Performance Analysis of the IEEE 802.11 MAC Protocol for DSRC Safety Applications

3882 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 8, OCTOBER 2011 Performance Analysis of the IEEE 802.11 MAC Protocol for DSRC Safety Applications Md. Imrul Hassan, Student Member, IEEE, Hai L. Vu, Senior Member, IEEE, and Taka Sakurai, Member, IEEE Abstract—In this paper, we evaluate and improve the perfor- mance of the medium-access control (MAC) protocol for safety applications in a dedicated short-range communication (DSRC) environment. We first develop an analytical model to study the IEEE 802.11 distributed coordination function (DCF) MAC pro- tocol that has been adopted by the IEEE 802.11p standard for DSRC. Explicit expressions are derived for the mean and standard deviation of the packet delay, as well as for the packet delivery ratio (PDR) at the MAC layer in an unsaturated network formed by moving vehicles on a highway. The proposed model is validated using extensive simulations, and its superior accuracy is compared with that of other existing models is demonstrated. Insights gained from our model reveal that the principal reason for the low PDR of the DCF protocol is packet collision due to transmissions from hidden terminals. We then present a novel protocol based on the DCF that uses an out-of-band busy tone as a negative acknowl- edgment to provide an efficient solution to the aforementioned problem. We extend our analytical model to the enhanced protocol and show that it preserves predictive accuracy. Most importantly, our numerical experiments confirm that the enhanced protocol improves the PDR by up to 10% and increases the supported vehicle density by up to two times for a range of packet arrival rates while maintaining the delay below the required threshold level. Index Terms—Dedicated short-range communication (DSRC), medium-access control (MAC), performance analysis, safety applications. I. INTRODUCTION D EDICATED short-range communication (DSRC)refers to the use of vehicle-to-vehicle and vehicle-to- infrastructure communications to improve road safety and increase transportation efficiency. While there are no commercial DSRC systems yet, recent years have seen a dramatic increase in research and development activity in the DSRC field [1]. An important DSRC application is cooperative collision avoidance (CCA), where moving cars form a network to wirelessly communicate and warn each other of changing conditions or dangers ahead on the road to avoid accidents [2]. This application requires timely communication of safety messages between vehicles with high reliability, and the Manuscript received August 16, 2010; revised January 27, 2011 and May 16, 2011; accepted July 7, 2011. Date of publication July 29, 2011; date of current version October 20, 2011. The review of this paper was coordinated by Prof. R. Jäntti. M. I. Hassan and H. L. Vu are with the Centre for Advanced Internet Archi- tectures, Faculty of Information and Communication Technologies, Swinburne University of Technology, Hawthorn, Vic. 3122, Australia. T. Sakurai is with the Department of Electrical and Electronic Engineering, The University of Melbourne, Melbourne, Vic. 3010, Australia. Digital Object Identifier 10.1109/TVT.2011.2162755 medium-access control (MAC) protocol has a vital role to play. In this paper, we develop an accurate performance model for the IEEE 802.11 distributed coordination function (DCF) MAC protocol that has been adopted by the IEEE 802.11p standard for DSRC applications [3]. We find that the standard broadcast protocol yields a low probability of successful message delivery for CCA, and we respond by proposing and modeling an enhanced protocol involving retransmissions to improve the reliability of message delivery. In the survey paper [4], different MAC protocols for vehicle-to-vehicle communication networks were compared. Borgonovo et al. [5] proposed a distributed access technique called Reliable Reservation ALOHA (RR-ALOHA), which can dynamically establish a reliable single-hop broadcast chan- nel on a slotted/framed structure. The authors presented the mechanisms that compose the new MAC, namely, the basic RR-ALOHA protocol, which is an efficient broadcast service and the reservation of point-to-point channels that exploit par- allel transmissions. A directional antenna-based MAC protocol called D-MAC is proposed in [6], which uses directional an- tennas to direct transmission in specific directions. In D-MAC, by using a narrow beam, interference with parallel ongoing transmissions can be reduced. Su and Zhang [7] introduced a clustering-based DSRC architecture that takes into account both reliability and delay. The authors analyzed a cluster- based multichannel communication scheme consisting of sev- eral MAC protocols to reduce data congestion and to support QoS for real-time delivery of safety messages. In particular, most intracluster safety messages in [7] are exchanged using time-division multiple-access broadcast, whereas intercluster safety messages are aggregated by the cluster-head vehicles and sent using a contention-based access protocol. The aforementioned MAC protocols use time scheduling for multiple access, which is sensitive to mobility and topology changes and requires significant reconfiguration time. In ad- dition, coordination among vehicles requires knowledge of all neighboring vehicles, and it takes a few time cycles to agree on a stable schedule. As a result, the access delay in such a case is relatively high. A way of possibly achieving lower delay is to use a decentralized MAC protocol, such as the IEEE 802.11 DCF protocol [8] used in wireless local area networks (LANs). DCF is based on carrier sense multiple access (CSMA) and can operate with a variety of traffic loads and does not require much reconfiguration upon a change in topology. DCF has both unicast and broadcast operating modes. Broad- cast mode is appropriate for time-critical applications such as CCA because, unlike unicast, broadcast does not require the establishment of an association context between stations 0018-9545/$26.00 © 2011 IEEE HASSAN et al.: ANALYSIS OF THE IEEE 802.11 MAC PROTOCOL FOR DSRC SAFETY APPLICATIONS 3883 Fig. 1. Total delay and PDR using the following parameter set: (data rate [in megabits per second], packet arrival rate [in packets per second], packet size [in bytes]). before data communications can commence. Broadcast could use multihop transmissions to enhance coverage, but recent studies have suggested that a single-hop transmission is suf- ficient in most situations to reach all neighboring vehicles in an accident’s vicinity [1]. In the rest of this paper, we use the term “broadcast” to refer to single-hop broadcast in contrast to multihop broadcast or flooding known in the literature of wire- less ad hoc networks. The problem with broadcast mode is that it is less reliable, since it cannot support any request–response handshaking procedures that improve reliability such as con- ventional acknowledgement (ACK) or virtual carrier sensing [request to send/clear to send (RTS/CTS)], due to the risk of a “storm” of response packets. To illustrate the delay performance and reliability of broad- cast 802.11 DCF for CCA, we conducted ns-2 simulations [9] on a highway scenario. In this scenario, vehicles are represented as a collection of random and statistically identical stations in a 1-D mobile ad hoc network and are stationary during the communication interval (further details of the simulation setup can be found in Section IV). In Fig. 1, we plot the mean of the total delay and the packet delivery ratio (PDR) (the probability of successful packet delivery) versus the vehicle density, with different curves parameterized by the triplet (data rate [in megabits per second], packet arrival rate [in packets per second], packet size [in bytes]). We define PDR as the probability of delivering the packet to all intended receivers within the transmission range of a given transmitting node. It has been suggested in [10] that a suitable maximum delay requirement for cooperative collision warning and intersection collision warning applications is 100 ms. The same maximum allowable delay is also considered in [11] and [12]. It has been specified in [13] that the PDR should be not less than 90%. With respect to these performance targets, we see from Fig. 1 that, in the simulated scenarios, the delay requirement can be comfortably met, but the PDR requirement is comprehensively violated except for low packet arrival rates and very low vehicle densities. It is apparent that the conventional DCF broadcast protocol may have difficulty supporting the CCA application. To understand the underlying reasons for poor performance, we are motivated to develop an analytical model in this paper to study the behavior and to improve the performance of the protocol in unsaturated broadcast networks. The performance of the DCF has been extensively studied in the wireless LAN environment. Bianchi [14] analyzed the performance of a saturated network using a Markov chain model. In [15], Malone et al. extended the model to the un- saturated case. Tickoo and Sikdar [16] developed an alternative unsaturated model by modeling each station as a G/G/1 queue. These papers all considered unicast communications rather than broadcast communications. Rao et al. developed an analytical model to determine the probability of packet collision in the broadcast scenario [17]. However, all the aforementioned pa- pers assume a fully connected network (i.e., there are no hidden terminals). Although there exists a wide body of literature analyzing the hidden-terminal problem, several limitations of those models were highlighted in [18] for the case of unicast communication. The model in [19] attempted to capture the characteristics of the DSRC safety communications, where broadcasting takes place in an unsaturated network with hidden terminals. However, the renewal theory-based argument used in this model is not entirely suitable for hidden-terminal analysis, as also pointed out in [18]. In particular, the model in [19] predicts a nonzero successful transmission probability with arbitrary hidden collisions. However, it is shown that, when nodes are always backlogged with packets to send (i.e., in a heavy traffic scenario), the probability of successful transmis- sion can be zero. This is because the transmitting node and other hidden nodes always use the same backoff window (i.e., no retransmission is enabled), and the so-called vulnerable period1 of a node could actually be larger than its backoff window, thus guaranteeing a hidden collision. Note that the model is also inaccurate when there is very little traffic on the channel. 1The vulnerable period of a tagged node is the time period during which, if any hidden node commences a packet transmission, it will be colliding with the transmission from the tagged node. 3884 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 60, NO. 8, OCTOBER 2011 TABLE I COMPARISON OF OUR MODEL WITH THE EXISTING MODELS IN THE LITERATURE Furthermore, the IEEE 802.11 DCF protocol was not properly modeled in [19] since the analysis assumes that a backoff process is initiated for each packet at a node, irrespective of whether the channel is idle or busy. The first major contribution in this paper is an accurate analytical model for the DCF in unsaturated broadcast networks both with and without hidden terminals. Table I summarizes the scope of our model relative to that of the other models described earlier; it shows that only [19] attempts to cover the same aspects, albeit the hidden-terminal problem is not modeled accurately and is therefore marked with a “‡” in the table. We focus on the packet delay and the PDR as the two main performance metrics of interest in our study. Our model uses a mean-value decoupling approximation for the collision probability, and we apply an M/G/1/∞ queuing model for each station to obtain the total packet delay. While we make necessary assumptions to keep the model simple, we show via comparison with simulation that the results are nevertheless accurate. We also provide a comparison with results from the existing model in [19] to demonstrate the superior accuracy of our model. The numerical experiments with our model reveal that the principal reason for the low PDR of the DCF is packet collision due to transmissions from hidden terminals. Our second major contribution is a set of modifications to the DCF to improve the PDR when hidden terminals are present. The essential idea is to use retransmissions to trade increased delay for decreased packet loss (i.e., higher PDR). We propose that, in the event of detection of an errored packet, the receiving stations transmit a negative ACK (NACK) in the form of a busy tone signal in a narrow out-of-band channel. We assume that all nodes are equipped with an additional transceiver to detect such an out-of-band signal. Note that a transceiver with a simple energy detector would be enough to detect the NACK signal. Senders of recent packets that hear the NACK (but that could be hidden from each other) shall then execute a backoff process and resend their last packet, and this can be repeated up to a maximum number of retransmissions. To reduce the chance of successive collisions between hidden terminals, we distinguish between senders according to the delay in receiving the busy tone and assign different backoff contention windows depending on whether the delay is short or long. We extend our DCF analytical model to the enhanced protocol and show that our extended model preserves predictive accuracy. Most importantly, our numerical experiments confirm that, compared with the DCF, the enhanced protocol gives a higher PDR for a wider range of vehicle densities and packet arrival rates while maintaining the delay below the required threshold level. The results demonstrate that our enhancement can bring the standard a step closer to the ultimate solution for safety ap- plications, which is to meet the strict QoS requirements under arbitrary vehicle densities and network conditions. The rest of this paper is organized as follows: In Section II, we provide an overview of DSRC challenges and standardiza- tion activities. In Section III, we describe our analytical model for the DCF. We verify the accuracy of our DCF model by com- parison with simulation in Section IV. Then, in Section V, we describe our enhanced protocol and illustrate its performance. Finally, we conclude this paper in Section VI. II. DEDICATED SHORT-RANGE COMMUNICATIONS CHALLENGES AND STANDARDIZATION ACTIVITIES The primary goals of emerging DSRC systems are to enhance road safety and to improve transportation efficiency. In this paper, we focus on the safety aspect. Road safety is supported by the transmission of routine status messages and event-driven emergency messages. Routine status messages are periodically sent to neighboring vehicles to inform them of the current status of the originating vehicle (e.g., location, speed, and direction), whereby the receiving vehicles/drivers can then anticipate any potential hazards (e.g., traffic jam ahead) and take necessary action. Event-driven safety messages are triggered by rapid changes in vehicle behavior such as a hard brake or an airbag explosion. To enable preventative action, it is essential that both types of safety messages are correctly received by surrounding vehicles in a timely fashion. One of the main challenges to achieve that objective is the loss of packets due to the presence of hidden terminals. This occurs when a node is transmitting to a target node while a third node that is unaware of the transmitter also starts its transmission and causes interference at the receiver. The hidden-terminal problem can afflict all decentralized wireless networks but is particularly severe in broadcast scenarios. In the broadcast case, there are multiple receivers for each message, which are scattered in the transmission range of the sender. Any node that is within the sensing range of any receiver but outside the transmission range of the sender is a potential hid- den terminal. Therefore, the potential hidden-terminal region is significantly larger than that for unicast communication. The distinctive demands of DSRC applications, as well as the unique operating environment involving fast moving vehicles, mean that specifically tailored communication protocols are required for DSRC systems. The following sections describe the protocols being standardized by the IEEE standards body for DSRC. A. IEEE WAVE In 1999, the U.S. Federal Communication Commission (FCC) allocated 75 MHz of spectrum in the 5.9-GHz band for DSRC use. The DSRC spectrum is divided into seven 10-MHz- wide channels and a reserved 5-MHz channel. One of the 10-MHz channels, which is called the control channel, is restricted to safety communications only, whereas the other channels are available for both safety and nonsafety usage. HASSAN et al.: ANALYSIS OF THE IEEE 802.11 MAC PROTOCOL FOR DSRC SAFETY APPLICATIONS 3885 Fig. 2. IEEE 802.11 DCF basic access [8]. The initial effort at standardizing DSRC radio technology took place in the American Society for Testing and Materials (ASTM) 2313 working group [13]. Recently, the IEEE Wireless Access in Vehicular Environment (WAVE) project has pub- lished specifications for the FCC DSRC spectrum based on an orthogonal frequency-division multiplexing (OFDM) air inter- face. IEEE WAVE encompasses the IEEE 802.11p standard [3] for MAC and physical layers and the IEEE 1609 family of stan- dards, which define the higher layer protocols and the protocol architecture [20]. IEEE 802.11p is based on IEEE 802.11a but with modifications to support vehicular communications with low latency. The 802.11p MAC protocol, such as other 802.11 variants, uses the DCF for channel access. B. IEEE 802.11 DCF Protocol Description In the IEEE 802.11 DCF, nodes contend for the channel using a CSMA mechanism with collision avoidance, as shown in Fig. 2. When a node has a packet to send, the channel must be sensed idle for a guard period, which is known as the distributed interframe space (DIFS). If during that period of time the channel becomes busy, then the access is deferred until the channel becomes idle again and a backoff process is initiated. Backoff intervals are slotted, and stations are only permitted to commence transmissions at the beginning of slots. The discrete backoff time is uniformly distributed in the range [0, CW − 1], where CW is called the contention window. At the first transmission attempt, CW is set equal to W , which is the minimum contention window. The backoff time counter is decremented by 1 at the end of each idle slot. It is frozen when a packet transmission is detected on the channel and reactivated after the channel is sensed idle again for a guard period. The guard period is equal to a DIFS if the transmitted packet was error free, and it is equal to the extended interframe space if the packet was in error. The station transmits when the backoff counter reaches zero. A collision occurs when the counters of two or more stations reach zero in the same slot. After every successful data packet transmission, a station initiates a post- transmission random backoff process. If the next packet was already enqueued when the previous packet was sent, its defer time will span the entire backoff period, whereas a packet that arrives at the MAC layer after the previous packet was sent would experience only part of the backoff period, or none at all, if the backoff period has already elapsed. As already mentioned, DSRC safety messages are transmit- ted in broadcast mode, which is different in several ways from unicast communication for the IEEE 802.11 DCF. First, there is no ACK sent after the successful reception of a data packet; thus, the sender is unaware of any packe