Taherkhani 2016
Taherkhani 2016
Taherkhani 2016
Abstract—In an urban environment, intersections are critical surrounding environment, by exchanging some information.
locations in terms of road crashes and number of killed or injured Therefore, VANets can play a vital role to ensure safer urban
people. Vehicular ad hoc networks (VANETs) can help reduce environments for road users [3], [4].
the traffic collisions at intersections by sending warning messages
to the vehicles. However, the performance of VANETs should VANet is employed by Intelligent Transportation Sys-
be enhanced to guarantee delivery of the messages, particularly tems (ITS) for vehicle-to-vehicle (V2V) and vehicle-to-
safety messages to the destination. Data congestion control is an infrastructure (V2I) communications. Communications in
efficient way to decrease packet loss and delay and increase the VANets rely on standards and protocols defined in Dedicated
reliability of VANETs. In this paper, a centralized and localized Short Range Communication (DSRC) and Wireless Access in a
data congestion control strategy is proposed to control data con-
gestion using roadside units (RSUs) at intersections. The proposed Vehicular Environment (WAVE). IEEE 802.11p and IEEE1609
strategy consists of three units for detecting congestion, clustering are two WAVE standards. These standards are used to manage
messages, and controlling data congestion. In this strategy, the resources, network services, multi-channels operations, secu-
channel usage level is measured to detect data congestion in the rity services and so on. VANets employs Road-Side Units
channels. The messages are gathered, filtered, and then clustered (RSUs) and On-Board Units (OBUs) to conduct V2I and V2V
by machine learning algorithms. K-means algorithm clusters the
messages based on message size, validity of messages, and type communications. RSUs are fixed at the roadsides, while OBUs
of messages. The data congestion control unit determines appro- are fixed on vehicles [3]–[6].
priate values of transmission range and rate, contention window The applications developed for VANets can be classified
size, and arbitration interframe spacing for each cluster. Finally, into three main categories: 1) safety applications (e.g., road
RSUs at the intersections send the determined communication hazard control notification and emergency electronic break
parameters to the vehicles stopped before the red traffic lights to
reduce communication collisions. Simulation results show that the light), 2) convenience applications (e.g., parking availability
proposed strategy significantly improves the delay, throughput, notification and congested road notification,), and 3) commer-
and packet loss ratio in comparison with other congestion control cial applications (e.g., service announcements and content map
strategies using the proposed congestion control strategy. database download) [5], [7]. These applications generate two
Index Terms—Congestion control, machine learning algo- types of messages for communications in VANets, that includes
rithms, K-means algorithm, quality of service, vehicular ad hoc safety and non-safety messages. The safety messages including
networks. beacon and emergency messages are transferred in the con-
trol channel. The non-safety messages including the messages
I. I NTRODUCTION generated by convenience and commercial applications are
transferred in service channels [6], [8], [9].
Data congestion is one of the problematic issues in VANets. The open-loop solutions prevent congestion before it happens
From now on we will refer to data congestion just as con- in the. In the closed-loop solutions, however, the congestion is
gestion. Congestion occurs in the networks when the channels controlled after being detected [28]. Detection of congestion
are overloaded in high dense network conditions, ultimately can be carried out by employing measurement methods that
resulting in increased packet loss and delay, and reduced the sense the number of messages in queue, the channel usage level
performance of network. Therefore, congestion control needs and the channel occupancy time [29].
to be conducted to support QoS, as well as to ensure the safety As mentioned in introduction, the congestion control strate-
and reliability in vehicular environments [13]–[16]. gies in VANets are classified into rate-based, power-based,
Basically, congestion control strategies for VANets can be CSMA/CA-based, prioritizing and scheduling-based, and
classified into five categories including rate-based, power- hybrid strategies [17]. In the following, these strategies are
based, CSMA/CA-based, prioritizing and scheduling-based, discussed.
and hybrid strategies [17]. In rate-based strategies, transmission
rate is decreased when the channels are congested [18], [19].
A. Rate-Based Strategies
In power-based strategies, transmission power or range is dy-
namically adjusted to decrease the channels loads [20], [21]. The rate-based strategies are type of closed-loop solutions
In CSMA/CA-based strategies, congestion is controlled by that control congestion after being detected in the networks.
modifying the CSMA/CA protocol and adjusting the contention These strategies dynamically reduce the transmission rate or
window size and/or Arbitration inter-frame spacing (AIFS) to packet generation rate to reduce the packet collision rate in
decrease the channel access [22], [23]. The prioritizing and the congested channels. Ye et al. [18] measured optimal packet
scheduling-based strategies define a priority for each message transmission rate based on the vehicle density in order to
and schedule them in control and service channel queues such increase the broadcast efficiency and reliability. They modified
that the emergency and safety massages get higher priorities to the WAVE standard for adding a congestion control layer that
transfer with less delay [24], [25]. Finally, in hybrid strategies, communicates with MAC layer. Then, they investigated the
all or some of the proposed solutions in previous categories are packet reception rate of beacon messages by considering the
combined to control congestion [12], [26], [27]. impact of fading on one-dimension broadcasting. However,
In this work, a Machine Learning Congestion Control one-dimension broadcasting is not usual in real applications of
(ML-CC) strategy is introduced. The proposed strategy is a VANets.
hybrid centralized and localized strategy that employs Road He et al. [19] proposed a cross-layer strategy to control
Side Unites (RSUs) to control congestion. Indeed, this strat- congestion in control channel, and guarantee delivery of event-
egy centrally performs in the RSU set at each intersection driven safety messages. In this strategy, first, the occupancy
instead of in the vehicles to locally control the congestion time of control channel in MAC layer is measured. The channel
of that intersection. In this strategy, the messages are clus- is considered to be congested if the occupancy time exceeds
tered using machine learning algorithms in each RSU inde- a predefined threshold. Then, MAC layer sends a signal to
pendently. The transmission range and rate, contention window application layer for blocking all beacon messages. By blocking
size, and AIFS parameters are the effective communication beacon messages, the control channel is reserved only for
parameters for congestion occurrence. Thus, the congestion emergency messages, and consequently load of control channel
can be controlled by adjusting the communication parameters is reduced. In this strategy, however, measuring of the channel
for different classes of messages rather than all the messages, occupancy time in MAC layer is difficult.
which, as a result, increases the efficiency of the process. The
communications parameters for each class of the messages
B. Power-Based Strategies
are determined based on the minimum delay for transferring
the messages of each cluster. Then, the determined commu- The power-based strategies control congestion by tuning
nication parameters are sent to the vehicles located in the transmission power (range). The transmission power is one of
congestion area at each intersection. Controlling congestion in the most important factors in occurrence of channel collision.
these areas helps to reduce the number of packets lost and When many nodes in the same communication rage compete
delay and consequently increases the safety and reliability in to acquire the channel, channel collision and consequently
VANets. congestion occurs. The power-based strategies are open-loop
The rest of this paper is structured as follows. Section II strategies that avoid congestion by tuning transmission power
reviews the background of congestion control and the best and reducing channel loads.
congestion control strategies in VANets. Section III proposes Torrent-Moreno et al. [20] proposed a Distributed-Fair
a strategy to control congestion centrally and locally. Finally, Power Adjustment for Vehicular environment (D-FPAV)
Section IV evaluates the performance of the proposed strategy strategy. This strategy dynamically adjusts the beaconing trans-
based on QoS parameters. mission range based on the vehicle density to reduce the
channel loads. However, by shrinking the transmission range
of beacon messages, the probability of delivering the beacon
II. BACKGROUND AND R ELATED W ORKS
messages in far distances is reduced. Therefore, the VANets’
Generally, congestion control strategies are classified into applications using beacon information, face some difficulties to
two groups of solutions: open-loop and closed-loop solutions. obtain essential information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
TAHERKHANI AND PIERRE: CENTRALIZED AND LOCALIZED DATA CONGESTION CONTROL STRATEGY FOR VANets 3
Sahu et al. [21] proposed Network Coding Congestion Bouassida et al. [25] proposed a new strategy that controls
Control (NC-CC) strategy that uses network coding to control the congestion by defining priority for messages based on
beacon overhead. The proposed strategy tunes the transmission static and dynamic factors. In this strategy, the messages are
range of beacon messages by network coding at the packet scheduled in control and service channels. The static and dy-
level. In this strategy, the number of forwarded beacon mes- namic factors are defined based on the content of messages and
sages is reduced by forwarding the coded beacon messages condition of network, respectively. This strategy can improve
only 2-hops over a predefined forwarding zone. Therefore, this the delivery delay of safety messages.
strategy decreases channels overhead by reducing the transmis-
sion range of beacon messages. The proposed strategy is also
E. Hybrid Strategies
scalable due to its ability to forward the beacons messages to a
large number of receivers. The hybrid strategies combine all or some of the solutions
employed in the previous strategies for solving congestion
in VANets. Djahel et al. [26] proposed a three-phase hybrid
C. CSMA/CA-Based Strategies
strategy. In the first phase, the messages are prioritized based
CSMA/CA protocol is considered as the default congestion on the messages content and number of hops between senders
control protocol in VANets. This protocol determines the chan- and receivers to avoid congestion. In second phase, the average
nel access ability for each node in MAC layer. CSMA/CA- waiting time, beacon reception rate and collision rate metrics
based strategies adjust the channel access ability by modifying are measured to detect congestion. If the values of these metrics
the channel access parameters such as contention widow size exceed predefined thresholds, the congestion is considered to
and AIFS, and consequently control the congestion in the occur in the VANet. After detecting congestion, in third phase,
channels [30]. the transmission rate and transmission range of beacon mes-
Hsu et al. [22] proposed an Adaptable Offset Slot (AOS) sages are adjusted to make an efficient usage of the channel
strategy for reducing the channel load and delay. AOS uses the bandwidth. Although the delay of the proposed strategy is
number of neighbor vehicles to obtain the minimum contention significant, the reliability and safety of VANets are guaranteed
window size. AOS strategy linearly increases the contention using this strategy.
window size by increasing the number of vehicles. In this Taherkhani and Pierre [12], [27] introduced two hybrid con-
strategy, however, the delay of emergency messages increases gestion control strategies called Uni-Objective Tabu Search
when the contention window size increases in high vehicle (UOTabu) and Multi-Objective Tabu search (MOTabu). These
density condition. strategies are closed-loop strategies that detect the congestion
Jang et al. [23] provided a detection-based MAC strategy. by measuring the channel usage level. If the channel usage
This strategy detects the congestion by exchanging RTS/CTS level exceeds 70%, the congestion is considered to occur. Then,
messages to predict the number of message collisions. Then, by tuning transmission rate and transmission range, the con-
the contention window size is dynamically adapted according gestion is controlled. The optimal values for these parameters
to predicted network status. In other words, to reduce the are obtained by a Tabu search algorithm. UOTabu determines
channels overloads, the contention window size is increased transmission rate and range by considering the minimum delay
by increasing the number of collisions. Although using this [27], while MOTabu considers minimum delay and jitter to
strategy, throughput is improved and the number of collision control congestion [12]. The results showed that UOTabu and
is reduced, it is not a real-time strategy for broadcasting the MOTabu reduced the delay and the packet loss, and conse-
messages. quently improved the performance of VANets.
Despite the advantages of the introduced congestion control
strategies, some drawbacks can be observed. Some of the
D. Prioritizing and Scheduling-Based Strategies
strategies need extra interactions between the vehicles to detect
The congestion control strategies in this category assign the congestion in the network. These extra interactions increase
priority to the messages such that more chance are given to the channel loads and the possibility of collision [17], [31],
the more important messages (e.g., emergency messages) for [32]. In some strategies, by reducing beaconing rate to control
being transferred over the channels without delay. Bai et al. [24] load of channels, the applications using the beacon information
introduced Context Awareness Beacon Scheduling (CABS) face to lack of information to operate efficiently [33]–[35].
strategy to schedule the beacon messages dynamically. CABS Tuning the transmission power and rate to control congestion
strategy solves congestion resulted from high rate of bea- are affected by various parameters such as vehicle density,
coning in high dense vehicular networks. This strategy is a distance between sender and receiver, message size and so on.
distributed strategy that piggybacks the information in beacon However, in a large scale network, tuning transmission rate and
messages (e.g., velocity, direction and position of vehicles). transmission power are faced to many challenges due to the
Then, a unique time slot is assigned to each vehicle based on large number of influential parameters [12], [31], [36].
TDMA-like transmission. CABS strategy improves the packet The CSMA/CA protocol employs the exponential back-off
reception rate and channel access delay. However, the internet- mechanism. However, this mechanism is not efficient for broad-
working in MAC layer that needs to be considered to allocate casting the beacon messages [37]. This mechanism cannot work
proper time slots for different transmissions is not taken into properly in high rate message situations, especially when the
account. messages have a time-out that lead to the packets to be dropped
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
TAHERKHANI AND PIERRE: CENTRALIZED AND LOCALIZED DATA CONGESTION CONTROL STRATEGY FOR VANets 5
of all members of each cluster. Finally, all the data are clustered
based on the new centroid. K-means repeats this process until
the members of each cluster do not move to the other clusters
anymore [46]–[49].
Basically, K-means consists of three main steps: 1) selecting
initial centroids for k clusters; 2) computing squared
Euclidean distance of each data to the centroids; 3) computing
the new centroids cluster to find closest centroids. Steps 2 and 3
should be repeated until the cluster members no longer change
[46]–[48].
The initial centroids for K-means can be chosen by Forgy
and Random methods [50]. However, there is no guarantee that,
using these methods, K-means converges [50], [51]. Therefore,
the researchers use various methods for determining the initial
centroids. In this work, the initial centroids for k clusters are
assumed to be the first k messages received by RSUs.
K-means has three inputs including features, number of
clusters and number of iterations. Clustering algorithms classify
a set of objects based on identified features; thus, features
have a significant impact on performance of the clustering
algorithms. In K-means, the features should be transformed to
the dimensional values. Generally, there is no efficient strategy
for determining the features. In fact the features should be de-
termined specifically for each problem based on the knowledge
about the domain of problem [46], [47], [49]. In this work,
the features of K-means are defined based on the features of
messages including the message size, validity of messages,
distances between vehicles and RSUs, type of message and
direction of message sender.
The number of clusters is the second input for K-means
clustering algorithm. The best number of cluster for each prob-
lem can be defined by executing the clustering algorithm for
different numbers of clusters [46], [47], [49]. In this work, the
number of clusters (k) for K-means is obtained by conducting Fig. 2. Pseudocode of the proposed K-means algorithm.
a set of preliminary simulations shown in Section IV-B.
The clustering algorithm is terminated when a predefined increase the number of vehicles that can receive these types of
convergence is obtained. That means there are not any changes messages. However, the number of collisions increases when
in the clusters’ members. However, if the convergence is not the transmission range of messages is high. The transmission
obtained in the acceptable time, the clustering algorithm should rate also impacts the saturation of the channels. High transmis-
be terminated after a predefined number of iterations. Theoreti- sion rate improves the performance of VANets’ applications
cally, K-means does not rapidly converge especially for the big due to the more frequently sending the information to these
data sets [46]. In this work, the number of iterations is assumed applications. However, high transmission rate may saturate
to be 100. the channels increasing the load of channels [53]. Contention
The complexity of K-means depends on the number of data window size and AIFS also impact the condition of channels.
in each data set, the number of features, number of clusters and To define the priority of the messages for transferring in the
number of iterations. Therefore, proper initial conditions can channels, the contention window size and AIFS need to be
result in a better clustering [52]. Fig. 2 shows the pseudocode determined for each type of messages [54]. Prioritizing and
of the proposed K-means used for clustering in this work. scheduling the messages help prevent the channels saturation
and congestion occurrence in the networks [55].
For adjusting the communication parameters for each cluster
C. Congestion Control Unit
of messages, the proposed strategy selects the proper values of
Congestion control unit adjusts the communication param- these parameters among the range of values defined by DSRC
eters for each cluster determined in data control unit. The standard [6], [8]. DSRC defines the transmission rate and range
communication parameters considered in this unit are trans- between 3–27 Mbps and 10–1000 m, respectively. Based on this
mission rate, transmission range, contention window size and standard, the possible values for transmission rate are 3, 4.5 6,
AIFS. The performance of VANets is considerably affected by 9, 12, 18, 24, and 27 Mbps [56]–[58], while the possible values
transmission range and rate. The messages, especially safety for transmission range are 10, 50, 100, 126, 150, 210, 300,
messages, are usually sent with high transmission range to 350, 380, 450, 550, 650, 750, 850, 930, 971, and 1000 m [59].
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
TABLE I
S IMULATION PARAMETERS U SED IN THE U RBAN S CENARIO
TAHERKHANI AND PIERRE: CENTRALIZED AND LOCALIZED DATA CONGESTION CONTROL STRATEGY FOR VANets 7
Fig. 7. Impact of the number of vehicles on average throughput. Fig. 8. Impact of the number of vehicles on the (a) number of packets lost and
(b) packet loss ratio.
TAHERKHANI AND PIERRE: CENTRALIZED AND LOCALIZED DATA CONGESTION CONTROL STRATEGY FOR VANets 9
types of messages, and adjusting communication parameters [19] J. He, H.-H. Chen, T. M. Chen, and W. Cheng, “Adaptive congestion
for each cluster. Therefore, using ML-CC strategy, congestion control for DSRC vehicle networks,” IEEE Commun. Lett., vol. 14, no. 2,
pp. 127–129, Feb. 2010.
can be controlled properly, and subsequently a safer environ- [20] M. Torrent-Moreno, P. Santi, and H. Hartenstein, “Distributed fair trans-
ment can be provided for road users especially at the urban mit power adjustment for vehicular ad hoc networks,” in Proc. 3rd Annu.
intersections. IEEE SECON, 2006, pp. 479–488.
[21] P. K. Sahu, A. Hafid, and S. Cherkaoui, “Congestion control in ve-
To practically implement the proposed strategy in real ve- hicular networks using network coding,” in Proc. IEEE ICC, 2014,
hicular networks, an RSU needs to be set at each intersection. pp. 2736–2741.
Also, since congestion control is a real-time process, RSUs may [22] C.-W. Hsu, C.-H. Hsu, and H.-R. Tseng, “MAC channel congestion con-
trol mechanism in IEEE 802.11 p/WAVE vehicle networks,” in Proc.
need to be equipped with Graphic Processing Units (GPUs) IEEE VTC Fall, 2011, pp. 1–5.
for quickly executing machine learning algorithms. Note that [23] H.-C. Jang and W.-C. Feng, “Network status detection-based dynamic
the machine learning algorithms conduct a large number of adaptation of contention window in IEEE 802.11 p,” in Proc. IEEE 71st
VTC Spring, 2010, pp. 1–5.
calculations and operations that takes a lot of time. [24] S. Bai, J. Oh, and J.-I. Jung, “Context awareness beacon scheduling
scheme for congestion control in vehicle to vehicle safety communica-
tion,” Ad Hoc Netw., vol. 11, no. 7, pp. 2049–2058, Sep. 2013.
[25] M. S. Bouassida and M. Shawky, “On the congestion control within
VANET,” in Proc. 1st IFIP WD, 2008, pp. 1–5.
R EFERENCES [26] S. Djahel and Y. Ghamri-Doudane, “A robust congestion control scheme
[1] P. Vetrivelan, P. Narayanasamy, and J. J. Charlas, “A multi-constraint for fast and reliable dissemination of safety messages in VANETs,” in
real-time vehicular (MCRV) mobility framework for 4G heterogeneous Proc. IEEE WCNC, 2012, pp. 2264–2269.
vehicular ad-hoc networks,” in Proc. Int. Multi Conf. Eng. Comput. Sci., [27] N. Taherkhani and S. Pierre, “Congestion control in vehicular ad hoc
2012, vol. 1, pp. 423–428. networks using meta-heuristic techniques,” in Proc. 2nd ACM Int. Symp.
[2] N. Nidhi and D. Lobiyal, “Performance evaluation of realistic vanet Des. Anal. Intell. Veh. Netw. Appl., 2012, pp. 47–54.
using traffic light scenario,” Int. J. Wireless Mobile Netw., vol. 4, no. 1, [28] A. S. Tanenbaum, Computer Networks, 5th ed. Englewood Cliffs, NJ,
pp. 237–249, 2012. USA: Prentice-Hall, 2010.
[3] K. Golestan et al., “Vehicular ad-hoc networks (VANETs): Capabilities, [29] M. Y. Darus and K. Abu Bakar, “A review of congestion control algorithm
challenges in information gathering and data fusion,” in Autonomous for event-driven safety messages in vehicular networks,” Int. J. Comput.
and Intelligent Systems. Berlin, Germany: Springer-Verlag, 2012, Sci. Issues, vol. 8, no. 5, pp. 49–53, Sep. 2011.
pp. 34–41. [30] IEEE Draft Standard for Wireless LAN Medium Access Control (MAC)
[4] G. Karagiannis et al., “Vehicular networking: A survey and tutorial on and Physical Layer (PHY) Specifications, IEEE Std. 802.11p/D8.0,
requirements, architectures, challenges, standards and solutions,” IEEE 2009.
Commun. Surveys Tuts., vol. 13, no. 4, pp. 584–616, 4th Quart. 2011. [31] C.-L. Huang, Y. P. Fallah, R. Sengupta, and H. Krishnan, “Information dis-
[5] J. Guerrero-Ibáñez, C. Flores-Cortés, and S. Zeadally, “Vehicular ad- semination control for cooperative active safety applications in vehicular
hoc networks (VANETs): Architecture, protocols and applications,” ad-hoc networks,” in Proc. IEEE GLOBECOM, 2009, pp. 1–6.
in Next-Generation Wireless Technologies. London, U.K.: Springer- [32] M. Barradi, A. S. Hafid, and J. R. Gallardo, “Establishing strict
Verlag, 2013, pp. 49–70. priorities in IEEE 802.11 p WAVE vehicular networks,” in Proc. IEEE
[6] S. Zeadally, R. Hunt, Y.-S. Chen, A. Irwin, and A. Hassan, “Vehicular ad GLOBECOM, 2010, pp. 1–6.
hoc networks (VANETS): Status, results, and challenges,” Telecommun. [33] J. Camp and E. Knightly, “Modulation rate adaptation in urban and
Syst., vol. 50, no. 4, pp. 217–241, Aug. 2012. vehicular environments: Cross-layer implementation and experimental
[7] M. Jabbarpour Sattari, R. M. Noor, and S. Ghahremani, “Dynamic con- evaluation,” IEEE/ACM Trans. Netw., vol. 18, no. 6, pp. 1949–1962,
gestion control algorithm for vehicular ad hoc networks,” Int. J. Softw. Dec. 2010.
Eng. Appl., vol. 7, no. 3, pp. 95–108, May 2013. [34] T. Tielert, D. Jiang, Q. Chen, L. Delgrossi, and H. Hartenstein,
[8] Y. Liu, F. Dion, and S. Biswas, “Dedicated short-range wireless communi- “Design methodology and evaluation of rate adaptation based congestion
cations for intelligent transportation system applications: State of the art,” control for vehicle safety communications,” in Proc. IEEE VNC, 2011,
Transp. Res. Rec., J. Transp. Res. Board, no. 1910, pp. 29–37, 2005. pp. 116–123.
[9] Y. Qian and N. Moayeri, “Design of secure and application- [35] M. Torrent-Moreno, “Inter-vehicle communications: Assessing informa-
oriented VANETs,” in Proc. IEEE Veh. Technol. Conf. Spring, 2008, tion dissemination under safety constraints,” in Proc. 4th Annu. Conf.
pp. 2794–2799. WONS, 2007, pp. 59–64.
[10] M. Johnson and E. Howard, “Road safety vision 2010,” Canadian Council [36] R. Baldessari, D. Scanferla, L. Le, W. Zhang, and A. Festag, “Joining
Motor Transp. Admin., Ottawa, ON, Canada, 2007. forces for VANETs: A combined transmit power and rate control algo-
[11] “Pedestrian Safety Plan,” 2010. [Online]. Available: https://www. rithm,” in Proc. 7th Int. WIT, 2010, pp. 1–5.
brampton.ca/EN/residents/Roads/pedestrian-driver-safety/Documents/ [37] R. Stanica, E. Chaput, and A.-L. Beylot, “Congestion control in CSMA-
Pedestrian%20Safety%20Plan%20-%20Final.pdf. based vehicular networks: Do not forget the carrier sensing,” in Proc. 9th
[12] N. Taherkhani and S. Pierre, “Improving dynamic and distributed con- Annu. IEEE SECON, 2012, pp. 650–658.
gestion control in vehicular ad hoc networks,” Ad Hoc Netw., vol. 33, [38] J. Guo and J. Zhang, “Safety message transmission in vehicular commu-
pp. 112–125, Oct. 2015. nication networks,” presented at the 17th ITS World Congress, Busan,
[13] T. Ghosh and S. Mitra, “Congestion control by dynamic sharing of band- Korea, 2010.
width among vehicles in VANET,” in Proc. 12th Int. Conf. ISDA, 2012, [39] W. Guan, J. He, L. Bai, and Z. Tang, “Adaptive congestion control of
pp. 291–296. DSRC vehicle networks for collaborative road safety applications,” in
[14] Y. P. Fallah, C. Huang, R. Sengupta, and H. Krishnan, “Congestion control Proc. IEEE 36th Conf. LCN, 2011, pp. 913–917.
based on channel occupancy in vehicular broadcast networks,” IEEE 72nd [40] Y. Zang et al., “Congestion control in wireless networks for vehicular
Veh. Technol. Conf. Fall (VTC 2010-Fall), Ottawa, ON, pp. 1–5, 2010. safety applications,” in Proc. 8th Eur. Wireless Conf., 2007, p. 7.
[15] M. Sepulcre, J. Mittag, P. Santi, H. Hartenstein, and J. Gozalvez, “Con- [41] C. M. Bishop, Pattern Recognition and Machine Learning. New York,
gestion and awareness control in cooperative vehicular systems,” Proc. NY, USA: Springer-Verlag, 2006.
IEEE, vol. 99, no. 7, pp. 1260–1279, Jul. 2011. [42] D. E. Goldberg and J. H. Holland, “Genetic algorithms and machine
[16] R. K. Singh and N. Tyagi, “Challenges of routing in vehicular ad hoc learning,” Mach. Learn., vol. 3, no. 2, pp. 95–99, Oct. 1988.
networks: A survey,” Int. J. Electron. Commun. Comput. Eng., vol. 3, [43] S. B. Kotsiantis, I. Zaharakis, and P. Pintelas, “Supervised machine learn-
no. 1, pp. 126–132, Jan. 2012. ing: A review of classification techniques,” Informatica, vol. 31, no. 3,
[17] X. Shen, X. Cheng, R. Zhang, B. Jiao, and Y. Yang, “Distributed conges- pp. 249–268, 2007.
tion control approaches for the IEEE 802.11 p vehicular networks,” IEEE [44] J. Dougherty, R. Kohavi, and M. Sahami, “Supervised and unsupervised
Intell. Transp. Syst. Mag., vol. 5, no. 4, pp. 50–61, Winter 2013. discretization of continuous features,” in Proc. 12th Int. Conf. Mach.
[18] F. Ye, R. Yim, S. Roy, and J. Zhang, “Efficiency and reliability of one-hop Learn., 1995, pp. 194–202.
broadcasting in vehicular ad hoc networks,” IEEE J. Sel. Areas Commun., [45] Y. Kodratoff, Introduction to Machine Learning. San Mateo, CA, USA:
vol. 29, no. 1, pp. 151–160, Jan. 2011. Morgan Kaufmann, 2014.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
TAHERKHANI AND PIERRE: CENTRALIZED AND LOCALIZED DATA CONGESTION CONTROL STRATEGY FOR VANets 11
[46] A. K. Jain, “Data clustering: 50 years beyond K-means,” Pattern Recognit. [63] T. Issariyakul and E. Hossain, An Introduction to Network Simulator NS2.
Lett., vol. 31, no. 8, pp. 651–666, Jun. 2010. New York, NY, USA: Springer-Verlag, 2012.
[47] J. A. Hartigan and M. A. Wong, “Algorithm AS 136: A k-means clustering [64] F. K. Karnadi, Z. H. Mo, and K.-C. Lan, “Rapid generation of re-
algorithm,” Appl. Stat., vol. 28, no. 1, pp. 100–108, Jan. 1979. alistic mobility models for VANET,” in Proc. IEEE WCNC, 2007,
[48] D. Kaur and K. Jyoti, “Enhancement in the performance of K-means pp. 2506–2511.
algorithm,” Int. J. Comput. Sci. Commun. Eng., vol. 2, no. 1, pp. 29–32, [65] K. Z. Ghafoor, J. Lloret, K. A. Bakar, A. S. Sadiq, and S. A. B. Mussa,
Feb. 2013. “Beaconing approaches in vehicular ad hoc networks: A survey,” Wireless
[49] J. MacQueen, “Some methods for classification and analysis of multivari- Pers. Commun., vol. 73, no. 3, pp. 885–912, Dec. 2013.
ate observations,” in Proc. 5th Berkeley Symp. Math. Stat. Probability, [66] C. Campolo, Y. Koucheryavy, A. Molinaro, and A. Vinel, “Characterizing
1967, pp. 281–297. broadcast packet losses in IEEE 802.11 p/wave vehicular networks,” in
[50] G. Hamerly and C. Elkan, “Alternatives to the k-means algorithm that find Proc. IEEE 22nd Int. Symp. PIMRC, 2011, pp. 735–739.
better clusterings,” in Proc. 11th Int. Conf. Inf. Knowl. Manage., 2002, [67] J. Li and C. Chingan, “Delay-aware transmission range control for
pp. 600–607. VANETs,” in Proc. IEEE GLOBECOM, 2010, pp. 1–6.
[51] A. Zollmann and S. Vogel, “A word-class approach to labeling PSCFG
rules for machine translation,” in Proc. 49th Annu. Meet. Assoc. Comput.
Linguistics, Hum. Lang. Technol., 2011, vol. 1, pp. 1–11.
[52] B. W. Ng, “Wavelet based image texture segmentation using a modi-
fied K-means algorithm,” Ph.D. dissertation, Dept. Elect. Electron. Eng., Nasrin Taherkhani received the B.Sc. and M.Sc.
Univ. Adelaide, Adelaide, SA, Australia, 2003. degrees in software engineering from Qazvin
[53] M. Torrent-Moreno, J. Mittag, P. Santi, and H. Hartenstein, “Vehicle- Islamic Azad University, Qazvin, Iran, in 2006 and
to-vehicle communication: Fair transmit power control for safety-critical 2010, respectively. She is currently working to-
information,” IEEE Trans. Veh. Technol., vol. 58, no. 7, pp. 3684–3703, ward the Ph.D. degree with École Polytechnique de
Sep. 2009. Montréal, Montreal, QC, Canada. Her research in-
[54] M. Torrent-Moreno, D. Jiang, and H. Hartenstein, “Broadcast reception terests include handling challenges and routing al-
rates and effects of priority access in 802.11-based vehicular ad-hoc gorithms in wireless sensor networks and quality of
networks,” in Proc. 1st ACM Int. Workshop Veh. Ad Hoc Netw., 2004, service and congestion control in vehicular ad hoc
pp. 10–18. networks.
[55] M. D. Felice, A. J. Ghandour, H. Artail, and L. Bononi, “Enhancing the
performance of safety applications in IEEE 802.11 p/WAVE vehicular
networks,” in Proc. IEEE Int. Symp. WoWMoM, 2012, pp. 1–9.
[56] V. D. Khairnar and S. N. Pradhan, “Simulation based evaluation of high-
way road scenario between DSRC/802.11 p MAC protocol and STDMA Samuel Pierre received the B.Eng. degree in
for vehicle-to-vehicle communication,” J. Transp. Technol., vol. 3, no. 1, civil engineering from École Polytechnique de
pp. 88–104, Jan. 2013. Montréal, Montreal, QC, Canada, in 1981; the B.Sc.
[57] J. Kakarla and S. S. Sathya, “A survey and qualitative analysis of multi- and M.Sc. degrees in mathematics and computer
channel MAC protocols for VANET,” Int. J. Comput. Appl., vol. 38, science from Université du Québec a Montréal,
no. 6, pp. 38–42, Jan. 2012. Montreal, in 1984 and 1985, respectively; the
[58] K. Bilstrup, E. Uhlemann, and E. G. Ström, “Medium access control in M.Sc. degree in economics from University of
vehicular networks based on the upcoming IEEE 802.11 p standard,” in Montreal, Montreal, in 1987; and the Ph.D. degree
Proc. 15th World Congr. ITS, Nov. 2008, pp. 408–413. in electrical engineering from École Polytechnique
[59] D. B. Rawat, D. C. Popescu, G. Yan, and S. Olariu, “Enhancing de Montréal in 1991. He is currently a Professor of
VANET performance by joint adaptation of transmission power and con- computer engineering with the École Polytechnique
tention window size,” IEEE Trans. Parallel Distrib. Syst., vol. 22, no. 9, de Montréal, where he is the Director of the Mobile Computing and Networking
pp. 1528–1535, Sep. 2011. Research Laboratory and the NSERC/Ericsson Industrial Research Chair in
[60] H. Hartenstein and K. Laberteaux, VANET: Vehicular Applications and next-generation mobile networking systems. He has authored or coauthored
Inter-Networking Technologies. Hoboken, NJ, USA: Wiley, 2010. more than 500 technical publications, including papers in refereed archival
[61] M. Behrisch, L. Bieker, J. Erdmann, and D. Krajzewicz, “Sumo- journals, textbooks, patents, and book chapters. His research interests include
simulation of urban mobility-An overview,” in Proc. 3rd Int. Conf. Adv. mobile computing and networking, cloud computing, and electronic learning.
Syst. SIMUL, 2011, pp. 55–60. Prof. Pierre is a Fellow of the Engineering Institute of Canada. He is a Regional
[62] D. Krajzewicz, G. Hertkorn, C. Rössel, and P. Wagner, “SUMO (simula- Editor of Journal of Computer Science and an Associate Editor of IEEE
tion of urban mobility),” in Proc. 4th Middle East Symp. Simul. Model., Communications Letters, IEEE Canadian Journal of Electrical and Computer
2002, pp. 183–187. Engineering, and IEEE Canadian Review.