WIRELESS COMMUNICATIONS AND MOBILE COMPUTINGWirel. Commun. Mob. Comput. 2016; 16:3–17
Published online 3 June 2014 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/wcm.2492
RESEARCH ARTICLE
CSMA/CA-based medium access control for indoormillimeter wave networksJian Qiao1, Xuemin (Sherman) Shen1*, Jon W. Mark1, Bin Cao2, Zhiguo Shi1 andKuan Zhang1
1 Department of Electrical and Computer Engineering, University of Waterloo, Waterloo, ON, Canada, N2L 3G12 CERC, Harbin Institute of Technology Shenzhen Graduate School, Shenzhen, Guangdong, China
millimeter wave; medium access control; CSMA/CA; backoff mechanism; contention window
*Correspondence
Xuemin (Sherman) Shen, Department of Electrical Computer Engineering, University of Waterloo, Waterloo, ON, Canada, N2L 3G1.E-mail: [email protected]
1. INTRODUCTION
Millimeter wave (mmWave) technology in the 60 GHzband is one of the most promising technologies to pro-vide high data rate (e.g., multi-Gbps) for indoor applica-tions of wireless personal area networks (WPANs) [1–4]and wireless local area networks (WLANs) [5]. Becauseof the abundant bandwidth (around 7 GHz) in the unli-censed 60 GHz band, mmWave communications enablemulti-Gbps wireless connections for high-speed wirelessmultimedia services such as uncompressed high-definitionTV and high speed downloading services. Because freespace propagation loss is proportional to the square ofcarrier frequency, the propagation loss in 60 GHz bandis much higher than that in lower frequency bands, forexample, 28 dB higher than in 2.4 GHz [4]. High direc-tional antennas are used to combat the severe propagationloss and achieve multi-Gbps throughput for bandwidth-intensive applications. The high propagation loss and the
use of directional antennas can enable more efficientspatial reuse.
Although mmWave communications can achieve instan-taneous transmission rate of multi-Gbps, the transmissionthroughput of each node cannot support multimedia appli-cations requiring multi-Gbps throughput, when a largenumber of nodes contend for the wireless channel [5]. Toprovide multi-Gbps throughput for each node and supportmultimedia applications, it is important to improve the net-work capacity. Increasing the capacity of wireless networksrequires increasing the concurrency with which sharedchannels are accessed or increasing the amount of informa-tion sent with each transmission [6]. Multipacket receptionconsists of the ability of allowing multiple nodes to trans-mit their packets simultaneously to the same receiver,and the receiver can decode all such packets successfully.Multipacket reception can be implemented by allowing anode to decode multiple concurrent packets using mul-tiuser detection [6,7] or distributed multiple input multipleoutput techniques. It has been demonstrated that a gain of
CSMA/CA-based MAC for indoor millimeter wave networks J. Qiao et al.
Figure 1. Normalized signal power over distance.
‚.log.n// (n is the number of nodes in the network) can beachieved to increase mmWave network capacity by usingmultipacket reception instead of single packet reception[8,9]. The directional antennas for mmWave communica-tions are helpful to enable multipacket reception in termsof reduced interferences because of the directivity.
Multipacket reception and the unique features ofmmWave communications bring more challenges onmmWave medium access control (MAC) design. CurrentMAC protocols designed for narrowband systems cannotbe directly applied to mmWave networks. Indoor mmWavenetworks are based on the hybrid multiple access of carriersensing multiple access/collision avoidance (CSMA/CA)and time division multiple access (TDMA) [4,10,11]. Pre-vious work has addressed the MAC design and analysison the part of TDMA-based MAC in indoor mmWavenetworks [1,2,5,12]. In this paper, we focus on design-ing efficient CSMA/CA-based MAC protocol for mmWavenetworks with multipacket reception capability. A basicunderlying assumption in the design and evaluation oflegacy CSMA/CA-based MAC protocols was that any con-current transmissions of two or more packets result incollision, which leads to a failure in reception of all thepackets. The actual situation in many wireless communi-cation systems is that the packet with the strongest powerlevel can be received successfully (captured) in the pres-ence of contending transmissions. As shown in Figure 1,the mmWave signal power significantly degrades over dis-tance, thus the received signal power of nearby nodes aremuch stronger than that of distant nodes. Therefore, thechannel is always captured by nearby nodes, resulting inserious unfairness. Moreover, for a receiver with multi-packet reception capability, much stronger power of thenearest nodes also leads to significant degradation of sys-tem throughput. As shown in Figure 2, there are five nodestransmitting to the receiver (network controller) simultane-ously. The stronger received signal power from the nearbynode results in failure of the other four transmissions from
the distant nodes. With multipacket reception capability atthe receiver, the system can successfully receive four pack-ets instead of one if the nearest node does not transmit.Therefore, we propose a backoff mechanism giving highertransmission priority to the distant nodes to achieve bet-ter system throughput and fairness, considering the highpropagation loss of mmWave communications.
The main contributions of this paper are threefold. First,by changing the contention window size, we propose anovel backoff mechanism to adjust the transmission prob-ability according to network congestion status and nodetransmission status. Second, we theoretically analyze thethroughput of mmWave system with multipacket recep-tion and demonstrate that the system throughput can beimproved by giving a higher transmission probability to thenode with a transmission failure than that with a transmis-sion success. Finally, extensive simulations are conductedto demonstrate that the proposed mechanism is effectiveand efficient on improving system throughput and fairness.
The remainder of the paper is organized as follows.In Section 2, the system model is described. A novelbackoff mechanism is proposed in Section 3. The systemthroughput is analyzed using a Markov model in Section 4.The performance of the proposed mechanism is evaluatedby extensive simulations in Section 5. Related works arereviewed in Section 6 followed by concluding remarks inSection 7.
2. SYSTEM MODEL
We consider indoor mmWave networks for the scenar-ios such as large office, conference room, and airport, inwhich many active nodes contend for the mmWave chan-nel and share the medium. Consequently, it is important toincrease network capacity, in order to satisfy the transmis-sion demands of these nodes in the network. Thus, mul-tipacket reception is implemented to increase the network
J. Qiao et al. CSMA/CA-based MAC for indoor millimeter wave networks
Figure 2. Indoor mmWave network architecture.
capacity in this paper. For the scenarios with a smallnumber of active nodes in mmWave indoor networks,the transmission demands can be satisfied even withoutefficient resource utilization. Therefore, it is more interest-ing to consider the challenging cases where the networkresources are limited compared with the traffic demandsof nodes.
2.1. System architecture
Since mmWave indoor networks (e.g., WPANs/WLANs)are centralized in nature, we consider a network composedof multiple wireless nodes and a single network controlleras shown in Figure 2. All wireless nodes are equippedwith electronically steerable directional antennas. Withbeamforming technologies [13–15], the wireless nodes areable to select the best transmission beam and receptionbeam and direct the beams toward each other for transmis-sion and reception. The communication mode for indoormmWave networks is the hybrid of ad hoc mode and cellu-lar mode. Specifically, there are two types of connectionsin the network: end-to-end transmission within the network(ad hoc mode) and the transmission between a node in thenetwork and another node external to the network via net-work controller (cellular mode). CSMA/CA is mainly fornon-delay sensitive applications, such as web browsing,which need to make connections external to the networkvia a network controller. Therefore, in this paper, the com-munication is based on the cellular mode. We consider n asspatially distributed nodes that communicate with a singlenetwork controller (e.g., a piconet controller or an accesspoint) over a slotted channel.
2.2. MAC structure
As indicated in the standards [10,11] for mmWave WLANand WPAN, the networks are based on hybrid multi-ple access of CSMA/CA and TDMA. Different applica-tions have various transmission requirements, for example,applications such as video streaming and wireless displayhave stringent QoS requirements on delay and transmis-sion rate, whereas applications such as web browsing aresensitive to response time and may not require bandwidthguarantees. Therefore, CSMA/CA is used for a burst typeof application such as web browsing because of the loweraverage latency, whereas TDMA is more desirable forvideo transmission due to its better quality of service (QoS)and efficiency. As shown in Figure 3, we consider the IEEE802.15.3 superframe structure as the MAC structure in thispaper. A superframe is composed of three phases: the Bea-con period for network synchronization and control mes-sages among the network controller and wireless nodes, thecontention access period (CAP) based on CSMA/CA fornon-delay sensitive applications, and the channel time allo-cation period (CTAP) composed of M channel time slotsfor bandwidth-intensive and delay-sensitive applications,respectively.
CSMA/CA-based MAC for indoor millimeter wave networks J. Qiao et al.
We have proposed concurrent transmission schedulingalgorithms to exploit the spatial reuse for TDMA-basedCTAP considering both interfering links [5] and non-interfering links [2,16] for delay-sensitive applications.This work focuses on enabling efficient multipacket recep-tion in CAP for mmWave indoor networks.
2.3. Antenna model
The antenna model used in this paper is the one consideredin the previous work [2,5,12]. The side lobes of directionalantennas are generally small enough compared with themain lobe. For instance, the gain of the main lobe of typ-ical directional antennas is more than 100 times the gainof the largest side lobe. Thus, the interference region ofan antenna is principally determined by its main lobe. Thesimplified antenna model, considering the main lobe, willnot result in a fundamental change on the results of thispaper. We define the radiation pattern in a two-dimensionalplane, and the gain of the antenna G.�/ is a function of theazimuth angle �. Specifically, the antenna gain is constantwithin the beamwidth and zero outside the beamwidth,
G.�/ D
�C, j � j� ��
20, otherwise
(1)
where �� D 2�=B is the antenna beamwidth, whereasB is the number of beams. The network controlleris equipped with multiple antennas to receive multiplepackets simultaneously.
2.4. Packet capture model
Because the throughput is one important aspect of thesystem performance and it is counted by the num-ber of received packets, we describe the packet cap-ture model to decide whether a packet is successfullyreceived. Two packet capture models are adopted inthis paper, namely, signal-to-interference-plus-noise ratio(SINR) capture model and vulnerability circle capturemodel. The SINR capture model describes the real situ-ation of packet capture and is used to evaluate the per-formance of the proposed backoff mechanism while thevulnerability circle capture model is a simplified modelbased on the SINR capture model and is used for systemthroughput analysis.
2.4.1. SINR capture model.
The channel time is slotted and the transmission dura-tion for packets is one slot long. The propagation modelincludes path loss and fading. The received power of atransmission from wireless node i, located at distance di
from the receiver, is given by
PR.di/ D PT Kd��i F (2)
where PT is the transmission power, � is the path lossexponent, which is usually determined using a measure-ment approach (typically in the range of 2 to 6 for indoorenvironments [17]), K is the attenuation constant, and Fis the fading factor. The received SINR from node i at thereceiver is
SINR.i/ DPR.di/
N0W CPn
jD1,j¤i PR.dj/(3)
where N0 is the one-sided power spectral density of noiseand W is the system bandwidth. For SINR packet capturemodel, given a set of simultaneous transmission packets,the packet from node i is successfully received if
SINR.i/ > h (4)
where h is the packet capture threshold ratio. For sin-gle packet reception narrowband systems, 1 � h � 10,whereas for wideband multipacket reception systems, suchas UWB and mmWave, h < 1 [18]. Let M denote the mul-tipacket reception capability, then the maximum numberof packets will be captured if there are M equal received-power packets at the receiver with M D d1=he [18,19],where de is the ceiling function.
2.4.2. Vulnerability circle capture model.
This model was proposed in [20] and studied in [21]. Inthis model, the node closest to the receiver can capture thechannel due to its larger power at the receiver. Specifically,a node i with distance di from the receiver captures thechannel if there are no simultaneous transmissions withina disk of radius ˛di.˛ > 1/. The parameter ˛ is the vul-nerability circle capture ratio. We extend the vulnerabilitycircle capture model to the case of multipacket reception.Based on SINR capture model, a transmission from node iis successful if
SINR.i/ > h, .h < 1/ (5)
To achieve this, the received power of the successful pack-ets should be similar to each other. To make the analysistractable, the successful packets are assumed to have sim-ilar distances to the receiver. Let ˇ denote the number ofsuccessful transmission packets. For a receiver with mul-tipacket reception capability, a transmission from a nodewith distance di is successful if there are ˇ.ˇ � M/ simul-taneous transmissions around the circle with radius di andthe other simultaneous transmissions (if they exist) are out-side of the disk of radius �di, where � is the vulnerabilitycircle capture ratio for the case of multipacket reception.� > 1/. Because mmWave signal power degrades sig-nificantly over distance, the values of ˛ and � could berelatively smaller than those at lower frequency bands.
J. Qiao et al. CSMA/CA-based MAC for indoor millimeter wave networks
3. BACKOFF MECHANISM DESIGN
In CSMA/CA-based MAC, the backoff mechanism con-trols the transmission probability of each node. In a generalbackoff mechanism, each node i sets an integer backoffcounter Bi,j randomly generated from a contention win-dow Wi,j, that is, Bi,j 2 f0, 1, : : : , Wi,j � 1g. The subscript jrepresents the backoff stage of node i. The contention win-dow size is reset after a transmission attempt and node iretransmits packet after Bi,j time slots.
In this section, we first present the design considera-tions for the proposed backoff mechanism. Then, a detailedbackoff mechanism is proposed to adjust the contentionwindow size considering the unique features of indoormmWave systems, aiming to improve the system through-put and fairness.
3.1. Design considerations
A transmission failure may occur in two independent sce-narios. (i) Both the nearby nodes and distant nodes sendpackets to the receiver. Because of shorter transmissiondistance, the nearby nodes usually have stronger receivedpower to satisfy the SINR capture model SINR.i/ > h,thus the transmissions from distant nodes fail. (ii) Thetransmissions arriving at the receiver are only from dis-tant nodes and the received SINR of each node is notlarger than the packet capture threshold ratio h in (4) ifthere are too many transmissions. Thus, no packet cap-tures the channel. For the first scenario, we can mitigatethe channel capture effect of nearby nodes by giving highertransmission probability to distant nodes, and the receivercan receive more packets because of its multipacket recep-tion capability. The second scenario indicates that muchhigher transmission probability of the distant nodes canalso result in system throughput reduction. It would makethe case worse if the contention window size were fur-ther reduced after a transmission failure of distant nodes.Consequently, to achieve better system throughput, thetransmission probability needs to be properly adjusted.
The transmission probability controlled by the backoffmechanism has a strong impact on both system through-put and fairness. Giving higher priority to the distant nodescan achieve higher system throughput due to the multi-packet reception capability at the receiver. Meanwhile, thenearby nodes would have less chance to access the chan-nel and may even suffer starvation problem. During theCAP in each superframe, the transmission requests are alsosent from the nodes that are active during the next CTAP.The number of received transmission requests has signifi-cant impact on the overall system throughput of mmWaveindoor networks based on the hybrid multiple access ofCSMA/CA and TDMA [22]. Thus, the proposed backoffmechanism needs to avoid starvation to nearby nodes if wegive higher transmission priority to distant nodes.
We propose a novel backoff mechanism for CSMA/CA-based MAC with the following considerations. (i) Identifydifferent scenarios for packet transmission failure and use
different backoff strategies to deal with them in order toincrease the number of successful packets. (ii) Try to avoidstarvation problem in case there is large delay for thetransmission requests of those flows operating in CTAP.In addition, the performance of response time would beimproved by dealing with the starvation problem. (iii)The unique features of mmWave communication shouldbe considered, for example, the mmWave signal strengthis very sensitive to transmission distance due to its highpropagation loss.
3.2. Proposed backoff mechanism
Initially, with neighbor discovery [23] and beamformingtechnologies [13], the wireless nodes register with the net-work controller and train their antenna array in the direc-tion of the receiver to maximize the received signal powerat the receiver. With the accurate localization service pro-vided by the mmWave indoor system [24,25], the networkcontroller has the valid network topology information.Considering that all the nodes are randomly distributedin a circular region, there are more nodes located in thedistant area, thus, the transmission probability of distantnodes could have great impacts on the number of transmis-sions arriving at the receiver. To attain more transmissionsuccesses, the contention window size of distant nodesshould be adjusted more moderately compared with nearbynodes. In addition, mmWave signal power degrades signif-icantly over transmission distance and the received signalpowers of all the transmissions determine whether a trans-mission is successful or not. Therefore, we use d�� as thecoefficient of the interval to adjust the contention windowsize. Specifically, the adjustment interval for node i after atransmission failure is
˙d��i Ww
�, whereas the adjustment
interval for node i after a transmission success is˙
d��i Wr�
,where Ww and Wr denote the basic backoff intervals aftera transmission failure and a transmission success, respec-tively. To give a higher transmission probability to the nodewith a transmission failure than that with a transmissionsuccess, we have Wr > Ww.
Both much higher transmission probability and lowertransmission probability for distant nodes can result insystem throughput reduction; therefore, we adjust the con-tention window after a transmission attempt according tothe network congestion status. If the scenario in whichtransmission fails with no packet successfully received atthe receiver occurs frequently, it is very likely that thetransmission probability of distant nodes is much higher.Then we enlarge their contention window to reduce thetransmission probability. Otherwise, the channel is cap-tured by nearby nodes and we reduce the transmissionprobability of nearby nodes by increasing their contentionwindow size while the transmission probability of distantnodes needs to be increased by reducing their contentionwindow size.
During the kth slot of the CAP period in the mth
superframe, the network controller receives a number oftransmission packets and determines which transmission
CSMA/CA-based MAC for indoor millimeter wave networks J. Qiao et al.
Algorithm 1 Backoff Mechanism for CSAM/CABEGIN:1: for slot k of CAP in mth superframe do2: Several nodes send packets to network controller3: if Multiple packets arrive but no successful packets then4: Set fa.k/ D 1
5: Update Fk,m DPk
lDk�TC1 fa.l/
T6: if Fk,m > Fthr then7: Update contention window Wi D Wi C
˙d��i Ww
�8: else9: Update contention window Wi D Wi �
˙d��i Ww
�10: end if11: else12: Set fa.k/ D 013: if Transmission of node i is successful then14: Update contention window Wi D Wi C
˙d��i Wr
�15: else16: Update contention window Wi D Wi �
˙d��i Ww
�17: end if18: end if19: if Wi � Wmax then20: Update contention window Wi D Wmax �
˙d��i Ww
�21: end if22: Update k D kC 123: if No more slots in mth CAP then24: FROZEN25: Until the CAP of next superframe26: Update m D mC 127: end if28: Go to line 129: end for
END;
packets are captured, based on the SINR packet capturemodel in (4). Then it broadcasts a feedback packet to all thenodes in the network to indicate the transmission failuresand transmission successes. We use packet capture failurefrequency .Fk,m/ to determine the contention window size.Fk,m is defined as the number of time slots (during the pre-vious T time slots from the current kth time slot in the mth
superframe), in which there are transmissions arriving atthe receiver but no packet is captured successfully, dividedby T . Fk,m is given as
Fk,m D
PklDk�TC1 fa.l/
T(6)
where fa.l/ is
fa.l/ D
�1, MT .l/ > 1 and MR.l/ D 00, otherwise
(7)
and MT .l/ and MR.l/ indicate the number of transmissionsand successful receptions in the lth time slot, respectively.If Fk,m is larger than a specific threshold Fthr, the networkbecomes congested because of the higher transmissionprobability of distant nodes, thus the contention windowof each transmission node i in the current time slot isincreased by
˙d��i Ww
�to reduce transmission probability.
Otherwise, the contention window of node i is increased by˙d��i Wr
�following a transmission success and decreased
by˙
d��i Ww�
after a transmission failure. The contentionwindow size can be adjusted until it reaches the maximumvalue Wmax. To deal with the starvation problem, when
the contention window of node i is larger than or equal toWmax, it is set as Wmax �
˙d��i Ww
�to increase its trans-
mission probability. When all the transmissions are notsuccessful at the receiver, although the contention windowsize is increased, it does not mean that the transmissionprobability after a transmission failure .pw/ is less than thatafter a transmission success .pr/. In this case, the trans-mission probability of distant nodes is much higher, thus,we reduce it moderately to achieve better system through-put. It is very likely that pw is still larger than pr afterthe contention window size is increased for the nodes withtransmission failure.
As shown in Figure 3, the CAP is not continuous in dif-ferent superframes. Therefore, the backoff status of eachnode is frozen at the end of each CAP and restarts at thebeginning of the next CAP in the following superframe.The detailed procedure of the proposed backoff mechanismis described by the pseudocode in Algorithm 1.
4. SYSTEM THROUGHPUTANALYSIS
In this section, we analyze the system throughput forindoor mmWave networks with multipacket receptioncapability. The packet capture model is based on thevulnerability circle capture model described earlier. Theperformance analysis of this section is to theoreticallydemonstrate that the transmission probability adjustment inthe proposed backoff mechanism can significantly improvethe system throughput, compared with traditional backoffmechanisms.
The transmission states of a wireless node can bedescribed as the Markov state diagram shown in Figure 4.A node can be in two states: after success (AS) and afterfailure (AF). State transition may take place after a trans-mission attempt. A node moves into the AS (AF) state aftera transmission success (failure). Because transitions takeplace after an attempt, a node does not change its statefollowing an idle slot. A node in the AS state transmitswith probability pr at each slot, disregarding the status ofthe channel while the transmission probability is pw for anode in the AF state. The values of pr and pw representthe size of contention window in the backoff mechanism.For example, a high value of pr corresponds to maintaininga small contention window following a transmission suc-cess in the traditional backoff mechanisms as the one used
J. Qiao et al. CSMA/CA-based MAC for indoor millimeter wave networks
in IEEE 802.11. Similarly, a low pw value corresponds tomaintaining a large contention window following a trans-mission failure. Therefore, traditional backoff mechanismshave a large value of pr and a small value of pw, which cor-responds to the event the contention window is increasedafter a transmission failure and the contention window isreduced after a transmission success.
Generally, when a node transmits a packet, the loss prob-ability due to collisions depends on other transmissionsduring that slot. However, in most work on the performanceanalysis, the loss probability of a packet transmitted by anode at distance d, defined as PF.d/, is assumed to be inde-pendent of the number of retransmissions suffered. Thevalidity and accuracy of the assumption have been recentlyverified [26,27]. The state transition probability from stateAS to state AF is prPF.d/. Similarly, the state transitionprobability from state AF to AS is pw.1 � PF.d//. Hence,the transition probability matrix is given as
With the transition probability matrix, steady-state prob-abilities of states AS .�AS/ and AF .�AF/ for anode at distance d can be obtained by solving thefollowing equations:
��AS C �AF D 1
�P D �(9)
where � D f�AS,�AFg. Then, we have
�AS Dpw.1 � PF.d//
prPFd C pw.1 � PF.d//
�AF DprPF.d/
prPFd C pw.1 � PF.d//
(10)
The total transmission probability of a node at distance d is
p.d/ D �ASpr C �AFpw
Dprpw
pw � pwPF.d/C prPF.d/
(11)
According to the vulnerability circle capture model, for areceiver with single packet reception capability, a trans-mission from a node at distance d succeeds if there is nosimultaneous transmission among the other .n � 1/ nodeswithin a disk with radius ˛d around the receiver, wheren is the total number of transmission nodes in the net-work. For a receiver with multipacket reception capabilityM, a transmission of a node at distance d becomes suc-cessful if there is no simultaneous transmission among the.n � ˇ/.2 � ˇ � M/ nodes within a disk of radius �daround the receiver; meanwhile, the ˇ simultaneous trans-missions are distributed around the circle with radius d.ˇ is the number of simultaneous transmissions received
successfully. The probability that all the .n � ˇ/ nodes donot transmit packets within the disk of radius �d is given as
P 0 D"
1 �Z �d
0p.y/f .y/dy
#n�ˇ
(12)
where f .d/ is the probability density function of the dis-tance from a node to the receiver, and p.d/ is the trans-mission probability of a node at distance d. To achievesuccessful reception of ˇ packets simultaneously, the trans-mitting nodes of ˇ packets need to be distributed aroundthe circle of radius d. For example, the packet transmissionof a node located within the disk of radius d=� can resultin the transmission failure of a node at distance d accord-ing to the vulnerability circle capture model. To make theanalysis tractable, it is assumed that the other .ˇ�1/ simul-taneous transmitting nodes are distributed within the areabetween the circle of radius d.1C"/ and the circle of radiusd.1� "/. " is a parameter determining the location area forthe .ˇ � 1/ transmitting nodes. Thus, the correspondingprobability is
P 00 D"Z d.1C"/
d.1�"/f .y/dy
#ˇ�1
(13)
Therefore, the transmission failure probability for a nodeat distance d for multipacket reception case is
PF.d/ D 1 �
n � 1
ˇ � 1
!P 0P 00 (14)
Substituting (12) and (13) into (14), we have
PF.d/ D 1 �
n � 1
ˇ � 1
!"Z d.1C"/
d.1�"/f .y/dy
#ˇ�1
�
"1 �
Z �d
0p.y/f .y/dy
#n�ˇ(15)
The throughput of a node at distance d from the receiver is
C.d/ D p.d/.1 � PF.d// (16)
and the total system throughput is given as
C D nZ D
0C.y/f .y/dy
D nZ D
0f .y/p.y/.1 � PF.y//dy
(17)
where D is the radius of the whole communication area.Given the probability density function f .d/, the transmis-sion probability of a node at AS state pr, and the transmis-sion probability of a node at AF state pw, we can obtain thesystem throughput C numerically. If all the nodes are uni-formly distributed in the disk of radius D with the network
CSMA/CA-based MAC for indoor millimeter wave networks J. Qiao et al.
Figure 5. System throughput of different combinations of transmission probabilities.
controller in the center, we have
f .d/ D
� 2dD2 , .0 < d � D/0, otherwise
(18)
With multipacket reception capability for M D ˇ D 4,Figure 5 shows the numerical results of system through-put for different combinations of pr and pw. In traditionalCSMAC/CA, the transmission probability after transmis-sion failure pw is less than that after transmission suc-cess pr, in order to reduce the transmission collision. InFigure 5, we show the system throughput for the casesof both pw < pr and pw > pr. It can be seen that thenetwork with multipacket reception capability can achievemuch higher system throughput with the case of pw > pr,compared with the traditional case of pw < pr. Initially,the system throughput increases as more nodes trans-mit packets in the network and then it decreases becausemore collisions occur because of larger number of nodesinvolved in packet transmission. For the case of pw >
pr, much larger pw can result in fast degradation ofsystem throughput.
In summary, the aforementioned analysis demonstratesthat, in the case of multipacket reception, the systemthroughput can be improved by giving a higher transmis-sion probability to the node after a transmission failurethan that after a transmission success. Meanwhile, muchhigher transmission probability after transmission failurecan also result in system throughput reduction. The afore-mentioned analysis theoretically verifies the performanceof the proposed backoff mechanism.
5. PERFORMANCE EVALUATION
In this section, we describe the performance evaluation set-tings and present the simulation results for the proposed
backoff mechanism compared with two other exponentialbackoff mechanisms.
We evaluate the performance of the proposed backoffmechanism in terms of system throughput, fairness, andpower consumption in a typical mmWave indoor environ-ment (i.e., large office space). The network controller isplaced in the center of the room and all wireless nodes arerandomly distributed in the circular region with a radiusof 20 m. Each node is equipped with a directional antennawith a beamwidth of 60ı, corresponding to six beams ateach node. Each packet is one-slot long and the packet cap-ture is based on SINR packet capture model with packetcapture threshold ratio h D 0.25 (i.e., multipacket recep-tion capability is 4). The signal propagation model is basedon the free space Friis model with path loss exponent � D2. The reference distance is set to 1.5 m, which is also usedto bound the adjustment interval because it is proportionalto 1=d� . The main parameters used in our simulations arelisted in Table I.
We compare the proposed backoff mechanism with twoother exponential backoff mechanisms, namely, traditionalexponential backoff mechanism and alternative exponen-tial backoff mechanism. In traditional exponential backoffmechanism, after every transmission failure, the contentionwindow size is doubled while it is reduced half after atransmission success. The alternative exponential backoffmechanism doubles the contention window size after atransmission success and reduces the contention windowby half following a transmission failure. We use the per-formance of traditional exponential backoff mechanism asthe baseline for comparison. The alternative exponentialbackoff mechanism gives higher transmission probabilityafter a transmission failure. However, its exponential back-off would give much higher transmission probability aftertransmission failure and results in congestions.
Figure 6 shows the normalized system throughput of thethree backoff mechanisms as a function of the number of
J. Qiao et al. CSMA/CA-based MAC for indoor millimeter wave networks
Table I. Simulation parameters.
Parameters Symbol Value
System bandwidth W 1200 MHzTransmission power PT 0.1 mWBackground noise N0 �134 dBm/MHzReference distance dref 1.5 mPath loss at dref PL0 71.5 dBSlot time �T 10 �sBeacon period TBEA 50 �sRandom access period TRAP 80 msChannel time allocation period TCTAP 500 msMaximum contention window Wmax 10 000Basic success backoff interval Wr 300Basic failure backoff interval Ww 200Congestion threshold Fthr 0.3Congestion concern duration T 20
Figure 6. Normalized system throughput of three backoff mechanisms.
nodes in the network. The alternative exponential backoffmechanism and proposed backoff mechanism can achievehigher system throughput compared with traditional expo-nential backoff mechanism because it gives higher trans-mission probability after a transmission failure. As morenodes are involved in the network, the proposed backoffmechanism adjusts the contention window considering thenetwork congestion status and thus achieves higher systemthroughput in comparison with the alternative exponentialbackoff mechanism.
Figure 7 shows the average throughput per node atdifferent distances to indicate the fairness. We run thesimulation 100 times with 30 nodes randomly distributedin the network. With the proposed backoff mechanism,the throughput of each node does not decrease much withthe distance, whereas with traditional exponential backoffmechanism, the throughput of distant nodes is much lessthan that of nearby nodes. The proposed backoff mech-anism considers both the network congestion status and
the transmission status of each node to adjust thetransmission probability and achieves better fairness.Although the achieved gain on the throughput of each nodeis not that much, the overall system throughput would beimproved significantly because there are many nodes inthe network.
Transmission failure probability is the probability thata transmission attempt experiences a failure. Note that apacket can suffer multiple transmission failures before itis successfully received. Figure 8 shows the transmissionfailure probabilities of the three backoff mechanisms. Theyhave similar performance on transmission failure probabil-ity if there are not many nodes in the network. The alter-native exponential backoff mechanism gives much highertransmission probability to distant nodes and more nodesare distributed in the distant area. Thus, there are manytransmissions coming from distant area and the SINR ofeach cannot exceed the packet capture threshold ratio. Asa result, the transmission failure probability of the alterna-
CSMA/CA-based MAC for indoor millimeter wave networks J. Qiao et al.
Figure 7. Average node throughput with different distance for fairness.
Figure 8. Transmission failure probability.
tive exponential backoff mechanism increases rapidly withthe increased number of nodes. Although the transmissionfailure probability of the alternative exponential backoffmechanism exceeds that of the others for a dense network,the network throughput of it is still more than that of thetraditional exponential backoff mechanism because thereare more transmission attempts with the alternative expo-nential backoff mechanism due to its higher transmissionprobability for distant nodes.
The normalized average packet delays of the three mech-anisms are shown in Figure 9. The packet delay is definedas the time duration from the time the packet is trans-mitted to the time the packet is received successfully. Asthe number of nodes increases, the network becomes morecongested and the packet delay increases. The proposedbackoff mechanism has shorter packet delay correspondingto less congestions in the network.
We then compare the energy consumption of theproposed backoff mechanism with that of the other twoexponential backoff mechanisms. For fair comparison,we use the total transmission energy (consumed by bothsuccessful and unsuccessful packets) divided by the totalnumber of successful packets to obtain the energy con-sumption per successful packet. From Figure 10, we cansee that our proposed backoff mechanism is more energyefficient than the other two exponential backoff mecha-nisms. By adjusting the contention window, the proposedbackoff mechanism can reduce the number of transmis-sion failures (energy waste) and make the receiver acceptmore packets.
The proposed backoff mechanism can achieve better per-formance on system throughput, fairness, average packetdelay, and energy consumption. Meanwhile, it introducescommunication overheads and computational overheads
J. Qiao et al. CSMA/CA-based MAC for indoor millimeter wave networks
Figure 9. Average packet delay.
Figure 10. Normalized power consumption per packet.
as indicated in the proposed mechanism. The main com-munication overheads are the feedback packets from thenetwork controller to other nodes in the network. However,in traditional backoff mechanisms in CSMA/CA, thereare still ACK packets from the receiver to confirm thepacket reception success. In the proposed backoff mech-anism, the nodes need to compute packet capture failurefrequency .Fk,m/ and compare it with the threshold. It is asimple computation and does not consume much computa-tional power. Therefore, the proposed backoff mechanismcan significantly improve the network performance withlimited extra cost.
6. RELATED WORKS
A wide range of MAC layer protocols and algo-rithms have been proposed for mmWave indoor networks[1,2,5,12,13,16,22,28–31]. One line of research focuses ontransmission scheduling for TDMA-based MAC period to
improve the system throughput [2,5,12,16,22,29,30]. Thehigh propagation loss and the utilization of directionalantenna result in relatively low multi-user interference, sothat concurrent transmissions can be supported to exploitthe spatial reuse [2,12,16,29,30]. In [2], a multi-hop con-current transmission scheduling scheme is proposed toallow non-interfering transmission links to operate simul-taneously over mmWave channels. To further improve thespatial reuse, spatial-time division multiple access-basedconcurrent transmission scheduling schemes are proposedto allow both non-interfering and interfering links totransmit concurrently either to achieve suboptimal systemthroughput [5] or to make the accumulated interferencein each time slot below a specific threshold [30]. In[12,32], an exclusive region (ER)-based resource man-agement scheme is proposed to exploit the spatial reuseof mmWave WPANs with directional antenna, and theoptimal ER sizes are analytically derived.
Another line of research on mmWave MAC consid-ers the mutual impact on system throughput of both
CSMA/CA-based MAC for indoor millimeter wave networks J. Qiao et al.
CSMA/CA-based MAC period and TDMA-based MACperiod. A long period of CSMA/CA will cause a lowdata transmission time in TDMA period, whereas a shortlength of access time of CSMA/CA will cause a large num-ber of data transmission collisions in CSMA/CA. In [22],system throughput is optimized by adjusting the accessperiods of CSMA/CA and TDMA without consideringconcurrent transmissions.
To the best of our knowledge, there are fewworks addressing the CSMA/CA-based MAC period formmWave indoor networks. Most of existing works onmmWave MAC focus on TDMA-based MAC period whichis mainly for bandwidth-intensive multimedia applications.Burst type of applications such as web browsing maynot require bandwidth guarantees and uses CSMA/CAmechanism to access the channel. To apply applicationswithin much shorter CSMA/CA access period comparedwith TDMA access period, it is important to improve thenetwork capacity. Multipacket reception is implementedin wireless networks to significantly improve networkcapacity [19,33,34]. In [33], the proposed MAC protocoladaptively grants access to the MPR channel to severalusers to maximize the expected number of successfullyreceived packets in each slot. The system throughput ofnetworks with multipacket reception capability is analyzedin [19] considering spatially distributed nodes. [34] pro-poses a physical layer multipacket reception techniqueand the corresponding MAC layer which closely followsthe IEEE 802.11 distributed coordination function scheme.In this paper, we propose a new backoff mechanism forCSMA/CA-based MAC for wideband mmWave indoornetworks with multipacket reception capability, consider-ing the high propagation loss resulting significant systemthroughput reduction.
7. CONCLUSION
In this paper, we have proposed a backoff mechanismfor CSMA/CA-based MAC for mmWave indoor net-works with multipacket reception capability. Generally, theproposed backoff mechanism gives higher transmissionprobability to distant nodes to take the advantage ofmultipacket reception capability at the receiver. Thetransmission probability can be adjusted by changing thecontention window size according to the node’s transmis-sion status (failure or success) and network congestion sta-tus. With the proposed backoff mechanism, the mmWaveindoor networks can achieve higher system throughput andbetter fairness.
To the best of our knowledge, this work is one ofthe first attempts to design CSMA/CA-based MAC formmWave networks with multipacket reception capability,considering the severe signal power degradation overdistance in mmWave band. Because different parametersettings can impact the system performance, we intendto develop mechanisms selecting appropriate parameter
values to achieve optimal system performance, for exam-ple, system throughput. Moreover, mmWave links arehighly susceptible to blockage because of the limited abil-ity to diffract around obstacles such as moving peopleand furniture in indoor environment. Therefore, designingefficient CSMA/CA-based MAC for multi-hop scenariois an interesting and challenging problem in mmWaveindoor networks.
REFERENCES
1. Singh S, Ziliotto F, Madhow U, Belding EM,
Rodwell M. Blockage and directivity in 60 GHz
wireless personal area networks: from cross-layer
model to multihop MAC design. IEEE Journal of
Selected Areas in Communications 2009; 27 (8):
1400–1413.
2. Qiao J, Cai LX, Shen X, Mark JW. Enabling multi-hop
concurrent transmissions in 60 GHz wireless personal
area networks. IEEE Transactions on Wireless Commu-
J. Qiao et al. CSMA/CA-based MAC for indoor millimeter wave networks
10. IEEE 802.15 WPAN Millimeter wave alternative PHYtask group 3c (TG3c). Available: http://www.ieee802.org/15/pub/TG3c.html [accessed on July 2013].
11. IEEE 802.11 VHT Study Group. Available:http://www.ieee802.org/11/Reports/vht_update.htm[accessed on July 2013].
12. Cai LX, Cai L, Shen X, Mark JW. REX: a ran-domized exclusive region based scheduling schemefor mmWave WPANs with directional antenna. IEEETransactions on Wireless Communications 2010; 9(1):113–121.
13. Lee H, Ko Y. Low complexity codebook-based beam-forming for MIMO-OFDM systems in millimeter-wave WPAN. IEEE Transactions on WirelessCommunications 2011; 10(11): 3607–3612.
14. Shad F, Todd TD, Kezys V, Litva J. Dynamic slotallocation (DSA) in indoor SDMA/TDMA using asmart antenna basestation. IEEE/ACM Transactions onNetworking 2001; 9(1): 69–81.
15. Ward J, Compton RT. Improving the performance ofslotted ALOHA packet radio network with an adaptivearray. IEEE Transactions on Communications 1992;40: 292–300.
16. Qiao J, Cai LX, Shen X. Multi-hop concurrenttransmission in millimeter wave WPANs with direc-tional antenna. In IEEE ICC, Cape Town, 2010;1–5.
18. Nguyen G, Ephremides A, Wieselthier J. Comments on“capture and retransmission control in mobile radio”.IEEE Journal of Selected Areas in Communications2006; 24(12): 2340–2341.
19. Celik GD, Gil Z, Khan WF, Modiano E. MACfor networks with multipacket reception capabil-ity and spatially distributed nodes. IEEE/ACMTransactions on Mobile Computing 2010; 9(2):226–240.
20. Roberts LG. Aloha packet system with and withoutslots and capture broadcast channels. Computer Com-municationr Review 1975; 5: 28–42.
21. Lau CT, Leung C. Capture models for mobile packetradio networks. IEEE Transactions on Communica-tions 1992; 40: 917–925.
22. Pyo CW, Harada H. Throughput analysis and improve-ment of hybrid multiple access in IEEE 802.15.3cmmWave WPAN. IEEE Journal of Selected Areas inCommunications 2009; 27(8): 1414–1424.
23. An X, Prasad RV, Niemegeers I. Impact of antenna pat-tern and link model on directional neighbor discovery
in 60 GHz networks. IEEE Transactions on Wireless
Communications 2011; 10(5): 1435–1447.
24. Deparis N, Loyez C, Rolland N, Rolland PA.
UWB in millimeter wave band with pulsed ILO.
IEEE Transactions on Circuits and Systems 2008;
55(4): 339–343.
25. Winkler F, Fischer E, Grass E, Langendorfer P. An
indoor localization system based on DTDOA for differ-
ent wireless LAN systems. In Proceedings on WPNC,
Hannover, Germany, 2006; 117–122.
26. Cho JW, Jiang Y. Fundamentals of the backoff process
in 802.11 DCF. 2009 CoRR, abs/0904.41555.
27. Celik G. Distributed MAC protocol for networks
with multipacket reception capability and spatially
distributed nodes. Master’s Thesis, MIT, Cam-
bridge, MA, 2007. Available: http://web.mit.edu/
gcelik/MSThesis/Thesis.pdf [accessed on September
2013].
28. Ziliotto F, Madhow U, Belding EM, Rodwell MJW.
Millimeter wave, WPAN: cross-layer modeling and
multi-hop architecture. In IEEE INFOCOM, Anchorage,
AK, 2007; 2336–2340.
29. Yang Z, Cai L, Lu W. Practical concurrent transmis-
sion scheduling algorithms for rate-adaptive wireless
networks. In IEEE INFOCOM, San Diego, CA, 2010;
1–9.
30. Sum C, Lan Z, Funada R, Wang J, Baykas T,
Rahman MA, Harada H. Virtual time-slot allocation
scheme for throughput enhancement in a millimeter-
wave multi-Gbps WPAN system. IEEE Journal on
Selected Areas in Communications 2009; 27(8):
1379–1389.
31. Shen X, Zhuang W, Jiang H, Cai J. Medium
access control in ultra-wideband wireless networks.
IEEE Transactions on Vehicular Technology 2005;
54(5): 1663–1677.
32. Cai LX, Cai L, Shen X, Mark JW. Resource manage-
ment and QoS provisioning for IPTV over mmWave-
based WPANs with directional antenna. ACM
Mobile Networks and Applications (MONET) 2009;
14(2): 210–219.
33. Zhao Q, Tong L. A multiqueue service room MAC
protocol for wireless networks with multipacket recep-
tion. IEEE/ACM Transactions on Networking 2003;
11(1): 125–137.
34. Zheng P, Zhang Y, Liew S. Multipacket reception in
wireless local area networks. In IEEE ICC, Istanbul,
CSMA/CA-based MAC for indoor millimeter wave networks J. Qiao et al.
AUTHORS’ BIOGRAPHIES
Jian Qiao is currently working towardhis PhD degree at the Department ofElectrical and Computer Engineering,University of Waterloo, Canada. Hereceived his BE degree in BeijingUniversity of Posts and Telecom-munications, China in 2006 and hisMASc degree in Electrical and Com-puter Engineering from University
of Waterloo, Canada in 2010. His research interestsinclude millimeter wave WPANs, medium access control,resource management, and millimeter wave 5Gcellular networks.
Xuemin (Sherman) Shen (IEEEM’97-SM’02-F’09) received his BSc(1982) degree from Dalian Mar-itime University (China) and hisMSc (1987) and PhD degrees (1990)from Rutgers University, New Jersey(USA), all in Electrical Engineering.He is a professor and a universityresearch chair, Department of Electri-
cal and Computer Engineering, University of Waterloo,Canada. He was the associate chair for Graduate Stud-ies from 2004 to 2008. Dr. Shen’s research focuses onresource management in interconnected wireless/wirednetworks, wireless network security, social networks,smart grid, and vehicular ad hoc and sensor networks.Dr. Shen served as the technical program committeechair/co-chair for IEEE Infocom’14, IEEE VTC’10Fall, the Symposia Chair for IEEE ICC’10, the TutorialChair for IEEE VTC’11 Spring, and IEEE ICC’08, theTechnical Program Committee Chair for IEEE Globe-com’07, the General Co-Chair for Chinacom’07 andQShine’06, the Chair for IEEE Communications Soci-ety Technical Committee on Wireless Communications,and P2P Communications and Networking. He alsoserves/served as the editor-in-chief for IEEE Network,Peer-to-Peer Networking and Application, and IET Com-munications; a founding area editor for IEEE Transactionson Wireless Communications; an associate editor for IEEETransactions on Vehicular Technology, Computer Net-works, and ACM/Wireless Networks, and so on; and theguest editor for IEEE JSAC, IEEE Wireless Communica-tions, IEEE Communications Magazine, and ACM MobileNetworks and Applications, and so on. Dr. Shen receivedthe Excellent Graduate Supervision Award in 2006 and theOutstanding Performance Award in 2004, 2007, and 2010from the University of Waterloo, the Premier’s ResearchExcellence Award (PREA) in 2003 from the Provinceof Ontario, Canada, and the Distinguished PerformanceAward in 2002 and 2007 from the Faculty of Engineering,University of Waterloo. Dr. Shen is a registered Profes-sional Engineer of Ontario, Canada, an IEEE Fellow,an Engineering Institute of Canada Fellow, a Canadian
Academy of Engineering Fellow, and a Distinguished Lec-turer of IEEE Vehicular Technology Society and Commu-nications Society.
Jon W. Mark (LF) received hisPhD degree in Electrical Engineeringfrom McMaster University in 1970. InSeptember 1970, he joined the Depart-ment of Electrical and ComputerEngineering, University of Water-loo, Waterloo, Ontario, where he iscurrently a distinguished professoremeritus. He served as the depart-
ment chairman during the period July 1984–June 1990. In1996, he established the Center for Wireless Communica-tions (CWC) at the University of Waterloo and is currentlyserving as its founding Director. Dr. Mark had beenon sabbatical leave at the following places: IBM ThomasJ. Watson Research Center, Yorktown Heights, NY, as aVisiting Research Scientist (1976–1977); AT&T BellLaboratories, Murray Hill, NJ, as a resident consul-tant (1982–1983): Laboratoire MASI, UniversitlePierreet Marie Curie, Paris France, as an invited professor(1990–1991); and Department of Electrical Engineering,National University of Singapore, as a visiting professor(1994–1995).He has previously worked in the areas of adaptive equaliza-tion, image and video coding, spread spectrum communi-cations, computer communication networks, ATM switchdesign, and traffic management. His current research inter-ests are in broadband wireless communications, resourceand mobility management, and cross domain interworking.A life fellow of IEEE and a fellow of the CanadianAcademy of Engineering, Dr. Mark is the recipient of the2000 Canadian Award for Telecommunications Researchand the 2000 Award of Merit of the Education Founda-tion of the Federation of Chinese Canadian Professionals.He was an editor of IEEE Transactions on Communica-tions (1983–1990), a member of the Inter-Society SteeringCommittee of the IEEE/ACM Transactions on Network-ing (1992–2003), a member of the IEEE CommunicationsSociety Awards Committee (1995–1998), an editor ofWireless Networks (1993–2004), and an associate editor ofTelecommunication Systems (1994–2004).
Bin Cao received his MASc in Infor-mation and Communication Engineer-ing from Harbin Institute of Tech-nology Shenzhen Graduate School,China. Now, he is with his PhDdegree at Harbin Institute of Tech-nology Shenzhen Graduate School.He has been a visiting student atBBCR group from 2010 to 2012,
University of Waterloo, Waterloo, Canada. His researchinterests include signal processing for wireless com-munications, cognitive radio networking, and cross-layer optimization.
J. Qiao et al. CSMA/CA-based MAC for indoor millimeter wave networks
Zhiguo Shi (IEEE M’10) received theBS degree and PhD degree both inElectronic Engineering from ZhejiangUniversity, Hangzhou, China, in 2001and 2006, respectively. From 2006 to2009, he was an assistant professorwith the Department of Informationand Electronic Engineering, ZhejiangUniversity, where currently he is an
associate professor. From September 2011, he begins a 2-year visiting to the Broadband Communications Research(BBCR) Group, University of Waterloo. His research inter-ests include radar data and signal processing, wirelesscommunication, and security. He received the Best PaperAward of IEEE WCNC 2013, Shanghai, China, and IEEEWCSP 2012, Huangshan, China. He received the Scientific
and Technological Award of Zhejiang Province, China in2012. He serves as an editor of KSII Transactions onInternet and Information Systems. He also serves as TPCmember for IEEE VTC 2013 Fall, IEEE ICCC 2013, MSN2013, IEEE INFOCOM 2014, IEEE ICNC 2014, etc.
Kuan Zhang received his BSc andMSc degrees in Electrical and Com-puter Engineering from Northeast-ern University, Shenyang, China, in2009 and 2011. He is currently work-ing toward his PhD degree withthe Department of Electrical andComputer Engineering, University ofWaterloo, Canada. His research inter-
ests include security and privacy for mobile social network.
