Achieving Energy Efficiency in Wireless Sensor Networks Using Dynamic Channel Polling and Packet Concatenation

Shama Siddiqui,Anwar Ahmed Khan,Sayeed Ghani

1 Department of Computer Science,DHA Suffa University,Karachi,Pakistan

2 Department of Computer Science,Institute of Business Administration,Karachi,Pakistan

3 Department of Computer Science,Sindh Institute of Management&Technology,Karachi,Pakistan

Abstract:There has been a significant interest of researchers to combine different schemes focused on optimizing energy performance while developing a MAC protocol for Wireless Sensor Networks(WSNs).In this paper,we propose to integrate two cross-layer schemes:dynamic channel polling and packet concatenation using a recent asynchronous MAC protocol “Adaptive & Dynamic Polling MAC”(ADPMAC).ADP-MAC dynamically selects the polling interval distribution based on characterization of incoming traffic patterns using Coefficient of variation(CV ).Packet Concatenation(PC)refers to combining the individually generated data packets into a single super packet and sending it at the polling instant.Also,the Block Acknowledgement(BA)scheme has been developed for ADP-MAC to work in conjunction with the packet concatenation.The proposed schemes have been implemented in Tiny-OS for Mica2 platform and Avrora emulator has been used for conducting experiments.Simulation results have revealed that the performance both in terms of energy & packet loss improves when ADP-MAC is used in conjunction with the additional features of PC&BA.Furthermore,the proposed scheme has been compared with a stateof-art packet concatenation primitive PiP(Packet-in-Packet).It has been observed that ADP-MAC supersedes the performance of PiP in terms of PDR(Packet Delivery Ratio)due to better management of synchronization between source and sink.

Keywords:energy efficiency; concatenation; dynamic;polling

I.INTRODUCTION

Due to the rapid emergence and deployments of Internet of Things(IoT),the requirement of efficient operation of low power devices including sensor nodes has become crucial.Wireless Sensor Networks(WSNs)play a vital role in IoT through sensing and collecting data from the environment[1–4].WSNs have unique characteristics and challenges;one of the major issues being energy consumption[5].

Channel polling refers to the activity of listening to the channel;when the receiver listens to the channel,it could be referred as receiver“polls”the channel.Efficient selection of Channel polling(listening)intervals has been regarded as a major performance criterion for the MAC protocols in WSN,as they relate directly to other parameters such as duty cycle and network lifetime; it is the mechanism using which,the receiver nodes decide the instants at which they should listen to the channel[6].In this context,dynamic channel polling refers to polling or listening the channel at intervals dynamically set based on the network traffic conditions.Various dynamic polling mechanisms have been found in literature,such as deciding polling intervals based on the history of previous busy or idle polls[7]or applying machine learning algorithms for predicting the next polling interval[8].

Packet Concatenation(PC)is the process of combining several packets into one large(super)packet so that the energy of transmitting separate headers bytes can be reduced.There are three major reasons due to which packet concatenation has been advocated for WSNs[9].Firstly,the overhead at MAC and Physical layers remain fixed for the packets with varying payload lengths.In case of transmitting data packets with short payloads,higher channel resource gets wasted and lower throughput is achieved,Duty cycle mechanism of the MAC protocols provides the second reason of packet concatenation:while the radio remains inactive in the sleep period,the sensing activity continues and there must be some means for storing and queuing the data.Thirdly,in many applications,WSN data is often routed to the same sink;therefore,sending a concatenated packet improves both the delay and energy performance.

Packet concatenation can particularly be applicable for the WSN scenarios where the receiver may not be available or awake while the sender continues to generate packets.In such a situation,the sender may concatenate all the generated packets before the receiver becomes available and polls the channel.This would help the network to reduce the overhead bytes while the receiver may also know the historical data generated when it was not available.An example of such applications could be intrusion or event detection,where a burst of packet is often generated in case of event occurrence.If the receiver does not poll the channel immediately when the event occurred,the senders may concatenate all the packets being generated and sends a super packet.

In addition to packet concatenation,block acknowledgement(BA)transmission has also been regarded as an important energy saving mechanism as it further reduces the transmission bytes.In this method,instead of acknowledging each data packet separately with an explicit Acknowledgement(ACK)packet,the packets received together are acknowledged by the receiver using a single packet[10].Therefore,block acknowledgement schemes have been used widely in wireless and wireless sensor networks in the past.

This paper presents enhancement to our previous protocol ADP-MAC[11]by integrating the mechanisms of packet concatenation&block acknowledgement transmissions.ADP-MAC was designed based on the concepts of dynamic channel polling where the nodes poll(listen to)the channel based on the dynamic adaptation of polling interval distributions with the variations in incoming traffic patterns.Although ADP-MAC showed performance improvement in comparison to the previous protocol SCP-MAC[12],it transmits the data and acknowledgement packets separately,even if they are transmitted at the same instant.In this work,we propose performance enhancement for ADP-MAC by integrating it with additional features known in literature for improving the efficiency of MAC protocols[13,14].The performance evaluation of proposed schemes has been conducted by comparing the results of ADP-MAC simulations with and without implementation of PC&BA.

By integrating the above mechanisms,various real world applications of WSN and IoT requiring data collection from many and transmit to one,can be facilitated.For example,the WSN working on applications that require periodic monitoring and generating huge amounts of data such as continuous monitoring of field temperature or habitat monitoring,can benefit due to reduced energy consumption and improved throughput.However,it is to be noted that the proposed scheme may not serve as a good choice for applications with real time communication requirements because additional delay would incur due to transmission of long packets on the network.

For evaluating the performance of proposed integration of ADP-MAC with PC and BA,we conducted comparison with PiP(Packet-in-Packet),which is an energy efficient link layer paradigm[15].PiP has been developed based on the unique combination of Concurrent Transmission(CT),constructive interference and capture effect.Concurrent transmission refers to multiple nodes transmitting identical packet at the same instant(within 0.5µs);constructive interference refers to the superimposition of signals when identical packets are transmitted within 0.5µs;and,capture effect allows the receiver nodes to capture a signal from air and demodulate.Furthermore,the authors also integrated a novel but optional hardware operation of PA(power amplifier)with PiP.PA is a hardware component which is embedded within a WSN node and facilitates the receiver to convert a low power RF signal into high power,which in turn ensures effective capture effect.PiP has been claimed to outperform the Carrier Sense Multiple Access(CSMA)based link layer protocols since PiP does not need the mechanisms of collision avoidance and network layer routing.

In summary,major contributions of this article are detailed below:

•To propose performance enhancement for a dynamic polling-based MAC protocol ADP-MAC,by integrating it with schemes of packet concatenation and block acknowledgement transmissions.

•To implement the schemes of packet concatenation and block acknowledgement in Tiny-OS environment for the Mica2 platform.

•To compare performance of ADP-MAC with and without the proposed schemes of PC and BA.

•To compare performance of a CSMA-based protocol ADP-MAC with CT based paradigm PiP.

Rest of this paper has been organized as follows:Section II presents the brief review of existing MAC schemes designed for optimized energy performance.Section III details the proposed integration of packet concatenation and block acknowledgement with ADPMAC.Section IV presents the theoretical analysis for the proposed schemes.Section V details the experimental settings.Section VI discusses the results and presents analysis of the findings.Finally,section VII concludes the work and suggests the directions for future research.

II.RELEVANT WORK

There are various MAC schemes proposed in previous research that targets optimizing the energy efficiency of WSN.In this section,we summarize some of these schemes and describe the packet concatenation mechanisms used in this paper at the end.

2.1 Duty Cycle

Duty cycle management has been regarded as one of the most fundamental mechanisms for optimizing energy efficiency of WSN.In this mechanism,the nodes are sent to a periodic sleep,due to which energy consumption reduces which could occur due to the issues such as idle listening and overhearing.Various WSN applications are known to benefit from the duty cycle schemes; for example,two MAC schemes Duty Cycle Scheduling based on Residual Energy(RidE)&Duty Cycle scheduling based on Next Wake-up Time(NeWT)have been proposed for Underwater Sensor Networks[16].In RidE,the duty cycle and transmission schedules of nodes are decided based on the residual energy,whereas in NeWT,the nodes alternate their sleep periods based on their scheduled wake-up periods to avoid the collisions.

Dynamic Duty cycle based(D3)protocol has been developed in[17]for large scale WSN.To meet the real time transmission requirements,the sleep-wakeup schedules between the two communicating nodes are staggered.The nodes are assigned grades based on their distance from the sink,and the node nearest to sink is assigned grade zero; data is transmitted from the nodes of higher grades to the ones with lower.For the two nodes with adjacent grades,the periodic sleep-wake-up schedules are staggered and hence a data propagation pipeline is created to reduce the end-to-end delay.

One of the major challenges associated with the dynamic duty cycle in the asynchronous networks is the high computation complexity.The nodes have to maintain information about schedule of neighbours and also transmit theirs regularly.Furthermore,in most of the existing mechanisms,the packets farther from sink may get delayed channel access due to which urgent data from these may suffer.

2.2 Receiver Initiated Schemes

The receiver initiated MAC schemes have been advocated by the researchers in order to reduce the idle listening as well as overhearing time of receiver nodes.In this context,Adaptive Wake up intervals have been proposed for preamble sampling based MAC protocols in[18].This protocol has introduced 4 mechanisms of transmitting initial control frame messages,traffic estimation function,control frame message,and adaptive function.The receiver nodes initiates the control frame and traffic estimation function uses the information about previous cycles such as Idle Listening Times(ILTn,ILTk),Traffic Status Register(TSR)and number of wake-up without receiving beacon message(NWwbm)to reduce the wake-up frequency of receiving nodes in future.The adaptive function running at the receiver node predicts the next wake-up interval(WUI)of the transmitter.The protocol has shown to result in accurate convergence for the traffic which in turn reduces energy consumption and latency while improving the throughput and packet delivery ratio.

Another protocol Enhanced Receiver Initiated MAC protocol for facilitating uni-cast,multi-cast and broadcast traffic has been proposed in[19].Each receiver periodically wakes up and transmits a beacon.For the uni-cast traffic,the node holding data for a receiver,transmits the data in response to the beacon.However,since the nodes operate in asynchronous fashion,EnRI-MAC facilitates reducing the redundant transmission for the cases of broadcast and multi-cast.For the broadcast scenario,upon receiving a beacon,the sender transmits a wait request instead of data; this wait request packet specifies the time at which data shall be transmitted.The receiver nodes go to sleep immediately upon receiving this beacon and wake up at the packet arrival time.The waiting time has been kept equal to the beacon period so that the sender can receive beacons from all the nodes.Thus,the sender needs to transmit a single broadcast and all the nodes successfully receive it.On the other hand,for the case of multi-cast transmission,the sender node maintains the information about the multi-cast group.In case it receives a beacon from the node which is not in the required group,it ignores the message.For the other receivers,the wait request is transmitted.Subsequently,a single multi-cast message is sent similar to the broadcast case.EnRI-MAC contributes to reducing the energy consumption of WSN by reducing the re-transmitted packets.

2.3 Multi-Channel Protocols

Muti-channel protocols are developed with the focus on optimizing the energy efficiency and guaranteeing the QoS.An advertisement based distributed MAC protocol(Adv-MMAC)has been developed where the nodes switch their interfaces between multiple channels dynamically[20].The two phases:advertisement(ADV)and data,run in parallel; the nodes with data contend during ADV phase to access the Advertisement channel so they may transmit the ADV packet.Once the acknowledgement packet is received,the nodes switch to the free channel for communication.The parameters of end-to-end delay,data delivery ratio and energy efficiency all have shown improvement for Adv-MMAC as compared to other MAC variants.A schedule based Dual-Channel Dual-Slot(DCDS)MAC protocol for delay sensitive WSN has been proposed in[21].Using the designed algorithm,each node can select two time slots on either of the two channels.On the default channel,each slot is split into three periods Common Frequency(CF),Control Message(CM)and Data Message(DM).The frequencies to be used are shared by the nodes during CF,the control information such as wake-up instant is shared during CM and finally,the data is shared during DM period.The neighbor which is closer to the gateway is chosen by each sender.It has been shown that the use of two channels ensures efficient throughput,delay,energy consumption and network lifetime; the number of channels can further be increased based on the application requirements to cater to more nodes.

The major issue associated with the multi-channel protocols is that to-date,the multi-channel hardware platforms for WSN have not become common.Therefore,the protocols deploying single channel for all types of packet transmission have been more advocated.

2.4 Dynamic Polling

Similar to the other approaches,dynamic polling/listening to the channel offers energy advantage to the nodes by reducing their idle listening energy.Moreover,the delay also gets reduced as the nodes get an opportunity to receive the packet earlier as compared to if the channel is polled at fixed intervals only.Over the past two decades,several dynamic polling based protocols have been proposed for WSN.In AX-MAC[22],series of short preambles are used to create a synchronization between sender and receiver.The major novelty brought by AX-MAC was that the nodes adjust their polling intervals dynamically based on the network traffic conditions.Markov chain has been used to analyse the threshold parameters of U and D and make the decisions to adapt the polling intervals accordingly.Here,U refers to the number of times a node detects that the channel is idle before increasing the polling interval,whereas,D refers to the number of times a node should detect that the network is busy before it decreases the polling interval.Although,AXMAC made dynamic decisions,it was a challenge to select appropriate values of U and D as the protocol may become too sensitive or conservative.

Similarly,in most of the other dynamic polling schemes,the decision to poll the channel has been made on the history of past channel activity;however,mostly,the polling is conducted at constant intervals instead of using probabilistic distribution to decide the next polling instant.Based on this gap analysis,we developed Adaptive & Dynamic Polling based MAC protocol(ADP-MAC)in our previous work[11].This protocol has a novel contribution of dynamically selecting the‘polling interval distributions’based on the characterization of the incoming traffic patterns.The protocol identifies incoming traffic patterns by analyzing the coefficient of variation(CV)of the generation

time of incoming packets at the senders.Mainly,CVis used to evaluate whether packets are being generated at sender nodes at fixed intervals or at intervals which can be modeled through exponential distribution.The major aim for assessing theCVof incoming traffic is to be able to match the channel polling instants with the traffic arrival instants(to reduce both the delay and energy consumption).

ADP-MAC has been defined as“adaptive”as it has capability to monitor the variation in traffic pattern for a period specified initially(TAD)and can select the polling interval distribution which suits the arrival pattern.Moreover,the protocol has also been defined as“dynamic”because it could dynamically select the best suited polling interval distribution based on the arrivals during the operation.The detailed operation of the protocol under different network scenarios has already been covered in our previous work[11].In Figure 1,the basic operation of ADP-MAC has been illustrated by taking an example case of a single transmitter and receiver.Also,Table 1 lists the basic concepts and parameters used in the description of ADPMAC.

Table 1.Definitions and parameters used to described operation of ADP-MAC.

Table 2.Simulation parameters.

As illustrated by Figure 1 the sender and receiver follow same period for a cycleTcyclewith wake-up durationTW& sleep durationTS.The wake-up and sleep periods of both the nodes are of similar duration but are asynchronous to each other.In the first cycle,it is assumed that the sender has a data packet,hence,it transmits short strobe preambles of durationTprewith short pausesTpausewhich is acknowledged by the receiver using an Early Acknowledgment(EA).The data transmission takes place followed by the acknowledgment reception.TWhas shown to be extended both at the sender&receiver because once the data transmission initiates,the nodes do not sleep until the completion of the communication activity.The sender goes to sleep,whereas receiver remains awake for an additional durationTADD,expecting more queued packets to be received from other senders.

As the usual operation of ADP-MAC continues,the receiver nodes keep track ofCVof the generation time of the incoming data packets.The polling interval distribution is then decided based on the analysis ofCV.Initially,the receiver monitors the value ofCVfor the initial adaptive periodTADto select the polling interval distribution based on arrival patterns.After the initial adaptation,the protocol keeps on monitoring the value ofCVupon reception of each data packet and keeps on iteratively updating it for the Repetitive Dynamic periodTDY N; the new value ofCVcalculated at the reception of each packet is referred as incrementalCV.TADhas been kept longer thanTDY Nto ensure that the statistics collected for Adaptive period achieves stability.Here it is to be noted that although the value ofCVis monitored for each received data packet,the polling interval distribution is only switched after a periodTADorTDY Nbased on theCVcalculated during either of these periods.Hence,in thenthcycle shown in Figure 1,the polling intervals are not same as in the cycle 1.This is because deterministic polling intervals have been assumed in the first cycle,whereas for the second,it has been considered that the protocol has switched to exponential polling based on the value ofCV Tcalculated over the periodTDY N.

Figure 1.Basic operation of ADP-MAC.

Figure 2.Single hop operation of PiP[15].

The calculation ofCVhas been detailed previously in[11],however,for the sake of completion,we summarize the derivation here:The following standard formula for calculatingCVhas been used:

whereµis the mean &σis the standard deviation of the distribution[23]

Since we are interested to calculate and update the value ofCVas each packet arrives(to ensure synchronization between the packet arrival and channel polling instants),let x be the time interval between last two consecutive arrivals.The first time a packet arrives,CVis intended to be calculated using the following equations:

Standard equations for meanµand standard deviationσare given as

Substituting in Eq.(5)

This can be solved to obtain:

Finally,

where

Since the series at this stage contains only a single value of x,CVis assumed as 0.Subsequently,upon reception of each packet,the equations given below have been used for incremental update ofCV.The equation forCVis updated as:

where

After calculating each value ofCV,the following equations are iteratively updated:

Thus,at each instant of decision making i.e.afterTADorTDY N,the value ofCVis compared against the value ofCV T.IfCVis greater thanCV T,exponential Polling is selected for the next period and deterministic is selected otherwise.The optimal value ofCV Tcan be decided based on the traffic arrival patterns of the WSN application.

Two types of polling interval distributions have been used to illustrate the design of ADP-MAC:deterministic&exponential.The decision to select one of these depends on the selected threshold value ofCV(CV T).For the design of ADP-MAC,the value ofCV Tis set as 0.8,as the value ofCVfor Constant Bit Rate(CBR)traffic is 0,when there is no variation in the arrival patterns and it is 1 for the Poisson arrivals,where there is a significant variation between the inter arrival durations of packets[24].Moreover,asCVgoes from 0 to 1,the randomness of the traffic increases for CBR,all the way to being Poisson.Therefore,a value of 0.8 has been chosen as a threshold so that the channel is polled almost instantly when each Poisson arrival is generated.This implies that for values ofCV T <0.8,the receivers poll the channel at fixed intervals(referred to as deterministic polling),whereas for the values ofCV T >0.8,channel is polled at exponential intervals(referred to as exponential polling).

The experimental evaluation for the selection ofCV Tand its impact on the performance of ADP-MAC has been presented in our previous work[25].It was found that as we increasedCV Tfrom 0 upwards to 1,the performance in terms of energy consumption and delay of the CBR traffic improved while the equivalent performance of Poisson traffic degraded.However,at some point aroundCV Tof 0.8,the improvements in CBR levelled off while the degradation of Poisson continues.Hence,the optimal performance was expected somewhere for the values ofCV Tin between 0.6 and 0.8.Hence,for this article,we selected a value of 0.8 as a possible optimal point.

2.5 Packet Concatenation

Packet concatenation has also been an important research direction to reduce the energy cost of transmitting redundant bytes.Some of the innovative schemes being used for concatenating packets and reducing the transmission cost are detailed in this section.

Recently,dynamic packet aggregation has been proposed for Electromagnetic based Wireless Nano Sensor Networks(EM-WNSNs)operating in the range of THz[26].The focus of this scheme is to aggregate packets generated by nano-sinks in order to reduce the mismatch between the bandwidth requirements of EM-WNSNs and assignment by the IoT backhaul.The Narrowband-IoT(NB-IoT)and Low power wide area networks(LPWAN)have often been used as IoT backhaul for the EM-WNSNs.Due to the contention of EM-WNSNs and M2M and Human to Human communications,there is a bandwidth limitation from the perspective of backhaul network.Also,it is to be noted that in various applications of EM-WNSNs,it is better to discard the packets than buffer,as these packets become outdated very quickly.Hence,whenever the gateway is queried by the backhaul,the gateway adjusts the packet aggregation size of nano sinks using probabilistic process; this aggregation is performed based on the information about network connectivity and backhaul bandwidth.The proposed scheme results in a better energy efficiency for the WNSNs and better utilization of bandwidth for the IoT backhaul.

Multihop Adaptive with packet concatenation-MAC(MAC2)is also an adaptive protocol which adjusts the listening intervals based on the traffic arrival patterns[14].The protocol operates for a single sink environment and concatenates several packets(according to the predefined concatenation threshold)before adding the MAC header.The proposed scheme has been successful in improving the energy efficiency as well as latency when compared with the previous MAC variants that did not deploy concatenation.

Packet in Packet(PiP)is a packet concatenation mechanism that relies on the concept of Concurrent Transmission(CT)and performs concatenation of packets while they are in air[15].The Capture Effect and Constructive Interference have been deployed for concatenating the packets.The scheme has been developed both for the single and multi-hop topologies.The authors proposed that instead for waiting for the synchronization information specific to each sender,The sender nodes can transmit their packet such that their payload gets concatenated with the packets which have already been transmitted by their neighbors.The design of PiP in single hop environment has been illustrated by Figure 2[15].

In Figure 2,the sink H is shown to transmit the synchronization information initially.All the sender nodes A,B,C and D synchronize precisely based on the received information.Subsequently,the sink receives information from each source during inpacket slots.These inpacket slots are randomly chosen by the senders as there is no mechanism to preassign these slots.In addition,PiP also utilizes the hardware operation of Power Amplifier(PA);PA enhances the power of received RF signal ensuring that the packets received from long distances can also be captured and can thus interfere constructively instead of getting cancelled due to comprising of weak signals.The sender nodes need to switch the PA on when transmitting during inslots and switch it off when no transmission takes place.

Based on the above scheme,all the sources become able to transmit their information in a single transmission round instead of waiting for the next transmission cycle as is the case with conventional 802.15.4 based protocols.Also,PiP has the capability of collecting different packets from the neighboring nodes unlike other CT based protocols which require transmission of identical packets;it is to be noted that in the frame shown in Figure 2,the synchronization information is identical for all the network.

PiP offers advantage of reduced synchronization overhead and data collection time.However,the nodes must send their packets such that the payload can efficiently be integrated into larger packets.Furthermore,the protocol PiP heavily relies on the hardware operation of power amplification in order to ensure concatenation of packets regardless of transmission power and distance of source nodes.

III.PROPOSED SCHEME

3.1 Modifying ADP-MAC

The major contribution of this paper is to integrate the mechanisms of PC and BA with the functionality of ADP-MAC.When ADP-MAC would be used in conjunction with PC,the packets which were waiting before the arrival of EA from the sink would be concatenated into a single Super packet and would be transmitted as a single unit.Subsequently,the Super packet would be acknowledged by using a single BA packet.Although the advantage of PC and BA schemes is more visible for the cases where all nodes send packets to a central sink,the scheme would also bring performance improvement to the WSN otherwise.Even if the destination of all packets is not same for multiple sink environments,concatenation would still lead to energy advantage as all the packets to same destination would be concatenated into a single super packet.

The operation of ADP-MAC including the proposed features of packet concatenation&block acknowledgment has been illustrated in Figure 3.It has been assumed in the operation illustrated that when the receiver polls the channel,the sender had already generated 3 packets which are concatenated into a single Super packet.Since the destination of all the packets is assumed to be same,the header information is combined leading to energy advantage.Figure 3 has shown concatenation of only three packets for the purpose of illustration;however,the proposed concatenation scheme for ADP-MAC could entertain any number of packets.

Since sender includes the feildnumber_of_packetsin the preamble,the receiver sends the EA for the packets based on the space available in buffer.In Figure 3,it has been assumed that the receiver sends an EA for all the 3 packets(to be received as a super packet)and remains awake till the expected reception time,even if theTWexpires.Here,Time required to transmit data(TD)is flexible based on the number of packets contained in a single super packet.The sender transmits the super packet,for which the receiver transmits a block acknowledgement; this acknowledges all the packets that were successfully received in a Super packet.Similarly,in the second cycle exponential polling intervals are selected and 2 packets(4&5)are sent as a Super packet.The remaining operation of ADP-MAC then remains the same as discussed earlier.

Figure 3.ADP-MAC integrated with packet concatenation and block acknowledgement schemes.

Figure 4.Comparison of ADP-MAC(original)with ADP-MAC(modified using PC and BA).

The comparison of operations of ADP-MAC(original)with the modified ADP-MAC suggested in this article has been pictorially illustrated in Figure 4.The two scenarios of communication between sender and receiver have been shown,where the first illustration is for ADP-MAC without concatenation,and the second is for the modified ADP-MAC with concatenation.It has been shown that 3 packets were generated at the sender and could not be transmitted because the receiver was not available,for the first case,as the receiver gets available,the packets are individually transmitted with their separate header bytes.On the other hand,for the second case,all three packets were combined into a single super packet and transmitted while the header bytes were reduced.As a result,the super packet has been shown to complete transmission earlier as compared to the first case.Clearly,the energy consumption for the case of modified ADP-MAC would also be lower as lesser header bytes are transmitted.

3.2 Definition of Packet Structures

For integrating ADP-MAC with packet concatenation,the super packet has been designed to combine the individual packets into a single super packet.The sequence no.,payload & Cyclic Redundancy Check(CRC)bytes of each packet has been separately added within the super packet for the identification of each packet.Here,it is to be noted that Super Packet itself has a CRC which is calculated based on entire packet.Also,each internal packet within a Super Packet has a separate CRC calculated for it only.In case all internal packets are correctly received,the CRC of Super Packet will be received correctly.This would allow the receiver to quickly identify the correct reception instead of needing to check CRC of each internal packet.On the other hand,if one or more internal packets are not correctly received,the CRC of the Super Packet would not be receive correctly.Only for these situations,the receiver will check the CRC of each internal packet to identify the erroneous packets.This strategy reduces the computation cost at the receiver as packets can be quickly processed.

Generally,the limit of packets which can be concatenated into a single packet is decided based on the physical layer parameters.Since ADP-MAC is built on top of the open source implementation of CSMA in tiny-OS[27],we had a limit of transmitting 250B together.Therefore,in each super packet a maximum of 5 packets could be concatenated.The structures of Data & Super packets have been shown in Figure 5(a)&(b).The feildsGeneration_Time&Ready_for_TxTimehave been used for calculating theCVof the incoming traffci patterns.For the source node,both of these feilds contain same value of time; for the forwarding nodes,the value ofReady_for_TxTimeis used which comprises of the value of time at which the packet was received from the source node,because as soon as the node receives a packet,it becomes ready to be transmitted.

The difference between the number of transmitted bytes in case of transmitting the concatenated packet is clearly illustrated in Figure 5(a)&(b).The PHY&MAC headers remain same as in the data packet illustrated by Figure 5(a),but theSuperAppHeadervaries slightly from theAPP_headeras it does not contain the feildsData_Seq# &GenerationTime;these two have now been moved to individual data packets to identify them as well as to record their generation time.CRC bytes of each internal packet are separately added to detect the erroneous internal packets within a super packet,as previously explained.Each time a concatenated packet is received,the receiver transmits a block acknowledgement to confrim the correct or erroneous reception.

The structure of acknowledgement/block acknowledgement packet has been shown in Figure 5(c).A single type of ACK packet has been used for acknowledging the reception of a single data packet and Super packet.to_Addressis the address of node which sends or forwards the packet,whereas theoriginal_sourceis the node which generates the packet.In case of single hop network,both feilds contain the same address.On the other hand,the two feilds can be used in combination for the case of multi-sender(multiple source nodes),multi-hop networks where different nodes can generate same packet ID’s.Similarly,Pkt_Nohas also been included in the ACK packet so the sender can identify the Seq#of data packet for which the confrimation has been received.

The feild#_of_Pktsin the ACK packet differentiates whether the ACK packet is meant to acknowledge a single data packet or a Super packet.When ACK is sent for a single data packet,the feild contains 1.On the other hand,if ACK packet needs to acknowledge a Super packet,a block acknowledgment scheme has been developed by using the feild #_of_Pkts.As it is a single byte feild,it has been designed to contain a series of 0s&1s to represent the successful or negative acknowledgement for each packet concatenated in the received Super packet.

The 5 least signifciant bits of the feild#_of_Pktsin ACK packet are used for Selective Acknowledgement(SACK)as shown in Figure 5(c);the 3 most signifciant bits represent the number of packets for which the acknowledgement has been sent(there can be 5 or less packets concatenated into a Super packet at any given time,as previously discussed).The SACK is subsequently analyzed by the sender through matching the binary sequence with the sequence in which the data packets were concatenated into a Super packet.

Figure 5.Implemented packet structures in ADP-MAC(a)data packet(b)super packet(c)structure of block acknowledgement.

IV.THEORTICIAL ANALYSIS

In this section,we present a high level theoretical analysis to compare the performance of ADP-MAC with the proposed schemes of PC&BA against ADP-MAC without these schemes.

Let’s assume we need to transmit N data packets,each of size S.Also,the size of overhead/header bytes in each data packet is taken asβ.Originally(without PC&BA),transmitting N data packets would also imply the transmission of N acknowledgement packets(each of sizeα).

Thus,originally transmitted bytes would be:

When we integrate ADP-MAC with BA and PC,we essentially reduce the number of bytes transmitted,which can be computed in terms of percentage reduction in bytes transmitted,X as follows:

For the integration of BA with ADP-MAC,we acknowledge all the received packets with a single acknowledgement packet instead of N packets.Hence,the bytes reduced when BA is implemented can be given in terms of percentage reduction in bytes transmitted(with BA)as:

Finally,when we integrate the ADP-MAC with PC,header bytes of only one packet are transmitted instead of for N packets.Also,when applying PC,the BA is also implemented as all the concatenated packets are acknowledged with a single acknowledgement packet.Thus,the percentage reduction in bytes transmitted(with PC)is computed as:

By plotting Eq.(10),the reduction of bytes achieved by implementing BA while varying N(S constant at 50 Bytes)has been illustrated in Figure 6,whereas variation of S(N constant at 10)has been illustrated in Figure 6(b).

Figure 6(a)illustrates that the percentage reduction in bytes transmitted initially varies sharply with the increasing N.This is because initially,the impact of block acknowledgement is higher as the data packets comprised on lesser bytes(N was small)which implies that reduction in the acknowledgement bytes represents a higher impact ; however,as N continued to increase above 5,the percentage reduction becomes nearly constant and not as prominent because the bytes of data packets will be higher in number which reduces the impact of BA relatively.Moreover,Figure 6(b)illustrates the effect of varying data packet size(S)on the percentage reduction of bytes transmitted; here,it is seen that as S is increased,the reduction in bytes transmitted lowered.This is because when the size of data packet is smaller,the effect of reduction due to the block acknowledgements is more prominent;in comparison,when S increases,the effect of block acknowledgement reduces because more bytes would be transmitted for data and there would be a relatively lower impact of block acknowledgements.

Next,by plotting Eq.(11),the impact of PC has been seen on the percentage reduction of bytes transmitted by varying N and S in Figure 7(a)&(b)respectively.

Figure 6.Reduction of transmitted bytes due to block acknowledgement.(a)Varying number of data packets,(b)Varying size of data packet.

Again,it has been shown in Figure 7(a)that increasing N results in a higher reduction of percentage of transmitted bytes.Similarly,Figure 7(b)illustrates that with the increasing S,the reduction of bytes shows a lower impact.

Finally,the effect of PC and BA has been compared in Figure 8.

As indicated in both the Figure 8(a)and(b),the impact for bytes transmitted is higher when PC is applied as this scheme also integrates the mechanism of BA.For both the increase of S and N,the impact of PC is seen to be higher.

Now,let’s compare a simple example of transmitting 3 packets(N=3)with and without using the proposed methods of BA and PC.First,let’s assume that the packets were transmitted without BA and PC as shown in Figure 9:

Figure 9 illustrates that 3 packets have been transmitted,each comprising on overhead of sizeβand payload of size P.Each packet has been separately sent,and has been acknowledged by transmitting acknowledgement packet of sizeα.

Figure 7.Reduction of transmitted bytes due to packet concatenation.(a)varying number of data packets,(b)varying size of data packet.

Let S=50 Bytes,α=14 Bytes andβ=14 Bytes

If BA and PC both are not implemented,the original bytes transmitted can be computed using Eq.(8)

Thus,X0would be 192 Bytes

If BA is implemented for the same case,as shown in Figure 10,the transmitted bytes shall be reduced.

Now,as shown in Figure 10,there is no reduction in the bytes transmitted for the data packets as the packets are still transmitted separately; as a result,all the overhead as well as payload bytes will be transmitted.However,now instead of sending 3 Acknowledgement packets as in Figure 9,only single block acknowledgement packet is sent,due to which the transmitted bytes reduce.

Figure 8.Comparing impact of packet concatenation and block acknowledgements(a)varying number of data packets,(b)varying size of data packet.

Figure 9.Packet transmission without BA and PC.

Figure 10.Packet transmission with BA.

Hence,the percentage reduction in transmitted packets can be calculated using Eq.(10):

Thus,if block acknowledgement is used,14 percent reduction has been obtained for 3 data packets.In case,more than 3 packets are acknowledged using a single block ack,higher percentage reduction would be obtained as earlier depicted by Figure 6(a).

Figure 11.Packet transmission with PC.

Finally,if we implement the mechanism of PC along with the BA,as shown in Figure 11,the percentage of bytes transmitted shall further reduce.

Figure 11 illustrates that the 3 packets which were shown to be separately transmitted in Figure 9 and Figure 10 have been concatenated into a single super packet.As a result,the overhead bytes which were earlier sent for each data packet,are combined resulting in reduced number of transmitted bytes.

Using Eq.(11):

Since the scheme of PC also integrates BA,therefore,the percentage reduction is doubled as compared to when only BA was used.Therefore,the impact of BA and PC has been seen on reducing the transmitted bytes.This concept is further illustrated in terms of energy consumption and delay by conducting experiments as described in the next sections.

V.EXPERIMENTAL SETUP

ADP-MAC and the performance improving mechanisms suggested in this work(PC&BA)have been implemented in Tiny-OS for Mica2 platform and the experiments were conducted using Avrora emulator[28].The single hop star topology with multiple senders in a 30-node network has been set for this experiment,where all the nodes except sink generated and sent 50 packets to the centralized sink.

The message generation on each node was varied from 10 to 250 sec,whereas the polling interval was set constant at 50 msec.Two types of arrival distributions,CBR(CA)& Poisson(PoA)are used for evaluating the performance of ADP-MAC with and without applying concatenation.Here,CA represents the WSN applications with periodic monitoring requirements and PoA represents the applications which generate irregular/urgent traffic such as ehealth,intrusion detection,vehicular networks,etc.Similar assumptions have been made about the traffic arrivals by previous authors such as[29,30].It is to be noted that CBR and Poisson arrivals have been assumed based on theirCV.However,ADP-MAC could deal with any traffic distribution and will dynamically adapt its polling interval distribution in accordance with the change inCVof incoming traffic patterns.

Up to 5 data packets were concatenated(all the packets which were generated before the subsequent poll were concatenated)and one Super packet was sent in place of separate data packets.It is to be noted that although Mica2 mote is furnished with a 128Kbytes program flash memory[31],only up to 5 packets have been buffered in the ADP-MAC implementation,due to the limitation of Physical layer used in this work,as it could transmit a packet with a maximum length of 250 bytes[31].For the concatenated packets,the block acknowledgement with the Selective Acknowledgement(SACK)mechanism has been used.All the other major simulation parameters have been detailed in table.2.

VI.RESULTS&DISCUSSION

The performance evaluation of the proposed schemes packet concatenation and block acknowledgement transmission for the MAC protocol ADP-MAC has been conducted by studying the parameters,energy consumption and packet loss.All experiments were repeated 4 times and the results have been presented with 95%confidence intervals.

6.1 Energy Consumption

Energy Consumption Rate per node has been calculated by dividing the total energy consumption by the number of nodes and simulation duration.The performance results based on mean energy consumption rate per node for the variants of ADP-MAC(with and without packet concatenation&block acknowledgements)for both the CBR and Poisson Arrivals have been presented in Figure 12.

Considering the case of ADP-MAC without concatenation,it can be seen in Figure 12 that the energy consumption results for Poisson Arrivals are better as compared to the ADP-MAC operating with Poisson arrivals.The protocol has been specifically designed to work with varying traffic patterns and therefore,it provides a better support for Poisson traffic through efficiently matching the channel polling instants with the packet arrival instants.Therefore,ADP-MAC can support the applications of WSN efficiently as mostly these applications generate Poisson traffic.

As expected,the energy consumption for the cases of packet concatenation has been found to reduce for both the CBR and Poisson Arrivals.Since the number of transmitted packets and those a node can keep in buffer at any given time is same for all the generation intervals,the difference in energy consumption is solely obtained due to the reduced transmission of bytes as a result of packet concatenation and block acknowledgement transmission.

The impact of varying the internal parameter of ADP-MAC,TAD,has also been studied to evaluate its impact over the energy performance of protocol.The value ofTAD,was initially kept at 25 sec,however,in Figure 13,the value ofTADwas varied from 15 to 35 sec,while the message generation interval was held constant at 150 sec.These settings ofTADimply that now the receiver would listen to the channel for 30 sec(when set so)before making the decision to switch the polling distribution.The energy consumption for the cases of CBR arrivals do not change as compared to the Figure 12 because there would be of no difference in the arrival pattern regardless for how long the receiver listens to the channel.However,when the Poisson arrivals were evalued,the energy consumption for both the cases of without and with packet concatenation has found to be decreasing with the increasing value ofTAD.This is because now the receiver would be in a better position to make a switch of polling interval distribution,as it would analyse more values of the previous arrival instants.

From the above results and discussions,it has been observed that at each message generation interval,significant number of packets were concatenated.The percentage of concatenated packets as well as their corruption has been illustrated in Figure 14.Since concatenation is more expected for the Poisson arrivals,(due to generation of packets at irregular intervals),only this case has been depicted.As seen in Figure 14,the fraction of packets concatenated reduces with the increasing message generation interval.At the low generation intervals,there will be high number of packets generated before they could be transmitted at polling instants.Thus,more packets would concatenated due to frequent generation.Gradually,as the message generation interval is increased,there will be fewer packets at each poll,resulting in lesser packets being concatenated.

Figure 12.Energy consumption results for ADP-MAC with and without packet concatenation&block ACK.

Figure 13.Energy consumption results for ADP-MAC by varying TAD .

Figure 14.Percentage of packet concatenation and corruption.

Figure 15.Packet loss for ADP-MAC with and without packet concatenation&block ACK.

Figure 16.Comparison of packet loss rate due to buffer overflow and other reasons for the cases of ADP-MAC with and with-out packet concatenation.

It is important to note that the block acknowledgment transmission does not only conserve energy due to reduction in transmission of redundant overhead bytes,but it also reduces the re-transmission energy.The use of block ack ensures that there is no need of checking CRC of internal packets in case the CRC of Super packet is correctly received; furthermore,only selective packets are re-transmitted instead of resending the complete Super packet in case of corruption of one or more internal packets(within a Super packet).Figure 14 shows the fraction of super packets partially corrupted with varying message generation intervals.This observation reveals that at each message generation interval,there was a certain fraction of packets which were not fully acknowledged by a block acknowledgment.In such situation,the CRC of internal packets(within a super packet)were checked to identify the exact corrupted packet.As a result,only those packet/s were re-transmitted which were corrupted instead of transmitting a full super packet.Moreover,it has been shown in Figure 14 that upon increasing the message generation interval,the fraction of corrupted super packets reduces,due to lesser network load and packet collisions.

6.2 Packet Loss

Packet loss ratio has been calculated by using the following formula:

As illustrated in Figure 15,packet loss has been found to be slightly lower for ADP-MAC with packet concatenation for both the CBR and Poisson Arrivals.This is because for the cases of packet concatenation when a sender node wins the channel,it transmits a long Super packet in continuity without the need to wait for a single acknowledgement.In the absence of packet concatenation,even if the sender receives an EA for a certain number of packets,there are chances that it would not be able to transmit the second packet in the queue due to the channel getting occupied by some other node(for the cases of heavy traffic generation as the buffer would be expected to overflow).

The packet loss occurred due to being tail dropped(due to buffer overflow)and due to other reasons(collisions,non-availability of channel,failed retransmission attempts,non-reception of EA,etc)has been compared for the cases of with and without concatenation.Figure 16 reveals the comparison for these cases for Poisson arrivals.

As indicated by Figure 16,for the results of ADPMAC without PC,the packet loss rate due to buffer overflow is higher as compared to the packets dropped due to other reasons.When concatenation is applied,the loss happening due to packets being tail dropped has reduced significantly as the packets will be transmitting as a single larger packet.On the other hand,the losses happening due to other reasons have been found to be contributing a major proportion of packet loss which is expected due to the transmission of larger packets.

6.3 Comparison with PiP

ADP-MAC and PiP have been compared for their performance in terms of Packet Delivery Ratio(PDR).Here,we define PDR as the ratio of the number of packets successfully received at the sink to the number of packets generated by the source nodes.The single hop topology of 30 nodes was set up for these experiments.2 cases were simulated for PiP:PiP without PA and PiP with Power Amplification(PA).Since Avrora could not offer a chance to perform RF symbol level transmission,we ensured synchronization of the nodes in order to model the constructive interference.Moreover,for simulating PiP with PA,the packets were sent using high Transmit(Tx)power so all the packets can be received.

Figure 17(a)presents the PDR comparison of 2 protocols when source nodes are placed at unequal distance from the single sink,whereas Figure 17(b)shows the case where all the source nodes are placed at the same distance.As shown in Figure 17(a),when the source nodes are located at single hop but unequal distances from the sink(closest node at 0.5 m and farthest node at 3m),PDR appears to be significantly better for ADP-MAC at the farthest node when compared with that of PiP with or without PA.As reported by the authors in[15],PiP has the capability of reducing data collection time in single hop environment,however due to lack of effective synchronization between the source nodes,the packet loss remains significantly high for PiP without PA.Clearly,the randomness involved in the choice of inpacket slots increases the probability of collisions which results in significantly degrading the PDR of PiP.On the other hand,when PiP is coupled with PA,the PDR improves for the farthest node because now,better capture effect would take place as all the packets will be received at the sink with high power and effective superimposition would take place.Yet,ADP is shown to perform better at the farthest node due to better match of channel listening instant of sink and packet generation instant of the source nodes due to the dynamic polling.

Also,the PDR results for ADP-MAC and PiP at the closer node are presented in Figure 17(a).It has been found that for both variants of PiP,the PDR remains same(100%)for the closer node.This is because the closer nodes in PiP with or without PA will behave in a similar manner and even without the use of PA,a strong capture effect would take place.In contrast,there is not a significant difference on the performance of ADP-MAC as regardless of the distance,the nodes would win the channel based on the CSMA and dynamic listening intervals.Therefore,it has been observed that PiP performs well for the closer nodes in comparison to ADP-MAC,whereas ADP-MAC exhibits more consistent performance.

Figure 17.PDR for ADP-MAC and PiP for:(a)unequal distance(b)equal distance.

Subsequently,Figure 17(b)reveals the performance comparison of ADP-MAC and PiP when the source nodes were situated at an equal distance of 1m from the sink.The results have been presented for any 2 random nodes out of total 30 nodes in the network.It has been found that for both nodes,ADP-MAC results in lower PDR as compared to PiP with PA.However,the performance of ADP-MAC has been found to be nearly same for both the nodes.Similarly,PiP with PA also shows similar level of performance for both nodes.On the other hand,PiP without PA shows lower performance for node A and higher for node B,although the distance of both the sources was kept same.Therefore,it has been revealed that the effect of randomness in the selection of inpacket slots could heavily govern the performance of PiP,particularly when it is used without PA,whereas ADP-MAC shows more consistence performance for each source regardless of its distance from the sink.

VII.CONCLUSION&FUTURE WORK

This paper presented an extension to dynamic polling-based MAC protocol ADP-MAC by integrating the mechanisms of packet concatenation and block acknowledgement transmissions.It has been found that packet concatenation and block acknowledgements lead to significant performance improvements in terms of energy consumption and packet loss.When the packets are concatenated in WSN,the transmission cost reduces due to(i)reduced transmission of packet overhead bytes and(ii)reduced packet loss due to buffer overflow.Also,transmitting the block acknowledgement reduces the overhead bytes.Super packet is transmitted with an overall CRC along with a separate CRC for internal packets.This mechanism ensures that the transmission cost is further reduced by identifying the corrupted packet and only retransmitting that instead of retransmitting entire super packet.

Furthermore,ADP-MAC coupled with packet concatenation and block acknowledgement has been shown to outperform a state-of-art link layer paradigm PiP.This is because the design of PiP requires synchronization between the nodes to achieve efficient inpacket concatenation.Despite the initial synchronization between the nodes,each sender in PiP selects a random inpacket slot for further transmissions.As a result,there are certain cases where packets cannot be properly concatenated due to possibilities of clock drifts and other delays.On the other hand,ADP-MAC achieves a better match between the packet transmission and channel polling instants due to analysis ofCV.

In future,the proposed idea of packet concatenation can be applied to facilitate recent MACs of WSN and WBAN such as TAP-MAC.Also,real-world applications of WSN,particularly based on bursty transmission can be integrated with proposed packet concatenation scheme to reduce the transmission cost and achieve better energy performance.