1

Performance Evaluation of MIMO-Aware MediaAccess Control Protocol

Dan J. Dechene,Student Member, IEEE,Khalim Amjad Meerja,Student Member, IEEE,and Abdallah Shami,Member, IEEE

Abstract—Multiple-Input and Multiple-Output (MIMO) sys-tems are recently being recognized for achieving simultaneoustransmissions on wireless channels that overlap in time. Asaresult, new scheduling mechanisms are required that coordinatethese multiple transmissions among wireless stations which sharea single wireless channel. This paper proposes a new distributedMIMO-aware media access control (MA-MAC) scheme whichwill allow two simultaneous transmissions when stations areequipped with MIMO. The proposed MA-MAC scheme is com-patible with the existing IEEE 802.11 standard for wirelesslocalarea networks (WLANs). It uses weighted nulling and intelligentpacket fragmentation for its operation. Detailed performanceanalysis of MA-MAC scheme is carried out using the NS-2network simulator. Firstly, this paper presents the concept ofweighted nulling that is necessary for simultaneous transmissionson wireless channel, followed by a detailed description of theproposed MA-MAC scheme. Secondly, it presents the perfor-mance of MA-MAC scheme under various network scenarios thatincludes both saturated and unsaturated conditions. This paperpresents the performance improvements shown by MA-MACwhen compared with an existing media access control schemewhich is also proposed for scheduling simultaneous transmissionsfor MIMO.

Index Terms—MIMO, 802.11, Beamforming, Media AccessControl, Ad Hoc Networks

I. I NTRODUCTION

W IRELESS local area networks (WLANs) are now usingMultiple-Input Multiple-Output (MIMO) antenna tech-

nology to improve overall network performance by achievingdata rates beyond100 Mega bits per second (Mbps). The initialgoal of the upcoming IEEE 802.11n WLAN standard is toat least provide the data rate of Fast Ethernet [1]. MIMOis also commonly seen in all IEEE 802.11a/b/g compatiblewireless devices due to its various advantages. MIMO hasthe ability to increase signal-to-interference-noise ratio (SINR)performance using antenna selection mechanisms which is notpossible in Single-Input Single-Output (SISO) systems. MIMOsystems also have the ability to capitalize on the scatteringeffects of the wireless channel. This allows a physical layerdesigner to utilize the rich scattering phenomena in order toimprove data link layer throughput through the use of spatialmultiplexing. In addition, spectral efficiency can be improvedby fully exploiting MIMO at the physical layer. For exam-ple, spectrum can be efficiently used through beamformingcapability of MIMO. The new IEEE 802.11n standard exploitsMIMO capabilities using space-time block codes in additionto providing optional beamforming capability [1].

Media access control (MAC) layer initially perceivedMIMO as an improved physical layer link capable of providing

higher data rates with increased reliability. The design ofMACprotocols that exploit MIMO features is largely an unexploredproblem. Since IEEE 802.11n standard endorses the usageof beamforming with MIMO, MAC layer can utilize this toimprove throughput performance of WLAN. Beamformingenables any particular wireless station to selectively tunein or tune out transmissions from other wireless stations.This will allow simultaneous transmissions on the wirelesschannel. Design of efficient MAC protocols that can schedulesimultaneous transmissions are therefore essential to exploitthe MIMO beamforming feature. Also these MAC protocolshave to be backward compatible with IEEE 802.11 stan-dard. Therefore MAC layer designers have recently startedto propose MAC protocols for MIMO systems that schedulesimultaneous transmissions with in a single collision domainto improve throughput performance in WLANs.

In this work, a MIMO-aware MAC (MA-MAC) schemeis proposed that schedules two simultaneous transmissions1.These two simultaneous transmissions that overlap in timeare scheduled within a single collision domain to improvechannel utilization. This is achieved through a newly pro-posed MAC decision process along with intelligent packetfragmentation. The proposed MA-MAC scheme is compatiblewith the IEEE 802.11 standard. The scheme is proposed fora three element antenna array MIMO system. The selectionof a small antenna size is mainly due to space limitations inmobile phones, personal digital assistants (PDAs) and laptops.

The remainder of the paper is as follows. Section II providesa review of the MAC protocols proposed in the literature forMIMO. Then the motivation for pursuing this work is given inSection III. An overview of MIMO along with the concept ofweighted nulling that is necessary for simultaneous transmis-sion is given in Section IV. Detailed description of MA-MACscheme is then given in Section V. In Section VI performanceanalysis of the proposed MA-MAC scheme is carried out usingsimulations in NS-2 network simulator. The performance ofMA-MAC is compared with SPACE-MAC protocol whichis proposed in literature for IEEE 802.11 standard. Finallyconclusions are drawn on this work in Section VII.

II. L ITERATURE REVIEW

Recently several MAC protocols are proposed in literaturefor MIMO systems. All these protocols attempt to exploitMIMO capabilities to improve throughput performance in

1Part of this work is presented in the proceedings of the 32nd IEEEconference on Local Computer Networks (LCN) 2007, Dublin, Ireland.

2

wireless networks. In this section some MAC protocols thatschedule simultaneous transmissions using MIMO are brieflysummarized along with their limitations.

A. MIMA-MAC

Mitigating interference using multiple antennas MAC(MIMA-MAC) allows multiple stations to communicate withina contention region by utilizing zero-forcing MIMO receiversin each wireless station [5]. The number of simultaneoustransmissions available in MIMA-MAC is equal to the numberof antennas per station. In MIMA-MAC, the authors haveproposed a fixed size MIMA-MAC frame that is dividedinto a contention period and a contention-free period. Duringthe contention period, wireless stations compete for channelaccess for the contention-free period. The contention periodis divided into slots for multiple contentions. The order ofchannel acquisition during the contention period determinesthe order of transmission of training and acknowledgmentpackets in the contention-free period by the wireless stations.

In the contention-free period, stations first send trainingsequences for channel estimation. As mentioned, the orderin which stations acquired the channel during the contentionslots defines the order of the training sequences. Followingthese sequences, all transmitting stations will send data packetssimultaneously. Receivers decode data from their respectivetransmitters using the channel information resolved from thetraining sequences. Following data transmission, acknowledge-ment packets are also sent by the receivers to confirm success-ful transmission of the data packets. These are transmittedinthe same order as the training sequences.

The main disadvantage with MIMA-MAC is the require-ment of a fixed MIMA-MAC frame size which is not suitablefor networks with varying packet sizes.

B. NULLHOC

NULLHOC schedules simultaneous transmissions by apply-ing gains to each antenna element at both transmitting andreceiving stations [8]. In this protocol, by designing antennaweights appropriately, any station may listen or ignore (i.e.,tune in or tune out) any other station. This allows multiplepackets to be sent over the channel simultaneously. In thisprotocol, the channel is divided into two sub-channels namelydata and control sub-channels. The control sub-channel (CC)is used to monitor traffic levels on the network while thedata sub-channel (DC) is used for data transmission. TheNULLHOC protocol utilizes a five packet exchange sequence(RTS/CTS/DS/DATA/ACK). Request-To-Send and Clear-To-Send (RTS/CTS) packets are sent on the control channel.If these stations can successfully exchange these packets,then the transmitting station follows with a Data-Send (DS)packet on the data sub-channel to reserve the channel resource.This is followed by the data packet transmission by thetransmitting station and positive acknowledgement packetbythe receiving station on the data sub-channel. By exchangingantenna weights in the control channel, NULLHOC supportsmultiple transmissions.

The major limitation in NULLHOC is the need for channelpartition. This protocol imposes hardware complexity restric-tions. Although channel estimation is performed on the con-trol sub-channel, the channel information required for tuningshould be found for the data sub-channel which is impractical.The operation of NULLHOC is also not compatible with IEEE802.11 standard.

C. SPACE-MAC

SPACE-MAC also utilizes antenna weights to schedulesimultaneous transmissions on a single collision domain [9].However in SPACE-MAC, a wireless station uses the sameadjusted weights for both transmission and reception. As inNULLHOC, antenna weights are exchanged via control pack-ets (RTS and CTS). Stations should always transmit packets(including control packets) without interfering with existingactive transmissions. In SPACE-MAC, the first station thatgains access to the channel determines the silence period. Allother stations must remain idle following their transmissionuntil the completion of the silence period. In SPACE-MAC,the silence period is required because any station currentlyinvolved in transmission is unaware of any other transmissionsthat began during its data packet or acknowledgment packettransmission phase. Additionally, any station that wishestotransmit must not interfere with this ongoing transmissionaswell as not transmit if it cannot complete its entire packetexchange sequence before the end of the silence period.

The performance of the SPACE-MAC protocol is heavilydependant on the length of the silence period. The optimallength of silence period varies with the network size and thetraffic conditions. This is a major limitation with SPACE-MACas suboptimal silence periods drastically reduce the maximumachievable throughput for a particular network scenario.

D. Antenna Saturation

Examining these weighted MIMO MAC protocols (suchas NULLHOC and SPACE-MAC), it is easy to see thatthese protocols reach saturation quickly in terms of the gainassociated with an increasing number of antennas. In orderto provide proper antenna spacing, the maximum number ofantennas that can be supported in consumer devices such aslaptops and PDAs is limited. As such, in this work we consideronly a three antenna element MIMO system capable of achiev-ing two simultaneous transmissions. Furthermore, in manypractical networking scenarios, the number of stations withina given network are limited so that any additional transmissionresources may not be fully exploited and therefore reduce therequirement of additional antennas.

III. M OTIVATION

Though all the mentioned MAC protocols in the previoussection schedule simultaneous transmissions over a singlecollision domain, only SPACE-MAC is compatible with theIEEE 802.11 standard. But the maximum achievable through-put in SPACE-MAC is dependant on the optimal length ofsilence period. This optimal silence period varies with differentnetwork scenarios and traffic conditions.

3

For example, consider a network with20 wireless stationslocated closely to each other to share a single collision domain.All these stations are identical and use SPACE-MAC protocolfor scheduling their transmissions. It is assumed that MIMOphysical layer offers1 Mbps bandwidth to the MAC layer.With this assumption, the throughput performance of SPACE-MAC for two different packet sizes (512 and 1024 bytes)is shown in Figure 1. The length of the silence period isvaried to observe achievable throughput in SPACE-MAC. Thescheme achieves roughly up to a maximum of1.3 Mbpsoverall throughput when the packet size is1024 bytes. Thismaximum throughput is achieved only for a particular lengthof silence period which is optimal for this scenario. For anyother silence periods, the throughput is reduced considerably.The same phenomenon can be observed when the packet sizeis changed to512 bytes. When the size of packets is512bytes, SPACE-MAC barely achieves1 Mbps throughput foroptimal silence periods. Therefore, it is quite obvious that thethroughput performance of SPACE-MAC is highly dependanton the length of silence period.

Furthermore, optimal silence period for SPACE-MAC alsodepends on the contention level and dynamics of the networktraffic. This creates restrictions for designers in choosing anoptimal silence period for nondeterministic network scenar-ios. As the dependency of SPACE-MAC on silence periodis undesirable, more robust MAC protocols for MIMO areessential than SPACE-MAC. The primary motivation is toeliminate such dependance on silence period. The proposedMAC protocol should be compatible with IEEE 802.11 andcapable of scheduling simultaneous transmissions similartoSPACE-MAC. With all these considerations, a new MIMO-aware MAC (MA-MAC) scheme is proposed to not onlyimprove throughput but also delay performance in WLANs.

IV. MIMO S YSTEMS

A. MIMO Overview

The possibility of having more than one transmission ata time on the wireless channel is due to the MIMO offeredbeamforming feature. Also MIMO systems generally offerbetter performance compared to Single-Input Single-Output(SISO) systems [11]. Assuming that each station on a WLANhasM antennas and by neglecting additive noise, the receivedsignal at antenna elementj of a wireless station is given by

r(t)j =

M−1∑

i=0

s(t)ihij (1)

wheres(t)i is the signal transmitted fromith antenna elementon the transmitting wireless station andhij describes theamplitude and phase distortion between theith transmittingantenna element on the transmitting station and thejth antennaelement on the receiver station. More generally, the aboveequation can be written in a more condensed form as follows

−−→

r(t) =−−→

s(t)HH (2)

where−−→

r(t) and−−→

s(t) represent the signal vectors across allreceiver and transmitter antennas respectively andH is theM × M MIMO channel matrix with elementshij .

0 0.1 0.2 0.3 0.4 0.5 0.6 0.70.5

0.6

0.7

0.8

0.9

1

1.1

1.2

1.3

1.4

Silence Period % of (RTS+SIFS+CTS+SIFS+DATA+SIFS+ACK)

Thr

ough

put (

Mbp

s)

512 Byte Packets1024 Byte Packets

Fig. 1. Throughput vs. Silence Period - 20 Stations - 1Mbps Channel DataRate

A designer may exploit MIMO by using spatial multiplex-ing, spatial diversity and/or beamforming by pre and/or post-processing of the transmitted/received signal. The proposedwork in this paper exploits MIMO by applying gains toeach antenna element on the transmitting station for pre-processing similar to the work carried out in NULLHOC andSPACE-MAC. The antenna weights are applied post-receptionat the receiver wireless station, again similar to NULLHOCand SPACE-MAC. Denoting this vector of antenna gains (orweights) as−→wn, where n denotes the index of a particularstation, the output of an antenna array is

−−−→

sP (t) = s(t)−→wn (3)

wheres(t) is data signal to be transmitted and−−−→

sP (t) is trans-mitted vector by the MIMO antenna array of the transmittingwireless station. Using (2), the received signal vector at areceiving wireless stationm can be expressed as

−−−→

rP (t) = s(t)−→wnHHnm (4)

whereHnm is the MIMO channel matrix between transmittingstation n and receiving stationm, and

−−−→

rP (t) represents thesignal at the input of the receiver station MIMO antenna array.The superscriptH denotes the hermitian operation.

In this work assumption is made that the MIMO channelmatrix can be estimated accurately using pilot symbols thatareembedded in control packet headers. Also the channel betweenany two stations is assumed to remain static for the durationof one frame exchange sequence (RTS/CTS/DATA/ACK) inIEEE 802.11 distributed coordination function (DCF). Alsothe channel is assumed to be reciprocal, i.e.,Hmn = HH

nm.Finally, it is assumed that the receiver wirless station appliesits antenna weights to the received signal

−−−→

rP (t) in a mannerthat generates an overall received signal as

r(t) =−−−→

rP (t)−→wm = s(t)−→wnHHnm

−→wm (5)

where−→wm is the weights applied by the receiver station. The

4

antenna weight vectors−→wn and−→wm in the above equation arealways normalized to unity.

Using (5) along with the knowledge ofHnm, transmittingand receiving stations can designwn andwm in such a mannerto produce an overall complex gain across the channel. Oncontrary receiving stations can choose weights to tune outtransmissions from particular transmitting wireless stations.This can be achieved as follows

−→wnHHnm

−→wm = 0 (6)

B. Weight Adjustment

Stations can selectively tune in or tune out a particulartransmission from a station by properly adjusting their antennaweights. The transmitting stations on the WLAN can eithertransmit when the wireless channel is absolutely idle or whensome transmissions are already in progress. Therefore eachtransmitter-receiver pair would face interference from othertransmitter-receiver pairs. In order to limit the interference(for proper data transmission on the network), only a limitednumber of simultaneous transmissions are allowed at anyparticular time instance in a WLAN. Therefore, with anMelement MIMO system on each wireless station, the numberof simultaneous transmissions on the channel are limitedto M+1

2. This is necessary because each existing transmitter-

receiver pair consumes two degrees of freedom. Since thiswork considers a three element antenna array MIMO system, itis possible to achieve two simultaneous transmissions within asingle collision domain. For transmission on wireless channel,the stations adjust their weights during one of the followingtwo situations: 1) when the channel is absolutely idle, and 2)when the channel has existing transmissions.

1) Idle Channel Weight Adjustment:Wireless stations thatinitiate transmission when the channel is idle have the flexibil-ity to adjust their antenna weights to obtain the best possiblesignal-to-interference-noise ratio (SINR). For this a transmit-ting wireless station uses default antenna weights to initiateits transmission. The intended receiver station then adjusts itsantenna weights to maximize the SINR and responds to thetransmitter to allow for transmission to proceed. The receiverstation adjusts its antenna weights as follows

−→wR = (−→wTHHTR)H (7)

In order to design−→wR, the receiver station must be awareof both the channel matrixHTR and the transmitter antennaweight vector−→wT .

The weight adjustment in this scenario can be understoodfrom a simple ad hoc network scenario shown in Figure 2.The ad hoc network has six stations, Stations A, B, C, D, Eand F. Suppose that the channel is idle and Station A is thefirst station to transmit on the channel. Assuming that StationB is the intended receiver for Station A, Station B adjusts itsantenna array element weights using (7). Other stations on thechannel tune out the transmissions from both Station A andStation B using (6).

2) Busy Channel Weight Adjustment:Wireless stations canalso initiate a second transmission while the channel is busywith an ongoing data transmission. The stations wishing to

A

B

C D

E

F

Fig. 2. Ad hoc network scenario

initiate the second transmission should not interfere withtheexisting transmission. Therefore, they have to adjust theirantenna weights accordingly. For example, again consider thescenario shown in Figure 2. Assuming Station A and Station Bare already transmitting to each other, the rest of the stationsshould not interfere with Station A and Station B. Supposingthat Station C initiates a second transmission on the wirelesschannel to send data to Station D, it should adjust its weightssuch that

−→wAHHAC

−→wC = 0 and−→wBHHBC

−→wC = 0 (8)

Defining hxy = −→wxHHxy, the above equation can be ex-

pressed ashAC

−→wC = 0 andhBC−→wC = 0 (9)

Then the weights of Station C are adjusted as

−→wC = [hHAC hH

BC hHDC ]−1[0 0 1]H (10)

Since Station C is not currently aware of the channel orantenna weights used by station D, it assumeshDC as equal to[1 1 1]. Station D however will be able to properly adjustits antenna weights as it will know the channel informationfrom all the stations currently transmitting on the channel(Station A, Station B and Station C) as well as their antennaweights. Station D will adjust its antenna weights as follows

−→wD = [hHAD hH

BD hHCD]−1[0 0 1]H (11)

V. MA-MAC P ROTOCOL

The proposed MIMO-aware MAC (MA-MAC) schemein this section utilizes the beamforming feature in MIMOto schedule multiple transmissions on the wireless channel.Wireless stations adjust their antenna weights to selectivelytune in or tune out a particular transmission as governedby MA-MAC. The proposed scheme uses a three elementantenna array as MIMO physical layer to schedule up to

5

(a) RTS Frame Format

(b) CTS Frame Format

Fig. 3. Access Control Packets

two simultaneous transmissions in a single collision domain.This is achieved through a newly proposed MAC decisionprocess along with intelligent packet fragmentation. MA-MACutilizes the request-to-send/clear-to-send (RTS/CTS) accessmechanism used in the IEEE 802.11 distributed coordinationfunction (DCF). The antenna weights are conveyed throughRTS and CTS packets. The following subsections presentdetailed description of MA-MAC.

A. RTS/CTS Control Packet Format

For stations to selectively tune in or tune out a particulartransmission, they have to be aware of the antenna weights thatare in use by transmitting stations. This requires a mechanismfor conveying the antenna weights to all neighboring stations.MA-MAC uses RTS and CTS control packets to conveyantenna weights. The proposed format for RTS and CTScontrol packets is shown in Figure 3. A separate12 byte fieldis inserted in the payload of the RTS and CTS packets thatstores three antenna element weights currently in use. RTS andCTS packets are also used to perform channel estimation usingpilot symbols embedded in the physical (PHY) preamble.

B. Protocol Operation

The MA-MAC protocol allows up to two simultaneoustransmissions to proceed in a single collision domain. Wirelessstations in MA-MAC adjust their antenna weights and takeappropriate scheduling decisions depending on the channelstatus, i.e., depending on whether the channel is absolutelyidle or a transmission is already taking place on the channel.The network scenario given in Figure 2 will be used again forthe description of the proposed MA-MAC scheme.

Initially, all stations observe the channel for a certaintime which is the sum of partial weight sensing period anddistributed inter frame space (WSPpartial + DIFS). The timeWSPpartial is a small period for which a station can beassured that there are no active transmissions on the channeland is equal to the sum of short inter frame spacing andthe amount of time to send a special SPsrc packet (SIFS +SPsrc), while DIFS is a duration defined by the 802.11 DCFspecifications. If the channel is found idle for this time, stationsbegin decrementing their backoff counter. When a station’sbackoff counter expires, it sends an RTS packet containing itscurrently used antenna weights. This is shown in Figure 4,where Station A sends an RTS packet to Station B. Uponreception of RTS, all neighboring stations which are not theintended receivers (i.e., Stations C, D, E and F) store theduration of the transmission as well as the antenna weightsused by Station A. The destination station (Station B), receivesthe RTS packet and adjusts its antenna weights to maximizethe received signal SINR (as described in Section IV). Inaddition, this Station B responds with a CTS packet containingweight information. Stations A and B will now become theprimary stations as they have successfully exchanged RTS andCTS packets between them. Station A will now proceed withDATA packet transmission to Station B which will in turnrespond with positive acknowledgement (ACK) packet.

The primary stations transmit using the packet exchangesequence similar to the RTS/CTS scheme in DCF. When thepacket exchange sequence between the primary stations (e.g.,Stations A and B) is completed, they select a new weightvector and observe the channel for a certain time equal toWSPpartial. If the channel is found idle during this time,the stations continue to observe the channel to be idle foradditional DIFS time before resuming the backoff process.

Once a primary transmission has been established, otherstations may compete for the remaining channel resource.After the primary transmission is established and other stationsresume their backoff process which is shown in Figure 4.Assuming that Station C is the next station to complete itsbackoff process, it initiates a secondary transmission withoutinterfering the ongoing primary transmission provided it isefficient to do so. The station must perform a proper MACdecision process to determine what actions must be takenbased on the residual time remaining in the primary trans-mission (shown byΩ in Figure 4). Assuming that Station Cis successful in establishing transmission, the time takentocomplete a full frame exchange is given by the expression

RTS+ SIFS+ δ + CTS+ SIFS+ δ + DATA + SIFS+ δ + ACK

whereδ is the propagation delay experienced. In this case thereare four unique cases that the secondary station (Station C)can encounter based on the residual timeΩ if it beginstransmission, which are:

1) A secondary station will complete its full transmissionbefore the primary finishes.

2) The secondary station will be transmitting an ACKpacket when the primary transmission finishes.

3) The secondary station will be transmitting its DATApacket when the primary transmission finishes.

6

Fig. 4. MA-MAC scheme timing diagram.

(a) Scenario 1. (b) Scenario 2. !"" #$%$%&&'& ( !"" !)*$&% !"" %&#)+&,-. /012 /304

5/6789 :;< 789 =:> %&&'&? @&A#A!*B%!&##C1DEFG C1DEH

(c) Scenario 3.

IJKJL IMNOIPQ

ROSTUVWXUYJ JQZOMNOOMNO L[\]^__ `a[baQKOcKO Q de_eb` fJQde\g g hib^\e``

NbeejekJlm cKO QKOOMNONbeeje

Nbeejece`noece`noe

(d) Scenario 4.

Fig. 5. Timing diagram of MAC decision process for secondarystations.

4) The secondary station will be still performing a controlpacket exchange (RTS/CTS) when the primary transmis-sion finishes.

For each of the above cases, the secondary station makes aproper MAC decision to govern its transmission. For all casesthe timing diagrams are shown in Figure 5. The details ofoperation for each case is given below.

1) Scenario 1:When a secondary station determines thatthe residual time in the primary transmission is sufficient tocomplete the entire packet exchange, the station proceeds withits transmission. This is when

Ω >RTS+SIFS+δ+CTS+SIFS+δ

+DATA+SIFS+δ+ACK+δ(12)

where all quantities define the time taken to send the particularpacket, SIFS is short inter-frame spacing defined by the 802.11specifications andδ is the propagation delay. DATA denotesthe data packet transmission time of a secondary stationwishing to transmit.

In this case Station C proceeds with transmission usingantenna weights designed not to interfere with the primarystations (Station A and Station B) as shown in Figure 5(a).As mentioned before these weights are included in the RTSpackets in order to disseminate this information to all otherstations. Upon reception of this transmission, the intendedreceiver (i.e., Station D) readjusts its antenna weights tomaximize reception from Station C while still tuning out the

7

Fig. 6. Pseudocode for MA-MAC Operation

primary stations. Station D will subsequently respond withaCTS packet containing its antenna weights. This transmissionis now active and referred to as the secondary transmission.At this time, all other stations in the network update and storethe information (such as antenna weights and transmissionduration). Since it is not possible to accommodate more thantwo simultaneous transmissions, the other stations set theirnetwork allocation vector (NAV) to the expiration of theearliest transmission.

2) Scenario 2: Alternatively, if the secondary station de-termines the primary transmission completes during its ACKtransmission, then that station proceeds with this transmissionaccordingly. This is when

RTS+SIFS+δ+CTS+SIFS+δ+DATA+SIFS+δ

< Ω <

RTS+SIFS+δ+CTS+SIFS+δ+DATA+SIFS+δ+ACK+δ

(13)

Since the duration of ACK is small compared to thetransmission duration of the entire DATA packet exchangesequence, it is efficient to start the secondary transmission.In this case the responsibility is given to primary stationstoperform collision avoidance. The primary stations observethechannel for time equal to WSPpartial to identify any ongoingtransmission. When the channel is identified as busy and theprimary stations cannot decode a specific packet, they deferfor

ACK − WSPpartial (14)

This guarantees completion of the secondary transmission.

For fairness, all other observing stations set their networkallocation vectors (NAVs) to the duration of the primarytransmission plus the duration of an ACK. The timing diagramfor this is shown in Figure 5(b).

3) Scenario 3: When a primary transmission finishes,the primary stations are unaware of the ongoing secondarytransmission. To overcome this, secondary stations which arecurrently in the DATA packet transmission phase are re-quired to relay information regarding their current transmission(including duration and antenna weights) using the weightsensing period (WSP). The secondary stations (i.e., Station C,Station D) convey the information regarding secondary trans-mission to the primary stations (i.e., Station A, Station B)using coordinated intelligent packet splitting. This situationoccurs when

RTS+SIFS+δ+CTS+SIFS+δ

< Ω <

RTS+SIFS+δ+CTS+SIFS+δ+DATA+SIFS+δ

(15)

In this case, the secondary stations (i.e., Station C, Station D)perform the weight sensing procedure:

• At the instant the primary transmission completes, thesecondary station transmitter (Station C) halts transmis-sion of the DATA packet.

• After SIFS time, this station sends a SPsrc packet con-taining antenna weights, transmission duration and pilotsymbols for channel estimation.

• Following SPsrc and a small time to account for propa-gation delay, the secondary receiving station (Station D)

8

(a) SPsrc

(b) SPdst

Fig. 7. Weight Exchange Period packet formats

sends a SPdst packet containing antenna weights and pilotsymbols for channel estimation.

• Again after SIFS time, the secondary transmitter attachesa short PHY header to the remaining portion of the DATApacket (now referred to as DATAfrag,2).

• At this time, the secondary stations are now referred toas primary stations and are governed by MAC operationfor primary stations.

The timing diagram for WSP is shown in Figure 5(c).The packet formats for both SPsrc and SPdst are shown inFigures 7(a) and 7(b) respectively. Both packets contain theweights currently used by the secondary stations as well aspilot symbols embedded in the PHY preamble for channelestimation. Furthermore, SPsrc contains the duration of theremaining portion of transmission.

4) Scenario 4:Sometimes the residual timeΩ is insuffi-cient to complete a successful RTS/CTS exchange, meaning:

Ω < RTS+SIFS+δ+CTS+SIFS+δ (16)

In this case the station does not send an RTS packet andalternatively defers to the end of the primary transmissionaswell as increment its backoff counter. This action is taken asthere is insufficient time to establish a successful transmission(perform RTS/CTS packet exchange). The pseudo-code forMA-MAC scheme is given in Figure 6.

VI. SIMULATION RESULTS

This section provides detailed simulation results of the MA-MAC protocol. The results are presented for both saturated andunsaturated conditions under different network scenariosandnetwork parameters. Simulations of MA-MAC and SPACE-MAC are carried out using NS-2 network simulator [12]. Eachtime the performance of MA-MAC is compared with SPACE-MAC. Since SPACE-MAC performance relies heavily on itssilence period, in each simulation the SPACE-MAC silenceperiod is optimized before comparing its performance withthe proposed MA-MAC. The standard simulation parametersused throughout for both MA-MAC and SPACE-MAC aresummarized in Table I unless otherwise specified.

TABLE ISIMULATION PARAMETERS.

Parameter Value

SIFS 10µsDIFS 50µsSlot Time 20µsSPsrc 154 bitsSPdst 138 bitsDATAfrag2 PHY header 42 bitsWSPpartial 164µsWSP 360µsPropagation delay:δ 6µsData Rate 1MbpsCWmin 32CWmax 1024RTS (MA-MAC) 56 bytesCTS (MA-MAC) 50 bytesRTS (SPACE-MAC) 58 bytesCTS (SPACE-MAC) 52 bytes

A. Saturated Performance

1) Overall Improvements:To analyze the performance ofMA-MAC and SPACE-MAC schemes under saturated condi-tions, a network scenario is created with stations that sendfixed size packets at a constant bit rate equal to the maximumdata rate offered by the MIMO physical layer (i.e.,1 Mbps -as assumed). This forces the stations to always be in saturatedcondition as they always have packets to send in their buffer.In these simulations, all the stations are located within thetransmission range of each other and therefore exist in asingle collision domain. The performance of both MA-MACand SPACE-MAC is shown for varying number of stations inorder to provide an overall comparison of saturated throughputfor these protocols. Figure 8 shows the performance undersaturated conditions for a varying number of stations. Itcan be seen that MA-MAC performs better than the bestperformance achievable with SPACE-MAC. Figure 8(a) showsthat the achievable overall throughput using SPACE-MAC isonly around1.3 Mbps, whereas MA-MAC achieves morethan 1.5 Mbps in overall throughput. Though delay valuesunder saturated conditions have little significance, the averagedelay experienced by the transmitted data packets in MA-MACscheme is smaller compared to SPACE-MAC for all networksizes (see Figure 8(b)).

2) Effect on Network Parameters:The same saturated net-work scenario is utilized to understand the effect of the variousnetwork parameters on the performance of MA-MAC. In thiscase individual parameters are varied to study the effect onthe overall network throughput.

Figure 9 shows the performance of both schemes for varyingpacket sizes and presented for various networks sizes. Thethroughput of both protocols increases with respect to packetsize. However as the network size increases, the throughputis slightly reduced for both protocols. This is a result of thegreater number of collisions experienced in the dense network.SPACE-MAC experiences the same trend as MA-MAC as afunction of network size, however the overall throughput isless than that is achievable using MA-MAC. This is due tothe presence of large silence periods in SPACE-MAC.

9

4 6 8 10 12 14 16 18 201

1.1

1.2

1.3

1.4

1.5

1.6

Network Size (Number of Stations)

Sat

urat

ion

Thr

ough

put (

Mbp

s)

MA−MACOptimized SPACE−MAC

(a) Saturation throughput versus number of stations in network.

4 6 8 10 12 14 16 18 201000

2000

3000

4000

5000

6000

7000

Del

ay (

ms)

Network Size (Number of Stations)

MA−MACOptimized SPACE−MAC

(b) Average packet delay versus number of stations in network.

Fig. 8. Comparison of MA-MAC and SPACE-MAC schemes under saturated conditions.

In Figure 10 the packet size is fixed to1024 bytes, and thesize of the minimum contention window is varied for bothprotocols. The maximum number of backoff stages is alsoset to6. It is observed that the contention window size thatoffers the maximum throughput depends largely on the numberof stations in the network. For MA-MAC, the maximumthroughput is achieved when the minimum contention windowis set to64 with a network size of10 stations, however for a50 station network, the window size must be512 to achievethe best performance. Furthermore, increasing the windowfurther beyond the maximum point causes rapid degradation inthroughput. Varying the window size for SPACE-MAC revealsa major limitation in the protocol operation. As shown inFigure 10, there is a rapid reduction in throughput once athreshold window size is reached based on the number ofstations in the network. This occurs as the residual time inSPACE-MAC is small with such large contention windows,such that frequently stations cannot initiate a secondary trans-mission due to the restriction imposed by the SPACE-MACsilence period.

Next the effect of the maximum number of backoff stageson network throughput is studied. For this the minimumcontention window size is fixed to32 and the maximumnumber of backoff stages is varied. Figure 11 shows theresults of these simulations. For10 stations using MA-MAC,the change in throughput as a function of backoff stage isnegligible. This is due to the low number of stations involvedin collision. For 20 and 50 station scenarios, it is observedthat for low backoff stages, the throughput suffers degradation.At approximately a maximum backoff stage value of6, thethroughput gain associated with any additional increases isnegligible. SPACE-MAC experiences similar trends as thenumber of backoff stages increases with respect to MA-MAC,however achieves lower aggregate throughput.

Finally the effect of a varying number of stations on networkthroughput is shown. In this case, three sets of values are used

400 600 800 1000 1200 1400 1600 1800 2000 22001

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

Packet Size (bytes)

Thr

ough

put

MA−MAC − n=10MA−MAC − n=20MA−MAC − n=50Optimized SPACE−MAC − n=10Optimized SPACE−MAC − n=20Optimized SPACE−MAC − n=50

Fig. 9. Performance vs Packet Size

16 32 64 128 256 512 1024

0.6

0.8

1

1.2

1.4

1.6

Thr

ough

put

Initial size of the backoff window (W)

MA−MAC − n=10MA−MAC − n=20MA−MAC − n=50Optimized SPACE−MAC − n=10Optimized SPACE−MAC − n=20Optimized SPACE−MAC − n=50

Fig. 10. Performance vs Window Size

10

0 1 2 3 4 5 6 7 80.9

1

1.1

1.2

1.3

1.4

1.5

1.6

Maximum Backoff Stage (m)

Thr

ough

put

MA−MAC − n=10MA−MAC − n=20MA−MAC − n=50Optimized SPACE−MAC − n=10Optimized SPACE−MAC − n=20Optimized SPACE−MAC − n=50

Fig. 11. Performance vs Number of Backoff Stages

5 10 15 20 25 30 35 40 45 501.25

1.3

1.35

1.4

1.45

1.5

1.55

1.6

Number of Stations

Thr

ough

put MA−MAC − W=32, m=3

MA−MAC − W=32, m=5MA−MAC − W=128, m=3Optimized SPACE−MAC − W=32, m=3Optimized SPACE−MAC − W=32, m=5Optimized SPACE−MAC − W=128, m=3

Fig. 12. Performance vs Number of Stations

Fig. 13. Unsaturated network scenario.

400 600 800 1000 1200 1400 1600 1800 2000 22006

7

8

9

10

11

12

13

14

15

16

Downlink Packet Size (bytes)

Ave

rage

Ove

rall

Pac

ket D

elay

(m

s)

MA−MACOptimized SPACE−MAC

Fig. 14. Total average delay versus downlink packet size.

TABLE IITHROUGHPUT AND DELAY EXPERIENCED BY EACH STATION

(a) 6 Stations

Station ID Throughput (kbps) Delay (ms)

1 259.41 15712 256.39 15893 254.22 16034 259.29 15715 246.83 16516 249.10 1635

Standard Deviation 8.05 33.38

(b) 12 Stations

Station ID Throughput (kbps) Delay (ms)

1 128.12 31822 128.75 31623 123.04 33124 122.49 33255 131.79 30936 128.45 31707 125.06 32618 127.65 31949 119.44 341310 129.16 314611 121.90 334212 126.22 3225

Standard Deviation 3.65 95.38

for the contention window size and number of backoff stages.The results are shown in Figure 12. It can be observed thatthe throughput of both protocols reduces for an increasingnumber of stations. The window size of128 offers the highestthroughput for a large number of stations. It can be observedthat when there are a small number of stations, there aremany wasted idle slots causing a reduction in throughput. Fora window size of128 and a maximum backoff stage of3,SPACE-MAC however experiences a slight gain in throughputunlike MA-MAC. This is due in part to the modification ofthe silence period parameter in SPACE-MAC to allow it toachieve maximum throughput.

11

6 8 10 12 14 16 18 20 22 24 2635

40

45

50

55

60

65

Network Size (Number of Stations)

Thr

ough

put p

er S

tatio

ns (

kbps

)

MA−MACOptimized SPACE−MAC

(a) Throughput per station versus total number of stations in network.

6 8 10 12 14 16 18 20 22 24 260

500

1000

1500

2000

2500

3000

3500

4000

4500

5000

Network Size (Number of Stations)

Del

ay (

ms)

MA−MACOptimized SPACE−MAC

(b) Average packet delay of each station versus number of stations in network.

Fig. 15. Throughput and delay performance comparison of MA-MAC and SPACE-MAC.

3) Fairness:The fairness of the MA-MAC protocol is alsostudied thoroughly for numerous network sizes and underdifferent scenarios. For clarity the results are presentedforonly 6 and 12 stations scenarios. The throughput and delayobserved for each station in the scenario is tabulated tocompare fairness. In this scenario, all the stations are undersaturated conditions and send1024 bytes packets. All otherparameters are found in Table I. These results are presentedin tabular form in Tables II(a) and II(b) respectively. Fromthetables, it can be observed that the proposed MA-MAC schemeprovides a reasonable degree of fairness to all the wirelessstations on the channel.

B. Performance under Unsaturated Conditions

The performance of both protocols is also studied underunsaturated conditions. For this study, the networking scenarioshown in Figure 13 is created. All these stations are locatedin a single collision domain and have a queue length of50packets. Bi-directional transmission is enabled between thewireless stations as shown in the diagram. The transmissionsare denoted as uplinks and downlinks for identifying thedirection of data transmission between stations. For example,Station A uses downlink to transmit to Station B while StationB uses uplink to transmit to Station A. All wireless stationsare transmitting at64 kbps for both types of links.

1) Effect of Link Packet Size:Initially the effect of varyingthe data packet size is examined. For this, stations transmitat a constant bit rate. Uplinks transmit512 byte packetswhile the size of downlink packets is varied to study theeffect on network performance. The average throughput perstation for both MA-MAC and SPACE-MAC are shown inFigure 14. From this, it is observed that the overall averagepacket delay experienced in the network increases linearlyasthe data packet length of the downlink is varied. The averagedelay experienced per packet for MA-MAC is approximately20 − 25% less than that offered by SPACE-MAC.

2) Effect of Increasing Number of Stations:Next, theeffect of increasing the number of stations in the networkis examined. To study this effect on throughput and delayperformance for both schemes, more wireless stations areadded to the network scenario given in Figure 13. In thiscase, instead of having only three pairs (i.e.,6 stations intotal), the number of station pairs is increased on the network.Additionally, both types of links fix packet sizes to512 bytes.The throughput and delay performance of each station underMA-MAC and SPACE-MAC for an increasing number ofstations is shown in Figure 15. MA-MAC performs better thanSPACE-MAC as the network size increases. Also it can beseen that SPACE-MAC reaches saturation earlier than MA-MAC. When the number of stations in the network exceeds14 for SPACE-MAC, the average throughput per station beginsto drop whereas with MA-MAC, the network does not reachthese saturated conditions until approximately18 stations.

3) Packet Delay Distribution:Finally the distribution ofpacket delay is examined for the unsaturated network scenario.The previously used unsaturated network scenario is used with10 stations. The number of stations for this scenario is chosenafter studying the results presented in Figure 15 as at thisnetwork size all stations achieve64 kbps throughput usingboth MA-MAC and SPACE-MAC. The probability densityfunction (PDF) of the delay performance for both MA-MACand SPACE-MAC is shown in Figure 16. From this it can beobserved that the variation in delay experienced by packetsinMA-MAC is less than that experienced in SPACE-MAC.

PDF analysis is also performed for other network sizes inunsaturated conditions to verify the performance of MA-MAC.In all cases the delay performance achieved by MA-MAC isbetter than that achievable with SPACE-MAC.

VII. C ONCLUSION

In this paper a new MA-MAC scheme is proposed for threeelement antenna array MIMO systems. The proposed scheme

12

6 7 8 9 10 11 12 13 14 150

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Delay (ms)

PD

F

MA−MACOptimized SPACE−MAC

Fig. 16. PDF for packet delay. Network contains10 stations.

schedules two simultaneous transmissions at any instant oftime in a distributed manner. MA-MAC is compatible withIEEE 802.11 standard similar to SPACE-MAC. The perfor-mance of both MA-MAC and SPACE-MAC is studied underboth saturated and unsaturated conditions. It is observed thatMA-MAC achieves between 20 - 25% improvement comparedto optimized SPACE-MAC.

For the future work, the effect of increasing the number ofantennas in terms of antenna saturation will be studied. TheMA-MAC scheme will be used as the basis scheme to supportinfrastructure networks.

ACKNOWLEDGMENT

We would like to thank Dr. Serguei Primak for his valuableinsight and suggestions in pursuing this work.

REFERENCES

[1] Y. Xiao, “IEEE 802.11n: Enhancements for Higher Throughput inWireless LANs,” IEEE Wireless Communications Magazine, vol. 12,no. 6, Dec. 2005.

[2] J. Mirkovic, G. Orfanos, H. J. Reumerman, and D. Denteneer, “A MACProtocol for MIMO Based IEEE 802.11 Wireless Local Area Networks,”IEEE WCNC, 2007.

[3] M. Hu and J. Zhang, “MIMO Ad Hoc Networks with Spatial Diversity:Medium Access Control and Saturation Throughput,”43rd IEEE Conf.on Decision and Control, 2004.

[4] D. Zheng and J. Zhang, “Channel-Aware Weighted Proportional FairMedium Access Control in Wireless LANs with MIMO Links,”IEEEQShine’05, 2005.

[5] M. Park, C. Soon-Hyeok, and S. M. Nettles, “Cross-Layer MAC Designfor Wireless Networks Using MIMO,”IEEE GLOBECOM, 2005.

[6] M. Park, R. W. Health, and S. M. Nettles, “Improving Throughputand Fairness for MIMO Ad Hoc Networks Using Antenna SelectionDiversity,” IEEE GLOBECOM, 2004.

[7] K. Sundaresan, R. Sivakumar, M. A. Ingram, and C. Tae-Young,“Medium Access Control in Ad Hoc Networks with MIMO Links:Optimization Considerations and Algorithms,”IEEE Transactions onMobile Computing, vol. 3, no. 4, Oct-Dec 2004.

[8] J. C. Mundarath, P. Ramanathan, and B. D. V. Veen, “NULLHOC: AMAC Protocol for Adaptive Antenna Array Based Wireless Ad HocNetworks in Multipath Environments,”IEEE GLOBECOM 2004, 2004.

[9] J. Park, A. Nandan, M. Gerla, and L. Heechoon, “SPACE-MAC:Enabling Spatial Reuse Using MIMO Channel-Aware MAC,”IEEE ICC2005, 2005.

[10] D. J. Dechene, K. A. Meerja, A. Shami, and S. Primak, “A NovelMIMO-Aware Distributed Media Access Control Scheme for IEEE802.11 Wireless Local Area Networks,”To Appear in Proc. of IEEELocal Computer Networks 2007, 2007.

[11] A. Paulraj, R. Nabar, and D. Gore,Introduction to Space-Time WirelessCommunications. Cambridge University Press, 2003.

[12] S. McCanne and S. Floyd, “ns Network Simulator,” ONLINE, available:http://www.isi.edu/nsnam.

[13] International Standard [for] Information Technology - Telecommunica-tions and information exchange between systems-Local and metropoli-tan area networks - Specific Requirements - Part 11: WirelessLANMedium Access Control (MAC) and Physical Layer (PHY) specifications,1999th ed., IEEE 802.11 WG, Reference number ISO/IEC 8802-11:1999 (E) IEEE STD 802.11, 1999.