paper special section on network and system ......paper special section on network and system...

2976IEICE TRANS. COMMUN., VOL.E96–B, NO.12 DECEMBER 2013

PAPER Special Section on Network and System Technologies for Sustainable Society

IEEE 802.11ah Based M2M Networks Employing Virtual Groupingand Power Saving Methods

Kohei OGAWA†a), Student Member, Masahiro MORIKURA†, Fellow, Koji YAMAMOTO†, Senior Member,and Tomoyuki SUGIHARA††, Nonmember

SUMMARY As a promising wireless access standard for machine-to-machine (M2M) networks, the IEEE 802.11 task group ah has been dis-cussing a new standard which is based on the wireless local area network(WLAN) standard. This new standard will support an enormous numberof stations (STAs) such as 6,000 STAs. To mitigate degradation of thethroughput and delay performance in WLANs that employ a carrier sensemultiple access with collision avoidance (CSMA/CA) protocol, this pa-per proposes a virtual grouping method which exploits the random arbi-tration interframe space number scheme. This method complies with theCSMA/CA protocol, which employs distributed medium access control.Moreover, power saving is another important issue for M2M networks,where most STAs are operated by primary or secondary batteries. Thispaper proposes a new power saving method for the IEEE 802.11ah basedM2M network employing the proposed virtual grouping method. With theproposed virtual grouping and power saving methods, the STAs can savetheir power by as much as 90% and maintain good throughput and delayperformance.key words: M2M networks, wireless LAN, IEEE 802.11ah, EDCA, powersaving, CSMA/CA

1. Introduction

Machine-to-machine (M2M) networks, which are com-posed of power meters and various sensors, are currentlyattracting considerable attention for the realization of smartgrids and smart communities. M2M networks are charac-terized by a large number of stations (STAs). A wirelessmedium access standard for M2M networks has been dis-cussed by the IEEE 802.11 task group ah (hereafter we writethis abbreviated IEEE 802.11ah) [1], which is based on thecutting-edge wireless local area network (WLAN) standardsuch as the IEEE 802.11ac standard [2]. In IEEE 802.11ah,thousands of STAs are assumed to belong to an access point(AP). As the number of STAs increases, the throughput anddelay performances of the system are significantly degradedwhen the carrier sense multiple access with collision avoid-ance (CSMA/CA) protocol is exploited. In normal cases,this problem does not occur because traffic in M2M net-works supposed to be usually light. However, we shouldconsider that temporary traffic congestion occurs. For ex-ample, such instances occur when the AP transmits a cer-tain trigger frame to many STAs to collect STA information.

Manuscript received March 13, 2013.Manuscript revised June 27, 2013.†The authors are with the Graduate School of Informatics,

Kyoto University, Kyoto-shi, 606-8501 Japan.††The author is with the Allied Telesis Holdings K.K, Tokyo,

141-0031 Japan.a) E-mail: [email protected]

DOI: 10.1587/transcom.E96.B.2976

Many STAs respond to the trigger frame and transmit theirreply data frames. These events cause traffic congestion inan M2M network.

To solve this problem, several STA grouping methodshave been proposed, which can be classified into real andvirtual grouping. The first method allocates all STAs intosome groups and assigns a communication time period toeach group according to a centralized control scheme ofthe AP to realize the real grouping method [3]. It needsoverhead control information to support this scheme. Thesecond method realizes the grouping by a distributed con-trol scheme based on the CSMA/CA protocol. As a vir-tual grouping method, the distributed coordination func-tion with virtual group (DCF/VG) scheme is proposed [4],[5]. The DCF/VG scheme achieves good throughput per-formance even when traffic congestion occurs; however, itsuffers from a power consumption problem of STAs. Ac-cording to the DCF/VG scheme, each STA always searchesfor the allocated slot, which is composed of a channel busyperiod and an idle period. An STA can transmit its dataframe only at the allocated slot. Therefore, each STA al-ways needs to sense channel busy states in order to detectassigned slot boundaries.

To realize an M2M network composed of power-efficient STAs, this paper proposes the IEEE 802.11ah basedM2M networks employing virtual grouping and power sav-ing method. Virtual grouping is realized by the random ar-bitration interframe space (IFS) number (AIFSN) scheme toimprove the throughput performance under the condition ofmany active STAs. An active STA means that it stays at anawake state to transmit data frames. The random AIFSNscheme is proposed for the precise quality of service (QoS)control of WLANs to transmit several types of data frames[6]. We apply this method to the M2M network to realizevirtual grouping and improve the throughput performanceof a system with 6,000 STAs. Moreover, we propose a novelpower saving method using the virtual grouping method.

Most STAs in the M2M networks are supposed to bebattery-operated sensors. Therefore, power saving is a keyissue. The legacy IEEE 802.11 standard supports the powersave poll (PS-Poll) method as a power saving method. Inaddition to the PS-Poll method, the IEEE 802.11n standardsupports the power save multi-poll (PSMP) method [12].Various power saving methods for wireless sensor networksand WLANs have been proposed [7]–[11]. However, thereare few schemes for single hop networks with a large num-

Copyright c© 2013 The Institute of Electronics, Information and Communication Engineers

OGAWA et al.: IEEE 802.11AH BASED M2M NETWORKS EMPLOYING VIRTUAL GROUPING AND POWER SAVING METHODS2977

ber of STAs. Schemes proposed in [7]–[9] are intendedfor multi-hop and ad-hoc mode systems. Those proposedin [10], [11] are single hop WLANs with a small numberof STAs. For single-hop networks with a small number ofSTAs, the PS-Poll or PSMP methods are usually used to re-alize power saving.

For power saving of the STAs, the IEEE 802.11ahadopts the PS-Poll method, which enables non-active STAsto sleep except during the period when they are receivingselected beacon frames. The PS-Poll method is effectiveunder the usual light traffic conditions; however, it is in-effective under heavy traffic. An active STA that has dataframes in its transmission queue cannot enter into the sleepstate unless the queue becomes empty. Under heavy trafficconditions, STAs are likely to wait for a long period beforethe data frames are transmitted. Thus, the power consump-tion of STAs increases. In this case, the PSMP method, canconventionally be used. It is suitable for a small number ofSTAs, but not for M2M networks that have a large numberof STAs, because the PSMP frame length becomes very longowing to the large number of STAs. For example, the PSMPframe length of 6,000 STAs is estimated to be 48 kB, whichtakes approximately 1.3 s to send and receive at a transmis-sion rate of 300 kbit/s. The AP has to transmit and the STAsusually receive these extraordinarily long PSMP frames incase temporal traffic congestion occurs, which causes an in-crease in STAs’ power consumption.

We, therefore, propose a new power saving method em-ploying virtual grouping in addition to the legacy PS-Pollmethod. The proposed method enables the STAs to save ontheir power consumption in a distributed control manner.

This paper is organized as follows. Section 2 de-scribes the virtual grouping method with the random AIFSNscheme. The analysis of the virtual grouping is given inSect. 3. Section 4 provides the description of the proposedpower saving method employing virtual grouping. Sec-tion 5 evaluates the performance of the M2M network anddiscusses the effectiveness of the proposed power savingmethod. Finally, Sect. 6 presents the conclusion of this pa-per.

2. Virtual Grouping with Random AIFSN Scheme

This section describes the random AIFSN scheme and vir-tual grouping.

2.1 Random AIFSN Scheme

According to the random AIFSN scheme, an STArandomly selects its AIFSN value from the interval[AIFSNmin,AIFSNmax], prior to every data frame transmis-sion. AIFSN is an integral number, which decides the lengthof the AIFS as

AIFS time = SIFS time + AIFSN × slot time, (1)

where slot time = 52 µs [1]. AIFS is a type of IFS, whichis used in enhanced distributed channel access (EDCA) of

Fig. 1 Medium access control with the random AIFSN scheme. TheAIFSN value is randomly determined after the transmission.

the IEEE 802.11e standard [13] to realize QoS control. TheAIFS determined by Eq. (1) is expressed as AIFS(AIFSN).For example, AIFS(2) means that the AIFSN value is two. Itis equal to the period of the distributed IFS.

To explain the random AIFSN scheme the transmissionprocedure is shown in Fig. 1. In this figure, a cycle is com-posed of an idle channel period and the following busy chan-nel period. According to the scheme, each STA waits for theAIFS time and backoff time before its transmission as wellas the IEEE 802.11e channel access procedure. The trans-mission period is composed of a data frame period, a shortIFS (SIFS) period, and an ACK frame period. Two casescould occur during the transmission period; successful dataframe transmission and data frame collision.

For the installation of this method, the parameter ofAIFSNmax has to be shared among the AP and the STAs.It is managed by the AP. The AP distributes it to each STAby a data frame which includes the parameter of AIFSNmax

to prepare for traffic congestion, when the STAs associatewith the AP to start the data and control frame transaction.

2.2 Virtual Grouping

In [6], the random AIFSN scheme was applied to a systemwith a small number of STAs to realize the precise QoS con-trol for several types of data transmissions. We applied thisscheme to M2M networks with a large number of STAs. Asa result, we found that virtual grouping is feasible. Thethroughput performance is considerably improved by thevirtual grouping method, which limits the number of con-tending STAs in the case of simultaneous data frame trans-missions. We define two types of STAs. One is a contendingSTA that decreases its backoff counter value, and the otheris a non-contending STA that does not decrease its backoffcounter value during a cycle.

The flowchart of the EDCA scheme is shown in Fig. 2.As shown in the flowchart, the STA that does not decreasebackoff counter does not transmit any frames in the cycle.These events are shown as bold allows in Fig. 2. We callsuch STAs non-contending STAs. By this definition, the


Fig. 2 Flowchart of the EDCA scheme.

STAs are divided into two groups every cycle as follows:

IF (AIFS Ni > min1≤ j≤N

[AIFS Nj + backo f f j])

STAi is a non-contending STA;ELSE

STAi is a contending STA;

where STAi denotes the i-th STA, AIFS Nj is the AIFSNvalue of STA j, backoff j is the backoff counter value of STA j

and N is the total number of STAs.In Fig. 1, STA 1 decreases its backoff counter and de-

tects the channel busy state at cycle 1. STA 2 detects thechannel busy state before decreasing its backoff counter.STA 3 decreases its backoff counter to zero and then trans-mits its data frame. In this case, STA 1 and STA 3 may havetransmitted a data frame during the cycle, but STA 2 cannotbecause it have not decreased its backoff counter. Thus, dur-ing the cycle, STA 1 and STA3 are considered as contendingSTAs, and STA 2 is considered as a non-contending STA. Inthis case, STAs are divided into two groups, namely, con-tending and non-contending STAs by the random AIFSNscheme. With this scheme, virtual grouping of the STAs isrealized. Figure 3 shows the virtual grouping mechanism.After the data frame transmission, an STA obtains a newAIFSN value. If obtained AIFSN value is small, it is likelyto be a contending STA. By contrast if obtained AIFSN value

Fig. 3 Contending and non-contending STAs with virtual grouping whenrandom AIFSN scheme is employed.

is large, it is likely to be a non-contending STA.Figure 4 shows the state transition diagram of the

AIFSN distribution in the case of the random AIFSNscheme. For the sake of simplicity, AIFSNmin and AIFSNmax

are selected as 2 and 4, respectively, in this figure. Atthe first state, each STA selects its AIFSN value from theinterval [AIFSNmin, AIFSNmax]. The number of STAs foreach AIFSN value has a uniform distribution. Hereafter,we denote the STA of which the AIFSN value equals n asSTA(AIFSN = n). We also denote the STA of which theAIFSN value is greater than n as STA(AIFSN > n). Then,the state is moved to the second state where the number ofSTA(AIFSN = 2)s decreases and becomes small, whereasthe number of STA(AIFSN > 2)s increases. This transitionis done by the following procedure:

1. STA(AIFSN = 2)s are likely to be contending to trans-mit their frames.

2. After their frame transmissions, they reset their AIFSNvalues from the interval [2, 4] with the random AIFSNscheme, which causes a decrease of the number ofSTA(AIFSN = 2)s and an increase of the numberSTA(AIFSN > 2)s.

The similar transition occurs from the second state tothe third state. Finally, most of the STAs have theAIFSNmax value at the steady state. The majority be-come STA(AIFSN = AIFSNmax)s which correspond approx-imately to the non-contending STAs, the minority, but a cer-tain number, become STA(AIFSN < AIFSNmax)s which cor-respond approximately to the contending STAs. We callthis phase “reduction phase” at the steady state, since thenumber of STA(AIFSN < AIFSNmax)s decreases in thisphase. When the number of STA(AIFSN < AIFSNmax)s be-comes very small, STA(AIFSN = AIFSNmax)s also contends.Some of them with small backoff counter values transmittheir frames and reset their AIFSN values. The number ofSTA(AIFSN < AIFSNmax) increases. We call this phase“provision phase.”


Fig. 4 State transition diagram of AIFSN distribution. For the sake ofsimplicity, AIFSNmax = 4 is selected in this figure.

3. Performance Evaluation and Analysis of VirtualGrouping

The computer simulation results of virtual grouping are pre-sented in this section. To evaluate the throughput, delay, andpower consumption performance, we developed a simula-tion platform on a Monte Carlo simulation using C++ lan-guage, which emulates the IEEE 802.11 protocol as closelyas possible for real-world operation of each transmittingSTA. The simulation parameters are shown in Table 1.

3.1 Evaluation of Virtual Grouping

The evaluation of virtual grouping with the random AIFSNscheme is presented. A single-hop network, composed of

Table 1 IEEE 802.11 TGah parameters.

Parameters ValueChannel band width 1 MHzData rate 300 kbit/sMAC payload 200 BOFDM symbol 40 µsSlot time 52 µsSIFS 160 µsCWmin 15CWmax 1023

Fig. 5 Average number of contending STAs with the random AIFSNscheme (AIFSNmax = 20).

an AP and many STAs, is considered according to the IEEE802.11ah system model. For the sake of simplicity, a satu-rated traffic condition is assumed in which the transmissionqueues of all the STAs are always occupied by data frames.After a data frame transmission of an STA, the next dataframe is immediately supplied to the STA [14].

Figure 5 shows the average number of contendingSTAs with the random AIFSN scheme in the steady state.The average number of contending STAs is kept very smallwith many active STAs. In this case, most of the STAsare usually at a non-contending state. When the number ofSTAs is equal to 6,000, the average number of contendingSTAs is only 16. It enables the M2M networks to improvethe throughput performance drastically, which is operatedwith the CSMA/CA protocol under heavy traffic conditions.

Figure 6 shows the throughput performance under asaturated traffic condition versus the number of STAs. Thevalue of AIFSNmax is set 20 in order to maximize thethroughput performance with 6,000 STAs. The transmissiondata rate equals 300 kbit/s. The throughput performance ofthe conventional DCF system decreases with an increase inthe number of STAs. On the other hand, the virtual group-ing method with the random AIFSN scheme drastically im-proves the throughput performance.

When the STAs of different transmission data rates aremixed, the number of contending STAs is similarly limited;therefore, virtual grouping works effectively. The through-put performance is affected by transmission rates. However,


Fig. 6 Throughput performance under saturated traffic condition whenall the STAs transmit data frames at 300 kbit/s.

Fig. 7 Throughput performance under saturated traffic condition whenhalf of the STAs transmit data frames at 300 kbit/s and the others at150 kbit/s.

the virtual grouping scheme by the medium access con-trol employing EDCA with the random AIFSN scheme isnot affected. Figure 7 shows the throughput performancewhen half of the STAs transmit at 300 kbit/s and the oth-ers at 150 kbit/s. By including low transmission rate STAs,the throughput performance becomes less than that shownin Fig. 6; however, it is improved similarly when comparedwith that of the conventional DCF system.

Some of the STAs may have an error-prone link. Weassume that 10% of the links are error-prone, of which frameerror rate equals 0.1. Figure 8 shows the evaluation resultsof the total throughput performance. The difference in thethroughput performances shown in Figs. 6 and 8 is small.The virtual grouping method effectively works in the casewhere some error-prone links are included.

3.2 Analysis of Virtual Grouping

We described the mechanism of the virtual grouping methodin Sect. 2.2. In this section, we analyze the mechanism ofhow virtual grouping is achieved, using computer simulation

Fig. 8 Throughput performance under saturated traffic condition when10% of the STAs have the error-prone links whose average frame error rateis 0.1, and the data transmission rate is 300 kbit/s.

Fig. 9 AIFSN distribution of each state using the random AIFSN scheme(AIFSNmax = 11).

results. In this analysis, the value of AIFSNmax is 11, whichmeans that each STA chooses its AIFSN value as an integernumber in the range of 2 - 11.We assume that the number ofSTAs is 10,000.

Figure 9 shows AIFSN distribution of each state us-ing the random AIFSN scheme. At the first state, all theSTAs randomly choose their AIFSN values at the start of thecomputer simulation. Therefore, approximately 1,000 STAshave the same AIFSN value because of the range of AIFSNand the total number of STAs. The STA(AIFSN = 2)simmediately decrease their backoff counters and transmitthe data frames. They reset their AIFSN values; there-fore, the number of STA(AIFSN = 2)s decreases, and thenumber of STA(AIFSN > 2)s increases. If the number ofSTA(AIFSN = 2)s becomes small enough, STAs(AIFSN =3) acquire a transmission priority. This is the second


Fig. 10 Transition of the AIFSN distribution using the random AIFSNscheme (AIFSNmax = 11).

Table 2 Example of the steady distribution of AIFS N.

AIFSNCycle 2 3 4 5 6 7 8 9 10 1162,145 3 0 3 0 3 11 9 21 175 9,77562,146 3 0 3 0 2 12 9 21 175 9,77562,147 3 0 3 0 3 11 9 21 175 9,77562,148 4 0 3 0 3 10 9 21 175 9,77562,149 3 0 3 0 3 10 9 21 175 9,77662,150 59 57 55 54 35 59 55 70 232 9,32462,151 58 58 56 52 37 58 56 71 231 9,32362,152 56 58 56 52 37 59 56 71 232 9,32362,153 55 58 56 52 37 59 57 71 232 9,32362,154 52 58 56 52 38 59 57 71 234 9,323

state. As in the case of STA(AIFSN = 2)s, the num-ber of STA(AIFSN = 3)s decreases, and the number ofSTA(AIFSN > 3)s increases. Then, STA(AIFSN = 4)sfollow them. Finally, most STAs become STA(AIFSN =AIFSNmax)s at the steady state. This evolution process isshown in Fig. 10. The number of STA(AIFSN < AIFSNmax)sdecreases with a passage of cycles. The number of STAswith large AIFSN values decreases gradually because theymay choose smaller AIFSN values after their transmissions,and these STAs with smaller AIFSN values acquire the trans-mission priority. Then, they will have larger AIFSN val-ues due to the reset of their AIFSN values. Table 2 showsexamples of AIFSN distribution in the steady state. MostSTAs are STA(AIFSN = AIFSNmax)s at every cycle. Af-ter the 62,149th cycle, STA(AIFSN = AIFSNmax)s contend,and some of them transmit their data frames. In this case,a collision event occurs because more than one STAs trans-mit their data frames. Then, STAs(AIFSN < AIFSNmax) arenewly provided. There are two phases in the steady stateas described in Sect. 2.2. The reduction phase distributionallows the minor STAs that have smaller AIFSN values tocontend.

The number of contending STAs at the steady stateis shown in Fig. 11. The number of contending STAsstays small except at some discrete time instances. These

Fig. 11 Transition of the number of contending STAs at the steady state.The number of STAs is 10,000.

instances correspond to the provision phase in whichSTA(AIFSN < AIFSNmax)s are provided, such as the62,150th cycle in Table 2. After the provision phases,the number of contending STAs immediately gets small.The distribution of the reduction phase limits the numberof contending STAs where most STAs are STA(AIFSN =AIFSNmax)s, and there are some STAs with smaller AIFSNvalues. They prevent the majorities from contending.

4. Proposed Power Saving Method Employing VirtualGrouping

We proposes the power saving method which makes non-contending STAs, a large majority, to sleep in addition to thePS-Poll method. The conventional power saving methodswork well under light traffic conditions with a small numberof STAs. However, under traffic congestion conditions witha large number of STAs, these methods degrade the through-put and power consumption performance. The objective ofthe proposed method is to reduce the power consumptionof the system under traffic congestion conditions. The keyidea of the proposed method is to make the non-contendingSTAs sleep. The flowchart of this scheme is shown in Fig. 12for active STAs that have data frames in their transmissionqueues. First of all, they undergo a virtual grouping pro-cess with the random AIFSN scheme to transmit the dataframe as explained in Sect. 2. During the virtual groupingprocess, each active STA experiences either of the states –the contending or the non-contending state – at every cycle.Each STA counts the number of successive non-contendingstates and enters into the sleep state if the number exceeds athreshold rth. After sleeping for a certain sleep period Tsleep,the STA returns to the active state. When an active STA de-creases its backoff counter value, it stays in the active stateand tries to transmit its data frame. Each STA repeats thisprocedure until its queue becomes empty.

The proposed method realizes the power savingmethod in a distributed control manner. With a small valueof rth, it is easy for the STAs to sleep, although they arelikely to lose opportunities to transmit their data frames. The


Fig. 12 Flowchart of the proposed power saving method for active STAs.

Tsleep value is restricted by the IEEE 802.11 standard andthe characteristics of wireless sensor devices. With regardto the IEEE 802.11 standard, the STAs have to receive bea-con frames which include delivery traffic indication maps(DTIMs). Hence, Tsleep must be smaller than the beaconframe interval. The beacon interval of IEEE 802.11 basedWLANs is usually 100 ms. In the IEEE 802.11ah standard,1/10 down clocking is considered. Therefore we assume 1 sas the beacon interval. In order to receive the beacon framesfrom the AP, the STAs must wake up within 1 s. On theother hand, Tsleep must be greater than one cycle period, theaverage of which is approximately 3 ms. As a time unit forthe sleep mode, we select 100 ms as Tsleep.

In this proposed power saving method, data transmis-sions with the PS-Poll method are assumed. In this case,the STAs are usually in a sleep mode to save power con-sumption except for receiving the DTIMs that are includedin beacon frames from the AP. After receiving the DTIMs,the STAs know existence of their downlink data frames atthe AP. When an STA wakes up to request the downlinkdata frame, it transmits a PS-Poll frame to the AP. Then, theAP transmits the data frame to the STA a SIFS period aftersuccessful reception of the PS-Poll frame. This time chartis shown in Fig. 13. Therefore, only the STAs perform thechannel access procedure for the downlink data frames. Theproposed method effectively works when the AP transmitsdata frames to the STAs.

Fig. 13 Time chart of the downlink traffic channel access with thePS-Poll method.

Table 3 Power consumption parameters [15].

State ValueTransmitting 340 mWReceiving 260 mWData processing 56 mWChannel sensing 204 mWSleep 0.19 mW

5. Evaluation of the Proposed Power Saving Method

The computer simulation results of the proposed power sav-ing method are presented in this section. The simulationparameters are shown in Tables 1 and 3. In the sleep state inTable 3, some parts of the circuit are awake and the associa-tion with the AP is maintained [15].

5.1 System Description

A single-hop network with an AP and 6,000 STAs is consid-ered according to the IEEE 802.11ah system model. EachSTA is assumed to have one data frame in its transmissionqueue at the start of the simulation to evaluate the temporaryuplink traffic congestion state. In this case, as the simulationtime progresses, the traffic congestion is mitigated owing toan increase of the number of non-active STAs that have suc-cessfully transmitted their data frames. For the sake of sim-plicity, the AP transmits no data frames or beacon framesduring this period.

The system performance is evaluated using the two cri-teria. The first is the maximum delay, which is the requiredtime for all the STAs to transmit the data frames. The secondone is the consumed energy, which is the amount of energythat all the STAs consumes within this period.

5.2 Simulation Results

We shows the improved performance of the system with theproposed method and derive an optimum value for rth. Inthis subsection, Tsleep is fixed at 100 ms.

Figure 14 shows the consumed energy of the 6,000STAs versus rth. The conventional method indicates the ran-dom AIFSN scheme with the PS-Poll method for power sav-ing. This figure shows that the proposed method reducesthe consumed energy by 96% as compared with the conven-tional method at rth = 1. The consumed energy uniformly


Fig. 14 Total consumed energy of the STAs versus rth. rth is the sleepthreshold.

Fig. 15 Maximum delay time for 6,000 STAs to transmit all data framesversus rth. rth is the sleep threshold.

increases with value of rth, as the STAs are unlikely to sleepwith a large rth value. Considering only the consumed en-ergy of the STAs, rth = 1 is the best value.

Moreover, we should consider the delay performancewhen choosing the rth value. Figure 15 shows the maximumdelay time for the 6,000 STAs to transmit versus rth. Themaximum delay decreases with an increase in the rth valueand converges to the conventional value. Assuming that thepermissible increase in the maximum delay is less than 5%,rth should be greater than 1. Considering the foregoing ar-gument, rth = 2 is the optimal value. Using this value inFig. 14 the consumed energy is reduced to 4.6 kJ, which is10% of the conventional method.

We evaluate the total consumed energy of the STAsversus rth where the number of active STAs equals 100, andthe maximum delay time for 100 active STAs to transmit allthe data frames versus rth as shown in Figs. 16 and 17, re-spectively. These simulation results show that rth = 2 is theoptimal value not only in the heavy traffic condition but alsoin the light traffic condition.

Finally, we evaluate the consumed energy of the pro-

Fig. 16 Total consumed energy of the STAs versus rth. rth is the sleepthreshold. The number of active STAs is 100.

Fig. 17 Maximum delay time for 100 STAs to transmit all data framesversus rth. rth is the sleep threshold.

Fig. 18 Total consumed energy of the STAs versus number of activeSTAs.

posed method at different conditions where some STAs areactive and the others are not. Figure 18 shows the con-sumed energy versus the number of active STAs. In this


Fig. 19 Maximum delay time for all the STAs to transmit the data framesversus number of active STAs.

case, rth = 2 is selected according to the discussion previ-ously mentioned. The proposed method reduces the con-sumed energy of the STAs in all cases. On the other hand,the maximum delay time for all the active STAs to finishtheir transmissions is increased by less than 5%, as shown inFig. 19. These results show that the proposed power savingmethod can be applied to various conditions without param-eter adaptation.

For the conventional PS-Poll method, an STA requestsa downlink data frames by using a PS-Poll frame when theSTA wakes up. The PS-Poll method itself is common to boththe proposed method and the conventional method. If weassume the downlink data frames, there will be the similarincrease in maximum delay for the proposed method and theconventional method.

6. Conclusion

To realize a good throughput and power-efficient M2M net-work employing the CSMA/CA protocol, this paper has pro-posed the IEEE 802.11ah based M2M networks employingvirtual grouping and power saving methods. To solve thethroughput performance degradation caused by contentionof many STAs, this paper has shown that virtual groupingwith the random AIFSN scheme achieved a good throughputperformance. To reduce the power consumption of the STAswhen the CSMA/CA protocol is used, this paper proposedthe power saving method combined with virtual grouping. Itenabled power saving in a distributed control manner. Thesimulation results show that the proposed method reducedthe energy consumption of the STAs by 90% within a 5% in-crease in the delay, compared with the system using the vir-tual grouping method and the conventional PS-Poll method.

Acknowledgment

This work was supported in part by Grant-in-Aid for Scien-tific Research (B) (no. 21360149).

References

[1] “IEEE P802.11 task group ah — Meeting update,” http://www.ieee802.org/11/Reports/tgah update.htm

[2] IEEE P802.11ac/D2.2 Draft Std. Amendment 4: Enhancements forVery High Throughput for Operation in Bands below 6 GHz, 2012.

[3] M. Park, et al., “RAW slot assignment,” IEEE 802.11-12/1321r0,Nov. 2012.

[4] S.M. Kim and Y.J. Cho, “Performance evaluation of IEEE 802.11distributed coordination function with virtual group,” IEICE Trans.Commun., vol.E88-B, no.12, pp.4625–4635, Dec. 2005.

[5] S.M. Kim and Y.J. Cho, “A virtual grouping scheme for improvingthe performance of IEEE 802.11 distributed coordination function,”Proc. Wireless Networks and Emerging Technologies, July 2004.

[6] S. Gaur and T. Cooklev, “Introducing finer prioritization in EDCAusing random AIFSN,” Proc. TridentCom 2007, Orlando, Florida,USA, May 2007.

[7] C. fan Hsin and M. Liu, “Network coverage using low duty-cycledsensors: Random & coordinated sleep algorithms,” Proc. 3rd Inter-national Symposium on Information Processing in Sensor Networks,pp.433–442, 2004.

[8] J. Liu, X. Jiang, S. Horiguchi, and T.T. Lee, “Analysis of randomsleep scheme for wireless sensor networks,” Int. J. Sensor Networks,vol.7, no.1, pp.71–84, 2010.

[9] Y.S. Chen, M.K. Tsai, L.S. Chiang, and D.J. Deng, “Adaptivetraffic-aware power-saving protocol for IEEE 802.11 ad hoc net-works,” Proc. IEEE 17th International Conference on Parallel andDistributed Systems, pp.866–871, Dec. 2011.

[10] D. Qiao and K.G. Shin, “Smart power-saving mode for IEEE 802.11wireless LANs,” Proc. IEEE INFOCOM, pp.1573–1583, March2005.

[11] X. Hu, Z. Chen, and Z. Yang, “Energy-efficient scheduling strategiesin IEEE 802.11 wireless LANs,” Proc. IEEE Computer Science andAutomation Engineering, pp.570–572, May 2012.

[12] IEEE Std. 802.11n Amendment 5: Enhancements for HigherThroughput, 2009.

[13] IEEE Std. 802.11e Amendment 8: Medium Access Control (MAC)Quality of Service Enhancements, 2005.

[14] G. Bianchi, “Performance analysis of IEEE 802.11 distributed co-ordination function,” IEEE J. Sel. Areas Commun., vol.18, no.3,pp.535–547, March 2000.

[15] D.M. Dobkin and B. Aboussouan, Low power Wi-Fi (IEEE 802.11)for IP smart objects, GainSpan Corporation, 2009.

Kohei Ogawa received his B.E. degreefrom Kyoto University, Japan, in 2012. He iscurrently an M.E. student at Kyoto University,Japan. He has been engaged in the research ofthe protocol for Wireless M2M networks.


Masahiro Morikura received his B.E.,M.E., and Ph.D. degrees in electronics engi-neering from Kyoto University, Kyoto, Japan in1979, 1981 and 1991, respectively. He joinedNTT in 1981, where he was engaged in the re-search and development of TDMA equipmentfor satellite communications. From 1988 to1989, he was with the Communications Re-search Centre, Canada, as a guest scientist.From 1997 to 2002, he was active in the stan-dardization of the IEEE 802.11a based wireless

LAN. He received the Paper Award and the Achievement Award fromIEICE in 2000 and 2006, respectively. He also received the Education, Cul-ture, Sports, Science and Technology Minister Award in 2007 and MaejimaAward in 2008. Dr. Morikura is now a professor in the Graduate School ofInformatics, Kyoto University. He is a member of the IEEE.

Koji Yamamoto received the B.E. degree inelectrical and electronic engineering from KyotoUniversity in 2002, and the M.E. and Ph.D. de-grees in informatics from Kyoto University in2004 and 2005, respectively. From 2004 to2005, he was a research fellow of the Japan So-ciety for the Promotion of Science (JSPS). Since2005, he has been with the Graduate Schoolof Informatics, Kyoto University, where he iscurrently an associate professor. From 2008to 2009, he was a visiting researcher at Wire-

less@KTH, Royal Institute of Technology (KTH) in Sweden. His researchinterests include game theory, spectrum sharing, and M2M networks. Hereceived the PIMRC 2004 Best Student Paper Award in 2004, the EricssonYoung Scientist Award in 2006, and the Young Researcher’s Award fromthe IEICE of Japan in 2008. He is a member of the IEEE.

Tomoyuki Sugihara received his B.P. de-gree from Kyoto Sangyo University, Japan, in1985. He is currently a director of Allied Tele-sis Holdings, K.K., Japan. He has been engagedin the research & development of Ethernet andIP network products and protocols.

paper special section on network and system ......paper special section on network and system...

Documents