hp://www.diva-poral.org This is he published version of a paper presened a The 2013 IEEE Inernaional Conference on Inerne of Things, Beijing, China, 20-23 Augus 2013. Ciaion for he original published paper: Glaropoulos, I., Mangold, S., Vukadinovic, V. (2013) Enhanced IEEE 802.11 Power Saving for Muli-Hop Toy-o-Toy Communicaion. In: IEEE Compuer Sociey N.B. When ciing his work, cie he original published paper. Permanen link o his version: hp://urn.kb.se/resolve?urn=urn:nbn:se:kh:diva-128571
Enhanced IEEE 802.11 Power Saving for Muli-Hop Toy-o-Toy Communicaion Ioannis Glaropoulos, Sefan Mangold, Vladimir Vukadinovic Disney Research 8092 Zurich, Swizerland Absrac In he fuure Inerne of Things (IoT), baerypowered devices equipped wih shor range radios may need o communicae wih each oher over muli-hop links. This may significanly increase heir energy consumpion. Whereas mos research on IoT assumes ha he devices use energy-efficien IEEE 802.15.4 wireless ransceivers, we focus on IEEE 802.11 because of is wide peneraion in consumer elecronics such as oys. We exend he IEEE 802.11 power saving mode (PSM), which allows he devices o ener he low-power doze sae, wih a raffic announcemen scheme ha faciliaes muli-hop communicaion. The scheme propagaes raffic announcemens along muli-hop pahs o ensure ha all inermediae nodes remain awake o forward he pending daa frames wih minimum laency. Simulaion resuls show ha he proposed Muli-Hop PSM (MH-PSM) improves boh end-o-end delay and doze ime compared o he sandard PSM. MH-PSM is pracical and sofware-implemenable since i does no require changes o he pars of he IEEE 802.11 medium access conrol ha are ypically implemened in hardware. I. INTRODUCTION Many objecs, such as consumer elecronics and oys, are becoming equipped wih sensors and wireless communicaion capabiliies. They will connec o each oher locally using ad hoc neworks and globally using IP infrasrucure o creae he Inerne of Things (IoT). Ad hoc neworks enables wireless devices o exchange daa wih one anoher when a fixed nework infrasrucure (access poins, cellular base saions) is no available. In ad hoc neworks, communicaion beween devices ha are ou of each oher s ransmission range is esablished over muli-hop roues. Hence, each device is no only receiving and ransmiing is own daa, bu i also serves as a muli-hop relay for oher devices. This increases he energy consumpion and decreases he baery lifeime of he devices. Therefore, one of he major challenges for ad hoc neworking of baery-powered devices is he energyefficiency of radio communicaion. Mos of he on-going research on IoT assumes ha he devices are equipped wih low-power IEEE 802.15.4 (Zigbee) ransceivers. However, for consumer elecronics, such as radio-enabled oys, he wide peneraion of IEEE 802.11 (Wi-Fi) dicaes he choice of he wireless echnology. The energy consumpion of 802.11 is high compared o 802.15.4 and oher low-power radios. To alleviae his problem, he 802.11 sandard [1] already specifies Power-Saving Mode (PSM) ha allows idle 802.11 saions o ransiion o a low-power doze sae. An 802.11 saion in PSM mode wakes up periodically from he doze sae, lisens for raffic announcemens coming Fig. 1. Applicaion scenario: Muli-hop communicaion beween oys in an oudoor game. Devices communicae direcly wih each oher, wihou nework infrasrucure. from oher saions ha have daa packes desined for i, and announces is own daa packes desined for oher saions. If a STA does no receive any raffic announcemens and i does no have buffered packes ha can be ransmied, i reurns o he doze sae. The sandard [1] specifies he deails of PSM mechanism for boh infrasrucure/bss mode (Basic Service Se wih an access poin) and ad hoc/ibss mode (Independen Basic Service Se wihou an access poin). In he ad hoc mode, especially in mui-hop neworks, he PSM is known o perform poorly, causing undesirable energy consumpion and/or long packe delivery delays [2, 3, 4]. The reason is because PSM has been originally designed for single-hop communicaion in he infrasrucure mode (from he access poin o a saion and vice versa). When a daa frame is forwarded over muliple hops, sandard 802.11 PSM may significanly increase is delivery delay because only he nex-hop saion is noified abou he pending frame via raffic announcemens he saions on subsequen hops may remain in he doze sae. Therefore, in each beacon inerval he frame is forwarded over a single hop and has o be buffered before being forwarded furher. In his paper, we propose a mechanism ha wakes up downsream saions so ha daa frames can be forwarded over muliple hops in a single beacon inerval. This is achieved by insrucing each saion along he pah o send a raffic announcemen o is downsream neighbor. The proposed mechanism significanly reduces he end-o-end laency, es-
Beacon Beacon inerval Awake or Doze Beacon inerval ATIM window Daa TX/RX window ATIM window Daa TX/RX window Beacon Awake or Doze Fig. 2. 802.11 PSM divides each beacon inerval ino an ATIM window and a daa TX/RX window. pecially for bursy raffic where inermediae saions may move o he doze sae beween wo consecuive raffic burss. The mechanism enhances he sandard PSM o wha we call muli-hop PSM (MH-PSM). MH-PSM does no prevens saions o iner-operae wih hose ha run sandard PSM i does no aler he sae machine, frame formas, and oher imporan elemens of he proocol. MH-PSM is also sofwareimplemenable i does no require modificaions o he pars of he 802.11 MAC proocol ha are usually implemened in hardware, such as he CSMA/CA and handling of conrol frames (RTS, CTS, ). The res of he paper is organized as follows: Secion II summarizes he sandard 802.11 PSM. In Secion III we describe MH-PSM and discuss pracical implemenaion issues. The performance of he mechanism is evaluaed in Secion IV using simulaions. In Secion V, we provide an overview of relaed work. Finally, Secion VI concludes he paper. II. POWER-SAVING MODE FOR 802.11 AD HOC NETWORKS In he sandard 802.11 PSM for ad hoc/ibss neworks, ime is divided ino periods called beacon inervals. Each saion wakes up a he beginning of each beacon inerval and sars a back-off procedure in an aemp o ransmi a beacon. If a saion receives a beacon from anoher saion before is back-off imer expires, i cancels he pending beacon ransmission. The Timing Synchronizaion Funcion (TSF) uses he ime-samped beacons o synchronize clocks among saions o ensure ha all saions wake up a he same ime. Following he beacon exchange, each saion says awake for an ATIM window inerval, as shown in Fig. 2. During he ATIM window, saions announce pending daa frames o heir neighbors using unicas announcemen raffic indicaion messages (ATIMs). ATIMs are sen using 802.11 disribued coordinaion funcion (DCF), which implemens CSMA/CA channel access procedure. A saion ha receives an ATIM should respond wih an. Successful exchange of ATIM- packes beween wo saions implies ha hey can now exchange any pending daa frames and hus boh should say awake unil he nex beacon inerval. Saions ha do no send nor receive any ATIM frame during an ATIM window will move o he doze sae for he res of he beacon inerval. Afer he end of ATIM window, all saions ha remain awake will follow he normal DCF procedure o ransmi and receive daa frames. The described PSM proocol has many drawbacks. For example, when a saion successfully ransmis or receives an ATIM frame during an ATIM window, i mus say awake during he enire beacon inerval. A low loads, his approach Beacon resuls in a much higher energy consumpion han necessary. Anoher shorcoming is ha all saions in an IBSS mus use he same fixed ATIM window size, which is se a he ime when he IBSS is creaed, as well as idenical beacon inervals. Since he ATIM window size criically affecs he hroughpu and energy consumpion, he fixed ATIM window does no perform well in all siuaions, as shown in [5]. Some of hese drawbacks have been addressed in previous works, which are menioned in he relaed work secion. This paper, however, addresses he problem of end-of-end delay on muli-hop pahs, which is described in he following. Consider a scenario where saion A needs o send a single frame/message o saion D using inermediae saions B and C as relays (Fig. 3). In he firs beacon inerval, saion A announces he daa frame o saion B using an ATIM frame. Saion B acknowledges he ATIM an remains awake so ha i can receive he daa following he ATIM window. Saion C has no received any ATIMs and, herefore, i eners he doze sae. Since saion B is no able o forward he frame o C in he curren beacon inerval, i has o wai for he sar of he nex beacon inerval o send an ATIM o saion C. Following a successful ATIM- exchange, he frame is forwarded o C. Saion D will receive he frame in he hird beacon inerval. The resuling end-o-end delay may considerably affec nework applicaions wih sric laency consrains. Therefore, enabling PSM in muli-hop ad hoc neworks mus be combined wih effecive mechanisms for miigaing is effec on he resuling packe delays. III. ENHANCED 802.11 PSM FOR MULTI-HOP COMMUNICATION In he above described scenario, he daa frame sen by A mus be buffered a B before i is relayed o C in he following beacon inerval. This could have been avoided if here was a way for B o, upon receiving he ATIM from A, send an early ATIM o C and D o inform hem abou he pending daa frame a A. This is wha our low-laency muli-hop PSM (MH-PSM) aims o achieve. Before inroducing MH-PSM, we describe he forma of ATIM frames. An ATIM frame includes a MAC header, whose srucure shown in Fig. 4 is common o all managemen frames. The header includes hree address fields: Address 1 conains he MAC address of he ATIM receiver. Address 2 conains he MAC address of he ATIM sender. Address 3 may conain differen informaion depending on he ype of he managemen frame and nework (BSS, IBSS, or mesh). In case of an ATIM frame, Address 3 conains he BSSID (BSS idenifier) of he IBSS, bu his idenifier is no used. The frame body of an ATIM is empy. A. Proposed Exension: Muli-Hop PSM (MH-PSM) We propose ha, in order o inform all saions along he pah o D abou he pending daa frame, he saion A wries he MAC address of D ino he Address 3 field of he ATIM frame ha are sen o B. The mehods ha A can use o resolve he MAC address of D from is IP address are discussed laer
Beacon inerval 1 Beacon inerval 2 Beacon inerval 3 A B C A B C ATIM Daa ATIM Daa ATIM Daa D D ATIM window Daa TX/RX window ATIM window Daa TX/RX window ATIM window Daa TX/RX window Fig. 3. Muli-hop forwarding in sandard 802.11 PSM may cause a delay of several beacon inervals. in his Secion. Upon receiving he ATIM, B inspecs he Address 3 field o derive he final desinaion of he daa frame announced by ha ATIM. I rerieves he MAC address of D from he Address 3 field, resolves i o he IP address of D, and consuls he rouing able o find ou ha C is he nex hop on he pah o D. Then B creaes an ATIM frame for C wih he MAC address of D inside he Address 3 field. When C receives he ATIM from B, i uses he same procedure o creae an ATIM for D. In his way, a wave of ATIMs is creaed along he pah o wake up all relays and he desinaion of he daa frame. Following he end of he ATIM window, he daa frame is forwarded end-o-end since all saions on he pah are in he awake sae. The procedure is illusraed in Fig. 5. The ATIM wave may no reach he end desinaion: i may erminae a he end of he ATIM window or upon reaching a saion ha canno resolve he MAC address of he desinaion. In ha case, he daa frame will be forwarded in he curren beacon inerval as far as he saion ha has received he las ATIM in he sequence. Neverheless, MH-PSM may significanly decrease he end-o-end delay because he probabiliy ha daa frames are forwarded over muliple hops in a single beacon inerval is higher han wih he sandard PSM. B. Address 3 Resoluion The sending saion A needs o sore he MAC address of he desinaion D ino he Address 3 field of ATIMs sen o B. Therefore, A needs o resolve he MAC address of D from is IP address. Since he paper is argeing Inerne of Things (IoT) and smar oy communicaion scenarios, we assume ha IPv6 is used. IPv6 proocol suie uses Neighbor Discovery (ND) proocol [6] for address resoluion, nexhop deerminaion, and duplicae address deecion. Address resoluion enables saions o deermine MAC addresses of heir neighbors given only heir IP addresses. The neighbor soliciaion messages, which are used for address resoluion, are sen via mulicas. The ND proocol is no designed wih muli-hop ad hoc neworks in mind. A node in such nework is able o broadcas o oher nodes wihin is radio range, bu he communicaion is non-ransiive. Therefore, a wireless ad hoc nework is a non-broadcas muli-access (NBMA) srucure wih generally no nework-wide mulicas capabiliies. The nework soliciaion messages are no forwarded in an IBSS. Hence, saion A is only able o resolve MAC addresses of is immediae neighbors, bu no of D, which is muliple hops away. There are several proposals o exend he capabiliies of he ND proocol o muli-hop ad hoc neworks [7] and 6LoWPAN neworks in paricular [8]. These proposals include mechanisms for muli-hop duplicae address deecion (DAD), which allows a saion o check he uniqueness of an IP address in an n-hop neighborhood. The muli-hop DAD can also be used for muli-hop address resoluion: saion A may iniiae muli-hop DAD for he IP address of D. Upon receiving a DAD reques, D will respond wih a DAD confirmaion message ha conains is MAC address. I his way, A can resolve he MAC address of D based on is IP address. Noe ha each saion mainains a cache of resolved addresses, which reduces he need for nework-wide muli-hop address resoluion. A B C D ATIM ATIM D Frame Conrol Duraion Address 1 Address 2 Address 3 Sequence Conrol FCS D Mul67hop*ATIM*forwarding ATIM D Daa Daa Daa ATIM*window Daa*TX/RXwindow Fig. 4. Srucure of he ATIM frame. The Address 3 field can be used for he MAC address of he end desinaion. Fig. 5. The proposed muli-hop forwarding mechanism allows daa frames o be forwarded end-o-end in a single beacon inerval.
TABLE I DEFAULT SIMULATION PARAMETERS. Parameer Value Grid size 7 7 saions Grid spacing 50 m Channel model uni disk IEEE 802.11 PHY mode 11 Mb/s (802.11b) Shor / long rery limi 4 / 7 (Threshold: 500 B) MAC buffer size 100 frames Beacon inerval 50 ms ATIM window 10 ms Traffic model Poisson (λ) Daa frame size Uniform [50,1500] C. Backward-Compaibiliy and Sofware Implemenaion Backward-compaibiliy wih he sandard PSM is ensured since MH-PSM does no violae neiher frame formas nor proocol operaions. Saions ha implemen sandard PSM will no check he Address 3 field of received ATIMs and, herefore, he wave of ATIMs will erminae a such saions. This diminishes some of he delay improvemens, bu oherwise does no preven or impair communicaion. MH- PSM is also sofware-implemenable: Parsing and creaion of ATIM frames are no ime-criical operaions ha have o be implemened in hardware. This enables driver-level implemenaion of MH-PSM wihou modificaions o he lowlevel MAC operaions. We are currenly implemening MH- PSM in an Aheros AR9170 driver. IV. PERFORMANCE EVALUATION We compared he performance of sandard PSM and MH- PSM using simulaions. The performance is measured in erms of end-o-end delay, doze ime raio, ATIM overhead, and packe delivery raio, as defined below: End-o-End Delay is he average ime required o forward a daa frame from a source o is desinaion over muliple hops. I is averaged over all successfully delivered daa frames. Doze Time Raio is he percenage of beacon inervals in which a saion eners doze sae, which closely correlaes wih he energy consumpion. I is averaged over all saions ha paricipae in raffic forwarding. ATIM Overhead is he average number of ATIM frames sen per one successfully delivered daa frame. The relaive ATIM overhead of MH-PSM is he raio of ATIM overheads obained wih MH-PSM and sandard PSM. Packe Delivery Raio (PDR) is he percenage of daa frames ha are successfully delivered o he end desinaion. A saion may drop a daa frame if i exceeds he maximum number of reransmissions. The simulaion seup and he resuls are described in he following: A. Simulaion Seup We implemen and esed MH-PSM in Jemula802 [9], which is a Java-based 802.11 proocol simulaor developed in our group. We consider a regular 7 7 grid of saic 802.11 saions. Adjacen nodes are 50 m apar from each oher. We Fig. 6. Simulaed nework opology wih a single flow. The ransmission range is se o 50 m, 100 m, and 150 m o produce pahs wih 2, 3, and 6 hops, respecively. assume a simple uni disk radio propagaion model. We varied he radio range from 50 m o 150 m o influence he number of hops beween source-desinaion pairs. The beacon inerval and ATIM window size are 50 ms and 10 ms, respecively, unless saed oherwise. The daa raffic is Poisson (exponenial inerarrival imes) wih uniformly disribued frame sizes. The number of acive flows and mean frame inerarrival ime are varied o conrol he load in he nework. The raffic is roued over shores pahs; he rouing is saic. We ensured ha he simulaion duraion is sufficien o make he variaions in ime-moving averages insignifican. The defaul simulaion parameers are summarized in Table I. B. Simulaion Resuls Consider firs he simple single-flow scenario shown in Fig. 6, where he saion in he firs column of he grid is sending daa frames o he saion in he las column over muliple hops. Noe ha non-forwarding nodes in he grid affec he performance of forwarding nodes: The 802.11 sandard mandaes ha a saion ha ransmis a beacon should remain awake for he res of he beacon inerval. As he number of is neighbors decreases, he probabiliy ha a saion ransmis a beacon before i receives one increases. The radio ransmission range is se o 50 m, 100 m, and 150 m in differen simulaion runs, which produces pahs wih 2, 3, and 6 hops, respecively. On average, he sender is generaing λ = 10 frames per second (0.5 frames per beacon inerval). The resuls for he average end-o-end frame delay are shown in Fig. 7 (lef). As expeced, he delay increases wih he number of hops. For he sandard PSM i akes almos N beacon inervals o forward a frame over N hops. I may happen ha a frame is forwarded over muliple hops in a single beacon inerval: if is nex-hop neighbor is awake, a saion may immediaely forward he frame o i, wihou waiing for he nex ATIM window o send a raffic announcemen. In a lighly loaded nework, however, i is likely ha he nex-hop saion is in he doze sae, and herefore, he daa frame has o be announced wih an ATIM in he nex beacon inerval. The resuls show ha he delay is significanly
Fig. 7. End-o-end delay, doze ime raio, and ATIM overhead for differen numbers of hops. shorer for MH-PSM. Alhough i slighly increases wih he number of hops (due o processing in inermediae saions and increasing probabiliy of collisions/reransmissions caused by hidden saions) he average delay is well below 50 ms, which is he duraion of he beacon inerval. As he number of hops increases from wo o six, he percenage of frames ha are forwarded end-o-end wihin a single beacon inerval deceases from 88% o 86%, bu sill remains excepionally high compared o sandard PSM (28% and 0%). The average doze ime raio is shown in in Fig. 7 (middle). The resuls show ha MH-PSM significanly increases he energy efficiency by allowing he saions o move o he doze sae more ofen han he sandard PSM. The reason for his is ha MH-PSM prevens excessive buffering of frames in inermediae saions, which effecively decreases he raffic load and he probabiliy of collisions/reransmissions. The sixhop packe delivery raio for MH-PSM is 99.4% versus 91.5% for he sandard PSM. The resuls presened so far show ha MH-PSM provides boh shorer delay and lower energy consumpion, which is a major improvemen over he sandard PSM whose parameric adjusmens/opimizaions may only rade shorer delay for higher energy consumpion and vice versa. In Fig. 7 (righ), we show he ATIM overhead for boh PSM schemes. While he overhead for MH-PSM is slighly higher for pahs wih few hops (i.e. wo or hree), he opposie is rue for he six-hop pah where i resuls in 25% overhead reducion compared o he sandard PSM. To undersand he reasons for he rend reversal, consider a five-hop pah from saion A o saion E via B, C, and D, as shown in Fig. 8. Assume ha one frame is buffered a saion A and one a saion C. In he bes-case scenario, i will ake four beacon inervals and six ATIMs o deliver boh frames o he desinaion wih he sandard PSM. Wih MH-PSM however, i will only one beacon inerval and four ATIMs o achieve he same because i creaes a wave of ATIMs ha flushes all buffered frames o he desinaion, as shown in Fig. 9. There are however scenarios where he ATIM overhead of MH-PSM is higher han ha of he sandard PSM, even for pahs wih many hops. In he sandard PSM, a saion sends a single ATIM o is neighbor o announce all daa frames ha i inends o forward o his neighbor, regardless of heir end desinaions. In MH-PSM, he saion may send muliple ATIMs wih differen Address 3 fields o he neighbor if he pending daa frames have differen end desinaions. For example, consider wo flows whose eigh-hop pahs conain a common subse or relays, as shown in Fig. 10. In MH-PSM, he common relays may need o forward wo ATIMs wih differen Address 3 fields o heir nex-hop neighbors in he same ATIM window. This is no he case in sandard PSM, where only one ATIM is sen. The resuls in Fig. 11 show ha he ATIM overhead of MH-PSM is almos 60% higher in his scenario. However, MH-PSM ouperforms sandard PSM in all oher respecs: he end-o-end delay is close o enfold shorer, he doze ime raio is slighly higher, and he packe delivery raio is significanly improved. Therefore, he relaive ATIM overhead of MH-PSM had no bearing o he key performance merics. Fig. 8. Sandard PSM requires 4 BIs and 6 ATIMs o deliver he frames buffered a A and C. Fig. 9. MH-PSM requires only 1 BI and 5 ATIMs o deliver he frames buffered a A and C.
Fig. 10. An example of wo flows whose pahs parially overlap. TABLE II PERFORMANCE OF STANDARD PSM AND MH-PSM FOR DIFFERENT BEACON INTERVALS. THE TRANSMISSION RANGE IS 50 M FRAMES ARE FORWARDED OVER SIX HOPS. Bcn. In. Delay (ms) Doze ime (%) PDR (%) PSM MH-PSM PSM MH-PSM PSM MH-PSM 50 ms 269.08 35.44 0.29 0.40 91.54 99.36 100 ms 514.84 47.15 0.15 0.26 85.43 99.17 We nex invesigae he impac of beacon inerval on he performance of sandard PSM and MH-PSM. The resuls presened so far assume a beacon inerval of 50 ms. We increased he beacon inerval o 100 ms and repeaed he simulaions for he basic scenario shown in Fig. 6 wih he ransmission range of 50 m (i.e. six hops). The average frame inerarrival ime is 100 ms. The resuls are summarized in Table II. As expeced, he frame delay for PSM doubles because he ime ha frames say buffered in he inermediae nodes is proporional o he beacon inerval. The delay for MH-PSM also increases, bu he increase is comparably modes. The increase is due o he fac ha MH-PSM does no guaranee ha all frames will be delivered end-o-end in a single beacon inerval. Some of he frames have o be buffered along he pah as in he case of sandard PSM. Anoher observaion is ha wih he sandard PSM packe delivery raio decreases significanly for he longer beacon inerval (from 91.5% o only 85.4%), while wih MH-PSM i decreases only slighly (from 99.4% o 99.2%). Wih he sandard PSM, he number of buffered frames along he pah increases wih he duraion of he beacon inerval, which effecively increases he raffic load in he nework and he probabiliy of collisions. Wih MH-PSM, mos frames are delivered end-o-end wihou buffering in he inermediae nodes. In our final se of simulaions, we consider muliple inersecing flows in he grid. The scenarios wih 2, 4, and 8 flows shown in Fig.12 complemen he single-flow scenario in Fig. 6. The ransmission range is se o 50 m and, herefore, frames are forwarded over six hops. The resuls in Table III show ha he performance deerioraes wih he number of flows. Transmissions of inersecing nodes are especially prone o collisions because hey are surrounded by four acive/forwarding saions ha do no hear each ohers ransmissions ( hidden saions ). The impac of collisions on he performances of he sandard PSM and MH-PSM is somewha differen: While he frame delay for he sandard PSM remains unaffeced by he number of flows, he delay for MH-PSM increases considerably (ye sill remains relaively low). The reason is ha collisions in inersecing nodes may disrup he cu-hrough forwarding of daa frames in MH-PSM. In he single-flow scenario, 88% of frames are forwarded end-o-end in a single beacon inerval. In he eigh-flow scenario, his percenage drops o 79%. The addiional hold-up in inersecing nodes does no affec he frame delay in he sandard PSM so prominenly because mos frames are anyway forwarded only one hop per beacon inerval. C. Ongoing Work and Open Issues We are currenly implemening he proposed MH-PSM on a hardware plaform shown in Fig. 13. The plaform consiss of an Arduino Due board wih ARM Corex-M3 microprocessor and 9 KB of SRAM [10] and an 802.11n ransceiver based on Aheros AR9170 chipse [11]. The plaform runs Coniki operaing sysem [12]. MH-PSM will be implemened as a par of a Coniki Wi-Fi driver for AR9170. We are planning o validae he simulaion resuls on a esbed of 25 devices. We will furher invesigae he impac of ATIM window size and beacon inerval on delay and energy consumpion of MH- PSM. According o he 802.11 sandard, he beacon inerval and ATIM window are deermined a he ime when an IBSS is creaed and shall be saic for he lifeime of he IBSS. A shor ATIM window reduces he energy spen while lisening TABLE III PERFORMANCE OF STANDARD PSM AND MH-PSM FOR DIFFERENT NUMBERS OF FLOWS. THE TRANSMISSION RANGE IS 50 M FRAMES ARE FORWARDED OVER SIX HOPS. Fig. 11. Performance of sandard PSM and MH-PSM for he scenario wih wo flows whose pahs parially overlap. Num. flows Delay (ms) Doze ime (%) PDR (%) PSM MH-PSM PSM MH-PSM PSM MH-PSM 1 269.08 35.44 0.29 0.40 91.50 99.36 2 269.07 39.78 0.21 0.37 83.72 92.44 4 269.34 42.28 0.19 0.34 81.74 86.88 8 273.37 46.73 0.17 0.24 80.01 85.80
Fig. 12. Simulaed nework opology wih 2, 4, and 8 simeric flows. for ATIMs. However, if i is oo shor, i migh no provide enough ime o announce all pending frames, which decreases he hroughpu. If he ATIM window is oo long, here migh be no enough ime o ransmi all announced daa frames in he pos-atim window. The size of he ATIM window can be se based on he expeced raffic load in he nework lower load implies shorer ATIM window o minimize he energy consumpion. MH-PSM inroduces an addiional radeoff: Even a a low raffic load, a longer ATIM window migh be needed in order o propagae he wave of ATIMs end-oend, which means ha he remaining pos-atim window may be oo shor o forward he announced daa frames end-o-end. Hence, some of he downsream saions migh be awaken for no reason. Therefore, in MH-PSM, he choice of he ATIM window size depends no only on he expeced raffic load, bu also on he expeced number of hops o he desinaion. We are also planning o invesigae he ineracions of MH-PSM wih upper layers (rouing, ranspor) and heir join performance under node mobiliy. V. RELATED WORK The IEEE 802.11ah proposal [13] defines a low power medium access mehod ha opimizes sandard 802.11 PSM for baery-powered devices used in smar meering and machine-o-machine communicaion. However, he opimizaion focuses on BSS (infrasrucure) neworks where PSMenabled saions communicae wih an access poin. Opimizaion of PSM for IBSS (ad-hoc) neworks has araced considerable aenion in he research communiy. A number of approaches focuses on minimizing he duraion of idle lisening by inroducing mechanisms for early ransiion o he doze sae [2, 3, 4]. In [2], he explici announcemen of he number of pending frames in ATIMs is proposed in order o allow he receiving saion o move o he doze sae afer i receives he las frame, insead of waiing for he end of he beacon inerval. In [14], he auhors propose a scheme where ATIMs conain informaion abou he naure of he inended raffic, so saions can differeniae beween broadcas and muli-cas raffic; in he laer case hey can immediaely ransi o doze sae if hey are no members of he mulicas group. In various approaches, he early ransiion o he doze sae is combined wih he dynamic adjusmen of he ATIM window duraion, depending on he raffic condiions in he IBSS [15]. In [3] he auhors propose an algorihm for a saion o dynamically adjus he remaining ATIM window duraion as a response o ATIM recepions in order o ransi o sleep earlier in case of low nework raffic.to furher decrease he energy wased for idle lisening, [16] proposes a scheme where ransmiing saions announce heir inenion of sending ATIM frames in a shor ime period a he beginning of he beacon inerval. Saions ha do no send or receive any announcemens do no have o say awake for he enire ATIM window. Considering a similar low-raffic scenario, [17] proposes a scheme where he absence of raffic is declared by ransmiing a delayed beacon, so ha saions can skip idle lisening during he ATIM window. In [18, 19], he auhors propose a opologyaware power-saving algorihm based on he overhearing of he ATIM frames ransmied by he neighbors. By exracing he source addresses from he received ATIM acknowledgmens, a saion can defer from ransmiing ATIMs o saions known o remain awake afer he expiraion of he ATIM window. This scheme can efficienly decrease he required ATIM window size in a fully-conneced IEEE 802.11 mesh nework, bu i is less effecive in muli-hop IBSS nework opologies. Opimizaions of PSM for muli-hop IBSS neworks have also been proposed in several papers. For example, [4] inroduces an ad-hoc clusering scheme where maser nodes form a backbone ha relays he muli-hop raffic beween PSMenabled slave nodes and proposes a disribued algorihm for dynamical and fair selecion of maser nodes in an IBSS. In [20], saions increase heir energy saving by waking-up a muliples of he beacon period and uilize an adapive nex-hop selecion framework in order o decrease he muli-hop packe delays caused by he longer wake-up duy cycles. Laency opimizaion for non-psm saions was addressed in [21], where waves of RTS/CTS frames are proposed o reserve radio resources along he roue for laency-opimized mulihop communicaion. VI. CONCLUSIONS The Fuure Inerne of Things will connec no only Zigbeeenabled devices, such as sensors, bu also consumer elecronics ha predominanly uses Wi-Fi for nework conneciviy. The power saving mechanisms of he IEEE 802.11 MAC have o be
furher opimized o enable low-cos baery-powered devices, such as elecronic oys, o connec o each oher direcly wihou infrasrucure suppor. In his paper, we proposed MHPSM, an exension of he sandard IEEE 802.11 PSM ha enables low-laency communicaion over muliple hops. A he same ime, MH-PSM increases he doze ime raio and, herefore, exends he baery lifeime of he devices. Using simulaions, we showed he effeciveness of he proposed scheme. MH-PSM is sofware implemenable since i does no require changes o he lower MAC. I is also backwardcompaible wih he sandard PSM, which guaranees ineroperabiliy wih legacy devices. VII. NOWLEDGEMENT Fig. 13. Hardware plaform for MH-PSM evaluaion consiss of an Arduino Due board and an Aheros AR9170-based Wi-Fi ransceiver. This work was parially funded by he European Union Sevenh Framework Programme (FP7-ICT/2007-2013) under gran agreemen number 288879 (Calipso; see hp://www.iccalipso.eu/). R EFERENCES [1] Par 11: Wireless LAN Medium Access Conrol (MAC) and Physical Layer (PHY) Specificaions, IEEE Sd., Rev. IEEE Sd 802.11-2012, 2012. [2] D.-Y. Kim and C.-H. Choi, Adapive power managemen for IEEE 802.11-based ad hoc neworks, in Proc. 5h World Wireless Congress, San Francisco, USA, May 2004. [3] E.-S. Jung and N. H. Vaidya, Improving IEEE 802.11 power saving mechanism, Wireless Neworks, vol. 14, no. 3, pp. 375 391, 2008. [4] S. Yongsheng and T. A. Gulliver, An energy-efficien MAC proocol for ad hoc neworks, Wireless Sensor Nework, vol. 1, no. 5, pp. 407 416, 2009. [5] H. Woesner, J.-P. Eber, M. Schlager, and A. Wolisz, Power-saving mechanisms in emerging sandards for wireless LANs: he MAC level perspecive, IEEE Personal Comm., vol. 5, no. 3, pp. 40 48, 1998. [6] T. Narem, E. Nordmark, W. Simpson, and H. Soliman, Neighbor discovery for IP version 6 (IPv6), RFC 4861, Sep. 2007. [7] M. Grajzer, Nd++ an exended IPv6 Neighbor Discovery proocol for duplicae address deecion o suppor saeless address auoconfiguraion in IPv6 mobile ad-hoc neworks, Inerne-Draf, March 2011. [8] Z. Shelby, S. Chakabari, and E. Nordmark, Neighbor discovery opimizaion for Low Power and Lossy Neworks (6LoWPAN), InerneDraf, Aug. 2012. [9] S. Mangold, Jemula802, hps://gihub.com/schmis/jemula802. [Apr2013]. [10] Arduino Due, hp://arduino.cc/en/main/arduinoboarddue. [Apr2013]. [11] Aheros AR9001U, hp://wikidevi.com/files/aheros/specshees/ AR9001U.pdf. [Apr-2013]. [12] Coniki OS, hp://www.coniki-os.org/. [Apr-2013]. [13] S. Aus, R. Prasad, and I. G. M. M. Niemegeers, IEEE 802.11ah: Advanages in sandards and furher challenges for sub 1 GHz Wi-Fi, in Proc. IEEE In. Conf. Comm. (ICC), Oawa, Canada, 2012. [14] A. K. Sharma, A. Gupa, and A. Misra, Opimized power saving mechanism for wireless ad hoc neworks, in Proc. 1s In. Conf. Recen Advances in Inf. Tech. (RAIT), Dhanbad, India, Mar. 2012. [15] E.-S. Jung and N. Vaidya, An energy efficien MAC proocol for wireless LANs, in Proc. IEEE Infocom, New York, USA, 2002. [16] N. Rajangopalan and C. Mala, Modified power save model for beer energy efficiency and reduced packe laency, American Journal of Engineering and Applied Sciences, vol. 5, no. 3, pp. 237 242, 2012. [17] J.-M. Choi, Y.-B. Ko, and J.-H. Kim, Enhanced power saving scheme for IEEE 802.11 DCF based wireless neworks, in Personal Wireless Comm., ser. Lecure Noes in Compuer Science, 2003, vol. 2775, pp. 835 840. [18] W. Akkari, A. Belghih, and A. Ben Mnaouer, Enhancing power saving mechanisms for ad hoc neworks using neighborhood informaion, in Proc. In. Wireless Comm. and Mobile Comp. Conf. (IWCMC), Cree, Greece, Aug. 2008. [19] A. Belghih and W. Akkari, Neighborhood aware power saving mechanisms for ad hoc neworks, in Proc. IEEE Conf. Local Compuer Neworks (LCN), Monreal, Canada, Oc. 2008. [20] R.-H. Hwang, C.-Y. Wang, C.-J. Wu, and G.-N. Chen, A novel efficien power-saving MAC proocol for muli-hop MANETs, In. Journal of Comm. Sysems, vol. 26, no. 1, pp. 34 55, Jan. 2013. [21] G. Hierz, J. Habeha, E. Weiss, and S. Mangold, A cu-hrough swiching echnology for IEEE 802.11, in Proc. IEEE Circuis and Sysems Symp. on Emerging Technologies, Shanghai, China, 2004.