Experimental and Field Trial Results of

Enhanced Routing Based on LOAD for G3-PLC

Kaveh Razazian, Afshin Niktash,

Victor Loginov, Jim LeClare

Smart Grid Solutions

Maxim Integrated

Irvine, CA, USA

Thierry Lys

Metering Division

ERDF

Nanterre, France

Cédric Lavenu

MIRE Department

EDF R&D

Clamart, France

Abstract— G3-PLC is a power line communication standard

using OFDM technology for data communication defining both

physical and data link layers. The data link layer is derived from

IEEE 802.15.4 and Low-Power Wireless Personal Area Network

(6LoWPAN) standards. 6LowPAN Ad hoc Distance-vector

protocol (LOAD) is used as an efficient routing protocol over

varying network topologies and link conditions. Since LOAD was

originally developed for wireless networks, several unique

enhancements were added to tailor it to power line

communications. The enhanced LOAD protocol handles typical

issues such as asymmetrical links and ghost nodes. In addition,

link repair functionalities are provided and control traffic is

reduced. On the other hand, this paper describes how fair

channel access is granted to all nodes, providing a sound medium

access mechanism which is crucial for a power line

communication network. In a second experimental part of the

paper, some pressing issues are replicated within a test platform

to explore the routing enhancements in a controlled environment.

Finally, this paper focuses on the results obtained for a cluster of

132 nodes operating on an actual low voltage grid in Ste Maure

de Touraine in the Loire Valley, France. This cluster is part of a

2000 node Advanced Metering Infrastructure (AMI) field trial

led by the French Distribution Network Operator (DNO) ERDF.

For the first time, results showing enhanced LOAD routing

protocol performance within a power line communication AMI

deployment are presented. LOAD performs as expected,

providing connectivity to all meters installed in a typical

residential area.

Keywords— G3-PLC; Routing; LOAD; LOADng; 6LoWPAN; Ad-

hoc; AODV; AMI

I. INTRODUCTION

The use of the distribution grid for data communication has gained heightened interest over the past several years. It is well known that a power line channel is far from being an ideal channel for data transmission. Power line channel characteristics and parameters typically vary with frequency, location, time and the type of equipment connected to the channel. While the high-frequency regions above 1 MHz exhibit significant multipath frequency selective fading accompanied by severe attenuation on distances above several tens of meters, the lower frequency regions from 20 kHz to 500 kHz are especially susceptible to narrowband interference and impulsive noise [1].

G3-PLC is one of the standards adopted by ITU [2][3] using OFDM technology in different available bandplans ([35.990.6 kHz] within the CELENEC A band and [154.7487.5 kHz] within the FCC band). It consists of concatenation of a Reed-Solomon(RS) encoder with a ½ rate convolutional encoder, followed by a 4x repetition coding block for Robust mode (ROBO) cascaded with a two-dimensional time and frequency interleaver [4]. The G3-PLC system supports three different modulations, namely DBPSK, DQPSK, and D8PSK . Furthermore, the MAC layer which provides an interface between the Logical Link Control (LLC) layer and the physical layer is derived from IEEE 802.15.4. The MAC layer regulates the access to the medium using Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA), provides feedback to upper layers in form of positive and negative acknowledgements (ACK/NACK) and performs packet fragmentation and reassembly.

G3-PLC also provides IPv6 connectivity at network layer level and adopted 6LoWPAN adaptation layer for the efficient support of IPv6. 6LowPAN Ad hoc Distance-vector protocol (LOAD) was selected as an effective routing protocol to handle changing link conditions [5]. LOAD is a simplified on-demand routing method based on Ad-hoc On-demand Distance Vector routing protocol (AODV) [6][7]. Drafted within the 6LoWPAN working group in IETF, LOAD is one of the most promising routing schemes with low complexity which allows easy implementation and interoperability.

Since LOAD was originally designed for wireless networks, it did not consider unique power line channel characteristics in its routing strategy. In this paper an enhanced LOAD routing scheme tailored to power line communication is added to the G3-PLC data link layer to address asymmetrical link issues and to provide link repair functionalities. Furthermore, the information acquired during the bootstrapping procedure is utilized to avoid unnecessary control message flooding and to handle ghost node issues.

Carrier sense is a fundamental part of the distributed access procedure used to share the communication medium. Proper CSMA/CA operation is essential for routing and overall performance of G3-PLC due to the nature of physical layer (the frame size increases with the robustness of the modulation

2013 IEEE 17th International Symposium on Power Line Communications and Its Applications

978-1-4673-6016-6/13/$31.00©2013 IEEE 149

(a) (b)

Figure 1. LOAD routing, a) RREQ broadcast, b)RREP unicast

scheme). Enhanced channel access fairness is specified in the G3-PLC MAC layer and is described in section III.

This paper aims to replicate some pressing issues within a test platform to explore the routing enhancements in a controlled environment. It also reports G3-PLC field trial results of the proposed enhancements, obtained for a 132 meter cluster covering a low voltage grid in Sainte Maure de Touraine, France. This cluster is part of a larger 2000 node AMI field trial led by the French DNO ERDF.

The rest of the paper is organized as follows: section II gives an overview of the LOAD routing protocol and related enhancement for power line communication, section III summarizes the enhancements to the CSMA/CA medium access mechanism, section IV focuses on laboratory and field test results and section V concludes the paper.

II. LOAD ROUTING PROTOCOL AND

ENHANCEMENTS

LOAD is a reactive on-demand and ad hoc routing protocol for mesh networks. Within a network, it finds a route between any source node and any destination node (ad hoc protocol). The installation of a route towards a destination node is triggered by the request of a source node and does not require periodic signaling (reactive on-demand protocol). From a network layer point of view, a route computed by the LOAD is considered as a single IP hop at 6LoWPAN level.

Fig. 1 demonstrates the messaging in LOAD. Routes are established and maintained using three types of control messages: route requests (RREQ), route replies (RREP) and route errors (RERR).When a source node intends to send data to a destination node for which no valid route is available in its routing table, it generates a broadcast RREQ message carrying route cost information according to the metric used. The RREQ message is broadcasted through the network towards the destination. The reception of RREQ messages allows the destination and intermediate nodes to identify the neighbor through which the source node can be reached (reverse route).

Upon reception of a RREQ message, the destination node unicasts a RREP message back to the source node along the reverse route installed previously. The reception of RREP messages allows the source as well as intermediate nodes to identify the neighbor through which the destination node can be reached (forward route). This procedure installs a symmetrical link between source and destination. Moreover, should a node detect a broken link, a Route Error (RERR) is broadcasted to initiate new source discovery process [5].

In order to improve LOAD performance for power line communication, a number of enhancements were proposed for the field trial. Unlike wireless networks, the propagation channel is characterized by a location and time dependent impedance. The presence of different loads along the power line creates unbalanced signal propagated downwards and upwards the emitter plane, resulting in asymmetrical links between devices. This behavior is specific to power line communication and leads to unequal link quality with respect to the communication direction. Hence, a route is required to be sufficiently robust for both forward and reverse directions. The

G3-PLC specification proposes a mathematical formula to compute the link cost based on several parameters, including Modulation (MOD) and Link Quality Indicator (LQI). While forward LQI and Modulation are measured from the received RREQ, there is no provision in LOAD allowing a node to send reverse channel quality information to an intermediate node to be taken into account in link cost calculation. In order to include reverse channel quality into the equation, it is proposed to use neighbor table information as defined in G3-PLC [3]. Using a tone map request accompanied by channel estimation, a node can calculate reverse channel condition and send them back by tone map response. The information is used to update the neighbor table and serves as reverse LQI and MOD parameters. A conservative approach is followed to choose the worst of the forward and reverse channel quality in link cost calculation to ensure bidirectionality of the route.

In the LOAD protocol, if an intermediate node finds the same RREQ identifier from the same originator in its Route Request Table, it does not rebroadcast it. This implies that the first RREQ received at any intermediate node is assumed to have the better route cost, but this is not necessarily true as a RREQ from a better route might have been delayed. In order to address this issue, late RREQs are also rebroadcasted, only if their associated route cost is significantly lower than the last broadcasted one. This approach improves the route computation while it still reduces the number of RREQ broadcasts if they do not allow the installation of significantly better routes.

The LOAD protocol specifies that a RREP has to be generated in response to every RREQ if its route cost is lower than the RREQs previously received from the same source and with the same identifier. This results in sending multiple RREPs through different routes. In order to reduce RREP control traffic, the waiting period for the source node to collect all RREPs is moved to the destination node. After the first RREQ is received, the destination will wait for a pre-defined period to make sure that no other RREQs carrying a better route cost is in transit. At the end of this waiting period, the best route is chosen and a RREP is generated and sent towards the source along the selected route.

Ghost (or orphan) nodes are another issue observed in power line communication. A ghost node refers to a device that had been successfully associated and was operating in the

150

B

D

A

C

X

Figure 2.Failed RREP causing local link repair

network for some time, but bidirectional communication is no longer possible due to dynamic changes in the behavior of the channel. A ghost node always responds to beacon requests from new associating devices and may be selected as a forwarder while it does not have a viable link to the network coordinator. This could cause a series of failures in association procedure since the ghost node can be repeatedly selected as the forwarder for the same associating device. To address this shortcoming, a payload is defined and added to the beacon response to inform the associating node about communication problem with the network coordinator. This information is used to avoid selecting the ghost node for subsequent attempts.

Another important feature for power line communication resides in link repair. Let’s consider a case where multiple routes are available towards a destination, one of which includes a complete unidirectional forward route but with a better overall route cost. There is a chance for this broken route to be selected as the best route, triggering the transmission of a RREP along it. If the RREP never arrives at the originator, the route discovery fails. Subsequent route discoveries could follow the same route and fail again while an alternative bidirectional route with worse overall route cost could have been selected. Fig. 2 shows this condition. The local link repair mechanism tries to find another route during the forward route establishment procedure (RREP transmission), which could result in the selection of that alternative route and hence avoids deadlocks.

Authors along with several academic and industry leaders, are drafting the Lightweight On-demand Ad hoc Distance-vector Routing Protocol – Next Generation (LOADng) [8], which aims to enhance the LOAD routing protocol with additional features and capabilities. LOADng’s basic operations are based on LOAD (RREQ, RREP, RERR signalling) with new extensions. In particular, LOADng addresses the challenges of power line communication previously outlined, such as asymmetrical link handling and control traffic optimization. Indeed, the native inclusion of blacklists and hop by hop acknowledgement guarantees the computation of bidirectional routes, since a node is able to identify and tag its neighbors for which only a unidirectional link is available. In addition, the design of LOADng provides support for optimized flooding mechanisms for RREQ broadcasting, but also enables features such as the support of multiple metrics within the same network. Furthermore, RFC 5444 [9] guidelines can be employed which enable easy addition of future extensions, and broaden application.

III. CSMA/CA ENHANCEMENTS

The design of an effective CSMA/CA is crucial to regulate

the simultaneous access of the medium in typical smart grid

applications with hundreds of devices. Channel access is

accomplished by CSMA/CA mechanism with a random

backoff time. The random backoff mechanism spreads

transmission attempts over time, thereby reducing the

probability of collisions. This section describes how channel

access fairness can be improved for devices competing for the

channel.

The channel access mechanism in G3-PLC uses a

truncated binary exponential backoff mechanism similar to the

un-slotted mode of the CSMA/CA algorithm as in [10]. The

algorithm is implemented using units of time called

aSlotTime. Each device maintains two variables for each

transmission attempt, NB and BE. NB is the number of times

the CSMA/CA algorithm has been used to backoff while

attempting the current transmission; this value is initialized to

0 before each new transmission attempt. BE is the backoff

exponent, which is related to the number of backoff periods a

device will wait before attempting to access the channel. BE is

initialized to the value of minBE.

Fig. 3 illustrates the steps of the CSMA/CA algorithm.

After initialization of NB and BE, the transmission is delayed

for a random number of complete backoff periods in [0:2BE

-1]

range:

Backoff Time = random(2BE

–1)×aSlotTime (1)

If the channel is determined to be busy as indicated by

Physical Carrier Sense (PCS), NB and BE are incremented

and random backoff is repeated, provided that BE is not

exceeding maxBE. If the channel is idle, the transmission can

be started immediately.

Figure 3. Improved CSMA/CA algorithm

151

One main deficiency of this approach derived from [10] is

the fairness. If multiple devices try to transmit at the same

time, the device which acquired the channel first has a higher

chance to continue to access the channel for subsequent

transmissions as its random backoff window is the smallest.

On the other hand, other devices that could not access the

channel increase their random backoff time up to the

maximum value to (2maxBE

–1)×aSlotTime which translates to

lower probability of accessing the channel. Therefore, devices

with longer waiting time for pending transmissions have a

lower chance to acquire the channel which results in unfair

allocation of channel access to devices.

In order to improve the channel access fairness, a method

is introduced to give equal chance to other devices to compete

for the channel by reducing the random backoff window size

after a number of unsuccessful channel access attempts

(macCSMAFairnessLimit) as highlighted in Fig. 3. For

example, if we choose macCSMAFairnessLimit=15 and

maxCSMABackoffs=50, as soon as the number of backoffs

(NB) reaches 15, 30 and 45, BE is rolled back to minBE. In

large deployments, it is very important to properly select

CSMA/CA parameters to ensure the fair and competitive

opportunity of channel access for all devices.

IV. EXPERIMENTAL AND FIELD TRIAL RESULTS

This section elaborates on laboratory tests and field trial results of the G3-PLC with the previously introduced enhancements.

A. Laboratory Test Platform

1) Maximum number of hops and CSMA/CA tests

This experiment verifies data communication with the maximum number of hops (14 as specified in [3]) in a network. The experimental network includes only one branch with 60dB attenuation between devices and they are connected with cold wire as shown in Fig. 4. The attenuation is selected such that each device can successfully receive only packets from its immediate neighbor to make sure multiple hops are created. The PAN coordinator (DC) can associate devices in the order N1, N2, N3,…. Then request and response packets are sent successfully between DC and each node. Successful data communication between DC and N15 is not possible as it requires 15 hops to be established, which is beyond the maximum number of hops allowed in the network.

The CSMA/CA is investigated by sending packets from DC to N14 and at the same time from N14 to DC. Since two ends are several hops away, the do not see each other. As packets are being forwarded, CSMA algorithm is expected to handle the effective channel access to reduce packet loss due to collisions. Furthermore, a noise source may also be added to a middle node to create a hidden node scenario. In a simple test with three nodes, a noise source is added to N1 and N2 sends packets to DC while DC also sends packets to N2. The noise level is selected such that Physical Carrier Sense in N2 or DC cannot detect each other’s packets, but N1 can successfully detect and receive packets from N2 and DC. Under this

Figure 4. Maximum number of hops and CSMA/CA test

condition, N2 and DC are considered hidden nodes from each other and therefore the CSMA/CA cannot avoid collisions.

2) Asymmetrical Link Avoidance Test

This experiment demonstrates avoidance of asymmetrical links in routing. A unidirectional link is created by noise insertion between DC and node N2. The noise level is selected such that successful communication is possible from DC to N3, but not from N3 to DC. The link between DC and N3 is unidirectional as shown in Fig. 5. In this setup, the PAN coordinator (DC) associates devices in the following order: N2, N3, N4 and N5. Then, data is successfully exchanged between DC and N5. It appears that the route discovery procedure between DC and N5 resulted in a 3-hop route and intermediate node N2 is used to forward data between DC and N3. Hence, the selected route does not use the unidirectional link between DC and N3.

By disconnecting the noise generator as shown in Fig. 6, the channel between DC and N3 becomes symmetrical, resulting in a bidirectional link. The PAN coordinator (DC) associates devices in the following order: N3 and N5. Then, data can be successfully exchanged between DC and N5. The route discovery procedure selects a route using the bidirectional link between DC and N3.

Figure 5. Asymmetrical channel

Figure 6. Symmetrical channel

152

Figure 7. Weak link simulation

3) Weak Link Avoidance Test

This experiment verifies that a route discovery procedure will avoid using weak communication links when selecting a route. The PAN coordinator (DC) associates N1 and then N2. By inserting noise and proper selection of noise level, the direct communication link between DC and N2 becomes marginal. The computed route between DC and N2 uses device N1 as intermediate hop as shown in Fig. 7. When the noise generator is removed, it is verified that a direct communication between DC and N2 is established.

B. Field Trial Results and Analysis

ERDF deployed a 2000 meter pilot project in both rural and urban environments to assess the performance of the G3-PLC technology in the field. This section focuses on a cluster installed in Sainte Maure de Touraine, France. The area includes a number of 2 to 3 story residential apartment buildings as well as single family houses in a geographically scattered neighborhood of about 0.3 km

2. The cluster consists

of a data concentrator (DC) installed in a secondary substation, communicating with 132 meters spread over the low voltage (LV) grid. The distance between the farthest meter and the data cconcentrator is 450m. The average distance of meters with the data concentrator is 200m. The considered part of the distribution grid was comprised of a combination of aerial lines (naked lines, insulated twisted lines) and underground cables. The data concentrator is connected to the low voltage (LV) side of a 400 kVA MV/LV transformer. The communication between the DC and the monitoring center is established via a GPRS link. The metering information and topology-related data (including the established routes between all G3-PLC devices) is periodically read by the monitoring center.

Meters are configured to request network association upon start-up after an increasing random delay. Upon failure, the delay is increased by several minutes to avoid network congestion due to the transmission of multiple beacon requests and beacons.

The DC sends periodic reading requests to the meters and they respond with the appropriate metering data. On the other hand, meters can also initiate the communication when sending alarm information. This mode of operation requires the establishment of bidirectional routes. The route discovery procedure can be invokes the data concentrator or meters.

The presented results were captured on November 12th

2012 at 7:15PM when most of the inhabitants were at home with electric appliances turned on.

Fig. 8 shows the distribution of number of hops observed in this installation and Fig. 7 depicts the obtained logical

topology. One half of the meters are able to communicate with the data concentrator directly (single hop meters). This demonstrates the robustness of the G3-PLC physical layer in the field. About 30% of the meters are distributed relatively evenly at a 2-hop distance from the DC. Only few nodes are acting as forwarders for the remaining 20% of the meters which are located at a logical distance exceeding 2 hops (3 to 6 hops).

The communication quality can be analyzed by studying the modulation distribution in each topological layer. The type of modulation selected by a node to communicate with its neighbor is determined by the channel estimation function defined in G3-PLC. For low quality links, the ROBO mode is used. As the quality improves, higher modulations DBPSK, DQPSK or D8PSK are selected. Fig. 10 shows the distribution of the modulation as the logical distance from the data concentrator (number of hops) changes. Most of the links are capable of D8PSK modulation. In particular, most of the nodes located at one or two hops away from the data concentrator are able to communicate in D8PSK or DQPSK.

This can be further analyzed in Fig. 11 which shows the modulation distribution across the network. About 56% and 39% of all communications between nodes are performed using D8PSK and DQPSK modulations respectively. This means that the routing algorithm has successfully established high quality links in the network and use of links with bad channel which requires BPSK or ROBO modulation is insignificant.

Figure8. Distribution of number of hops (%)

DC

Figure9. Logical network topology

153

Figure 10. Distribution of modulation with number of hops

Figure 11. Distribution of Modulation in the network

V. CONCLUSION

Due to the dynamic and time varying conditions of power lines, the efficiency of an adaptive routing algorithm capable of dealing with varying network topology and link condition is a key factor. The LOAD routing protocol was originally developed based on wireless mesh networking requirements. A number of enhancements are presented to make it more efficient for power line communication. Unidirectional links, as one of the major contributors to network failure, are avoided by using reverse link quality (LQI) and modulation information. Further improvements are described to reduce the RREQ and RREP control traffic and to avoid the occurrence of deadlocks during link establishment. In addition, an effective CSMA/CA method which is crucial to improve channel access fairness is presented. The method allows other devices to compete for the channel by reducing their random backoff contention windows after several unsuccessful attempts. The effect of proposed schemes is investigated in a controlled environment with test platforms. Finally, the result of deployment of 132 digital meters in Sainte Maure de Touraine, France is presented. The result shows about half of the meters are able to communicate directly with the data concentrator and another 30% of the meters are communicating at a 2-hop distance. In addition, majority of all communications between nodes are performed using D8PSK and DQPSK modulations. The new enhanced routing strategies ensure creation of routes with high quality links and minimize the usage of marginal ROBO links.

ACKNOWLEDGMENT

The authors would like to express their great appreciation to ERDF and EDF management for supporting G3-PLC deployment in France and enabling the first 2000 meter field trial in Tours and Lyon areas. The authors would also like to thank the Gateways and Energy Efficiency Department at Sagemcom for design, development and successful deployment of meters and data concentrators. Finally, the authors would like to thank Dr. Javad Yazdani for his support, contribution and technical expertise.

REFERENCES

[1] K. Razazian, M.Umari, A. Kamalizad, V. Loginov, M. Navid, " G3-PLC specification for powerline communication: overview, system simulation and field trial results," IEEE Intl. Symposium on Power Line Communications and Its Applications, 2010.

[2] G3-PLC physical layer specification and G3-PLC MAC Layer specification, G3-PLC Alliance, Aug. 2012..

[3] Recommendation ITU-T G.9903 Narrowband orthogonal frequency division multiplexing power line communication transceivers for G3-PLC networks, Oct. 2012.

[4] K. Razazian, M.Umari, A. Kamalizad, “Error correction mechanism in the new G3-PLC specification for powerline communication,” IEEE Intl. Symposium on Power Line Communications and Its Applications, 2010.

[5] http://tools.ietf.org/html/draft-daniel-6lowpan-load-adhoc-routing-03.

[6] C.Perkins, E. Belding-Royer, S. Das, “Ad hoc on-demand distance vector (AODV) routing,” RFC 3561.

[7] G. Lakshmikanth, M.A Gaiwak, P.D Vyavahare, “Simulation based comparative performance analysis of adhoc routing protocols,” TENCON IEEE Region 10 Conference, pp. 1-5, 2008.

[8] T. Clausen, A. Colin de Verdiere, J. Yi, A. Niktash, Y. Igarashi, H. Satoh, U. Herberg, C. Lavenu, T. Lys, C.Perkins, J. Dean, “The lightweight on-demand ad hoc distance-vector routing protocol – next generation (LOADng),” https://tools.ietf.org/html/draft-clausen-lln-loadng-06.

[9] T.Clausen, C.Dearlove, J.Dean, C.Adjih, “Generalized mobile ad hoc network (MANET) packet/message format,” RFC 5444.

[10] IEEE 802.15.4, Part 15.4: Wireless Medium Access (MAC) and physical layer (PHY) specifications for low-rate wireless personal area networks (WPANs).

154

本文献由“学霸图书馆-文献云下载”收集自网络，仅供学习交流使用。

学霸图书馆（www.xuebalib.com）是一个“整合众多图书馆数据库资源，

提供一站式文献检索和下载服务”的24 小时在线不限IP

图书馆。

图书馆致力于便利、促进学习与科研，提供最强文献下载服务。

图书馆导航：

图书馆首页 文献云下载 图书馆入口 外文数据库大全 疑难文献辅助工具