GHz frequency band and has 2 Mbps data rate. To the best of our knowledge, this is the most energy efficient com-mercial off-the-shelf transceiver available today having 6.25 nJ/bit energy consumption in RX mode and 3.5 nJ/bit energy consumption in TX mode with -18 dBm power level. The transceiver has a carrier detect functionality, but lacks RSSI. In the analysis, a beacon transmission power level is adjusted between -18 dBm and 0 dBm. A beacon length is 24 B to comply with the LR-WPAN standard [6].

During a beacon interval, one set of one to four beacons are transmitted and received. This corresponds to the operation of a routing node in WSN. Transceiver start-up transients, crys-tal inaccuracy (20 ppm), and sleep mode power consumption are included in the analysis. For comparison, the same analy-sis is performed for a LR-WPAN compliant Chipcon CC2420 transceiver. Since the transceiver has an RSSI mechanism, only one beacon transmission and one reception during a bea-con interval are required for a basic MAC operation.

The power consumption results are presented in Figure 6. The beacon exchange power consumption is typically far be-low 10 µW. With 4 s beacon interval (BI) a basic MAC op-eration with single beacon transmissions and receptions con-sumes 3.51 µW with the nRF24L01 transceiver. If two bea-

111123 4

4

41

1

1 1 1

2

22

1

3 344 4

33

2 211 1 1

1

342

131

4

4

32

2 12 4

2 2 2 2 2 2 2 3 444 43

1 2 3 4

TX

-20 dBm -10 dBm -5 dBm 0 dBm Out of range

40 m 70 m

Glass doorGlass door

X

XXX

XX X

X

X

X

The minimum transmission power of the received beacons:

Figure 5: Test points and the minimum transmission powers of received beacons in office environment.

-20

-15

-10

-5

0

5

0 100 200 300 400 500Distance (m)

min

(PT(succeed)

) (dB

m)

RoadsideOpen space

Out of range

Figure 4: The minimum beacon transmission powers as a function of transmission distance in outdoor environment.


cons are sufficient for the path loss metering method, the in-crease for power consumption is 2.78 µW resulting 6.29 µW beacon exchange power. With four beacons, beacon exchange power is 13.6 µW. For comparison, the CC2420 transceiver beacon exchange power consumption with single beacons and RSSI measurements is 20.08 µW. Clearly, the increase of transmitted beacons increase measurement accuracy, but raise beacon exchange energy consumption. Energy can be saved by transmitting two beacons, but randomizing the transmis-sion power of the lower power beacon. For maintaining net-work connectivity, the range of beacon transmissions must not vary, and the higher power beacon is always transmitted with the highest power. Thus, a node can accurately measure path loss during the time of few beacon intervals.

For demonstrating the energy efficiency of an entire WSN node utilizing the path loss metering method, the power con-sumption of a routing LR-WPAN coordinator is mathemati-cally analyzed in a large-scale WSN application. Network energy consumption is minimized by selecting a cluster-tree network topology with a beacon-enabled mode. The analyzed coordinator routes data from ten associated devices (nD) and one child coordinator (nC), which forwards data from 200 child and grandchild nodes (nDL). A low data rate sensing application is executed, where each node transmits a sensor value to uplink direction at 30 s intervals (IU). Each sensing item (LS) is 8 B long containing a sensor value, a node ID, and a time stamp. In addition, control and configuration data is broadcasted to downlink direction at 120 s intervals (ID). This results totally 460 bits/s throughput requirement through the coordinator. A MAC superframe order is fixed to 2 result-ing 57.6 ms Contention Access Period (CAP) length. Net-work dynamics and link failures are included in the analysis by averagely 1 hour network scan interval (INS), where each scan consumes energy ENS. A probabilistic number of beacon reception attempts (pB) is 75% of transmitted beacons (nB), as more than one beacons are transmitted.

The performance of LR-WPAN is analyzed in a star net-work in [9]. We complement the performance analysis with contention models in a cluster-tree network. LR-WPAN MAC is prone to two types of collisions; collisions caused by the hidden node problem, and the selection of a same backoff slot with another node. The probability of a hidden node collision (ph) between two nodes associated with a same coordinator,

but outside the range of each other is modelled with the dura-tions of a data (tDATA) frame, an acknowledgement (tACK) frame and CAP (tCAP) as

2 .DATA ACKh

CAP

t tpt

+= (2)

Nodes randomize backoff delays using a backoff exponent (BE) initialized with macMinBE (default is 3). The probabil-ity (pd) that two nodes select the same backoff delay is

1 .2 1d BEp =

− (3)

For modelling the probability of a successful transmission (ps), models for the average number of data transmissions (d) and contenting nodes (C) in CAP are defined as

21 2 CDLD

U D U D

nnd BI n uI I I N I

= + + +

, (4)

21 2min ,1 min ,1 ,CDLD

U D U D

nnC BIu n BIuI I I N I

= + + +

(5)

where u is the expected number of retransmissions. ID is di-vided by 2 for approximating the effect of indirect communi-cation. Coordinator aggregates data from nDL nodes by trans-mitting N (defined as 12) sensing items in a frame. Thus, ps can be modelled as

( ) ( )1 1 ,hd Cs h dp s p p= − − (6)

where h is the probability that two nodes in the range of a coordinator have a hidden node relationship (41% according to [10]), and s [9] is the probability of a successful clear channel assessment after macMaxCSMABackoffs attempts.

A frame is transmitted a = aMaxFrameRetries+1 times be-fore declaring a transmission failure. The probability of a successful transmission (v) and the average number of trans-mission attempts per frame (u) are

( ) 1

11

k ak

s sk

v p p=

−

== −∑ , (7)

( ) ( ) 1

11 1 .

k ak

s sk

u v a kp p=

−

== − + −∑ (8)

Finally, coordinator power consumption can be modelled as

( )

( ) ( )

( )

max 1,0

,

TXB B RXB B B RXBU

TXD RXA D DLCAP RX

U

TXD RXA RXDD TXA NS

D NS

E n E n p EP

BIE E n n u

t PI N

E E E E u EI I

+ + −=

+ ++ +

+ + ++ +

(9)

where ETXB, ERXB, ERXBU, ETXD, ETXA, ERXA, and ERXDD are the energy consumptions of a beacon transmission and reception, an unsuccessful beacon reception, a data transmission, an

1

10

100

0 5 10 15 20Beacon interval (s)

Bea

con

exch

ange

pow

er (µ

W)

. CC2420 (1 beacon, RSSI)nRF24L01 (4 beacons)nRF24L01 (3 beacons)nRF24L01 (2 beacons)nRF24L01 (1 beacon)

Figure 6: Beacon exchange power consumption as a function

of a beacon interval with one to four beacons.


acknowledge transmission and reception, and a data reception after a data request, respectively. PRX is the coordinator power consumption in reception mode. Moreover, the achieved goodput (G) through a coordinator can be modeled with a requested throughput and v as

( )2

.D CD DLS

U D

n nn nG L vI I

+ += +

(10)

Power consumptions and goodputs are separately analyzed for three low power 2.4 GHz transceivers, presented in Table 2. Chipcon CC2400 and CC2420 transceivers have RSSI mechanism, while Nordic Semiconductor nRF24L01 utilizes the implemented path loss metering method. Since the CC2400 transceiver has RSSI and high 1 Mbps data rate, high energy efficiency is expected. All the transceivers have on-chip data buffers for transmitted and received data, which allows interfacing with low-speed microcontrollers. Data buffer sizes are 32 B for nRF24L01 and CC2400 transceivers, and 128 B for the CC2420 transceiver. The rest of the ana-lyzed hardware is similar to the TUTWSN prototype plat-form, presented in Figure 3.

Resulted coordinator power consumption as a function of achieved goodput is plotted in Figure 7. Results are calculated with 2.0 V supply voltage. According to the analysis, CC2420, CC2400 and nRF24L01 transceivers can achieve over 435 bits/s goodput with 2.25 mW, 1.45 mW, and 0.703 mW power consumptions, respectively. The beacon intervals in these points are 1.97 s, 3.93 s and 3.93 s, respectively (not shown in the figure). By increasing the number of beacons to four, the nRF24L01 achieves the same goodput with 0.711 mW power. The results indicate that energy saving using the lowest complexity transceiver is very considerable, and the energy consumption of the path loss metering method is in-significant. According to the analysis, the minimum power consumption with the nRF24L01 transceiver is 0.316 mW, which is achieved with 333 bits/s goodput and 15.7 s beacon interval. At longer beacon intervals the loading of CAP in-creases significantly, which increases collisions and re-transmissions. This further increases CAP loading reducing achieved goodput and increasing power consumption, as seen with the nRF24L01 results. Also, the energy for network scan is very significant at long beacon intervals.

IV. CONCLUSIONS

The implementation of a simple transmission power based path loss metering method is presented. According to per-formance measurements, the method can provide the func-tionality of RSSI for the lowest complexity and power trans-ceivers with very low power consumption. In the analyzed large-scale LR-WPAN application, the path loss metering

consumes less than 0.5% of the entire node power consump-tion. In addition, achieved power saving by using the nRF24L01 transceiver and the path loss metering method is 69% compared to the CC2420 LR-WPAN compliant trans-ceiver with RSSI. Compared to the CC2400 low power RSSI equipped transceiver, the power saving is still over 51%.

REFERENCES

[1] H. Karl and A. Willig. Protocols and Architectures for Wireless Sensor Networks, John Wiley & Sons Ltd., Chichester, pages 1-13. 2005.

[2] F. Jin, H.-A. Choi and S. Subramaniam. Hardware-aware communica-tion protocols in low energy wireless sensor networks. In Proceeding of the IEEE Military Communications Conference, volume 22, no. 1, Bos-ton, USA, pages 676-681, 2003.

[3] M. Sikora, J.N. Laneman, M. Haenggi, D.J. Costello, Jr. and T. Fuja. On the Optimum Number of Hops in Linear Wireless Networks. In Pro-ceedings of the IEEE Information Theory Workshop, Texas, USA, pages 165-169, 2004.

[4] T.S. Rappaport, Wireless Communications – Principles and Practice, Prentice Hall, 1996.

[5] M.L. Sichitiu, V. Ramadurai and P. Peddabachagari. Simple Algorithm for Outdoor Localization of Wireless Sensor Networks with Inaccurate Range Measurements. In Proceedings of the International Conference on Wireless Networks, Nevada, USA, pages 300-305, 2003.

[6] IEEE Std 802.15.4-2003, wireless medium access control (MAC) and physical layer (PHY) specifications for low-rate wireless personal area networks (WPANs).

[7] Y. Wei, J. Heidemann and D. Estrin. Medium access control with coor-dinated adaptive sleeping for wireless sensor networks. IEEE/ACM Transactions on Networking, 12(3): 493-506, June 2004.

[8] N. Reijers, G. Halkes and K. Langendoen. Link Layer Measurements in Sensor Networks. In Proceedings of the 1st IEEE International Confer-ence on Mobile Ad-hoc and Sensor Systems, Florida, USA, pages 224-233, 2004.

[9] N.F. Timmons and W.G. Scanlon. Analysis of the performance of IEEE 802.15.4 for medical sensor body area networking. In Proceedings of the 1st Annual IEEE Communications Society Conference on Sensor and Ad Hoc Communications and Networks, California, USA, pages 16-24, 2004.

[10] Y.-C. Tseng, S.Y. Ni and E.Y. Shih. Adaptive approaches to relieving broadcast storms in a wireless multihop mobile ad hoc network. IEEE Transactions on Computers, 52(5): 545-557, May 2003.

Table 2: Comparison of analyzed transceivers.

Radio data rate [kbps] RSSI

sleep [µA]

RX [mA]

TX (0dBm) [mA]

TX (-10dBm)

[mA] CC2400 1000 Yes 1.5 24 19 13 CC2420 250 Yes 0.02 18.8 17.4 11

nRF24L01 2000 No 0.9 12.3 11.3 7.5

0

1

2

3

4

0 100 200 300 400 500Goodput (bits/s)

Coo

rdin

ator

pow

er (m

W) CC2420 (RSSI)

CC2400 (RSSI)nRF24L01 (4 beacons)nRF24L01 (2 beacons)

Requested throughput: 460 bits/s

Figure 7: LR-WPAN coordinator power consumption as a

function of achieved goodput.

PUBLICATION 3

M. Kohvakka, M. Kuorilehto, M. Hännikäinen, and T. D. Hämäläinen, “NetworkSignaling Channel for Improving ZigBee Performance in Dynamic Cluster-Tree Net-works,” EURASIP Journal on Wireless Communications and Networking, vol. 8, no.3, pp. 1–15, January 2008.

Hindawi Publishing CorporationEURASIP Journal on Wireless Communications and NetworkingVolume 2008, Article ID 456535, 15 pagesdoi:10.1155/2008/456535

Research ArticleNetwork Signaling Channel for Improving ZigBee Performancein Dynamic Cluster-Tree Networks

Mikko Kohvakka, Jukka Suhonen, Mauri Kuorilehto, Marko Hannikainen, and D. Hamalainen

Institute of Digital and Computer Systems, Tampere University of Technology, P.O. Box 553, Tampere 33101, Finland

Correspondence should be addressed to Mikko Kohvakka, [email protected]

Received 30 April 2007; Revised 26 October 2007; Accepted 19 January 2008

Recommended by Biao Chen

ZigBee is one of the most potential standardized technologies for wireless sensor networks (WSNs). Yet, sufficient energy-efficiencyfor the lowest power WSNs is achieved only in rather static networks. This severely limits the applicability of ZigBee in outdoor andmobile applications, where operation environment is harsh and link failures are common. This paper proposes a network channelbeaconing (NCB) algorithm for improving ZigBee performance in dynamic cluster-tree networks. NCB reduces the energy con-sumption of passive scans by dedicating one frequency channel for network beacon transmissions and by energy optimizing theirtransmission rate. According to an energy analysis, the power consumption of network maintenance operations reduces by 70%–76% in dynamic networks. In static networks, energy overhead is negligible. Moreover, the service time for data routing increasesup to 37%. The performance of NCB is validated by ns-2 simulations. NCB can be implemented as an extension on MAC andNWK layers and it is fully compatible with ZigBee.

Copyright © 2008 Mikko Kohvakka et al. This is an open access article distributed under the Creative Commons AttributionLicense, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properlycited.

1. INTRODUCTION

Wireless sensor network (WSN) is an emerging technologyenabling fully autonomous self-configuring ad-hoc networks[1]. WSN may consist of thousands of small nodes, whichsense their environment, communicate wirelessly with eachother, and share collaborative tasks. Nodes route sensed dataand events to a sink node, which forms a gateway to othernetworks. WSN nodes may be embedded deeply in our liv-ing environment or operate under harsh conditions in out-doors or a machinery hall. Thus, a high tolerance against un-reliable radio links, variable network size, and mobile nodesis required [2, 3]. Due to the large number of nodes, bat-tery replacements are difficult. Hence, nodes may have toscavenge supply energy solely from their operation environ-ment [4], or operate up to several years with small batteries.This necessitates very high energy-efficiency in communica-tion protocols, algorithms, and hardware platforms. WSNshave a vast number of potential applications [5], for examplemonitoring and controlling in home, office, and industrialenvironments, monitoring of remote or hostile geograph-ical regions, tracking of animals and objects, and surveil-lance.

The wireless personal area networks (WPANs) workinggroup was initially focused on creating the IEEE 802.15.1standard for physical (PHY) and medium access control(MAC) layers based on Bluetooth technology [6]. The work-ing group soon formed two other subgroups, firstly IEEE802.15.3 focusing on high-speed WPAN [7] for multimediaapplications. In December 2000, IEEE 802.15.4 [8] low-rateWPAN was initiated for providing low-complexity, low-cost,and low-power wireless connectivity among inexpensive de-vices, such as WSN nodes [9]. ZigBee [10] is an open spec-ification for low-power wireless networking, which comple-ments IEEE 802.15.4 with a network layer, security modes,and application profiles. The first version of the ZigBee spec-ification was announced in December 2004.

IEEE 802.15.4 together with ZigBee is one of the mostpotential standardized technologies for enabling WSNs. Asavailable energy is scarce, the beacon-enabled network is es-sential, since time synchronized sleep and wakeup mech-anism can be adopted [11]. Moreover, large network sizewith widely located nodes is enabled by a cluster-tree net-work topology. By these settings, ZigBee can achieve veryhigh energy-efficiency in static networks [12]. However,even an immobile WSN has a dynamic behavior caused by

mailto:[email protected]

2 EURASIP Journal on Wireless Communications and Networking

low-transmission power levels combined with dynamic oper-ating environment, such as opened and closed doors, movingobjects, and interferences from other networks that all affectradio frequency (RF) propagation. In addition, many envi-sioned WSN applications, such as asset tracking and interac-tive games, require mobility support for nodes. In these con-ditions, ZigBee performance is unsatisfactory due to energy-hungry passive scan operations. Hence, techniques for reduc-ing passive scan energy are one of the key challenges.

In this paper, we propose a network channel beaconing(NCB) algorithm for improving the performance and reduc-ing the energy consumption of ZigBee in dynamic networks.NCB utilizes frequent beacon transmissions on a dedicatednetwork signaling channel. A passive scan is performed by re-ceiving these beacons resulting in dramatically reduced pas-sive scan duration and energy consumption in low-data-rateapplications. NCB can be implemented as a manufacturerspecific extension on MAC and NWK layers and it is fullycompatible with standard ZigBee devices.

The rest of this paper is organized as follows. The re-lated work is discussed in Section 2. An introduction to Zig-Bee and IEEE 802.15.4 standard is presented in Section 3.Section 4 presents the design of NCB. The performance ofNCB equipped ZigBee is analyzed and compared againststandard ZigBee in Section 5. In Section 6, the beacon trans-mission rate of NCB is energy optimized for further energysaving. Simulations for validating the results are presented inSection 7. Finally, this paper is concluded in Section 8.

2. RELATED WORK

Most of researches about the IEEE 802.15.4 beacon-enablenetwork have been restricted to a star topology. The perfor-mance of a star network with 10 nodes has been analyzed in[13]. The analysis focused on the effect of crystal tolerance, aframe size, and the usage of guaranteed time slots (GTSs) ona node lifetime. Bougard et al. [14] have presented a mathe-matical analysis of a large-scale star network. A special con-tribution is bit error rate measurements with two evaluationboards connected through a set of calibrated attenuators. Theoperational analysis considers mainly the effect of path lossand packet size on energy consumption. The performancesimulations of IEEE 802.15.4 in a star network have been pre-sented in [15]. It has been found that a significant energysaving is achieved by a low-duty-cycle operation.

An analysis and experimental measurements of IEEE802.15.4 in a cluster-tree network has been presented in [11].It has been found that beacon-enabled cluster-tree networksare prone to beacon collisions, which will lead to synchro-nization failures. The beacon collisions can be reduced byutilizing a rather long beacon interval. Clearly, another op-tion is to use more frequency channels for the network. Yet,both of these options increase the energy consumption ofpassive scans proprotionally.

Beacon collisions can also be reduced by scheduling. Awakeup scheduling scheme presented in [16] utilizes syn-chronized superframe timing such that coordinators beginbeaconing at the same time, and active periods are fully over-lapping. Since entire inactive period can be used for sleep-

ing, energy consumption is reduced. To reduce collisions,beacon transmission period is extended to contain a num-ber of subslots during which beacons can be sent. Yet, theselection of a noninterfering subslot has not been specified.Another scheduling scheme presented in [17] organizes theentire active periods of different coordinators in a nonover-lapping manner. This minimizes the changes to the currentIEEE 802.15.4 specification. Scheduling has also been pro-posed for reducing hidden node collisions that are typicalfor the carrier sense multiple access with collision avoidance(CSMA-CA) mechanism of IEEE 802.15.4 [18].

While the above scheduling schemes utilize software-based time synchronization; also hardware-based synchro-nization mechanisms exist. RT-Link [19] is a time divi-sion multiple access (TDMA) MAC protocol for multihopWSNs. The protocol employs two out-of-band synchroniza-tion sources: atomic clock broadcasts for outdoors and am-plitude modulation (AM) transmissions for in indoor con-ditions. The latter utilizes building’s power grid as an an-tenna to radiate time sync pulses. According to experimen-tal measurements, the hardware-based synchronization is ro-bust, scalable, and energy-efficient option to software-basedtechniques. Yet, the hardware cost and complexity are in-creased.

In our previous work [12], we have analyzed the per-formance of IEEE 802.15.4 and ZigBee in large-scale WSNapplications. An energy analysis on a cluster-tree multi-hop network indicated that the highest energy-efficiency ina low-data-rate WSN application is achieved by moderate(30.7 milliseconds) superframe duration and by scaling bea-con interval according to requested throughput. Accordingto simulations, random error situations are energy-hungry,since they usually cause the reconstruction of a completesubtree. During the reconstruction, leaf nodes may need toperform several passive scans until the above network hier-archy is initialized. Simulations also depicted that the proba-bility that two coordinators randomize the same slot for theirsuperframes is significant. For better scalability, it would beadvantageous to divide clusters into several frequency chan-nels.

Furthermore, our previous work [20] has proposed theuse of frequent beacon transmissions to improve the per-formance of synchronized low-duty-cycle MAC protocols indynamic networks. The work focused on flat networks, forwhich energy optimized beacon rate was determined. It hasbeen found that optimized beacon rate results in a very sig-nificant energy saving.

In this paper, we will extend our idea of frequent bea-con transmissions to ZigBee. In contrast to [20], we willutilize a cluster-tree network topology and a dedicated net-work signaling channel for frequent network beacons. Net-work beacons allow energy-efficient passive scans on the net-work channel, independently from the utilized IEEE 802.15.4beacon interval and the amount of frequency channelsfor superframes. This enables the optimization of energy-consumption and scalability in dynamic networks. As thenetwork beacons are transmitted on the network channelonly; collisions with data and control frames are eliminated.We will also present performance analysis of the network

Mikko Kohvakka et al. 3

Application ZDO

SSPApplication support (APS)

NWK

MAC

868/915 MHzPHY

2.4 GHzPHY

IEEE 802.15.4

ZigBee alliancedefined

Figure 1: IEEE 802.15.4 and ZigBee.

channel approach. Performance models are validated by sim-ulations.

3. ZigBee AND IEEE 802.15.4 OVERVIEWS

3.1. ZigBee protocol stack

A ZigBee protocol stack is presented in Figure 1 [21]. An ap-plication layer at the top of the stack determines node re-lationships, and supervises network initiation and associa-tion functions. Overall node management is performed by aZigBee device object (ZDO). Application endpoints may callZDO in order to discover other ZigBee nodes on the networkand the services they offer, and to define security and net-work settings. A security service provider (SSP) offers secu-rity functions including encryption, key generation, key dis-tribution, authentication, and access control lists.

A network (NWK) layer provides network self-organ-ization and multihop routing capability. NWK performsroute discovery and maintenance, and message relayingfunctions. NWK can initiate a new network and assign net-work addresses to new nodes associating with the networkfor the first time.

An application support (APS) sublayer connects NWK,SSP, and endpoints, and routes messages to different end-points.

MAC and PHY layers are defined by IEEE 802.15.4. MACis responsible for the channel access mechanism, acknowl-edged frame delivery, network association, and disassocia-tion.

IEEE 802.15.4 supports two direct sequence spread spec-trum (DSSS) PHY layers operating in industrial, scientific,and medical(ISM) frequency bands. A low-band PHY oper-ates in the 868 MHz or 915 MHz frequency band and has araw data-rate of 20 Kbps or 40 Kbps, respectively. A high-band PHY operating in the 2.4 GHz band specifies a data-rate of 250 kbps and has nearly worldwide availability. The2.4 GHz frequency band is the most potential for large-scaleWSN applications, since the high-radio data-rate reducesframe transmission time and usually also the energy pertransmitted and received bit.

3.2. Node types

IEEE 802.15.4 defines three types of logical devices, a per-sonal area network (PAN) coordinator, a coordinator, and adevice. ZigBee denotes them as ZigBee coordinator, ZigBee

router, and ZigBee end-device, respectively. For clearness, weutilize the naming of IEEE 802.15.4 from now on.

PAN coordinator is the primary controller of PAN, whichinitiates the network and operates often as a gateway to othernetworks. Each PAN must have exactly one PAN coordina-tor. Coordinators collaborate with each other for executingdata routing and network self-organization operations. De-vices do not have data routing capability and can communi-cate only with coordinators.

Due to the low-performance requirements of devices,they may be implemented with very simple and low-costhardware. The standard designates these low-complexitynodes as reduced function devices (RFD). Nodes with thecomplete set of MAC services are called as full function de-vices (FFDs).

3.3. Network topologies

The standard supports two network topologies, star, andpeer-to-peer, presented in Figure 2. In the star topology, alldata exchanges are controlled by a PAN coordinator that op-erates as a network master, while devices operate as slaves andcommunicate only with the PAN coordinator. This single-hop network is most suitable for delay critical applications,where large network coverage is not required.

A peer-to-peer topology allows “mesh” type of networks,where any coordinator may communicate with any othercoordinator within its range, and have messages multihoprouted to coordinators outside its range. This enables theformation of complex self-organizing network topologies.The network may contain also RFDs as devices. Peer-to-peertopologies are suitable for industrial and commercial appli-cations, where efficient self-configurability and large cover-age are important. A disadvantage is the increased networklatency due to message relaying.

One special type of peer-to-peer topology is a cluster-treenetwork, defined by ZigBee. The network consists of clusters,each having a coordinator as a cluster head and multiple de-vices as leaf nodes. A PAN coordinator initiates the networkand serves as the root. The network is formed by parent-childrelationships, where new nodes associate as children with theexisting coordinators. This well-defined structure simplifiesmultihop routing and allows effective energy saving; eachnode maintains synchronization of data exchanges with itsparent coordinator only. The rest of time, nodes may saveenergy in a sleep mode. This is not possible in the peer-to-peer networks, where coordinators need to receive continu-ously to be able to receive data from any node in the range.A disadvantage is that a coordinator failure may cause a largeamount of orphaned child and grandchild nodes causing en-ergy wasting during network reassociations [12].

3.4. MAC layer

The MAC layer can operate on both beacon-enabled andnonbeacon modes. In the nonbeacon mode, a protocol is asimple CSMA-CA. This requires a constant reception of pos-sible incoming data. The power saving features that are criti-cal in WSN applications are provided by the beacon-enabled


Star topology Peer-to-peer topology Cluster tree topology

ZigBee coordinator

FFD

RFD

Figure 2: Star, peer-to-peer, and cluster-tree topology examples.

mode. Hence, we concentrate on the beacon-enabled modefrom now on.

In the beacon-enabled mode, all communications areperformed in a superframe structure presented in Figure 3.A superframe is bounded by periodically transmitted beaconframes, which allows network nodes to synchronize them-selves to the network. An active part of a superframe is di-vided into three parts: the beacon, contention access period(CAP), and contention-free period (CFP). At the end of thesuperframe is an inactive period, when nodes may enter to apower saving mode.

CAP is a mandatory part of a superframe during whichchannel is accessed using a slotted CSMA-CA scheme. Coor-dinators are required to listen to the channel the whole CAPto detect and receive any data from their child nodes. Thechild nodes may only transmit data and receive an optionalacknowledgement (ACK) on demand, which increases theirenergy-efficiency.

CFP is an optional part of a superframe. Nodes requir-ing dedicated bandwidth and low-latency transmissions [22]may request GTS from a PAN coordinator. CFP does not uti-lize any collision avoidance mechanism. Due to inter-clustercollisions, the applicability and benefits of GTS are very lim-ited in large peer-to-peer or cluster-tree networks. In addi-tion, CFP can be utilized only for a direct communicationwith a PAN coordinator.

The beacon interval (BI) and the active superframe dura-tion (SD) are adjustable by IEEE 802.15.4 parameters beaconorder (BO) and superframe order (SO) as

BI = aBaseSuperframeDuration× 2BO,

SD = aBaseSuperframeDuration× 2SO,(1)

where 0 ≤ SO ≤ BO ≤ 14, and aBaseSuperframeDurationequals 960 radio symbols or 15.36 milliseconds in the2.4 GHz band. Hence, BI and SD range between 15.36 mil-liseconds and 251.7 seconds. The superframe structure ismaintained by a PAN coordinator. In cluster-tree networks,all coordinators transmit beacons for assisting other nodes tomaintain synchronization with them.

For minimizing inter-cluster interferences, it is desirableto concatenate superframes of neighboring clusters. ZigBeespecifies that BI is divided into BI/SD slots. During a start-

BeaconCAP CFP Inactive

BI

SD

GT

S

GT

S

Figure 3: Superframe structure in beacon-enabled mode.

up, each coordinator randomizes a free slot for its super-frame.

4. THE DESIGN OF NETWORK CHANNEL BEACONING

The designed algorithm should improve ZigBee energy-efficiency, throughput, and scalability in dynamic low-duty-cycle cluster-tree networks, while the energy overhead instatic networks should be insignificant.

A starting point for the design is IEEE 802.15.4 config-ured to a beacon-enabled mode with inactive period. Thesesettings provide the highest energy-efficiency in static net-works. For maintaining the energy-efficiency also in dynamicnetworks, we focus on minimizing the energy consumptioncaused by link failures.

4.1. Network scan in IEEE 802.15.4

A network scan over a given list of channels is initiatedby a MLME-SCAN.request primitive. The most importantparameters are ScanType, ScanChannels, and ScanDuration.ScanType is used to select a suitable scan type among en-ergy detection (ED) scan, active scan, passive scan, or orphanscan. ScanChannels is a bitmap indicating which channels areto be scanned. ScanDuration is used to calculate the lengthof time to spend scanning each channel. ED scan is used todetermine channel usage, active or passive scan to locate bea-con frames containing any PAN identifier, or an orphan scanto locate a PAN to which the device is currently associated.Passive and orphan scans are mandatory requirements for alldevices. ED and active scans are optional for an RFD.

The ED scan is used by a prospective PAN coordinatorto select a suitable channel for a new PAN. The ED scan es-timates the received signal power within the bandwidth of achannel, while no attempt is made to identify or decode sig-nal on the channels. Based on the detected energy, the chan-nel with the lowest energy can be selected.

The active and passive scans allow a device to locatingany coordinator transmitting beacon frames within its per-sonal operating space (POS). These scan types are used by aprospective PAN coordinator to select a PAN identifier priorto starting a new PAN, or they could be used by a device forselecting a suitable coordinator prior to association.

The active scan is performed by sending a beacon requestcommand and then by receiving for possible beacon framesin return. Upon receiving the beacon request, each node in


Beacon...

Beacon

Beacon...

Beacon

Beacon...

Beacon

PD-DATA.indication...

PD-DATA.indication


PD-DATA.indication

...


PD-DATA.indication

Device PHYDevice MACDevice nexthigher layer

MLME-SCAN.request(passive)

Set 1st channel

ScanDuration

Set 2nd channel

ScanDuration

Set last channel

ScanDuration

MLME-SCAN.confirm(success)

Figure 4: Passive scan message sequence chart of ZigBee.

a range transmits a beacon. This necessitates that each coor-dinator is continuously in reception mode.

The passive scan is performed by receiving beaconswithout transmitting any requests. This necessitates thatall coordinators transmit beacons at predetermined inter-vals in some of the specified frequency channels. However,coordinators are not required to be constantly in recep-tion mode allowing the inactive time, which is crucial forenergy-efficiency. The message sequence chart of the IEEE802.15.4 passive scan is presented in Figure 4. For detect-ing all coordinators in POS, passive scan should be con-ducted on each selectable channel at least a beacon trans-mission interval. In ZigBee, passive scan duration equalsBI + aBaseSuperframDuration.

The orphan scan allows a device to attempt to relocateits coordinator following a loss of synchronization. The scanis performed in each of specified set of channels by send-ing an orphan notification command and then by receiv-ing for a possible coordinator realignment command until amacResponseWaitTime (491.2 milliseconds at 2.4 GHz band)expires. The scan is terminated, if the realignment commandis received, or all the specified channels have been scanned.

The orphan scan is suitable for networks, where coordi-nators are constantly in reception mode. In energy-efficientnetworks utilizing the inactive period, coordinators are typ-ically most of time in sleep mode, and the reception of theorphan notification command cannot be guaranteed. Thus,the orphan scan should be replaced by a MAC sublayer resetfollowed by a passive scan, and a network reassociation, asspecified by IEEE 802.15.4.

As the focus of this paper is on networks where all nodesutilize the inactive period, only ED scan and passive scan areapplicable. Since ED scan is utilized only at the startup of aPAN coordinator, we can focus purely on the minimizationof passive scan energy consumption.

4.2. Minimization of passive scan energy

Passive scan energy consumption is a product of transceiverpower consumption in reception mode and passive scan du-ration. Since it is difficult to affect the transceiver power char-acteristics, we focus on minimizing the scanning time.

As found in [19], the energy consumption of a onesecond long network (passive) scan equals the energy ofnearly 3000 beacon transmissions by a typical low-powertransceiver. One can easily conclude that it is worthwhile toincrease beacon transmissions rate by reducing BI. Yet, thecost of a shorter BI is increased beacon reception energy dueto beacon synchronization. Due to time drift, the synchro-nized reception of beacons is also more energy hungry thanthe transmission of them.

For eliminating this drawback, the designed NCB algo-rithm utilizes additional beacon transmissions on a dedi-cated network signaling channel. These network beacons aretransmitted without collision avoidance independently fromthe IEEE 802.15.4 specified beacons (standard beacons) andreceived during passive scans only. Since the dedicated net-work channel eliminates collisions with data transmissions,the transmission rate of network beacons can be significantlyhigher than the rate of standard beacons. This effectivelyminimizes the required passive scan duration reducing bothenergy consumption and data routing breaks. Moreover, thepassive scan duration is independent from BI and the num-ber of utilized channels for superframes, which is a majorimprovement for the standard passive scan. Thus, the scala-bility can be improved by increasing BI length and the num-ber of utilized frequency channels, while maintaining rapidand low-energy passive scans. As the 2.4 GHz and 915 MHz,frequency bands have 16 and 10 selectable channels, respec-tively, the use of a separate network signaling channel is fea-sible.

In principle, the network beacons are used for search-ing a suitable coordinator with which to associate. Then,the association is performed on the cluster channel of a de-sired coordinator, where standard beacon, data, and controlframes are exchanged. In practice, occasional collisions be-tween beacons are possible. However, they do not interferewith data and control frame exchanges, or the standard bea-con synchronization. Assuming that the transmission inter-val of network beacons is around two orders of magnitudelonger than a beacon length, the probability of collisions isvery low in sparse networks. In dense networks, we can as-sume that adequate number of beacons can be received cor-rectly even though some collisions occur.

Due to frequency selective channel fading, the link qual-ity of the network and cluster channels may differ [23]. Thiscauses two consequences: all coordinators in the range of thecluster channel are not detected, or the signal strength of a se-lected coordinator is too low for communication, when op-eration is switched to the cluster channel. However, we canassume that an adequate number of network beacons can bereceived during the passive scan for ensuring network con-nectivity. According to the schedule and frequency channelinformation of received network beacons, a node attemptsto receive the standard beacons, one after another, until


a suitable coordinator with adequate signal strength is found.Standard beacons are received accurately at the specified mo-ments minimizing the energy consumption of idle listening.

If the network beacons of two coordinators collide, colli-sions are also probable with beacons transmitted later on, be-cause beacons are sent periodically at constant intervals. Toprevent beacon collisions without the need of network bea-con scheduling, we propose delaying network beacon trans-mission by a random jitter (J), defined as

J = ϕ[0 · · · JMAX

] · tTXB, (2)

where ϕ[a, b] is a random function on the interval [a, b],JMAX is the maximum jitter, and tTXB is the time required tosend a network beacon.

As presented in Figure 5, network beacons are broad-casted by all coordinators during inactive periods. This is eas-ily managed by a single radio transceiver. Normally, networkbeacons are transmitted at rate fN . However, assuming oneradio per station, a coordinator may not send a network bea-con, while maintaining an active period or communicatingwith another coordinator. Then, the transmission of a bea-con must be delayed until the end of the active period, caus-ing at most 1/ fN + SD interval between network beacons.While it is possible to avoid delaying the transmission byselecting suitable network beacon interval and active periodboundaries, the presented interval is considered as a practicalnetwork scan time with NCB.

The algorithm operates similarly on PANs operating ona single channel and PANs, where coordinators are dividedinto several frequency channels. Additional information car-ried in network beacons in respect to IEEE 802.15.4 bea-cons are an exact time to the beginning of the next super-frame, and the frequency channel of the coordinator. Thepresented design does not have any effect on the standardbeacons transmitted at the beginning of superframes.

The message sequence chart of a passive scan utilizingNCB is presented in Figure 6. As network beacons are trans-mitted at rate fN, passive scan is performed by the MLME-SCAN.request primitive by setting: ScanType = passive, Scan-Channels = network channel, and ScanDuration = 1/ fN + SD.

If the network will be deployed in an environment havinghigh-RF interferences, some frequency channels may be lo-cally jammed [23]. The robustness of network channel can beimproved by defining two network channels, where networkbeacons are transmitted consecutively. Correspondingly, apassive scan is conducted on the both channels. The networkchannels should be selected before deployment according toRF spectrum measurements. It should also be noted that itis always possible to fall back into the standard passive scan.However, in the rest of this paper we assume rather low-RFinterferences and utilize one network channel.

4.3. Implementation guidelines

The management functions of NCB algorithm should be im-plemented on the NWK layer. In addition, IEEE 802.15.4MAC should be modified by adding functionality for net-work beacon transmissions. The format of network beacons

is a slight variation of an IEEE 802.15.4 beacon frame. Wesuggest replacing the fields GTS fields and pending addressesof the IEEE 802.15.4 beacon frame with time to next super-frame and coordinator channel, as illustrated in Figure 7. Foradequate resolution, the space allocated for these fields are 4B and 1 B, respectively. The initialization and transmissionsof network beacons are very similar to standard beacons.

For maintaining all the benefits of standardized technol-ogy, interoperability with standards is essential. In principle,a ZigBee node should be able to find a NCB equipped co-ordinator by a passive scan and to associate with it and viceversa.

We suggest adding a NCB specification field in the pay-load of standard beacons specifying the utilized networkchannel, and the transmission rate of network beacons. TheNCB field that can be utilized for determining is NCB sup-ported in the PAN.

At a startup, a new PAN coordinator should perform ac-tive/passive and an ED scans, as specified by IEEE 802.15.4.According to the scans, free channels for the PAN and net-work beacons are selected, and beacon transmissions are ini-tiated.

At a startup, other nodes perform a passive scan and se-lect a suitable PAN to associate with, as specified in IEEE802.15.4. According to the NCB specification field, nodes de-termine the usage of network channel in PAN. If NCB is sup-ported, network beacon transmissions are initiated. For al-lowing compatibility with nodes lacking the NCB function-ality, the standard specified passive scan should be performedeach time the NCB passive scan cannot find suitable parents.Since the NCB algorithm does not affect standard beacontransmissions, the ordinary passive scan easily finds all nodesresulting in cross-compatibility with ZigBee.

For the highest energy-efficiency, it is important to en-ergy to optimize network beacon rate according to the levelof network dynamics [19]. A successful operation of NCB ne-cessitates that the network beacon rate is uniform and glob-ally known in entire network. It is possible to predeterminethe network beacon rate according to presumable networkdynamics in a given application. However, a more efficientoption is to adjust the network beacon rate dynamically. Wesuggest that each node observes and maintains a record ofan average link lifetime. The maintained values are transmit-ted to a PAN coordinator in data frames upon a request. Itis suggested that the PAN coordinator broadcasts the requestat regular intervals, for example once per hour. According tothe gathered link lifetimes, the PAN coordinator determinesan energy optimal network beacon rate for the PAN. The net-work beacon rate is flooded through the network by utilizingthe NCB specification field of standard beacons minimizinga control frame overhead.

5. PERFORMANCE ANALYSIS

To be able to analyze purely the effect of NCB on the en-ergy of network maintenance operations, we first divide theenergy consumed by the wireless communication into threeclasses [19]: node startup, network maintenance, and dataexchange energies, as illustrated in Figure 8. Node startup



Time

Time

Time

Time

Network channel receptionfor network scan

A C A C

D B D B

F E F E

Networkchannel

Channel 1

Channel 2

...

Channel NBI

Superframe ofcoordinator D

Superframe ofcoordinator B

Superframe ofcoordinator F

Superframe ofcoordinator E

Network beacons of:

Coordinator ACoordinator BCoordinator C

Coordinator DCoordinator ECoordinator F

Figure 5: Principle of frequent network beacons.

Beacon...

Beacon


PD-DATA.indication

Device PHYDevice MACDevice nexthigher layer

MLME-SCAN.request(passive)

Set network channel

ScanDuration= 1/ fN + SD

MLME-SCAN.confirm(success)

Figure 6: Passive scan message sequence chart of NCB equipped ZigBee.

energy is composed of a passive scan for detecting coordi-nators in a range, and a network association for connectingthe node to the network. The network association consistsof control packet exchange according to the IEEE 802.15.4specification. Network maintenance and data exchange oper-ations are executed after the startup period. Network main-tenance energy consists of beacon transmissions and recep-tion, passive scans, and network reassociations according tooccurred link failures. Data exchange energy is consumedby upper-layer payload data and acknowledgement transmis-sions and receptions. We define that network beacon trans-missions have lower priority than data exchanges. Hence,network beacon broadcasts do not affect data exchange en-ergy. From now on, we can focus on the minimization ofnetwork maintenance energy.

Next, the performance analysis of NCB algorithm in aWSN application will be presented. The analysis begins byan energy analysis of a reference hardware platform, and pro-ceeds to network maintenance power analysis through MACoperation models.

5.1. Hardware platform

To obtain realistic results, steady-state power consump-tions of a commercial IEEE 802.15.4 compliant Chip-con CC2420EM/EB [24] transceiver evaluation board aremeasured. In addition, the power consumption of aPIC18LF8722 [25] microcontroller is measured to estimatethe energy consumption of a low-power MAC protocolprocessor. The measured power consumptions of the plat-form in different operation modes are presented in Table 1.Highest power of 56.5 mW is consumed in reception mode(PRX). In transmission mode, power consumption variesfrom 26.6 mW to 48.0 mW. We define that frames are al-ways transmitted with the highest transmission power levelresulting in 48.0 mW transmission power (PTX). Sleep morepower consumption is 30 μW. Table 2 presents the measuredtransient times as the transceiver switches from sleep to idlemode and from idle to active (RX or TX) mode. The tran-sient times from active to idle and from idle to sleep modeare negligible.


MACMAC

headerSuperframespecification

Time to nextsuperframe

Coordinatorchannel Payload MAC

footer

PHYSynch.header

PHYheader

MAC protocol data unit (MPDU)

Figure 7: Suggested format of an NCB network beacon frame.

Start-up period Node lifetime

Star

t-u

pen

ergy

:(i

)Pa

ssiv

esc

an(i

i)N

etw

ork

asso

ciat

ion Data exchange energy:

(i) Upper layer payload data andacknowledgement exchange

Network maintenance energy:(i) Beacon transmissions and receptions

(ii) Passive scans and network re-associationsafter link failures

Time

Figure 8: Node energy consumption classification.

Table 1: Measured static platform power consumptions at 3 V sup-ply voltage.

Symbol MCU mode Radio mode Power consumption

PTX

Active TX (0 dBm) 48.0 mW

Active TX (−1 dBm) 45.0 mW







PRX Active RX 56.5 mW

PI Active Idle 2.79 mW

PS Sleep Sleep 30 μW

Table 2: Measured chipcon CC2420 transient times.

Symbol Description Time (μs)

tST Sleep to idle 970

tIA Idle to active 192

5.2. Network configuration

We analyze the energy-efficiency of NCB in a cluster-tree net-work, where each cluster contains nD (11) devices and nC (2)child coordinators. Network depth (d) is 4 levels of hierarchy.Coordinators broadcast standard beacons once per BI result-ing a standard beacon rate fC = 1/BI, where BI gets valuesbetween 0.246 second and 15.7 seconds. Network beacon ratefN is varied from 0.1 Hz to 100 Hz. The network operates onnCH channels. The utilized parameters and their values arepresented in Table 3.

5.3. MAC operation models

To be able to analyze the network maintenance energy, MACoperation models are defined for a beacon transmission, abeacon reception, and a passive scan.

A beacon frame transmission consists of a sleep-to-idle(tSI) transient, idle-to-active (tIA) transient, and actual datatransmission defined as the ratio of beacon frame length (LB)and radio data-rate (R). No CCA analysis is needed for a bea-con frame. During the sleep-to-idle transient, radio powerconsumption equals PI, after which power consumption risesto PTX. The beacon transmission energy (ETXB) is

ETXB = tSIPI + (tIA + LB/R)PTX. (3)

The resulting energy consumption per transmitted frameis ETXB = 39.6 μJ, which equals 190 nJ per a PHY layer databit.

A beacon reception begins with radio startup transients.The radio is in reception mode until a beacon frame has beenreceived including a time margin required due to synchro-nization inaccuracy (tI), and the time drift between a trans-mitting and a receiving node. As synchronization is obtainedby standard beacon receptions; the time drift caused by crys-tal tolerance (ε) is directly proportional to BI. The beaconreception energy (ERXB) is

ERXB = tSIPI + (tIA + 2εBI + tI + LB/R)PRX. (4)

The resulting energy consumption per received frame isERXB = 70.2 μJ (BI = 0.96 second). This equals 338 nJ per aPHY layer data bit.

A passive scan begins with the startup transients. Then, aradio is in RX mode during passive scan duration (tNS). Theenergy required for message exchanges during a reassociationis negligible compared to the energy of the passive scan, andthus it is ignored in the following analysis. Thus, the passivescan energy (ENS) is

ENS = tSIPI + (tIA + tNS)PRX. (5)

The passive scan duration depends on the utilization ofNCB. For ZigBee, the durationis a function of the number ofscanned cluster channels and a beacon order as

tNS = nCH · aBaseSuperframeDuration (2BO + 1). (6)

As defined above, practical network scan duration forNCB equipped ZigBee is

tNS = 1/ fN + SD. (7)


Table 3: Utilized parameters and their values.

Symbol Parameter Value

fC Standard beacon rate 63.8 mHz–16.4 Hz

fN Network beacon rate 0.1 Hz–100 Hz

LB Beacon frame length 26 B

nD Number of devices associated with each coordinator 12

nC Number of child-coordinators associated with each coordinator 3

nCH Number of cluster channels 1 or 10

R Radio data-rate 250 Kbps

r Radio range 20 m

tI Synchronization inaccuracy 0.10 ms

v Node mobility 0.01–10 m/s

ε The crystal tolerance 20 ppm

5.4. Network maintenance power analysis

Network maintenance operations are performed continu-ously during entire network lifetime. Hence, it is most con-venient to consider a long-time energy consumption dividedby the elapsed time resulting in average power consumption.Average network maintenance power PM is defined as a sumof passive scan power PNS and beacon exchange power PB.

The interval of passive scans depends on the average linklifetime (tLF). In should be noted that a link failure some-where along the routing path between a given node and asink may cause a passive scan and network reassociation. Forsimplicity, we omit these link failures in this analysis, and theinterval between passive scans equals a link lifetime. We alsoomit occasional frame errors, and determine link lifetime ac-cording to node mobility. Assuming a random mobility forall nodes, average link lifetime can be approximated by a ra-dio range (r) and node mobility (v) as

tLF = r/v. (8)

The resulted link lifetime as a function of node mobilityis plotted in Figure 9. As mobility increases, node quicklymoves out of the communication range of its coordinator,and the lifetime of a link drops rapidly. According to the linklifetime, the average power consumption of network scans(PNS) is obtained as

PNS = ENS/tLF. (9)

We determine average beacon exchange power accordingto beacon transmission and reception energies. All nodes re-ceive beacons, but they are transmitted by coordinators only.In standard ZigBee, beacons are transmitted and receivedfrom a parent at rate fC. As beacon exchange power is av-eraged over all the nodes of a cluster; beacon exchange powerconsumption (PB) for ZigBee is

PB =fC

nD + 1ETXB + fCERXB. (10)

When NCB is utilized, beacon transmissions are in-creased. Besides the standard beacons, each coordinator

0.01 0.1 1 10

Mobility (m/s)

1

10

100

1000

10000

Lin

klif

etim

e(s

)

Figure 9: The effect of mobility to link lifetime in the simulatedscenario.

transmits network beacons on the network channel at ratefN . For NCB equipped ZigBee, average beacon exchangepower consumption is

PB =fN + fCnD + 1

ETXB + fCERXB. (11)

5.5. Energy analysis results

The network maintenance powers are compared betweenstandard ZigBee and NCB equipped ZigBee. BI and v arefixed to 3.96 seconds and 0.01 m/s, respectively. As the net-work operation is divided into 1 and 10 channels, the main-tenance power of the standard ZigBee equals 132 μW, and1.13 mW, respectively, as presented in Figure 10. When NCBis used, the network maintenance power PM (the sum of PB

and PN) has a minimum of 42 μW at 2.6 Hz network bea-con rate. At low-beacon rates below 1 Hz, PM typically dou-bles as the beacon rate halves and the power consumption aredominated by the passive scan power. The effect is reversedat high-beacon rates above 10 Hz, when the beacon transmis-sions dominate the power consumption. At the energy opti-mum 2.6 Hz network beacon rate, NCB algorithm reducesthe network maintenance power up to 96%. Yet, the BI mayalso be energy optimized reducing the network maintenancepower of standard ZigBee.

For finding an energy optimal value for BI, the networkmaintenance power of ZigBee is analyzed as the functionof BI. From now on, the number of channels is fixed to 1.


0.1 1 10 100

fN (Hz)

0

1

10

100

1000

1000

PM

(μW

)

Beacon exchange power (NCB)Network scanning power (NCB)Network maintenance power (NCB)Network maintenance power (ZigBee, 1 channel)Network maintenance power (ZigBee, 10 channels)

Figure 10: Average beacon exchange, passive scan, and networkmaintenance power consumptions as the function of network bea-con transmission rate.

Average node mobility gets values 0.01 m/s, 0.1 m/s, and1 m/s, ranging from a nearly static network to a quitedynamic network, respectively. The results are presentedin Figure 11. At shorter values of BI, network mainte-nance power is dominated by beacon exchange power con-sumption. At longer BI values, passive scans dominate thepower consumption. An energy optimal BI depends onthe node mobility. The energy optimal values of BI at0.01 m/s, 0.1 m/s, and 1 m/s mobility levels are 1.97 second,0.49 second, and 0.12 second, respectively. The resulted min-imum network maintenance power consumptions at thesepoints are 94 μW, 288 μW, and 936 μW, respectively.

Next, the network maintenance power of NCB equippedZigBee is analyzed as the function of network beacon trans-mission rate. Node mobility gets values 0.01 m/s and 1 m/s,and BI is varied from 0.246 second to 15.7 seconds. The re-sults are presented in Figure 12. In contrast to ZigBee, thenetwork maintenance power is minimized by maximizing BI.This is logical, since passive scans are performed by receivingnetwork beacons independently of the transmission rate ofstandard beacons. Hence, the increase of BI reduces beacontransmission and reception power consumption. The min-imum network maintenance power is achieved by optimiz-ing network beacon transmission rate. At 0.01 m/s and 1 m/snode mobility levels, energy optimal network beacon ratesare 2.6 Hz and 28 Hz, respectively. Achieved network main-tenance power consumptions at these points are 28.1 μW and220 μW, respectively, as BI = 15.7 seconds. Compared to Zig-Bee at these link lifetimes, NCB equipped ZigBee achieves70%–76% lower-network maintenance power. The resultsalso indicate that NCB algorithm operates most efficiently indynamic networks, and an energy optimal network beaconrate depends significantly on a link lifetime. A shorter link

0.01 0.1 1 10 100

BI (s)

0.01

0.1

1

10

100

PM

(mW

)

Mobility: 1 m/sMobility: 0.1 m/sMobility: 0.01 m/s

Figure 11: Average network maintenance power of ZigBee as thefunction of BI, while node mobility varies.

0.1 1 10 100

fN (Hz)

0.01

0.1

1

10

100

PM

(mW

)

v = 1 m/s, BI = 0.246 sv = 1 m/s, BI = 0.983 sv = 1 m/s, BI = 3.93 sv = 1 m/s, BI = 15.7 s

v = 0.01 m/s, BI = 0.246 sv = 0.01 m/s, BI = 0.983 sv = 0.01 m/s, BI = 3.93 sv = 0.01 m/s, BI = 15.7 s

Figure 12: Average network maintenance power of NCB equippedZigBee as the function of network beacon rate, as BI, and node mo-bility varies.

lifetime increases passive scan power consumption and thus,shifts the PM minimum to higher-network beacon rates. Ac-cording to [12], the typical power consumption of a ZigBeenode in a static low-data-rate network is below 1 mW. Hence,network maintenance power has a very significant effect onentire network lifetime and the optimization of network bea-con rate is very important.

6. NETWORK BEACON RATE OPTIMIZATION

An optimal network beacon rate ( f ∗N ) is determined by min-imizing the network maintenance power with respect to the


0.01 0.1 1 10

Node mobility (m/s)

0.1

1

10

100

f∗ N(H

z)

nD = 16nD = 8nD = 4

nD = 2nD = 0

Figure 13: Network maintenance power of NCB equipped ZigBeeat f ∗N as the function of node mobility, and comparison with stan-dard ZigBee.

network beacon rate. An optimization function can be writ-ten as

Pm =fN + fCnD + 1

ETXB + fCERXB

+tSIPI + (tIA + 1/ fN + SD)PRX

rv.

(12)

It can be shown that there exists a unique minimum at f ∗Nthat is obtained by setting dPm/dfN = 0 in (12). This yields

f ∗N =√

PRXv(nD + 1)ETXBr

. (13)

The optimal network beacon rate is determined by radioparameters ETXB,PRX, and r, and network parameters nD andv.

The effect of a link lifetime and a cluster size on the opti-mal network beacon rate is presented in Figure 13. The opti-mal rate is nearly directly proportional to node mobility. Asthe number of devices per cluster (nD) increases from 0 to 16,the energy optimal network beacon rate increases 312%. Al-though, a higher-network beacon rate increases PB in coor-dinators, overall network maintenance power consumptionis reduced due to shorter passive scans.

The network maintenance powers of ZigBee and NCBequipped ZigBee are compared as the function of node mo-bility. In the comparison, BI varies from 0.246 second to15.7 seconds, and NCB is operating at the energy optimalnetwork beacon rate. The results are presented in Figure 14.In static networks with very low mobility, obtained networkmaintenance powers are nearly the same. Hence, the energyoverhead of NCB is small. The energy saving of NCB com-pared to standard ZigBee increases rapidly with higher-nodemobility. At 1 m/s node mobility, achieved energy saving us-ing NCB is from 44% to 99.5%.

0.001 0.01 0.1 1 10

Node mobility (m/s)

0.01

0.1

1

10

100

PM

(mW

)

Standard, BI = 15.7 sStandard, BI = 3.93 sStandard, BI = 0.983 sStandard, BI = 0.246 s

NCB, BI = 15.7 sNCB, BI = 3.93 sNCB, BI = 0.983 sNCB, BI = 0.246 s

Figure 14: Energy optimal network beacon rate as the function ofnode mobility, and as the number of devices per a cluster varies.

Finally, we analyze the effect of NCB on a service time,during which a node is connected to a network and capa-ble for routing data. We consider a time period equaling totLF. The active operation time between passive scans equalstLF reduced by the durations of a passive scan (tNS) and net-work association (tA) operations. After a passive scan, a nodeselects a suitable parent and waits for the beginning of par-ent’s next superframe requiring on average a half BI. Then,the node transmits an association request and waits for a re-sponse to the next superframe. Hence, on average tA = 1.5BI.The service time (S) is determined as the percentual durationof active operation time in proportion to the entire period as

S = INS − tNS − tAtLF

. (14)

The resulted service time as the function of a link lifetimeis plotted in Figure 15. Due to the time required for associa-tion, the service time depends significantly on BI. However,NCB algorithm improves the service time of ZigBee up to37%. The improvement is the most significant at node mo-bility levels above 0.1 m/s.

7. SIMULATIONS

The performance of the network channel beaconing wascompared against standard ZigBee using ns-2 simulation tool(version 2.31) [26]. Next, the NCB implementation for ns-2and obtained performance results are discussed.

7.1. Implementation

For enabling simulations in ZigBee network, few modifica-tions for the IEEE 802.15.4 implementation provided by ns-2were made. First, the code was modified to support a cluster-tree topology with inactive time and to perform synchroniza-tion to the coordinator prior to each association attempt.


0.01 0.1 1 10

Node mobility (m/s)

0

10

20

30

40

50

60

70

80

90

100

Serv

ice

tim

e(%

)

Standard, BI = 0.246 sStandard, BI = 0.983 sStandard, BI = 3.93 sStandard, BI = 15.7 s

NCB, BI = 0.246 sNCB, BI = 0.983 sNCB, BI = 3.93 sNCB, BI = 15.7 s

Figure 15: Service time versus node mobility at f ∗N and compari-son with ZigBee.

Otherwise, the synchronization procedure would have in-cluded a long-term idle listening. Second, a beacon schedul-ing was implemented, as required by the ZigBee. Because thestandard does not specify the exact method to determine thebeacon schedules, a custom algorithm was used. The algo-rithm utilizes precalculated beacon transmission times to en-sure unique time slots for each coordinator within an inter-ference range. It should be noted that while end-to-end delayvaries in different schedules, the energy, throughput, and re-liability results are the same, assuming that a nonconflictingschedule is found.

Moreover, the simplest version of ZigBee routing was im-plemented by omitting the optional routing tables and routediscovery procedure. Instead, the packets were routed alongthe formed cluster tree. While the route discovery procedureallows communication between two arbitrary nodes in thenetwork, it causes high-initial delay, since a node must com-municate with its destination. Therefore, the route discov-ery is impractical for networks having high degree of mobil-ity.

For NCB, the NWK layer was modified to issue the pas-sive scan command with the network channel instead of theusual channel range. The received network beacon frameswere distinguished from other frames by a unique frametype value. The received beacons were handled similarly tothe standard beacons, expect for the time of next super-frame which was calculated with time to next superframefield. Each station was preconfigured with network channeland network beacon interval, thus the beacon interval wasthe same during a simulation run. Coordinator transmittednetwork beacons periodically by briefly switching to the net-work channel. For these simulations, network beacon trans-mission jitter JMAX was set to the value of eight, which wasfound to reduce collisions adequately.

20 m

PAN coordinatorCoordinatorMobile device

Movement pathCommunication range

Figure 16: Simulated network topology. Coordinators are locatedon a grid around a PAN coordinator, while devices move around thegrid.

The simulated scenario consisted of 36 stationary coordi-nators and 8 mobile devices. The coordinators were formedon a grid around a PAN coordinator, while mobile devicesmoved along a circle having a radius of 45 m, as illustrated inFigure 16. Each coordinator was within the communicationrange of its adjacent nodes. That is, the coordinators in thecenter had four neighbors. Maximum hop count (networkdepth) to the PAN coordinator was six. The simulation uti-lized a two-ray ground propagation model allowing commu-nication range of 20 m. This was enough for reliable commu-nication between immediate neighbors, but prevented inter-ference from other nodes. To simulate normal network oper-ation, each mobile device transmitted 40 B packets periodi-cally to the PAN coordinator with 30 s as a period. Simulationtime was one hour.

IEEE 802.15.4 was simulated without CFP and batterylife extension options. During CAP, channel access modewas slotted CSMA-CA. For evaluating power consumption,the measured power consumption and transient times ofChipcon CC2420 were used, as presented in Tables 2 and 3.Frames were always sent with the highest transmission powerlevel (0 dBm).

7.2. Simulation results

The simulations produced power consumption and servicetime results for each node in the network. The obtained re-sults are averaged among all nodes of the same type resultingin average node behavior. It should be noted that the powerconsumption results include also the power consumptions ofa startup and all data exchanges. Thus, the presented powerconsumptions are higher than that in the analysis.

We present first the simulated service time as the func-tion of node mobility. The results are presented in Figure 17.In the results, the maximum relative error with 95% confi-dence is 7%. The service time depends heavily on the bea-con interval, as long interval increases association delays. Be-cause shorter network scan time allows faster neighbor dis-covery and thus, minimizes the time in which a node is not


0 0.5 1 1.5 2 2.5 3

Node mobility (m/s)

0

20

40

60

80

100

120

Serv

ice

tim

e(%

)

Standard, BI = 0.984 sNCB, BI = 0.984 s, fN = 0.2 sNCB, BI = 0.984 s, fN = 0.4 s


Figure 17: Simulated service time of a mobile device (SO = 1).

0 0.5 1 1.5 2 2.5 3

Node mobility (m/s)

0.01

0.1

1

10

En

dde

vice

pow

er(m

W)


Standard, B I = 3.93 sNCB, BI = 3.93 s, fN = 0.2 sNCB, BI = 3.93 s, fN = 0.4 s

Figure 18: Simulated power consumption of mobile devices (SO =1).

connected to the network, NCB increases service time up to14%. The simulations results slightly lower service time thanthe analysis. This is mostly caused by occasional unsuccessfulassociations.

Power usage of a mobile device is presented in Figure 18.In the results, the maximum relative error with 95% confi-dence is 12%. When a device is stationary, only an initial net-work scan is performed. As the normal operation is identicalbetween standard ZigBee and NCB, the difference on energyusage is minimal. When mobility increases, links break oftenand more network scans are required. Then, NCB performssignificantly better than standard due to its short networkscan time.

0 2 4 6 8

Beacon interval (s)

0.01

0.1

1

10

Coo

rdin

ator

pow

er(m

W)

Standard, SO = 1NCB, fN = 0.2 s, SO = 1NCB, fN = 0.4 s, SO = 1

Standard, SO = 2NCB, fN = 0.2 s, SO = 2NCB, fN = 0.4 s, SO = 2

Figure 19: Simulated coordinator power consumption.

On low mobility and at 3.93 seconds beacon interval,NCB has order of magnitude lower power consumption thanthe standard. Yet, the power consumption increases signifi-cantly between 0.6 m/s and 0.8 m/s due to increased failuresduring synchronization. As beacon interval is long, a devicemay try to synchronize to a neighbor that has already movedoutside its communication range. The device tries to trackbeacons and enables its radio for several beacon intervals be-fore failing. Thus, after 0.6 m/s, NCB results are nearly thesame with the standard. At shorter, at 0.98 second beacon in-terval, the same effect occurs after 2.5 m/s mobility.

Figure 19 shows the average power usage on a station-ary ZigBee coordinator. The increased power requirement ofNCB is only 2%–4% higher. This is expected, as networkbeacons are short and do not require CSMA-CA mecha-nism. Thus, the transceiver receptions dominate the powerconsumption. This is evident in Figure 19, as the super-frame length (beacon order) has the most profound effect onpower.

Generally, the simulations show higher-power consump-tion than the analysis. The reason is that simulations in-cluded also power consumption due to data exchange, whileanalysis focused on the network maintenance power only.However, the results and findings are comparable and provethe energy-efficiency of proposed network channel beacons.

8. CONCLUSIONS

The design and performance analyses of the NCB algorithmfor ZigBee are presented. A dedicated frequency channel forfrequent network beacons reduces the energy consumptionof IEEE 802.15.4 passive scans. Energy-efficiency is furtherimproved by optimizing the network beacon transmissionrate according to observed link lifetimes.

According to the energy analysis, NCB algorithm reducesthe network maintenance power of ZigBee in dynamic net-works 70%–76%. This equals even milliwatts absolute power


saving using low-power hardware platforms. As coordinatorpower consumption in a static network and a low- data-rateapplication is typically below 1 mW [12], achieved powersaving is dramatic. In addition, the NCB algorithm improvesnetwork performance by increasing the service time for datarouting up to 37%. Simulations validate the energy-efficiencyof NCB, although the energy saving is slightly lower than inthe analysis.

As the NCB algorithm minimizes passive scan energy re-gardless the number of utilized PAN channels, the divisionof clusters into several frequency channels becomes feasibleand energy-efficient. This would improve ZigBee scalabil-ity and performance especially in dense and large WSNs byeven eliminating inter-cluster interferences. It should be alsonoted that the network beacons provide an efficient way tosignal network maintenance, neighborhood, and routing in-formation for additional algorithms and protocols. NCB canbe implemented as an extension on MAC and NWK layers,and it is fully compatible with ZigBee.

REFERENCES

[1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci,“Wireless sensor networks: a survey,” Computer Networks,vol. 38, no. 4, pp. 393–422, 2002.

[2] J. Polastre, R. Szewczyk, A. Mainwaring, D. Culler, and J. An-derson, “Analysis of wireless sensor networks for habitat mon-itoring,” in Wireless Sensor Networks, C. S. Raghavendra, K.M. Sivalingam, and T. Znati, Eds., pp. 399–425, Springer, NewYork, NY, USA, 2004.

[3] J. Suhonen, M. Kohvakka, M. Hannikainen, and T. D.Hamalainen, “Design, implementation, and experiments onoutdoor deployment of wireless sensor network for environ-mental monitoring,” in Proceedings of the 6th InternationalWorkshop on Embedded Computer Systems: Architectures, Mod-eling, and Simulation (SAMOS ’06), pp. 109–121, Samos,Greece, July 2006.

[4] S. Roundy, P. K. Wright, and J. Rabaey, “A study of low-level vi-brations as a power source for wireless sensor networks,” Com-puter Communications, vol. 26, no. 11, pp. 1131–1144, 2003.

[5] M. Haenggi, “Opportunities and challenges in wireless sensornetworks,” in Handbook of Sensor Networks: Compact Wirelessand Wired Sensing Systems, M. Ilyas and I. Mahgoub, Eds., pp.1.1–1.14, CRC Press, Boca Raton, Fla, USA, 2004.

[6] IEEE Std 802.15.1-2002, wireless medium access control(MAC) and physical layer (PHY) specifications for wirelesspersonal area networks (WPANs).

[7] IEEE Std 802.15.3-2003, wireless medium access control(MAC) and physical layer (PHY) specifications for high ratewireless personal area networks (WPANs).

[8] IEEE Std 802.15.4-2006, wireless medium access control(MAC) and physical layer (PHY) specifications for low-ratewireless personal area networks (WPANs).

[9] J. A. Gutierrez, E. H. Callaway Jr., and R. L. Barrett Jr., Low-Rate Wireless Personal Area Networks—Enabling Wireless Sen-sors with IEEE 802.15.4, IEEE Press, New York, NY, USA, 2003.

[10] “ZigBee Alliance Document 053474r06; ZigBee Specification,Version 1.0,” December 2004.

[11] J. Ha, W. H. Kwon, J. J. Kim, Y. H. Kim, and Y. H. Shin, “Feasi-bility analysis and implementation of the IEEE 802.15.4 multi-hop beacon enabled network,” in Proceedings of the 15th Joint

Conference on Communications & Information (JCCI ’05), pp.1–5, Korea, April 2005.

[12] M. Kohvakka, M. Kuorilehto, M. Hannikainen, and T. D.Hamalainen, “Performance analysis of IEEE 802.15.4 and Zig-Bee for large-scale wireless sensor network applications,” inProceedings of the 3rd ACM International Workshop on Perfor-mance Evaluation of Wireless Ad Hoc, Sensor and UbiquitousNetworks (PE-WASUN ’06), pp. 48–57, Terromolinos, Spain,August 2006.

[13] N. F. Timmons and W. G. Scanlon, “Analysis of the perfor-mance of IEEE 802.15.4 for medical sensor body area network-ing,” in Proceedings of the 1st Annual IEEE ICommunicationsSociety Conference on Sensor and Ad Hoc Communications andNetworks (SECON ’04), pp. 16–24, Santa Clara, Calif, USA,October 2004.

[14] B. Bougard, F. Catthoor, D. C. Daly, A. Chandrakasan, andW. Dehaene, “Energy-efficiency of IEEE 802.15.4 standard indense wireless microsensor networks: modeling and improve-ment perspectives,” in Proceedings of Design, Automation andTest in Europe (DATE ’05), vol. 1, pp. 196–201, Munich, Ger-many, March 2005.

[15] G. Lu, B. Krishnamachari, and C. S. Raghavendra, “Perfor-mance evaluation of the IEEE 802.15.4 MAC for low-rate low-power wireless networks,” in Proceedings of the 23rd IEEE In-ternational Performance Computing and Communications Con-ference (IPCCC ’04), pp. 701–706, Phoenix, Ariz, USA, April2004.

[16] D. Mirza, M. Owrang, and C. Schurgers, “Energy-efficientwakeup scheduling for maximizing lifetime of IEEE 802.15.4networks,” in Proceedings of the 1st International Conference onWireless Internet (WICON ’05), pp. 130–137, Budapest, Hun-gary, July 2005.

[17] A. Koubaa, A. Cunha, and M. Alves, “A time division beaconscheduling mechanism for IEEE 802.15.4/Zigbee cluster-treewireless sensor networks,” in Proceedings of the 19th EuromicroConference on Real-Time Systems (ECRTS ’07), pp. 125–135,Pisa, Italy, July 2007.

[18] L.-J. Hwang, S.-T. Sheu, Y.-Y. Shih, and Y.-C. Cheng, “Group-ing strategy for solving hidden node problem in IEEE 802.15.4LR-WPAN,” in Proceedings of the 1st IEEE International Confer-ence on Wireless Internet (WICON ’05), pp. 26–32, Budapest,Hungary, July 2005.

[19] A. Rowe, R. Mangharam, and R. Rajkumar, “Rt-link: a timesynchronized link protocol for energy constrained multi-hopwireless networks,” in Proceedings of the 3rd Annual IEEE Com-munications Society Conference on Sensor, Mesh and Ad HocCommunications and Networks (SECON ’06), vol. 2, pp. 402–411, Reston, Va, USA, September 2006.

[20] M. Kohvakka, M. Hannikainen, and T. D. Hamalainen, “En-ergy optimized beacon transmission rate in a wireless sensornetwork,” in Proceedings of the 16th IEEE International Sym-posium on Personal, Indoor and Mobile Radio Communications(PIMRC ’05), vol. 2, pp. 1269–1273, Berlin, Germany, Septem-ber 2005.

[21] B. Heile, “Wireless sensors and control networks: enabling newopportunities with ZigBee,” ZigBee Alliance Tutorial, Decem-ber 2006, http://www.zigbee.org.

[22] A. Koubaa, M. Alves, and E. Tovar, “GTS allocation analysis inIEEE 802.15.4 for real-time wireless sensor networks,” in Pro-ceedings of the 14th International Workshop on Parallel and Dis-tributed Real-Time Systems (WPDRTS ’06), pp. 1–8, Rhodes,Greece, April 2006.

http://www.zigbee.org


[23] D. Sexton, M. Mahony, M. Lapinski, and J. Werb, “Radio chan-nel quality in industrial wireless sensor networks,” in Proceed-ings of the Sensors for Industry Conference (SIcon ’05), pp. 88–94, Houston, Tex, USA, February 2005.

[24] Data Sheet for CC2420 2.4 GHz IEEE 802.15.4/ZigBee RFTransceiver, http://focus.ti.com/lit/ds/symlink/cc2420.pdf.

[25] Data Sheet for PIC18LF8722 Microcontroller, http://focus.ti.com/lit/ds/symlink/cc2420.pdf.

[26] The Network Simulator - ns-2, version 2.31, http://www.isi.edu/nsnam/ns.

http://focus.ti.com/lit/ds/symlink/cc2420.pdf



http://www.isi.edu/nsnam/ns

http://www.isi.edu/nsnam/ns

PUBLICATION 4

M. Kohvakka, J. Suhonen, M. Kuorilehto, V. Kaseva, M. Hännikäinen, and T. D. Hä-mäläinen, “Energy-Efficient Neighbor Discovery Protocol for Mobile Wireless Sen-sor Networks,” Ad Hoc Networks, vol. 7, no. 1, pp. 24–41, January 2009.

© 2007 Elsevier B.V. Reprinted with Permission.

Available online at www.sciencedirect.com

Ad Hoc Networks 7 (2009) 24–41

www.elsevier.com/locate/adhoc

Energy-efficient neighbor discovery protocol formobile wireless sensor networks

Mikko Kohvakka *, Jukka Suhonen, Mauri Kuorilehto, Ville Kaseva,Marko Hannikainen, Timo D. Hamalainen

Tampere University of Technology, Institute of Digital and Computer Systems, FI-33720 Tampere, Finland

Received 14 November 2006; received in revised form 2 November 2007; accepted 27 November 2007Available online 15 December 2007

Abstract

Low energy consumption is a critical design requirement for most wireless sensor network (WSN) applications. Due tominimal transmission power levels, time-varying environmental factors and mobility of nodes, network neighborhoodchanges frequently. In these conditions, the most critical issue for energy is to minimize the transactions and time con-sumed for neighbor discovery operations. In this paper, we present an energy-efficient neighbor discovery protocol targetedat synchronized low duty-cycle medium access control (MAC) schemes such as IEEE 802.15.4 and S-MAC. The protocoleffectively reduces the need for costly network scans by proactively distributing node schedule information in MAC pro-tocol beacons and by using this information for establishing new communication links. Energy consumption is furtherreduced by optimizing the beacon transmission rate. The protocol is validated by performance analysis and experimentalmeasurements with physical WSN prototypes. Experimental results show that the protocol can reduce node energyconsumption up to 80% at 1–3 m/s node mobility. 2007 Elsevier B.V. All rights reserved.

Keywords: Algorithm/protocol design and analysis; Low power design; Mobile computing; Wireless sensor networks

1. Introduction

Wireless sensor network (WSN) is the mostpotential technology for very low power ubiquitousnetworks. Foreseen applications fields include mon-

1570-8705/$ - see front matter 2007 Elsevier B.V. All rights reserved

doi:10.1016/j.adhoc.2007.11.016

* Corresponding author. Tel.: +358 3 3115 4702; fax: +358 33115 4786.

E-mail addresses: [email protected] (M. Kohvakka),[email protected] (J. Suhonen), [email protected] (M.Kuorilehto), [email protected] (V. Kaseva), [email protected] (M. Hannikainen), [email protected] (T.D.Hamalainen).

itoring of remote or hostile geographical regions,tracking of animals and objects, and monitoring insmart building and industries [1,2]. WSN may con-sist of even thousands of small and fully autono-mous nodes, which gather sensor information,perform data processing, and communicate witheach other. Nodes route data by multiple low-energy hops to sink nodes, which may act as gate-ways to other networks. Network management isfully decentralized eliminating the need for a fixedcontroller and infrastructure, which prevents asingle point of failure.

.








M. Kohvakka et al. / Ad Hoc Networks 7 (2009) 24–41 25

In most of these envisioned applications, nodesmust operate with a scarce energy budget. Theymay have to scavenge supply energy solely fromtheir operation environment, or operate up to sev-eral years with small batteries [3]. Hence, the energyconsumption is one of the key performance metricsfor WSN realizations.

To reach the energy budget, WSN nodes operatewith constrained communication and computationresources. Thus, a typical node performs data pro-cessing using a MicroController Unit (MCU) hav-ing a couple of million instructions per second(MIPS) processing speed and tens of kbytes pro-gram and data memories [4,5].

Although the advances in radio frequency (RF)circuits have been significant, a radio transceiver isstill the most power-consuming component in aWSN node. The power consumption of currentradios is nearly the same in transmission and recep-tion modes, and energy is saved only in a sleepmode, in which the radio circuitry is completelyswitched off. The nodes should come out of the sleepmode only to transmit or receive packets that arevital for node operation, thus avoiding idle listeningof unnecessary traffic [6]. Therefore, a desired pairof nodes should be active simultaneously. As a glo-bal synchronization is very difficult to reach in largenetworks using low power and low cost components[7], nodes may reduce idle listening and energy con-sumption by forming locally synchronized commu-nication links with their neighbors.

WSNs are often considered stationary in a waythat the established communication links are unal-tered for entire network lifetime. In practice, evenan immobile network has a dynamic behavior [8,9]due to random node failures and dynamic operatingenvironment caused by opened and closed doors,moving objects, changing weather conditions, andinterferences from other networks, which all affectRF propagation. In addition, numerous envisionedWSN applications, such as access control, assetstracking, and interactive games necessitate themobility of nodes causing dynamics for networktopology [6,10]. Thus, WSN nodes should be ableto recognize topology changes and update commu-nication links rapidly, energy-efficiently, and withminimum interruptions on application data routing.In current WSN proposals, new neighbors are typi-cally discovered by listening on available RF chan-nels for their transmissions (network scan), lastingup to tens of seconds [11–13]. As calculated in[14], each second of a network scan consumes the

energy equaling to 2800 frame transmissions usinga typical low power transceiver. In dynamic net-works, scans may increase node energy consump-tion one order of magnitude [14]. Thus, we needan energy-efficient neighbor discovery to substitutefor network scans.

In this paper, we present a new energy-efficientneighbor discovery protocol (ENDP) for synchro-nized low duty-cycle medium access control (MAC)schemes. The presented protocol reduces the needfor network scans by distributing synchronizationinformation from nodes in two-hop neighborhood.This information is carried in the beacon payloadsof underlying MAC protocol and utilized for estab-lishing new communication links. In addition,ENDP introduces an efficient network beacon sig-naling scheme to make network scans more energy-efficient. To the best of our knowledge, ENDP isthe first protocol that can effectively minimize net-work energy consumption in dynamic WSNs. Theenergy efficiency and operation fidelity are verifiedby analytical performance models and experimentalmeasurements using real WSN prototypes.

This paper is organized as follows. Section 2 pre-sents design approaches for low duty-cycle MACprotocols and discusses how neighbor discovery iscurrently performed. The design of ENDP is pre-sented in Section 3. For obtaining realistic results,we introduce a WSN prototype platform and deter-mine its radio energy models in Section 4. Accord-ing to the energy models, the performance ofENDP is determined analytically and optimized inSection 5. Section 6 presents experimental measure-ments by the prototypes. The paper is concluded inSection 7.

2. Related research

2.1. Low duty-cycle MACs

The MAC sublayer operating on top of a physi-cal layer (PHY) manages radio transmissions andreceptions on the shared wireless medium, andhence has a major effect on network performanceand energy consumption. The design objectives ofWSN MAC differ completely from traditional wire-less voice and data networks, which pursue to max-imize wireless medium utilization. In WSNs, MACis pursuing to provide energy-efficient, adaptiveand error tolerant operation, and adequate scalabil-ity for large and dense networks having only fewkbit/s network throughput requirement [15].

26 M. Kohvakka et al. / Ad Hoc Networks 7 (2009) 24–41

WSN MACs achieve energy-efficient low duty-cycle operation by dividing time into consecutiveshort active data exchange periods and relativelylong inactive periods. These low duty-cycle MACscan be divided into two categories: unsynchronizedand synchronized.

Unsynchronized low duty-cycle MACs utilizefrequent channel sampling, as presented in Fig. 1.Transmissions may be random and on-demandbasis. Each transmitted frame typically begins witha preample (Tpreample) lasting longer than the chan-nel sampling interval (Tinterval) that is typically set totens of milliseconds [16]. Hence, the periodic chan-nel sampling can detect the preample of each trans-mission prior to the actual data transmission. Afterthe detection of a preample, a node simply listens onthe channel until data are received. UnsynchronizedMACs are relatively simple and require less memorycompared to synchronized approaches. Commonfeature to all of these is that the energy consumptionis directed to the transmitting node. Thus, it is moreexpensive for a node to transmit a data frame thanto receive one. Because of the frequent channel sam-pling and long preample transmission prior to eachdata frame, unsynchronized MACs cannot reachthe energy efficiency of the lowest power synchro-nized MACs in low data-rate applications. Themost common unsynchronized MACs designed for

Channel sampling for checking channel activity

c

Preample

Tinterval Tpreample

Clear channel assessment

Source

Destination

Rx

Tx

Rx

Tx

Unsynchronized:

Twakeup_period

Source

Destination Rx

Tx

Rx

Tx

Synchronized:

Tactive

Tsleep

Beacon transmission / reception

c c

c

Data Data

Data

Fig. 1. Operation principle of unsynchronized (B-MAC) andsynchronized (IEEE 802.15.4) low duty-cycle MAC protocols.

WSNs are B-MAC [17], SpeckMACB/D [16], andWiseMAC [18].

Synchronized low duty-cycle MACs utilize aperiodic wakeup scheme, which divides time into asleep period (Tsleep) and active period (Tactive)repeated at Twakeup_period intervals, as presented inFig. 1. The active period is denoted as a superframe,where a node first broadcasts a beacon frame, andthen listens on the RF channel for possible incom-ing data during a short channel access period. Abeacon is a periodic broadcast of network signaling,which contains a node address, synchronizationinformation with the utilized RF channel and asuperframe schedule, and other network manage-ment information. Upon receiving one beacon, anode learns the exact timing of upcoming beaconsand active periods, and can then form synchronizedlinks to neighbors. Ideally, beacons are receivedwithout idle listening. In practice, slight idle listen-ing is caused by a synchronization granularity anda crystal inaccuracy clocking wake-up timers. Formaximizing energy efficiency, data frames areexchanged time-accurately in pre-determined timeslots or a short contention access period right afterthe beacon.

Current synchronized low duty-cycle MACs aremost suitable for applications where data are routedcontinuously from a static sensor field to sink nodes.Since sleep periods typically last several seconds,routing delay is higher than in unsynchronizedMACs. Another weakness of synchronized lowduty-cycle MACs is the energy consumption ofneighbor discovery procedures, which are requiredfor establishing synchronized communication links.In current proposals, neighbor discoveries are per-formed by network scans, which reduce networkenergy efficiency and data routing performance indynamic networks. In static networks, synchronizedMACs can achieve even one order of magnitudelower energy consumption than unsynchronizedMACs [16,19,20].

According to our experiments, synchronized lowduty-cycle MAC schemes can achieve the highestenergy efficiency also in dynamic networks, whenan efficient neighbor discovery protocol is used.This is proved in this paper. Hence, we focus on syn-chronized low duty-cycle MACs from now on.

2.2. Mobility support

Several studies have analyzed the effect of mobil-ity on WSN energy consumption [21] and data routing


performance [21,22]. These studies focus on routingand transport sublayers, while assuming an idealMAC. However, very few studies exist, which aimto improve MAC sublayer operation and energyefficiency on mobile WSN applications.

A mobility-aware sensor MAC (MS-MAC) [10]protocol adjusts the interval of a network scansaccording to observed node mobility. If no mobilityis observed, a 10 s long neighbor discovery isrepeated every 5 min. When a node detects mobilityby a change in the radio signal strength, a neighbordiscovery is performed instantly reducing the dis-connection time after actual link breaks. In dynamicnetworks, neighbor discovery is performed at mosttwice a minute. A major disadvantage of the proto-col is the high energy consumption of mobile nodesdue to the long and frequent network scans.

A study presented in [23] proposes the adjust-ment of frame sizes according to a predicted biterror rate. The prediction utilizes an Extended Kal-man Filter (EKF), which deals with the time-vary-ing wireless channel quality and the effect of framesize on network performance. A smaller frame sizeis selected when the signal characteristics are poorreducing frame error rate and retransmissions. Lar-ger frames are used when the quality of the channelimproves increasing obtained bandwidth. Accord-ing to simulations, EKF improves the energy-effi-ciency of a synchronized low duty-cycle MACprotocol about 24% at 50–75 mph node speed.

A traffic adaptive medium access (TRAMA) [24]MAC proposal utilizes three assisting protocolsfor signaling neighborhood information: a neighborprotocol (NP), a schedule exchange protocol (SEP),and an adaptive election algorithm (AEA). NPgathers a list of node identities from a two-hopneighborhood utilizing random access periods. Thisinformation is used for eliminating hidden node col-lisions, since only one transmitter per a two-hopneighborhood is selected to transmit at a time.SEP is used for signaling the set of receivers forthe traffic originating at the node. This provides effi-cient unicast, multicast and broadcast transmis-sions. AEA selects transmitters and receivers toachieve collision free transmission, and to optimizechannel utilization and energy efficiency using theinformation obtained by NP and SEP.

TRAMA updates the two-hop neighborhoodinformation periodically by exchanging special sig-naling frames with all neighbors during long-termrandom access periods. Since an entire networkenters the random access period simultaneously sus-

pending data routing and requiring a constantchannel reception, the performance of TRAMAdecreases rapidly with increased network dynamics.

The IEEE 802.15.4 low-rate wireless personalarea network (LR-WPAN) [13] is a multi-purposestandard specifying PHY and MAC layers. The Zig-Bee Alliance [25] builds on this foundation, by pro-viding a network layer and a framework for theapplication layer. LR-WPAN has an option forbeacon-enabled mode, which together with a clus-ter-tree network topology option enables a synchro-nized low duty-cycle scheme with high energyefficiency [19].

LR-WPAN beacons contain a beacon payloadfield for carrying network signaling data. ZigBee rou-ters (cluster heads) utilize this payload for signalingtime offsets to their parent routers. Each router uti-lizes this information for selecting a non-overlappingtime slot for its superframe. This is required, since allclusters of a network operate in a single RF channel.In the low duty-cycle mode, neighbor discoveriesare performed by network scans, which suspenddata routing. Thus, the performance of the ZigBeedecreases rapidly, as network dynamics increases.

The operation of mobile nodes in a stationarysensor field is improved in a self-organizing mediumaccess control for sensor networks (SMACS) [12]by an eavesdrop-and-register (EAR) algorithm.SMACS assigns a locally unique time slot and fre-quency channel for each communication link. Forinviting new nodes to join the network, a fixed glo-bal signaling channel is used. Mobile nodes executethe EAR algorithm. They register stationary nodesin their range by receiving broadcast invitation mes-sages on the signaling channel. Yet, the algorithm isenergy inefficient, since mobile nodes must listen onthe channel nearly constantly.

The stationary nodes perform a neighbor discov-ery at semi-regular intervals by entering a randomaccess BOOTUP mode [26]. In the BOOTUP mode,a node broadcasts invitation messages on the signal-ing channel, and listens on the channel for possibleresponses and other invitation messages. Since theBOOTUP mode suspends data exchanges andrequires a nearly continuous radio reception, net-work performance reduces rapidly with increasednetwork dynamics.

Our earlier work with Tampere University ofTechnology Wireless Sensor Network (TUTWSN)[9] shows increased energy efficiency of a dynamicWSN by frequent beacon broadcasts on a fixednetwork signaling channel (network channel). As a

M

L

F

E

K

GH

J

DC

I

A

B

Ch: 12

Ch: 35

Ch: 56

Ch: 55

Ch: 16

Offset:+220 ms

Offset:+150 ms

Offset:+100 ms

Offset:+390 ms Offset:+450 ms

Offset:+120 ms

Ch: 21

Node I, ch55, neighbors[ch35,+150ms; ch12,+100ms]

Node L, ch21, neighbors[ch56,+450ms; ch16,+390ms]

Ch:07

Node M, ch07, neighbors[ch55,+220ms; ch21,+120ms]

Fig. 2. Connectivity and distributed neighbor information.Arrows indicate received beacons (k = 2).


communication link fails or a new node arrives in asensor field, a node can perform neighbor discoveryby listening to the network channel for one beaconinterval. With current TUTWSN prototypes, a net-work beacon interval equals to 500 ms, while theinterval between superframes is several seconds.The transmission of frequent beacons on the net-work channel reduces significantly the energy con-sumption in dynamic networks [14].

TUTWSN forms a multi-cluster-tree topology,where several cluster-tree structures are super-posi-tioned allowing efficient multi-path routing, and hightolerance against link failures. Each cluster consistsof a single cluster head (headnode) and several clus-ter members (subnodes, or child headnodes). Eachheadnode associates with several parents, which havedifferent routing performance metrics and providepaths to different sinks. Hence, the multi-cluster-treetopology can combine the advances of a well-orga-nized and energy-efficient cluster-tree topology withthe flexibility of a mesh topology.

3. Energy-efficient neighbor discovery protocol design

ENDP distributes neighborhood informationand uses the information for establishing new com-munication links. The protocol permits mobility forall nodes without significant increase of energy con-sumption or performance degradation. ENDP con-sists of three parts: proactive distribution of neighbor

information, neighbor discovery algorithm for search-ing new neighbors, and energy-efficient network scan

facilitated by additional beacon transmissions. Theformer two parts minimize the need for networkscans, while the third part reduces the energy neededby each scan.

3.1. Proactive distribution of neighbor information

ENDP operates above a synchronized low duty-cycle MAC and utilizes the payloads of existingMAC beacons for distributing neighbor informa-tion. ENDP supports both clustered and non-clus-tered network topologies. In clustered networks,cluster heads distribute the schedules of other clus-ter heads by transmitting beacons. Yet, all nodescan receive the beacons and exploit the neighbordiscovery algorithm and the energy-efficient net-work scan. In non-clustered networks, all nodestransmit beacons, distribute the schedule of anyother node, and execute ENDP similarly. In bothcases, the operation principle of ENDP is similar.

The utilization of beacons for distributing neigh-bor information has several benefits. First, beaconexchanges are synchronized by a MAC protocol.This provides very low idle listening and energyoverhead due to additional control signalexchanges. Second, beacons are quite short, whichprovides energy-efficient implementation. In addi-tion, there is no need for a new frame type. Mostimportantly, the beacons are transmitted in any casefor maintaining link synchronization.

The distributed neighbor information consists ofsynchronization information about all parents withwhich synchronization is currently maintained.Complete synchronization information referring toone node is called a synchronization data unit(SDU). SDU consists of two parts: the channel onwhich the node is operating, and the time base dif-ference (offset) between the node and the sender ofSDU. The latter is used to avoid the need for a glo-bal time and to reduce the required value range,thus fitting the schedule information into fewer bits.

The distribution of SDUs is presented in Fig. 2.Nodes I and L are parents to node M. Thus, nodeM receives beacon transmissions from nodes I andL. Similarly, node M is a parent to node K, whichreceives beacons from M. We assume that even ifit would be physically possible for M to receive bea-con transmissions from J, the node should onlymaintain synchronization with a maximum of k

parents, which in this case equals to two.Each node that transmits beacons derives SDUs

from its parents and piggybacks them in a beaconpayload. In this case, node I transmit the SDUs


about nodes A and B. Similarly, L transmits SDUsabout nodes H and G. Due to the beacon receptionsfrom nodes I and L, M has accumulated SDUsabout nodes A, B, G and H.

Next, we look at the processing of SDUs thatnode K receives from node M. Clearly, it would bepossible that node M sends all SDUs it knows.Yet, the number of nodes from which SDUs accu-mulates will rise to a power on each step towardsfurther nodes. At the same time, the inaccuracy ofoffset values cumulates degrading the reliability ofthe protocol. Thus, it is feasible to send SDUs refer-ring to direct parents only, and the beacon transmis-sion of node M contains SDUs referring to nodes I

and L. Node K determines the beacon time tbeaconI of

node I by employing the SDU sent by neighbor M

as

tbeaconI ¼ tbeacon

M þ toffsetSDU ; ð1Þ

where tbeaconM is the known beacon transmission time

of node M and toffsetSDU is the offset stored in the SDU.

Since beacons are broadcast, some of the SDUsnode M receives from I and L may be duplicatesor refer to already known nodes. However, these sit-uations can be detected, since SDU determines thefrequency and time moment unambiguously. Uponreceiving a new SDU, the timing is comparedagainst other known neighbors. Because beaconsare sent periodically, two beacon times (t1, t2) arethe same, if

9i; i 2 Z : t1 t2 þ i tAC < d; ð2Þ

where tAC is the access cycle length and d denotesthe inaccuracy of the offset value. The SDU is ig-nored, if a neighbor having the same beacon timeand channel exists. Otherwise, the neighbor infor-mation is updated. As neighbors may move out ofcommunication range, stored information is period-ically checked and references to entries that are notrecently updated are removed.

Since a SDU consist of only few bytes of data,one beacon can contain several SDUs causing onlya minimal energy overhead. By receiving one bea-con from each parent, a node gets valid neighbor-hood information in two-hop radius rapidly andenergy-efficiently.

3.2. Neighbor discovery algorithm

The neighbor discovery algorithm operates ontwo main principles: maintaining redundant com-munication links and using the distributed informa-

tion for energy-efficient neighbor discovery. Forassuring a continuous data routing in a dynamicnetwork, adequate redundancy of communicationlinks is essential. The use of several parallel commu-nication links is ideal for multi-cluster-tree networktopology, like TUTWSN. In the presented design,up to k communication links are maintained. Thevalue of k depends on the frequency of link changesand thus, the physical network topology and thedegree of network dynamics. Proper k-values areexperimentally evaluated later in this paper.

The operation of the neighbor discovery algo-rithm is presented as a pseudocode as follows. Thealgorithm employs neighbor lists that are updatedon received beacons and SDU information. N* is alist of all known neighbors, whereas NS containsneighbors from which periodic beacons are received.The algorithm requires data memory for storingtwo-hop neighborhood information. As the numberof neighbors is at most k, the total number of neigh-bor entries is upper bounded to k + k2. The requiredprogram memory space for the algorithm is typi-cally less than 1 kbytes.

Symbols:

k: preferred number of nodes that synchroniza-tion is maintained to N*: list of all known neighbors NS: list of synchronized neighbor nodes, NS N*

q0: limit for minimum link quality required forreception q+: limit for good link quality rssi(i): link quality to the node i

source(i): set of nodes that have advertised node i,source(i) NS

ts: network scan timer

NEIGHBOR_DISCOVERY()

1: if NS = £ then

2: if timer tS has expired then

3: perform network scan until high RSSI

value or k nodes are found

4: NS nodes found during the scan5: reset timer tS

6: end if

7: end if

8: if |NS| < k then

9: N = £

10: for all i 2 (N* NS) do

11: receive a beacon from i

12: if reception was successful then


13: update rssi(i) based on the beacon

reception

14: if rssi(i) P q+ then

15: NS NS [ i16: if |NS| P k then

17: return

18: end if

19: else if rssi(i) P q0 then

20: N N [ i21: end if

22: end if

23: end for24: NS NS [ select k |NS| nodes from N that

have the highest RSSI

25: end if

Initially, a node does not know any neighborsand a network scan is required (rows 1–7). For min-imizing the network scan duration, the scan is fin-ished when either a neighbor with high RSSI or kneighbors are found. Since a neighbor with a highlink quality has also a high probability of advertis-ing other neighbors within nodes communicationrange, the rest of the k nodes are searched bySDUs.

The network scan utilizes second resolution timerts to prevent constant scanning. For saving energy,ts is gradually increased if no neighbors are found.For fast neighbor discovery, the timer is set toexpire immediately when all parents are lost. Asbeacons provide enough information for synchroni-zation, network scans are rare. Thus, the selectionof timeout value ts is not critical in normal opera-tion, when suitable neighbors in a range exist.

SDUs are utilized for obtaining connectivity upto k neighbors (rows 8–25). The algorithm iteratesthrough all known neighbors (N*) and attempts toreceive beacons from the neighbors, which the nodeis not already synchronized with. If the RSSI valueof received beacon is equal to or higher than q+, thenode is added to the list NS and synchronizationwith the node is maintained. Otherwise, if the signalstrength is adequate for communication, the neigh-bor information is cached and the algorithm contin-ues for discovering a high quality link.

If the number of detected high quality links is notadequate, the rest of the links are selected amongthe cached neighbor information by selecting thehighest RSSI values (row 24). When several choiceswith near equal link quality are available, a set ofneighbors having together the highest number ofunique SDUs is selected. This creates comprehen-

sive neighborhood information improving protocolperformance and network robustness.

As beacon schedules are exactly known, synchro-nization attempts are made by short beacon recep-tion periods. Thus, the neighbor detection usingSDU information has an additional benefit overthe traditional network scan, as a node can sleepor communicate with other neighbors while discov-ering its neighborhood. As all nodes exploit theneighbor discovery algorithm similarly, randommobility for all nodes is allowed. The maximumdegree of mobility is mostly limited by a routingprotocol.

3.3. Energy-efficient network scan

Neighbor discovery by a network scan isunavoidable at a start-up, when waking up from along-term sleep, or when all links are lost. For per-forming the scan efficiently, the frequency channelsand transmission interval of beacons must be pre-determined. If nodes operate on several frequencychannels, the neighbor discovery may be appliedon each possible frequency, as in IEEE 802.15.4.A drawback is that the time and energy consumedby the scan is multiplied by the number of scannedchannels. Another alternative is to use a network-wide fixed channel for neighbor discovery. In bothcases the duration and energy of a network scanreduces by minimizing beacon transmission interval.As beacon transmissions also consume power, anoptimal beacon interval is achieved when the sumof beacon transmission and network scan energyconsumptions is minimized [14].

As compared in [27], narrow band and high data-rate transceivers operating on the 2.4 GHz Indus-trial, Scientific and Medical (ISM) band are themost suitable for WSNs due to their low energy con-sumption per a transmitter or received bit of PHYlayer data. With a typical 1 MHz bandwidth, thetheoretical maximum number of separate channelsis 83. According to our experiments with NordicSemiconductor nRF24L01 and nRF2401A trans-ceivers, the number of non-interfering (orthogonal)channels is 21. In addition, the 2.4 GHz spread spec-trum transceiver of IEEE 802.15.4 has 16 channelsand several orthogonal channels.

If beacons share the same frequency channel withdata transmissions, while beacon transmission rateis high, the probability of collisions with dataframes is significant. For minimizing idle listening,the transmission times of beacons should be


accurate. Thus, collision avoidance by channel sens-ing prior to beacon transmissions is impractical.Due to the above, the use of a separate global net-work channel for network scan is feasible and wellreasoned.

In the following example implementation with amulti-cluster tree topology, two types of beaconsare utilized: cluster beacons on a cluster channeland network beacons on a network channel. As pre-sented in Fig. 3, network beacons are broadcast byall cluster heads during inactive periods. This is eas-ily managed by a single radio transceiver, since itcan be tuned to a different frequency channel veryrapidly. The network scan is performed by receivingthe network beacons on the network channel. Themaximum network scan duration that is requiredfor receiving a network beacon from all neighboringcluster heads equals to one network beacon interval.

Due to the dedicated network channel, the net-work beacons can collide with other network bea-cons only. However, we can assume that thetransmission interval of network beacons is severalorders of magnitude longer than the length of a sin-gle network beacon transmission. Thus, the collisionprobability of network beacons can be consideredvery low. In addition, we can assume that an ade-quate number of SDUs can be obtained by a net-work scan even though some collisions happen.Hence, the effect of occasional beacon collisionson network performance is minimal. Thus, the net-work beacons are transmitted without collision

time

time

Superframes

time

time

time

Reception time forneighbor discovery

Network beacons of:

Twakeup_period

Network beacon

Networkchannel

Clusterhead D (Ch: 4)

Clusterhead C (Ch: 3)

Clusterhead B (Ch: 2)

Clusterhead A (Ch: 1)

Clusterhead A:

Clusterhead B:

Clusterhead C:

Clusterhead D:

tpre


Fig. 3. Principle of frequent network beacons.

avoidance. This improves the achieved energy-effi-ciency, since energy consumption of beacon trans-missions is minimized.

For improving scalability and reducing colli-sions, the presented implementation utilizes severalcluster channels such that superframes are dividedby time and frequency. In practice, each cluster uti-lizes a locally unique frequency and time slot for itssuperframe. The frequency and time slot areselected and dynamically adjusted according to theneighborhood information obtained from networkscans and ENDP protocol.

The network and cluster beacons may be similar,but the utilization of different frame types allowsshorter beacons and more energy-efficient imple-mentation. While cluster beacons contain vitalinformation for data exchanges in superframes, net-work beacons contain node status information forthe selection of an appropriate neighbor for associ-ation. The transmission of network beacons can beimplemented as an extension in a MAC protocol.The network scan with the network channel exten-sion is implemented simply by configuring the fre-quency channel and scan duration parameters.

Network beacons perform best, when their trans-mission rate is uniform and globally known in entirenetwork. In the current implementation, the trans-mission rate of network beacons is determinedbefore deployment according to statistically esti-mated link lifetimes. However, a more efficientoption is to adjust the network beacon rate dynam-ically. We propose that each node observes andmaintains a record of an average link lifetime. Themaintained values are transmitted to a sink period-ically, or on request. According to the gathered linklifetimes, the sink determines an energy optimal net-work beacon rate for the network and floods theresulted rate through the network in a controlframe.

Although the utilization of a network channel isnot necessitated by ENDP, we discuss shortly itseffects on ENDP. For improving the efficiency ofnetwork scans, it is recommended to transfer SDUsin network beacons. Then, each node can detect allneighbors and their parents by the short networkscan.

Generally, nodes maintain synchronization withtheir parents by receiving cluster beacons. Forupdating SDU information, nodes should occasion-ally receive also network beacons from theirparents. The reception of network beacons can beperformed energy-efficiently, if the transmission


times of at least some of the beacons are known. Inour implementation, one network beacon is alwaystransmitted exactly tpre before each cluster beacon.Hence, a network beacon can be received with aminimum idle listening and the complexity of syn-chronization does not increase. If the beacon recep-tion fails, due to a collision or other reason, thereception is retried prior to the next cluster beacon.The transmission moments of other network bea-cons are coarsely defined and they always have alower priority than data exchanges. An optimalinterval of network beacon receptions is a tradeoff between energy consumption of network beaconreceptions and the freshness of SDU information.

4. TUTWSN prototype platform

TUTWSN prototype is a development platformdesigned for testing and analyzing the operation ofreal WSN applications. In this paper, the platformsare utilized for experimentally measuring the perfor-mance of ENDP. In addition, measured staticpower consumptions of the platform are used asan input for a performance analysis.

The TUTWSN physical platform contains aMicrochip PIC18LF4620 MicroController Unit(MCU), which integrates an 8-bit processor corewith 64 kbytes FLASH program memory, 4 kbytesRAM data memory, and 1 kbyte EEPROM. Theclock frequency is 4 MHz resulting in 1 MIPSperformance.

For wireless communication, the platform hasa Nordic Semiconductor nRF2401A 2.4 GHzradio with selectable 250 kbps or 1 Mbps transmis-sion data-rate and 83 available frequency channels.

Battery

M

Temperaturesensor

Voltage re

Push button and LED

Fig. 4. TUTWSN pro

Transmission power level is selectable between 20and 0 dBm with four steps. The radio has an inter-face for low speed MCU, which consists of 32 bytesdata buffers for transmission and reception, andintegrated logic for address recognition and cyclingredundancy check (CRC) error detection. Thesesimplify data processing in MCU and allow lowspeed data exchange between MCU and radio.The radio does not provide RSSI functionality. Aloop antenna is implemented by a PCB trace, whichhas a dipole radiation pattern and around +4 dBidirectivity (simulated).

The prototype is powered by a 3.0 V and1600 mAh CR123A lithium battery. The batteryvoltage is converted to 2.25 V supply voltage by aMAX1725 linear regulator.

The implemented TUTWSN prototype is pre-sented in Fig. 4. The prototype is 124 mm 21 mmsized, and encapsulated in a waterproof plasticenclosure. The prototype consists of two separateboards, one for MCU, radio, voltage regulation,and temperature sensor, and the other for battery,push button, LED, and I/O connector.

4.1. Prototype power analysis

For the performance analysis, static power con-sumptions of the implemented platform are mea-sured at utilized operation modes. The results arepresented in Table 1. The supply voltage used inthe measurements is 3.0 V. The minimum achiev-able power consumption is 37 lW, when the wholesystem is in a sleep mode. The power consumptionincreases to 3.17 mW, when MCU is in active modeand running at 1 MIPS speed. The transmission

AntennaTransceiver

CU

gulation

totype platform.

Table 2Symbol descriptions and defined values

Symbol Description Defined value

d Inaccuracy of SDU time offset 250 lsd Node density 0.1 nodes/m2

e Crystal tolerance 20 ppmfC Cluster beacon transmission rate 0.5 HzfN Network beacon transmission rate 0.01–100 HzfU Network beacon reception rate 0.5 Hzk Number of nodes with which

synchronization is maintained1–4

LF Frame length 256 bitsn Nodes within a radio range Variabler Radio range 10 mq Range of sufficient signal strength

compared to maximum radio range0.5

R Radio data-rate 1 MbpstI Synchronization inaccuracy 50 lstST Transmitter and receiver start-up time 200 ls

Table 1Measured TUTWSN prototype power consumptions

Symbol MCU Transceiver Power (mW)

PRX 1 MIPS RX 60.171 MIPS TX (0 dBm) 42.17

PTX 1 MIPS TX (6 dBm) 34.671 MIPS TX (12 dBm) 31.371 MIPS TX (20 dBm) 29.571 MIPS Stand-by 3.291 MIPS Sleep 3.17Sleep Sleep 0.037


(TX) of a frame with active MCU consumes thepower between 29.57 and 42.17 mW, depending onthe transmission power level (20 to 0 dBm). Inthe following analysis, we approximate that theaverage transmission power level is 6 dBm, whichconsumes 34.67 mW power (PTX). The highestpower (PRX) of 60.17 mW is consumed in receptionmode (RX). Depending on the operation mode, thetransceiver consumes 10 and 20 times the power ofMCU in active mode.

4.2. Radio energy models

For analyzing the performance of ENDP, we firstmodel the behavior of the utilized platform. Radioenergy models define the energy consumptions ofrequired transceiver operations, including a frametransmission and reception, and a network scan.In addition, the non-idealities of PHY layer are con-sidered incorporating radio start-up time (tST), crys-tal tolerance (e), and synchronization inaccuracy(tI). All essential phases of the radio operationsare identified by oscilloscope measurements. In thecurrent implementation, the time base error duesynchronization inaccuracy and time drift is typi-cally within ±50 ls. When attempting to receive abeacon from a parent’s parent according to SDUinformation, inaccuracy is higher due to two sepa-rate synchronizations and higher time drift. Accord-ing to the measurements, the offset inaccuracy (d)over two hops is typically within ±250 ls.

While the frame formats vary according to proce-dures, the radio frame in TUTWSN has a fixedlength of 256 bits, which is the maximum data buffersize of the radio. In the following models, weassume that the active times of MCU and radioare equal. Hence, the MCU power is included inthe power consumptions of transmission and recep-tion modes. The symbols with defined values arepresented in Table 2.

A frame transmission consists of a radio start-uptransient and the actual data transmission defined asthe ratio of frame length (LF) and radio data-rate(R). During a start-up transient, the radio powerconsumption is equal to the transmission modepower (PTX). Thus, the frame transmission energyETX is modeled as

ETX ¼ tST þLF

R

P TX: ð3Þ

A frame reception begins with the radio start-uptransient. The radio consumes the reception modepower (PRX) until a frame has been completelyreceived. An average idle listening time caused bysynchronization inaccuracy and crystal tolerance isincluded. The frame reception energy ERX is

ERX ¼ tST þ tI þ2efC

þ LF

R

P RX: ð4Þ

The energy consumption of a beacon receptionaccording to SDU information is modeled sepa-rately due to a higher synchronization error. Anaverage reception time includes radio start-up tran-sient, time offset inaccuracy (d), and the framelength. Thus, the beacon reception energy utilizingSDU information ESDU

RX is

ESDURX ¼ tST þ dþ LF

R

P RX: ð5Þ


Since some of the two-hop neighbors signaled inSDUs are out of the nodes range, beacon receptionmay result in a failure. Due to the time offset inac-curacy, the reception margins are wide and theireffect on network energy consumption may be sig-nificant. Since both positive and negative time offseterrors are considered, the energy consumption of anunsuccessful beacon reception based on SDU infor-mation ESDU

RXU is

ESDURXU ¼ tST þ 2dþ LF

R

P RX: ð6Þ

A network scan begins with a radio start-up tran-sient and continues by a channel reception until tNS

is elapsed. Thus, the network scan energy ENS canbe modeled as

ENS ¼ ðtST þ tNSÞP RX: ð7Þ

Because the energy consumption of individual bea-con receptions is negligibly compared to the net-work scan energy, they are ignored in the model.

5. Performance analysis

For making the analysis of ENDP independentof data exchanges, we divide the energy consumedin a wireless sensor node in three classes [14]: node

start-up, network maintenance, and data exchangeenergies as presented in Fig. 5. The node start-upenergy consists of neighbor discovery and networkassociation operations. The network maintenanceand data exchange operations are executed afterthe start-up period during the node lifetime. Thenetwork maintenance energy consists of beacontransmissions and receptions (beacon exchange),network scans, and possible re-associations. Thedata exchange energy is consumed by payload datatransmissions and receptions, and the MAC signal-

Star

t-up

ener

gy:

• Net

wor

k sc

an•N

etw

ork

asso

ciat

ion Data exchange energy:

• Upper layer payload data andacknowledgement exchange

Network maintenance energy:• Beacon transmissions and receptions

• Network scans

Node lifetimeStart-upperiod

time

Fig. 5. WSN node energy classification.

ing frames related to data transmissions, such asacknowledgments.

As the lifetime of nodes is expected to be frommonths to years, the start-up energy is negligiblecompared to the total node energy consumption.Furthermore, we assume that data exchange opera-tions are not affected by network maintenance oper-ations. Hence, from now on we focus on thenetwork maintenance operations.

In the following analysis, we first model the per-formance of ENDP, and then energy optimize thetransmission rate of network beacons. For simplic-ity, we analyze the energy consumption of a singlemobile node moving in a stationary sensor field.In addition, a non-clustered network, where allnodes transmit beacons is considered. Energy con-sumption is determined over 1 s periods of opera-tion, which equals to average power consumption.The evaluation of average power instead of energyis more convenient due to the time independency.

5.1. Network scan interval

Assuming a uniform node distribution, let nodedensity be d nodes/m2. This results that n = dpr2

nodes are located in a radio range (r). Let us con-sider a node that maintains synchronization andreceives SDUs from k neighboring nodes. As a nodemoves in a stationary WSN field at speed v, a com-munication link failure occurs when any of the k

links fails. Hence, a link failure rate (fF) is

fF ¼kvr: ð8Þ

Next, we model the probability of a successfulneighbor discovery by utilizing SDUs. Let us con-sider a situation presented in Fig. 6, where the nodeA maintains synchronization and receives SDUsfrom nodes B and D. The distance between nodesA and B is b, and their radio ranges form circleswith a radius r. Moreover, node B maintains

SA

A B

CD E

SA B SB

r rb

Fig. 6. Intersections of the communication ranges of nodes Aand B. Arrows show the direction of synchronization.


synchronization with nodes C and E located in itsrange. Since the node E is in the intersection areaof the ranges of A and B (SA\B), A can receive its(beacon) transmissions and the signaled SDU isvalid. Since the node C is outside the area SA\B,node A cannot detect its transmissions resulting inan invalid SDU. The size of the intersection areaSA\B is calculated by the radius r and the distanceb as [28]

AA\BðbÞ ¼ 4

Z r

b=2

ffiffiffiffiffiffiffiffiffiffiffiffiffiffir2 x2p

dx: ð9Þ

Thus, the probability that a node located in SB isalso in the intersection area SA\B equals toAA\B(b)pr2. Moreover, let nodes A and B be ran-domly located neighbors, where b gets a value inthe range [0, r], and A receives SDUs from B. Theprobability pI that a received SDU is valid is deter-mined by integrating the probability AA\B(b)pr2

over the circle of radius b centered at A for b in[0, r] we obtain [28]:

pI ¼Z r

0

2bAA\BðbÞpr4

db 59%: ð10Þ

Each node maintains synchronization with k

other nodes, which all generate k SDUs. We modelthe probability (q) that all of the received k2 SDUsare invalid and a network scan is required. Clearly,the probability of invalid SDU is (1 PI) on thefirst attempt. Then, on each attempt (a) from twoto k, the number of nodes outside the intersectionarea SA\B is reduced by one. The result is raisedto the power of k according to the number of nodesfrom which SDUs are received. Thus, q is

q ¼Yk

a¼1

1 npI

n ða 1Þ

k

: ð11Þ

Finally, the required network scan interval (INS) is

INS ¼1

fFq: ð12Þ

5.2. Number of beacon receptions

After a link failure, a node attempts to receivebeacons according to SDUs until a new neighborwith sufficient signal strength is detected, or all(k2) SDUs are examined. We define that the rangeof sufficient signal strength in proportion to maxi-mum radio range (r) is q. The probability of findinga suitable neighbor on a first attempt is pIq

2. If the

attempt fails, the number of nodes outside the areaof sufficient signal strength is decreased. The proba-bilities of finding a suitable neighbor using a

attempts, ranging from one to k2 are determined.Then, the expected number (u) of beacon receptionattempts is modeled by a weighted average as

u ¼ pIq2 þ

Xk21

a¼2

aYa1

b¼1

1 npIq2

n ðb 1Þ

!npIq

2

n ða 1Þ

þ k2Yk21

a¼1

1 npIq2

n ða 1Þ

!: ð13Þ

Since a dense network is assumed, duplicate SDUsand SDUs referring to a current parent are not con-sidered in the model.

If no new neighbors are detected by SDUs, a net-work scan is performed. Similarly, the network scanis performed until a node within a range of qr isdetected. The number of beacon receptions (nB)until a beacon with sufficient signal strength isreceived is

nB ¼ q2 þXn1

a¼2

aYa1

b¼1

1 nq2

n ðb 1Þ

!nq2

n ða 1Þ

þ nYn1

a¼1

1 nq2

n ða 1Þ

!: ð14Þ

Finally, the required network scan duration (tNS)is

tNS ¼nB

fNn: ð15Þ

The resulted network scan duration is presented inFig. 7, while the interval of network beacon trans-missions varies. If the number of nodes within radiorange (n) is high, tNS is reduced even one order ofmagnitude compared to the entire network beaconinterval.

5.3. Network maintenance power consumption

A network maintenance power PM is defined asthe sum of a network scan power PNS and a beaconexchange power PB. The power consumed by net-work scans PNS is defined as the energy of a singlescan procedure ENS divided by the average intervalof scans INS. The long-time average of network scanpower is obtained by

P NS ¼ fFq tST þnB

fNn

P RX: ð16Þ

0.01

0.1

1

10

100

0.01 0.1 1 10 100Network beacon transmission rate (Hz)

Net

wor

k m

aint

enan

ce p

ower

(mW

)

Conventional scanWithout ENDPENDP, k=1ENDP, k=2ENDP, k=3ENDP, k=4

v = 0.1 m/s

Fig. 8. Network maintenance power at 0.1 m/s node velocity, ask varies.

0.01

0.1

1

10

100


Net

wor

k m

aint

enan

ce p

ower

(mW

)

Conventional scanWithout ENDPENDP, k=1ENDP, k=2ENDP, k=3ENDP, k=4v = 10 m/s

Fig. 9. Network maintenance power at 10 m/s node velocity, as k

varies.

0.001

0.01

0.1

1

10

100


Req

uire

d ne

twor

k sc

an ti

me

(s) d=0.013 nodes/m^2

d=0.025 nodes/m^2d=0.064 nodes/m^2d=0.32 nodes/m^2

d=0.013 nodes/m2

d=0.025 nodes/m2

d=0.064 nodes/m2

d=0.32 nodes/m2

Fig. 7. The effect of node density and network beacon transmis-sion rate to the required network scan time.


The transmission rates of cluster and networkbeacon are fC and fN, respectively. Cluster beaconsare received at the beginning of superframes at ratefC from k neighbors with which synchronization ismaintained. For updating two-hop neighborhoodinformation, network beacons carrying SDUs arereceived at rate fU from k neighbors. Moreover, linkfailures occur at rate fF after which averagely u clus-ter beacons are received employing SDU informa-tion. The average power consumed by the beaconexchange is

P B ¼ ðfN þ fCÞETX þ kðfC þ fUÞERX

þ fF ðu 1ÞESDURXU þ ESDU

RX

: ð17Þ

For ensuring valid SDU information also invery dynamic networks, fU is fixed to fC. Thenetwork is configured with the parameter valuesdefined in Table 2. First, the network maintenancepower is analyzed as the function of networkbeacon rate. The results with 0.1 and 10 m/s nodevelocities are presented in Figs. 8 and 9. Asconventional network scans without the networkbeacon signaling scheme are employed, the net-work maintenance power consumptions are 1.23and 60.17 mW, respectively. By adding the net-work beacons, the minimum network maintenancepowers are achieved at 2.5 and 18 Hz networkbeacon rates resulting in 95 and 742 lW powerconsumptions, respectively. The lowest power con-sumption is achieved by combining networkbeacons with ENDP. The resulted minimumpower consumptions at 0.1 and 10 m/s node veloc-

ities are 87 lW (k = 1) and 397 lW (k = 2),respectively. At these settings, the energy savingcompared to conventional network scans is 93–99%.

For evaluating the effect of synchronization erroron ENDP performance, network maintenancepower is analyzed, while the SDU offset inaccuracy(d) varies from 50 to 1000 ls. In this analysis, k isfixed to two. First, v is set to 0.1 m/s and the resultsare plotted in Fig. 10. An energy optimal networkbeacon rate is around 0.43 Hz, when the networkmaintenance power is between 94 and 98 lW. Thus,the effect of synchronization inaccuracy is only 4%in rather static networks.

0.01

0.1

1

10

100


Net

wor

k m

aint

enan

ce p

ower

(mW

)

Without ENDPENDP, = 50 µsENDP, = 250 µsENDP, = 500 µsENDP, = 1000 µs

v = 0.1 m/s

Fig. 10. The effect of SDU time offset inaccuracy on the networkmaintenance power at 0.1 m/s node velocity.

0.01

0.1

1

10

100


Net

wor

k m

aint

enan

ce p

ower

(mW

)

Without ENDPENDP, = 50 µsENDP, = 250 µsENDP, = 500 µsENDP, = 1000 µs v = 10 m/s

Fig. 11. The effect of SDU time offset inaccuracy on the networkmaintenance power at 10 m/s node velocity.

0.001

0.01

0.1

1

10

100

0.01 0.1 1 10Node mobility (m/s)

Ener

gy o

ptim

al b

eaco

n ra

te (H

z)

Without ENDPENDP, k=1ENDP, k=2ENDP, k=3ENDP, k=4

Fig. 12. Energy optimal beacon transmission rate as the functionof node mobility.


Then, v is set to 10 m/s. Obtained results arepresented in Fig. 11. An energy optimal networkbeacon rate is 4.8 Hz. As d gets values 50, 250,500 and 1000 ls, network maintenance powers are333, 397, 478 and 640 lW, respectively. WithoutENDP, the power consumption is 742 lW. Theresults depict that ENDP is susceptible for synchro-nization errors in dynamic networks. However, evenwith the highest 1000 ls inaccuracy, the resultedpower consumptions are significantly lower thanwithout ENDP.

5.4. Energy optimal network beacon rate

As seen in the previous analysis, the networkmaintenance power depends significantly on the rate

of network beacon transmission. For obtaining thehighest energy-efficiency, the network beacon rateshould be energy optimized according to networkdynamics.

An energy optimal network beacon rate f N isdetermined by minimizing the network maintenancepower (PM) with respect to the beacon transmissionrate. It can be shown that there exists a unique min-imum for f N that is obtained by setting oPM/ofN = 0[14]. This yields:

f N ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffinBP RX

ETXINSn

r: ð18Þ

An optimal network beacon rate is determinedby beacon frame transmission energy (ETX), radiopower consumption in reception mode (PRX) andnumber of nodes in range (n). Furthermore, nB

and INS are functions of the range of sufficient signalstrength (q), node speed (v), radio range (r) and thenumber of nodes with which synchronization ismaintained (k).

The energy optimal beacon transmission rate isplotted as the function of node speed in Fig. 12.As seen in the figure, the optimal rate increases pro-portionally to the node speed. In addition, ENDPreduces the optimal beacon rate even three ordersof magnitude. Fig. 13 presents the energy optimalbeacon rate as a function of nodes in a range. Athigher node densities the optimal beacon transmis-sion rate decreases, since the required network scantime (tNS) reduces. In very sparse networks, theoptimal beacon rate also decreases, especially when

0.001

0.01

0.1

1

10

100

0 50 100 150 200Nodes in a range

Ener

gy o

ptim

al b

eaco

n ra

te (H

z) Without ENDP ENDP, k=1ENDP, k=2 ENDP, k=3ENDP, k=4

Fig. 13. Energy optimal beacon transmission rate as the functionof nodes in a range (v = 1 m/s).


k is 3 or 4. This is caused by fewer beacon receptions(u) until one beacon with sufficient signal strength isreceived. In addition, ENDP reduces the optimalbeacon rate up to four orders of magnitude. Theseresults indicate that ENDP can effectively reducethe need for network scans.

6. Experimental measurements

The performance of the implemented ENDP ismeasured experimentally with TUTWSN prototypeplatforms and TUTWSN protocol stack. As theprototype platform lacks the RSSI functionality, asimple path-loss metering mechanism is imple-mented by replacing each beacon transmission withtwo consecutive beacons transmitted at differenttransmission power levels. A path-loss is interpretedfrom successfully received beacons, as presented in[29]. Totally eight nodes are configured to head-nodes and deployed in an office environment, as

120 m

Stationary nodes Sink Mobility pattern

Fig. 14. Network topology for performance measurements.

presented in Fig. 14. One node is mobile in the sta-tionary sensor field as presented in the figure. Theradio range varies from 10 to 120 m depending onthe node position. The maximum range is achievedalong a long corridor, while walls attenuate RFpropagation very significantly.

The average power consumption is determinedby powering the mobile node by a 0.2 F (C) supercapacitor and measuring the slope of capacitor ter-minal voltage. In practice, the elapsed time (t) whencapacitor terminal voltage (U) decreases from 5.0 to4.0 V is measured. Since the prototype platforms areequipped with linear voltage regulators, the rela-tively high input voltage does not affect prototypecurrent consumption, except a negligible smallincrease of a quiescent current. Thus, average powerconsumption (P) is calculated as

P ¼ dUdt

CU 0; ð19Þ

where U0 is the nominal battery voltage (3.0 V) ofthe prototype platform. Thus, resulted power con-sumptions are applicable for lithium battery pow-ered nodes.

For the measurements, the mobile node is config-ured to a subnode operation. This minimizes thedata exchange power consumption, which is essen-tial when focusing to the power consumption of net-work maintenance operations. The mobile nodetransmits a 32 bytes diagnostics data frame on acontention-free time slot at 8 s intervals. In addi-tion, control frames for association and slot reserva-tion requests are transmitted after communicationlink changes. The transmission interval of networkbeacons is fixed to 500 ms, thus defining the maxi-mum duration of a single network scan.

The power consumption of the mobile node ismeasured while executing ENDP and varying k

between 2 and 4. For comparison, power consump-tion is also measured without ENDP, when newneighbors are searched by network scans. Measure-ments are performed with node velocities of 1 and3 m/s, and without mobility.

The results are presented in Fig. 15. The resultsprove that ENDP effectively reduces the powerconsumption of a mobile node. In this case, ENDPoperates most effectively when k is three. Then, theresulting energy saving is 57–80% compared tothe situation without ENDP. In contrast toanalysis, the network beacon rate is not adjustedand energy optimized according to network

0

1

2

3

4

5

Mobility: no Mobility: 1 m/s Mobility: 3 m/s

Mea

sure

d po

wer

(mW

)

without ENDPENDP, k = 2ENDP, k = 3ENDP, k = 4

Fig. 15. Measured power consumption as a function of nodemobility as k varies and without ENDP.


dynamics. This results in higher energy consump-tion of network scans. When all nodes are station-ary, ENDP increases node power consumption300–640 lW.

The experimental measurements result in higherpower consumptions than mathematical analysis.This is partly caused by the data exchanges andthe path-loss metering method substituting forRSSI. The method increases significantly the num-ber of unsuccessful beacon receptions. In addition,the energy consumptions of data processing andsensing are not included in the analysis. Moreover,the office environment caused significant fading,gaining and shadowing on RF propagation makinglink changes more frequent and protocol operationmore difficult than in an open space. However, theexperimental measurements clearly proved the per-formance of ENDP.

7. Conclusions

The design and implementation of a novel neigh-bor discovery protocol are presented. The proposedENDP significantly reduces the frequency of high-energy network scans and supports high nodemobility by effectively distributing synchronizationinformation over two-hop neighborhood in MACbeacon payloads. Furthermore, a unique beaconsignaling scheme is presented and optimized toallow energy-efficient network scans needed on astart-up and when ENDP based neighbor discoveryfails.

According to performance results, the beaconsignaling scheme achieves 92–99% energy savingcompared to conventional network scans. ENDPcan further reduce energy consumption up to 46%.In static networks, energy overhead is only slight.Experimental measurements indicate up to 80%energy saving compared to network scan basedneighbor discovery with the beacon signalingscheme. ENDP can operate above any synchronizedlow duty-cycle scheme, such as S-MAC and the bea-con-enabled mode of IEEE 802.15.4.

References

[1] M. Haenggi, Opportunities and challenges in wireless sensornetworks, in: M. Ilyas, I. Mahgoub (Eds.), Handbook ofSensor Networks: Compact Wireless and Wired SensingSystems, CRC Press, Florida, 2004, pp. 1.1–1.14.

[2] I.F. Akyildiz, W. Su, Y. Sankarasubramaniam, E. Cayirci,Wireless sensor networks: a survey, Elsevier Comput. Net-work 38 (4) (2002) 393–422.

[3] S. Roundy, P.K. Wright, J. Rabaey, A study of low-levelvibrations as a power source for wireless sensor networks,Comput. Commun. 26 (11) (2003) 1131–1144.

[4] J. Hill, M. Horton, R. Kling, L. Krishnamurthy, Theplatforms enabling wireless sensor networks, Commun.ACM 47 (6) (2004) 41–46.

[5] J. Hill, R. Szewczyk, A. Woo, S. Hollar, D.E. Culler, K.S.J.Pister, System architecture directions for networked sensors,in: Proceedings of the Ninth International Conference onArchitectural Support for Programming Languages andOperating Systems, 2000, pp. 93–104.

[6] H. Karl, A. Willig, Protocols and Architectures for WirelessSensor Networks, John Wiley & Sons Ltd., Chichester, 2005,pp. 1–13.

[7] C.R. Lin, M. Gerla, Adaptive clustering for mobile wirelessnetworks, IEEE J. Select. Areas Commun. 15 (1997) 1265–1275.

[8] J. Polastre, R. Szewczyk, A. Mainwaring, D. Culler,J. Anderson, Analysis of wireless sensor networks forhabitat monitoring, in: C.S. Raghavendra, K.M. Sivalingam,T. Znati (Eds.), Wireless Sensor Networks, Springer, NewYork, 2004, pp. 399–425.

[9] J. Suhonen, M. Kohvakka, M. Hannikainen,T.D. Hamalainen, Design, implementation, and outdoordeployment for wireless sensor network for environmentalmonitoring, in: Proceedings of the Embedded ComputerSystems: Architectures, Modeling, and Simulation, 2006,pp. 109–121.

[10] H. Pham, S. Jha, Addressing mobility in wireless sensormedia access control protocol, in: Proceedings of theIntelligent Sensors, Sensor Networks and Information Pro-cessing, 2004, pp. 113–118.

[11] W. Ye, J. Heidemann, D. Estrin, Medium access controlwith coordinated adaptive sleeping for wireless sensornetworks, IEEE/ACM Trans. Network 12 (3) (2004) 493–506.


[12] K. Sohrabi, J. Gao, V. Ailawadhi, G.J. Pottie, Protocols forself-organization of a wireless sensor network, IEEE Person.Commun. 7 (5) (2000) 16–27.

[13] IEEE Std 802.15.4-2003, Wireless Medium Access Control(MAC) and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal Area Networks (WPANs).

[14] M. Kohvakka, M. Hannikainen, T.D. Hamalainen, Energyoptimized beacon transmission rate in a wireless sensornetwork, in: Proceedings of the 16th Annual IEEE Interna-tional Symposium Personal Indoor and Mobile RadioCommunications, 2005, pp. 1269–1273.

[15] A. Woo, D.E. Culler, A transmission control scheme formedia access in sensor networks, in: Proceedings of theSeventh Annual International Conference Mobile Comput-ing and Networking, 2001, pp. 221–235.

[16] K.-J. Wong, D.K. Arvind, SpeckMAC: low power decen-tralized MAC protocols for low data-rate transmissions inspecknets, in: Proceedings of the Second InternationalWorkshop Multi-hop Ad Hoc Networks: From Theory toReality, 2006, pp. 71–78.

[17] J. Polastre, J. Hill, D. Culler, Versatile low power mediaaccess for wireless sensor networks, in: Proceedings of theSecond ACM Conference Embedded Networked SensorSystems, 2004, pp. 95–107.

[18] A. El-Hoiyi, J.-D. Decotignie, J. Hernandez, Low powerMAC protocols for infrastructure wireless sensor networks,in: Proceedings of the Fifth European Wireless Conference,2004, pp. 563–569.

[19] M. Kohvakka, M. Kuorilehto, M. Hannikainen, T.D.Hamalainen, Performance analysis of IEEE 802.15.4 andZigBee for large-scale wireless sensor network applications,in: Proceedings of the Third ACM International WorkshopPerformance Evaluation of Wireless Ad Hoc, Sensor, andUbiquitous Networks, 2006, pp. 48–57.

[20] M. Kohvakka, M. Hannikainen, T.D. Hamalainen, Ultralow energy wireless temperature sensor network implemen-tation, in: Proceedings of the 16th Annual IEEE Interna-tional Symposium Personal Indoor and Mobile RadioCommunications, 2005, pp. 801–805.

[21] A.A. Somasundara, A. Kansal, D.D. Jea, D. Estrin, M.B.Srivastava, Controllably mobile infrastructure for lowenergy embedded networks, IEEE Trans. Mobile Comput.5 (8) (2006) 958–973.

[22] M. Grossglauser, D.N.C. Tse, Mobility increases the capac-ity of ad hoc wireless networks, IEEE/ACM Trans. Network10 (4) (2002) 477–486.

[23] P. Raviraj, H. Sharif, M. Hempel, C. Song, H.H. Ali, J.Youn, A mobility based link layer approach for mobilewireless sensor networks, in: Proceedings of the IEEEInternational Conference Electro Information Technology,2005, 6 pp.

[24] V. Rajendran, K. Obraczka, J.J. Garcia-Luna-Aceves,Energy-efficient, collision-free medium access control forwireless sensor networks, Wireless Network 12 (1) (2006) 63–78.

[25] ZigBee Alliance Document 053474r06; ZigBee Specification,Version 1.0, December 2004.

[26] K. Sohrabi, G.J. Pottie, Performance of a novel self-organization protocol for wireless ad hoc sensor networks,in: Proceedings of the Vehicular Technology Conference,vol. 2, 1999, pp. 1222–1226.

[27] M. Kohvakka, M. Hannikainen, T.D. Hamalainen, Wirelesssensor prototype platform, in: Proceedings of the IEEEInternational Conference Industrial Electronics, Control andInstrumentation, 2003, pp. 1499–1504.

[28] Z.Y.-C. Tseng, S.Y. Ni, E.Y. Shih, Adaptive approachesto relieving broadcast storms in a wireless multihop mobilead hoc network, IEEE Trans. Comput. 52 (5) (2003) 545–557.

[29] M. Kohvakka, J. Suhonen, M. Hannikainen, T.D. Hamalai-nen, Transmission power based path loss metering forwireless sensor networks, in: Proceedings of the 17th AnnualIEEE International Symposium Personal, Indoor andMobile Radio Communications, 2006.

Mikko Kohvakka (M.Sc.’01, TUT)received the M.Sc. in electrical engineer-ing from Tampere University of Tech-nology (TUT), Finland. He is currentlypursuing his Ph.D. in the Institute ofDigital and Computer Systems at TUT.His research interests include hardwareplatforms and MAC layer issues inwireless sensor and ad hoc networks.

Jukka Suhonen (M.Sc.’04, TUT)
received the M.Sc. in computer sciencefrom the Tampere University of Tech-nology (TUT), Finland. He is currentlypursuing his Ph.D. in the Institute ofDigital and Computer Systems at TUT.His research interests include wirelessnetworking, network protocol and algo-rithm design, and Quality of Serviceissues in wireless networks.
Mauri Kuorilehto (M.Sc.’01, TUT)
received the M.Sc. in Computer Sciencefrom Tampere University of Technology(TUT), Finland. He is currently pursuinghis Ph.D. in the Institute of Digital andComputer Systems at TUT. His researchinterests include network simulation,operating systems, and application dis-tribution in wireless sensor and ad hocnetworks.

oc N

Ville Kaseva is a fifth year student atTampere University of Technology
(TUT), and pursuing his M.Sc. in theInstitute of Digital and Computer Sys-tems. His research interests are userinterfaces, sensor integration and loca-tioning services for wireless sensornetworks.
M. Kohvakka et al. / Ad H

Marko Hannikainen (M.Sc.’98, Ph.D.’02,
TUT) acts as a senior research scientistat the Institute of Digital and ComputerSystems at TUT, and a project managerin the DACI research group. Hisresearch interests include wireless localand personal area networking, wirelesssensor and ad hoc networks, and novelweb services.
Timo D. Hamalainen (M.Sc.’93,
Ph.D.’97, TUT) acted as a seniorresearch scientist and project manager atTUT in 1997–2001. He was nominatedto full professor at TUT/Institute ofDigital and Computer Systems in 2001.He heads the DACI research group thatfocuses on three main lines: wireless localarea networking and wireless sensornetworks, high-performance DSP/HWbased video encoding, and interconnec-
tion networks with design flow tools for heterogeneous SoCplatforms.

etworks 7 (2009) 24–41 41

PUBLICATION 5

M. Kohvakka, M. Hännikäinen, and T. D. Hämäläinen, “Energy Optimized BeaconTransmission Rate in a Wireless Sensor Network,” in Proceedings of the 16th Inter-national Symposium on Personal Indoor and Mobile Radio Communications (PIMRC2005), Berlin, Germany, Sep. 11–14, 2005, pp. 1269–1273.

© 2005 IEEE. Reprinted, with permission, from the proceedings of the 16th Interna-tional Symposium on Personal Indoor and Mobile Radio Communications 2005.


Energy Optimized Beacon Transmission Rate in a Wireless Sensor Network Mikko Kohvakka, Marko Hännikäinen, and Timo D. Hämäläinen

Tampere University of Technology Institute of Digital and Computer Systems P.O.Box 553, FI-33101 Tampere, Finland

[email protected], [email protected], [email protected]

Abstract—Resource constrained Wireless Sensor Networks (WSN) require an energy efficient Medium Access Control (MAC) protocol that minimizes the radio active time (duty cycle). Time slotted MAC schemes provide lowest duty cycles by dividing time into consecutive data exchange and sleep periods. Synchronization for data exchange and network maintenance is achieved by exchanging beacons. For detecting changes in network topology, nodes periodically perform scanning during which beacons are received from neighbors. This is energy consuming, and the energy required equals to the transmission of thousands of packets. This paper shows that the energy consumption is mainly depending on the beacon transmission rate, and that an optimal rate is a function of three parameters: a network scanning interval, beacon transmission energy, and radio reception power. The optimal beacon transmission rate is derived for a TUTWSN prototype by power analysis and energy models. According to the analysis, the optimization decreases the average network energy consumption up to an order of magnitude. For the prototype, the optimal beacon transmission rate is 3.7 Hz, when network scanning is performed with 2 minutes intervals.

I. INTRODUCTION Wireless Sensor Networks (WSN) are a relatively new

family of ad hoc networks [1]. They are developed for low data rate monitoring in home, outdoor and industrial environments [2]. WSN nodes combine wireless communication, sensor elements and application computing with ultra-low energy consumption, size and cost. Nodes are autonomous and they operate without any kind of maintenance. Nodes are powered by tiny batteries for months to years, or they scavenge energy from environment [3].

The energy consumption in WSN nodes is typically dominated by transceiver circuitry [4]. Thus, energy conservation requires the minimizing of the radio transmitting and receiving times (duty cycle) while meeting the delay and throughput requirements [5]. Radio transmissions and receptions on the shared wireless channel are controlled by a Medium Access Control (MAC) protocol. Time slotted MAC protocols are the most energy efficient due to small idle listening time [6], [7]. Data is exchanged periodically using either contention based or dedicated assignment channel access, while the rest of time radios are in a sleep mode.

Dynamic WSN with multi-hop data routing needs network maintenance information exchange between nodes. In this

paper, this maintenance traffic is generally denoted as beacons. Especially, time slotted protocols utilize beacon transmissions at the beginning of a data exchange period for scheduling transmissions, obtaining time synchronization, and informing node status. Beacons are also employed when associating or leaving a network. In network scanning, nodes actively search new neighbors by listening beacons on the wireless channel.

At best, time slotted protocols can reduce the duty cycle significantly, reaching far below 1%. In practice, the duty cycle is higher because of network scanning and for reducing data transfer delays. In very low data rate WSNs, the duty cycle is reduced by extending the sleep times between data exchange periods. This usually reduces the beacon transmission rate, but increases the required network scanning time and energy proportionally. The network scanning may consume energy equal to the transmission of thousands of data packets. Hence, techniques for controlling the network scanning energy are very important.

This paper proves that a significant energy reduction is obtained by optimizing the beacon transmission rate. Furthermore, the transmission of additional beacons during node sleep time for reducing the network scanning time is proposed.

This paper is focused on ad-hoc networks with homogeneous nodes. The analysis employs TUTWSN prototype power measurements and energy models. Yet, results are general and applicable for different time slotted WSNs, such as Sensor-MAC (S-MAC) [5], the Timeout-MAC (T-MAC) [8], and IEEE 802.15.4 Low Rate Wireless Personal Area Network (LR-WPAN) standard [9].

To the best of our knowledge, there is no previous work that considers the effect of beacon transmission rate on time slotted WSN energy consumption.

The rest of this paper is organized as follows. Section 2 presents the classification of energy consumptions in the MAC layer. The TUTWSN prototype platform with power consumption analysis and radio energy models is presented in Section 3. The network maintenance power consumption is analyzed in Section 4. The optimization model for beacon transmission rate is presented in Section 5. Section 6 concludes the paper.

II. WSN ENERGY CLASSIFICATION For reasoning the effect of the beacon transmission rate on

node energy consumption, we first divide the energy consumed by wireless communication into three energy classes:

• node start-up, • network maintenance, and • data exchange.

These are presented in Fig. 1. The node start-up energy consists of network scanning for detecting nodes in a range, and network association for connecting a node to a network. The network maintenance and data exchange operations are executed after the start-up period during the node lifetime. The network maintenance energy consists of beacon transmissions and receptions (beacon exchange), network scanning, and possible re-associations. This periodic scanning is needed in dynamic WSNs with node failures, mobility, and radio environment changes. The data exchange energy is consumed by payload data transmissions and receptions, and MAC signaling, such as acknowledgement.

As node lifetimes are expected to be from months to years, the start-up energy is negligible small compared to the total node energy consumption. Furthermore, we assume that data exchange operations have a higher priority than beacon transmissions. Thus, data exchange energy is not affected by the beacon transmission rate. Hence, we can focus on analyzing the network maintenance energy.

A. Network Maintenance Power Network maintenance operations are performed

continuously during the entire network lifetime. Hence, we consider the energy consumption over 1 s period of operation, which equals to the average power consumption. The evaluation of power instead of energy is more convenient due to the independence of time.

The average network maintenance power Pm is defined as the sum of the network scanning power Pns and the beacon exchange power Pb. The network scanning power Pns depends on the energy of a single scanning procedure Ens, and the network scanning interval Ts. Ts determines the maximum delay for detecting changes in a network neighborhood. Shorter Ts improves network robustness, but also increases the

average network scanning power consumption. A long time average network scanning power is obtained by

nsns

s

EPT

= . (1)

Beacons are transmitted at rate fbtx and received at fbrx.

Beacons may be received from multiple neighbors, which increases fbrx proportionally. The average power consumed by the beacon exchange is

b tx btx rx brP E f E f= + x . (2) The network maintenance power in a practical case is

estimated by combining this theoretical analysis with the measurements and energy models of a real TUTWSN prototype platform.

III. TUTWSN PROTOTYPE PLATFORM TUTWSN is a low energy WSN developed at Tampere

University of Technology. TUTWSN design goal has been minimized radio duty cycle, while maintaining the data transfer requirements of applications. Despite of low duty cycle operation, TUTWSN achieves rapid and low energy network start-up and maintenance operations.

The TUTWSN prototype platform is an improved version of the first version presented in [10]. The platform is controlled by Xemics XE88LC02 MicroController Unit (MCU), which integrates a CoolRisc 816 processor core with a 16-bit Analog-to-Digital Converter (ADC), 22 KB program memory and 1 KB data memory. An external 8 KB EEPROM provides a non-volatile data storage. In the prototype, ADC samples data from a temperature sensor, but nearly any type of sensor may be used.

For wireless communication, the platform utilizes a NordicVLSI nRF2401 2.4 GHz radio with selectable 250 kbps or 1 Mbps transmission data rate. The radio has integrated pattern recognition and CRC functions, which simplify communication protocol processing in MCU. In addition, the radio has an interface for low speed MCU, which consists of data buffers for transmission and reception. Radio is used with 256-bit fixed length packets, which enables the use of the higher 1 Mbps transmission data rate improving overall energy efficiency.

The prototype is powered by a 0.22 F super-capacitor and a linear voltage regulator. An interface for an ambient energy scavenging circuit is also available.

The prototype platform is implemented on a 31 mm x 23 mm x 5 mm sized printed circuit board presented in Fig. 2. The top side of the board contains the radio, antenna, EEPROM, sensor, and connectors. MCU and regulator are mounted on the bottom side.

Nod

e st

art-u

p en

ergy

:•N

etw

ork

scan

ning

•Net

wor

k as

soci

atio

n Data exchange energy:• Upper layer payload data and

acknowledgement transmissions and receptions

Network maintenance energy:• Beacon transmissions and receptions

• Periodic network scanning for updating network topology information

Node lifetimeStart-up period

time

Fig. 1. The classification of the WSN node energy consumption.

TABLE II SYMBOL DESCRIPTIONS AND MEASURED OR DEFINED VALUES FOR

TUTWSN PROTOTYPE PLATFORM.

Symbol Description Defined value

Lf frame length 256 bits R radio data rate 1 Mbps Ti idle listening time 100 µs

Tst transmitter and receiver start-up time 250 µs

Symbol Description Measured value

Ptx power consumption in TX mode, MCU in active mode

0 dBm: 30.68 mW -20 dBm: 20.07 mW

Prx power consumption in RX mode, MCU in active mode

44.98 mW

El energy for exchanging 1 bit data between transceiver and MCU

2.3 nJ/bit

Antenna

EEPROM

Temperature sensor

Programming connector

Test pins

Radio

MCURegulator

Top side Bottom side

0 10 20 mm

Fig. 2. TUTWSN prototype platform.

A. TUTWSN Prototype Power Analysis The static power consumptions of the main TUTWSN

platform components are measured at different operation modes. The supply voltage used in the measurements was 2.4 V. The results are presented in Table I. The minimum power consumption is 17 µW, when all components are in the sleep mode. The maximum power consumption is 46.08 mW, when the radio is in the receive mode (RX). The MCU power consumption is 1.33 mW at 1.8 MIPS processing speed. Clearly, the radio is the most power consuming component, taking nearly 33 times the power of MCU.

B. TUTWSN Radio Energy Models Energy consumption in TUTWSN prototype is modeled

according to the MAC protocol procedures (reception, transmission, and scanning) and the frame formats. While the frame formats for the procedures vary, the radio frame in TUTWSN has a fixed length of 256 bits. The symbols with defined or measured values are presented in Table II. Also, the symbols defined in the equations are summarized in Table III.

A frame transmission consists of data loading from MCU to the radio data buffer, a radio start-up transient and a data transmission. During the start-up transient, the radio power consumption is approximated to be equal to the transmission mode (TX) power consumption. Thus, the frame transmission energy Etx is modeled as

txf

stlftx PR

LTELE ⎟⎟

⎠

⎞⎜⎜⎝

⎛+= + . (3)

The resulting energy consumptions are Etx(high) = 16.11 µJ at the higher transmission power (0 dBm) and Etx(low) = 10.74 µJ at the lower transmission power (-20 dBm). This equals to 63 nJ and 42 nJ per transmitted physical layer data bit at the higher and lower transmission power, respectively.

The frame reception begins with the radio start-up transient. The radio remains in the RX mode until a frame has been received including the idle listening time due to TDMA synchronization inaccuracy. Then, a received frame is loaded from the radio to MCU. The frame reception energy Erx is modeled as

frx st i rx f l

LE T T P L

R⎛ ⎞

= + + +⎜ ⎟⎝ ⎠

E . (4)

The resulting energy consumption per received packet is Erx = 27.85 µJ. This yields 109 nJ per received physical layer data bit.

The network scanning begins with the radio start-up transient. To be able to receive beacons from all neighboring nodes, the radio is in the RX mode for the whole beacon interval, which is an inverse of the beacon transmission rate fbtx. Each received beacon is loaded from the radio to MCU,

TABLE III SYMBOL DESCRIPTIONS FOR ENERGY CONSUMPTION MODELS.

Symbol Description

Etx packet transmission energy (256-bit packet)

Erx packet reception energy (256-bit packet)

Ens network scanning energy

Es node start-up energy

fbtx beacon transmission energy

fbrx beacon reception energy

Pm average network maintenance power consumption

Pns average power consumption for network scanning

Pb average power consumption for beacon transmissions

Ts average network scanning interval

TABLE I MEASURED TUTWSN PROTOTYPE POWER CONSUMPTION.

MCU ADC Sensor Radio Power (mW)

active active active RX 46.08 active active active TX (0 dBm) 31.78 active active active TX (-20 dBm) 21.17 active active active data loading 3.76 active active active sleep 2.45 active active off sleep 2.41 active off off sleep 1.35 sleep off off sleep 0.017

and an association request may be transmitted. These energies are difficult to estimate, but they are negligibly small compared to the channel reception energy and therefore ignored in the model. Thus, the network scanning energy Ens can be modeled as

1ns st rx

btxE T P

f⎛ ⎞

= +⎜⎝ ⎠

⎟ . (5)

When fbtx is set to 1 Hz, the resulting network scanning energy is Ens = 44.99 mJ. This equals to the energy of 2793 beacon transmissions at the higher transmission power. Clearly, the effect of scanning on the whole energy consumption of a node is very significant, and becomes more critical at slower beacon transmission rates.

IV. EFFECT OF THE BEACON TRANSMISSION RATE ON POWER CONSUMPTION

The network maintenance power consumption is estimated for the TUTWSN prototype platform by combining the energy models with the real measurements. In this way, we are able to separate the network maintenance power from other power classes and obtain results that are independent on the TUTWSN MAC protocol.

The network scanning, beacon transmission and network maintenance powers in the function of the beacon transmission rate fbtx are presented in Fig. 3. Ts and fbrx are fixed to 500 s and 0.25 Hz. The figure clearly shows how the network scanning power Pns decreases rapidly as fbtx increases. A decrease from 900 µW to 90 µW is obtained as the beacon transmission rate increases from 0.1 Hz to 1 Hz.

In contrast to Pns, the beacon transmission power Pb can be minimized by minimizing fbtx. Pb decreases from 269 µW to 26.9 µW as fbtx decreases from 10 Hz to 1 Hz. As shown in the figure, the network maintenance power Pm has the minimum

(98 µW) at the 1.9 Hz beacon transmission rate. At low beacon transmission rates below 1 Hz, Pm typically doubles as the beacon transmission rate halves. The effect is reversed at high beacon transmission rates above 10 Hz.

Next, we consider the effect of the network scanning interval on network maintenance power consumption. Clearly, the increase of network scanning interval reduces the power for network scanning and shifts the Pm minimum to lower beacon transmission rates. This is plotted in Fig. 4, where Pm varies with fb, while Ts equals to 20 s, 100 s, and 500 s. The network scanning interval affects the maintenance power most significantly at lower beacon transmission rates. When fb is 1 Hz, Pm increases from 137 µW to 2.3 mW as Ts decreases from 500 s to 20 s. Next, we will determine optimal beacon transmission rate in respect of the network scanning interval.

V. BEACON TRANSMISSION RATE OPTIMIZATION An optimal beacon transmission rate (fbtx

*) is determined by minimizing the network maintenance power with respect to the beacon transmission rate. The whole network maintenance power can be summarized as

1rxm st tx btx rx

s btx

PP T E f ET f

⎛ ⎞= + + +⎜ ⎟

⎝ ⎠brxf . (6)

It can be shown that there exists a unique minimum at fbtx

* that is obtained by setting ∂Pm/∂fbtx = 0 in (6). This yields

* rxbtx

tx s

PfE T

= . (7)

An optimal beacon transmission rate is determined by a network parameter (scanning interval) Ts, and the radio parameters Etx and Prx. Fig. 5 presents the variation of fbtx

* with Ts. For the TUTWSN prototype platform, the optimal

0

100

200

300

400

500

600

700

800

900

0.1 1 10 100f b (Hz)

Pm

(µW

)

Beacon transmission powerNetwork scanning powerNetwork maintenance power

Fig. 3. Beacon exchange, network scanning and network maintenance power consumptions as a function of beacon transmission rate (Ts = 500 s, fbrx = 0.25 Hz).

100

1000

10000

0.1 1 10 100f btx (Hz)

Pm

(µW

)

= 20 s = 100 s = 500 s

T s

T s

T s

Fig. 4. Network maintenance power in function of beacon transmission rate with different network scanning intervals (fbrx = 0.25 Hz).

beacon transmission rate increases from 1.3 Hz to 12.9 Hz, as Ts decreases from 1000 s to 10 s.

VI. CONCLUSIONS The estimations and measurements with the TUTWSN

prototype platform depict that the network maintenance power ranges from below 100 µW to several milliwatts, for which the beacon transmission rate has the most significant effect. To enable WSN operation solely from ambient energy, a node power consumption should be in the order of 200 µW [11]. Hence, the optimization of the beacon transmission rate is critical.

The optimal beacon transmission rate is a function of three parameters: the network scanning interval, beacon frame transmission energy and radio power consumption in a RX mode. For the TUTWSN prototype, the optimal beacon transmission rate with a typical temperature sensing application and quite stable network with 120 s network scanning interval is 3.7 Hz. For a highly dynamic network with 20 s network scanning interval, the optimal beacon transmission rate is about 10 Hz.

The beacon rate optimisation results of this paper are not dependent on the data exchange interval. Therefore, we propose the use of additional beacons in the middle of the data exchange interval during idle periods. This lowers the energy consumption at new nodes scanning for a network and at

existing nodes that detect changes in the network neighborhood. The utilized optimization function does not consider collisions on the RF channel, which become noteworthy at higher beacon transmission rates and dense networks. Hence, we propose using a separate network signalling channel for beacon transmissions.

0

2

4

6

8

10

12

14

10 100 1000T s (s)

f btx

* (H

z)

Fig. 5. Energy optimal beacon transmission rate in function of network scanning interval.

The findings of this paper are generic and can be applied on at least LR-WPAN standard, and S-MAC and T-MAC protocols. Especially for LR-WPAN, the proposed additional beacons transmitted during the coordinator inactive period would effectively reduce the network scanning time when over 1 s beacon intervals are used.

REFERENCES [1] K. Sohrabi, J. Gao, V. Ailawadhi, and G. J. Pottie, “Protocols for self-

organization of a wireless sensor network,” IEEE Personal Communications, vol. 7, no. 5, pp. 16-27, Oct. 2000.

[2] E. H. Jr. Callaway, and E. H. Callaway, Wireless sensor networks: architectures and protocols, Auerbach Publications, 2003.

[3] S. Roundy, P. K. Wright, and J. Rabaey, “A study of low level vibrations as a power source for wireless sensor nodes,” Computer Communications, vol. 26, no. 11, pp 1131-1144, Jul. 2003.

[4] F. Jin, H.-A. Choi, S. Subramaniam, “Hardware-aware communication protocols in low energy wireless sensor networks,” in Proc. IEEE Military Communications Conf., vol. 22, no. 1, Boston, MA, Oct 2003, pp. 676-681.

[5] W. Ye, J. Heidemann, and D. Estrin, “Medium access control with coordinated, adaptive sleeping for wireless sensor networks,” ACM/IEEE Trans. Networking, vol. 12, no. 3, pp. 493-506, Jun. 2004.

[6] I. F. Akyildiz, W. Su, Y. Sankasubramaniam, and E. Cayirci, “Wireless sensor networks: a survey,” Computer Networks, vol. 4, no. 38, pp. 393-422, Mar. 2002.

[7] W. R. Heinzelman, A. P. Chandrakasan, and H. Balakrishnan, “An Application-Specific Protocol Architecture for Wireless Microsensor Networks,” IEEE Trans. Wireless Communications, Vol. 1, No. 4, pp. 660-670, Oct. 2002.

[8] T. van Dam, and K. Langendoen, “An adaptive energy-efficient MAC protocol for wireless sensor networks,” in Proc. 1st ACM Conf. on Embedded Networked Sensor Systems, Los Angeles, CA, Nov. 2003, pp. 171-180.

[9] Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal Area Networks (LR-WPANS), IEEE Std 802.15.4-2003 edition.

[10] M. Kohvakka, M. Hännikäinen and T. Hämäläinen, “Wireless sensor prototype platform,” in Proc. IEEE Int. Conf. Industrial Electronics, Control and Instrumentation, Roanoke, VA, Nov. 2003, pp. 1499-1504.

[11] S. Roundy, P. K. Wright, and J. Rabaey, “A study of low level vibrations as a power source for wireless sensor nodes,” Computer Communications, vol. 26, no. 11, pp 1131-1144, Jul. 2003.

PUBLICATION 6

M. Kohvakka, M. Kuorilehto, M. Hännikäinen, and T. D. Hämäläinen, “PerformanceAnalysis of IEEE 802.15.4 and ZigBee for Large-Scale Wireless Sensor Network Ap-plications,” in Proceedings of the 3rd ACM International Workshop on PerformanceEvaluation of Wireless Ad Hoc, Sensor, and Ubiquitous Networks (PE-WASUN 2006),Torremolinos, Spain, Oct. 6, 2006, pp. 48–57.

©ACM, 2006. Reprinted with Permission.

Performance Analysis of IEEE 802.15.4 and ZigBee for Large-Scale Wireless Sensor Network Applications

Mikko Kohvakka Tampere University of

Technology Institute of Digital and Computer Systems

33720 Tampere, Finland +358 3 3115 4702

[email protected]

Mauri Kuorilehto Tampere University of


33720 Tampere, Finland +358 3 3115 4567

[email protected]

Marko Hännikäinen Tampere University of


33720 Tampere, Finland +358 3 3115 3837

[email protected]

Timo D. Hämäläinen Tampere University of


33720 Tampere, Finland +358 3 3115 3385

[email protected]

ABSTRACT This paper analyses the performance of IEEE 802.15.4 Low-Rate Wireless Personal Area Network (LR-WPAN) in a large-scale Wireless Sensor Network (WSN) application. To minimize the energy consumption of the entire network and to allow adequate network coverage, IEEE 802.15.4 peer-to-peer topology is selected, and configured to a beacon-enabled cluster-tree structure. The analysis consists of models for CSMA-CA mechanism and MAC operations specified by IEEE 802.15.4. Network layer operations in a cluster-tree network specified by ZigBee are included in the analysis. For realistic results, power consumption measurements on an IEEE 802.15.4 evaluation board are also included. The performances of a device and a coordinator are analyzed in terms of power consumption and goodput. The results are verified with simulations using WIreless SEnsor NEtwork Simulator (WISENES). The results depict that the minimum device power consumption is as low as 73 µW, when beacon interval is 3.93 s, and data are transmitted at 4 min intervals. Coordinator power consumption and goodput with 15.36 ms CAP duration and 3.93 s beacon interval are around 370 µW and 34 bits/s.

Categories and Subject Descriptors C.2.1 [Computer-Communication Networks]: Network Architecture and Design – wireless communication.

General Terms Measurement, Performance, Standardization, Theory

Keywords Wireless networks, IEEE 802.15.4, ZigBee, WPAN, cluster-tree, performance evaluation, simulation.

1. INTRODUCTION Wireless Sensor Networks (WSN) are fully autonomous self-configuring ad-hoc networks. Typical emerging applications are monitoring and control in home, office, industrial, and outdoor environments. WSNs may consist of thousands of tiny and very energy constrained nodes, which communicate wirelessly with each

other, sense their environment, and share collaborative tasks. Due to the large number of nodes, frequent battery replacements are difficult. Hence, the network lifetime should be in the order of years requiring a very careful design of communication protocols, algorithms, and hardware platforms.

The Wireless Personal Area Networks (WPAN) Working Group was initially focused on creating the IEEE 802.15.1 standard for Physical (PHY) and Medium Access Control (MAC) layers based on Bluetooth technology in 1999 [11]. In the following year the Working Group formed two other subgroups, firstly IEEE 802.15.3 for high-speed WPAN [12] for multimedia applications, and in December 2000 IEEE 802.15.4 low-rate WPAN (LR-WPAN) [13] providing low-complexity, low-cost and low-power wireless connectivity among inexpensive devices [2]. Thus, LR-WPAN has been one of the most foreseen technologies enabling WSNs.

ZigBee [16] is an open specification for low-power wireless networking, which complements the LR-WPAN standard with network and security layers, and application profiles. For security and reliability, ZigBee supports access control lists, packet freshness timers, and 128-bit Advanced Encryption Standard (AES). Different stack profiles are defined for home control, building automation, and plant control applications. The first version of ZigBee specification was announced in December 2004.

In this paper we present mathematical performance analysis and simulations of IEEE 802.15.4 LR-WPAN in a large-scale WSN application with up to 1560 nodes. The network is formed in a beacon enabled cluster-tree topology according to ZigBee specification [16]. While few studies considering only star networks exist, this is the first paper that covers a cluster-tree topology enabling low energy and large area implementations.

The performance of a device and a coordinator are analyzed in terms of the average power consumption and throughput. The analysis results are verified by simulations with WIreless SEnsor NEtwork Simulator (WISENES) [4]. These performance models form a framework that can be used as such for designing real LR-WPAN deployments.

The paper is organized as follows. Section 2 discusses the state of related research on LR-WPAN performance analysis. Section 3 gives an overview of the LR-WPAN standard. The analyzed application, network topology, and hardware platform are presented in Section 4. Section 5 presents the models for required MAC operations. The analysis results are given in Section 6 followed by

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. PE-WASUN’06, October 6, 2006, Torremolinos, Malaga, Spain. Copyright 2006 ACM 1-59593-487-1/06/0010...$5.00.

48

the simulations results in Section 7. Finally, Section 8 concludes this paper.

2. RELATED RESEARCH According to our best knowledge, there exist only few articles that analyze mathematically or simulate the performance of IEEE 802.15.4. The performance of IEEE 802.15.4 in a star network with 100 nodes is analyzed in [1]. The paper contains a compact mathematical analysis of average power consumption and transmission failure rate. The analysis is complemented with real measurements of steady state powers and transient energy, and switch times from a standard compliant evaluation board. A special contribution is bit error rate measurements with two evaluation boards connected through a set of calibrated attenuators. The operational analysis considers mainly the effect of path loss and packet size on energy consumption.

In [6], the performance of IEEE 802.15.4 is analyzed for medical sensor body area networking. The analysis considers quite extensively a very low data rate star network with 10 body implanted sensors transmitting data 1 to 40 times per hour. The analysis focuses on the effect of a crystal tolerance, a frame size, and the usage of IEEE 802.15.4 Guaranteed Time Slots (GTS) on a node lifetime. For analyzing the standard performance in WSN applications, further analysis with larger and more complex network topologies and other IEEE 802.15.4 MAC parameters is required.

In [5], the performance simulations of IEEE 802.15.4 in a star network are presented. The network consists of 49 nodes configured to IEEE 802.15.4 beacon-enabled mode. The evaluation considers latency and energy with different amounts of background traffic. Also, the performance of IEEE 802.15.4 GTS and beacon tracking are simulated. Still, the applicability of the results for WSN applications is insufficient, since larger network sizes with cluster-tree network topologies are required.

3. IEEE 802.15.4 OVERVIEW An IEEE 802.15.4 network may operate in one of three Industrial, Scientific, Medical (ISM) frequency bands, presented in Table 1. The 2.4 GHz frequency band is the most potential for large-scale WSN applications, since the high radio data rate reduces the frame transmission time and thus the energy per transmitted and received bit. Also, network scalability is improved, since a higher number of nodes may communicate with each other within a given time period. In addition, the band is available in most countries worldwide. Hence, in this analysis we focus on the 2.4 GHz band.

IEEE 802.15.4 defines three types of logical devices, a Personal Area Network (PAN) coordinator, a coordinator and a device. The PAN coordinator is the primary controller of PAN, which initiates the network and operates often as a gateway to other networks. Each PAN must have exactly one PAN coordinator. Coordinators collaborate with each other for executing data routing and network self-organization operations. Devices do not have data routing capability and can communicate only with coordinators.

Due to the low performance requirements of devices, they may be implemented with very simple and low-cost platforms. The standard designates these low complexity platforms as Reduced Function Devices (RFD). Platforms with the complete set of MAC services are referred to as Full Function Devices (FFD).

3.1 Network Topology The standard supports two network topologies, star and peer-to-peer, which are presented in Figure 1. In the star topology, communication is controlled by a PAN coordinator that operates as a network master, while devices operate as slaves and communicate only with the PAN coordinator. The devices may be either FFDs or RFDs. This single-hop network is most suitable for delay critical applications, where a large network coverage area is not required.

A peer-to-peer topology allows “mesh” type of networks, where any coordinator may communicate with any other coordinator within its range, and have messages multi-hop routed to coordinators outside its range. The network may contain also RFDs as devices. This enables the formation of complex self-organizing network topologies. Peer-to-peer topologies are suitable for industrial and commercial applications, where efficient self-configurability and large coverage are important. A disadvantage is the increased network latency due to the message relaying.

One special type of peer-to-peer topology is a cluster-tree network, which is highly static once it has been formed. The network consists of clusters, each having a coordinator as a cluster head and multiple devices as leaf nodes. A PAN coordinator initiates the network and serves as the root. The network is formed by parent-child relationships, where new nodes associate as children with the existing coordinators. A PAN coordinator may instruct a new child FFD to become the cluster head of a new cluster. Otherwise, the child operates as a device. Node operations in a cluster-tree network are specified by ZigBee [16].

An example of the cluster-tree network structure is presented in Figure 2. This well-defined structure simplifies multi-hop routing and allows considerable energy savings; each node maintains the synchronization of data exchanges with its parent coordinator only. The rest of time, nodes may save energy in a sleep mode. This is not possible in the “mesh” peer-to-peer networks, where coordinators need to receive continuously to be able to receive data from any node in the range. A disadvantage is that a coordinator failure may cause a large amount of orphaned child and grand-child nodes and energy consuming re-associations.

3.2 Medium Access Control (MAC) Layer The MAC protocol in IEEE 802.15.4 can operate on both beacon-enabled and non-beacon modes. In the non-beacon mode, a protocol is a simple Carrier Sense Multiple Access with Collision Avoidance (CSMA-CA). This requires a constant reception of possible

PAN Coordinator FFD RFD

Star Topology Peer-to-Peer Topology

Figure 1. Star and peer-to-peer topology examples.

Table 1. IEEE 802.15.4 frequency bands and data rates. Band 868 MHz 915 MHz 2.4 GHz

Region EU, Japan US Worldwide Channels 1 10 16 Data rate 20 kbps 40 kbps 250 kbps

49

incoming data. The power saving features that are critical in WSN applications are provided by the beacon-enabled mode. Hence, we concentrate on the beacon-enabled mode from now on.

In the beacon-enabled mode, all communications are performed in a superframe structure presented in Figure 3. A superframe is bounded by periodically transmitted beacon frames, which allow nodes to synchronize to the network. An active part of a superframe is divided into 16 contiguous time slots that form three parts: the beacon, Contention Access Period (CAP) and Contention-Free Period (CFP). At the end of the superframe is an inactive period, when nodes may enter to a power saving mode. The Beacon Interval (IB) and the active Superframe Duration (SD) are adjustable by Beacon Order (BO) and Superframe Order (SO) parameters as

2BOBI aBaseSuperframeDuration= × , (1)

2SOSD aBaseSuperframeDuration= × , (2)

where aBaseSuperframeDuration = aBaseSlotDuration x aNumSuperframeSlots, and 0 14SO BO≤ ≤ ≤ .

aBaseSlotDuration equals to 60 radio symbols resulting 15.36 ms minimum SD (SO = 0) at the 2.4 GHz band with 16 superframe slots. Hence, IB and SD may be between 15.36 ms and 251.7 s. The superframe structure is maintained by the PAN coordinator. In cluster-tree networks, all coordinators may transmit beacons in order to maintain the synchronization with their children.

3.2.1 Contention Access Period In addition to the beacon, CAP is a mandatory part of a superframe. Coordinators are required to listen to the channel the whole CAP to detect and receive any data from their child nodes. On the other hand, the child nodes may only transmit data and receive an optional acknowledgement (ACK) when needed, which increases their energy efficiency.

Downlink data from a coordinator to its child node are sent indirectly requiring totally four transmissions. The availability of pending data is signaled in beacons. First, a child requests the pending data by transmitting a data request message. The coordinator responds to the request with ACK frame, and then transmits the requested data frame. Finally, the child transmits ACK if the data frame was successfully received.

IEEE 802.15.4 utilizes a modified slotted CSMA-CA scheme during the CAP, except for ACK frames that are transmitted without carrier sensing. The scheme is modified from the IEEE 802.11 DCF protocol [10]. The major differences to the legacy CSMA-CA are that a channel is not sensed during a backoff time, and that a new random backoff is selected if a channel is busy during the carrier sensing. In dense networks this may lead to inefficient channel utilization and long channel access delay [3]. The protocol is prone to hidden node collisions, when two nodes outside the range of each other are transmitting data to a common coordinator. The protocol does not provide any type of protection against the hidden node problem [3].

For accessing a channel, each node maintains three variables: NB, BE and CW. NB is the number of CSMA-CA backoff attempts for the current transmission, initialized to 0. BE is the backoff exponent, which defines the number of backoff periods a node should wait before attempting a Clear Channel Assessment (CCA). CW is the contention window length, which defines the number of consecutive backoff periods a channel needs to be silent prior to a transmission. The backoff period length (tBOP) is defined as 0.32 ms in the 2.4 GHz band. Default values for BE and CW are 3 and 2, respectively.

Before a transmission, a node locates a backoff period boundary by the received beacon, waits for a random number of backoff periods (0 to 2BE – 1), and senses the channel by CCA for CW times. If the channel is idle, a transmission begins. Otherwise NB and BE are increased by one and the operation returns to the random delay phase. When NB exceeds macMaxCSMABackoffs (default is 4), transmission terminates with a channel access failure. A node may try to retransmit the frame at maximum aMaxFrameRetries (default is 3) times before MAC issues a frame transmission failure.

The standard defines that each frame is followed by an interframe spacing. The spacing depends on the length of a MAC Protocol Data Unit (MPDU). A Long InterFrame Spacing (LIFS) defined as 640 μs is used for frames containing longer than 18 B MPDU. Shorter frames are followed by a Short InterFrame Spacing (SIFS) defined as 192 μs.

3.2.2 Contention-Free Period CFP is an optional feature of IEEE 802.15.4 MAC, in which a channel access is performed in allocated time slots. A node may reserve bandwidth for delay critical applications by requesting GTS from a PAN Coordinator. The GTS allocations are signaled in beacon frames. In star networks, a device may obtain better Quality-of-Service (QoS) by the use of GTS, since contention and collisions are avoided. Yet, the applicability of GTS in peer-to-peer or cluster-tree networks is poor, since GTS may be used only between the PAN Coordinator and its one-hop neighbors. Moreover, inter-coordinator collisions degrade QoS in GTS, since no collision avoidance mechanism is used in CFP.

PAN Coordinator Cluster Coordinator Device

Cluster

Figure 2. Analyzed cluster-tree network topology.

Beacon CAP CFP Inactive

GTS

GTS

SDBI

Figure 3. Superframe structure in beacon-enabled mode.

50

3.2.3 BatteryLifeExtension IEEE 802.15.4 supports a BatteryLifeExtension option, which may reduce coordinator energy consumption. When the option is selected, BE is initialized to 2, and any transaction should begin during macBattLifeExtPeriods backoff periods (default is 6 equaling 1.92 ms at 2.4 GHz band) after the beacon. This allows the coordinator to sleep the rest of CAP and conserve energy. In practice, only one uplink or downlink transmission attempt is possible per CAP, which significantly reduces available throughput. In addition, a collision probability is high due to the shorter backoff time. Thus, this option is suitable only for small and very low data rate networks.

4. ANALYSIS SETUP 4.1 Network Parameters We analyze IEEE 802.15.4 MAC performance in a large-scale WSN application. In order to obtain the lowest power consumption, a cluster-tree type beacon-enabled network topology is selected. In addition, CFP option is not used and all data exchanges are performed during CAP. Thus, the CAP length (tCAP) is approximated to be equal to the superframe length. In the following analysis SO and BO are varied between 0 - 2 and 6 - 10, resulting 15.36 ms - 61.44 ms CAP length, and 0.96 s - 15.73 s beacon interval. This setting corresponds to a very low duty cycle operation.

The selected cluster-tree network, where each coordinator has three child coordinators (nC) and 12 devices (nD) is presented in Figure 2. The depth of the network is four resulting totally 1560 nodes. Each node transmits data frames, which are routed through the network to a PAN coordinator (uplink data). The transmission interval of data frames is normalized to the beacon interval. In the analysis, the uplink transmission interval is varied between 1 and 100 beacon intervals. For updating data gathering settings, the PAN coordinator broadcasts downlink data frames to the network at the intervals of 100 beacons. In the beacon-enabled network, the broadcasts are implemented by indirect unicast transmissions for each parent-child link.

Inter-cluster interferences are minimized by interlacing superframes of neighboring nodes. According to ZigBee [16] specification time is divided into periodic slots, the length of which equals to the superframe length. During start-up each coordinator performs a passive scan for searching neighbors and free slots for periodic superframes. A slot is selected randomly from founded free slots for reducing collisions with sibling coordinators. Due to the interlacing, inter-cluster interferences are ignored in the analysis.

In the following analysis, each node transmits 6 B sensing item (LI) consisting of 16-bit sensor sample and 32-bit timestamp. In addition, each data frame contains 2 B application layer, 8 B network layer, 9 B data link layer, and 8 B physical layer headers. A 33 B short frame length (LS) is used for transmitting sensing items separately from devices to coordinators, and for transmitting downlink data. Coordinators aggregate received sensing items, and utilize 105 B long frames (LL) containing totally 12 sensing items. Data aggregation requires the use of a 6 B aggregate header. ACK frame length (LA) is 11 B.

4.2 Reference Hardware Platform To obtain realistic results, the IEEE 802.15.4 performance analysis is combined with the real measurements of a commercial IEEE 802.15.4 compliant Chipcon CC2420EM/EB [8] transceiver evaluation board. To estimate the energy consumption of a low power MAC protocol processor, the power consumption of a Microchip PIC18LF8720 [9] nanoWatt series microcontroller is also measured. The measurements include static power consumptions and switch times between various operation modes. The results are presented in Table 2 and Table 3.

5. PERFORMANCE MODELS The LR-WPAN protocol is modeled according to IEEE 802.15.4 standard and ZigBee specification. All the significant run time operations are defined for a device and a coordinator operating in a cluster-tree network by analytical models. The models are presented in Table 4. Due to the long expected network lifetime, models concerning node start-up operations are outside of this analysis.

The models are grouped into CSMA mechanism (contention models), MAC layer operations (MAC operation models), and node operations. Realistic results are obtained by combining these models with the measurements from the hardware platform.

5.1 Contention Models The contention models are used for analyzing the CSMA-CA mechanism including backoff time, collisions and retransmissions. For simplification, the slotted structure is not considered in the equations. Hence, the following equations utilize CAP more efficiently than in reality, giving also slightly higher performance. The following models are based on the analysis of IEEE 802.15.4 in a star network, presented in [6]. The utilized parameters and their values are summarized in Table 5.

Table 2. Measured platform power consumptions. Symbol MCU mode Radio mode Power @ 3 V

Active TX (0 dBm) 48.0 mW Active TX (-1 dBm) 45.0 mW Active TX (-3 dBm) 42.1 mW Active TX (-5 dBm) 39.1 mW Active TX (-7 dBm) 36.0 mW Active TX (-10 dBm) 32.9 mW Active TX (-15 dBm) 29.8 mW

PTX

Active TX (-25 dBm) 26.6 mW PRX Active RX 56.5 mW PCCA Active CCA 55.8 mW PI Active Idle 2.79 mW

PMCU Active Sleep 1.50 mW PS Sleep Sleep 30 μW

Table 3. Measured Chipcon CC2420 transient times. Symbol Description Time (μs)

tSI sleep to idle 970 tIT idle to transmit 192 tIR idle to receive 192 tRT receive to transmit 220 tTR transmit to receive 200

51

5.1.1 Backoff Time and Energy The probability (q) that a single data transmission with the consecutive ACK is detected by CCA at any time in CAP is estimated by the RF transmission times of data and ACK frames. q is modeled separately for short (qS) and long (qL) data frames as

,S AS

CAP

L Lq

t R+

= (3)

.L AL

CAP

L Lq

t R+

= (4)

For estimating the average CSMA backoff time and energy, we first model the amount of data transmissions during CAP. As a uniform cluster-tree network structure is used, the number of nodes (nDL) hierarchically below an analyzed coordinator is calculated as

( )1

1 ,a k

aDL C D

an n n

=

=

= +∑ (5)

where k is the network depth below the analyzed coordinator. The devices associated directly with the coordinator are excluded. As k increases from 1 to 4, nDL gets values from 39 to 1560.

The average amount of data transmissions during CAP (d) is modeled by the number of nodes whose data is routed through the coordinator, the number of retransmissions (u), and the interval of uplink (IU) and downlink (ID) data transmission normalized to a beacon interval (IB). For approximating the effect of indirect communication, ID is divided by 2. As downlink data utilizes flooding through the cluster tree, downlink data transmissions are not increased proportionally to nDL. d is modeled separately for short (dS) and long (dL) data frames as

( )2,D CD

SU D

n nnd u

I I+⎛ ⎞

= +⎜ ⎟⎝ ⎠ (6)

.DL SL

U L

n L ud

I L= (7)

The equations (3), (4), (6) and (7) are further used for modeling the probability (pC) of a clear channel by CCA. According to [7], the probability that two randomly deployed nodes in the range of a coordinator have a hidden node relationship (h) is around 41%. This affects significantly the performance of CCA, since a portion of

ongoing transmissions are not detected. With the two consecutive CCA analyses, as specified in IEEE 802.15.4 standard, pC can be modeled as

2 (1 ) 2 (1 )(1 ) (1 )S Ld h d hC S Lp q q− −= − − . (8)

In case of unsuccessful CCA, the CSMA-CA backoff algorithm is repeated at maximum b times (macMaxCSMABackoffs, default is 4) before declaring a channel access failure. The probability (s) of a successful CCA with b backoff attempts, and the average number of backoffs (r) for each frame are modeled as

1

1(1 ) ,

a ba

C Ca

s p p=

−

=

= −∑ (9)

1

1(1 ) (1 )

a ba

C Ca

r s b ap p=

−

=

= − + −∑ . (10)

By combining r with average backoff time, total backoff time and energy for each transmission attempt are estimated. The average backoff time (tBO) as a function of a backoff exponent is given as

2 1( )2

BE

BO BOPt BE t−= . (11)

After each unsuccessful backoff attempt, BE is incremented by one until aMaxBE (default is 5) is reached. The total backoff time (tBOT) is obtained by summing the average backoff and CCA analysis (tCCA) times of each attempt. As CCA is performed twice only if the channel is assessed to be clear on the first attempt, we approximate

Table 4. IEEE 802.15.4 operation models.

Model Symbol Device power consumption PDEV

Coordinator power consumption PCOOR Maximum throughput TMAX

Node operation models

Goodput G Data frame transmission ETXD, tTXD Indirect data reception ERXDD, tRXDD

ACK transmission ETXA, tTXA ACK reception ERXA, tRXA

Beacon transmission ETXB, tTXB Beacon reception ETXB, tTXB

MAC operation models

Network scanning ENS, tNS Avg. number of retransmissions u

Transmission success rate v Contention models

Backoff energy and time EBOT , tBOT

Table 5. Utilized parameters and their values. Symbol Parameter Value

A Sensing items transmitted in a long frame 12 BO MAC beacon order 6 - 10 h Probability of a hidden node relationship 41 % ID Downlink data transmission interval (IB) 100 INS Network scanning interval 3 h IU Uplink data transmission interval (IB) 1 - 100 k Network depth below analyzed coordinator 1 - 4 LA ACK frame length 11 B LB Beacon frame length 26 B LI Sensing item length 6 B LL Long data frame length 105 B LO Data frame overhead (PHY, MAC, NWK) 25 B LS Short data frame length 33 B nC Number of child coordinators 3 nD Number of devices 12

nDL Number of the nodes hierarchically below a

coordinator 39 - 1560

R Radio data rate 250 kbps SO MAC superframe order 0 - 2 tA ACK frame RF time 352 μs tAW ACK wait duration 864 μs tBOP Backoff period length 320 μs tCCA CCA analysis time 128 μs tI Synchronization inaccuracy 100 μs

tLIFS Long InterFrame Spacing 640 μs tRES Response time for data request 19.52ms tSIFS Short InterFrame Spacing 192 μs εRX The crystal tolerance of a receiving node 20 ppm εTX The crystal tolerance of a transmitting node 20 ppm

52

that averagely 3/2 CCA analyses are performed for each backoff attempt. Hence, the total backoff time can be modeled as

( ) ( )( )1

0

3 min , .2

a r

BOT IR CCA BOa

t r t t t macMinBE a aMaxBE= −

=

= + + +∑ (12)

It is assumed that a radio is in the idle mode during the backoff time. The idle mode is used instead of the sleep mode, since radio start-up time from sleep mode is almost 1 ms. Hence, total backoff energy (EBOT) can be approximated as

( )3 ( )2BOT IR CCA CCA I BOT IE r t t P P t P= + − + . (13)

5.1.2 Collision Probability Collisions causing transmission failures are common in a highly loaded network. The CSMA-CA MAC is prone to two types of collisions; collisions caused by the hidden node phenomenon, and collisions caused by the selection of the same backoff slot with another node. Neither of these collisions can be avoided by the standard, since handshaking before a transmission is not used. The probability (ph) that two nodes in a hidden node position transmit simultaneously and collide is modeled as

( )2 L L S S

hS L

q d q dp

d d+

=+

. (14)

Due to the relatively long inactive part of superframes, nodes most probably gather data during the inactive time and start a backoff procedure simultaneously at the beginning of CAP. Since BE is initialized with macMinBE (default is 3) for each new transmission, nodes randomize backoff delays using a relatively low backoff exponent. The probability (pd) that two nodes select the same backoff delay and collide can be modeled as

12 1d BEp =

−. (15)

5.1.3 Retransmission Probability For approximating the amount of traffic in CAP, we determine the average number (C) of contenting nodes in CAP as

1 2 2min ,1 min ,1 .DL SD C

U D D U C L

n LC u n u n

I I I I n L⎡ ⎤ ⎡ ⎤⎛ ⎞ ⎛ ⎞

= + + +⎢ ⎥ ⎢ ⎥⎜ ⎟ ⎜ ⎟⎢ ⎥ ⎢ ⎥⎝ ⎠ ⎝ ⎠⎣ ⎦ ⎣ ⎦

(16)

For modeling the probability (ps) of a successful transmission, C is combined with ph and pd as

( ) ( )( )1 1S Lh d d Cs h dp s p p+= − − . (17)

A frame is transmitted c times before declaring a transmission failure. c is defined as aMaxFrameRetries (default is 3)+1. The probability of a successful transmission (v) after a attempts is

( ) 1

11

a ca

s sa

v p p=

−

=

= −∑ . (18)

Finally, the average number of transmission attempts per frame (u) can be modeled as

( ) ( ) 1

11 1

a ca

s sa

u v c ap p=

−

=

= − + −∑ . (19)

5.2 MAC Operation Models MAC operation models are used for analyzing energy and time required for the transmission and reception of data, ACK and beacon frames, and the network scanning. A frame transmission in IEEE 802.15.4 consists of a backoff time and the actual data transmission. The time and energy of a data transmission are modeled separately for short (tTXDS, ETXDS) and long (tTXDL, ETXDL) frames according to the above defined backoff models, radio transient times and power consumptions, frame lengths, and radio data rate (R) as

STXDS SI BOT IT

Lt t t t

R= + + + , (20)

STXDS SI I BOT IT TX

LE t P E t P

R⎛ ⎞= + + +⎜ ⎟⎝ ⎠

, (21)

LTXDL SI BOT IT

Lt t t tR

= + + + , (22)

LTXDL SI I BOT IT TX

LE t P E t PR

⎛ ⎞= + + +⎜ ⎟⎝ ⎠

. (23)

In indirect communication the maximum response time (tRES) for a data request is 19.52 ms. Hence, the data reception time and energy may be significantly higher than in direct communication. The coordinator response time is assumed to be an average of tBOT and tRES. In this analysis, indirect communication utilizes only short frames. The time (tRXDD) and energy (ERXDD) for an indirect data transmission after a data request are modeled as

2RES BOT S

RXDD I LIFSt t L

t t tR

+= + + + , (24)

( )RXDD RXDD LIFS RX LIFS IE t t P t P= − + . (25)

The ACK reception time (tRXA) and energy (ERXA) are modeled by assuming the wait duration for ACK being a half of the maximum allowed (tAW). The time and energy are modeled as

2AW A

RXA TR SIFSt Lt t t

R= + + + , (26)

( )RXA RXA SIFS RX SIFS IE t t P t P= − + . (27)

Moreover, ACK transmission time (tTXA) and energy (ETXA) are modeled as

2AW A

TXA RTt Lt t

R= + +

, (28)

2AWA

TXA RT TX ItL

E t P PR

⎛ ⎞= + +⎜ ⎟⎝ ⎠

. (29)

Due to relatively long beacon intervals, beacon reception models consider the synchronization inaccuracy and crystal tolerances in receiving and transmitting nodes (εRX, εTX). In addition, a node is typically in a sleep mode before the beacon reception. The beacon reception time (tRXB) and energy (ERXB) can be modeled as

( ) ,BRXB SI IR RX TX B I LIFS

Lt t t I t t

Rε ε= + + + + + +

(30) ( ) ( ) .RXB RXB SI LIFS RX SI LIFS IE t t t P t t P= − + + + (31)

Finally, beacon transmission time (tTXB) and energy (ETXB) are modeled without CCA analysis as

53

BTXB SI IT

Lt t tR

= + +, (32)

BTXB SI I IT TX

LE t P t P

R⎛ ⎞= + +⎜ ⎟⎝ ⎠

. (33)

If a node looses the contact with its associated coordinator (orphans) due to mobility or a radio link failure, it may perform either an orphaned device realignment procedure, or reset the MAC sublayer and re-associate with the network. In beacon-enabled networks the latter performs best and a node performs a passive channel scan on a single channel prior to the re-association (the operating channel of the network is known). The energy required for the message exchange during the association is negligible compared to network scanning energy, and thus it is ignored in the following analysis. The network scanning time (tNS) and energy (ENS) can be modeled as follows:

( )2 1 ,BONS IRt t aBaseSuperframeDuration= + + (34)

NS NS RXE t P= . (35)

6. ANALYSIS RESULTS In this section the performances of a LR-WPAN device and a coordinator are analyzed in terms of power consumption, requested throughput and achieved goodput.

6.1 Device Power Consumption To be able to operate in a beacon-enabled IEEE 802.15.4 network, a device receives beacons and exchanges data with a coordinator. If a communication link to the coordinator is lost a device performs a network scanning. The rest of time a device is in the sleep mode. The duty cycle of a device (DCDEV) is calculated with beacon receptions, uplink and downlink data exchanges, and network scanning. The average network scanning interval (INS) depends on device speed and radio link quality. DCDEV can be modeled as

( ).TXDS RXA RXDD TXATXDS RXA NSRXB

DEVB U B D B NS

t t t t ut t ttDC u

I I I I I I+ + ++

= + + + (36)

Similarly, the device power consumption is modeled as

( )

( )1 .

TXDS RXA RXDD TXARXBDEV

B D B

TXDS RXA NSDEV S

U B NS

E E E E uEP

I I IE E E

u DC PI I I

+ + += +

++ + + −

(37)

The average device power consumption as a function of the uplink data transmission interval with various IB and SO is plotted in Figure 4. The network scanning interval is approximated to be averagely 3 hours, which corresponds to a deployment with low dynamics and good link qualities. In general, the power consumption decreases with longer beacon and uplink data transmission intervals, since the energy required for beacon receptions and data transmissions diminishes. At longer beacon intervals, the network scanning power becomes significant, since the network scanning energy increases directly proportionally to the beacon interval. Hence, the device power consumption with BO = 10 is higher than when BO equals to 8. Moreover, small SO increases device power consumption slightly due to the increased contention and retransmissions.

6.2 Coordinator Power Consumption In order to operate as a coordinator in a beacon-enabled cluster-tree network, a node has to transmit beacons and receive the CAP for communicating with the nodes associated with it. In addition, a coordinator maintains synchronization with its parent by receiving beacons from it. The activity of a coordinator depends significantly on its location in the network and hence, the requested throughput. As the power consumptions of radio transmission and reception modes are quite similar, we estimate that the power consumption of a coordinator during CAP equals to the reception mode power consumption. The data flow to the uplink direction is performed by long MAC payloads containing A sensing items per each frame. In this analysis A is 12. The duty cycle of a coordinator (DCCOOR) is modeled as

( )( )

( )

1

.

TXDL RXA DL DTXB RXBCOOR

B U B

TXDS RXA RXDD TXA CAP NS

D B B NS

t t n n ut tDC

I I I At t t t u t t

I I I I

+ + ++= +

+ + ++ + +

(38)

Similarly, the average power consumption of a coordinator is

( ) ( )

( ) ( )

1

1 .

TXDL RXA DL DCAP RXTXB RXBCOOR

B B U B

TXDS RXA RXDD TXA NSCOOR S

D B NS

E E n n ut PE EP

I I I I AE E E E u t

DC PI I I

+ + ++= + +

+ + ++ + + −

(39)

The average coordinator power consumption as a function of the uplink data transmission interval with various IB and SO is plotted in Figure 5. As in the case of the device power consumption analysis, the network scanning interval equals to 3 hours. With very low data rate network, coordinator power consumption can go below 200 μW. In typical applications coordinator power consumption is between 1 and 10 mW.

10

100

1000

1 10 100IU (Beacon intervals)

Dev

ice

pow

er (μ

W)

BO = 6, SO = 0 BO = 10, SO = 1BO = 8, SO = 0 BO = 6, SO = 2BO = 10, SO = 0 BO = 8, SO = 2BO = 6, SO = 1 BO = 10, SO = 2BO = 8, SO = 1

Figure 4. Device power consumption as a function of data

transmission interval (k = 3).

54

6.3 Goodput The throughput (TREQ) requested by the sensor application depends on the data transmission intervals (IU, ID), beacon interval (IB) and the number of nodes (children, grand-children etc.) that route data through the coordinator (nDL). The requested throughput can be determined as:

( )21 D CD DL IREQ

U D B

n nn n LTI I I

+⎛ ⎞+ += +⎜ ⎟⎝ ⎠

. (40)

The requested throughput as a function of the uplink data transmission interval is plotted in Figure 6 in different depths (k) of the network. For generalizing the results the throughput is normalized to the beacon interval.

Furthermore, the achieved goodput (G) of a coordinator is modeled with the requested throughput and the probability of a successful transmission (v), as follows:

REQG T v= . (41)

The achieved goodput versus requested throughput with different CAP lengths (SO) is plotted in Figure 7. The results are presented both as a percentage of the requested throughput and in absolute values. Again the results are normalized to the beacon interval. The results indicate that a goodput of about 200 bits per a superframe

can be achieved with 90 % transmission success rate. As SO varies from 0 to 2, the maximum goodputs are 302, 545 and 897 bits per a beacon interval. These are obtained with goodput percanteges between 33% - 55%. With higher requested throughputs, the increased contention reduces goodput rapidly. A longer CAP length reduces contention and improves the goodput.

Network energy efficiency is analyzed by dividing the achieved goodput by the coordinator power consumption, which results to the energy per successfully transmitted bit of payload data over one hop. The achieved energy efficiency is presented in Figure 8. As seen in the figure, the energy efficiency depends on the requested throughput, SO, and BO. The highest energy efficiency is achieved, as SO gets value 1, and is around 5 μJ/bit. The figure also shows that for the energy efficiency, the optimization of BO and SO parameters is very important.

7. SIMULATIONS Provided models are verified by simulating complete ZigBee [16] protocol stack in WISENES [4]. WISENES is targeted for large-scale simulations of long-term WSN deployments. In WISENES, protocols are designed in high abstraction level with Specification and Description Language (SDL) [14]. SDL design tools allow the use of graphical notation in the implementation [15].

The characteristics of environment and sensor nodes are defined for

0.1

1

10

100

1 10 100IU (Beacon intervals)

Coo

rdin

ator

pow

er (m

W)


Figure 5. Coordinator power as a function of data

transmission interval (k = 3).

10

100

1000

10000

100000

1 10 100IU (Beacon Intevals)

T RE

Q (b

its/b

eaco

n in

terv

al)

k = 4k = 3k = 2k = 1

Figure 6. Requested throughput versus uplink data

transmission interval (IU) as k varies.

0 %10 %20 %30 %40 %50 %60 %70 %80 %90 %

100 %

100 1000 10000T REQ (bits/beacon interval)

G /

T REQ

100

1000

G (b

its/b

eaco

n in

terv

al)

SO = 0 SO = 1SO = 2 SO = 0 (right axis)SO = 1 (right axis) SO = 2 (right axis)

Figure 7. Coordinator goodput as a function of requested

throughput (k=2).

1

10

100

10 100 1000 10000T REQ (bits/s)

Ener

gy e

ffici

ency

(μJ/

bit/h

op)


Figure 8. Coordinator energy efficiency as a function of

requested throughput (k = 2).

55

WISENES using eXtensible Markup Language (XML). The detailed parameterization and the realistic modeling of wireless transmission medium, physical phenomena, and sensor node hardware capabilities result to accurate performance information of simulated protocols, nodes, and networks [4]. The output information is stored to logs, which provide detailed event information for post-processing. During runtime, the progress of the simulation is visualized through a graphical user interface.

7.1 Simulation Case Description For the WISENES simulations a node platform similar to the one in the analysis is parameterized, protocols initialized, and an application implemented. The application implemented for the simulations consists of three tasks: sensing, aggregation, and sink. The sensing task on the devices measures a temperature reading periodically (IU) and sends it to the parent coordinator. At the coordinators, the buffered temperature readings from the children are combined to an aggregate packet, which is sent to the PAN coordinator within IU time intervals. At the PAN coordinator, the sink task stores received aggregate packets. Further, it initiates the periodic (ID) downlink control packets. The network topology for the simulations is forced to follow that of analysis.

The IEEE 802.15.4 LR-WPAN MAC protocol implementation in WISENES conforms the standard. The synchronization inaccuracy among the nodes is balanced by a simple algorithm. The algorithm tracks the time drift between consequent beacons and compensates the drift with the averaged inaccuracy over ten last beacons. The functional parameters of the MAC protocol are set by the upper protocol layers. However, unlike specified by the standard, in the association a child device waits for the association response for macTransactionPersistenceTime instead of aResponseWaitTime symbols. The latter is not valid timeout in the beacon-enabled mode with BO values of five or higher.

In the presented simulations, WISENES ZigBee NWK implementation uses hierarchical addressing and routing scheme (nwkUseTreeAddrAlloc and nwkUseTreeRouting are set to 1), which is suitable for static cluster-tree topologies [16]. Each coordinator initiating its own superframe selects a random slot for its superframe among the free slots found out during the passive scan. The WPAN MAC is configured to the beacon-enabled mode with case dependent BO and SO values. The nwkTransactionPersistenceTime is set so that it results to 16 IB. Other parameters are set to the default values of IEEE 802.15.4 standard and ZigBee specification.

7.2 Simulation Results The simulations are done with IU value 60. Only one value of IU is simulated due to the long running time of simulations caused by the large-scale, long simulated network lifetime, and the amount of different parameters. Thus, the simulations show the validity of analysis, which extends to larger number of parameters.

7.2.1 Power Consumption Power consumption is evaluated separately for devices and coordinators. The simulated and analyzed power consumptions for devices with different CAP lengths (SO) are depicted in Figure 9 as a function of IB. The presented simulation results are averaged over randomly selected nodes. Similarly, the power consumptions for coordinators are depicted in Figure 10.

Both analysis and simulation results indicate that SO has only minimal effect on the device power consumption. The difference between the analysis and simulation results is mainly due to the higher dynamics of the simulations. Furthermore, due to the longer frames, a coordinator is able to send several frames during CAP more likely than a device after the first successful send. The overall average difference is 14.7%, which indicates that analysis gives accurate results. In the coordinator power consumption the validity of analysis results is also clearly visible. The main reasons for the deviation are similar to those of device power consumption. The overall average difference is 13.9 %.

7.2.2 Goodput As the uplink data transmissions are relative to IB, the goodput does not depend on the beacon interval. The simulated and analyzed goodputs for coordinators at depth 2 (k = 4) are given in Table 6 for different CAP lengths (SO). The simulated goodput results are averaged over BO values from 6 to 10. As shown, the analyzed and simulated goodputs are quite similar. Unlike in the analysis, SO has a minor effect on the goodput in simulations. As mentioned above, the coordinators tend to be able to send several frames during the CAP. With larger SO this is more beneficial.

7.2.3 Considerations In general, the analysis results correspond accurately to those obtained from the simulations. The main reason for the slight

60

90

120

150

180

0 2 4 6 8 10 12 14 16Beacon interval (s)

Dev

ice

pow

er (μ

W)

Simulation, SO = 2 Analysis, SO = 2Simulation, SO = 1 Analysis, SO = 1Simulation, SO = 0 Analysis, SO = 0

Figure 9. Simulated and analyzed device power consumption as

a function of beacon interval.

0.1

1

10

0 2 4 6 8 10 12 14 16Beacon interval (s)

Coo

rdin

ator

pow

er (m

W)

Simulation, SO = 2 Analysis, SO = 2Simulation, SO = 1 Analysis, SO = 1Simulation, SO = 0 Analysis, SO = 0

Figure 10. Simulated and analyzed coordinator power

consumption as a function of beacon interval (k = 2, IU = 60 IB).

56

deviation is the higher dynamics of the simulations. Random error situations and the reconstruction of the complete sub-tree is typically energy consuming. The leaf nodes may need to initiate several scan operations due to the time required to reinitialize the above network hierarchy. Further, as several nodes are attempting the association simultaneously, association failures are common.

Another cause for the goodput result deviation is the saturation of data throughput at the lower levels of the network hierarchy. This is not considered in the analysis, but in the simulations the goodputs of child coordinators affect to that of the parent coordinators. This is more evident with the small values of IU.

Some issues considering ZigBee specification should be noted based on the simulations. First, the hierarchical addressing and routing scheme is prone to errors due to its assumption of static network. The address space is exhausted, if there are nodes with random mobility pattern. Further, during associations duplicate addresses occur, if a coordinator removes the neighbor table entry of a device, association of which fails. However, if the device successfully received the association response but the acknowledgement was lost, duplicate addresses realize.

Another typical cause for errors is the time slot selection when a coordinator is started. The decision for the own time slot has to be made based on the information obtained during the passive scan. However, in dense networks the delay between the passive scan and the starting of own superframe may be considerable. Thus, a coordinator may not be aware of the time slots of its neighbors, especially if a complete sub-tree is reinitialized after an error situation. Thus, the possibility of the selection of a same time slot is high, if the slot randomization is limited to certain subset of slots, or if BO is small (e.g. 6) and SO large (e.g. 2).

8. CONCLUSIONS The performance of IEEE 802.15.4 standard MAC protocol is analyzed in a large-scale cluster-tree network. The analysis consists of models for MAC and node operations, and CSMA mechanism. According to the analysis, BO and SO parameters have a very significant effect on the network performance and energy efficiency. The results depict that minimum device power consumption is as low as 73 µW, when the beacon interval is 3.93 s, and uplink data are transmitted at 4 min intervals. With the shortest 15.36 ms CAP length and 3.93 s beacon interval achieved coordinator goodput and power consumption are around 34.4 bits/s (135 bits/IB) and 370 µW. Simulations show that the performance models are generally accurate, but some inaccuracy is caused by the random error situations detected by simulations, but which are not included in the analysis. In the future, the operation of IEEE 802.15.4 will be modeled and simulated in a dynamic WSN with error prone nodes and slight mobility.

9. REFERENCES [1] Bougard, B., Catthoor, F., Daly, D.C., Chandrakasan, A.,

Dehaene, W., Energy efficiency of IEEE 802.15.4 standard in dense wireless microsensor networks: modeling and improvement perspectives, in Proceedings of Design, automation and test in Europe (DATE’05) (Munich, Germany, March 7-11, 2005). IEEE, 2005, Vol. 1, 196-201.

[2] Gutierrez, J.A., Callaway, Jr, E.H., Barrett, Jr, R.L., Low-Rate Wireless Personal Area Networks – Enabling Wireless Sensors with IEEE 802.15.4, IEEE Press, NY, 2004.

[3] Hwang, L.-J., Sheu, S.-T., Shih, Y.-Y., Cheng, Y.-C., Grouping strategy for solving hidden node problem in IEEE 802.15.4 LR-WPAN, in Proceedings of the 1st IEEE international conference on Wireless Internet (WICON’05) (Budabest, Hungary, July 10-15, 2005). IEEE, 2005, 26-32.

[4] Kuorilehto, M., Kohvakka, M., Hännikäinen, M., Hämäläinen, T.D., High Level Design and Implementation Framework for Wireless Sensor Networks, In Proceedings of Embedded computer systems: architectures, modeling, and simulation (SAMOS V) (Samos, Greece, July 18-20, 2005). Springer Verlag, 2005, 384-393.

[5] Lu, G., Krishnamachari, B., Raghavendra, C.S., Performance evaluation of the IEEE 802.15.4 MAC for low-rate low-power wireless networks, in Proceedings of the 23rd IEEE international Performance computing and communications conference (IPCCC’04) (Phoenix, AZ, USA, April 15-17, 2004). IEEE, 2004, 701-706.

[6] Timmons, N.F., Scanlon, W.G., Analysis of the performance of IEEE 802.15.4 for medical sensor body area networking, in Proceedings of the 1st IEEE international conference. on Sensor and ad hoc communications and networks (SECON’04) (Santa Clara, CA, USA, 4-7. October, 2004). IEEE, 2004, 16-24.

[7] Tseng, Y.-C., Ni, S.Y., Shih, E.Y., Adaptive approaches to relieving broadcast storms in a wireless multihop mobile ad hoc network, IEEE Trans. on Computers, 52, 5 (May 2003), 545-557.

[8] Data Sheet for CC2420 2.4 GHz IEEE 802.15.4/ZigBee RF Transceiver, available online at http://www.chipcon.com/files/CC2420_Data_Sheet_1_3.pdf

[9] Data Sheet for PIC18LF8720 Microcontroller, available online at: http://ww1.microchip.com/downloads/en/ DeviceDoc/39609b.pdf

[10] IEEE Standard 802.11-1997, Wireless Lan Medium Access Control (MAC) And Physical Layer (PHY) Specifications.

[11] IEEE Standard 802.15.1-2002, wireless medium access control (MAC) and physical layer (PHY) specifications for wireless personal area networks (WPANs).

[12] IEEE Standard 802.15.3-2003, wireless medium access control (MAC) and physical layer (PHY) specifications for high rate wireless personal area networks (WPANs).

[13] IEEE Standard 802.15.4-2003, wireless medium access control (MAC) and physical layer (PHY) specifications for low-rate wireless personal area networks (WPANs).

[14] ITU-T Recommendation Z.100, Specification and description language (SDL), Aug 2002, available online at: http://www.itu.int/ITU-T/studygroups/com17/languages/

[15] Telelogic TAU SDL Suite, http://www.telelogic.com/corp/products/tau/sdl/index.cfm

[16] ZigBee Alliance Document 053474r06; ZigBee Specification, Version 1.0, Dec 2004.

Table 6. Analyzed and simulated goodputs of a coordinator (k = 2, IU = 60 IB).

SO Analysis (bits/ IB )

Analysis (%)

Simulation (bits/ IB )

Simulation (%)

SO = 0 135.6 91.3 132.5 89.2 SO = 1 136.4 91.9 142.0 95.6 SO = 2 136.7 92.0 147.1 99.1

57

PUBLICATION 7

M. Kohvakka, M. Hännikäinen, and T. D. Hämäläinen, “Wireless Sensor PrototypePlatform,” in Proceedings of the 29th Annual Conference of the IEEE Industrial Elec-tronics Society (IECON 2003), Virginia, USA, Nov. 2–6, 2003, pp. 1499–1504.

© 2003 IEEE. Reprinted, with permission, from the proceedings of the 29th AnnualConference of the IEEE Industrial Electronics Society 2003.


Wireless Sensor Prototype Platform

Mikko Kohvakka, Marko Hännikäinen, and Timo D. Hämäläinen Tampere University of Technology

Institute of Digital and Computer Systems Korkeakoulunkatu 1, FIN-33720 Tampere, Finland

Tel. +358 3 3115 2111, Fax +358 3 3115 4786 [email protected], [email protected], [email protected]

Abstract – This paper presents the design and performance

measurements of a wireless sensor prototype platform (UbiSensor). UbiSensor combines techniques used in wireless microsensors and radio frequency identification (RFID) resulting a wireless sensor having sensing, data processing, network protocol execution, and energy scavenging capabilities. The platform design is driven by energy consumption minimization of given tasks. A commercially available microcontroller, low power RF transceiver, and power generator circuits are used. The measurements indicate 430 µW power consumption in typical operating conditions, using 1 Hz sample and transmit rate, which can be scavenged by a 29 mm2 sized solar cell or using a transponder interface circuit in a near proximity to a RFID reader. UbiSensor performance results are promising, but further research is required.

I. INTRODUCTION

The rapid development of low power radio frequency circuits and microcontrollers has enabled new wireless applications in industrial and home automation. One class of applications is wireless microsensor networks, where numerous wireless low power sensors enable robust and powerful sensing even for automated control applications.

Several research projects, e.g. WINS [1] and SmartDust [2], are aiming to integrate sensing, computing, and wireless communication in small size and low-cost production. This paper presents design of a new type of wireless sensor called UbiSensor that has been developed at Tampere University of Technology. UbiSensor combines low power radio and microcontroller of wireless microsensors and power management of radio frequency identification (RFID). This results a prototype, which is able to scavenge and store energy from operating environment. Supported sensor types are e.g. light, temperature, acceleration, pressure, and gas sensors.

This paper is organized as follows. The design requirements for the UbiSensor network and platform are presented in Section 2. Section 3 gives a detailed description of the UbiSensor hardware platform design. Power performance measurements are presented in Section 4. Finally, conclusions are given and future work projected.

II. UBISENSOR DESIGN REQUIREMENTS

The aim of this research is to improve the quality of industrial manufacturing processes by utilizing wireless sensor networks. As one of the results, the UbiSensor has been developed.

The design of UbiSensor platform is driven by the requirements of sensing applications and operational environments. Wireless sensors can be embedded and distributed into the environment. Thus, battery replacement is a difficult option and sensors may have to generate their supply energy from environment. The small physical size and light weight of a sensor are central limitations. Also, due to the large number of sensors required, the production cost should be low.

While wireless sensors have scarce power budged, the performance requirements are considerable. Our UbiSensors are required to measure 16-bit samples with 1-10 Hz sample rates, perform different type of processing tasks on data, and transfer processed samples through the sensor network by executing network protocols.

The energy awareness should be incorporated into every stage on the network design and operation. In platform design, component selection is important and compromises have to be made between obtained performance and power consumption to minimize overall energy consumption of given tasks. Dynamic power consumption is reduced by minimizing supply voltage and operating frequency. In software design it is important to minimize the number of executed instructions in the given tasks. In network protocols, energy consumption is reduced by minimizing the number and duration of receive and transmit operations, while using effectively power save modes. Furthermore, the entire system must be able to dynamically adjust system performance and make tradeoffs between energy consumption and operational fidelity [2 - 5].

UbiSensor network is controlled by a developed ubiquitous network protocol. The protocol design is driven by the required support for dynamic ad-hoc routing. The protocol also defines a gateway node, which bridges the sensor network to outside resources. The network should be flexible to manage, which means that no manual configuration is needed for changes of network contents. Sensors are mobile and can join or leave the network in ad-hoc manner. The protocol should contain power saving support. Furthermore, the network should provide place and time information associated with the measurement data. The network protocol design is outside of the scope of this paper, and we are concentrating on the UbiSensor platform design.

III. UBISENSOR PLATFORM DESIGN

General platform architecture for a wireless microsensor can be divided into four subsystems [4]:

• Computing subsystem consisting of a Micro Controller

Unit (MCU) • Communication subsystem consisting of a short range

radio transceiver • Sensing subsystem consisting of a set of sensors and

actuators, which link the microsensor to the outside world, and

• Power subsystem consisting of a battery and a voltage converter, which provides the system supply voltage.

The general architecture is used as a base for the UbiSensor design. The developed UbiSensor platform architecture is presented in Fig. 1. The central component of the platform is MCU that includes the computing subsystem. In addition, part of the communication and sensing subsystems are executed on MCU. For communication, MCU includes the network protocol and drivers for accessing the radio. MCU also controls the Analog-to-Digital Converter (ADC) that gathers data from sensors. The power subsystem consists of power generators, a set of voltage converters and a capacitor. The power subsystem provides system supply energy, which is scavenged from operating environment.

A. Communication Subsystem

The communication subsystem consists of a radio transceiver, which enables wireless communication with neighbouring UbiSensors. In general, radio has four operation modes: transmit, receive, idle and sleep. In transmit and receive mode, radio requires minimum power levels on the order of multiple milliwatts due to analog mixers, filters and oscillators. Thus, it is important to minimize the communication time and maximize the low power time in idle and sleep modes. Furthermore, the operating mode change transients may dissipate a lot of energy, e.g. from sleep mode to transmit mode, which should be considered in the network protocol [2][4].

Frequency band affects significantly to the radio properties. The operating frequency determines radio operating range and interferences from the surrounding machinery. For industrial environment the interferences emitted from the machinery are significant below few hundred megahertz and drops rapidly above 1 GHz frequency [6]. Most commercial short-range radio transceivers in European region operate in the 433, 868 and 2400 MHz license-free Industrial Scientific Medicine (ISM) frequency bands. The characteristics of some of the most potential low power transceivers currently available are summarized in Table 1 [7 - 16].

The power efficiency of the radio can be determined by

dividing radio power consumption by radio bit rate in transmit or receive mode, which yields energy per bit ratio (E/bit). The E/bit is computed to the radios above and results are presented in Fig. 2. The supply voltage is fixed to 3.0 V for comparability. As seen in the figure, radios operating at the 2.4 GHz frequency band are the most energy efficient due to high data rates. Yet, these high data rates require also a high performance microcontroller with a high-speed serial bus.

This problem is solved in the NordicVLSI nRF2401 [9] transceiver by an on-chip buffer, which adapts a low data rate microcontroller interface with a high data rate (250 kbps or 1 Mbps) radio. The output power is adjustable between -20 and 0 dBm. The nRF2401 is utilized and analyzed in the UbiSensor prototype.

B. Computation Subsystem

The choice of MCU affects significantly the power consumption of a wireless sensor. Microcontroller power efficiency can be analyzed by dividing processor performance, using million instructions per second (MIPS), by the power consumption. Thus, MIPS/W ratio can be computed as

2

6

2

6 1010

ddSWddSW VCCPIVCfCPIf

WMIPS

⋅⋅=

⋅⋅⋅⋅=

−−

, (1)

TABLE I TRANSCEIVER FEATURES

Manufacturer Model Data Rate (kbps)

Band (MHz)

Sleep (uA)

Idle (mA)

RX (mA)

TX (mA)

NordicVLSI nRF2401 1000 2400 1 0.03 19 8.8Skyworks CX72303 1000 2400 1 0.02 24 11Conexant RF109 1200 2400 5 25 89 31Infineon PBA31305 1000 2400 70 35 50 60RFM TR1001 115.2 868 0.75 - 3.8 12Xemics XE1201A 64 433 0.2 0.06 6 5.5Infineon TDA 5250 64 868 9 0.75 8.6 4.9NordicVLSI nRF903 76.8 433-950 1 0.6 18.5 12.5Xemics XE1202 76.8 868/433 0.2 0.85 14 33NordicVLSI nRF401 20 433 - 0.01 11 8

Power Subsystem

Capacitor

Voltage Coverter

CommunicationSubsystem

Pow

er

gene

rato

r

Sens

or

ADC

Rad

ioMCU

SensingSubsystem

ComputingSubsystem

Fig. 1. UbiSensor architecture.

57

72

281

403

33

78

258

230723

1650

223

547

150

99

1200

1289

488

313

180

26

0 200 400 600 800 1000 1200 1400 1600 1800

nRF2401

CX72303

RF109

PBA31305

TR1001

XE1201A

TDA 5250

nRF903

XE1202

nRF401

Energy/bit (nJ/b)

TransmitReceive

Fig. 2. Transceiver energy consumption comparison.

where CSW is switch capacitance, Vdd system supply voltage, f operating frequency and CPI clock per instruction ratio [17]. The MIPS/W ratio is not dependent on the operating frequency. Power efficiency of some low power microcontrollers is compared in Fig. 3. The supply voltage is fixed to 3.0 V for fairness. We should note that the instruction set affects the performance to some extend. Thus, only orders of magnitude are important. As seen in the figure, CoolRisc processors are the most power efficient. The differences between 81, 88 and 816 models are basically in the amount of registers and memory [17].

The UbiSensor computation subsystem uses Xemics XE88LC01 [18] microcontroller, which integrates CoolRisc 816 processor core with ADC. The microcontroller has 8 k instructions program memory and 512 B data memory. Maximum clock frequency is 2 MHz and CPI ratio is 1. According to measurements, power consumption of the XE88LC01 is 1.8 mW at 1.8 MHz clock frequency and 3.0 V supply voltage yielding power efficiency of 1000 MIPS/W. Result is worse than in Fig. 3 due to the integrated peripheral components, i.e. ADC, UART, USRT, counters, timers and I/O ports.

C. Sensing Subsystem

The selection of ADC is driven by current consumption and sample resolution. Sample rate is typically not critical and is in the order of 10 Hz. UbiSensor sensing subsystem uses a simple low power temperature sensor, and a 16-bit ADC, which is integrated in the MCU. The ADC includes three pre-amplification stages for signal pre-processing and a switch for selecting input from four differential sources [18].

D. Power Subsystem

The UbiSensor power subsystem design is presented in Fig. 4. The power subsystem consists of two power generators, which scavenge system supply energy. Panasonic BP-376634 solar panel is 37 mm x 66 mm sized having 5.5 V open circuit voltage and 33 mA short circuit current. Atmel U3280 [19] transponder interface circuit is designed for RFID applications and is able to generate energy and communicate using 125 kHz magnetic field. Yet, only energy generator is used and communication is implemented using

higher performance radio transceiver. Output voltage of the generator is limited to 2.8 V and current to 8 mA. However, magnetic field of the operation environment is not sufficient for energy scavenging and extra field is generated by the Atmel TMEB8704 RFID reader application kit [20].

The voltage converters provide voltage conversions and regulation. An important issue is efficiency. Thus, switch-mode regulators are used. The UbiSensor power subsystem consists of Maxim MAX1676 [21] step-up and Texas Instruments TPS62200 [22] step-down regulators, which are selected due to small quiescent current and good efficiency at the required current range. The step-down regulator output voltage is minimized to 2.2 V to reduce system power consumption. This is below the specified supply voltage range of MCU, but the system reliability is verified.

A capacitor ensures the UbiSensor supply energy and applicability when energy cannot be generated from the operating environment. The UbiSensor uses Panasonic EECS0HD224H [23] capacitor rated to 0.22 F and 5.5 V. The capacitor voltage depends on the energy of the capacitor and varies between 2.2 V and 5.0 V. The lower 2.2 V limit is determined by the output voltage of the step-down regulator.

E. Implementation

Implemented UbiSensor prototype is presented in Fig. 5.

5071

125133

540625

1111

25003030

5000

0 1000 2000 3000 4000 5000

8051 (83L51FA)PIC17Cxx

68HCxxPIC16C5x

ARM7/ThumbPunch

Nordic uRISCCoolRisc 816

CoolRisc 88CoolRisc 81

MIPS / W

Fig. 3. MIPS/W comparison [17].

MAX1676 TPS

62200

TRANSPONDERINTERFACE

SOLAR PANEL

2.2V2.8V

5.0V

2.2 - 5.0V0.22F

Power Generator Voltage Converters

Capacitor

Fig. 4. Power subsystem architecture.

CapacitorTransponder Interface Radio

Voltage Regulators

Interconnection Board and Sensor

MCUSolar Panel

Fig. 5. Photograph of the UbiSensor prototype.

Computation and communication subsystems are implemented using the evaluation boards of component manufacturers. Power subsystem is implemented using separate circuit boards for each functional block, which are then optimized and measured separately. An interconnection board is used to connect the circuit boards together, and is also equipped with the temperature sensor.

IV. PROTOTYPE POWER PERFORMANCE

UbiSensor prototype power consumption is first calculated using data sheets of the selected components. Best case and lowest power operating conditions are used to find the lower limit of the consumption. Then, power consumption and power generator performance are measured in typical operating configuration. In both cases the following operation sequence is used.

MCU is powered up from idle mode and the temperature sensor is sampled. Then, the transceiver is powered up and measured data is transmitted in a packet of 96 bits. Finally, the system returns to idle mode. System wake up frequency is 1 or 10 Hz.

A. Calculated Minimum Power Consumption

Minimum power consumption is calculated using 2.2 V supply voltage and the following system configuration.

Radio – Transceiver power consumption is estimated according to the NordicVLSI nRF2401data sheet [9]. The packet is loaded into the radio transmit buffer using 50 kbps data rate and transmitted using 1 Mbps data rate and -20 dBm output power. Transceiver current consumptions in transmit and idle mode are 8.8 mA and 32 µA. The transmitter power up time is 195 µs.

MCU and ADC – The power consumption of CPU and ADC is estimated according to the Xemics XE88LC01 data sheet [18]. ADC current consumption is approximated using 16-bit conversion and 512 kHz input sampling frequency with 256 times oversampling. Thus, the conversion time is 500 µs. The MCU and ADC use 2 MHz clock frequency. Idle mode power consumption is 1.7 µA.

Voltage converter – The efficiency of the voltage converter (DC-DC) is estimated to be 80%.

The results of the calculation are presented in Fig. 6a. As

seen in the figure, calculated minimum power consumptions using 1 Hz and 10 Hz sample rates are 117 µW and 314 µW.

B. Measured Power Consumption

UbiSensor power consumption is measured in typical operating conditions. Operating frequency is set to 1.8 MHz due to correct timing of UART, which is used for diagnostics reasons. The measured power consumptions for each subsystem are presented in Fig. 6b. The system power consumptions using 1 Hz and 10 Hz sample rates are 430 µW and 1.32 mW. Results are significantly higher than the calculated minimum. This is probably due to the higher actual power consumption on the idle mode than estimated. As seen in the figure, the transceiver power consumption is remarkably high at 10 Hz transmit frequency. One reason for that is the energy consumption at the transmitter power down transient, which is unspecified and ignored in the calculations.

The current waveforms of microcontroller and transceiver are measured and presented in Fig. 7. and Fig. 8. Due to

b) Measured

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1 10Sample Rate (Hz)

Pow

er C

onsu

mpt

ion

(mW

) MCU&ADCRADIODC-DC

a) Calculated Minimum

0.0

0.1

0.2

0.3

0.4

1 10Sample Rate (Hz)

Pow

er C

onsu

mpt

ion

(mW

) MCU&ADCRADIODC-DC

Fig. 6. Calculated and measured power consumption.

Fig. 7. Measured current consumption waveforms.

Fig. 8. Zoomed current consumption waveforms.

decoupling capacitors on the device boards, the waveforms are slightly smoothed. As seen in the figure, system is powered up at 1000 ms intervals. Average current consumption of transceiver and microcontroller are 126 µA and 28 µA. Detailed current consumption waveforms are presented in Fig. 8. System wake up event is marked in the figure as A. During the first 150 µs after the wake up MCU and ADC are initialized (A - B), which is followed by 500 µs long ADC measurement. During this, MCU is in the idle mode and the current consumption is low. When conversion is ready (C), MCU is powered up and the radio is initialized. That lasts around 230 µs (C - D). The transmitted data is loaded into the radio transmit buffer using 55 kbps data rate, which lasts 1.8 ms (D - E). Then, the radio is powered up to transmit mode (F - G) and the packet is transmitted using 1 Mbps data rate (G). Finally, the system returns to idle mode (H).

UbiSensor power consumption is analyzed in typical operating conditions and supply voltages. The results are presented in Table 2. The analysis excludes the power consumption of the voltage converter. The highest power consumption of 61.98 mW is reached when the radio is receiving and MCU, ADC and the sensor are active. Reducing the supply voltage from 3.0 V to 2.4 V decreases system power consumption 24% - 77%, depending on the operation mode. Reducing the transmit power from the maximum (0 dBm) to the minimum (-20 dBm) decreases power consumption 33%. Furthermore, the system power consumption is 54% higher in the receive mode than in the transmit mode using the lowest (-20dBm) RF transmission power. That should be considered when designing network protocols.

C. Measured Power Generator Performance

The energy scavenging performance of the solar panel and transponder interface circuit are analyzed. The transponder interface circuit is implemented using the Atmel U3280 transponder interface chip. The circuit uses LC-resonance circuit tuned to 125 kHz using 2.2 nF capacitor and 737 µH coil. The coil area is 31 cm2.The Atmel TMEB8704 reader application kit is used to generate magnetic field and the output power of the U3280 is measured with various load resistances and distances between reader coil and transponder coil. The results are presented in Fig. 9. The maximum power is around 19 mW at 410 ohms load resistance. The chip has internal voltage and current limiters, which make a distinct peak to the output power plot and should be considered when designing voltage converter circuit. As seen in the right figure, the U3280 chip requires fairly strength magnetic field and practically an external field generator is required. When using TMEB8704 application kit, useful operating range is on the order of few centimeters. For longer operating range, a higher power field generator is required.

The Panasonic BP-376634 solar panel consists of small and serial connected solar cells gaining totally 5.5V output voltage. The output power of the solar panel is measured with various loading impedances at the same time as the light intensity is varied. The results are presented in Fig. 10. As seen in the figure, optimal loading impedance is 400 ohms and measured maximum output power is 36 mW at 80 klx

0

5

10

15

20

100 300 500 700Load Resistance (Ohms)

P / mWI / mAU / V

0

5

10

15

20

0 2 4 6 8Proximity (cm)

P / mWI / mAU / V

Fig. 9. Transponder interface performance.

0

5

10

15

20

25

30

35

40

100 300 500 700Load Resistance (Ohms)

P / mWI / mAU / V

0

5

10

15

20

25

30

35

40

1020304050607080Illuminance (klx)

P / mWI / mAU / V

Fig. 10. Solar panel performance.

TABLE II POWER ANALYSIS OF THE UBISENSOR

MCU Mode

ADC Mode

Sensor Mode Radio Mode Vcc (V) Power

(mW)Receive Data 3.0 61.98

2.4 46.28Transmit Data 3.0 43.41

(Power: 0 dBm) 2.4 31.98Transmit Data 3.0 29.82

Active Active Active (Power: -20 dBm) 2.4 21.37Configuration and 3.0 5.25

data loading 2.4 3.95Idle 3.0 3.70

2.4 2.703.0 3.632.4 2.653.0 3.392.4 2.35

Sleep 3.0 2.10 2.4 1.29

Idle 3.0 0.44Off 2.4 0.10

Idle 3.0 0.44 2.4 0.10

Not Not installed 3.0 0.14installed 2.4 0.02

light intensity, which correspond approximately a bright sun light. Generated energy corresponds to the power of 1.5 mW/cm2. The characteristics of the solar panel are promising. The selected solar panel is optimal for outdoor use, where sunlight enables adequate power generation.

V. CONCLUSIONS

A prototype for a wireless ubiquitous sensor network is designed and implemented. The design process includes specification of system requirements, analysis and selection of system components, and performance calculations. The prototype is implemented and its applicability according to power consumption verified by measurements. The power consumption measurements indicate promising results.

The prototype average power consumption using 1 Hz sample and transmit rate is 430 µW, which can be scavenged by the both power generator methods, e.g. using a 29 mm2 sized solar cell in a bright sunlight or using the transponder interface circuit in a near proximity to the RFID reader. The average power consumption depends significantly on the executed network protocol and the sample rate requirements of the sensor application. The measure-and-transmit type of operation presented in this paper requires a continuously receiving destination node. When power consumption should be minimized in the entire network, this type of communication is usually not practical.

UbiSensor is passive most of time and the power consumption depends strongly on the utilization of low power operating modes. Important issues in the UbiSensor development are the research of efficient energy scavenging methods and the design of an intelligent network protocol combined with a power management algorithm.

REFERENCES

[1] G. Asada, M. Dong, T. Lin, F. Newberg, G. Pottie, W. Kaiser, and H. Marcy, “Wireless Integrated Network Sensors: Low Power Systems on a Chip”, European Solid State Circuits Conference, (ESSCIRC’98), Sep 1998.

[2] B. Warneke, M. Last, B. Leibowitz, and K. S. J. Pister, "Smart Dust: Communicating with a Cubic-Millimeter Computer", Computer, vol. 34 no. 1, pp. 43-51, Jan 2001.

[3] R. Min, M. Bhardwaj, Seong-Hwan Cho, E. Shih, A. Sinha, A. Wang, and A. Chandrakasan, "Low-Power Wireless Sensor Networks", Fourteenth International Conference on VLSI Design, Jan 2001, pp. 205 – 210.

[4] V. Raghunathan, C. Schurgers, S. Park, M. B. Srivastava, "Energy-aware wireless microsensor networks", IEEE Signal Processing Magazine, vol. 19 no. 2, pp. 40-50, Mar 2002.

[5] A. Bellaouar, M. Elmasry, Low-Power Digital VLSI design Circuits & Systems, Kluwer Academic Publishers: 1995.

[6] T.S. Rappaport, “Indoor radio communications for

factories of the future”, IEEE Communications Magazine, vol. 27, no. 5, May 1989, pp. 15 –24.

[7] Xemics SA, “XE1201A 300-500MHz Low-Power UHF Transceiver” Data Sheet, 2002. Available: http://www.xemics.com/

[8] Xemics SA, “XE1202 433MHz/868MHz/915MHz Low-power, integrated UHF transceiver”, Data Sheet, 2003. Available: http://www.xemics.com/

[9] NordicVLSI ASA, “Single Chip 2.4 GHz Transceiver nRF2401”, Product Specification, Rev. 1.0, 2003. Available: http://www.nvlsi.no/

[10] NordicVLSI ASA, “433MHz-950MHz Single Chip RF Transceiver nRF903”, Product Specification, Rev. 3.1, 2002. Available: http://www.nvlsi.no/

[11] NordicVLSI ASA, “433 MHz Single Chip RF Transceiver nRF401”, Product Specification, Rev. 1.6, 2002. Available: http://www.nvlsi.no/

[12] Infineon Technologies AG, “PBA 313 05 Bluetooth Radio Transceiver”, Product Brief, 2003. Available: http://www.infineon.com/

[13] Infineon Technologies AG, “ASK/FSK 868MHz Wireless Transceiver TDA 5250 D2”, Data Sheet, Rev 1.6, 2002. Available: http://www.infineon.com/

[14] Skyworks Solutions, Inc, “CX72303: 1.8 V Ultra-Low Power Bluetooth RF Transceiver”, Product Summary, 2003. Available: http://www.skyworksinc.com/

[15] RF Monolithics, Inc., “TR1001 868.35 MHz Hybrid Transceiver”, Data Sheet, Rev. 2003.07.17, 2003. Available: http://www.rfm.com/

[16] Conexant Systems, Inc, “RF109 2400 MHz Digital Spread Spectrum Transceiver”, Data Sheet, 2000. Available: http://www.conexant.com/

[17] C. Piguet et al., “Low-Power Design of 8-b Embedded CoolRisc Microcontroller Cores”, IEEE Journal of Solid-State Circuits, vol. 32, no. 7, Jul 1997, pp. 1067-1078.

[18] Xemics SA, “XE88LC01 Sensing Machine Data Acquisition with 16+10 bit ZoomingADC”, Data Sheet, 2002. Available: http://www.xemics.com/

[19] Atmel Corp., “U3280M Transponder Interface for Microcontroller”, Data Sheet, Rev. A4, 2001. Available: http://www.atmel.com/

[20] Atmel Corp., “RFID Application Kit TMEB8704”, Application Note, 2001. Available: http://www.atmel.com/

[21] Maxim Integrated Products, Inc., “MAX1674 – MAX1676 High-Efficiency, Low-Supply-Current, Compact, Step-Up DC-DC Converters”, Data sheet, Rev. 3, 2000. Available: http://www.maxim-ic.com/

[22] Texas Instruments Inc., “TPS62200 – TPS62205 High-Efficiency, SOT23 Step-Down, DC-DC Converter”, Data sheet, Mar 2002. Available: http://www.ti.com/

[23] Matsushita Electric Corporation of America, “Electric Double Layer Capacitors (Gold Capacitor)/SD”, Data Sheet. Available: http://www.panasonic.com/

PUBLICATION 8

M. Kohvakka, M. Hännikäinen, and T. D. Hämäläinen, “Ultra Low Energy WirelessTemperature Sensor Network Implementation,” in Proceedings of the 16th Interna-tional Symposium on Personal Indoor and Mobile Radio Communications (PIMRC2005), Berlin, Germany, Sep. 11–14, 2005, pp. 801–805.

© 2005 IEEE. Reprinted, with permission, from the proceedings of the 16th Interna-tional Symposium on Personal Indoor and Mobile Radio Communications 2005.


2005 IEEE 16th International Symposium on Personal, Indoor and Mobile Radio Communications

Ultra Low Energy Wireless Temperature SensorNetwork Implementation

Mikko Kohvakka, Marko Hannikainen, and Timo D. HamalainenTampere University of Technology

Institute of Digital and Computer SystemsP.O.Box 553, FI-33101 Tampere, Finland

[email protected], marko.hannikainen@tutfi, [email protected]

Abstract-Condition monitoring in buildings is one of the mostpotential and foreseen applications for Wireless Sensor Networks(WSN). This paper presents the design and full scale prototypeimplementation of WSN for temperature monitoring. Theprototypes are implemented using low power commercial of-the-shelf components including a 2.4 GHz radio, microcontrollers,and a custom TUTWSN communication protocol. A userapplication provides a graphical data analysis. Measurementsindicate 183 ltW to 390 jIW average node power consumptions, astemperature is measured at 5 s intervals and data is multi-hoprouted to a gateway. Predicted lifetime with two AA batteries isup to 4.9 years. In addition, experiments indicate that timeaccuracy is extremely important in hardware prototypes.

I. INTRODUCTION

Wireless Sensor Networks (WSN) will soon enable variousintelligent applications in home, office, industrial and outdoorenvironments [1]. WSNs consist of thousands of fullyautonomous nodes, which communicate wirelessly with eachother, sense their environment and share collaborative tasks.Data is routed using multiple hops for robust and scalableoperations [2]. Although the network management is complex,available energy and radio bandwidth are very scarce.

Conditional monitoring in buildings, including temperature,humidity, and gas measurements is a straightforward and oneof the most potential applications for WSNs [1]. The data rateand latency requirements are quite low, but the required nodelifetime should be years without maintenance [3].

Only few experimental evaluations of conditionalmonitoring multi-hop WSNs have been published. One WSNimplementation presented in [4] employs totally 65 BerkeleyMica II Commercial Off-The-Shelf (COTS) motes, whichmonitor temperature in a vineyard. Temperature values aregathered at 5 minutes intervals, and the data is routed to a basestation node with at maximum 8 hops. The motes areimplemented with low power microcontrollers and 916 MHzradios. For the application, the motes are supplied by large 42Ah battery packs. The reported lifetimes of the routing nodesare about 3 months.

Another WSN implementation consisting of 25 COTS typePicoNodes is implemented and measured in [5]. The nodesemploy StrongARM SA-l 100 microprocessors and Bluetoothphysical layers with custom data link layer and higher layerprotocols. In a performance evaluation, the nodes measure

temperature values at 5 s intervals, and multi-hop route data toa base station node. Measured average power consumption pernode is about 6 mW resulting approximately 50 days lifetimewith two AA-batteries.

In this paper we present the implementation of a fullfeatured WSN for temperature sensing. Compared to currentproposals, significantly longer WSN node lifetimes withsmaller node size are pursued to achieve. The system consistsof sensor nodes and a gateway to connect the network to a PCfor data analysis. The network is based on our TUTWSNwireless sensor network design that provides very high energyefficiency.The case application is a home temperature monitor

network, where the nodes are mounted on the ceilings andwalls as depicted in Fig. 1. The nodes acquire temperaturevalues at 5 s intervals with at least 1°C accuracy. The numberof nodes is 100, but this is not limited by TUTWSN. Thenetwork is ranging up to 100 meters, where the proximitybetween the nodes is at most ten meters. There are no strictdelay requirements for the data collection, but the latency perhop should be below 5 s.

The paper is organized as follows. Section 2 gives anoverview of the TUTWSN topology and its MAC protocol.The design of the WSN prototypes and the user applicationare presented in Section 3. Section 4 presents the performanceevaluation, including component power analysis andprototype performance measurements. Section 5 concludes thepaper.

II. TUTWSN OVERVIEWThe TUTWSN protocol stack is targeted at low data rate

applications with unlimited scalability and extremely lowenergy consumption. An important TUTWSN design goal hasbeen minimized communication duty cycle, while providingrapid and low energy network configuration and maintenanceoperations. TUTWSN utilizes a clustered network topology,where each cluster consists of a cluster head (headnode) andseveral subnodes. The headnodes form a flat topology withmulti-hop data routing as depicted in Fig. 1. All nodes canmove freely and the network is created and reconfiguredautonomously.

978-3-8007-2909-8/05/$20.00 ©2005 IEEE 801


Headnode a Subnode 0 Gateway I , ClustersFig. 1. TUTWSN topology in temperature monitoring application.

A. TUTWSNMACTUTWSN is based on our previous QoS capable

TUTWLAN MAC protocol [6]. It is a time slotted protocol,which divides each access cycle into an active period(superframe) and idle time as depicted in Fig. 2. Data isexchanged in superframes, while on idle time the nodes are ina power save mode. The access cycle length is an adjustableMAC protocol parameter, which allows a trade off betweenenergy, data rate and latency.

For enabling scalability, each headnode controls itssuperframe structure asynchronously to each other. Inter-cluster interferences are eliminated by allocating different RFchannels for neighboring clusters. The applied localsynchronization within a cluster reduces headnode duty cycleand is much easier to control than maintaining a globalsynchronization [7].

Each superframe begins with a beacon, which specifiesslot reservations and synchronizes data exchanges. Thebeacon is succeeded by four ALOHA and at most eightreservable time slots. The ALOHA slots are used forassociation and slot reservation requests, and for low prioritydata. The reservable slots provide reliable data transfer with afixed throughput, but they require a slot reservationprocedure. The combination of ALOHA and TDMA slots isapplicable for different traffic types and operating conditions,such as radio interference. In addition, all headnodesbroadcast network beacons on a common network signalingchannel at 500 ms intervals. The network beacon contains acluster channel and an exact time to a following superframe,

Clusterchannel

Access cycle

Supert Super-i;frame: Idle frameI -0

which effectively reduces the required radio receiving timeand energy for neighbor discovery operations and inter-clustersynchronization.

Inter-cluster communication is performed during a sourcecluster idle time and a destination cluster superframe. Asource headnode acts similarly as a subnode and synchronizesitself to the schedule of a destination headnode. In addition, asource headnode has to keep track of time offsets to allutilized destination headnodes.

III. WIRELESS TEMPERATURE SENSOR PROTOTYPES

Two prototype nodes have been implemented for thetemperature network. The TUTWSN node prototype isdesigned for very high energy efficiency and adequateperformance for operating both as a subnode and headnode. Asimpler TUTWSN subnode prototype targets to lower cost andeven smaller size.

A. TUTWSNNode PrototypeThe node prototype hardware architecture is presented in

Fig. 3. The prototype has Xemics XE88LCO2 MicroControllerUnit (MCU) including a 16-bit ADC, 22 KB program memoryand 1 KB data memory. An external 8 KB Microchip25AA320 EEPROM is included for non-volatile data memory.MCU is selected due to adequate 2 MIPS performance andhigh 1344 MIPS/W energy efficiency.

Nordic Semiconductor 2.4 GHz nRF2401A radiotransceiver provides wireless communication. The highfrequency allows physically small antennas and enoughchannels for frequency spreading between clusters. The radiohas 1 Mbps data rate, which enables very high energyefficiency. The consumed energy per a physical layer bit isabout 109 nJ at RX mode and 63 nJ at TX mode with 0 dBmpower level [8]. Transmission power is selectable between -20dBm and 0 dBm. The radio has 84 selectable RF channels,which enables good network scalability. Furthermore, theradio has an internal Cyclic Redundancy Check (CRC) errordetection and a data buffer, which reduces MCU performancerequirements. The prototype uses a quarter-wave GigaAntRufa SMD antenna due to small size, omni-directionalradiation and good 68% efficiency. Measured radio range isabout 15 m.A temperature sensing is implemented with a Maxim

MAX6607 temperature sensor, a Maxim MAX6018 voltagereference, and the integrated ADC. ADC has pre-amplification and offset compensation stages, which improve

time

§i1. T~:~ 1 l. l:~I

Beacon/ f.0 - 20 msFig. 2. TUTWSN access cycle and superframe. Fig. 3. TUTWSN node prototype hardware architecture.

978-3-8007-2909-8/05/$20.00 ©2005 IEEE

i~ iI: IN

802


Antenna

EEPROM

Radio

Test pins- _-ATemperatureJsensor 5Programming Jconnector IPower switch -

l l Regulator MCU0 10 20 mmFig. 4. The TUTWSN node prototype.

accuracy. The maximum sample rate using 16-bit resolution is2 kHz.The supply voltage is stabilized to 2.5 V by a Texas

Instruments TPS71525 linear voltage regulator. The regulatoris selected due to very low electromagnetic interferences andquiescent current. The prototype has a connector for a batterythat is selected according to measured power consumption andthe required lifetime.The TUTWSN node prototype is implemented in a PCB

presented in Fig. 4. The top side of the prototype contains theradio, antenna, EEPROM, temperature sensor, some test pinsand a programming connector. The bottom side contains MCUand the voltage regulator. The size of the prototype is 31 mmx 23 mm x 5 mm.

B. TUTWSN Subnode PrototypeThe TUTWSN subnode prototype is based on a Nordic

Semiconductor nRF24E 1 chip, which integrates a radiotransceiver, MCU and ADC on a single package. The radioPHY layer is similar to the nRF2401A providing goodcompatibility with the TUTWSN node prototype.The hardware architecture of the subnode prototype is

presented in Fig. 5. The chip includes a 8051 compatibleMCU core running at 16 MHz. The processor performance isabout 2 MIPS and has 4 KB RAM for a program and 256 BRAM for data. In addition, there is a low power RC oscillatorfor a wake-up timer. An external EEPROM is used forprogram memory.

The RC oscillator provides low, 2 ,uA sleep mode currentconsumption, but is very inaccurate. The oscillator frequencyis typically between 1 kHz and 5 kHz, depending onprocessing, temperature and supply voltage. The prototypeutilizes a Maxim MAX6607 temperature sensor and the ADC

Antenna -

nRF24E1 _

Programmingconnector

Power switch 1

Voltageconverter

EEPROM . 3

Temperaturesensor

Fig. 6. The TUTWSN subnode prototype on a 5 cent coin.

embedded in the nRF24E 1 chip. For simplicity, the prototypeuses an internal bandgab voltage reference. ADC has atmaximum 100 kHz sample rate and 12-bit resolution.

The TUTWSN subnode prototype is presented in Fig. 6.The top side of the prototype contains the nRF24E 1 chip withcrystal, antenna and a programming connector. The bottomside contains voltage converter, EEPROM, temperature sensorand a power switch. The size of the prototype is 23 mm x 13mm x 5 mm.

C. TUTWSN Gateway PrototypeThe TUTWSN gateway bridges TUTWSN to a RS-232

serial bus. The gateway utilizes the same MCU and radio thanthe TUTWSN node prototype. The RS-232 bus is accessed bya Maxim MAX3381 RS-232 transceiver and the UART ofMCU. The implementation of the WSN gateway prototype ispresented in [9].

D. User ApplicationThe user application is executed in a PC for presenting

measured data graphically and for configuring themeasurements. The user application is implemented in Javaenabling independence of underlying PC platform. Theapplication uses the Java Communication ApplicationProgramming Interface (API) that supports RS-232 serialports, and provides software porting for Windows and Linux.Above the API, the Swing technology provides GraphicalUser Interface (GUI) components [10].A screenshot of the user application is presented in Fig. 7.

The application presents the current cluster topologies onTUTWSN, and the data routing paths. These are updatedconstantly as the topology changes. Nodes are presented asinformation boxes, which contain a node ID, a logical nodetype (subnode or headnode), a sensor type, and the lastmeasured value. Sensor data can be displayed as graphicalcharts, as presented in the figure.

Fig. 5. TUTWSN subnode prototype hardware architecture.

978-3-8007-2909-8/051$20.00 ©2005 IEEE

\ \

803


Fig. 7. A screenshot of the user application.

IV. PROTOTYPE PERFORMANCEThe energy performance is measured from the implemented

prototypes. The measurements contain the static powerconsumptions of the main components of prototypes, and theaverage consumption while executing the wirelesstemperature sensing application.

A. Power AnalysisThe measurements are carried out in states where the radio,

MCU, ADC and sensor are programmed to enter differentoperation modes. The MCU power consumption is measuredin an active mode, when the MCU core is executing the sensorapplication, and in a sleep mode, when only a wake up timeris active. The ADC and sensor power consumptions aremeasured in active and shut down (off) modes. The radiopower consumption is measured in sleep, reception andtransmission modes with a minimum and maximumtransmission power level.

TABLE ITHE POWER ANALYSIS OF THE TUTWSN SUBNODE AND TUTWSN NODE

PROTOTYPES.

MCU ADC Sensor Radio TUTWSN TUTWSNsubnode power node power

active active active RX 66.03 mW 59.55 mWactive active active TX (0 dBm) 44.10 mW 37.70 mWactive active active TX (-20 dBm) 32.90 mW 26.60 mWactive active active sleep 11.64 mW 3.04 mWactive active off sleep 11.60 mW 3.01 mWactive off off sleep 10.56 mW 1.69 mWsleep off off sleep 15 ,W 24 iW

The results are presented in Table I and include the powerdissipation in the voltage converter, while the prototypes aresupplied with 3 V supply voltage. The minimum powerconsumptions for the prototypes are 15 iW and 24 ,uW, whenall components are inactive. A higher power in the nodeprototype is mostly caused by an external 32 kHz crystal thatdrives the MCU wake up timer.

The maximum power consumptions are achieved when allcomponents are in active mode and the radio is in receivemode. Then, the subnode prototype and the node prototypepower consumptions are 66.03 mW and 59.55 mW,respectively. The analysis shows the Xemics core has a veryhigh energy efficiency compared to the 8051 core. Moreover,the radio clearly dominates the momentary powerconsumption in both prototypes. Particularly, radio receivingtime is costly. The node power consumption over doublesduring the reception compared to transmission with the lowest-20 dBm transmission power.

B. MeasurementsThe average power consumptions are measured from

implemented prototypes, which execute the TUTWSN stackand the sensor application. For the measurements, the routingprotocol employs a strategy where routing is dynamic, but thelowest cost route towards the TUTWSN gateway is selectedaccording to the physical positions of the nodes. Allprototypes sample their temperature sensors at 5 s intervalsand route data to a WSN gateway using reservable data slots.TUTWSN node prototypes utilize a power saving strategy,

where data frames are transmitted and received only at thebeginning of TUTWSN MAC slots. During the remainingtime, nodes are in a sleep mode. However, the TUTWSNsubnode prototype lacks the required timing accuracy andresolution of the wake-up timer. Also, the low memoryresources limit the accuracy of the wake-up timer calibrationalgorithm. Hence, the subnode prototypes do not enter thelowest power sleep mode during the measurements. Thus, thesubnode prototype power consumption with the full powersaving features is estimated.

For the measurements, prototypes are supplied by I Fcapacitors and the average current consumptions aredetermined by the slopes of the capacitor terminal voltages.From the obtained currents, power consumptions aredetermined using the 3.0 V supply voltage.

First, the power consumptions of both prototypes aremeasured operating as subnodes. The power results with 5 s

TABLE IIMEASURED SUBNODE POWER CONSUMPTIONS AND ESTIMATED LIFETIMESWITH A CR2540 AND TWO AA BATTERIES (5 S ACCESS CYCLE LENGTH).

Average Lifetime with a Lifetime withpower CR2450 battery 2 AA batteries

TUTWSN node as 88.2 ,uW 879 days 10.1 yearsa subnodeTUTWSN subnode 10.6 mW 6.6 days 30.5 daysTUTWSN subnode 80.7 ,W 960 days 11.0 yearswith power saving(estimated)

978-3-8007-2909-8/05/$20.00 ©2005 IEEE

:d..:i

804


600

400

3300Q.

200

100

0

0

6

E

-W

3 aD!L._

a)2 oz

1

1 2 3Subnodes per headnode

Fig. 8. Measured headnode power consumptions and estimated lifetimeswith two AA batteries (5 s access cycle length).

MAC access cycle length are presented in Table II. Accordingto the measurements, the node prototype and the subnodeprototype consume 88.2 jiW and 10.6 mW, respectively. Inaddition, the estimated TUTWSN subnode prototype powerconsumption with full power saving would reduce to 80.7

gW.In the basis of the power consumption results, node

lifetimes are estimated with two battery types: a CR2450lithium coin battery and two serially connected AA alkalinebatteries. With the full power saving, the lifetimes of bothprototypes with the CR2450 are about 900 days, and with thetwo AA batteries over 10 years.

In practice, the lifetimes are limited by the battery self-discharging. The subnode prototype lifetime with the limitedpower saving is below a month. This clearly shows theimportance for the MAC layer power save functionality withcompatible hardware.

Next, the power consumption of the TUTWSN nodeprototype is measured in the headnode mode. The MACaccess cycle length is fixed to 5 s, and the number ofassociated subnodes increases from 0 to 4, which is typical inthe case application.The resulted headnode power consumption increases from

390 jiW to 560 pW, as presented in Fig. 8. To distributepower consumption more equally between nodes, theheadnode operation is to be rotated between TUTWSN nodeprototypes. In this case, the average power consumption of theTUTWSN node prototypes reduces from 390 1iW to 183 gW.This is also presented in the figure. The average latency per

hop is about 2.5 s.

The lifetime of the TUTWSN node prototype with two AAbatteries is estimated for a headnode, and averagely for a

headnode and 0 to 4 subnodes. The results are presented in theright axis of Fig. 8. Without the rotation of the headnodefunctionality, the headnode lifetime reduces from 2.3 years to1.6 years as subnodes are added. If the headnode rotation isapplied, average node prototype lifetime increases from 2.3years to 4.9 years.

Practically, the upper limit for associated subnodes is set bythe radio range and the headnode loading. The proximitybetween headnodes should not be more than ten meters. Also,in practice the network scanning caused by node failures,interferences and mobility would increase power consumptionaround 100,tW.

V. CONCLUSIONSPhysical implementations for a temperature sensing WSN

has been presented. Network energy consumption isminimized by the time slotted MAC protocol and very lowpower COTS hardware components. The measurements showthe importance of accurate timing and proper power savingmodes in time slotted MAC protocols. The powerconsumption with the rotation of headnodes and a stablenetwork is as low as 183 ,uW resulting nearly 5 years lifetimewith two AA batteries. Compared to the Mica II typebackbone nodes in [4], a headnode lifetime is 100 to 300 timeslonger depending on the rotation of the headnodes. Comparedto the PicoNodes in [5], TUTWSN node average lifetime isover 30 times longer.A reasonable usability is achieved by the long node

lifetimes and the fully autonomous network operation. Thepower consumption is low enough for scavenging the supplyenergy purely from ambient energy, including low levelvibrations, ambient light and temperature gradients.

REFERENCES[1] E. H. Jr. Callaway, and E. H. Callaway, Wireless sensor networks:

architectures andprotocols, Auerbach Publications, 2003.[2] J. N. Al-Karaki, and A. E. Kamal, "Routing techniques in wireless

sensor networks: a survey," IEEE Wireless Communications, vol. 11, no.6, pp. 6-28, Dec. 2004.

[3] J. Weatherall and A. Jones, "Ubiquitous networks and theirapplications," IEEE Wireless Communications, vol. 9, no. 1, pp. 18-29,Feb. 2002.

[4] R. Beckwith, D. Teibel, and P. Bowen, "Report from the field: resultsfrom an agricultural wireless sensor network," In Proc. IEEE Conf: onLocal Computer Networks, Nov. 2004, Tampa, FA, pp. 471-478.

[5] J. M. Reason, and J. M. Rabaey, "A study of energy consumption andreliability in a multi-hop sensor network," Mobile Computing andCommunications Review, vol. 8, no. 1, pp. 84-97, Jan. 2004.

[6] M. Hannikainen, T. Lavikko, P. Kukkala, and T. Hiimalaiinen,"TUTWLAN - QoS supporting wireless network," TelecommunicationSystems - Modelling, Analysis, Design and Management, vol. 23, no. 3-4, Kluwer Academic Publishers, 2003, pp. 297-333.

[7] C. R. Lin, and M. Gerla, "Adaptive clustering for mobile wirelessnetworks," IEEE Journal on Selected Areas in Communications, vol. 15,no. 7, pp. 1265-1275, Sep. 1997.

[8] [14] M. Kohvakka, M. Hannikainen, T. D. Hamailainen, "Energyoptimized beacon transmission rate in a wireless sensor network," inProc. IEEE Int. Symp. on Personal Indoor and Mobile RadioCommunications, Gernany, Sep. 2005, accepted.

[9] M. Kohvakka, M. Hannikainen, T. D. Hamalaiinen, "Wireless sensornetwork implementation for industrial linear position metering," In Proc.IEEE Euromicro Conf: on Digital Systems Design, Aug. 2005, Portugal,accepted.

[10] Java Foundation Classes (JFC/Swing), ltt_: 'java.suncomfproductsjfc/

978-3-8007-2909-8/05/$20.00 ©2005 IEEE

TUTWSN node as a headnode (power)--TUTWSN node average (power)

TUTWSN node as a headnode (lifetime)-c TUTWSN node average (lifetime) J

500

805

PUBLICATION 9

M. Kohvakka, M. Hännikäinen, and T. D. Hämäläinen, “Wireless Sensor NetworkImplementation for Industrial Linear Position Metering,” in Proceedings of the 8thEuromicro Conference on Digital System Design – Architectures, Methods, and Tools(DSD 2005), Porto, Portugal, Aug. 30–Sept. 3, 2005, pp. 267–273.

© 2005 IEEE. Reprinted, with permission, from the proceedings of the 8th EuromicroConference on Digital System Design – Architectures, Methods, and Tools 2005.


Mikko Kohvakka, Marko Hännikäinen, and Timo D. Hämäläinen

Tampere University of Technology Institute of Digital and Computer Systems P.O.Box 553, FI-33101 Tampere, Finland

[email protected]

Abstract

This paper presents the design and performance

measurements of a prototype Wireless Sensor Network (WSN) for industrial linear position metering. Design includes two different prototype platforms and a user application. Prototypes combine energy efficient commercial off-the-shelf components including a 2.4 GHz radio, and the custom TUTWSN communication protocols resulting high robustness, autonomous operation and very low power consumption. The user application displays sensor data graphically and enables further data analysis. Measurements contain component power analysis and prototype performance measurements. The measurements indicate 200 µW to 400 µW average node power consumption, as 16-bit sample is measured with 1 Hz sample rate, and routed to a WSN gateway with 1 s latency per hop and 512 bps throughput between nodes. Predicted lifetime of implemented WSN is 2 months with a small rechargeable battery or over 2 years with two AA batteries. 1. Introduction

Wireless Sensor Networks (WSN) nodes are characterized as very small, low energy and cheap platforms that are able to sense, process data and communicate wirelessly. Suitable applications are found from home, outdoor and industrial environments [1].

A fundamental WSN research challenge has been the combination of a complex network management with very scarce energy resources. Wireless communication protocols and algorithms are usually very complex, allowing fully autonomous ad-hoc operation in large scale networks. Network control is distributed to the whole network, which enables high robustness against local radio path interferences and node failures. WSN nodes are

deployed in the area of interest as a sensor field, where data is routed using multiple low energy hops [2], [3].

For practical reasons, nodes should operate by small batteries for years, or they have to scavenge all supply energy from their operating environment [4]. This requires very high energy efficiency in both communication protocols and hardware implementation, which both have to be tailored according to application requirements.

WSN nodes have been designed by several research groups over the last decade. Most often they are implemented by commercial off-the-shelf (COTS) components, which provide sufficient performance with low development and production costs.

BTnode [5] is a demonstration platform for ad-hoc networks. BTnode combines an Ericsson ROK 101 007 Bluetooth module with an Atmel ATmega128L microcontroller. BTnode supports only a star network topology with one master and at maximum seven slave nodes. However, a multi-hop WSN can be implemented by dual-radio BTnodes [6]. Reported power consumption of a connected dual-radio BTnode is 445 mW. Due to the high energy consumption, BTnodes are most suitable for applications that are active over a limited time period, while requiring high network throughput.

PicoNode combines a Bluetooth physical layer with custom data link layer and higher layer protocols, which are executed on a StrongARM SA-1100 microprocessor. The performance of a temperature sensing WSN with 25 PicoNodes is evaluated in [7]. Temperature is measured at 5 s intervals and data is multi-hop routed to a base station node. Reported average power consumption per node is about 6 mW resulting approximately 50 days lifetime with two AA-batteries.

Berkeley Mica II mote [8] is a commercial platform, which consists of an Atmel ATmega128L microcontroller and a 916 MHz radio. A WSN implementation with 65 Mica II motes deployed in a vineyard is presented in [9]. The motes gather temperature values at 5 minutes intervals and multi-hop route data to a base station node. The

Wireless Sensor Network Implementation for Industrial Linear Position Metering

reported lifetimes of routing nodes are about 3 months with large 42 Ah battery packs.

In this paper we present the implementation of a full featured WSN for industrial linear position metering. The work includes two different prototype platforms and a user application. Linear position sensors are widely used in industry for measuring the displacements of moving parts. WSN eliminates the need of extensive cabling, thus reducing both material and installation costs. A new WSN design has been selected, as current proposals do not fulfill the performance and lifetime requirements of the metering application. The physical WSN prototypes are implemented using low power COTS components and custom TUTWSN communication protocols.

The rest of this paper is organized as follows. The design requirements for the new WSN are presented in Section 2. Section 3 gives an overview of the TUTWSN topology and its MAC protocol. The design of WSN prototype platforms is presented in Section 4. Section 5 presents the performance evaluation, including component power analysis and prototype performance measurements. The paper is concluded in Section 6.

2. Application Design Requirements

The WSN application design contains a complete linear

position metering system from a sensor to data analysis on a workstation. The design utilizes TUTWSN as a base technology, including its protocol stack and network architecture. Two different prototypes are needed, one for the linear position sensor, and one for a WSN gateway that connect the WSN to other networks and terminals.

The linear position metering system measures low frequency movements, where data is gathered from sensors at 1 Hz sample rate and 16-bit resolution. All samples are time stamped. The expected number of sensors is 50, but this should not be limited. The network range is one hundred meters, where sensors are at most ten meters apart from each other. WSN data is multi-hop routed in the network while the targeted latency per hop is 1 s. The required network throughput is 512 bits/s. There are no strict delay requirements for the data collection, but the reliability of the WSN is critical. Therefore, the node must be able to buffer and retransmit samples.

A wireless linear position sensor is assembled by two parts, as a WSN node is attached on the physical connector of a linear position sensor. Thus, the WSN nodes should not increase the size or weight of the existing physical sensor significantly. The sensors will operate in harsh physical environment containing fluids, dust, vibration and electromagnetic interferences. Hence, WSN nodes should be sealed on waterproof metal containers. The sensors should be powered by rechargeable batteries having at least 1 month lifetime.

3. TUTWSN Overview TUTWSN protocol stack has been developed for low

data rate applications, where good scalability and low energy consumption are required. An important TUTWSN design goal has been minimized communication duty cycle, while providing rapid and low energy network configuration and maintenance operations. TUTWSN is based on our previous experience on QoS supporting, proprietary WLAN MAC implementation [10].

TUTWSN utilizes clustered network topology, presented in Figure 1. Each cluster consists of two logical types of nodes: a single cluster head (headnode) and several members (subnode). Headnodes maintain cluster synchronization and participate in multi-hop data routing in network. Subnodes can communicate only with headnodes, thus having significantly lower performance and energy requirements than headnodes. Network contains also one or more WSN gateways, which request data from WSN nodes and form bridges to other networks.

The implemented linear sensing application employs homogenous nodes, which select their logical type (headnode or subnode) dynamically during operation. The selection is made according to heard neighboring clusters and their loading. The portion of subnodes is higher in densely deployed parts of the network for achieving energy saving. Network lifetime is extended by actively rotating the headnode functionality among nodes. 3.1 TUTWSN MAC

TUTWSN utilizes a time slotted MAC protocol, which divides each access cycle into an active period for data exchanging (superframe) and an idle time, when nodes are in a power saving mode. These are presented in Figure 2. The slotted structure allows a trade off between energy and data rate. The duty cycle is further reduced by applying slotted-ALOHA and Time Division Multiple Access (TDMA) mechanisms in superframes. For

Headnode Subnode Gateway Cluster

Analyzer PC

Wireless linearposition sensors

Figure 1. TUTWSN clustered ad-hoc topology.

enabling scalability, each headnode controls its superframe structure asynchronously to each other. Inter-cluster interferences are eliminated by allocating different RF channels for neighboring clusters. Applied local synchronization in clusters reduces headnode duty cycle and is much easier to control than a global synchronization [11].

Each superframe begins with a beacon, which is succeeded by four ALOHA and at most eight reservable time slots, each 20 ms long. These are presented in Figure 2. The beacon synchronizes a cluster, specifies time slot assignments and provides cluster status information. ALOHA slots are used for association and slot reservation requests, and for low priority data. Reservable slots provide reliable data transfer with fixed throughput, but require a slot reservation procedure. In addition, all headnodes broadcast network beacons on a network signaling channel at 500 ms intervals. The network beacon contains an exact time to a following superframe, which effectively reduces the required radio receiving time and energy for network maintenance operations.

Inter-cluster communication is performed during source cluster idle time and destination cluster superframe. A source headnode acts similarly as a subnode and synchronizes itself to the schedule of a destination headnode. Source headnode has to keep track of time offsets to utilized destination headnodes.

4. Prototype WSN Design

Prototype WSN design includes hardware prototype

implementations for a WSN node and a WSN gateway, and software implementations for a WSN application in nodes and a user application in the analyzer PC.

4.1 WSN Node

A WSN node prototype is an improved version of the

WSN prototype platform presented in [12]. The hardware architecture for the WSN node is presented in Figure 3. The architecture is divided into four subsystems: communication, computing, sensing and power subsystems.

The communication subsystem provides wireless communication links to neighboring nodes. The subsystem consists of a Nordic Semiconductor nRF2401A radio transceiver operating at 2.4 GHz license free Industrial Scientific Medicine (ISM) frequency band. The 2.4 GHz frequency band is high enough for avoiding interferences from machinery [13] and for allowing physically small and efficient antennas. The radio has 1 Mbps data rate, which enables very high energy efficiency. The consumed energy per a physical layer bit is about 109 nJ at RX mode and 63 nJ at TX mode with 0 dBm power level [14]. The transmission power level is selectable between -20 dBm and 0 dBm. The radio has 84 selectable RF channels, which allows very good network scalability. Radio has an internal Cyclic Redundancy Check (CRC) error detection logic and a data buffer, which reduces microcontroller speed and time accuracy requirements. Antenna is implemented with a quarter-wave external monopole antenna due to good efficiency and omni-directional radiation. Measured radio range is about 30 m.

The computing subsystem manages node operations and executes sensor application. The subsystem consists of a Xemics XE88LC02 MicroController Unit (MCU), which integrates CoolRisc 816 processor core with a 16-bit Analog-to-Digital Converter (ADC), 22 KB program memory and 1 KB data memory. Maximum instruction execution speed is 2 MIPS. MCU is connected to an 8 KB Microchip 25AA320 EEPROM for non-volatile data memory. MCU is selected due to low 2.4 V supply voltage, high 1344 MIPS/W energy efficiency, and accurate ADC. Moreover, low 5 µW sleep mode power consumption, versatile timers and good availability are the advances of the MCU.

The sensing subsystem gathers requested data from node environment. The subsystem consists of a linear position sensor and the ADC integrated with MCU. The sensor utilizes a 5 kΩ linear potentiometer. During measurement, a constant voltage is applied over the potentiometer and the deviation is determined by the wiper voltage. ADC has pre-amplification and offset compensation stages, which increase sampling accuracy. The maximum sample rate using 16-bit resolution is 2 kHz.

time

Super−frame

Beacon

Clusterchannel

Access cycle

20 ms

Idle

ReservableALOHA

Super−frame

Figure 2. TUTWSN access cycle and superframe.

Power Subsystem

VoltageCoverterTPS71525

CommunicationSubsystem

Battery3V150

RadionRF2401

MCUXE88LC02

SensingSubsystem

ComputingSubsystem

EEPROM25AA320

ADC16−bit

LinearPosition Sensor

Figure 3. WSN node prototype architecture.

The power subsystem utilizes three serially connected 150 mAh Varta V150H NiMH batteries achieving a 3.6 V nominal voltage. The NiMH chemistry is selected due to good energy density, cheap price and better robustness against high temperatures and currents than lithium chemistries. The battery voltage is converted down to 2.5 V supply voltage by a Texas Instruments TPS71525 linear voltage regulator. The converter is selected due to low electromagnetic interferences and quiescent current. These are important, since proximity to other components is short and the average current consumption is low. Moreover, the converter is small and cheap. For the required one month WSN node lifetime, the power subsystem can deliver 208 µA average current drain.

The WSN node prototype is implemented in a four-layer FR-4 Printed Circuit Board (PCB), presented in Figure 4. The top side of the prototype contains radio, battery, programming connector, antenna and two diagnostics LEDs. The bottom side contains MCU, voltage regulator and a connector for the linear position sensor. The prototype is enclosed in an aluminum casing presented in Figure 5. The size of the prototype is 120 mm x 25 mm x 25 mm. The connector for the linear position sensor is also used for battery recharging.

4.2. WSN Gateway

A WSN gateway forms a bridge between WSN and an analyzer PC using a RS-232 serial port.

The WSN gateway prototype architecture is presented in Figure 6. The gateway utilizes the same MCU, radio and antenna than the WSN node. These are selected due to the high energy efficiency and radio compatibility with the WSN node prototypes. RS-232 bus is accessed by a Maxim MAX3381 RS-232 transceiver and a Universal Asynchronous Receiver Transmitter (UART) of MCU.

The required supply power is obtained directly from the

serial port; the power is rectified and filtered from Data Terminal Ready (DTR) and Request to Send (RTS) signals of RS-232. According to measurements, the these signals deliver 12 V open circuit voltage and 5 mA to 10 mA short circuit current, depending on PC hardware. The obtained voltage is converted down to 2.7 V supply voltage by a Texas Instruments TPS71501 linear voltage regulator.

The WSN gateway prototype is implemented in a four-layer RF-4 PCB, presented in Figure 7. The upper side of the prototype contains antenna, MCU, RS-232 transceiver, RS-232 connector, filter capacitors and two diagnostics LEDs. The bottom side contains radio, voltage rectifier diodes, voltage regulator and programming connector. The size of the prototype is 67 mm x 31 mm x 37 mm including antenna.

4.3. User Application

The user application is executed in the analyzer PC for presenting measured data graphically and for configuring the measurements. There are no power switches available on the prototypes itself, but the network is activated by the user applications.

The user application is implemented in JAVA enabling independence of underlying PC platform. The user application is based on a Java Communication API. The API supports RS-232 serial ports, and provides software porting for Windows and Linux. Above the Java Communications API, Swing provides User Interface (UI) components, which are utilized for implementing a Sensor UI.

A screenshot of the user application is presented in

Antenna

EEPROM

Sensorconnector

Radio

MCUVoltageregulator

Battery


DiagnosticsLEDs

Figure 4. WSN node prototype.

Linear position sensor

WSN nodeprototype

Moving axis for position measuring

Figure 5. Complete wireless linear position sensor prototype consisting of a WSN node attached to a position sensor.

VoltageCoverterTPS71501

RadionRF2401

MCUXE88LC02

RS−232 Driver

MAX3381

RS

−232

Con

nect

or

Figure 6. WSN gateway prototype architecture.

Figure 8. The application shows the current cluster topologies on WSN, and the data routing paths. These are updated constantly as the topology changes. Nodes are presented as information boxes, which contain a node ID, a logical node type (subnode or headnode), a sensor type and a current sensor value. Sensor data can be presented graphically for individual sensors or as jointed graph, as presented in the figure.

5. Prototype Performance

Power performance is measured from the implemented

WSN node prototype. The measurements include static power consumptions of the main components and an average consumption while executing the wireless linear position metering.

5.1. Power Analysis

The static power consumptions of the radio, MCU, ADC and the linear position sensor are measured at different operating modes. The MCU power consumption is measured in an active mode, when the core is running on 1.8 MIPS speed, and in a sleep mode, when only wake up timer is active. The ADC and sensor power consumptions are measured in active and shut down (off) modes. The radio power consumption is measured in various operation modes, including data loading between MCU and the radio buffer.

Measurement were done at 3 V supply voltage

including power dissipation in the power subsystem. The analysis results are presented in Table 1. The minimum power consumption is 21 µW, which is achieved when all components are inactive. The maximum required power is 57.60 mW, when all components are active and radio is in receive mode. In the implemented prototype, the radio clearly dominates the momentary power consumption. Particularly, radio receiving time is costly. The node power consumption over doubles during reception compared to transmission with the lowest -20 dBm transmission power. The linear position sensor power consumption is 960 µW.

MCU

RS−232 connector

RS−232 transceiver

Radio

Antenna


Voltageregulator

DiagnosticsLEDs

Voltagerectifiers

Filtercapacitors

Figure 7. WSN gateway prototype.

Figure 8. A screenshot of the user application.

5.2. Measurements

The average power consumptions are measured from a set of implemented WSN node prototypes, which execute the TUTWSN stack and the sensor application. For the measurements, the logical types of nodes are fixed. The routing protocol employs a strategy where routing is dynamic, but the lowest cost route towards the WSN gateway is established according to the physical positions of the nodes. All WSN node prototypes sample their linear position sensors at 1 Hz rate and route data to the WSN gateway prototype using reservable data slots. For achieving accurate time stamps, nodes execute a time synchronization algorithm. The synchronization algorithm loads the MCU notably, especially in headnodes. An oscilloscope screenshot of the headnode and subnode current waveforms are presented in Figure 10. The waveforms present a 20 ms time period, which contains ADC sampling, and TUTWSN data transmission and

acknowledgement reception. A peak current during reception is about 20 mA.

For the measurements, a headnode and a subnode are supplied by 1 F and 0.22 F super capacitors. The average current consumptions are determined by measuring the slopes of the capacitor terminal voltages. From the obtained currents, power consumptions are calculated using the 3.0 V supply voltage.

First, the average subnode and headnode power consumptions were measured, while MAC access cycle length is increased from 1 s to 10 s. Only one subnode is associated on each headnode. The power consumption results with maximum throughputs between headnodes are presented in Figure 9. The minimum power is obtained with 10 s access cycle, when subnode and headnode power consumptions are 59.6 µW and 302 µW, respectively. The available throughput is 102 bits/s. The throughput is inversely proportional to the TDMA access cycle length. With 1 s access cycle, throughput is increased to 1024 bits/s, but the power consumptions of a subnode and a headnode are raised to 249 µW and 968 µW, respectively. The average latency per hop is about a half of the access cycle length.

Then, the average headnode power consumption was measured, while the number of subnodes per each headnode was increased from 1 to 5. The TUTWSN MAC access cycle length is fixed to 2 s, which fulfills application requirements in data rate and latency. As the number of subnodes increase, the subnode power remains in 162 µW, but the headnode power consumption increases. The resulted headnode and network average power consumptions, and predicted WSN node lifetimes are presented in Figure 11. The increase of subnodes per headnode from 0 to 5 reduces the average node power consumption from 400 µW to 299 µW and increase node

Table 1. WSN node prototype power analysis.

1.69Off

0.021Sleep

3.01

Off

3.06

Sleep

4.70Data loading

26.46Transmit data (Power: -20 dBm)

39.73Transmit data (Power: 0 dBm)

57.60Receive data

ActiveActive

Active

Power (mW)Radio modeSensor

modeADC mode

MCU mode

1.69Off

0.021Sleep

3.01

Off

3.06

Sleep

4.70Data loading

26.46Transmit data (Power: -20 dBm)

39.73Transmit data (Power: 0 dBm)

57.60Receive data

ActiveActive

Active

Power (mW)Radio modeSensor

modeADC mode

MCU mode

0

100

200

300

400

500

600

700

800

900

1000

1 2 3 4 5 6 7 8 9 10T ac (s)

P (µ

W)

0

200

400

600

800

1000

1200

Hea

dnod

e th

roug

hput

(bps

)

Headnode power consumptionSubnode power consumptionHeadnode throughput (right axis)

Figure 9. Power consumption and throughput in function of TUTWSN MAC access cycle length (1 subnode per headnode).

ADC sampling

Data transmission

Acknowledgereception

Acknowledgetransmission

Data reception

Subnode

Headnode

Figure 10. Oscilloscope screenshot of the headnode and subnode current waveforms during ADC sampling, data transmission and reception.

lifetime from 47 days to 63 days. The required 31 days battery recharging interval is clearly exceeded regardless the number of subnodes per headnode.

Practically, the number of subnodes per each headnode is determined dynamically. To minimize energy consumption and improve scalability, the number of subnodes per headnode is maximized. The upper limit is set by radio range and headnode loading. The proximity between headnodes should be no more than ten meters.

6. Conclusions

A prototype WSN is implemented for industrial linear

position metering including two different hardware prototypes and a user application. For reducing energy consumption and improving scalability, a clustered network topology is used. Yet, the rotation of the headnode operation among nodes is needed to distribute energy consumption more equally and increase network lifetime. The measurements are carried out on a stable network without network discovery operations. In practice, interferences and node failures require occasional network scanning operations, which increase headnode power consumption around 100 µW [14]. However, required 31 days lifetime is clearly exceeded. Significantly longer lifetime would be obtained using primary batteries. For instance, two AA-type alkaline batteries would extend node lifetime to 27 months.

Results indicate that WSN technology is matured enough for real applications with high reliability requirements. The design and tailoring of WSN protocols according to the application is required for adequate energy efficiency and performance. Also, careful platform design and component selection for supporting protocol operation is required.

References

[1] E.H.Jr. Callaway, and E.H. Callaway, Wireless sensor networks: architectures and protocols, Auerbach Publications, 2003. [2] J.M. Rabaey, J. Ammer, J.L.Jr. da Silva, D.Patel, and S.Roundy, “PicoRadio supports ad hoc ultra-low power wireless networking”, IEEE Computer Magazine, vol. 33, no. 7, Jul. 2000, pp. 42-48. [3] J. Weatherall and A. Jones, “Ubiquitous networks and their applications”, IEEE Wireless Communications, vol. 9, no. 1, Feb. 2002, pp. 18-29. [4] S. Roundy, P.K. Wright, and J. Rabaey, “A study of low level vibrations as a power source for wireless sensor nodes”, Computer Communications, vol. 26, no. 11, Jul. 2003, pp 1131-1144. [5] BTnodes homepage. http://www.btnode.ethz.ch/ [6] M. Leopold, M.B. Dydensborg, and P. Bonnet. ”Bluetooth and Sensor Networks: A Reality Check”, in Proc. 1st ACM Conf. Embedded Networked Sensor Systems, Los Angeles, Nov. 2003, pp. 103-113. [7] J.M. Reason, and J.M. Rabaey, “A study of energy consumption and reliability in a multi-hop sensor network”, Mobile Computing and Communications Review, vol. 8, no. 1, Jan. 2004, pp. 84-97. [8] Crossbow Technology homepage. http://www.xbow.com/ [9] R. Beckwith, D. Teibel, and P. Bowen, “Report from the field: results from an agricultural wireless sensor network”, In Proc. IEEE Conf. on Local Computer Networks, Tampa, Florida, Nov. 2004, pp. 471-478. [10] M. Hännikäinen, T. Lavikko, P. Kukkala, and T. Hämäläinen, ”TUTWLAN - QoS supporting wireless network”, Telecommunication Systems - Modelling, Analysis, Design and Management, vol. 23, no. 3-4, 2003, pp. 297-333. [11] C.R. Lin, and M. Gerla, “Adaptive clustering for mobile wireless networks”, IEEE Journal on Selected Areas in Communications, vol. 15, no. 7, Sep. 1997, pp. 1265-1275. [12] M. Kohvakka, M. Hännikäinen and T. Hämäläinen, “Wireless sensor prototype platform”, in Proc. IEEE Int. Conf. Industrial Electronics, Control and Instrumentation, Virginia, Nov. 2003, pp. 1499-1504. [13] T.S. Rappaport, “Indoor radio communications for factories of the future”, IEEE Communications Magazine, vol. 27, no. 5, May 1989, pp. 15-24. [14] M. Kohvakka, M. Hännikäinen, T. D. Hämäläinen, ”Energy optimized beacon transmission rate in a wireless sensor network”, in Proc. IEEE Int. Symp. on Personal Indoor and Mobile Radio Communications, Germany, Sep. 2005, accepted.

200

300

400

500

600

700

800

900

1000

0 1 2 3 4 5Subnodes per headnode

P (µ

W)

40

45

50

55

60

65

70

WSN

nod

e lif

etim

e (d

ays)

Heanode powerNetwork average powerNetwork lifetime (right axis)

Figure 11. Headnode, subnode and network average power consumptions in the function of the number of subnodes per headnode (TUTWSN access cycle length = 2 s).

PUBLICATION 10

M. Kohvakka, T. Arpinen, M. Hännikäinen, and T. D. Hämäläinen, “High-PerformanceMulti-Radio WSN Platform,” in Proceedings of the 2nd International Workshop onMulti-hop Ad Hoc Networks: from theory to reality (REALMAN 2006), Florence,Italy, May 26, 2006, Italy, pp. 95–97.

©ACM, 2006. Reprinted with Permission.

High-Performance Multi-Radio WSN PlatformMikko Kohvakka

Tampere University of Technology

Institute of Digital and Computer Systems

33720 Tampere, Finland +358 3 3115 4702

[email protected]

Tero Arpinen Tampere University of


33720 Tampere, Finland +358 3 3115 4588

[email protected]

Marko Hännikäinen Tampere University of


33720 Tampere, Finland +358 3 3115 3837

[email protected]

Timo D. Hämäläinen Tampere University of


33720 Tampere, Finland +358 3 3115 3385

[email protected]

ABSTRACT This paper presents the design and implementation of a unique multi-radio Wireless Sensor Network (WSN) platform compared to current WSN nodes that have only one radio interface. Four independent, low energy radio transceivers allow simultaneous reception and transmission to either another multi-radio WSN node or up to four different WSN nodes. This also enables high interference tolerance, low latency, and high mesh-networking performance. The platform is based on synthesizable multi-processor System-on-Chip implementation on FPGA. The radios are compatible with ultra-low energy microcontroller based WSN nodes, which can be freely mixed in the network. The platform applicability is demonstrated by an ultra low latency WSN router and high data rate file transfer applications. Theoretical hop delay is as low as 106 μs, while 3.3 Mbps network throughput is achievable. Performance measurements for up to four parallel Nios II softcore processors are also presented.

Categories and Subject Descriptors B.4.1 [Input/Output and Data Communications]: Data Communications Devices – Processors, Receivers (e.g., voice, data, image), Transmitters

General Terms Design, experimentation, performance

Keywords Wireless sensor network, platform, multi-radio, multi-processor.

1. INTRODUCTION Wireless Sensor Networks (WSN) are useful in many civil and military applications [1]. WSNs may consist of thousands of fully autonomous nodes, which gather and process data, and communicate with each other using a dynamic multi-hop mesh networking.

To save energy, it is reasonable to reduce the amount of transferred data by data fusion in WSN nodes [1]. In addition, applications like access control require data authentication and encryption, such as Advanced Encryption Standard (AES). Although WSNs are usually considered as very low data rate networks, there might occur areas with high data rate densities.

The combination of low power consumption with complex computing and networking set very high challenges for the design of hardware architecture, communication protocols and data processing algorithms.

In this paper we present the design and implementation of a new type of re-configurable WSN platform. It can be used as a router, a multi-link node, a gateway, or high-performance backbone network nodes for simpler WSNs.

The implemented platform utilizes up to four parallel radio transceivers. The use of multiple radios increases significantly network performance by allowing simultaneous data receptions and transmissions with several neighbors. The radios are compatible with simpler microcontroller based WSN nodes [2], which can be freely mixed in the network. We also present reference application architectures targeting to ultra low delay and high data rate implementations.

This paper is organized as follows. The platform hardware is presented in Section 2. Section 3 presents a multi-processor architecture implemented on FPGA. Reference application architectures for low delay and high throughput implementations are presented in Section 4. Finally, conclusions are given in Section 5.

2. HARDWARE IMPLEMENTATION The hardware architecture of the platform is presented in Figure 1. All the processing takes place on Altera Cyclone EP1C20 [3] FPGA. The FPGA is large enough for implementing several embedded processor cores, system memory, and custom hardware accelerators. The Cyclone FPGA contains 20,060 Logic Elements (LEs) and 294,912 bits of embedded dual-port RAM. These resources are suited for implementing versatile multi-processor architectures. Further, custom hardware accelerators can be synthesized into the FPGA for accelerating functions that are performed faster and with lower energy on dedicated logic than on a processor.

Normally the FPGA configuration is loaded from an on-board

Copyright is held by the author/owner(s). REALMAN'06, May 26, 2006, Florence, Italy. ACM 1-59593-360-3/06/0005.

RadionRF2401

Processors and custom logic

FPGAAltera Cyclone

User interface

I/O interfaces

Battery4xAA

Voltage Coverter2 x LM317

DC input

1.5V 3.3V

RadionRF2401Radio

nRF2401Radio

nRF2401 Figure 1. Platform hardware architecture.

95

flash memory on power up. Alternatively, Altera ByteBlaster is used to configure the FPGA.

Wireless communication use four 2.4 GHz Nordic Semiconductor nRF2401A [4] radio transceivers. The radios have 1 Mbps data rate and 83 selectable frequency channels.

A User Interface (UI) consists of two push-buttons and LEDs, and a 7-segment display. In addition, digital I/O interfaces are provided for sensors, actuators and other network interfaces that can be freely utilized for custom extensions.

The platform can operate as stand-alone node in the WSN, but it can also be connected to a PC computer for debugging and monitoring of internal operation. This is enabled by RS-232 serial port. PC can also accommodate some of the WSN application or protocol functionality especially during the development phase.

The platform is powered either by four AA-type batteries or a mains power adapter, which is converted to 1.5 V and 3.3 V supply voltages by two LM317 linear voltage regulators.

The multi-radio WSN platform is implemented with three different types of printed circuit boards named as a mother board, a radio board and an I/O board. The complete WSN platform is presented in Figure 2. The platform size is 125 mm x 112 mm x 30 mm.

3. MULTI-PROCESSOR ARCHITECTURE The implemented multi-processor architecture composed of multiple Nios softcore processors is presented in Figure 3. Each subsystem includes a Nios II core connected to a timer and UART modules, and local data memory. The system peripherals are connected together by Avalon switch fabric. The Inter-Processor Communication (IPC) is performed via a shared memory. The required memory footprint is reduced by using shared instruction memory, and thus, the same program code for all the processors. Data memory spaces are local for each processor. The UART modules are connected to a UART multiplexer for sharing the single physical RS-232 port. The architecture includes four custom logic radio interface modules for accessing radio front-ends on the platform board. The Nordic radios are used in ShockBurst mode, when data framing and error detection are covered by the radio circuitry. A centralized radio access controller forwards data and interrupts between radio interfaces and processors.

The processors are 32-bit Nios II fast cores with 512 bytes of

instruction cache. Table 1 lists the total number of used LEs and on-chip memory (excluding program codes), maximum operating frequency (Fmax), and measured current consumption, as the number of processors is increased from 1 to 4. Due to limited memory resources, architecture instances up to four processors are preferred, and the unutilized LEs are allocated for custom logic. The current consumption of the mother board is measured, while executing a simple ROM-monitor program on the CPUs at 50 MHz. According to the measurements, the current consumed by the FPGA is practically linearly dependent on the number of the processors used. Moreover, radio current consumption is 18 mA in reception mode, and 13 mA in transmission mode (0 dBm).

4. REFERENCE APPLICATIONS The multi-radio WSN platform enables radically new WSN applications that have not been possible in the past. Four independent, low energy radio transceivers allow simultaneous reception and transmission to either another multi-radio WSN node or up to four different WSN nodes. This gives low latency, and high mesh-networking performance. High tolerance against link errors can be achieved by transmitting redundant data in different routes and frequency channels. It is also possible to use even diverse radios in different frequency bands, e.g. 2.4 GHz and 433 MHz. The hardware re-configurability allows heterogeneous platforms, where each node can be optimized for a unique task by special hardware accelerators and processing architecture.

The applicability of the implemented platform is demonstrated by two reference applications, which are targeting to minimize routing latency, and to maximize routing throughput.

Ultra Low Latency Router The aim of an ultra low latency WSN router realization is to minimize the routing delays over established multi-hop paths. In current WSN implementations, a hop delay may be very significant due to data buffering and channel access delays. The use of multiple hops makes thing worse, since the delay is repeated on each hop. Hence, the realization of multi-hop WSNs with high real-time performance has been very difficult.

Radio boards

I/O boardMother board

Cyclone device

Programming connector

DC input

Battery terminals (bottom side)

Power switch

7-segment display

Analog and digital I/O

LEDs and push-buttons

Figure 2. Implemented multi-radio WSN platform.

Nios IIData memory

Timer module

Shared instruction memory

Uart module

IPC memory

Radio accesscontroller

Radiointerface

Radiointerface

Radiointerface

Radiointerface

Cyclone FPGA

Nios II subsystem 1 Nios II subsystem 2

Nios II subsystem 3

Nios II subsystem n

UARTMultiplexer

From subsystems 2 - n

Avalonconnections

:

Figure 3. Implemented multi-processor architecture.

Table 1. Resource utilization for 1 to 4 CPUs. NiosIICPUs

Area [LEs]

Area[%]

Memory [kB]

Memory [%]

Fmax [MHz]

I [mA]

1 3 966 19 2.5 7 85.58 149.9 2 6 988 34 5.5 15 71.92 233.3 3 9 686 48 8.5 24 67.78 280.7 4 12 463 62 11.0 31 60.00 338.3

96

Implemented WSN router operates in a mesh network, as presented in Figure 4. To achieve highest throughput, a Frequency Division Multiple Access (FDMA) MAC protocol is used. A dedicated frequency channel for each radio link allows purely simultaneous data exchanges with four neighboring nodes. Nodes may establish multi-hop routing paths over the network, as from Source node to Destination node presented in the Figure 4. We present a method for delay minimization for established routing paths by the use of multiple radios.

An auto-forwarding architecture that is implemented on a link layer identifies the packets that need to be forwarded [5], and redirects them to an appropriate parallel radio. The implementation is well reasoned, since in typical sensor network scenarios around 65% of all packets received by a sensor node need to be forwarded [5]. An overview of the proposed link layer operation is presented in Figure 5, where data from Node A is forwarded to Node C through Node B by using two parallel radio links (links 1 and 2). Data is transmitted in frames consisting of a protocol header (H) and data payload (D). After a successful reception of data, a recipient responds with an acknowledgement (A). Node B can identify a data frame forwarded to Node C within a time duration tdelay, which consists of a radio wave propagation time (tprop), a header reception time (tH), and a short header decoding delay (tdec). Since parallel computing and communication architecture is used, data frame forwarding may begin right after tdelay, thus, equaling the hop delay. With 96-bit protocol header, 1 Mbps radio data rate, 30 m hop distance, and 10 μs header decoding delay, a theoretical hop delay is 106 μs.

High Data Rate File Transfer The ultra low latency WSN router design can be extended for high data rate applications by the use of multiple parallel routes. By the use of four parallel transceivers the source node distributes data to four neighbors, which route frames towards destination node through four different routes, as presented in Figure 4. The intermediate nodes utilize the presented low latency auto-forwarding architecture.

Assuming FDMA-type MAC protocol with unique frequency channels for each radio link, and acknowledged data exchanges, the network throughput (T) can be estimated as

,2

f H

f A sw

L LT nR

L L t R−

=+ +

(1)

where Lf, LA and LH are the lengths of a data and acknowledgement frames, and a frame header, respectively, tsw is the switch time between transmission and reception mode, n is the

number of parallel routes, and R is radio data rate. By selecting 96-bit frame header and acknowledge frame lengths, 3000 bits frames, and 200 μs radio mode switch time, theoretical throughput without retransmissions is 3.3 Mbps.

As the both application examples require only two radios in the intermediate nodes, the remaining the remaining two radios can be used for another parallel route to improve routing in the mesh-network, or for the transmission of FEC data to improve tolerance against bit errors.

5. CONCLUSIONS A new unique multi-radio WSN platform has been presented. FPGA-based hardware allows re-configurability and performance optimization for various applications. High flexibility is achieved by the use of four independent and simultaneously accessible radio transceivers. Simultaneous data exchange with multiple neighbors can be used for minimizing hop delays, and for multiplying network throughput. The platform gives excellent opportunities to design and test new kind of WSN protocols, algorithms and applications.

6. ACKNOWLEDGMENTS The platform will be available for research and development purposes from Atific Oy, Finland.

REFERENCES [1] H. Karl and A. Willig, Protocols and Architectures for

Wireless Sensor Networks, John Wiley & Sons Ltd., Chichester, 2005, pp. 1-13.

[2] M. Kohvakka, M. Hännikäinen, and Timo D. Hämäläinen, ”Ultra low energy wireless temperature sensor network implementation,” in Proc. PIMRC’05, Germany, Sep 2005.

[3] Altera Corp. home page, “Section 1, Cyclone FPGA Family Data Sheet,” August 2005, v1.3, Available: http://www.altera.com, Feb 2006.

[4] Nordic Semiconductor ASA home page, “Datasheet nRF2401A_rev1_0,” December 2004, v1.0, Available: http://www.nordicsemi.no, Feb 2006.

[5] V. Tsiatsis, S. Zimbeck and M. Srivastava, “Architectural strategies for energy efficient packet forwarding in wireless sensor networks,” in Proc. ISLPED’01, California, USA, Aug 2001, pp. 92-95.

Source

Destination

Figure 4. Analyzed mesh network topology.

Link 2Link 1Node A Node B Node C

H

AH

D

D

A H

H

D

D

H

H

D

D A

A H

H

D

D

Link 1:Node A

Node B

Link 2:Node B

Node C

tdec tsw

tsw

tA tH

tdelay

tprop

tf

Figure 5. Link layer operation.

97

kohvakka-medium access control and hardware prototype designs for low-energy wireless sensor...

Documents

energy overhead

highest energy

lowenergy wsns

wsn research

energyefcient protocols

thesis possible

lowest energy consumption

energyefcient maclayer