WoR-MAC: Combining Wake-on-Radio withQuality-of-Service for Intelligent Environments

Dawud Gordon, Matthias Berning, Rayan El Masri, Michael BeiglKarlsruhe Institute of Technology

Karlsruhe, GermanyEmail: [Firstname.Lastname]@kit.edu

Johannes Blanckenstein, Jirka KlaueEADS Innovation Works

Munich, GermanyEmail: [Firstname.Lastname]@eads.net

Abstract—Intelligent environments (IE) leverage embeddedprocessing and wireless communication to assist users in a varietyof ways. Applications rely on low power consumption for longerlifetimes, though different applications require different Quality-of-Service (QoS) requirements from the MAC layer. Until now,low power has come at the cost of other QoS parameters such aslatency or packet loss. This paper presents WoR-MAC, a wirelessMAC protocol which allows pre-existing protocols to be combinedwith remote multi-node wake-ups. The protocols are embeddedinto Wake-on-Radio (WoR) frames, allowing nodes to sleep dur-ing periods of low activity and be woken asynchronously with asingle short RF signal. After waking, nodes begin communicationusing the embedded MAC protocol. Once the nodes have beenwoken, they maintain the QoS of the original MAC, with greatlyreduced power consumption. The results indicate that WoR-MAC maintains packet loss characteristics of CSMA-CA andTDMA, as well as latency after accounting for the duty cycleand collaborative parameter estimation, while reducing powerconsumption by up to 49%, very close to the lower bound givenby the duty-cycle.

I. INTRODUCTION

In the field of pervasive computing, intelligent environments(IE) refer to living or working spaces which have been instru-mented with pervasive technology. These IE’s work to assistusers in accomplishing their goals by leveraging embeddedprocessing and communication between smart objects. Main-taining user’s freedom of motion calls for wireless communi-cation and mobile power sources, making energy conservationto prolong system lifetime of the utmost importance.

Low-power characteristics of processing units, memory andcommunication have been constantly improving, but the bat-teries which power them are not subject to Moore’s Law andhave therefore remained the limiting factor. Wireless com-munication is one of the most power consuming operationswhich such systems typically perform [1], making low-powerwireless communication an important topic.

Recent advances in wireless communication have intro-duced the concept of remotely waking dozing nodes usingspecific signals on the wireless channel. This process iscalled Wake-on-Radio (WoR), and allows nodes to conserveenergy when they are not actively required for communication.This technology began with simple amplitude sampling ofsignal magnitude levels to periodically monitor the channelfor wake-up preambles [6]. Later this concept was expandedallowing single node wake-ups to be addressed using preamble

sampling [7], where nodes periodically listen on the channel,and are therefore aware not only of a wake-up signal, but alsoof the address contained within it. Further reductions in powerconsumption were achievable by using preamble strobing andacknowledged wake-ups with the introduction of X-MAC [4].

X-MAC is extremely power-efficient, and allows a nodeto select a sleeping node, wake it with strobed, addressed,preambles, transmit data to that node once it is awake andthen return to sleep. For many real scenarios [16] withn-to-n communication patterns, bidirectional communicationpatterns, cluster-based star topologies and Quality-of-Servicerequirements, X-MAC presents the following problems

• X-MAC only supports waking one node at a time, there-fore multicast is not possible.

• X-MAC only supports one-way communications perwake-up event, e.g. no command-response.

• X-MAC has very rigid QoS characteristics which cannotbe adapted to the scenario.

The result is that for scenarios which require multicast ortwo-way communication, multiple wake-ups and duty-cyclesare required when using X-MAC. In IE’s, many differentapplications for different scenarios have been presented [9],[3]. Each of these applications has different QoS requirementson the MAC layer of the communication stack. Reducingpower consumption using WoR protocols greatly increasesapplication lifetime, but comes at the cost of sacrificing QoSof other MAC protocols, which is unacceptable for manyIE applications. This can have disastrous effects on QoSparameters such as effective bitrate and latency, essentiallyforcing the application designer to select exclusively betweenlow-power and QoS.

In this work, WoR-MAC, a MAC protocol for groups ornetworks of wireless devices is presented. The protocol canbe combined with other MAC protocols in order to combinethe low-power attributes of WoR with the QoS parameters ofthe original protocol. The technical novelties of the protocolare as follows:

• WoR-MAC introduces group or cluster-addressed wake-ups allowing nodes to wake multiple neighbors at once.

• WoR-MAC allows woken nodes to collaboratively esti-mate the length of a communication period.

• WoR-MAC allows communication via unicast, multicast

or bi-directionally (send/receive), using an embeddedMAC protocol.

The separate embedded MAC protocol is activated for com-munication for a period of time which is calculated collabora-tively. During this period, the nodes can exchange whicheverdata they wish, before all nodes return to sleep mode. Thepremise here, is that for sporadic communication, WoR-MACcombines the properties of X-MAC and WoR with the QoSparameters of the embedded MAC (e.g. CSMA-CA). Theresulting behavior conserves energy as X-MAC does duringidle periods, but at the same time provides the QoS of theembedded MAC during communication.

WoR-MAC is evaluated in an IE scenario, which simu-lates users entering and exiting an intelligent environment.The simulation compares two standard protocols, TDMA andCSMA-CA, with the same protocols embedded in WoR-MAC,referred to as WoR-TDMA and WoR-CSMA respectively. Theprotocols are evaluated in terms of energy consumption, andQoS parameters latency and packet loss. The standard WoRprotocol X-MAC [4] was also modeled for comparison asthe lower boundary for power consumption and to evaluateits QoS characteristics. The results indicate that both WoR-TDMA and WoR-CSMA achieve power consumption valuesclose to, or better than that of X-MAC, while maintainingthe QoS characteristics of the original TDMA and CSMA-CAprotocols respectively.

The paper is structured as follows, the WoR-MAC protocolis introduced in detail in Sec. II, followed by description of thescenario in Sec. III. The OPNET simulation environment willbe detailed in Sec. IV, followed by the results of the simulationin Sec. V and a discussion of the implications of those resultsin Sec. VI. Related protocols are examined in Sec. VII, andSec. VIII concludes the paper followed a brief summary offuture work in Sec. IX and the acknowledgments.

II. WOR-MAC PROTOCOL

This section will describe the behavior of the WoR-MACprotocol in detail. The exact parameterization used for thesimulation results will be detailed in Tab. II in Sec. IV.

A. The X-MAC Protocol

X-MAC will be briefly explained as WoR-MAC is anextension of many of the concepts introduced in X-MAC [4].Nodes are duty-cycled in order to conserve energy in periodsof inactivity. A node which wishes to communicate with adifferent node begins communicating strobed preambles, anal-ogous to duty-cycling, containing the address of the intendedreceiver. When a node’s duty-cycle expires, that node wakesup and listens to the channel. If no preamble is received bythat node, or if a preamble addressed to a different node isreceived, the listening node returns to sleep for the rest of itsduty-cycle period. If it hears a preamble containing its ownaddress, that node ACKs the preamble and switches to receivemode to receive the message intended for it. This behavior canbe seen in the upper portion of Fig. 1.

Dawud Gordon Papers und Fragestellungen 2

Wake to Contention

SENDTx

RECVRx

PreambleACK

X-MAC

Tx

Rx

WoR-MAC

...

sleep sleep

Begin Communication

Alert

sleep

sleep sleep sleep sleep

...…

Communication

Communication

Fig. 1. MAC Protocol Integration into the WoR-MAC Frame

B. Wake-to-Contention

In WSN applications, communication may also be directedto other nodes in the network, and not just a single target node.For this reason, WoR-MAC implements a Wake-to-Contention(WTC) paradigm which allows nodes to contend for themedium using arbitration based on a different, embedded,MAC protocol which is wrapped in WoR strobed preamble-sampling frames. The point of this is to allow the selection ofa MAC protocol for the communication period with optimalQoS requirements for the scenario. In this sense WoR-MACis actually a meta-MAC protocol, although for the purpose ofthis paper and application, TDMA and CSMA-CA are usedfor medium arbitration. For comparison, the upper portion ofFig 1 shows the behavior as indicated by X-MAC, where thelower portion shows the WTC specification.

C. Cluster-Wide Wake-Ups

In X-MAC and other WoR protocols, a node responds towake-up preambles from a remote sender by waking up andpossibly sending an ACK, then proceeding to carry out asingle transmission or reception operation before returning tosleep. WoR-MAC builds on this concept, where the wake-upACK is then re-ACKed with a “begin communication” packetwhich specifies the parameters for the contention period whichimmediately follows.

The reason for this re-ACKing is simple, WoR-MAC in-corporates the group ID’s of the nodes into the wake-uppreamble, allowing a pre-defined group of nodes to be wokenwith a single preamble. This enables an initiator to wakethe selective group for contention-based communication for aspecific period of time. This creates the problem of avoidingcollisions, as all nodes in a cluster will suddenly want torespond to a single preamble at the same time. In the cluster,each node has a cluster or group ID as well as a subnet IDspecifying the node’s personal address in the group. Using thepersonal address, each node is assigned a single ACK slot ina TDMA driven acknowledge period directly following thewake-up preamble.

This behavior is shown in Fig. 2 where the initiator transmitswake-up preambles until all nodes in the cluster have ACKed,followed by a Begin Communication Alert (BCA) and thenthe contention period. The time reserved for this step tpoll isdependent on the number of nodes in the cluster n, the time

Dawud Gordon Papers und Fragestellungen 3

sleep

sleep

sleep

sleep

sleep

sleep

sleep

sleep

sleep

Cluster Kommandos WoR-MAC

Node 1

Node 2

Node 3

Node n

…

1 23 … n 1 2 3 … n

Hit

Hit

Hit

Hit

Preamble ACK Slots BCA

Communication

Period

Initiator

ACKListen

for Alert

…

Fig. 2. MAC Protocol Behavior in a Group of Nodes

Dawud Gordon Papers und Fragestellungen 4

sleep

sleep

Timing WoR-MAC

1 2 3 … n 1 2 3 … n

Preambles ACK Slots BCA

Communication

Period

Initiator

ACKListen

for Alert

…

tpolltpoll tcom_len

…

Node n

tdetect

tpre

tswitch tswitch

n x tack

tbca

Fig. 3. WoR-MAC Timing Terminology

needed for an ACK tack, the time needed for the preambletpre, the delay a transceiver needs to switch between RX andTX mode tswitch and duration of the BCA tbca as shown inEq. 1. This behavior can be seen graphically in Fig. 3 whichexplains the different timing terms and indicates their positionsaccording to the phases of the MAC protocol.

tpoll = tpre + tswitch + n× tack + tswitch + tbca (1)

Initially in Fig. 2 node 3 does not receive the first preambleand is therefore unaware of the request. The initiator is awareof this due to the slotted ACKs and repeats the process untilnode 3 responds before transmitting the BCA. When to begincommunication can be decided based on a threshold, i.e. when80% of a cluster has responded, or a timeout, i.e. after 10 WoRpreambles. Since the decision is made by the initiator, it is alsopossible for the initiator to update the list of group members ifa node has not made contact in a certain amount of time. Themaximum delay added before the communication starts canbe calculated by the formula 2 where tdetect is the maximumtime a certain node needs to detect a preamble, and m is themaximum number of retries, to collect all ACKs. In the caseof duty cycled preamble detection tdetect has an upper boundof the length the duty cycle period.

toffset = tdetect +m× tpoll (2)

The BCA also specifies the parameters for the comingcommunication period, in this scenario the MAC is either

TABLE IWOR-MAC PACKET CONTENTS AND LENGTHS

Packet name Contents LengthWoR preamble group and node addr, seq. no. 68 bits

Packet type 4 bitsGroup ID 16 bitsNode ID 16 bitsSequence No. 32 bits

WoR ACK Requested Access 68 bitsPacket type 4 bitsGroup ID 16 bitsNode ID 16 bitsRequested Tx Time 32 bits

BCA Comm. Period Length 68 bitsPacket type 4 bitsGroup ID 16 bitsNode ID 16 bitsComm. Period Length 32 bits

Data packets Sensor data 288 bits

TDMA or CSMA-CA, and the BCA only specifies the lengthof the period. Theoretically the BCA could specify any ofa set previously agreed upon MAC protocols as well as therequired parameters for that protocol. Since the location ofthe wake-up ACK in terms of time slot already identifies thesender, the ACK can be used to transmit information otherthan the address of the ACKing node. The ACK is used tocommunicate the amount of data which each node expects totransmit during the next period in order to provide the initiatorwith a method for estimating the appropriate length for thecommunication period tcom len. The period length is thencollaboratively calculated by the group based on the needsof all of the nodes, transmitted to the group in the BCA.The calculation for the length of the communication periodtcom len used here is given by Eq. 3 where tdatan

is themedium access time estimate by node n that it requires andtmac is the MAC overhead per packet, pchan is the probabilitythat a packet will corrupted due to the channel, and pcoll isthe probability of a collision based on the contention.

tcom len =∑n

(tdatan+ tmac)×(1+pchan)×(1+pcoll) (3)

For specific applications, this calculation can be carried outby the application layer for a more accurate estimation usingcross-layer optimization. An overview of the packets, theircontents and their lengths can be found in Tab. I.

In the original publication, X-MAC is specified as push-onlyprotocol, where each node asynchronously wakes the node itwants to communicate with. In cluster scenarios as is oftenthe case with in IE application, there are several argumentsfor using pull in stead of push. For one, push is more energyefficient [10], and often times the cluster head must coordinatecommunication and therefore asynchronous push by end nodesrequires time synchronization. WoR-MAC does not fix thedirection of communication, as it is compatible with a myriadof different MAC protocols, therefore widening the applicationfield of WoR technology.

III. WIRELESS COMMUNICATION FOR INTELLIGENTENVIRONMENTS

The concept of an intelligent environment comes from theareas of Pervasive and Ubiquitous Computing. Intelligent envi-ronments combine embedded computing with communicationbetween devices with the “aim to enhance user productivityand facilitate everyday tasks” [15]. Devices are characterizedby often being objects with which users interact, and haveboth processing and communication capabilities.

Due to the fact that interaction with devices and smartobjects in intelligent environments usually involves movement,wireless communication is often selected as it allows usersto interact with the devices as they normally would. Thecombination of mobility and wireless communication requiresthe use of mobile power sources such as batteries, whichmakes power consumption an important issue for the devicesin question. Low battery lifetimes increase maintenance costs,as well as the total cost of ownership, which if too high wouldnegate any advantages the IE applications would offer. For thisreason reducing the power consumption of individual smartobjects is of the utmost importance.

IE approaches focus on environments in which humans live,work and collaborate, such as offices [5], factories [3] andhomes [12]. Applications for IEs leverage distributed pro-cessing and communication properties in order to assist usersin accomplishing tasks, reaching goals and general comfort.Often times such applications proactively adapt themselves,where environments learn to recognize activities, situationsand goals of users using machine learning techniques [9].The devices then proactively adapt and configure themselvesbased on this information in order to do what the users wishor require without needing explicit instruction. Focus is onimproving human-computer interaction, either implicitly usingproactive methods, or explicitly, for example by augmentinghuman perception using distributed sensors [3].

IE applications require distributed sensing, processing andcommunication in order for devices to assist users [12], [15],although the QoS parameters may differ between applications.For example, augmenting reality with sensory data calls foroverlaying visualizations of sensor measurements onto themeasurement location in the field of vision. In this case latencyis less critical, but packet loss is a serious issue for visualiza-tion [3]. Group activity and goal recognition requires sensoryinformation to be present with hard time synchronization forrecognition (low latency required), but some missing datacan be accounted for by the recognition algorithms [9]. Bothapplications have multiple consumers for each data packet andtherefore use either broadcast or multicast communication.

Finally, since the applications in IE’s are meant to helpthe users in the room, the functionality is not required whenthe users are not present. When users are elsewhere, devicescan conserve a significant amount of energy by reducingtheir communication and processing tasks. However, whenusers enter the environment, the environment needs to remainresponsive and react immediately, providing the QoS required

Dawud Gordon Papers und Fragestellungen 4

Transceiver Power Model

Standby10mA

RX ON85mA

TX ON120mA

Sleep250µA

Hibernate0.1µA

Off0mA

0.2ms

0.21ms

0.2ms0.50ms

0.50ms

Fig. 4. Power States Model of the CC2420 Transceiver in OPNET

by the specific application.

IV. SIMULATION ENVIRONMENT

The evaluation of the WoR-MAC protocol was carried outwith the help of the popular OPNET Modeler 15.0 simulationenvironment. The included process models for CSMA-CAfound in IEEE 802.11 provided the basis for the implementa-tion of our new MAC protocol. This underlines the statementthat the WoR-MAC is able to act as a meta-protocol andother MAC protocols can be easily embedded. The includedmodules only needed small adaptations to generate statisticsfor power consumption using a custom transceiver model.Additionally X-MAC was implemented for comparison withstandard protocols as well as the novel WoR-MAC protocol.The power consumption model is based on a real transceivermodule [8] and the respective values can be seen in the Fig. 4.In addition to RX and TX the Sleep state was invoked by theMAC processes to reduce power consumption in idle phases.

The covered area spanned 2m x 3m, therefore all peerswere able to communicate with each other in a single hoptopology. Every 5 seconds, an initiating node representing amobile device carried activates the WSN by beginning a newcommunication period. Theoretically, the presence of the userin the IE could be sensed or recognized by the environmentand does not need to be explicitly signaled. This requires acertain amount of complexity at the application level howeverwhich does not make sense in the context of evaluating a MACprotocol. In order to simulate the communication behavior ofan intelligent environment, nodes randomly generate a numberof communication packets from a normal distribution between1 and 5. Each packet is addressed to 1 to 5 other nodes, wherethe number of nodes as well as their identities are selected atrandom. The simulation varies the number of nodes in theroom from 5 to 100 in increments of 5 nodes. Every 10minutes the test subject was simulated to have entered theintelligent environment, staying there for 5 minutes, and thenleaving again. This was simulated over a timeframe of 12hours, which was then repeated 5 times with different random

TABLE IIWOR-MAC PARAMETERIZATION FOR THE SIMULATION

Parameter name Value UnitsBitrate 250 Kbits/sPreamble length 68 bitsACK period length variable (0.5 ms/node) msReceiver duty cycle 50 %Duty cycle period 100 msTx strength 5 mWRx Threshold -95 dBm

seeds for each protocol and each network size in order togenerate sufficient statistics.

As mentioned before, the application should not be activewhen the user is not in the room. To create this behavior,CSMA-CA and TDMA use an initial packet, from the user’sdevice to all other nodes, to signal the start of the communica-tion phase. Therefore the sensor nodes have to stay in receivemode until receiving this packet, announcing the presence ofthe user. X-MAC and WoR-MAC use preambles to initiatecommunication, therefore this type of behavior is inherent andthe protocols do not need to be modified.

The performance of WoR-MAC was evaluated using thefollowing metrics as a function of the number of nodes in thenetwork:

Packet Loss Since the number of packets transmitted isgenerated by a random process, the estimated packet loss canbe described stochastically as well. The number of packetstransmitted during the simulation can be output by the simu-lator but calculating how many of these were received is notso straight forward. Each node generates a certain number ofpackets, and a certain number of destination nodes for eachpacket for each communication period. Probabilistically, thenumber of packets which each node should receive is theexpectation of the probability distributions generating thosepackets, in this case the mean, or the number of sent packets.For this reason the packet loss in percent can be estimated by:

Ploss = 100 ∗ (Psent −Avg(Precv))/Psent (4)

where Psent and Precv are the number of packets sent andreceived by each node in the network respectively. For X-MAC this estimation is not necessary, as the protocol is notcapable of broadcast and therefore the number of packets sentand received are exactly equal.

Latency We define latency, or transmission delay, as thetime between the start of the communication period until thepacket is received at the destination node. The values shownhere are the averages of all packets during 12 hours of all 5simulation runs for a fixed number of nodes and a specificprotocol.

Energy Consumption The energy consumption representsthe electrical energy spent by each node during the completesimulated time. Only the power consumed by the transceiveris considered, since the utilization of the other parts of thesensor node, e.g. sensors, actuators, memory, etc., is nearlyidentical for all protocols.

0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0

10.0

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100

Late

ncy

in S

.

Number of Nodes

CSMA-CA TDMA X-MAC WoR-CSMA WoR-TDMA

0.0

0.1

0.2

0.3

0.4

0.5

0.6

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100

Late

ncy

in S

.

Number of Nodes

CSMA-CA TDMA X-MAC WoR-CSMA WoR-TDMA

Fig. 5. Average Latency

V. RESULTS

A. Latency

Fig. 5 shows the results of the cluster-based simulation withrespect to packet latency caused by the protocols examined. Asindicated by the figure, the latency of CSMA-CA and TDMAis comparable to that of their embedded counterparts, WoR-CSMA and WoR-TDMA respectively. WoR-CSMA slightlyoutperformed CSMA-CA in terms of latency, which at firstglance appears to be counterintuitive but will be explainedin Sec. VI. WoR-TDMA performed slightly only worse thanTDMA initially, although the latency per packet for WoR-TDMA increases more rapidly over the number of nodescommunicating that for TDMA. This is due to the fact that asthe number of nodes increases, so too does tpoll, or the timerequired for all nodes to respond to the wake-up command.X-MAC on the other hand, performed significantly worse thanthe other two protocols, with 1.23 seconds of latency for only 5nodes in the cluster, up to almost 9 seconds for large numbersof nodes.

B. Packet Loss

Fig. 6 shows that in terms of packet loss, both WoR-CSMAand WoR-TDMA performed similarly to the original proto-cols CSMA-CA and TDMA. Both TDMA and WoR-TDMAincurred losses very close to 0, around 0.01% on average.CSMA-CA and WoR-CSMA also performed similarly, withlosses for WoR-CSMA approximately 4.5% higher than forCSMA-CA on average. The causes of this discrepancy will bediscussed in detail in Sec. VI.

0

10

20

30

40

50

60

70

80

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100

Pac

ket

Loss

in %

Number of Nodes

CSMA-CA TDMA X-MAC WoR-CSMA WoR-TDMA

Fig. 6. Average Packet Loss in Percent

1300

1500

1700

1900

2100

2300

2500

2700

2900

5

10

15

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

10

0

Avg

. En

erg

y/N

od

e (

mW

s)

Number of Nodes

CSMA-CA TDMA X-MAC WoR-CSMA WoR-TDMA

1300

1400

1500

1600

1700

1800

1900

5

10

15

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

10

0

Avg

. En

erg

y/N

od

e (

mW

s)

Number of Nodes

X-MAC WoR-CSMA WoR-TDMA

Fig. 7. Average Energy Consumed by Each Node

C. Energy Consumption

The energy consumption values for the simulated scenarioare depicted in Fig. 7. CSMA-CA and TDMA have a constantconsumption of approx. 2680 mW per node, due to the fact thatboth protocols do not support duty-cycling and must thereforemaintain the transceiver on at all times. WoR-TDMA incurredthe lowest power consumption scaling linearly from 1375 mWper node for 5 nodes to 1488 mW for 100 nodes. WoR-CSMA also scales linearly from 1381 mW for 5 nodes to amaximum of 1788 mW for 70 nodes, at which point the energyconsumption remains constant up to 100 nodes. The reason forthis behavior will be discuss in Sec. VI. X-MAC consumes1434 mW on average for 5 nodes, up to 1710 mW for 100nodes, rising almost linearly but flattening out as collisionsand packet loss increase.

VI. DISCUSSION

A. Packet Loss

As indicated in Fig. 6, both CSMA-CA and TDMA pro-duced very similar packet loss values when compared withtheir counterpoarts embedded in WoR-MAC. For TDMA andWoR-TDMA, these values can be approximated to 0, withpacket loss resulting only from minor instability of the chan-nel. The 4.5% increase in packet loss from CSMA-CA toWoR-CSMA is due to the fact that WoR-CSMA discards asmall amount (on average 4.5%) of packets which remain inthe queue at the end of the communication period.

The standard OPNET CSMA-CA module was used forcommunication, natively and embedded in WoR-MAC. Oc-casionally, some packets are assigned a very long back-offtime with very low probability due to random exponentialback-off times. WoR-CSMA discards these packets at the endof a communication period, slightly increasing packet loss(and reducing latency) for WoR-CSMA, whereas CSMA-CAtransmits these few packets with very high latency.

B. Latency

Fig. 5 indicates that X-MAC, while being the standard forlow-power WoR communication, performs poorly in terms oflatency in an IE. This is mainly due to the design of theprotocol, which is optimized for point-to-point, one-way, uni-cast communication. Since only a single node can be awokenat one time, a packet addressed to multiple nodes must betransmitted to each node separately. Incorporating multi-nodewake-ups into WoR-MAC means the cluster can be awokenall together, making broadcast communication possible andtherefore greatly reducing latency.

While greatly reducing latency when compared to X-MAC,WoR-MAC does increase latency of the embedded protocolwhen compared to the native protocol. Comparing WoR-TDMA with TDMA for 5 nodes yields an offset of 17.8ms, whereas the difference between the two protocols for 100nodes is 208 ms. This indicates that there are two factors whichaffect the latency of the embedded protocol. First there is alatency offset caused by all nodes having to wake up beforecommunication can begin. Second there is the fact that tpollscales with the number of nodes in the environment, as canbe observed in the slope of latency curves for WoR-TDMAand TDMA respectively.

This same effect can be observed with WoR-CSMA andCSMA-CA, where latency for WoR-CSMA climbs faster overthe number of nodes than that of CSMA-CA. In this casehowever, the offset is negative, and embedding CSMA-CA inWoR-MAC actually improves latency. This is however due tothe fact that outlier packets with extremely high latency arediscarded, as is reflected in the packet loss statistics in Fig. 6.

C. Power Consumption

CSMA-CA and TDMA perform comparably badly in termsof power consumption as would be expected, since bothprotocols do not duty cycle the transceiver. X-MAC, whileoptimized for low-power unicast communication, begins to

break down for large numbers of nodes as wake-ups and com-munication begin to collide, increasing power consumption, aswell as packet loss and latency. WoR-TDMA outperforms allother protocols, consuming 4.1% less energy than X-MAC and48.9% less than CSMA-CA/TDMA for 5 nodes, and 13.0%less than X-MAC and 44.5% less than CSMA-CA/TDMA for100 nodes.

The top-out for power consumption for WoR-CSMA oc-curs when the collaborative estimation function for tcom len

surpasses the amount of time available for communication, inthis case 5 seconds. The estimation of tcom len is liberal forthis simulation, which is why neither latency nor packet lossare dramatically affected. Eventually as the communicationsaturates tcom len, latency will flatten out as all further packetsare discarded, and loss will increase dramatically. The fact thatthis does not occur means that there is room for optimizationin the estimation algorithm, specifically in the time requiredfor retries due to back-offs, which would considerably reducepower consumption for a real application.

VII. RELATED WORK

One of the first solutions to the overhearing problem inwireless networks was S-MAC which synchronizes wake andsleep periods of all nodes in the network to effectively reducepower consumption due to communication [17]. The nextadvance came with Low-Power Listening (LPL) [6] whichallows a node to be awoken by a remote source, and remainin sleep mode until it is needed, either remotely or due tosome internal process. This was accomplished by listeningto the wireless channel and receiving a wake-up messageby monitoring the RSSI values of the channel, which costsfar less than communication. LPL was further improved withthe development of B-MAC [13] which added several usefulfeatures. While effective, LPL does not allow targeting ofremote wake-ups and therefore causes overhearing as all nodeswho are in range are woken by a message.

WiseMAC [7] improves on LPL by using preamble sam-pling to periodically sample the channel in receive mode, andthereby being able to receive information within the wake-upmessage, effectively reducing overhearing by allowing targetedwake-ups. The concept of strobed preambles was presentedin the X-MAC protocol [4] which reduces preamble lengthand response time. X-MAC also further reduces overhearingdue to long preambles by ACKing the first preamble receivedand therefor beginning communication at the earliest possiblemoment. In the original publication, X-MAC is defined as apush-only protocol, which is not optimal for IE applications.

Several protocols have previously been introduced whichextend the functionality of X-MAC, making it more practicalin specific scenarios and situations. BEAM-MAC [2] adaptsX-MAC by appending the payload directly to the wake-up preamble and aggregating payloads into a single packet.This approach is optimized for multi-hop routing and WoR-MAC builds on this work, using a similar method for piggy-backing information onto preambles and ACKs. MaxMAC[11] dynamically changes the duty-cycle based on the network

load, eventually converging to CSMA when load peaks. Whilethis method is not optimal for a pull scenario where loads canbe precalculated, we build on this work using a load-adaptivecontention frame length as well. BurstMAC [14] adapts tobursts in traffic by utilizing multiple channels for parallelcommunication and uses a similar method of slotted ACKschannel allocation.

WoR-MAC can be viewed as a meta MAC which allows itto be used as a wrapper for frames of other MAC protocols.The WoR principles are an extension of X-MAC with severalmajor changes. First, wake-ups can be addressed to multiplenodes at once, allowing group wake-ups, and the ACK periodis slotted using TDMA in order to allow multiple nodes toACK a single wake-up. Once awoken, nodes operate on anovel Wake-to-Contention (WTC) paradigm in which accessto the media is arbitrated by a separate MAC protocol. Finally,the length of the WTC MAC period is specified collaborativelyby using the wake-up ACKs to convey load estimations.

VIII. CONCLUSION

This work began by introducing the wireless communicationscenarios of intelligent environments (IE). Example scenarioswhere introduced which demonstrated different requirementson the MAC layer in terms of quality-of-service (QoS). Anovel MAC protocol, WoR-MAC was introduced which allowsapplication developers to combine QoS capabilities of otherMAC protocols with the power-saving WoR abilities of X-MAC. The protocol allows pre-existing MAC protocols tobe embedded inside duty-cycled wake-up preambles to allowcommunicating nodes to reduce their power consumptionwhile maintaining communication QoS. The advances of theWoR-MAC protocol over the standard Wake-on-Radio proto-col X-MAC are three-fold:

• WoR-MAC introduces group or cluster-addressed wake-ups allowing nodes to wake multiple neighbors at once.

• WoR-MAC allows woken nodes to collaboratively esti-mate the length of a communication period.

• WoR-MAC allows communication via unicast, multicastor bi-directionally (send/receive), using an embeddedMAC protocol.

In order to evaluate the capabilities of the novel protocol,it was simulated based on the parameters of an example IEscenario. Two example IE applications were presented, namelygroup activity recognition and augmented reality based onsensory information, each of which has a different focus anddifferent requirements on the MAC layer of the IE. Twoprotocols, CSMA-CA and TDMA, were used as examples,and were embedded into WoR-MAC creating WoR-CSMAand WoR-TDMA respectively for comparison. The simulationmodeled communication using a standard transceiver and thesensing and communication requirements taken from a realscenario. During the simulation, packet loss, latency andenergy consumption where observed as QoS parameters forcomparing all protocols.

The results of the simulation indicated that embeddinga MAC protocol in WoR-MAC significantly lowered power

consumption by introducing duty-cycling. The amount ofpower saved is dependent on implementation and scenario,but is close to 49% in some cases, which is the upper limitgiven by the duty-cycle. The simulation results also indicatedthat the QoS parameters of the original protocol are generallypreserved after embedding it in WoR-MAC. The general ef-fects on latency are a slight increase due to the wake-up period,which grows as the number of nodes in the network increases.Packet loss, in general, remains unaffected. The exception isWoR-CSMA, where the random and exponential nature of thebackoffs causes high-latency packets to be discarded, creatinga slight decrease in latency and increase in loss.

In total, WoR-MAC allows IE applications to conserve largeamounts of energy, in the simulated scenario up to 49%.At the same time, WoR-MAC maintains the QoS propertiesprovided by the original, non-duty-cycled MAC protocols withonly minor sacrifices. This novel protocol provides developerswith a tool to conserve energy during periods of sporadicactivity or inactivity, without the QoS sacrifice which normallyaccompanies such approaches.

IX. FUTURE WORK

WoR-MAC has been implemented on wireless sensor nodesdesigned for wireless monitoring in aeronautical and automo-tive application areas. A full evaluation of the protocol underreal conditions has yet to be conducted. Also current imple-mentations only support embedding a single MAC protocolinto WoR-MAC. Theoretically, many different MACs couldbe integrated in parallel, allowing almost seamless switchingbetween embedded protocols. This would allow new researchinto cross-layer optimization to be conducted, and possiblyopen new doors to further improving the tradeoff between QoSand energy consumption.

ACKNOWLEDGMENTS

The authors would like to acknowledge funding by theEuropean Commission under the ICT project “CHOSeN -Cooperative Hybrid Objects Sensor Networks” (Project No.224327, FP7-ICT-2007-2) within the 7th Framework Pro-gramme.

REFERENCES

[1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci. Wirelesssensor networks: a survey. Computer Networks, 38(4):393 – 422, 2002.

[2] M. Anwander, G. Wagenknecht, T. Braun, and K. Dolfus. Beam: Aburst-aware energy-efficient adaptive mac protocol for wireless sensornetworks. In Networked Sensing Systems (INSS), 2010 Seventh Interna-tional Conference on, pages 195 –202, 2010.

[3] Matthias Berning, Till Riedel, Yong Ding, Takashi Miyaki, NicolaieFantana, and Michael Beigl. Augmented service in the factory of thefuture. In Adjunct Proceedings of the Eigth International Conferenceon Autonomic Computing (ICAC), 2011.

[4] Michael Buettner, Gary V. Yee, Eric Anderson, and Richard Han. X-mac: a short preamble mac protocol for duty-cycled wireless sensornetworks. In SenSys ’06: Proceedings of the 4th international conferenceon Embedded networked sensor systems, pages 307–320, New York, NY,USA, 2006. ACM Press.

[5] Anind K. Dey, Daniel Salber, Gregory D. Abowd, and MasayasuFutakawa. The conference assistant: Combining context-awareness withwearable computing. In Proceedings of the 3rd IEEE InternationalSymposium on Wearable Computers, ISWC ’99, pages 21–, Washington,DC, USA, 1999. IEEE Computer Society.

[6] A. El-Hoiydi. Aloha with preamble sampling for sporadic traffic in adhoc wireless sensor networks. In IEEE International Conference onCommunications (ICC) [1], pages 393 – 422.

[7] A. El-Hoiydi and J.-D. Decotignie. Wisemac: An ultra low power MACprotocol for multi-hop wireless sensor networks. In First Int. Workshopon Algorithmic Aspects of Wireless Sensor Networks (ALGOSENSORS2004), Lecture Notes in Computer Science, LNCS 3121, pages 18–31.Springer-Verlag, July 2004.

[8] Chipcon Products from Texas Instruments. Cc2420 2.4ghz ieee 802.15.4 / zigbee-ready rf transceiver datasheet.http://www.ti.com/lit/ds/symlink/cc2420.pdf, January 2012.

[9] Dawud Gordon, Jan-Hendrik Hanne, Martin Berchtold, Takashi Miyaki,and Michael Beigl. Recognizing group activities using wearable sen-sors. In 8th International ICST Conference on Mobile and UbiquitousSystems: Computing, Networking and Services (MobiQuitous). Springer-Verlag, December 2011.

[10] David Hasenfratz, Andreas Meier, Matthias Woehrle, Marco Zimmer-ling, and Lothar Thiele. If you have time, save energy with pull.In Proceedings of the 8th ACM Conference on Embedded NetworkedSensor Systems, SenSys ’10, pages 423–424, New York, NY, USA, 2010.ACM.

[11] Philipp Hurni and Torsten Braun. Maxmac: A maximally traffic-adaptivemac protocol for wireless sensor networks. In EWSN, pages 289–305,2010.

[12] Cory Kidd, Robert Orr, Gregory Abowd, Christopher Atkeson, IrfanEssa, Blair MacIntyre, Elizabeth Mynatt, Thad Starner, and WendyNewstetter. The aware home: A living laboratory for ubiquitouscomputing research. In Norbert Streitz, Jane Siegel, Volker Hartkopf,and Shinichi Konomi, editors, Cooperative Buildings. Integrating Infor-mation, Organizations, and Architecture, volume 1670 of Lecture Notesin Computer Science, pages 191–198. Springer Berlin / Heidelberg,1999.

[13] J. Polastre, J. Hill, and D. Culler. Versatile low power media access forwireless sensor networks. In SenSys04, pages 95–107, Baltimore, MD,November 2004.

[14] M. Ringwald and K. Romer. Burstmac - an efficient mac protocol forcorrelated traffic bursts. In Networked Sensing Systems (INSS), 2009Sixth International Conference on, pages 1 –9, 2009.

[15] M. Roman, C. Hess, R. Cerqueira, A. Ranganathan, R.H. Campbell, andK. Nahrstedt. A middleware infrastructure for active spaces. PervasiveComputing, IEEE, 1(4):74 – 83, oct-dec 2002.

[16] K. Romer and F. Mattern. The design space of wireless sensor networks.Wireless Communications, IEEE, 11(6):54–61, Dec. 2004.

[17] Wei Ye, John Heidemann, and Deborah Estrin. An energy-efficient macprotocol for wireless sensor networks. In the 21st International AnnualJoint Conference of the IEEE Computer and Communications Societies(INFOCOM 2002), pages 1567–1576, 2001.