Performance Analysis and Improvement on DSRC Application for V2V Communication
PPerformance Analysis and Improvement on DSRCApplication for V2V Communication
Liu Cao ∗ , Hao Yin ∗ , Jie Hu † , Lyutianyang Zhang ∗ ∗ Department of Electrical and Computer Engineering, University of Washington † Department of Electrical and Computer Engineering, North Carolina State UniversityEmail: ∗ { liucao, haoyin, lyutiz } @uw.edu, † [email protected] Abstract —In this paper, we focus on the performance ofvehicle-to-vehicle (V2V) communication adopting the DedicatedShort Range Communication (DSRC) application in periodicbroadcast mode. An analytical model is studied and a fixedpoint method is used to analyze the packet delivery ratio (PDR)and mean delay based on the IEEE 802.11p standard in afully connected network under the assumption of perfect PHYperformance. With the characteristics of V2V communication, wedevelop the Semi-persistent Contention Density Control (SpCDC)scheme to improve the DSRC performance. We use Monte Carlosimulation to verify the results obtained by the analytical model.The simulation results show that the packet delivery ratio inSpCDC scheme increases more than 10% compared with IEEE802.11p in heavy vehicle load scenarios. Meanwhile, the meanreception delay decreases more than 50%, which provides morereliable road safety.
Index Terms —V2V, DSRC, IEEE 802.11p, DCF, MAC design
I. I
NTRODUCTION
Vehicle-to-vehicle (V2V) communication is a cornerstone ofconnected vehicles (CVs) which are emerging as an importantcomponent of the next generation intelligent transportationsystems (ITS) [1]. As an effort to deploy CVs, technologiesand standards have been actively developed. Dedicated short-range communications (DSRC) has been tested as an enablingtechnology for V2V and V2I communications [2]. DSRCis a high-efficiency wireless communication technology usedin the smart transportation system. In V2V communicationsfor CV applications, the most important component is thebroadcast of the Basic Safety Messages (BSMs) [3]. TheBSMs are single-hop, periodic, and carry safety-related statusinformation of vehicles such as their speed, acceleration,position, and direction. Through the broadcast of BSMs byDSRC, vehicles can be aware of each other’s status, and trafficaccidents can be reduced.In DSRC, IEEE 802.11 distributed coordination function(DCF) MAC protocol has been adopted by the IEEE 802.11pstandard for DSRC applications. The MAC performance ofDSRC has been studied in some papers. The authors in [4] es-tablished a quantitative approach to describe the characteristicsof DSRC safety communication. The model in [5] providedan analytical model for the MAC protocol of DSRC underaperiodic broadcast mode. In [6], the authors also presentedthe performance of IEEE 802.11p considering both MAC andPHY layers. However, most of them only focused on analyzingthe performance where IEEE 802.11p was applied while not emphasizing how to improve the DSRC performance. Theperformance such as the PDR and packet delay will degradeheavily in high vehicle load scenarios.In this paper, we study an analytical model for IEEE802.11p in periodic broadcast mode to analyze the DSRCperformance. Since each vehicle collects the information fromother vehicles through the received BSMs from the previousperiods, they can obtain a timeline of packet generation fromothers. Each vehicle can determine its backoff counter basedon the historical information rather than randomly choosinga number in the range of a fixed contention window utilizedby IEEE 802.11p. Using this characteristics of V2V commu-nication, we develop the SpCDC scheme which shows betterperformance than IEEE 802.11p especially in heavy vehicleload scenarios.This paper is organized as follows: Section II providesan analytical model for IEEE 802.11p adopting the DCFfor channel access in periodic broadcast mode. Section IIIdevelops a new distributed scheme that enhances the DSRCperformance. Section IV compares the results of the MonteCarlo simulation with the results obtained by their analyticalmodels. Section V draws the conclusions.II. ANALYTICAL MODEL FOR IEEE 802.11 P In this section, we study an analytical model for IEEE802.11p in periodic broadcast mode. We assume perfect PHY-layer performance to simplify the analysis, i.e., any packet sentwithin a given radius can be heard perfectly if not interferedby others. Besides, we use a fixed point model to charac-terize the DSRC performance for V2V communications. Themechanism of DCF employing entire carrier sense multipleaccess with collision avoidance (CSMA/CA) procedure. Eachvehicle prepared to send a packet first senses the channel fora period, which is known as the distributed inter-frame space(DIFS). If the channel is sensed busy during this period, theaccess will be deferred and wait for a complete transmissionfrom the other vehicle. A backoff process will initiate after thechannel becomes idle again for a DIFS. Before the backoffprocess, the vehicle needs to choose a random number withina fixed contention window as the initial backoff counter whichdecrements by one every time. The counter during the backoffprocess is suspended when a transmission is detected in thechannel and will be reactivated after the channel is sensed idleagain for a DIFS. When the counter reaches zero, the vehicle a r X i v : . [ c s . N I] F e b ends the packet instantly. Otherwise, if a vehicle senses thechannel idle in the first whole DIFS period, it will occupythe channel and send the packet directly. In broadcast mode,the transmitter vehicle doesn’t need acknowledgements(ACKs)from other vehicles since gathering the information from allvehicles will lead to a prohibitively high overhead. Thus, thereis no re-transmission or increment of the contention windoweven if a packet collision occurs. A. Packet delivery ratio
PDR is defined as the probability that a BSM (we willalways refer BSM as packet) from the tagged vehicle is suc-cessfully broadcasted to all other vehicles in its transmissionrange. We define ρ as the probability of a packet staying atthe buffer for each vehicle, which can be expressed as ρ = λ E [ S ] , (1)where E [ S ] is the average service time for a packet stayingat the buffer. λ is the packet transmission frequency, whichindicates each vehicle regularly generates packet every /λ seconds. Define p b as the probability that the channel is sensedbusy when a new packet arrives, which is given by p b = ( N tr − λT tr (cid:18) − ( n c − n c p c (cid:19) , (2)where N tr is the number of vehicles in the network. n c is theaverage number of collided packets if a collision occurs. p c is the collision probability which will be introduced later. Afixed transmission delay T tr is T tr = E [ P ] R d + T H + δ, (3)where E [ P ] is the mean packet length of payload and R d is thedata rate. T H is the duration of transmitting the packet headersincluding physical layer header and MAC layer header. δ is thepropagation delay, in this paper, δ = 0 . In periodic broadcastmode, two cases will happen when a packet arrives: • Case 1: Vehicle immediately sends the packet withoutperforming a backoff process if the channel is sensedidle for a DIFS period. • Case 2: The packet will performance a backoff processbefore transmission if the channel is sensed busy. Thecorresponding probability is given by p b .Here we don’t consider the case where a previous packetwaiting for a long time due to the backoff process will bereplaced by a new arriving packet since the packet inter-arrival time is deterministic ( /λ seconds) and it is muchlonger than any possible packet delay. DCF employs a discretetime slot backoff scheme, if a backoff process is involved, thetransmission is synchronized to the beginning of a time slot[7]. Therefore, packet collision only occurs in the second casein this paper.We construct a model to characterize the backoff counterfor the IEEE 802.11p broadcast network. The backoff counter,which indicates the counter value of a broadcast vehicle, isa one-dimensional discrete time Markov Chain. The state transition diagram describing the decrements of a backoffcounter is shown in Fig.1. The non-null one-step transition Fig. 1: Markov Chain for backoff counter probabilities are (cid:26) P ( M k +1 = m − | M k = m ) = 1 m ∈ [1 , CW − P ( M k +1 = m | M k = 0) = 1 /CW m ∈ [0 , CW − (cid:27) , (4)where CW is a fixed contention window, and M k is the valuethat backoff counter reaches at discrete time k . The followingrelations can be derived according to Eq. (4) (cid:26) π m = CW − mCW π (cid:80) CW − m =0 π m = 1 (cid:27) , (5)where m ∈ [0 , CW − . π m is the probability that the counterreaches m during the backoff process. π is the probability thata vehicle starts to transmit the packet since the counter reacheszero. We can obtain π by solving Eq. (5) π = 21 + CW . (6)For any vehicle other than the tagged vehicle, the probabilityof transmitting a packet is ρπ given that the second casehappens. A collision occurs when at least one vehicle sendpacket in the same slot as the tagged vehicle. Thus, thecollision probability can be written as p c = p b (cid:16) − (1 − ρπ ) N tr − (cid:17) , (7)meanwhile PDR is P DR = 1 − p c . (8) B. Mean delay
The service time S includes the access delay T A and thetransmission delay T tr . The access delay is defined as theinterval between the instant the packet reaches the head of thequeue and the instant when the packet transmission begins.The end-to-end delay T s experienced by a packet is T s = Q + S = Q + T A + T tr , (9)where Q and T A are random variables(r.v.) indicating thequeuing delay and the access delay. For periodic broadcastmode, the queuing delay is zero for each packet. The accessdelay is classified as: • For case 1, the access delay is a DIFS period since packetdoesn’t perform a backoff process. • For case 2, the packet needs to wait for ongoing packettransmission and then performs a backoff process.ore precisely, the access delay can be summarized as fol-lows: T A = (cid:26) DIF S w.p. − p b T res + DIF S + T B w.p. p b , (10)where T res is the residual lifetime of an ongoing packettransmission, and T B is the backoff duration.Each slot in the backoff process can be interrupted bya transmission from other packets. During the interruption,the backoff counter is suspended. When the backoff counteris resumed, it starts from the beginning of the interruptedslot after deferring for a DIFS period. Therefore, the backoffduration T B is T B = M (cid:88) n =1 ( σ + T I ) , (11)where σ is the duration of a time slot, r.v. T I is the interruptionduration per slot, and r.v. M is the backoff counter value. Ifno other vehicle send packets in a given slot, an interruptiondoes not occur, which indicates T I is equal to zero. The slotwill be interrupted when at least one another vehicle sends apacket in that slot. T I can be expressed as T I = (cid:40) w.p. (1 − ρπ ) N tr − T tr + DIF S w.p. − (1 − ρπ ) N tr − (cid:41) . (12)Since M an T I are two r.v.s, the backoff duration T B is sumof a random number of r.v.s. The mean of T B is found readilyby using conditional expectation and as a result it is E [ T B ] = ( σ + E [ T I ]) E [ M ] . (13)As M is a r.v. which is uniformly distributed in the range [0 , CW − , we can get E [ M ] = CW − . (14)Meanwhile, the mean of interruption time T I is E [ T I ] = (cid:16) − (1 − ρπ ) N tr − (cid:17) ( T tr + DIF S ) . (15)From Eq. (10), the mean of access delay T A is obtained by E [ T A ] = DIF S + p b ( E [ T B ] + E [ T res ]) , (16)where T res follows the uniform distribution. Thus, the meanof T res is E [ T res ] = T tr DIF S. (17)Now we can get the mean delay E [ T s ] which is actually equalto the average service time: E [ T s ] = E [ S ] = E [ T A ] + T tr . (18)The reception delay T re describes how long other vehiclescan receive a packet from the tagged vehicle, which includesthe service time S and the possible collision delay T c . Thecollision delay is caused by the packet loss when collision occurs. Since the collision probability is p c , the mean collisiondelay follows a geometric distribution and is given by E [ T c ] = 1 λ ∞ (cid:88) n =1 np nc (1 − p c ) = p c (1 − p c ) λ . (19)Thus, the mean reception delay is E [ T re ] = E [ S ] + E [ T c ] . (20)III. IMPROVEMENT ON DSRC PERFORMANCEThe objective of this section is to develop a dis-tributed scheme - Semi-persistent Contention Density Control(SpCDC) aiming to improve the DSRC performance especiallyin heavy vehicle load scenarios. The tagged vehicle maintainsa timeline and marks the slots when other vehicles generatetheir packets through the received packets in the previousperiods. In a new transmission period, when the tagged vehiclereceives packets from neighbor vehicles before it generates apacket, it will know the packets from those vehicles are nolonger contending for channel access in this current period.The scenario of contending for channel access happens whenthe neighbor vehicles have generated packets but the taggedvehicle hasn’t yet received them at the instant it generates apacket. By counting the number of these packets, the taggedvehicle will know the instantaneous contention density anddetermine its backoff counter [8]. A. Analytical model for SpCDC scheme
Denote the number of packets contending for channel accessmeasured at the beginning of slot k as c ( k ) . Let S ( k ) = 1 and S ( k ) = 0 represent the events that slot k is sensed busyand idle. If slot k is sensed busy, the initial backoff counterof new generated packets arriving at slot k [ m ] (There are T tr σ mini-slots in slot k, m ∈ V , and V = (cid:8) , , ..., T tr σ (cid:9) ) willbe stopped until the ongoing transmission ends. If slot k isidle, m = 1 . The initial backoff counter b ( k [ m ]) of a packetarriving at slot k [ m ] depends on the instantaneous contentiondensity. Denote the number of packets that arrives at slot k [ m ] measured at the m th mini-slot in slot k as n a ( k [ m ]) . Denotethe number of packets with their backoff counters reducingto 0 at slot k as n t ( k ) . The framework of SpCDC is givenin Algorithm 1 where C is SpCDC protocol parameter, and R indicates the state whether the vehicle enters a new semi-persistent period. ω is the changed amount of the backoffcounter value based on contention density at the beginningof each semi-persistent period. It is randomly selected fromset {− , , } with equal probability.Since the expected change of c ( k ) in one slot is E { ∆ c ( k ) } = (cid:26) λN tr σ if S ( k ) = 0 λN tr T tr − n b if S ( k ) = 1 (cid:27) , (21)where σ is the duration of a time slot. T tr is the transmissiondelay. n b is the average number of packets in a busy slot.The probability of a slot being sensed idle and busy are givenrespectively by P ( S ( k ) = 0) = P ck (0) + (cid:0) − P ck (0) (cid:1) (1 − γ ) (22) lgorithm 1 Framework of Semi-persistent Contention Den-sity Control
Require:
Maintaining a list of timeline of packet generationsbased on the previous transmission periods.
Ensure:
A packet of tagged vehicle is generated and justarrives at the buffer, waiting to be sent. if S ( k ) = 1 then b ( k [ m ]) = C · ( c ( k ) + (cid:80) ms =1 n a ( k [ s ])) , m ∈ V if R = 1 then b ( k [ m ]) = b ( k [ m ]) + ω, ω ∈ {− , , } else b ( k [ m ]) = b ( k [ m ]) end if S ( k + b ( k [ m ])) = 1 c ( k + 1) = c ( k ) + (cid:80) s : s(cid:15)V n a ( k [ s ]) − n t ( k ) else b ( k [1]) = C · ( c ( k ) + n a ( k [1])) if R = 1 then b ( k [1]) = b ( k [1]) + ω, ω ∈ {− , , } else b ( k [1]) = b ( k [1]) end if S ( k + b ( k [1])) = 1 c ( k + 1) = c ( k ) + n a ( k [1]) end if P ( S ( k ) = 1) = (cid:0) − P ck (0) (cid:1) γ, (23)where P ck (0) is the probability of no packet contending forchannel access at slot k, i.e., c ( k ) = 0 . γ indicates how manypackets each backoff slot accommodates in average, and it isan approximate probability that the slot k is sensed busy givenat least one contending packet. Since the expected change of c ( k ) should be equal to 0 in the steady state, it holds E { ∆ c ( k ) | S ( k ) = 0 } P ( S ( k ) = 0)+ E { ∆ c ( k ) | S ( k ) = 1 } P ( S ( k ) = 1) = 0 . (24)Therefore, γ can be obtained by plugging Eq. (21), (22) and(23) into Eq. (24) γ = λN tr σ (cid:0) − P ck (0) (cid:1) ( n b − λN tr ( T tr − σ )) . (25)The average number of packets in a busy slot n b is greaterthan 1 due to packet collision. Assume each collision onlyinvolves two packets with collision probability P c , n b is givenby n b = 1 + P c . (26)Now we start to derive the mean delay for channel accesswhich includes the busy slots and idle slots during the backoffprocess. Suppose the arrival of a new packet is uniformly distributed in a busy slot, the mean of duration of busy slots T db is E [ T db ] = N tr − (cid:88) j =1 P ck ( j ) (cid:18) j − (cid:19) T tr = (cid:18) c s + 12 (cid:0) P ck (0) (cid:1)(cid:19) T tr , (27)where c s is the mean contention density, and P ck ( j ) = P ( c ( k ) = j ) , i.e., the probability of j packets contendingfor channel access. According to the computed initial backoffcounter and the number of busy slots, the mean of duration ofidle slots T di is E [ T di ] = ( C · ( c s + 1) − c s ) σ. (28)Given E [ T db ] and E [ T di ] , we can get the mean delay E [ T d ] E [ T d ] = E [ T db ] + E [ T di ] , (29)while the mean reception delay is E [ T re ] = E [ T d ] + E [ T c ] . (30)Since each packet arrives every /λ seconds, the probabilityof a packet staying at the buffer is E [ T d ]1 /λ . The mean contentiondensity should satisfy c s = ( N tr − E [ T d ]1 /λ = λ ( N tr − E [ T d ] . (31)We can also obtain the mean contention density c (cid:48) s in IEEE802.11p based on Eq. (14) and (15): c (cid:48) s = ( CW − (cid:16) − (1 − ρτ ) N tr − (cid:17) . (32)Given the probability of one packet staying at the buffer, theprobability that no packet is contending for channel accessover N tr − vehicles is P ck (0) = (1 − λ E [ T d ]) N tr − = (cid:18) − c s N tr − (cid:19) N tr − . (33)For an arbitrary k , we consider the worst case so that we canderive the upper bound of collision probability. In the worstcase, the initial backoff counter of an incoming packet alwaysholds b ( k [ m ]) < CW ( k ) where CW ( k ) is the contentionwindow at slot k . If a collision occurs in the initial slot inthe backoff process, the collision probability will be γ . If nocollision occurs in the initial backoff slot, the collision mayoccur in the remaining C ( c s + 1) − slots. Suppose slotsare independent with each other, the collision probability ineach slot is given by − (1 − γ ) c s , Thus, the upper bound ofcollision probability is P upperc = (1 − P ck (0) )( γ + (1 − γ )(1 − (1 − γ ) c s ) C ( c s +1)) − ) , (34)and the lower bound of PDR is P DR lower = 1 − P upperc . (35) ABLE I: DSRC communication parameters
Parameters Values
Packet length (payload), E [ P ] λ
2, 10 ppsSlot time, σ
16 usPropagation delay, δ CW R d
6, 12, 24 Mbps
IV. RESULTS OF ANALYTICAL MODEL ANDSIMULATIONIn this section, we present a simulation setup used tovalidate our analytical model and give validation results. Thecomputation for analytic models with corresponding simula-tions are conducted in Matlab. All assumptions are the samein the simulation and analytical models. Each vehicle on thelanes is equipped with DSRC wireless capability with perfectPHY-layer performance. Since vehicle can communicate witheach other in a fully connected network, the location of eachvehicle doesn’t impact their performance. We also use TableI’s parameters for the simulation in the SpCDC scheme, wherethe protocol parameter C = 3 and a complete semi-persistentperiod is 1 second for each vehicle.Fig.2 shows the DSRC performance in IEEE 802.11p asa function of number of vehicles, with different curves pa-rameterized by data rate R d (in megabits per second), packettransmission frequency λ (packets per second) and meanpacket length E [ P ] (in bytes). The analytic model agrees wellwith the simulation results. In the plotted range, the averagedelay increases almost linearly with the vehicle density exceptfor the case of 6 Mbps/10 packets per second/ 200 bytes.Meanwhile, the PDR in this case also drops markedly with theincreasing vehicle load. The reason why this case differs fromother cases is caused by more interruptions during backoffprocess and higher transmission delay.We observe the improvement on DSRC performance ac-cording to the developed model. First, as Fig.3 shows, theanalytical model matches well with the simulation results ofthe mean delay, which validates our model. Next, we onlyfocus on a typical case (6 Mbps/10 pps/200 bytes), which bestapproximates the parameters in a real situation. Fig.4(a) showsthe mean delay in IEEE in 802.11p with different contentionwindows and SpCDC scheme. The mean delay in SpCDCscheme is always below that in 802.11p with CW=128 whilebeing very close to that in 802.11p with CW=16. Fig.4(b)shows the contention density among them. When the numberof vehicles is 200, the contention density in SpCDC scheme isaround seven fewer than that in IEEE 802.11p with CW=128.Since the transmission delay is around 0.5 ms, the meandelay difference between 802.11p with CW=128 and SpCDCscheme will be more than 3 ms. (a) Mean delay(b) PDR Fig. 2: Performance in IEEE 802.11pFig. 3: Mean delay in SpCDCFig.5(a) presents the simulation results for PDR in IEEE802.11p and SpCDC scheme and the analytical lower boundof PDR for SpCDC scheme. The PDR in SpCDC increasesnearly 15% compared with that in IEEE 802.11p with CW=16and 10% with CW=128 in heavy vehicle loads. Besides, wecan also observe the analytical lower bound is not very tightespecially in heavy vehicle loads since the lower bound ofPDR is derived under the assumption of worst case. Nev-ertheless, even the lower bound lies above the performanceof PDR in IEEE 802.11p. Fig.5(b) provides the simulationresult for the mean reception delay between SpCDC and a) Mean delay(b) Contention density
Fig. 4: Comparison of the contention densityIEEE 802.11p. As the result shows, the mean reception delayin SpCDC scheme is much lower than IEEE 802.11p withdifferent contention windows, even the upper bound of meandelay in SpCDC is lower nearly 50% than IEEE 802.11pwith CW=128. This result indicates DSRC adopting SpCDCscheme can receive more timely BSMs in a long periodcompared with IEEE 802.11p. In other words, SpCDC schemeprovides more reliable road safety than IEEE 802.11p bylowing down the mean reception delay for each vehicle.V. CONCLUSIONIn this paper, we first focused on the performance analysisof DSRC performance adopting IEEE 802.11p in periodicbroadcast mode. With the assumption of a perfect PHY perfor-mance and the fixed point method, we presented the PDR andpacket delay in a fully connected network. Our analytic modelprovided a good match with simulation results. Then we devel-oped the SpCDC scheme to improve DSRC performance. Bycomparing the SpCDC scheme with IEEE 802.11p with somemetrics such as PDR and mean reception delay, we can verifythat SpCDC improves DSRC performance. Furthermore, it ispossible to partially adjust this scheme which can be appliedin the Listen Before Talk protocol based short-term sensing inNR V2X. (a) PDR(b) Mean reception delay
Fig. 5: Comparison of the PDR and reception delayACKNOWLEDGEMENTSThe authors would like to express the special thanks to Prof.Randall Berry and Prof. James Ritcey for helping discuss andreview this work! R
EFERENCES[1] N. Lu, N. Cheng, N. Zhang, X. Shen, and J. W. Mark, “Connectedvehicles: Solutions and challenges,”
IEEE internet of things journal ,vol. 1, no. 4, pp. 289–299, 2014.[2] J. B. Kenney, “Dedicated short-range communications (dsrc) standards inthe united states,”
Proceedings of the IEEE , vol. 99, no. 7, pp. 1162–1182,2011.[3] B. Kloiber, J. H¨arri, T. Strang, S. Sand, and C. R. Garcia, “Randomtransmit power control for dsrc and its application to cooperative safety,”
IEEE Transactions on Dependable and Secure Computing , vol. 13, no. 1,pp. 18–31, 2015.[4] X. Chen, H. H. Refai, and X. Ma, “A quantitative approach to evaluatedsrc highway inter-vehicle safety communication,” in
IEEE GLOBECOM2007-IEEE Global Telecommunications Conference , pp. 151–155, IEEE,2007.[5] M. I. Hassan, H. L. Vu, and T. Sakurai, “Performance analysis of the ieee802.11 mac protocol for dsrc safety applications,”
IEEE Transactions onvehicular technology , vol. 60, no. 8, pp. 3882–3896, 2011.[6] A. Bazzi, B. M. Masini, A. Zanella, and I. Thibault, “On the performanceof ieee 802.11 p and lte-v2v for the cooperative awareness of connectedvehicles,”
IEEE Transactions on Vehicular Technology , vol. 66, no. 11,pp. 10419–10432, 2017.[7] X. Yin,
Performance and reliability evaluation for DSRC vehicular safetycommunication . PhD thesis, Duke University, 2013.[8] J. Gao, M. Li, L. Zhao, and X. Shen, “Contention intensity baseddistributed coordination for v2v safety message broadcast,”