nW-MAC: multiple wake-up provisioning in asynchronously...

Ann. Telecommun. (2011) 66:567–582DOI 10.1007/s12243-010-0226-7

nW-MAC: multiple wake-up provisioningin asynchronously scheduled duty cycle MACprotocol for wireless sensor networks

Md. Obaidur Rahman · Muhammad Mahbub Alam ·Muhammad Mostafa Monowar · Choong Seon Hong ·Sungwon Lee

Received: 5 January 2010 / Accepted: 5 November 2010 / Published online: 25 November 2010© Institut Télécom and Springer-Verlag 2010

Abstract To reduce the energy cost of wireless sen-sor networks (WSNs), the duty cycle (i.e., periodicwake-up and sleep) concept has been used in severalmedium access control (MAC) protocols. Althoughthese protocols are energy efficient, they are primarilydesigned for low-traffic environments and thereforesacrifice delay in order to maximize energy conser-vation. However, many applications having both lowand high traffic demand a duty cycle MAC that isable to achieve better energy utilization with minimumenergy loss ensuring delay optimization for timely andeffective actions. In this paper, nW-MAC is proposed;this is an asynchronously scheduled and multiple wake-up provisioned duty cycle MAC protocol for WSNs.The nW-MAC employs an asynchronous rendezvousschedule selection technique to provision a maximumof n wake-ups in the operational cycle of a receiver.The proposed MAC is suitable to perform in both low-and high-traffic applications using a reception window-

Md. O. Rahman · M. M. Alam · M. M. Monowar ·C. S. Hong (B) · S. LeeDepartment of Computer Engineering,College of Electronics and Information,Kyung Hee University (Global Campus),Seoul, South Koreae-mail: [email protected]

Md. O. Rahmane-mail: [email protected]

M. M. Alame-mail: [email protected]

M. M. Monoware-mail: [email protected]

S. Leee-mail: [email protected]

based medium access with a specific RxOp. Further-more, per cycle multiple wake-up concept ensures op-timum energy consumption and delay maintaining ahigher throughput, as compare to existing mechanisms.Through analysis and simulations, we have quantifiedthe energy-delay performance and obtained results thatexpose the effectiveness of nW-MAC.

Keywords Wireless sensor network (WSN) ·Medium access control (MAC) · Wake-up schedule ·Operational cycle · Duty cycle · Energy · Delay

1 Introduction

Currently, one of the main challenges for wireless sen-sor networks (WSNs) is the handling of large variationsin traffic loads (i.e., low and high) for various applica-tions. Additionally, many monitoring applications (e.g.,fire alarm, intruder detection, tracking, etc.) demandtimely and effective actions with minimum delay. How-ever, the energy-efficiency constraint often makes itchallenging for WSNs to perform in real-time; in suchapplications a very low periodic traffic primarily existsfor a longer period of time in the absence of eventtraffic. Therefore, the WSN design protocols need toconsider an energy-delay trade-off in order to fulfillsuch application requirements.

Imperative and co-related research have shown en-ergy as the most decisive resource for any WSN, includ-ing monitoring networks; the demand for extended net-work life time motivates the design of energy efficientmedium access control (MAC) protocols [1, 2]. Con-sequently, the periodic wake-up and sleep schedule(namely duty-cycle)-based MAC is introduced to reduce

568 Ann. Telecommun. (2011) 66:567–582

energy consumption [1–4]. A sensor node in these pro-tocols has its own wake-up schedule, i.e., maintains awake-up cycle (hereafter, we interchangeably refer tothis as either an operational cycle or a cycle). The cycleduration is more adaptable for existing duty cycle MACprotocols in working energy efficiently in low trafficonly. Whereas, the blow of the of event data or heavytraffic is still unrevealed in designing such protocols.

To perform in real-time it is important to ensurea minimum level of delay. Unfortunately however,conventional sensor MAC design places the emphasison maximizing energy conservation [1, 2]. In typicalprotocols, each node wakes-up once per cycle to receiveany potential data and sleeps for the remainder of thecycle period. Senders that cannot transmit their data inthat wake-up time have to wait for at least a receiver’ssleep period; this is defined as a sleep delay in D-MAC [5]. Thus, in a multi-hop path, the worst-possibledelay is bounded by either the cycle or sleep periodsof the intermediate nodes. Moreover, we argue thatthe per cycle single wake-up strategy for the receiveris not suitable for event traffic; when an event occursthe delay increases if the nodes cannot forward theevent-triggered heavy traffic due to the sleep periods.Although the concept of multiple packet reception ateach wake-up is introduced in [1, 3], in heavy trafficenvironment their unbounded reception endures morecollisions resulting in very low throughput and largedelay due to retransmissions.

So far, based on rendezvous selection betweensender and receiver, duty cycle MAC protocols aremainly categorized into synchronous and asynchronousprotocols. In a synchronous MAC [4, 6], nodes in thesame neighborhood are time-synchronized for simulta-neous wake-up at the beginning of a common cycle andtransmit their data using the basic carrier sense multipleaccess with collision avoidance (CSMA/CA). However,nodes simultaneous wake-up in synchronous protocolsresults in higher contentions and collisions, and theclock synchronization overhead makes it difficult toimplement in a multi-hop WSN.

Conversely, in asynchronous protocols, nodes main-tain their individual cycles at random and the sender-receiver rendezvous is selected either through thesender or receiver initiated medium access. In senderinitiated MAC [2, 3], a long preamble (equivalent to atleast a receiver’s cycle or sleep period) from the senderis used to rendezvous with the receiver. Such a pream-ble causes a significant amount of energy waste and de-lay. In comparison to its sender initiated counterparts,a receiver initiated MAC [1] provides better energyand delay performance, by avoiding the unnecessarymedium occupancy time for a preamble transmission.

However, the existing receiver initiated schemes stillsuffers due to idle listening [4] and sleep delay at thesender end. Furthermore, recent asynchronous MACprotocols [1, 3] lags with traffic variations and rarelyhandle heavy traffic scenarios. Hence, improvementsare still needed for the issues discussed.

In this paper, motivated by the above observa-tions, nW-MAC is proposed. This is an asynchronouslyscheduled receiver- initiated duty cycle MAC proto-col for WSNs comprising n wake-up provisions in thereceiver cycle. This protocol attempts to satisfy thefollowing:

– Minimize energy consumption with optimal delayunder low traffic conditions.

– Maximize energy utilization with improvedthroughput and minimum delay in the handling ofheavy traffic.

The major contributions of nW-MAC can be sum-marized as follows: (1) An innovative multiple wake-up provisioning in sensor MAC is introduced, utilizingan energy and delay efficient asynchronous rendezvousschedule selection by the receiver. (2) A receptionwindow-based bounded channel access is used, limit-ing the reception opportunity (RxOp) of a receiverin receiving multiple packets at each wake-up. (3)Adaptability is ensured on-demand basis in using then wake-up provisions by the receiver at every cycle. (4)Performance of the nW-MAC is analyzed in detail andevaluated through extensive simulations.

The rest of the paper is organized as follows: Section 2presents a background study on existing sensor MACprotocols, Section 3 introduces the preliminary designconsiderations, Section 4 includes detailed protocoloperations, Section 5 analyzes nW-MAC in terms ofenergy consumption and delay, Section 6 presents per-formance evaluations by simulation efforts and com-parisons, and Section 7 provides our conclusions withfuture considerations.

2 Related work

In recent years, a number of MAC protocols have beenproposed for WSN, and as mentioned in Section 1,existing asynchronous MAC protocols can be catego-rized into sender- and receiver-initiated MAC. Thevery early senderinitiated MAC protocol, B-MAC [2],optimizes the WSN energy performance by employinga preamble sampling technique. In recent research,[1, 3, 7], it has been determined that the detection of apreamble forces a node to receive the entire preamblecausing large overhearing [4] oriented energy loss at

Ann. Telecommun. (2011) 66:567–582 569

the non-intended receivers. Although B-MAC is anenergy-efficient protocol and works well in very lowtraffic, it is unable to avoid preamble overhead andsleep delay.

The X-MAC [3] overcomes the overhearing prob-lem with the long preamble transmission in B-MAC.In X-MAC, strobe preambles (having receiver IDs)are transmitted by the sender, as shown in Fig. 1a.At the asynchronous wake-up (i.e., after the wake-upinterval), if the receiver detects the strobe preamble(s),it simply sends an acknowledgment (ACK), otherwiseit goes to sleep. Upon receiving an early ACK fromthe receiver, the sender then transmits the data andreceives the data ACK; hence, X-MAC partially re-duces the preamble transmission time. However, wehave identified that X-MAC still has two basic prob-lems. Firstly, it wastes energy due to the strobe pream-bles since, on average for half of the receiver’s cycle(i.e., wake-up interval), a sender needs to transmit thepreambles to be in rendezvous with the receiver, asshown in Fig. 1a. Secondly, at the sender end, thesleep delay due to the receiver’s wake-up interval (orcycle) is very large; on average, this delay is half of thecycle period at each hop. Additionally, X-MAC doesnot consider the consequences of high traffic, which isthe primary concern in the handling of a monitoringnetwork.

The asynchronous protocol WiseMAC [7] is lim-ited to work only for the downlinks of infrastructureWSNs, and adaptively minimizes the preamble length.At each reception, the receiver informs the senderabout the next wake-up schedule. Senders store thereceiver’s wake-up time (offset) and accordingly reducethe preamble length for further transmissions for that

Wake-up Interval / Cycle

Idle ListenStrobe

Preamble

Wake-up to send

Tx

Rx

Tx

Rx

Data

Data

P Sleep

Sender(S)

Receiver(R)

P P P P P P

A

A

P

P

A ACK

A

A

(a)Beacon Interval / Cycle

Wake-up to send

Tx

Rx

B B

Data

Data

BSender

(S)

Receiver(R)

Idle ListenBeacon / ACKB Sleep

Tx

RxB

B

(b)

Fig. 1 Protocol operations: a X-MAC and b RI-MAC

specified wake-up. nW-MAC uses the offset storingspecification of WiseMAC where each sender-receiverpair asynchronously selects their rendezvous.

The most recent protocol, RI-MAC [1], brings thefirst receiver initiated MAC concept into WSNs. Theoperation of RI-MAC is shown in Fig. 1b; a receiver ateach wake-up cycle (or beacon interval) sends a requestmessage (i.e., beacon) for potential data reception.Thus, the receiver initiated concept is able to decreasethe energy loss and medium occupancy time associatedwith a preamble as well as the delay that occurs inthe sender initiated protocols [2, 3, 7]. A sender inRI-MAC listens and checks the medium for a requestbeacon from the receiver. Upon its receipt, the sendertransmits the data and receives the data ACK (i.e.,also a beacon for more data request). Although withlow traffic RI-MAC performs efficiently, the mediumaccess mechanism of this protocol has a number ofshortcomings. We have observed that in RI-MAC theidle listening and sleep delay period at the sender endare very long due to the receiver’s beacon interval (as inFig. 1b). On average, the idle listening and sleep delayare lengthened to half of the receiver’s cycle at eachhop. Therefore, RI-MAC still causes unnecessary en-ergy loss and delay. Furthermore, with heavy traffic theunbounded multiple packets reception at each wake-up of the receiver increases the overall network con-tentions. Hence, collision oriented retransmissions inRI-MAC minimize throughput causing significant de-lay in multi-hop path.

Over the years, several synchronous MAC protocols,(e.g., S-MAC [4], D-MAC [5], T-MAC [6], R-MAC[8], DW-MAC [9]) have been proposed for WSNs.The energy-efficient S-MAC [4] is the first synchronousMAC protocol for WSNs, and several extensions ofthis innovative work are introduced later on. D-MAC[5] is such a protocol with a precisely synchronizedstaggered solution to achieve a minimum delay overa multi-hop path, where nodes first receive a packetand immediately transmit it to downstream. R-MAC[8] provides another staggered solution using a controlpacket (pion frame). In nW-MAC, we also propose asimilar but asynchronously scheduled staggered solu-tion, where network wide synchronization is avoidedand unlike [5, 8], receivers can receive multiple packetsat each wake-up

There exists a few well-known hybrid MAC proto-cols such as SCP-MAC [10] and Funneling-MAC [11].These protocols use the advantageous properties ofboth the synchronous and asynchronous schemes. InSCP-MAC, nodes are precisely synchronized like S-MAC, whereas in Funneling-MAC, only the nodes nearthe sink are synchronized.

570 Ann. Telecommun. (2011) 66:567–582

Despite the energy and delay efficiency, existingsynchronous and hybrid MAC protocols are not goodchoices for WSNs due to the synchronization over-head and their implementation complexity. nW-MACuses a determined asynchronously scheduled duty cycleapproach; each receiver maintains an individual cycleat random and senders follow the time differences topredict the scheduled rendezvous with the receiver con-sidering the potential clock drifts.

3 System model and preliminaries

3.1 Network model

nW-MAC is designed for a typical WSN scenario, asshown in Fig. 2a; it comprises multiple unidirectionaldata flow converging toward a single point or sink,resulting in a tree topology. The research for D-MAC[5] justifies our assumption of the tree topology fora duty cycle MAC protocol; the network consists ofstationary sensor nodes with a single routing path fromeach node to the sink.

nW-MAC is a receiver initiated MAC, there are Nreceiving nodes in the network, and the ith receiver isdenoted as ri. Each receiver has one downstream nodeand multiple upstream nodes. The downstream node ofri is denoted as di and the jth upstream node of ri isdenoted as uij.

E

CB

F

A

G H

D

I J

(a)

w0 w1 w2 w3 w0

Operational Cycle ( )

SoC SoC

iT

∨

4iT

∨

2iT

∨

3 4iT

∨

00

(b)

Fig. 2 a Network model and b in operational cycle (Ti) of noderi, the provisions of n wake-ups {w0, w1, w2, · · · , w(n−1)}; heren = 4

A flow is defined as the traffic from a particularsource node (either periodically, or upon the detectionof any event, or a combination of both). nW-MACconsiders the unicast transmission of a flow.

3.2 Assumptions

We assume that each receiver has an operational cycleand each cycle has an active operational period whichdetermines the duty cycle of a node [1, 3, 4]. However,the operational cycle and duty cycle for nW-MAC aredefined as follows:

Definition 1 Operational Cycle: an operational cycle ora node cycle is the time interval that includes multiplewake-up provisions. Let Ti denote the cycle of ri.

Definition 2 Duty Cycle: the duty cycle of a node isdefined as the ratio between a node’s active time to itsentire cycle time. Active time includes all of the actionsand activities of a node (i.e., channel access at single ormultiple wake-ups, medium listening, transmission andreception etc.).

In a cycle, nW-MAC provisions a number of wake-ups for a receiving node. Let n denote the number ofwake-ups in a cycle, and wk denote the identification(ID) of the kth wake-up where {k = 0, 1, · · · , (n − 1)},as shown in Fig. 2b. The value of n is set equal to 4based on extensive simulations; it produces the opti-mum energy-delay trade-off for a wide variation of sim-ulation environments. All of the subsequent sectionsof this paper use this same value for n. We furtherassume that each data transmission uses a request-to-receive (RTR) packet initiated by the receiving node ateach wake-up. RTR is similar to the beacon concept inRI-MAC [1], however, besides a request for data andacknowledgment a successful reception, a receiver innW-MAC also uses it for wake-up scheduling.

Definition 3 Wake-up Of fset: The n wake-up offsets ofa receiver originate from the start-of-cycle (SoC), tSoC

and the offset of the wk wake-up of ri is obtained by

tiwk

= tSoC + (k mod n)Ti

n, (1)

Therefore, in a cycle, the interval between consecu-tive wake-up provisions of ri is

Tin .

3.3 Initialization

We assume that each node chooses a random cycleduration at startup and remains active for a number

Ann. Telecommun. (2011) 66:567–582 571

of cycles. Furthermore, each node announces its cy-cle length to the upstream nodes using the basicCSMA/CA mechanism. However, the active period ofan upstream node uij initially depends on the cyclelength of the receiver ri which uij extracts from theannouncement of Ti. The active window of uij, denotedas Tactive, is given by

Tactive =Ti

n+ Tr, (2)

where, Tr is referred to as the RTR window which isdefined as the duration of the transmission of the RTRpacket and the receipt of its acknowledgment. Duringinitialization, node uij asynchronously wakes-up at ran-dom; after this point it wakes-up at a fixed intervalof Ti (i.e., the cycle length of uij’s receiver ri). Afterwake-up, uij checks the medium for a period of Tactive

before going back to sleep. Therefore, if ri transmits anRTR packet in each of its n wake-ups, it is guaranteedthat uij receives one of the n RTR packets of ri. Thisassumption can be verified by the dominating-awake-interval technique [12] and overlapping principle [13].

4 nW-MAC protocol operation

4.1 nW-MAC overview

In nW-MAC, an asynchronous scheme derives maxi-mum n wake-up schedules per cycle for each receiver.The schedule selection is mainly designed to optimizethe energy consumption and delay in accessing themedium.

After initialization and schedule selection, nodes fol-low the wake-up and sleep schedules. In basic nW-MAC, a receiver wakes-up at a scheduled receptiontime and sends an RTR in order to receive data fromthe senders. In contrast, a sender only wakes-up priorto the receiver’s specified wake-up time, and sendsdata upon receiving the RTR. Furthermore, to receivemultiple packets at each wake-up, a RxOp limit is usedfor the receiver.

To fulfill the demand of an event monitoring network,an adaptive version of nW-MAC is also proposed;under very low periodic traffic, a receiver maintains aper cycle single wake-up-based asynchronous staggeredschedule (see Section 4.4.1). Conversely, in response toevent-triggered heavy traffic, each receiver makes anon-demand adaptive and additive use of the provisionsfor n wake-ups in every cycle.

4.2 Asynchronous schedule selection

The proposed asynchronous scheme first selects a ren-dezvous schedule for each sender-receiver pair at oneof the n wake-ups of the receiver. To maximize energyconservation, both sender and receiver store their se-lected rendezvous, and accordingly, they wake-up tosend and receive, respectively.

To select the schedule, a receiver ri wakes-up at w0,w1, w2, and w3 within the cycle (Ti). At each wake-up,it transmits an RTR packet to its upstream node(s) uij ifthe medium is found free. Otherwise, the receiver waitsfor an RTR contention window period, denoted asCWRTR, and transmits the RTR afterward. On the con-trary, in every Ti interval, all uij asynchronously listento the medium for a period Tactive (derived in Section3.3). Therefore, if there is no collision originated loss,it can be guaranteed that any of the RTR packets (outof 4) transmitted by ri at different wks, is received bythe upstream nodes [12, 13]. Upon receiving the RTR,the upstream nodes wait for a SIFS and contend with arandom back-off (0 to CWRTR). An uij with an earlyback-off expiration replies with an acknowledgment(ACK) and stores that wk wake-up offset of ri as itsscheduled transmission (Tx) rendezvous in every Ti.

After receipt of the first ACK from any uij, ri waitsfor a maximum back-off (CWRTR) to receive furtherACK from another node, if any. The upstream node(s)that looses the previous contention pauses the back-off counter; when the medium becomes free it resumescounting to send an ACK. Accordingly, the wk wake-ups, at which ri receives an ACK from its upstreamnodes, are selected as the scheduled reception (Rx)rendezvous; therefore, for ri the maximum number ofscheduled wake-ups during Ti are less than or equal to n.

As in Fig. 3, node A transmits an RTR at each of itsfour wake-ups (w0, w1, w2, and w3), for every cycle; atw1 one of the RTRs is received and acknowledged by

A

B

C

D

w0 w1 w2 w3 w0

Active Window Sleep WindowRTR Window

Tactive

Tactive

Tactive

Tactive

RTR RTRACK

ACK

ACK

Fig. 3 Duty cycle after initialization. Node A’s upstream nodesB, C and D (as in Fig. 2a) have the active time Tactive in every TAinterval

572 Ann. Telecommun. (2011) 66:567–582

node B. Therefore, w1 is selected as the Tx rendezvousfor B with the receiver A. Again, nodes C and Dreceive the RTR at w3 of A. Now, if D sends an ACKfirst, C pauses the back-off and resumes whenever nodeD finishes transmission and vice versa. Hence, w3 isselected as the Tx rendezvous for C and D with theirreceiver A. Finally, in subsequent cycles node A wakes-up only at w1 and w3, since these are the selected Rxrendezvous for A.

An upstream node further selects a start-of-cycle(tSoC) for its own w0 wake-up. Let an uij have a Txrendezvous at ti

wk(i.e., wk wake-up of ri). Thus, the tSoC

for uij is chosen as

t1 = tiwk

−Ti

2n, t2 = ti

wk− (g1 + g2),

tSoC = rand (t1, t2), (3)

where rand(t1, t2) is a function that generates a ran-dom start-of-cycle ranging from the time offset t1 to t2.Hence, for n = 4, the tSoC or w0 wake-up of uij can be,at most, a Ti

8 unit time earlier than the wk wake-up ofri. Moreover, the guard time g1 is considered for thepotential clock drift [7], and the guard time g2 providesthe opportunity for receiving at least one data packet(from uij’s upstream node) by uij at its w0.

g2 = (slotTime × CWRTR) + TRTR + SIFS

+ (slotTime×CWData)+TData+SIFS+TACK, (4)

here, TRTR, TData, TACK are the required time peri-ods for transmitting RTR, data, and ACK packets,respectively. The CWData is the contention window forthe data packet, and slotTime is used to calculate themaximum back-off times.

Considering Fig. 4, suppose that node B has Tx ren-dezvous at w1 of A; hence, tSoC of B is selected betweenTA8 and (

TA4 − (g1 + g2)). After selecting tSoC, node B

commences sending an RTR at each of its w0, · · · , w3

wake-ups.

A

B

w0 w1 w2 w3 w0

w0 w1 w2

Transmission (Tx) Rendezvous

SoC

4AT

∨

2AT

∨

3 4AT

∨

00 8AT

∨

Reception (Rx) Rendezvous

w3

g2 g1

Fig. 4 Start-of-Cycle (SoC) selection for an upstream node of A;here, upstream node B has the Tx rendezvous at w1 wake-up of A

The proposed scheduling is adaptive and scalable toany topology changes in a WSN. To join the network,a new node or node having link failures listens to themedium for at least a maximum cycle period, Tmax,a system parameter. Therefore, within the coveragearea, that node is guaranteed to hear RTR(s) from itsneighbor(s) and is then able to select its receiver andschedules during regular MAC operations.

4.3 Basic nW-MAC

In basic nW-MAC, a receiver follows a receptionwindow-based MAC, where a reception window isdefined as

Definition 4 Reception Window: A variable length ac-tive period, Ti

rw, which reflects the RxOp limit of thereceiver ri in a specific reception rendezvous.

In a scheduled Rx rendezvous at wk (or tiwk

), node ri’smaximum Ti

rw length is bounded by either of two otheroffsets: (1) ri’s own Tx rendezvous at a downstreamnode di’s wake-up offset tdi

wk, and (2) ri’s another Rx

rendezvous at offset tiwl

, where l �= k. Thus, receiver ri

calculates Tirw as

T1 = tiwl

− tcurrent, T2 = tdiwk

− tcurrent,

Tirw = min(T1, T2) − g1, (5)

where tcurrent is the current time and the guard time,g1, is excluded for potential clock drift. Receiver ri

continuously update the instantaneous Tirw. As shown

in Fig. 5, in the cycle TB, node B has a scheduled Rx

A

B

E

F

w0

w1

w3w1

G

xRxR Tx

Tx

Tx

Tx

Rx

Rx

TBRW TB

RW

RTR Packet Data Packet

Fig. 5 Reception window-based channel access by node B, whereB has the downstream node A and upstream nodes E, F, G

Ann. Telecommun. (2011) 66:567–582 573

rendezvous at w0, w1 and w3 for the upstream nodes E,F and G, respectively. Here, the maximum T B

rw lengthat w0 is bounded by B’s Tx rendezvous with receiver A(at w1 of A). The maximum T B

rw at w1 is limited by B’sother Rx rendezvous (at w3 of B).

Algorithm 1 represents the core operations of theproposed reception window-based basic nW-MAC. Inevery cycle, a receiver ri periodically wakes-up at itsscheduled Rx rendezvous and transmits an RTR (if themedium is idle). Prior to sending the first RTR at eachwake-up, ri performs a back-off (0 to CWRTR) to avoidcollisions (line 3). However, if the medium is busy, ri

waits until the minimum of either the reception window(Ti

rw) or a duration ofTi2n (line 2). Meanwhile, it sends

the RTR whenever the medium becomes idle. Aftersending the RTR, ri waits for a SIFS plus a maximumback-off (CWData) (line 5) in order to receive data fromthe intended sender(s). The expiration of this periodforces the receiver to sleep (line 22) since it indicates nodata or a busy medium at the sender end. Conversely,upon receiving any data packet within that time period(line 6), ri waits for a SIFS, and transmits another RTR(line 8 or 12) as a data acknowledgment (i.e., the ACKbit is set to 1).

Moreover, a data request bit is set to 1 in the sameRTR (line 8), when, in addition to receive a further

data packet, Tirw is found greater than g1 plus the time

required to send the last RTR (i.e., the ACK) (lines 7-10). Hence, the reception window allows a receiver toreceive additional, as well as multiple packets, at eachwake-up. However, when Ti

rw goes below a specifiedduration (lines 11-18), the data request bit is set to 0in the last RTR (line 12), and instead of sleeping, areceiver is awakened until its next scheduled Rx or Txrendezvous (either till ti

wlor tdi

wk). This assumption consi-

ders a trade-off for saving the radio initiating energy [2].In contrast, to adjust the clock drift a sender wakes-

up g1 earlier than its Tx rendezvous. The sender listensto the medium for a Tr + 2g1 period to receive a po-tential RTR from the receiver, where Tr is the RTRwindow. However, due to the busy medium at the re-ceiver end, sender(s) might not get an RTR within thatduration and thereafter, waits for another

Ti2n period. If

there is still no RTR packet for the sender, it goes tosleep until the receiver’s next cycle.

Upon receiving an RTR, the sender(s) contends witha random back-off (0 to CWData). A node with an earlyback-off expiration sends a data packet and waits foran ACK from the receiver, while another sender(s) (ifany) pauses the back-off and waits for more data re-quest (in another RTR) from the receiver. The same ora different sender(s) further transmits data in the sameway as if the receiver requests more data, otherwise itgoes to sleep.

As shown in Fig. 5, at w0 wake-up, receiver B sendsthe RTR packets until the reception window allows andreceives multiple data packets from E in response toeach RTR. As soon as the T B

rw expires, B remains activeuntil the w1 wake-up of downstream node A.

In basic nW-MAC, the overall energy cost may in-crease as a result of the additional active period (

Ti2n )

for both the receiver and sender(s). However, to avoidthe sleep delay this energy trade-off is deemed neces-sary. Only in the worst-possible case does a receiveror sender have to lengthen the active period till

Ti2n .

Note that for n = 4, this period is only one eighth ofthe receiver’s cycle (Ti), and potentially happens onlywhen the medium is highly loaded [14].

4.4 Adaptive nW-MAC

The unpredictable nature of monitoring networks leadsto the mixing of high volume of event traffic with themostly durable low periodic data. Existing sensor MACprotocols [1, 3] rarely handle the event traffic phenom-ena. Therefore, adaptive nW-MAC is designed to workwith event data using the same reception window-basedconcept of Section 4.3.

574 Ann. Telecommun. (2011) 66:567–582

4.4.1 Asynchronous staggered schedule

In the proposed asynchronous schedule selection (Sec-tion 4.2), there is the possibility that in every cycle anode might experience a fixed delay prior to forwardingany data to downstream. The consequences of such adelay at every hop may result in a larger end-to-enddelay. As in Fig. 6a, node B has the Tx rendezvousat w1 wake-up of receiver A, whereas, it has the Rxrendezvous for upstream nodes C and D at its w0 andw1 wake-ups, respectively. Thus, packets received by Bat w1, experience a fixed delay (greater than TA

8 ), sinceit can only send the received packets at the w1 wake-upof A. Similar delay occurs for the packets received byC at its w1 and w2 wake-ups, as C can only send at w0

wake-up of B.Moreover, with very low periodic traffic the use

of multiple wake-ups in every cycle appears carelessfrom an energy- efficiency point of view. Therefore, toavoid these problems nW-MAC proposes an adaptivestaggered scheduling, which is able to maintain theoptimum energy consumption and delay for nodes withdifferent hops on a data gathering tree. To accomplishthis during regular MAC operations (using RTRs), areceiver ri announces a single Rx rendezvous at thewkmin of a cycle, where wkmin has a minimum forwardingdelay toward the downstream node di.

In subsequent cycles, the senders adhere to thenew rendezvous schedule with the receiver. Nodes ondifferent hops adaptively build up an almost staggereddata forwarding which we refer as an asynchronousstaggered schedule. In Fig. 6c, node B selects the newRx rendezvous at w0, since w0 has a minimum delaywith the w1 wake-up of A; upstream nodes C and Dfollow the new Tx rendezvous at the w0 wake-up of B.

Similarly, node C chooses the Rx rendezvous at w0 dueto its minimum delay with the w0 wake-up of receiver B.

Adaptation of this approach at the boundary nodesresults in a single wake-up-based asynchronous stag-gered solution for each cycle, as shown in Fig. 6c.Analysis shows that for a chain topology, the asyn-chronous staggered schedule limits the per hop averagedelay within an approximate maximum bound of

Ti2n ; for

n = 4, this delay is only one eighth of the receiver’scycle (Ti).

4.4.2 Multiple wake-up schedule

An increase in the traffic load may occur due to eventdetection, variation in the data generation rate, etc. Asmentioned, nW-MAC has the provisions for n wake-ups for each cycle of a receiver; whenever it is necessaryreceivers utilize these provisions. Note that the trafficload detection [14, 15] is out of the scope of this paper,rather, the nW-MAC protocol predicts and performsadaptively to deal with varying loads.

While operating in the single wake-up-based stag-gered schedule, a receiver ri decides that another wake-upin the same cycle is necessary, based on the deter-mination of full utilization of the reception window.More specifically, when the Ti

rw of a receiver ri expires,due to the downstream node di’s wake-up offset tdi

wk, ri

presumes that senders still have data to send. However,due to the bounded reception window the receivercannot receive the additional data.

A receiver adapts to the prediction of more data,and interjects an additional wake-up in the same cycle.To accomplish this, in addition to the wkmin wake-up,ri announces (using RTR) another Rx rendezvous atthe next minimally delayed (i.e., in terms of forwarding

A

B

C

D

E

w1

w0 w1

w1 w2

F

w1

w0

w1

Delay > ( )

Delay > ( )8AT∧

8BT∧

(a)

A

C

B

D

FE

(b)

A

B

C

D

E

w1

w0

w0

F

w1

w0

w0

Operational Cycle ( )AT∧

Operational Cycle ( ) BT

∧

Operational Cycle ( )CT∧

(c)

Fig. 6 a Rx and Tx Rendezvous schedules for the nodes of the tree in Fig. 6b, b A data gathering tree, and c Single wake-up-basedasynchronous staggered schedule for the nodes of the tree in Fig. 6b

Ann. Telecommun. (2011) 66:567–582 575

toward di) wake-up of the same cycle. Senders withmore data are informed (snoop the RTR) of the ad-ditional wake-up; hence, they immediately go to sleepuntil the receiver’s new wake-up. Thus, by an on-demand basis, a receiver additively utilizes all of its nwake-ups when necessary, due to the full utilization ofthe reception window at any newly added wake-up. Incontrast, under utilization of the reception window atany new wake-up, drives a receiver to deduct it fromthe wake-up list.

Therefore, a per cycle multiple wake-up strategy de-creases contentions and collisions at the subsequent Rxrendezvous of a receiver. With high traffic, this strategyis able to ensure a higher throughput with minimum de-lay, which are the critical goals of nW-MAC. Althoughmultiple wake-ups of a receiver require additional radioinitiating energy [2], this additional energy would beworthwhile in order to minimize delay and maximizethroughput.

5 Analysis of nW-MAC

To verify the performance of nW-MAC, several ob-servations on the energy consumption and delay opti-mization are analyzed in this section. Throughout theanalysis, the notations contained in Table 1 are used,

Table 1 Notation used for analysis

Notation Meaning

Etx Energy to send a data packetErx Energy to receive multiple data packetsPtx Power to transmita

Prx Power to receivea

Pidle Power in idle mode a

Psleep Power in sleep mode a

Pwake−up Power to wake-upa

Ti Cycle duration of a receiver ri

Tr RTR window periodTi

rw Maximum reception window of ri

Tiidle Idle period at receiver ri

Tijidle Idle period at sender uij

TData Time to transmit a data packetTRTR Time to transmit a RTR packetCWData Contention window for dataCWRTR Contention window for RTRslotTime Time duration of a single sloth Number of hopsn Number of wake-up in a cycle Ti

Dmax Maximum end-to-end delayDmin Minimum end-to-end delayDavg Average end-to-end delay

aPower levels of CC2420 radio

where the power levels of CC2420 radio are consideredin the energy measurements.

5.1 Energy consumption

The performance of nW-MAC depends on the energyconsumption for a number of RTRs, data and ACKsending and receiving, the number of wake-ups percycle, the idle period in medium listening, etc. The ex-pected energy to transmit a data packet by an upstreamsender uij of receiver ri, is calculated as

Etx = (m × TRTR) × Prx + TData × Ptx + TACK × Prx

+ Tijidle × Pidle, (6)

We assume that prior to sending a packet, uij has tolisten for an expected duration of Tr + Ti

2n and receive mnumber of RTRs from ri. The later assumption is truewhen uij fails to transmit the data packet in responseto a maximum (m − 1) number of RTRs; here, (m −1) other transmissions are considered prior to uij’s datatransmission. Thus, the idle listening time (Tij

idle) at uij

solely depends on m, such that

Tijidle = Tr + Ti

2n

2+ (SIFS × m)

+(

SIFS + slotTime × CWData

2+ TData

)

× (m − 1), (7)

With a very low traffic, the value of m is expected tobe small since the possibility of simultaneous senders ata given instant of time is minimal, and therefore thereexists fewer contention and collision possibilities. In asingle wake-up-based asynchronous staggered sched-ule, a sender listens to the medium for a maximumaverage bound of

Ti2n . When n = 4, this becomes ap-

proximately one eighth of the receiver’s cycle Ti. It isworth mentioning that for a sender and a receiver inRI-MAC [1] and X-MAC [3], respectively, the averagemedium listening time is, analytically, one half of thecycle length; this is greater than that using nW-MAC.

Conversely, a receiver can receive multiple pack-ets within the reception window. Thus, m data pack-ets received at each Rx rendezvous require at least(m + 1) RTR transmissions, including the ACK for thelast received packet. Hence, the energy for m packetsreceived is

Erx = ((m + 1) × TRTR) × Ptx + (m × TData) × Prx

+ Tiidle × Pidle, (8)

576 Ann. Telecommun. (2011) 66:567–582

Tiidle includes the average waiting time for a receiver

prior to sending the first RTR. Additionally, an averageof data contention back-off is considered after eachof the m RTR packet transmissions. A maximum databack-off is also added to Ti

idle when ri does not receiveany data after sending the last RTR (with an additionaldata request).

Tiidle =

(

Tirw

2or

Ti2n

2

)

+(

SIFS + slotTime × CWData

2

)

× m + (SIFS + slotTime × CWData), (9)

The observations based on the energy analysisdemonstrate that nW-MAC is able to maintain optimalenergy consumption; in low traffic a receiver primar-ily wakes-up once per Ti interval. Moreover, the en-ergy cost is minimal since receivers avoid unnecessarymedium listening [4], with the exception of the idlewaiting time for data reception.

5.2 Delay optimality

To quantify the delay of nW-MAC, we analyze theminimum and maximum end-to-end delay distributionof a chain topology. In asynchronous staggered sched-ule, receiver ri utilizes only one wake-up per cyclewhich is at the minimum delayed wake-up (wkmin ). Insuch a schedule, the per hop delay is bounded by themaximum possible duration of the reception windowTi

rw. According to the start-of-cycle (SoC) selection inSection 4.2, the wkmin wake-up of a receiver has, at most,a

Ti2n long reception window. Additionally, the Ti

rw of ri

cannot exceed the downstream node di’s wake-up offset(tdi

wk). Therefore, the maximum end-to-end delay from a

source node toward the sink is derived as

Dmax = Ti + (h − 1) ×(

Ti

2n− Ti

rw

2

)

, (10)

The assumption is made here that after generating apacket a source must wait, at most, for the duration ofTi before sending it to the receiver. For the remainderof the (h − 1) hops the maximum delay should be ( Ti

2n −Ti

rw2 ).

Subsequently, if any source obtains an immediaterendezvous with the receiver after receiving the packetfrom the upper layer, the minimum delay would be

Dmin = h ×(

Ti

2n− Ti

rw

2

)

, (11)

The delay approximation for adaptive multiplewake-up schedules is complex. Considering that to

handle heavy traffic a receiver might need to wake-up a maximum of n possible times at each cycle (i.e.,the worst-possible case) and accordingly announce itswake-up schedules or Rx rendezvous (using RTR).Being aware of the schedules, the upstream sendersselect all of the n wake-ups of the receiver as their Txrendezvous. Eventually, all of the senders have an equalprobability for sending data at each of the n wake-upsof the receiver. Hence, the per hop delay is equal to thecycle average which is Ti

2 and the average end-to-enddelay is

Davg =Ti

2+ h − 1

2× (

Ti − Tirw

)

, (12)

The delay analysis of nW-MAC demonstrates thatin an asynchronous staggered schedule the per hopaverage delay is bounded by Ti

2n . Thus, for a chain themaximum per hop delay is approximately one eighthof the receiver’s cycle time (Ti) (n = 4 for this paper’sexample). This causes a significant delay reduction ascompared with the existing asynchronous RI-MAC [1]and X-MAC [3]; the per hop average delay for thoseprotocols is one half of the cycle. Moreover, for amultiple wake-up schedule the average per hop delayof nW-MAC is equivalent to

Ti2 , which contributes to

minimizing the delay even under heavy traffic condi-tions.

6 Performance evaluations

6.1 Simulation setup

The simulations of nW-MAC are extensively per-formed in NS-2. The nW-MAC behavior for twodifferent network conditions is studied: (1) single-flowperformance in a chain topology, and (2) multipleflow performance in a tree topology. Routing trafficis avoided and a static path from each sensor node tothe sink is used. The link bandwidth is set as 250 Kbpsand all of the nodes in the network have a transmissionand carrier sensing range of 25 and 55 m, respectively.Source nodes generate data either periodically or basedon the detection of an event, or both.

Table 2 provides the details concerning the nW-MAC simulation parameters; the majority of them havebeen extracted from the NS-2 implementation of RI-MAC [1]. Furthermore, we use the power levels ofCC2420 radio to measure the energy consumption. Togather knowledge on nW-MAC performance, asyn-chronous sender initiated X-MAC [3] and receiver

Ann. Telecommun. (2011) 66:567–582 577

Table 2 Simulation Parameters

Parameter Value

Channel bandwidth 250 KbpsTransmission range 25 mCarrier sensing range 55 mData packet size 32 bytesRTR packet size 10 bytesSIFS 192 μsRTR window (Tr) 8 msContention window (CWData) 16Contention window (CWRTR) 8Slot time 320 μsRetry limit 5Simulation time 100 s

initiated RI-MAC [1] protocols are implemented in thesame simulation environment and compared with nW-MAC. We use a 6 byte long strobe preamble for X-MACwhich embeds the receiver’s ID; X-MAC receiver re-mains active for 20 ms at each wake-up [3] and doesa clear channel assessment (CCA) check for possiblestrobe preamble(s) from the sender. The CCA checkcontinues for a time period longer than the intervalbetween two consecutive preamble transmissions; amedium idle status at the expiration of the CCA checkforces the receiver to sleep.

The cycle duration (Ti) of the receivers varies from0.5 to 2.0 s. A single wake-up followed by a sleepperiod, in each cycle, is considered for the receivers inX-MAC and RI-MAC. However, for nW-MAC, pro-visions for multiple (n = 4) wake-ups are assumed foreach receiver in every

Tin interval. In order to provide

more stable results, an average of ten simulation exe-cutions is performed, each for 100 s, under the samenetwork conditions.

For the simulation results, the following metrics areused to realize the performance of the proposed nW-MAC:

– Energy Consumption: the average energy consump-tion, in Joules, of all the nodes of the network.

– Average Duty Cycle: the ratio between per cycle av-erage active time to the entire cycle time, expressedin percentage.

– End-to-End Delay: the time interval between apacket generated at the source and received at thesink.

– Per Hop Delay: it is calculated by dividing the end-to-end delay with the total number of hops.

– Delivery Ratio: the ratio between the total numbersof packets received by the sink to the number ofpackets generated by the source nodes.

– Throughput: it is calculated as the amount of re-ceived data (per second) by the sink over the simu-lation time.

– Signaling Overhead (per packet): it is measured inbytes by dividing the total amount of bytes trans-mitted as control packets to the total number ofsuccessfully received packets at the sink node.

The error bars in the presented graphs show thevariations of the obtained delay, and demonstrate theminimum and maximum values among the simulationruns.

6.2 Chain topology and single-flow performance

To attest to the concept of nW-MAC, it is very impor-tant to analyze the multi-hop performance in a chaintopology. In this work’s implementation, eight-nodechain (seven hops) with one source and a sink at eachend are used; the distance between two adjacent nodesis set to 20m.

Figure 7 shows the performance of the protocols forboth energy and duty cycle. As we observe in Fig. 7a,the average energy consumption (measured at a rateof 2 pkts/s with a 1 second cycle) of X-MAC is higherthan the other protocols, due to the preamble trans-mission causing more energy waste. The RI-MAC alsoconsumes relatively more energy mainly for the hugeidle listening at the sender end. However, the asynchro-nously scheduled rendezvous mechanism ensures lessenergy consumption for nW-MAC, as it reduces theidle listening at both the sender and receiver ends.

The graph of Fig. 7b presents the energy performanceunder different source reporting rates with a one secondcycle. Although the energy consumption minimally in-creases at lower rates, as the rate increases the energyconsumption becomes high for each of the protocols.At a high traffic rate, X-MAC has the highest energyconsumption due to the frequent sending and receivingof strobe preambles. RI-MAC also has a high energyconsumption due to the collision effect and unnecessarymedium listening. At a high traffic rate, nW-MACclearly outperforms the other protocols because of thescheduled wake-up and reception window strategy.

As shown in Fig. 7c, the duty cycle rises in proportionto the data rate. The X-MAC and RI-MAC result in amaximum of 52% and 34% average duty cycle, respec-tively. These are both much larger than the nW-MAC’s20% average duty cycle. The active time frequentlyincreases for X-MAC, when multiple senders contendfor medium access and only one of them is able to sendthe preamble or data. Moreover, in RI-MAC, receiversare mostly found awaken at high rates, though due

578 Ann. Telecommun. (2011) 66:567–582

1 2 3 4 5 6 70

1

2

3

4x 10

Ene

rgy

Con

sum

ptio

n (jo

ule)

Number of Hops

X-MAC RI-MAC nW-MAC

(a)

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0

1

2

3

4

5

x 10

Ene

rgy

Con

sum

ptio

n (jo

ule)

Source Data Rate (pkts/s)

X-MAC RI-MAC nW-MAC

(b)

0 1 2 3 40

10

20

30

40

50

60

Ave

rage

Dut

y C

ycle

(%

)


X-MAC RI-MAC nW-MAC

(c)

Fig. 7 Performance on energy and duty cycle for chain topology: a average energy consumption for different number of hops, b averageenergy consumption, and c average duty cycle for varying source data rate

to collisions their active time seems to be ineffectivefrom the energy point of view. In contrast, nW-MACallows a receiver to receive till the reception windowthat eventually bounds the active period of a node.

In Fig. 8, the end-to-end and per hop delay areevaluated. As shown in Fig. 8a, for all the protocols(with 2 pkts/s and one second cycle) the end-to-enddelay increases as the number of hops in the chainincreases. However, the delay of X-MAC and RI-MACare severely affected by the sleep periods of the nodes.In nW-MAC, the delay is less than the other two proto-cols since all of the nodes are able to avoid sleep delayand adaptively use the provisions of multiple wake-upschedules for packet reception and sending.

Figure 8b demonstrates that in X-MAC, the overalldelay increases due to the receiver’s sleep schedule andmedium occupancy time for the preamble transmission.Whereas, the sleep delay and increased retransmission

due to uncoordinated multiple packet receipts mostlyaffect the RI-MAC’s delay performance. Conversely,in nW-MAC, nodes receive multiple packets until thereception window limit and deliver data using the speci-fied staggered or multiple wake-up schedule. The coor-dinated and bounded medium access mainly results inreduced end-to-end delay for nW-MAC.

The per hop average delay of the protocols areplotted in Fig. 8c; cycle durations 0.5 to 2 seconds areused with a low data rate of 0.5 pkts/s. The delay ofnW-MAC is nearly identical to one-eighth ( 1

8 ) of thecycle, which is possible only for asynchronous staggeredscheduling. In contrast, due to the randomness in thereceiver’s wake-up, the per hop average delay of RI-MAC and X-MAC are both approximately equal to onehalf of the cycle. The effectiveness of nW-MAC for amonitoring application is then verified with the delayperformances demonstrated by the simulations.

1 2 3 4 5 6 70

2

4

6

8

End

-to-

End

Del

ay (

seco

nd)

Number of Hops

X-MAC RI-MAC nW-MAC

(a)

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.00

2

4

6

8

End

-to-

End

Del

ay (

seco

nd)


X-MAC RI-MAC nW-MAC

(b)

0.5 1.0 1.5 2.00.0

0.5

1.0

1.5

2.0

Per

Hop

Del

ay (

seco

nd)

Operational Cycle (second)

X-MAC RI-MAC nW-MAC

(c)

Fig. 8 Performance on delay for chain topology: a End-to-end delay for different number of hops, b end-to-end delay for varyingsource data rate, and c per hop average delay with respect to different operational cycle durations

Ann. Telecommun. (2011) 66:567–582 579

6.3 Tree topology and multiple flow performance

A network of a 100 × 100 m area is used for the treetopology, where 100 nodes are deployed in a uniformrandom distribution. On average, most of the sensorsare found within four to seven hops away from the sink.

In Fig. 9, the performances of basic nW-MAC andadaptive nW-MAC are compared with X-MAC andRI-MAC. To analyze the energy consumption, a lowperiodic traffic of 0.5 pkts/s is considered, along witha diverse number of sources. Figure 9a representsthe general incremental trend of energy usage withan increased number of sources. When the numberof sources is larger, X-MAC experiences significantlyhigher energy consumption than the other protocols.Note that for fewer sources, the basic nW-MAC con-sumes more energy, as compared with X-MAC, due toits initial scheduling having multiple wake-ups. How-ever, as the number of sources increases, the am-plification of the preamble transmission mostly affectsX-MAC. Among the protocols, adaptive nW-MACappears to be the most energy efficient since it en-sures better energy utilization with minimum waste. RI-MAC consumes less energy for fewer sources; whenthe number of sources increases it consumes and wastesmore energy due to a large number of collisions.

Adaptive nW-MAC achieves a lower duty cycle thanX-MAC and RI-MAC, as shown in Fig. 9b. However, incase of fewer sources, basic nW-MAC has the highestduty cycle due to its fixed multiple wake-ups in eachcycle. In X-MAC, prior to a data transmission a senderhas to send the preamble while other senders in thesame neighborhood wait in active mode; hence, theaverage duty cycle becomes longer. As the number ofsources increases node participation in data forwardingalso increases, resulting in a maximum 60% duty cycle

for X-MAC. Conversely, in RI-MAC senders have tolisten to the medium for a long period before sendingthe data; hence, the duty cycle is typically large. How-ever, adaptive nW-MAC has the smallest duty cycle,30% for 60 source nodes. The load-dependent schedul-ing and reception window-based MAC helps adaptivenW-MAC to achieve a superior duty cycle even withmultiple wake-ups.

According to the results of Fig. 9c, the delay of nW-MAC surpasses each of the other protocols. Actually,the use of the multiple wake-ups as well as the multiplepacket reception at each wake-up, results in steadydelay behavior for nW-MAC, even with more sources.In contrast, the per cycle single wake-up strategy andsleep delay degrades the delay performances of bothRI-MAC and X-MAC.

In Fig. 10, the tree topology performance of the pro-tocols are compared for different source rates, whereten sources are randomly chosen. As shown in Fig. 10a,X-MAC responds with a greater energy consumptionsince the senders always attempt to be in rendezvouswith their corresponding receivers. The basic nW-MACalso has a high energy consumption due to its unneces-sary use of multiple wake-ups lacking traffic adaptation.Although RI-MAC maintains low energy consumptionat low a rate, as the rate increases the energy consump-tion also increases for idle listening and collision relatedretransmission. In contrast, avoidance of such wasteassures less energy consumption for nW-MAC even ata higher traffic rate. In simulations, it is observed thatthe number of contenders (or senders) decreases withan increase in the number of wake-ups in the same nodecycle, resulting in fewer collisions in nW-MAC.

As demonstrated in Fig. 10b, during low rates, adap-tive nW-MAC has a marginal duty cycle differencefrom both X-MAC and RI-MAC. However, as the rate

10 20 30 40 50 600

1

2

3

4

Ene

rgy

Con

sum

ptio

n (jo

ule)

Number of Sources

X-MAC RI-MAC Basic nW-MAC Adaptive nW-MAC

x 103

(a)

10 20 30 40 50 600

10

20

30

40

50

60

70

Ave

rage

Dut

y C

ycle

(%

)

Number of Sources


(b)

10 20 30 40 50 60

2

4

6

8

10

12

End

-to-

End

Del

ay (

seco

nd)

Number of Sources

X-MAC RI-MAC nW-MAC

(c)

Fig. 9 Performance comparison for tree topology with different number of sources: a energy consumption, b average duty cycle, and cend-to-end delay

580 Ann. Telecommun. (2011) 66:567–582

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.01

2

3

4

5

Ene

rgy

Con

sum

ptio

n (jo

ule)



x 102

(a)

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.00

20

40

60

80

100

Ave

rage

Dut

y C

ycle

(%

)



(b)

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.00

4

8

12

16

20

End

-to-

End

Del

ay (

seco

nd)


X-MAC RI-MAC nW-MAC

(c)

Fig. 10 Performance comparison for tree topology with different source traffic rates: a energy consumption, b average duty cycle, andc end-to-end delay

increases nW-MAC performs better due to its adapt-ability and efficiency in using n wake-up schedules.It is worthy to observe that at high traffic rates theduty cycle of basic nW-MAC is more concise than bothX-MAC and RI-MAC, since the nodes in the basicprotocol maintain a fixed wake-up(s) in each cycle.

The delay effectiveness of nW-MAC for differentdata generation rates is shown in Fig. 10c. As expected,the delay of nW-MAC is lower for the single wake-upstrategy at a low traffic rate and the multiple wake-up strategy at a high traffic rate. In contrast, RI-MACexperiences a relatively high delay for typical sleepdelay and collision related retransmissions at low andhigh traffic rates, respectively.

Figure 11 shows the delivery ratio of the protocols.Although at lower traffic rates the delivery ratio of RI-MAC and X-MAC demonstrate better performance, athigh traffic rates their delivery ratio sharply decreases

0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.00.4

0.6

0.8

1.0

Del

iver

y R

atio

(no

rmal

ized

)


X-MAC RI-MAC nW-MAC

Fig. 11 Normalized delivery ratio for varying source rate

due to the increasing number of collision losses. Forboth low and high traffic, nW-MAC achieves a betterdelivery ratio; a maximum of approximately 95% ofthe generated packets is received by the sink duringsimulation. Thus, the results reflect the efficient use ofmultiple wake-ups in handling traffic variations.

Figure 12 compares the signaling overhead of theprotocols as a function of the source rate, where it isobserved that the signaling overhead of nW-MAC isless than the other protocols. In nW-MAC, the amountof contentions and collisions are always maintained ata sustainable level due to the adaptive and additiveuse of multiple wake-up provisions by a receiver ateach cycle. Hence, it is possible to restrain the perpacket signaling overhead (use of RTR and ACK) atan optimal level; which guarantees better throughputas well (shown in Section 6.4). In contrast, the lack oftraffic adaptability incurs more collisions in RI-MAC,

Fig. 12 Per packet signaling overhead

Ann. Telecommun. (2011) 66:567–582 581

resulting in lesser throughput, especially at high trafficrates. Intuitively the contentions as well as per packetsignaling overhead (use of beacon and ACK) are alsoincreased for retransmissions. The overhead of X-MACproduces the poorest results, having more variations inthe graph. This is because for a single data transmissionthe sender uses a variable series of strobe preamblesuntil it receives an ACK from the receiver.

6.4 Multi-flow and multi-event performance

In order to investigate the applicability of nW-MAC foran event monitoring application, we further considera simulation environment with regular traffic (peri-odic observation) at a very low rate of 0.2 pkts/s andevent traffic (event detection) at a rate of 4 pkts/s. Allof the nodes in the same tree topology (Section 6.2)periodically generate the regular traffic. In the net-work, ten sources in two randomly selected locationsare considered for event data generation. The eventtraffic locations, triggering time, and maximum dis-tance of the event sources from the sink are given inTable 3.

In Fig. 13, the throughput observations of both nW-MAC and RI-MAC are plotted over the simulationtime, and the bar line shows the average throughputachieved at the sink. It can be seen that nW-MACmaintains a better throughput for both traffic environ-ments. At low traffic, the use of asynchronous stag-gered schedule ensures an on time delivery of 88%of the packets every second. Whereas, for RI-MACit is approximately 65% of the total generated pack-ets. Furthermore, for event traffic, multiple wake-upschedules report 68% of the packets every second fornW-MAC, and RI-MAC ensures 45% reception of thepackets.

We also observe the average end-to-end delay ofthe received packets during the event occurrence time.Figure 14 shows a sustainable average delay of 1.22 sfor the packets of event 1, and 1.51 s for the packets ofevent 2. We believe that the achieved throughput anddelay for nW-MAC are well suited for a monitoringnetwork in order to fulfill the target of timely andeffective actions.

Table 3 Event traffic settings

Event Location Time Distance(id) (x, y) (s) (hop)

Event 1 (60, 80) 20th–35th 3Event 2 (70, 15) 60th–70th 4

0 20 40 60 80 1000

2

4

6

8

10

12x 103

Thr

ough

put (

bps)

Simulation Time (second)

(a)

0 20 40 60 80 1000

2

4

6

8

10

12x 103

Thr

ough

put (

bps)


(b)

Fig. 13 Throughput performance: a nW-MAC and b RI-MAC

20 25 30 60 65 700

1

2

3

Ave

rage

Del

ay (

seco

nd)


Event 1 Event 2

Fig. 14 Average delay of received event traffic

582 Ann. Telecommun. (2011) 66:567–582

7 Conclusions

In this paper, we have proposed an innovative asyn-chronously scheduled duty cycle MAC protocol forWSN having n number of multiple wake-up provisions.Unlike previous MAC protocols, nW-MAC considersboth energy consumption and delay, in parallel, andachieves a better energy-delay trade-off with multiple(n = 4) wake-ups per cycle for the receivers. Further-more, reception window-based adaptive nW-MAC isable to cope with dynamic traffic environments, i.e.,WSNs having monitoring applications. The analysis andsimulation results demonstrate that nW-MAC has alower energy consumption and better delay optimiza-tion, as well as throughput, than other currently existingprotocols.

It is very challenging to incorporate broadcast trafficin any asynchronous duty cycle MAC, since each nodemaintains periodic wake-up and sleep schedules atrandom. Furthermore, the hidden terminal problem isa prospective research issue to be handled in sensorMAC protocols. Like existing asynchronous schemes,the nW-MAC protocol also carries these limitations,and we leave them for future research.

Acknowledgements “This research was supported by the MKE,Korea, under the ITRC support program supervised by theNIPA” (NIPA-2010-(C1090-1021-0003)). Dr. CS Hong is thecorresponding author.

References

1. Sun Y, Gurewitz O, Johnson DB (2008) Ri-mac: a receiver-initiated asynchronous duty cycle mac protocol for dy-namic traffic loads in wireless sensor networks. In: SenSys’08: proceedings of the 6th ACM conference on Embed-ded network sensor systems. New York, NY, USA, ACM,pp 1–14

2. Polastre J, Hill J, Culler D (2004) Versatile low power mediaaccess for wireless sensor networks. In: SenSys ’04: proceedingsof the 2nd international conference on Embedded networkedsensor systems. Baltimore, MD, USA, ACM, pp 95–107

3. Buettner M, Yee GV, Anderson E, Han R (2006) X-mac:a short preamble mac protocol for duty-cycled wireless sen-sor networks. In: SenSys ’06: proceedings of the 4th interna-tional conference on Embedded networked sensor systems.Boulder, Colorado, USA, ACM, pp 307–320

4. Ye W, Heidemann J, Estrin D (2004) Medium access controlwith coordinated adaptive sleeping for wireless sensor net-works. IEEE/ACM Trans Netw 12(3):493–506

5. Lu G, Krishnamachari B, Raghavendra CS (2007) An adap-tive energy-efficient and low-latency mac for tree-based datagathering in sensor networks: Research articles. Wirel Com-mun Mob Comput 7(7):863–875

6. van Dam T, Langendoen K (2003) An adaptive energy-efficient mac protocol for wireless sensor networks. In: Sen-Sys ’03: proceedings of the 1st international conference onEmbedded networked sensor systems. New York, NY, USA,ACM, pp 171–180

7. El-Hoiydi A, Decotignie J-D (2004) Wisemac: an ultra lowpower mac protocol for the downlink of infrastructure wire-less sensor networks. In: ISCC ’04: proceedings of the ninthinternational symposium on computers and communications2004, vol 2 (ISCC”04). Washington, DC, USA, IEEE Com-puter Society, pp 244–251

8. Du S, Saha A, Johnson D (2007) Rmac: a routing-enhancedduty-cycle mac protocol for wireless sensor networks. In:INFOCOM 2007. 26th IEEE international conference oncomputer communications. Washington, DC, USA, IEEEComputer Society, pp 1478–1486

9. Sun Y, Du S, Gurewitz O, Johnson DB (2008) Dw-mac: alow latency, energy efficient demand-wakeup mac protocolfor wireless sensor networks. In: MobiHoc ’08: proceedingsof the 9th ACM international symposium on Mobile ad hocnetworking and computing. New York, NY, USA, ACM,pp 53–62

10. Ye W, Silva F, Heidemann J (2006) Ultra-low duty cycle macwith scheduled channel polling. In: SenSys ’06: Proceedingsof the 4th international conference on Embedded networkedsensor systems. New York, NY, USA, ACM, pp 321–334

11. Ahn G-S, Hong SG, Miluzzo E, Campbell AT, Cuomo F(2006) Funneling-mac: a localized, sink-oriented mac forboosting fidelity in sensor networks. In: SenSys ’06: pro-ceedings of the 4th international conference on Embeddednetworked sensor systems. New York, NY, USA, ACM,pp 293–306

12. Tseng Y-C, Hsu C-S, Hsieh T-Y (2003) Power-saving pro-tocols for IEEE 802.11-based multi-hop ad hoc networks.Comput Netw 43(3):317–337

13. Thakur S, Nandi S, Bhattacharjee R, Goswami D (2007)An asynchronous wakeup power-saving protocol for multi-hop ad hoc networks. Int J High Perform Comput Appl21(4):429–442

14. Wan C-Y, Eisenman SB, Campbell AT (2003) Coda: con-gestion detection and avoidance in sensor networks. In: Sen-Sys ’03: proceedings of the 1st international conference onEmbedded networked sensor systems. New York, NY, USA,ACM, pp 266–279

15. Rangwala S, Gummadi R, Govindan R, Psounis K (2006)Interference-aware fair rate control in wireless sensor net-works. SIGCOMM Comput Commun Rev 36(4):63–74

nW-MAC: multiple wake-up provisioning in asynchronously...

Documents

Transcript of nW-MAC: multiple wake-up provisioning in asynchronously...