[ieee 2011 seventh international conference on mobile ad-hoc and sensor networks (msn) - beijing,...

6
Maximizing Saturation Throughput of Control Channel in Vehicular Networks Shan Wang, An Song, Jibo Wei School of Electronics Science and Engineering National University of Defense Technology Changsha, China {chinafir, ansong, wjbhw}@nudt.edu.cn Abdelhakim Hafid Network Research Laboratory University of Montreal Montreal, Canada [email protected] Abstract—Described in the specifications of WAVE (Wireless Access in Vehicular Environments) standards, broadcast is the main traffic in vehicular networks when all vehicles monitor the control channel. In this paper, we show a simple Markov model to analyze the saturation throughput of control channel broadcast; our analysis reveals that existing IEEE 802.11p parameter settings can result in degraded network performance. In particular, the argument that the contention window size determines the performance of broadcast is concluded. Moreover, we propose a novel scheme which can achieve an optimal throughput by adapting the contention window size to the networks size. Both theoretical analyses and simulation results show the effectiveness of our proposal. Keywords- MAC; saturation throughput; contention window; networks size; broadcast; vehicular networks I. INTRODUCTION In recent years, the vehicular communication system has been one of the most important realizations of ad hoc networks. The latest IEEE 802.11p-2010 [1] was formed as a draft amendment to the IEEE 802.11 standard to specify the wireless access in vehicular environments. Furthermore, almost all Vehicular Ad hoc Networks (VANETs) adopt Dedicated Short-Range Communication (DSRC) [2] to provide transportation- oriented services. Its Media Access Control (MAC) protocol based on the legacy CSMA/CA is a distributed random access scheme with priorities, called the Enhanced Distributed Channel Access (EDCA) [3]. In [1][3], the IEEE 802.11p spectrum band is divided into seven 10MHz channels, composed of one control channel (CCH) and six service channels (SCH). The control channel is assigned for control and safety messages and the service channels are for both safety and non-safety usage. All devices participating in the networks must monitor the CCH for safety and private service advertisements during specific intervals. Outside these intervals, user data can be exchanged in the relevant SCH. Thus, the communication period, in VANETs, is divided into two intervals: control channel interval (CCHI) and service channel interval (SCHI). Whatever the ratio of these two intervals is, all vehicles must tune to CCH during CCHI (see Fig. 1). Figure 1. Channel access process of the IEEE 802.11p/1609.4 The main traffic, when vehicular stations use CCH, consists of WAVE Short Messages (WSMs) and WAVE Service Announcements (WSAs) [3]; the common feature of these two kinds of packets is that both mainly use broadcast transmission. Thus, the throughput performance of control channel broadcast determines the number of WSMs and WSAs exchanged during CCHI; this represents the efficiency and the capacity of vehicular networks. In this paper, we analyze the saturation throughput of vehicular networks using a Markov model which are developed for broadcast communication. Our analysis shows that the saturation throughput is far smaller than the capacity of the channel because of the use of a fixed contention window (CW) size. In order to maximize this throughput, we propose a scheme that adapts the contention window size to the networks size which is defined as the number of vehicles competing for the same channel (i.e., the number of one-hop neighboring stations). This paper is organized as follows. Section II presents related work. In Section III, we show a Markov model to analyze the saturation throughput of control channel; we also discuss the deficiency of existing MAC 2011 Seventh International Conference on Mobile Ad-hoc and Sensor Networks 978-0-7695-4610-0/11 $26.00 © 2011 IEEE DOI 10.1109/MSN.2011.44 298 2011 Seventh International Conference on Mobile Ad-hoc and Sensor Networks 978-0-7695-4610-0/11 $26.00 © 2011 IEEE DOI 10.1109/MSN.2011.44 297 2011 Seventh International Conference on Mobile Ad-hoc and Sensor Networks 978-0-7695-4610-0/11 $26.00 © 2011 IEEE DOI 10.1109/MSN.2011.44 297

Upload: abdelhakim

Post on 16-Mar-2017

217 views

Category:

Documents


0 download

TRANSCRIPT

Maximizing Saturation Throughput of Control Channel in Vehicular Networks

Shan Wang, An Song, Jibo Wei School of Electronics Science and Engineering

National University of Defense Technology Changsha, China

{chinafir, ansong, wjbhw}@nudt.edu.cn

Abdelhakim Hafid Network Research Laboratory

University of Montreal Montreal, Canada

[email protected]

Abstract—Described in the specifications of WAVE (Wireless Access in Vehicular Environments) standards, broadcast is the main traffic in vehicular networks when all vehicles monitor the control channel. In this paper, we show a simple Markov model to analyze the saturation throughput of control channel broadcast; our analysis reveals that existing IEEE 802.11p parameter settings can result in degraded network performance. In particular, the argument that the contention window size determines the performance of broadcast is concluded. Moreover, we propose a novel scheme which can achieve an optimal throughput by adapting the contention window size to the networks size. Both theoretical analyses and simulation results show the effectiveness of our proposal.

Keywords- MAC; saturation throughput; contention window; networks size; broadcast; vehicular networks

I. INTRODUCTION In recent years, the vehicular communication system

has been one of the most important realizations of ad hoc networks. The latest IEEE 802.11p-2010 [1] was formed as a draft amendment to the IEEE 802.11 standard to specify the wireless access in vehicular environments. Furthermore, almost all Vehicular Ad hoc Networks (VANETs) adopt Dedicated Short-Range Communication (DSRC) [2] to provide transportation-oriented services. Its Media Access Control (MAC) protocol based on the legacy CSMA/CA is a distributed random access scheme with priorities, called the Enhanced Distributed Channel Access (EDCA) [3]. In [1][3], the IEEE 802.11p spectrum band is divided into seven 10MHz channels, composed of one control channel (CCH) and six service channels (SCH). The control channel is assigned for control and safety messages and the service channels are for both safety and non-safety usage. All devices participating in the networks must monitor the CCH for safety and private service advertisements during specific intervals. Outside these intervals, user data can be exchanged in the relevant SCH. Thus, the communication period, in VANETs, is divided into two intervals: control channel interval (CCHI) and service channel interval (SCHI).

Whatever the ratio of these two intervals is, all vehicles must tune to CCH during CCHI (see Fig. 1).

Figure 1. Channel access process of the IEEE 802.11p/1609.4

The main traffic, when vehicular stations use CCH, consists of WAVE Short Messages (WSMs) and WAVE Service Announcements (WSAs) [3]; the common feature of these two kinds of packets is that both mainly use broadcast transmission. Thus, the throughput performance of control channel broadcast determines the number of WSMs and WSAs exchanged during CCHI; this represents the efficiency and the capacity of vehicular networks.

In this paper, we analyze the saturation throughput of vehicular networks using a Markov model which are developed for broadcast communication. Our analysis shows that the saturation throughput is far smaller than the capacity of the channel because of the use of a fixed contention window (CW) size. In order to maximize this throughput, we propose a scheme that adapts the contention window size to the networks size which is defined as the number of vehicles competing for the same channel (i.e., the number of one-hop neighboring stations).

This paper is organized as follows. Section II presents related work. In Section III, we show a Markov model to analyze the saturation throughput of control channel; we also discuss the deficiency of existing MAC

2011 Seventh International Conference on Mobile Ad-hoc and Sensor Networks

978-0-7695-4610-0/11 $26.00 © 2011 IEEE

DOI 10.1109/MSN.2011.44

298

2011 Seventh International Conference on Mobile Ad-hoc and Sensor Networks

978-0-7695-4610-0/11 $26.00 © 2011 IEEE

DOI 10.1109/MSN.2011.44

297

2011 Seventh International Conference on Mobile Ad-hoc and Sensor Networks

978-0-7695-4610-0/11 $26.00 © 2011 IEEE

DOI 10.1109/MSN.2011.44

297

parameter settings in VANETs. In Section IV, we study the impact of the contention window size on the throughput performance; then, we propose a scheme that adjusts the initial window size based on the networks size with the objective to maximize the throughput of control channel. Section V evaluates the proposed scheme via simulations. Finally, Section VI concludes the paper.

II. RELATED WORK The performance analysis of CSMA/CA-based

schemes has been thoroughly studied in [4]. Bianchi proposed a classical analytical model to calculate the saturation throughput of the legacy 802.11. After the availability of the 802.11p standard, some research on the performance of VANETs has been developed. Robinson et al. [5] proposed a numerical model to study the throughput of the enhanced distributed coordination function. They found that the contention window size had a pronounced effect on the network performance; however, they did not provide enough details to support this finding. Eichler [6] evaluated the delay and throughput of the IEEE 802.11p via simulation; he reported that the throughput decreases and the delay increases in dense scenarios. Wang et al. [7] proposed to use a threshold to update the backoff size; the proposed scheme takes a considerable time to achieve a better performance; furthermore, the resulting performance is far from optimal. Other studies, such as [8][9], improved the service channel utilization (i.e., throughput) by adjusting the length of SCHI; however, the considered performance consists mainly of the throughput of unicast traffic.

In this paper, we focus on the performance of control channel, where broadcast becomes the main source of application traffic; this results in our study is very different from the existing contributions.

In the open literature, Mansouri et al. [10] studied the performance of the network coverage versus the forwarding probability of broadcast. Tourrilhes [11] improved the reliability of CSMA/CA-based broadcast by the regional collision detection and feedback mechanism. Ma et al. [12] analyzed the saturation throughput of the IEEE 802.11 broadcast, but only the performance of the fixed large window size was discussed and evaluated.

III. ANALYSIS OF CCH COMMUNICATION In this section, we present an analysis, inspired from

[4], that concentrates on the saturation throughput of broadcast traffic; the objective is to develop a scheme that maximizes the throughput of control channel in vehicular networks. Table I shows the list of symbols used in formulating the model and in analyzing the performance.

TABLE I. LIST OF SYMBOLS

Symbol Description n Number of one-hop neighboring stations

0W Initial contention window size

� Transmission probability p Collision probability � Slot period � Propagation delay S Normalized saturation throughput

sT Duration of one successful transmission

cT Duration of one collision

H Duration of MAC & PHY header

A. Transmission Probability for Broadcast According to the IEEE 802.11 standard, whether the

last broadcast transmission is successful or not, the backoff window slot number for pending next packet is independent of the state of the previous broadcast. Thus, the probability coming into one backoff state is just related to the size of the initial contention window 0W . Let ( )s t and ( )b t be the stochastic process representing the backoff stage and the backoff time counter respectively; it is obvious that there always is ( ) 0s t � for broadcast. Thus, in broadcast, the bi-dimensional process { ( ), ( )}s t b t turns into 1-dimensional discrete Markov chain, which is depicted in Fig.2.

Figure 2. Markov Model for broadcast “Backoff Window”

Let� �0, 0lim ( ) 0, ( ) , (0, 1)k t

b Prob s t b t k k W��

� � � � � be the stationary distribution of the Markov chain. For the above Model, there are:

00, 1 0,0 0

0, 0, 1 0,0 0 0, (0, 2)W

k k

b b W

b b b W k W�

����� � � �� (1)

Also, exists 0 1

0,01

Wkk

b�

��� (2)

Considering that any broadcast transmission occurs when the backoff timer equals zero, the transmission probability � can be expressed as follows:

0,00

21

bW

� � �

(3)

We observe that, for broadcast, the transmission probability is independent of the status of collision but

299298298

related to the initial contention window size; this conclusion is consistent with [4].

B. Throughput Model for CCH The medium access control protocol of WAVE uses

the EDCA which is derived from the IEEE 802.11e, but redefined in the IEEE 1609.4. Varied Arbitrary Inter-Frame Space (AIFS) and contention window values are chosen for different access categories (ACs). Table II shows the default parameter settings used in the IEEE 802.11p for different ACs [3].

TABLE II. EDCA CONTENTION WINDOW SIZES

CWmin CWmax AIFSN AC[3] 3 7 2 AC[2] 3 7 3 AC[1] 7 15 6 AC[0] 15 1023 9

AIFS can be calculated as follows:

[ ] iAIFS i SIFS AIFSN �� � (4)

where, SIFS (Short Inter-Frame Space) is the shortest time gap defined in the MAC and physical layer specifications of the IEEE 802.11.

According to [4], the normalized throughput can be expressed as

(1 ) (1 )tr s data

tr tr s s tr s c

P PTS

P P PT P P T��

� � (5)

where trP is the probability that there is at least one transmission in the considered slot time and sP is the probability that there exists a single transmission in this slot.

Let in and i� denote the number of vehicles and the transmission probability of frames with access category i respectively; we can express trP and sP in EDCA as follows:

3 310,0

3

03

03

0

[ (1 ) (1 ) ]

1 (1 )

1 (1 )

ji

i

i

nni i i jj j ii

s nii

ntr i

i

ii

nP

P

n n

� � �

� ��

� � ��� ���� � ������ � � �������� ����

� ��

(6)

Since RTS/CTS is not adopted in broadcast, we have

[ ] [ ]

max[ ] [ ]

bcs databc

c data

T H E T AIFS iT H T AIFS i

� � ��� � � (7)

Here, superscript bc means under the condition of broadcast, and the access category is reflected in [ ]AIFS i .

If we consider the scenario where we have only fixed size packets with a same access category, the saturation throughput for control channel broadcast can be expressed as follows:

1

(1 )

(1 ) (1 ) ( )

bc s databc

tr tr cn

data bc n bcc c

P TSP P T

nTT T

� �� �

��

�� �

� � �

(8)

C. Problems with CCH Broadcast The objective of this subsection is to show the poor

performance of 802.11p-based networks during CCHI. Both the analytical (using MATLAB) and simulation results (using NS-2) adopted the default settings of the IEEE 802.11p parameters (see Table III).

For better clarity, we only consider the results of two categories, i.e., AC[3] and AC[0], which represent the highest and the lowest priority respectively. Fig.3 shows the theoretical saturation throughput versus the transmission probability.

TABLE III. PARAMETERS IN THE IEEE 802.11P

Slot Period ( � ) 13 s� SIFS 32 s� Header Duration 40 s� Symbol Duration 8 s� Propagation Delay ( � ) 1 s�

Packet Payload ( dataT ) 300 Bytes

Figure 3. Saturation throughput versus transmission probability

We observe that the throughput varies considerably with the variation of the transmission probability. As mentioned above, � is only related to the initial contention window size in broadcast, we can easily get

300299299

the results theoretically achieved by the IEEE 802.11p. They are labeled with the discrete-symbols (circle and square markers) in Fig.3.

Fig.3 also shows that the throughput of control channel broadcast is far smaller than the capacity of the channel; for example, in the case of medium density networks ( 10n� ) and an initial window size of 16, the normalized throughput for AC[0] is smaller than 50% of the channel capacity; when the number of vehicles increases to 20, the normalized throughput sharply falls to around 0.2.

Fig. 4 shows the throughput variation with the networks size (obtained using NS-2) for different access categories. This confirms the analytical results (see Fig.3) and the conclusion that the performance for any access category is very poor during CCHI, especially under heavy load conditions (in terms of the networks size).

Figure 4. Simulation throughput of AC[3] and AC[0]

It is worth noting that the performance shown in Fig.4 is slightly better than that shown in Fig.3; this can be explained by the fact that “capture effects” exist in NS-2 simulation, but is not considered in our analytical calculations. Analyses of capture effects have been discussed widely in [13-15]; and it is out of scope of this paper due to space limitations.

IV. SELECTION OF CONTENTION WINDOW SIZE We believe that the reason behind the poor

performance, during CCHI, is related to the contention window size. In the case of a fixed window size 0W , whatever the random seed, the probability that a same backoff count is assigned to different stations increases with the number of vehicles; this phenomenon will certainly cause more collisions.

Fig. 5 shows that using a bigger value of 0W (and thus a smaller value of � ) results in bigger throughput because using a large contention window size reduces

the channel access scheme’s sensitivity to the variation of the number of vehicles. The analytical (solid-lines) and the simulation results (discrete-symbols) shown in Fig. 5 confirm our observations effectively.

Figure 5. Saturation throughput using larger CWs

For example, the saturation throughput is nearly always more than 0.5 when we use 256 as the contention window size; when this size is 1024, the throughput does not change much while the number of vehicles varies from 40 to 100.

However, VANETs is a dynamic wireless networks; using a fixed (even big) value for the contention window size is not suitable and the performance during CCHI cannot be optimal. Indeed, in this case it is very likely that vehicles will unnecessarily wait longer backoff periods to broadcast packets; this will definitely causes mediocre bandwidth utilization especially in the case of sparse networks, such as suburb highways or country roads. Another shortcoming of using a big fixed window size for broadcast is that it results in a considerable system delay [12].

In order to get an optimal throughput performance, the contention window size should be adjusted according to the size of vehicular networks. Combining (3) and (8), we can express the saturation throughput, during CCHI, as a function of the contention window size:

10

0 0

2 ( 1)( 1) ( 1) ( )

nbc

data n bc n bcc c

n WS T

W T W T �

��� �

� � � (9)

To achieve the maximum throughput, we can compute the derivative of (9) with respect to 0W , then make it equal 0. After some simplifications, we obtain the following expression:

02

1cTW n

� �� �� �� �� �� �

(10)

where, � ��� � means the computation of the product ceiling.

301300300

Equation (10) shows that, for broadcast, the optimal selection of contention window depends on the number of vehicles and the probability distribution function of the packet size. If we consider that the packet size is fixed, then the optimal window size, 0W , will be a function of only the networks size n . The number of one-hop neighbors can be easily obtained via some schemes mentioned in [16][17].

Considering the characters of the communication in VANETs (see Fig.1), we proposed an optimized scheme to maximize the saturation throughput of control channel. Fig.6 shows the flow diagram of the proposal.

Figure 6. Flow diagram of the optimization

Figure 7. Comparison of the optimization with the original

Fig 7 shows the analytical results of the throughput variation with the networks size. We observe that our proposal leads to some considerable improvement of the

throughput during CCHI compared to using a fixed window size (as suggested by 802.11p). We also observe that the improvement is not impacted much by the increase of the number of vehicles.

V. PERFORMANCE EVALUATION The 802.11p parameter settings, which have been

shown in Table III, are used in both theoretical analysis and simulations. Here, the efficiencies of two access categories (i.e., AC[3] and AC[0]) are evaluated respectively.

Fig.8 shows that the simulation results (discrete-symbols) are consistent with analytical results (lines) quite well; it also illustrates that, even for AC[0], the proposed window adaptation scheme increases the throughput up to more than 0.63. Compared to the results shown in Fig.4, the improvement of throughput is considerable. Furthermore, we observe that the increase is maintained even if the number of vehicles increases. This means the optimization can make the media access scheme be more suitable for the scalability of vehicular networks.

Figure 8. Evaluation of the optimization

VI. CONCLUSION In this paper, we studied the throughput performance

of vehicular networks during CCHI. We did show, analytically and via simulation, that the saturation throughput of control channel broadcast in 802.11p is considerably low. To improve the performance of control channel, we proposed a scheme that adapts the contention window to the networks size; indeed, the optimization provides the computation of the contention window size that achieves the maximum throughput at any given time. The evaluation results, via NS-2, confirmed the findings of the proposal. In future work, we will extend the proposal to heterogeneous VANETs scenarios, in which applications with different access categories exist simultaneously.

Optimization:

Step 1: All stations start with default parameters during the initialization;

Step 2: When the first CCHI is coming, the RSU (Road-Side Unit) senses all active receptions (successful transmissions and collisions) and estimates the networks size;

Step 3: During the SCHI in a row, the RSU adds the estimation into the reserved field of a high-prioritized WSM which will be broadcasted in the next CCHI;

Step 4: When switching back to the immediately following CCHI, all OBUs (On-Board Units) receive the modified WSM sent from the RSU; before the CCHI is finished, an update timer (e.g. 1s) begins;

Step 5: When the timer is up, redo from Step 2.

302301301

REFERENCES [1] IEEE P802.11p/D10.0. “Wireless LAN Medium Access Control

(MAC) and Physical Layer (PHY) Specifications – Amendment 6: Wireless Access in Vehicular Environments”, 2010.

[2] W. Fisher, “Development of DSRC/WAVE Standards”, IEEE 802.11–07/2045r0, June 2007.

[3] IEEE Std 1609.4–2006. Trial-Use Standard for Wireless Access in Vehicular Environments(WAVE) -- Multi-channel Operation,2006.

[4] G. Bianchi, “Performance analysis of the IEEE 802.11 distributed coordination function,” IEEE Trans. Commun., vol.18, no.3, pp.535-547, Mar. 2000.

[5] J. Robinson, and T. Randhawa, “Saturation throughput analysis of IEEE 802.11e enhanced distributed coordination function,” IEEE JSAC, vol. 22, no. 5, pp. 917-928, June 2004.

[6] S. Eichler, “Performance evaluation of the IEEE 802.11p WAVE communication standard,” in IEEE VTC’07, Baltimore, MD, USA, Oct. 2007, pp. 2199-2203.

[7] Y. Wang; A. Ahmed, B. Krishnamachari, K. Psounis, “IEEE 802.11p performance evaluation and protocol enhancement,” in IEEE ICVES’08, Columbus, USA, Sep. 2008, pp. 317-322.

[8] S. Y. Wang, et al., “Improving the channel utilization of IEEE 802.11p/1609 networks,” in IEEE WCNC’09, Budapest, Hungary, Apr, 2009, pp. 1209-1214.

[9] Q. Wang, S. Leng, H. Fu, Y. Zhang, H. Weerasinghe, “An enhanced multi-channel MAC for the IEEE 1609.4 based vehicular ad hoc networks”, in IEEE INFOCOM’10, San Diego, CA, USA, Mar. 2010.

[10] H. Shah-Mansouri, B. Khalaj, M. Pakravan, A. Khodaian, “Counter-based broadcasting: modeling and performance analysis in CSMA-based wireless networks”, in IEEE PIMRC’09, Tokyo, Japan, Sep. 2009.

[11] J. Tourrilhes, “Robust broadcast: improving the reliability of broadcast transmissions on CSMA/CA”, in IEEE PIMRC’98, Boston, USA, Sep.1998.

[12] X. Ma, and X. Chen, “Performance Analysis of IEEE 802.11 Broadcast Scheme in Ad Hoc Wireless LANs”, IEEE Trans. Vehicular Technology, vol.57, no.6, Nov. 2008, pp.3757-3768.

[13] G. J. Sutton, R. P. Liu, X. Yang, I. B. Collings, “Modelling Capture Effect for 802.11 DCF under Rayleigh Fading”, in IEEE ICC’10, Cape Town, South Africa, May 2010, pp.1-6.

[14] Z. Hadzi-Velkov and B. Spasenovski, “Capture effect in IEEE 802.11 basic service area under influence of Rayleigh fading and near/far effect,” in IEEE PIMRC’02, vol. 1, pp. 172-176, Sept. 2002.

[15] F. Daneshgaran, M. Laddomada, F. Mesiti, M. Mondin, and M. Zanolo, “Saturation throughput analysis of IEEE 802.11 in the presence of non ideal transmission channel and capture effects,” IEEE Trans. on Comms., vol.56, no.7, Jul. 2008.

[16] C. Budianu, S. Ben-David, and L. Tong, “Estimation of the number of operating sensors in large-scale sensor networks with mobile access,” IEEE Trans. Signal Processing, vol. 54, May 2006.

[17] M. Howlader, M. Frater, and M. Ryan, “Estimating the number and distribution of the neighbors in an underwater communication network”, in SENSORCOMM’08, Washington, DC, USA, 2008, pp. 693-698.

303302302