research article stochastic polling interval...

Research ArticleStochastic Polling Interval Adaptation in Duty-Cycled WirelessSensor Networks

Sungryoul Lee

The Attached Institute of Electronics and Telecommunications Research Institute (ETRI) PO Box 1 YuseongDaejeon 305-600 Republic of Korea

Correspondence should be addressed to Sungryoul Lee srlee0525ensecrekr

Received 29 October 2014 Accepted 16 January 2015

Academic Editor Andrei Gurtov

Copyright copy 2015 Sungryoul LeeThis is an open access article distributed under theCreative CommonsAttribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

In past decades to achieve energy-efficient communication manyMACprotocols have been proposed for wireless sensor networks(WSNs) Particularly asynchronous MAC protocol based on low power listening (LPL) scheme is very attractive in duty-cycledWSNs it reduces the energy wasted by idle listening In LPL scheme a sensor node wakes up at every polling interval to samplethe channel If the channel is busy the sensor node will stay in wake-up mode for receiving the data packet Otherwise it goesto sleep and saves power However wrong choice of polling interval in LPL scheme causes unexpected energy dissipation Thispaper focuses on the polling interval adaptation strategy in LPL scheme with the aim of maximizing energy efficiency defined asthe number of packets delivered per energy unit We propose a novel polling interval adaptation algorithm based on stochasticlearning automata where a sensor node dynamically adjusts its polling interval Furthermore our simulation results demonstratethat the polling interval asymptotically converges to the optimal value

1 Introduction

In recent years wireless sensor networks (WSNs) have beendeveloped and implemented to realize automated homesystem combined with consumer electronics technologies[1] The design principle of such a system is very simple thelow-cost sensor nodes with low-power processing narrowcommunication range and small battery size are scatteredin a sensing field Therefore it is unreasonable to replaceor recharge the batteries of sensor nodes For example areal-time intrusion detection system detects the humidityand transmits an alert message to a remote base stationthat is a sink immediately [2] In this application it isdesirable to retain for long operational time of sensor nodebut the network lifetime is substantially constrained due tothe limited capacity of the battery

Recently a number of research efforts have been under-taken to save power of sensor node in communication activ-ities The main direction of these researches is duty-cyclingwhich reduces power wasted through idle listening that isthe time spent in wake-up mode with receiving any radiopackets Note that idle listening is a dominant reason that

drains the energy of sensor node [3ndash5] In duty-cycledWSNsthe sensor node alternates between sleep mode and wake-upmode to reduce energy consumption caused by idle listeningThe sensor node wakes up in small portion of operating timeand it goes to sleep in most of time The transition betweensleep mode and wake-up mode is decided by a predefinedschedule In sleep mode the sensor node turns off its radioto save power In wake-up mode the sensor node turns on itsradio to communicate with its neighbors If the sensor nodehas a packet destined to its neighbors or it is an intendedreceiver it transmits or receives a data packet in wake-upmode

Low power listening (LPL) scheme has motivated recentadvances in asynchronous MAC protocol design for duty-cycledWSNs In [6 7] the authors introduced the concept ofLPL scheme in a perspective of duty-cycled WSNs to reducethe idle listening where a sender transmits a 119862-sized pream-ble to a receiver which is supposed to wake up at every pollinginterval 119862 for preamble detection In B-MAC [6] a sendertransmits a long preamble to make its neighbors wake upbefore transmitting a data packet Upon detecting the pream-ble all neighbors keep wake-up mode until the transmission

Hindawi Publishing CorporationInternational Journal of Distributed Sensor NetworksVolume 2015 Article ID 486908 7 pageshttpdxdoiorg1011552015486908

2 International Journal of Distributed Sensor Networks

The sender

The intended receiver

The receiver

Wake-upperiod

Wake-upperiod

Wake-upperiod

Sending ACK

Sending continuousshort preambles Receiving ACK

Receiving adata packet

Sending adata packet

Figure 1 The time diagram of XMAC

of data packet has been finishedHowever B-MAC inherentlysuffers from the excessive long preamble accompanied withthe transmission of data packet and the overhearing ofnonintended receivers X-MAC [7] has been proposed as anenhanced LPL scheme fixing the drawbacks of B-MAC X-MAC adopts a series of short preambles to avoid overhearingIn X-MAC a sender transmits a short preamble and waitsfor a response from a receiver for a short time as shown inFigure 1 And the sensor node samples the channel whenit wakes up If it is the intended receiver upon receivingthe short preamble it replies with ACK to trigger the trans-mission of data packet Otherwise it goes to sleep to avoidoverhearing As compared with synchronousMAC protocolssuch as S-MAC [3] and T-MAC [4] these asynchronousMAC protocols have the advantage of not requiring any timesynchronization among sensor nodes

In LPL scheme the best performance can be obtainedby appropriately selecting the polling interval 119862 which is akey parameter in determining energy used to communicatebetween sensor nodes The optimal polling interval heavilydepends on traffic loads of the sensor node In low trafficloads a large polling interval takes advantage of saving energyof sensor node while a relatively small polling interval con-sumes less energy spent for communication between sensornodes in high traffic loads Also the polling interval must beadapted during runtime since traffic condition may not beknown a priori [8] If a priori assumption about traffic loadsis not accurate energy of sensor node will be wasted and theintended goal of application cannot be achieved For instanceelectronic security system which is installed in home detectssuspicious objects To achieve this requirement sensor nodesshould be in sleep mode for most of the time when there areno events of interest

Recently there have been a few studies on pollinginterval adaptation in duty-cycled WSNs The representativetechnique employed by existing polling interval adaptationschemes such as [9 10] guesses the traffic condition basedon previous results of channel sampling Note that thisapproach is referred to as Dynamic LPL (DLPL) scheme InDLPL scheme the sensor node obtains information about thenumber of consecutive busy (idle) sampling and dynamically

adjusts the polling interval by adopting the following Thepolling interval is increased after 119880 idle polling Similarly119863 consecutive busy pollings induce a decrease of pollinginterval This approach is simple and intuitive but has someproblems Most importantly it can make a scheme either tooaggressive or too conservative depending on 119880 and 119863 Forexample BoostMAC [9] which employs 119880 = 1 and 119863 = 1

reacts well to traffic changes as it immediately increasesdecreases the polling interval But it introduces unnecessarypolling interval fluctuations in stable environments whichcauses deterioration of performance Additionally it cannotbe expected to converge to optimal polling interval

In this paper we propose a stochastic polling intervaladaptation algorithm based on learning automata techniquewhere a sensor node dynamically adjusts its polling intervalaccording to the probability distribution of selecting thepolling intervals and selects the optimal polling intervalby learning In the proposed algorithm we adopt energyefficiency which is defined as the number of packets deliveredper energy unit to update the probability distribution associ-ated with the polling intervals Additionally our evaluationsverify that the proposed algorithmmakes the polling intervalconverge to the optimal value

Our main contributions are summarized as follows

(i) We consider the problem of runtime adaptation ofpolling interval in duty-cycled WSNs Fixed pollinginterval may degrade the performance of LPL schemewhen traffic loads are not known a priori

(ii) We investigate the problem of adapting the pollinginterval via stochastic learning automata To the bestof our knowledge our work is the first attempt toprovide a learning automation based solution foradaptation of polling interval in duty-cycled WSNs

(iii) We propose a novel polling interval adaptation algo-rithm which dynamically adjusts the polling intervaland the polling interval asymptotically converges tothe optimal value Our simulation results show thatthe proposed algorithm converges toward the optimalvalue

International Journal of Distributed Sensor Networks 3

The reminder of this paper is organized as follows InSection 2 we present a review of some previous related workSection 3 describes the proposed polling interval adaptationalgorithm based on stochastic learning automata Section 4discusses the simulation results and we conclude this paperin Section 5

2 Related Work

In this section we overview current researches on MACprotocol design in duty-cycled WSNs

Basic LPL scheme that is B-MAC [6] which uses along preamble to establish rendezvous between a senderand its receiver has overhearing problem In B-MAC sincethe sender does not indicate the intended receiver of thepacket all neighbor nodes must wait for finishing the longpreamble To solve this problem several works [7 11 12]adopt an aggressive short preamble to replace the longpreamble which divides into a series of short packets In placeof the long preamble SpeckMAC-B [11] uses the wake-uppacket and B-MAC+ [12] sends the chunks which containinformation about remaining chunks before transmitting thedata packet Note that the authors of [12] also presentedan extension of B-MAC+ that adapts the polling interval ofthe transceiver to the traffic loads experienced by differentsensors [13] Therefore the sensor node which receives theearly preamble can sleep during waiting for the data packetOther approaches [7 14] reply with an early ACK for stoppingthe transmission of excessive short preamble The sendertransmits a series of short preambles including ID of intendedreceiver ACK sent in response to the short preamble triggersthe transmission of data packet

In duty-cycled WSNs static approach which adopts afixed polling interval cannot be adapted to various networkconditions To overcome this problem several works [9 1315ndash20] allow dynamically changing sensorrsquos polling intervalThe role of these polling interval adaptation mechanism isto select the optimal polling interval according to trafficloads of sensor node PMAC [15] adopts an adaptive duty-cycled scheme instead of having a fixed duty cycle as inS-MAC to improve energy efficiency PMAC allows thesensor node to adaptively determine the sleep-wake upschedules based on its own traffic and the traffic patternsof its neighbors Dynamic LPL (DLPL) scheme [9 16] is awidely adopted and well-known polling interval adaptationalgorithm This works as follows if 119880 consecutive samplingsare idle the sensor node increases its polling interval if119863 consecutive busy samplings are busy the polling intervalis increased In BoostMAC [9] changes of polling intervalin sensor node are accomplished by using AIMD (AdditiveIncreaseMultiplicative Decrease) mechanism in response tothe results of channel sampling Also in [16] the authorspresented a Markov model which evaluates the performanceof DLPL scheme in terms of energy consumptionThe analyt-ical model enables us to investigate the effect of the updownthreshold (119880 and119863)mdashthese parameters determine how longa sensor node should stay at a certain polling intervalbefore it concludes the traffic condition has changedmdashon theperformance of DLPL scheme Obviously DLPL scheme is

easy to deploy but itmay produce unnecessary fluctuations orcannot react quickly to traffic changes according to updownthreshold Meanwhile the queue state is a useful informationfor guessing the network condition implicitly In [17 18]the authors proposed an adaptive control mechanism basedon the queue management where the controller changesthe polling interval dynamically by constraining the queuelength Similarly TA-MAC [19] adjusts sleep interval adap-tively according to state of sendingreceiving buffer trafficloads and battery lifetime In [20] the authors presenteda cross-layer design approach for joint optimization at theMAC and routing layers To address this challenge theyproposed an adaptation of listening modes according to localstate of each sensor and it enables the sensor node to learnlisteningmode of its neighbors in order to ensure correct datadelivery

Learning automata have been applied to study a widerange of solving optimization problems in wireless networksIn [21] the authors adopted a stochastic learning automatamodel to find the optimal channel selection for secondaryusers in cognitive radio networks Since primary userrsquos trafficpatterns are unknown and unpredictable they claim that thesecondary users must select the statistically optimal channelwhich maximizes the probability of successful transmissionand propose an estimator automata model to pursue globaloptimum with minimal number of iterations In [22] theauthors proposed a novel congestion control algorithm basedon learning automata in healthcare WSNs The primaryobjective of this approach is that the processing rate in sensornode is equivalent to the transmitting rate In each sensornode better data rate may be chosen on the basis of pastexperience with congestion with the other data rates In[23] the authors adjusted the threshold parameters of AutoRate Fallback (ARF) in IEEE 80211 WLANs using learningautomata

3 Stochastic Polling Interval Adaptation

31 Learning Automata At the beginning learning automatatechniques were introduced to find a solution in controlliteratures Recently they have been recognized as one ofthe most powerful methods to select the best action in astochastic environment And they have been adopted to solvethe stochastic optimal control problems in a wide range ofresearch fields

The purpose of stochastic learning automata which keeptrack of possible actions and their probabilities is tomaximizethe expected reward or to minimize the expected penaltybased on the response of possible actions In learning processan action from finite set of possible actions is applied to astochastic environment and then learning system records aresponse associated with an action as depicted in Figure 2The response reflects the condition of the stochastic environ-ment Learning can be described as follows

Let us denote the finite set of possible actions as 120572 =

[1205721(119905) 1205722(119905) 120572

119899(119905)] at time 119905 and selection probabilities

of their actions as 119875 = [1198751(119905) 1198752(119905) 119875

119899(119905)] at time 119905

respectively And let 120573 = [1205731(119905) 1205732(119905) 120573

119899(119905)] be the set

of automation output at time 119905 At every time iteration a


Input 120572

A stochastic environment

Learning automata

Output 120573

Figure 2 Overview of learning automata

stochastic environment takes the set of action 120572 as input andgenerates the output vector 119875 related to input action basedon the response 120573 Therefore probability distribution overactions is updated based on the response of the environmentand is reinforced to select the optimal action This processcontinues until a predefined condition

As previously mentioned a wireless sensor network is apart of stochastic environments where traffic loads are notknown a priori In our work we formulate the problem ofpolling interval adaptation as an environment that stochasticlearning automata select the best action

32 The Proposed Algorithm Motivated by the above dis-cussion we devise a control algorithm of maximizing theexpected reward for selecting the polling interval 119862

Now let us describe the basic operation of our proposedalgorithm which aims to converge to the optimal action thatis selecting the optimal polling interval We consider that asensor node selects the polling interval in119870 polling intervals1198621lt 1198622lt sdot sdot sdot lt 119862

119870 And let us denote action probability

vector 119875 associated with each polling interval The sensornode adopts LPL scheme with X-MAC [7] Additionallywhen a sender has a data packet to be sent it transmits aseries of short preambles longer than polling interval 119862

119870to

ensure asynchronous rendezvous between the sender and itsreceiver Since the receiver wakes up every polling intervaland samples the channel it detects the short preamble If thereceiver is the intended recipient it replieswithACKAnd thereception of ACK at the sender triggers the transmission ofdata packet Otherwise the receiver goes into sleep and waitsuntil the next polling interval to sample the channelNote thatthe operation of X-MAC is illustrated in Figure 1

Remark 1 Our algorithm operates at receiver-side If thesensorwhich acts as the sender has the data packet to transmitregardless of its operation mode (sleep or wake-up mode) itoperates inwake-upmode immediately and transmits a seriesof short preambles to check whether an intended receiverwakes up Therefore when the sensor acts as the sender itdoes not affect our algorithm

The sensor node selects the polling interval in eachdecision time We denote by 119905 = [119905

0 1199051 ] and 119905(119894) =

[119905119894minus1

119905119894] 119894 = 1 2 the set of decision time and the length of

119894th iteration of our algorithm respectively In the followingwe assume that there is enough time between consecutivedecision times that is 119905(119894) = [119905

119894minus1 119905119894] 119894 = 1 2 to get

the correct answer This means that the sensor node needs

Table 1 Simulation properties

Symbol Meaning Value119875119905119909

Power in transmitting 522mW119875119903119909

Power in receiving or listening 564mW119875sleep Power in sleeping 3 120583W119905csl Average carrier sense time 2ms119905119861

Time to transmit or receive a byte 32 120583s119871data Data packet length 50 B119871Preamble Preamble packet length 15 B119871ack ACK packet length 11 B119878119901

Time to transmit the preamble at the sender 048ms119878119886

Time to listen the ACK at the sender 0352ms119878119889

Time to transmit the data at the sender 16ms119877119897

Wake-up time at the receiver 10ms119877119886

Time to transmit the ACK at the receiver 0352ms119877119889

Time to receive the data at the receiver 16ms

enough time to receive one or more packets At time 119905 thesensor node selects the polling interval 119862

119894 119894 = 1 le sdot sdot sdot le 119870

based on the probability vector 119875 Note that during 119894thiteration polling interval is not changedThepolling intervalsare taken as input to stochastic learning automation and thesensor node updates the probabilities of polling intervals as afunction of the output given by a stochastic environment Inourmodel we adopt the energy efficiency defined as the ratioof total amount of packets delivered to total energy consump-tion as output metric This process continues until stoppingcondition

In our learning model the following measures are con-sidered 119873

119894 119894 = 1 le sdot sdot sdot le 119870 as the number of received

data packets with polling interval119862119894in each iteration and 119864

119894

119894 = 1 le sdot sdot sdot le 119870 as power consumption with polling interval119862119894in each iteration respectively 119864

119894is obtained as follows

(the parameters are shown in Table 1)

119864119894= 119864119904+ 119864119903

= (119875119905119909119878119901+ 119875119903119909119878119886)(

119862119894+ 119877119897

2 (119878119901+ 119878119886)

) + 119875119905119909119878119889

+ 119875sleep119862119894 + 119875119903119909119877119897+ 119875119905119909119877119886+ 119875119903119909119877119889

(1)

where 119864119904and 119864

119903indicate power consumption at the sender

and the receiver respectively Note that if the channel is idlethe sensor (receiver) considers only 119864

119903as 119864119894 Hence the

number of data packets delivered per energy unit119883119894= 119873119894119864119894

119894 = 1 le sdot sdot sdot le 119870 can be calculated At every decision timethe sensor node calculates energy efficiency 119883

119894 and updates

the accumulated energy efficiency Our algorithm finds thepolling interval in terms of maximizing energy efficiency atdecision time

We define 119863119894(119905) 119894 = 1 le sdot sdot sdot le 119870 which is the

deterministic estimation vector of polling interval 119862119894at time

119905 119894 = 1 le sdot sdot sdot le 119870 would be

119863119894(119905) =

119869119894(119905)

119867119894(119905)

(2)


where119867119894(119905) 119894 = 1 le sdot sdot sdot le 119870 is the count of how many times

polling interval 119862119894has been selected up to time 119905 and 119869

119894(119905)

119894 = 1 le sdot sdot sdot le 119870 is the accumulated energy efficiency withpolling interval 119862

119894up to time 119905 respectively

Next let 119880119894(119905) denote the stochastic estimator vector at

time 119905 119894 = 1 le sdot sdot sdot le 119870 which means the reward probabilityof polling interval 119862

119894

119880119894(119905) = 119863

119894(119905) + 119877

119894(119905) (3)

where 119877119894(119905) 119894 = 1 le sdot sdot sdot le 119870 is a random number which

is uniformly distributed in the interval [minus120574119867119894(119905) +120574119867

119894(119905)]

where 120574 is a perturbation system parameter set by the sensornode Other parameters of our algorithm are defined asfollows

119875119894(119905) the probability of selecting polling interval 119862

119894at

time 119905 119894 = 1 le sdot sdot sdot le 119870119877 the resolution parameter of learning automatonthat is a positive number which has property thatit determines the stepsize on the basis of probabilityvector 119875119865 the predefined convergence threshold119872 the maximum value of energy efficiency

In the following we develop the proposed algorithmbased on stochastic learning automata According to ourproposed algorithm the sensor node processes its obser-vation and updates probabilities of selecting the pollingintervals based on the response before selecting the newpolling interval Indeed in our algorithm the probabilityof selecting the optimal polling interval is increased whilethe probabilities of others are decreased Additionally ouralgorithm achieves asymptotic convergence

Algorithm 2

(1) Initialization

(i) Set 119875(0) = [1198751(0) 119875

119870(0)] where 119875

119894(0) = 1119870 119894 =

1 le sdot sdot sdot le 119870(ii) Initialize119867

119894(0) 119869119894(0)119863

119894(0)119880

119894(0) and119872 to zeros for

119894 = 1 le sdot sdot sdot le 119870(iii) Select the polling interval 119862

119899according to 119875(0)

(iv) Set119873119899and 119864

119899to zeros

(v) Start to sample the channel with 119862119899

(vi) Maintain the current polling interval 119862119899and then

record119873119899and 119864

119899until 119905(1)

(2) At time 119905

(i) Compute 119883119899and update the maximum of energy

efficiency119872

if 119883119899gt 119872 119872 = 119883

119899 (4)

(ii) Update the accumulated energy efficiency 119869119899(119905) as

119869119899(119905) = 119869

119899(119905) + (119883

119899119872)

(iii) Update119867119899(119905) by adding one as119867

119899(119905) = 119867

119899(119905) + 1

(iv) Compute 119899th element in the deterministic estimationvector119863

119899(119905) by setting119863

119899(119905) = 119869

119899(119905)119867119899(119905)

(v) For 119894 = 1 le sdot sdot sdot le 119870 compute the stochastic estimatorvector 119880

119894(119905) by setting 119880

119894(119905) = 119863

119894(119905) + 119877

119894(119905)

(vi) Find the optimal polling interval 119862119898

which is thehighest value in 119880(119905)

(vii) Update probability vector 119875 as follows

119875119894(119905) = max (119875

119894(119905) minus (

1

119877119870) 0) if 119894 = 119898 119894 = 1 119870

119875119894(119905) = sum

119894 =119898

119875119894(119905) if 119894 = 119898

(5)

(viii) Set119873119899 119864119899 and119883

119899to zeros

(ix) If119875119898(119905) gt 119865 then converge to optimal polling interval

119862119898and stop

(x) Otherwise select the new polling interval accordingto 119875(119905) and start to sample the channel for communi-cation

In our algorithm the update of probability vector 119875

depends on the deterministic reward vector 119863 and therandom number 119877 Initially during a few iterations thesensor node selects the polling interval mainly depending onrandom number 119877 This implies that all of polling intervalshave the chance to be selected as the optimal value Note thatlike other learning automata based optimization methodsthe computational complexity of our algorithm depends onthe number of possible actions 119870 Therefore our algorithmrequires time complexity 119874(119870) for updating the probabilityvector 119875 Also with increasing iteration our algorithmmainly depends on the deterministic reward Therefore thepolling interval with high probability is selected more fre-quently and will be the optimal value Moreover the asymp-totic process of our proposed algorithm is 120598-optimal whichis proved in [24]

Theorem 3 The proposed polling interval adaptation algo-rithm is 120598-optimal for stationary duty-cycled WSNs For anyarbitrarily small 120598 gt 0 and 120574 gt 0 there exists a 1199051015840 satisfying

119875119903 10038161003816100381610038161 minus 119875

119898

1003816100381610038161003816 lt 120598 gt 1 minus 120574 forall119905 gt 1199051015840 (6)

where 119898 is the index of optimal polling interval in terms ofenergy efficiency

4 Simulation Results

In order to evaluate the effectiveness of our proposed algo-rithm we performed extensive simulation experiments Weemploy a single sender-receiver pair in order to monitorreceiverrsquos polling interval according to its variable traffic loaddefined by node topological distance from the sink Note thatthe sensor nodes near the sink have more traffic loads andshorter polling interval than those far away from the sinkTheresults of simulation showed that our algorithm dynamically


1

2

3

4

5

6

7

10 20 30 40 50 60Number of iterations

Traj

ecto

ry o

fCi

Figure 3 Trajectory of polling interval

adjusts the polling interval according to probabilities ofpolling intervals to adopt traffic condition and then tracks theoptimal polling interval

In the simulation we assume that the sensor node hasseven polling intervals (119870 = 7) that is [119862

1 1198622 119862

7] =

20 40 80 160 320 640 1280 (msec) Also let us assume thatthe sender generates the data packets following the Poissonprocess with rate one (1 packetsec) The resolution param-eter 119877 is set by 2 and the convergence threshold 119865 is setby 099 Also 1 and 10 (sec) are used for 120574 and the length of119905(119894) The parameters used in simulations are summarized inTable 1

In Figure 3 we show the trajectory of polling intervalwith respect to the optimal value at run time It illustratesthat the proposed algorithm adjusts the polling interval Ininitial period the polling interval is fluctuated Also ouralgorithm accommodates the fluctuation and the pollinginterval 4 (160msec) ismore often selected than other pollingintervals at around the 20th iteration Note that the pollinginterval 4 (160msec) is the optimal value where it achievesminimal energy consumption in simulation experimentsCorrespondingly it can be seen that the polling intervalconverges within 50 iterations as depicted in Figure 3

In the proposed algorithm the probability of selectinga polling interval is updated to search the optimal value InFigure 4 the seven curves represent the probability of eachpolling interval and each point in the curve results fromeach iteration at runtime In initial period the probabilitiesof polling intervals are equal and selecting polling intervalis dependent on randomness As shown in Figure 4 ourproposed algorithm increases the choice probability of select-ing the optimal polling interval that is polling interval 4(160msec) at every iteration and achieves convergencetowards the optimal polling interval

Furthermore Figure 5 compares the proposed algorithmwith DLPL scheme Here as well we use fixed updownthreshold (119880119863) = (1 1) which is used in BoostMAC [9]AMAC [10] and PMAC [15] From the figure we can seethat the proposed scheme is more energy efficient than DLPLscheme This is because the proposed scheme converges to

0

01

02

03

04

05

06

07

08

09

1

10 20 30 40 50 60

Prob

abili

ty

Number of iterations

P1

P2

P3

P4

P5

P6

P7

Figure 4 Updating history of probability vector 119875

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

100 200 300 400 500

Ener

gy co

nsum

ptio

n (m

W)

Simulation time (s)

Proposed algorithmDLPL

Figure 5 Total energy consumption

the optimal polling interval with probability one while thepolling interval of DLPL scheme is fluctuated

5 Conclusion

To achieve energy efficient communication in duty-cycledWSNs one of major issues of MAC protocol design isdynamic adaptation of polling interval against network con-ditions Additionally control algorithm is adapted at runtimein response to local observation for each polling interval

In this paper we proposed a novel stochastic pollinginterval adaptation algorithm to tackle this issue To the bestof our knowledge this paper is the first attempt to applystochastic learning automata for control of polling interval inpractice In our algorithm the sensor node dynamically


adjusts its polling interval based on response which is thenumber of packets delivered per energy unit Using simula-tion experiments we observe that our proposed algorithmcan adjust the polling interval to converge to the optimalvalue gradually

Conflict of Interests

The author declares that there is no conflict of interestsregarding the publication of this paper

References

[1] I F Akyildiz W Su Y Sankarasubramaniam and E Cayirci ldquoAsurvey on sensor networksrdquo IEEE Communications Magazinevol 40 no 8 pp 102ndash114 2002

[2] G Song ZWeiW Zhang andA Song ldquoDesign of a networkedmonitoring system for home automationrdquo IEEE Transactions onConsumer Electronics vol 53 no 3 pp 933ndash937 2007

[3] W Ye J Heidemann and D Estrin ldquoAn energy-efficient MACprotocol for wireless sensor networksrdquo in Proceedings of theIEEE INFOCOM pp 1567ndash1576 June 2002

[4] T van Dam and K Langendoen ldquoAn adaptive energy-efficientMAC protocol for wireless sensor networksrdquo in Proceedings ofthe 1st International Conference on Embedded Networked SensorSystems (SenSysrsquo03) pp 171ndash180 Association for ComputingMachinery November 2003

[5] Q Yu C Tan and H Zhou ldquoA low-latency MAC protocolfor wireless sensor networksrdquo in Proceedings of the Interna-tional Conference on Wireless Communications Networkingand Mobile Computing (WiCOM rsquo07) pp 2816ndash2820 IEEESeptember 2007

[6] J Polastre J Hill and D Culler ldquoVersatile low power mediaaccess for wireless sensor networksrdquo in Proceedings of the2nd International Conference on Embedded Networked SensorSystems (SenSys rsquo04) pp 95ndash107 November 2004

[7] M Buettner G V Yee E Anderson and R Han ldquoX-MAC ashort preamble MAC protocol for duty-cycled wireless sensornetworksrdquo in Proceedings of the 4th ACM International Confer-ence on Embedded Networked Sensor Systems (SenSys rsquo06) pp307ndash320 November 2006

[8] Y Zhang N Meratnia and P Havinga ldquoOutlier detectiontechniques for wireless sensor networks a surveyrdquo IEEE Com-munications Surveys and Tutorials vol 12 no 2 pp 159ndash1702010

[9] K Stone and M Colagrosso ldquoEfficient duty cycling throughprediction and sampling in wireless sensor networksrdquo WirelessCommunications and Mobile Computing vol 7 no 9 pp 1087ndash1102 2007

[10] S H Lee J H Park and L Choi ldquoAMAC traffic-adaptivesensor network MAC protocol through variable duty-cycleoperationsrdquo in Proceedings of the IEEE International Conferenceon Communications (ICC rsquo07) pp 3259ndash3264 June 2007

[11] K J Wong and D Arvind ldquoSpeckMAC low-power decen-tralised MAC protocol low data rate transmissions in speck-netsrdquo in Proceedings of the 2nd IEEE International WorkshopMulti-Hop Ad Hoc Networks From Theory to Reality (REAL-MAN rsquo06) May 2006

[12] M Avvenuti P Corsini P Masci and A Vecchio ldquoIncreasingthe efficiency of preamble sampling protocols for wirelesssensor networksrdquo in Proceedings of the Mobile Computing and

Wireless Communications International Conference pp 117ndash122September 2006

[13] M Avvenuti and A Vecchio ldquoAdaptability in the B-MAC+protocolrdquo in Proceedings of the International Symposium on Par-allel and Distributed Processing with Applications (ISPA rsquo08) pp946ndash951 Sydney Australia December 2008

[14] S Mahlknecht andM Bock ldquoCSMA-MPS a minimum pream-ble sampling MAC protocol for low power wireless sensornetworksrdquo in Proceedings of the IEEE International Workshopon Factory Communication Systems (WFCS rsquo04) pp 73ndash80Vienna Austria September 2004

[15] T Zheng S Radhakrishnan andV Sarangan ldquoPMAC an adap-tive energy-efficient MAC protocol for Wireless Sensor Net-worksrdquo in Proceedings of the 19th IEEE International Parallel andDistributed Processing Symposium (IPDPS rsquo05) April 2005

[16] S Lee J Choi J Na and C-K Kim ldquoAnalysis of dynamiclow power listening schemes in wireless sensor networksrdquo IEEECommunications Letters vol 13 no 1 pp 43ndash45 2009

[17] H Byun and J So ldquoQueue-based adaptive duty cycle control forwireless sensor networksrdquo in Algorithms and Architectures forParallel Processing Proceedings of the 11th International Confer-ence ICA300 2011 Melbourne Australia October 24ndash26 2011Part II vol 7017 of Lecture Notes in Computer Science pp 205ndash214 Springer Berlin Germany 2011

[18] H Byun and J Yu ldquoAdaptive duty cycle control with queuemanagement inwireless sensor networksrdquo IEEETransactions onMobile Computing vol 12 no 6 pp 1214ndash1224 2013

[19] T-H Hsu T-H Kim C-C Chen and J-S Wu ldquoA dynamictraffic-aware duty cycle adjustment MAC protocol for energyconserving in wireless sensor networksrdquo International Journalof Distributed Sensor Networks vol 2012 Article ID 790131 10pages 2012

[20] R Jurdak P Baldi and C V Lopes ldquoAdaptive low powerlistening for wireless sensor networksrdquo IEEE Transactions onMobile Computing vol 6 no 8 pp 988ndash1004 2007

[21] Y Song Y Fang and Y Zhang ldquoStochastic channel selectionin cognitive radio networksrdquo in Proceedings of the 50th AnnualIEEEGlobal Telecommunications Conference (GLOBECOM rsquo07)pp 4878ndash4882 November 2007

[22] S Misra V Tiwari and M S Obaidat ldquoLACAS learningautomata-based congestion avoidance scheme for healthcarewireless sensor networksrdquo IEEE Journal on Selected Areas inCommunications vol 27 no 4 pp 466ndash479 2009

[23] Y Song X Zhu Y Fang and H Zhang ldquoThreshold optimiza-tion for rate adaptation algorithms in IEEE 80211 WLANsrdquoIEEE Transactions onWireless Communications vol 9 no 1 pp318ndash327 2010

[24] G I Papadimitriou M Sklira and A S Pomportsis ldquoA newclass of 120576-optimal learning automatardquo IEEE Transactions onSystemsMan and Cybernetics Part B Cybernetics vol 34 no 1pp 246ndash254 2004

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics


Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks



The sender

The intended receiver

The receiver

Wake-upperiod

Wake-upperiod

Wake-upperiod

Sending ACK

Sending continuousshort preambles Receiving ACK

Receiving adata packet

Sending adata packet

Figure 1 The time diagram of XMAC

of data packet has been finishedHowever B-MAC inherentlysuffers from the excessive long preamble accompanied withthe transmission of data packet and the overhearing ofnonintended receivers X-MAC [7] has been proposed as anenhanced LPL scheme fixing the drawbacks of B-MAC X-MAC adopts a series of short preambles to avoid overhearingIn X-MAC a sender transmits a short preamble and waitsfor a response from a receiver for a short time as shown inFigure 1 And the sensor node samples the channel whenit wakes up If it is the intended receiver upon receivingthe short preamble it replies with ACK to trigger the trans-mission of data packet Otherwise it goes to sleep to avoidoverhearing As compared with synchronousMAC protocolssuch as S-MAC [3] and T-MAC [4] these asynchronousMAC protocols have the advantage of not requiring any timesynchronization among sensor nodes

In LPL scheme the best performance can be obtainedby appropriately selecting the polling interval 119862 which is akey parameter in determining energy used to communicatebetween sensor nodes The optimal polling interval heavilydepends on traffic loads of the sensor node In low trafficloads a large polling interval takes advantage of saving energyof sensor node while a relatively small polling interval con-sumes less energy spent for communication between sensornodes in high traffic loads Also the polling interval must beadapted during runtime since traffic condition may not beknown a priori [8] If a priori assumption about traffic loadsis not accurate energy of sensor node will be wasted and theintended goal of application cannot be achieved For instanceelectronic security system which is installed in home detectssuspicious objects To achieve this requirement sensor nodesshould be in sleep mode for most of the time when there areno events of interest

Recently there have been a few studies on pollinginterval adaptation in duty-cycled WSNs The representativetechnique employed by existing polling interval adaptationschemes such as [9 10] guesses the traffic condition basedon previous results of channel sampling Note that thisapproach is referred to as Dynamic LPL (DLPL) scheme InDLPL scheme the sensor node obtains information about thenumber of consecutive busy (idle) sampling and dynamically

adjusts the polling interval by adopting the following Thepolling interval is increased after 119880 idle polling Similarly119863 consecutive busy pollings induce a decrease of pollinginterval This approach is simple and intuitive but has someproblems Most importantly it can make a scheme either tooaggressive or too conservative depending on 119880 and 119863 Forexample BoostMAC [9] which employs 119880 = 1 and 119863 = 1

reacts well to traffic changes as it immediately increasesdecreases the polling interval But it introduces unnecessarypolling interval fluctuations in stable environments whichcauses deterioration of performance Additionally it cannotbe expected to converge to optimal polling interval

In this paper we propose a stochastic polling intervaladaptation algorithm based on learning automata techniquewhere a sensor node dynamically adjusts its polling intervalaccording to the probability distribution of selecting thepolling intervals and selects the optimal polling intervalby learning In the proposed algorithm we adopt energyefficiency which is defined as the number of packets deliveredper energy unit to update the probability distribution associ-ated with the polling intervals Additionally our evaluationsverify that the proposed algorithmmakes the polling intervalconverge to the optimal value

Our main contributions are summarized as follows

(i) We consider the problem of runtime adaptation ofpolling interval in duty-cycled WSNs Fixed pollinginterval may degrade the performance of LPL schemewhen traffic loads are not known a priori

(ii) We investigate the problem of adapting the pollinginterval via stochastic learning automata To the bestof our knowledge our work is the first attempt toprovide a learning automation based solution foradaptation of polling interval in duty-cycled WSNs

(iii) We propose a novel polling interval adaptation algo-rithm which dynamically adjusts the polling intervaland the polling interval asymptotically converges tothe optimal value Our simulation results show thatthe proposed algorithm converges toward the optimalvalue



2 Related Work










[1205721(119905) 1205722(119905) 120572



119899(119905)] at time 119905


119899(119905)] be the set



Input 120572


Learning automata

Output 120573








119870to




0 1199051 ] and 119905(119894) =

[119905119894minus1



119894minus1 119905119894] 119894 = 1 2 to get



















119894




119864119894= 119864119904+ 119864119903

= (119875119905119909119878119901+ 119875119903119909119878119886)(

119862119894+ 119877119897

2 (119878119901+ 119878119886)

) + 119875119905119909119878119889

+ 119875sleep119862119894 + 119875119903119909119877119897+ 119875119905119909119877119886+ 119875119903119909119877119889

(1)

where 119864119904and 119864



119903as 119864119894 Hence the



119894 and updates





119863119894(119905) =

119869119894(119905)

119867119894(119905)

(2)




119894(119905)





119894

119880119894(119905) = 119863

119894(119905) + 119877

119894(119905) (3)



119894(119905)]



119894at



Algorithm 2

(1) Initialization

(i) Set 119875(0) = [1198751(0) 119875

119870(0)] where 119875

119894(0) = 1119870 119894 =


119894(0) 119869119894(0)119863

119894(0)119880



119899according to 119875(0)

(iv) Set119873119899and 119864

119899to zeros



record119873119899and 119864

119899until 119905(1)

(2) At time 119905


efficiency119872

if 119883119899gt 119872 119872 = 119883

119899 (4)


119869119899(119905) = 119869

119899(119905) + (119883

119899119872)


119899(119905) = 119867

119899(119905) + 1


119899(119905) by setting119863

119899(119905) = 119869

119899(119905)119867119899(119905)


119894(119905) by setting 119880

119894(119905) = 119863

119894(119905) + 119877

119894(119905)




119875119894(119905) = max (119875

119894(119905) minus (

1

119877119870) 0) if 119894 = 119898 119894 = 1 119870

119875119894(119905) = sum

119894 =119898

119875119894(119905) if 119894 = 119898

(5)

(viii) Set119873119899 119864119899 and119883

119899to zeros


119862119898and stop





119875119903 10038161003816100381610038161 minus 119875

119898






1

2

3

4

5

6

7


Traj

ecto

ry o

fCi




1 1198622 119862

7] =





0

01

02

03

04

05

06

07

08

09

1

10 20 30 40 50 60

Prob

abili

ty


P1

P2

P3

P4

P5

P6

P7


0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

100 200 300 400 500

Ener

gy co

nsum

ptio

n (m

W)

Simulation time (s)




5 Conclusion







References




























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











2 Related Work










[1205721(119905) 1205722(119905) 120572



119899(119905)] at time 119905


119899(119905)] be the set



Input 120572


Learning automata

Output 120573








119870to




0 1199051 ] and 119905(119894) =

[119905119894minus1



119894minus1 119905119894] 119894 = 1 2 to get



















119894




119864119894= 119864119904+ 119864119903

= (119875119905119909119878119901+ 119875119903119909119878119886)(

119862119894+ 119877119897

2 (119878119901+ 119878119886)

) + 119875119905119909119878119889

+ 119875sleep119862119894 + 119875119903119909119877119897+ 119875119905119909119877119886+ 119875119903119909119877119889

(1)

where 119864119904and 119864



119903as 119864119894 Hence the



119894 and updates





119863119894(119905) =

119869119894(119905)

119867119894(119905)

(2)




119894(119905)





119894

119880119894(119905) = 119863

119894(119905) + 119877

119894(119905) (3)



119894(119905)]



119894at



Algorithm 2

(1) Initialization

(i) Set 119875(0) = [1198751(0) 119875

119870(0)] where 119875

119894(0) = 1119870 119894 =


119894(0) 119869119894(0)119863

119894(0)119880



119899according to 119875(0)

(iv) Set119873119899and 119864

119899to zeros



record119873119899and 119864

119899until 119905(1)

(2) At time 119905


efficiency119872

if 119883119899gt 119872 119872 = 119883

119899 (4)


119869119899(119905) = 119869

119899(119905) + (119883

119899119872)


119899(119905) = 119867

119899(119905) + 1


119899(119905) by setting119863

119899(119905) = 119869

119899(119905)119867119899(119905)


119894(119905) by setting 119880

119894(119905) = 119863

119894(119905) + 119877

119894(119905)




119875119894(119905) = max (119875

119894(119905) minus (

1

119877119870) 0) if 119894 = 119898 119894 = 1 119870

119875119894(119905) = sum

119894 =119898

119875119894(119905) if 119894 = 119898

(5)

(viii) Set119873119899 119864119899 and119883

119899to zeros


119862119898and stop





119875119903 10038161003816100381610038161 minus 119875

119898






1

2

3

4

5

6

7


Traj

ecto

ry o

fCi




1 1198622 119862

7] =





0

01

02

03

04

05

06

07

08

09

1

10 20 30 40 50 60

Prob

abili

ty


P1

P2

P3

P4

P5

P6

P7


0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

100 200 300 400 500

Ener

gy co

nsum

ptio

n (m

W)

Simulation time (s)




5 Conclusion







References




























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










Input 120572


Learning automata

Output 120573








119870to




0 1199051 ] and 119905(119894) =

[119905119894minus1



119894minus1 119905119894] 119894 = 1 2 to get



















119894




119864119894= 119864119904+ 119864119903

= (119875119905119909119878119901+ 119875119903119909119878119886)(

119862119894+ 119877119897

2 (119878119901+ 119878119886)

) + 119875119905119909119878119889

+ 119875sleep119862119894 + 119875119903119909119877119897+ 119875119905119909119877119886+ 119875119903119909119877119889

(1)

where 119864119904and 119864



119903as 119864119894 Hence the



119894 and updates





119863119894(119905) =

119869119894(119905)

119867119894(119905)

(2)




119894(119905)





119894

119880119894(119905) = 119863

119894(119905) + 119877

119894(119905) (3)



119894(119905)]



119894at



Algorithm 2

(1) Initialization

(i) Set 119875(0) = [1198751(0) 119875

119870(0)] where 119875

119894(0) = 1119870 119894 =


119894(0) 119869119894(0)119863

119894(0)119880



119899according to 119875(0)

(iv) Set119873119899and 119864

119899to zeros



record119873119899and 119864

119899until 119905(1)

(2) At time 119905


efficiency119872

if 119883119899gt 119872 119872 = 119883

119899 (4)


119869119899(119905) = 119869

119899(119905) + (119883

119899119872)


119899(119905) = 119867

119899(119905) + 1


119899(119905) by setting119863

119899(119905) = 119869

119899(119905)119867119899(119905)


119894(119905) by setting 119880

119894(119905) = 119863

119894(119905) + 119877

119894(119905)




119875119894(119905) = max (119875

119894(119905) minus (

1

119877119870) 0) if 119894 = 119898 119894 = 1 119870

119875119894(119905) = sum

119894 =119898

119875119894(119905) if 119894 = 119898

(5)

(viii) Set119873119899 119864119899 and119883

119899to zeros


119862119898and stop





119875119903 10038161003816100381610038161 minus 119875

119898






1

2

3

4

5

6

7


Traj

ecto

ry o

fCi




1 1198622 119862

7] =





0

01

02

03

04

05

06

07

08

09

1

10 20 30 40 50 60

Prob

abili

ty


P1

P2

P3

P4

P5

P6

P7


0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

100 200 300 400 500

Ener

gy co

nsum

ptio

n (m

W)

Simulation time (s)




5 Conclusion







References




























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation












119894(119905)





119894

119880119894(119905) = 119863

119894(119905) + 119877

119894(119905) (3)



119894(119905)]



119894at



Algorithm 2

(1) Initialization

(i) Set 119875(0) = [1198751(0) 119875

119870(0)] where 119875

119894(0) = 1119870 119894 =


119894(0) 119869119894(0)119863

119894(0)119880



119899according to 119875(0)

(iv) Set119873119899and 119864

119899to zeros



record119873119899and 119864

119899until 119905(1)

(2) At time 119905


efficiency119872

if 119883119899gt 119872 119872 = 119883

119899 (4)


119869119899(119905) = 119869

119899(119905) + (119883

119899119872)


119899(119905) = 119867

119899(119905) + 1


119899(119905) by setting119863

119899(119905) = 119869

119899(119905)119867119899(119905)


119894(119905) by setting 119880

119894(119905) = 119863

119894(119905) + 119877

119894(119905)




119875119894(119905) = max (119875

119894(119905) minus (

1

119877119870) 0) if 119894 = 119898 119894 = 1 119870

119875119894(119905) = sum

119894 =119898

119875119894(119905) if 119894 = 119898

(5)

(viii) Set119873119899 119864119899 and119883

119899to zeros


119862119898and stop





119875119903 10038161003816100381610038161 minus 119875

119898






1

2

3

4

5

6

7


Traj

ecto

ry o

fCi




1 1198622 119862

7] =





0

01

02

03

04

05

06

07

08

09

1

10 20 30 40 50 60

Prob

abili

ty


P1

P2

P3

P4

P5

P6

P7


0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

100 200 300 400 500

Ener

gy co

nsum

ptio

n (m

W)

Simulation time (s)




5 Conclusion







References




























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










1

2

3

4

5

6

7


Traj

ecto

ry o

fCi




1 1198622 119862

7] =





0

01

02

03

04

05

06

07

08

09

1

10 20 30 40 50 60

Prob

abili

ty


P1

P2

P3

P4

P5

P6

P7


0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

100 200 300 400 500

Ener

gy co

nsum

ptio

n (m

W)

Simulation time (s)




5 Conclusion







References




























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation













References




























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









research article stochastic polling interval...

Documents