Vous êtes sur la page 1sur 14

Reliable Intersection-Based

Traffic Aware Routing


Protocol for Urban Areas
Vehicular Ad Hoc Networks
Tasneem S. J. Darwish and Kamalrulnizam Abu Bakar
Are with the Pervasive Computing Research Group, Faculty of Computing,
Universiti Teknologi Malaysia (UTM), 81310 Skudai, Johor D. T, Malaysia
Email: tasneem83darwish@gmail.com (tasneem).
Khalid Haseeb
Is with the Computer Science Department,
Islamia college (Chartered University), Peshawar, Pakistan
Email: khalid.haseeb@icp.edu.pk

Abstract­—Nowadays, providing reliable vehicle to vehicle communications in


vehicular ad hoc networks is essential to serve the emerging intelligent trans-
portation systems. Traffic aware routing is considered a promising approach for
delivering data packets between geographically separated vehicles. However,
most of the existing traffic aware routing protocols greedily select the next-hop
based on distance. Consequently, high packet loss ratio occurs especially with
the vehicles high speed mobility. Some existing forwarding schemes attempted to
improve the next-hop selection by considering inter-vehicle communication qual-
ity or neighbours predicted positions. Nevertheless, such schemes depend on the
neighbours mobility information received through beacons regardless of its fresh-
ness. In addition, the differentiation between road and intersection areas while for-
warding packets was neglected. Therefore, a reliable traffic aware routing protocol is
proposed, which selects the next-hop based on roads structure, neighbours predicted po-
sitions and their received signal strength, and the recentness of the mobility information
received from neighbours. The simulation results show significant improvements in terms
of packet delivery ratio and end-to-end delays.

image licensed by ingram publihing


Digital Object Identifier 10.1109/MITS.2017.2776161
Date of publication: 19 January 2018

IEEE Intelligent transportation systems magazine • 60 • spring 2018 1939-1390/18©2018IEEE


I. Introduction

V
ehicular ad hoc networks (VANETs) are a core component of the future intelligent transportation
systems. VANETs consist of vehicles with on-board wireless communication facilities that are able
to establish ad hoc communications with their peers, as well as with infrastructure networks [1].
Moreover, the evolution of the standard communication protocols such as IEEE 802.11p and IEEE
1609 contributed towards realizing the concepts of VANETs. In addition, the huge investment of automo-
bile manufacturers is pushing towards the real-world deployment of vehicular networks. For instance,
due to the applied regulations, all new cars in USA are expected to be communication-enabled by 2017 [2].
Although most of VANETs early applications were to enhance traffic safety, the number of third-party
providers offering non-safety applications is growing recently [3]. For example, VANETs can be utilized
for real-time data collection in traffic control and roads maintenance systems [4], automated toll pay-
ment, enhanced navigation, location-based services [5], infotainment applications and Internet access
[6]. Thus, in the near future the demand for reliable and efficient communication, among vehicles (V2V)
and between vehicles and infrastructure (V2I), will rise dramatically. To enable geographi-
cally separated vehicles to communicate, VANETs employ multi-hop communica-
tions by relying on intermediate vehicles to forward packets. However, the
high speed mobility of vehicles causes intermittent inter-vehicle com-
munication [7]. Therefore, delivering data packets through multi-
hop communications in VANETs environment requires reliable
and efficient routing protocol.
As geographical routing forwards packets based on
vehicles positions, it is considered a promising routing
approach in dynamic environments for its scalability
and robustness against frequent topology changes
[8]. Greedy Perimeter Stateless Routing (GPSR)
is the basic geographical routing protocol pro-
posed for ad hoc networks [9]. However, GPSR
forwards packets greedily based on vehicles
positions only, and it has no consideration for
network and traffic status. Therefore, pack-
ets might be forwarded through roads with
low vehicular density or high level of net-
work disconnections. In addition, choosing
the furthest neighbour as a next forwarder
might result in forwarding data packets to
unreachable neighbours.
To eliminate conventional geographi-
cal routing protocols limitations geographic
routing was integrated with traffic aware-
ness resulting in traffic aware routing (TAR)
protocols, which adapt to variable traffic con­­
ditions. Although there is a variety of TAR
protocols, intersection-based TAR protocols are
considered the most adaptable to urban VANETs
dynamic environment [10]. Intersection-based TAR
protocols make routing decisions at each inter-
section after giving each adjacent road a weighted
score, where the road with the highest score is cho-
sen for packet forwarding.
Unfortunately, many of the existing TAR protocols em-
ploy the greedy forwarding method to select a next forwarder
within roads and at intersections, which highly affect the routing
performance. Although exploiting greedy-based forwarding reduces
the number of hops, it results in higher packet loss ratio, especially with

IEEE Intelligent transportation systems magazine • 61 • spring 2018


er [14]. In [15], the full path to-
wards destination is dis­covered,
then data packets are forwarded
In the near future the demand for reliable and efficient greedily between anchor points.
communication, among vehicles (V2V) and between vehicles In such protocols, vehicles ob-
tain their neighbours positions
and infrastructure (V2I), will rise dramatically. via beacon messages. Neverthe-
less, as vehicles tend to change
their directions rapidly near to
or at intersection areas, the used
high speed mobility. Therefore, some forwarding approach- neighbours positions might be inaccurate. Thus, greedy
es restricted the next forwarder selection area or predicted forwarding causes high packet loss in VANETs environ-
neighbours positions before selecting the next forwarder. ment since it has no consideration for the roads structure,
However, such approaches have no consideration for roads recentness of neighbours mobility information, and com-
structure and the possibility of changing vehicles direc- munication links quality. In particular, a vehicle might
tion due to approaching intersection areas. In addition, as a be chosen as a next-hop even after it turns into a different
subcategory of position-based routing protocols, TAR pro- road or is no longer located in the coverage area of the cur-
tocols depend on the beacon messages exchanged between rent forwarder.
vehicles to get neighbours mobility information. Neverthe- To tackle the limitations of classical greedy forward-
less, the vehicle might change its direction and speed after ing, the authors in [16] proposed to forward packets on
sending a beacon message, which results in using outdated paths constructed based on distance and communication
neighbour mobility information that causes deceptive next link lifetime between vehicles. Although considering link
hop selection. Therefore, this research contributes towards lifetime reduces the impact of greedy forwarding as stable
eliminating the previously mentioned problems by pro- links are chosen, the proposed mechanism does not con-
posing a Reliable Traffic Aware Routing (RTAR) protocol, sider roads structure and the quality of received signals. In
which selects the next forwarder based on multiple crite- addition, link lifetime estimation was based on the neigh-
ria. Basically, RTAR differentiates between the next for- bours mobility information received in beacon messages,
warder selection in roads and intersection areas. Moreover, which may result in inaccurate estimation especially when
RTAR considers neighbours predicted positions, Received outdated information is used. A different approach was in-
Signal Strength Indicator (RSSI) values, and the received troduced in [17], where in low vehicular density the next-
neighbours mobility information recentness while select- hop is chosen as the closest to the destination and moving
ing next forwarders. towards it. On the other hand, in high density situations the
This paper is structured as follows: Section II discusses next-hop is the closest to the destination and has the low-
several TAR approaches proposed in the context of VANETs est load compared to other neighbour vehicles. However,
and details the motivations of this paper. Section III, intro- this approach neglects the recentness of the neighbours
duces RTAR functionality, design and algorithms. Section mobility information and the signal strength, which might
IV presents and discusses the performance evaluation of result in choosing an unreachable next-hop. A restricted
RTAR. Finally, Section V concludes this article. forwarding algorithm was applied in [18], which chooses
a neighbour within a specific distance range to reduce
II. Motivations and Related Work packet loss caused by distant next-hop selection. However,
Traffic aware routing is a recent approach for delivering the restricted forwarding range is a fixed range which does
data in VANETs environment, which endeavours to solve not adapt to roads structure or transmission links quality.
the limitations of geographical routing protocols by in- Moreover, the next-hop is selected based on neighbours
volving the traffic awareness routing metrics in the rout- mobility information without considering its validity
ing decisions. However, TAR protocols performance is or recentness.
seriously influenced by the utilized next-forwarder se- To alleviate the impact of vehicles mobility, the authors
lection scheme. in [19], [20] proposed the improved greedy forwarding,
Several TAR protocols adopted the greedy forward- where the current positions of neighbour vehicles are
ing mechanism, where the selected next-hop is the one predicted, then the closest neighbour to next intersec-
closest to the destination or to next intersection (i.e., fur- tion is selected as a next-hop. In [21], [22] the authors en-
thest neighbour from the current forwarder) [11]–[13]. hanced the mechanism of improved greedy forwarding
As the wireless signal strength decreases by increasing by selecting the neighbour with a high RSSI value (i.e.,
transmission distance, the furthest neighbour might not RSSI value above 60% of the maximum recorded RSSI
receive a good quality signal from the current forward- value). However, neglecting roads’ structure may lead to

IEEE Intelligent transportation systems magazine • 62 • spring 2018


­ eceptive next-hop selection, es-
d
pecially near to intersection areas
where vehicles tend to cha nge
t hei r directions rapidly. Moreover, Delivering data packets through multi-hop communications
the neighbours positions prediction in VANETs environment requires reliable and efficient
is done based on the instantaneous
mobility i nfor mat ion rec eived routing protocol.
via beacons, which does not de-
scribe vehicles movement over a
period of time. Thus, there is a
high probability of carrying out inaccurate neighbours po- the required quality of service while routing packets. Un-
sition prediction that leads to unreliable next-forwarder se- fortunately, depending on neighbours mobility informa-
lection. Therefore, high packet loss might occur, especially tion for calculating link duration without considering the
near intersection areas, as the aforementioned forward- recentness of such information may result in a deceptive
ing schemes have low reliability and may select unreach- next-hop selection. In [27], a contention-based forwarding
able neighbours. is employed, where the closest neighbour to destination
In a different approach, the virtual distance for all further forwards the data packet. Nevertheless, conten-
neighbours which are closer to the destination than cur- tion-based forwarding might generate redundant packet
rent forwarder is calculated [23], [24]. Here, the virtual replications and transmissions, which may cause broad-
distance refers to the largest two-hop distance, which cast storms and network congestions.
equals to the distance between the current forwarder and In TAR protocols, next-hop selection has a significant
its neighbour vehicle plus the distance between the neigh- impact on the reliability of the routing process. As can
bour vehicle and its closest neighbour to the destination. be deduced from the previously discussed related work,
Eventually, the neighbour with the largest virtual distance many of the existing TAR forwarding schemes suffer the
is chosen as the next hop. Although this mechanism de- limitations of greedy forwarding. Some TAR schemes at-
pends on two-hop distance for selecting a next-hop, com- tempted to address the problem by exploiting neighbours
munication links quality and stability were not considered. positions prediction, restricted forwarding area or RSSI
Moreover, the freshness of neighbours mobility informa- values as an indication of communication link quality.
tion was not considered in this approach, which makes dis- However, the aforementioned solutions depend on the
tance estimation more error-prone as it is calculated over neighbours mobility information received through bea-
two hops. con messages without considering the validity and fresh-
In [25], the authors improved the greedy forward- ness of such information, which may result in selecting
ing performance by increasing neighbour tables ac- a deceptive next-hop. In addition, the existing schemes
curacy. Actually, two problems caused by conventional have no consideration for roads structure, even though
hello messages scheme with fixed broadcasting intervals vehicles tend to change their direction rapidly within or
were identified. First, when old neighbours are no longer near intersection areas. To overcome the aforementioned
within transmission range, their entries are not deleted problems, RTAR introduces a reliable next-hop selection
on time. Second, some new neighbours are not added to scheme, which considers neighbours current position,
the neighbour table as soon as they enter the transmission communication links quality, roads structure and the re-
range since they need to wait for the next hello message. centness of neighbours mobility information. Moreover,
To address such problems, for each neighbour recorded in RTAR gives priority for the neighbours that their mobil-
the neighbour’s table, the lifetime is estimated. More- ity information is received recently to be next-forwarders
over, when two vehicles are approaching each other the at junctions, which contributes towards avoiding choos-
time after which they can directly communicate is esti- ing neighbours that turned to different roads. In addition,
mated by an intermediate vehicle. Next, without waiting RTAR utilizes the traffic and network status measurement
for the next hello message both vehicles are informed for adjacent roads to forward packets through the best
with the estimated time. Nevertheless, communication road, thereby improving the routing performance of traf-
links lifetime estimation is error-prone, due to the ef- fic aware routing.
fect of variable vehicles speed and the mobility patterns
at intersections. In addition, the introduced enhancement III. Reliable Traffic Aware Routing (RTAR) Protocol
generates extra network overhead that rises with high This section explains in details the proposed RTAR proto-
driving speeds. col for V2V communication in VANETs environment. RTAR
On the other hand, the authors in [26] selects the next- makes its routing decision when the data packet reaches an
hop by favouring either distance or link stability to satisfy intersection area to choose the best next road for ­forwarding

IEEE Intelligent transportation systems magazine • 63 • spring 2018


siders traffic and network status on adjacent roads, while
Algorithm 1 Road Area Reliable Routing (RARR)
making routing decisions at intersections to select the best
road for packet forwarding. Basically, for each adjacent road
Require: CF neighbour table, CF position, destination position
Ensure: Forward data packet through a road a score is calculated using its status evaluation results, then
 1: Set NF = Null the score is used while making routing decisions. Thus, pro-
 2: Set N = The CF neighbours table viding accurate road status evaluation results leads to better
 3: if CF and destination vehicle are in the same road then routing decisions that dynamically adapt to variable roads
  4:   TP = destination position
conditions. However, the network overhead generated by
 5: else
  6:   TP = TI.center the road evaluation process may degrade the routing proto-
 7: Endif cols performance. Therefore, to employ accurate and light-
 8: PFG = {n: n ! N / n.RSSI # RSSI threshold / n.predictedPosition in CF.coverage / weight road status evaluations, RTAR exploits the Real-time
 9: n.predictedPosition closer to TP than CF } Traffic and Network Status Measurement (RTNSM) process
10: PFG sameRoad = {n ! PFG: n is in same road as CF }
introduced in [29].
11: NF = CF
12: if PFG sameRoad .size! = 0 then //Beginning of Case 1 The main idea of RTNSM is based on creating and for-
13:    for i = 1; i 1= PFG sameRoad .size; i ++ do warding control packets between each two consecutive
14:       if PFG sameRoad [i ] closer to TP than NF then intersections to evaluate the road status in terms of both di-
15:           NF = PFG sameRoad [i ] rection vehicular density, inter-vehicle communication
16:       Endif
link lifetime among neighbours and the road’s network load,
17:    Endfor
18: Endif //End of Case 1 which provides a comprehensive road status description.
19: if NF == CF then Afterwards, the roads evaluations results are announced
20:    if PFG.size! = 0 then //Beginning of Case 2 at intersections areas to be available for routing purposes.
21:       Set NF1 = Null The RTNSM process evaluates roads when changes in roads
22:       Set NF2 = Null
status are expected to occur, resulting in detecting and re-
23:       Set Z = CF
24:       for i = 1; i 1= PFG.size; i ++ do flecting the roads real-time status more accurately with
25:          if PFG[i] closer to TP than then minimized network overhead. This is achieved by control-
26:             if CF received PFG[i] beacon in the last beacon interval then ling and adapting the generation frequency of control pack-
27:                 NF1 = PFG [i ] ets based on changes in road status.
28:                 Z = NF1
To improve the packet forwarding process, RTAR se-
29:             else if CF received PFG[i] beacon in the last two beacon
30: intervals then lects the next forwarder (NF) vehicle in a reliable way,
31:                 NF2 = PFG [i ] whereas multiple criteria related to current forwarder
32:                 Z = NF2 (CF) position (i.e., located at a road or an intersection) and
33:             Endif the conditions of its available neighbours are considered.
34:             Endif
To achieve this goal, RTAR introduces two algorithms for
35:          Endif
36:       Endfor selecting the NF. The first algorithm is the Road Area Re-
37:       if NF1! = Null then liable Routing (RARR), which is applied when the CF is
38:          NF = NF1 within a road area. However, if the CF vehicle is currently
39:       else if NF2! = Null then within an intersection area, the Intersection Area Reliable
40:          NF = NF2
Routing (IARR) algorithm is applied. The following subsec-
41:       Endif
42:    Endif tions explain the reliable packet forwarding procedure of
43: Endif //End of Case 2 RTAR in more details.
44: if NF == CF then
45:    for i = 1; i 1= PFG.size; i ++ do //Beginning of Case 3 A. The Road Area Reliable Routing (RARR)
46:       if PFG[i] closer to TP than NF then
When the CF is located within a road area and the destina-
47:           NF = PFG [i ]
48:       Endif tion vehicle is not the CF’s direct neighbour, RTAR selects
49:    Endfor the NF vehicle based on Algorithm 1 (RARR). To explain
50: Endif the workflow of Algorithm 1, references to its line numbers
51: if NF == CF then are used in this section.
52:    CF carries the data packet
First, if the destination vehicle is located on the same
53: Endif //End of Case 3
road as the CF, then the destination vehicle position is
considered the Target Point (TP). In case the destination
based on roads status. As vehicles are the main network en- vehicle and the CF are not on the same road, then the TP
tities in VANETs, routing protocols performance is highly is the center of the Target Intersection (TI). The TI is the
affected by the availability of vehicles and their inter-vehicle first closest intersection to the destination, where the data
communication quality [28]. Therefore, RTAR protocol con- packet should reach in order to carry out Algorithm 2 and

IEEE Intelligent transportation systems magazine • 64 • spring 2018


make the intersection-based routing decision. To forward
Algorithm 2 Intersection Area Reliable Routing (IARR)
the data packets towards TP, a NF vehicle needs to be cho-
sen. To select a NF vehicle in a reliable way, the CF forms
Require: CF neighbour table and road information table
the Potential Forwarders Group (PFG) which includes the Ensure: Forward data packet towards the best road
neighbours that fulfil the following conditions:  1: Set NF = Null
■■ The average RSSI value obtained by calculating the RSSI  2: Set N = The CF neighbours Table
mean of the received beacon messages/data packets  3: Set AR = Adjacent roads to CI
from the neighbour must be greater than the predefined  4: if CF does not have LREresult f or all AR then
 5:    Send CPs to get LREresult
threshold as formulated in Equation 1. The RSSI max val-  6:    wait to receive LREresult
ue is the maximum value among all the recorded RSSI  7: Endif
values of the received beacons and data packets from  8: for i = 1; i 1= AR.size; i ++ do //calculate adjacent roads scores
all neighbours.  9:    Roads [i ] .score = b 1 ·AR [i ] .LRE result + b 2 ·AR [i ] .PD
■■ Neighbour’s predicted position is in the CF transmis-
10: Endfor
11: RankedRoads = sortRoads(Roads) //Rank adjacent roads
sion coverage area. Predicted positions are calculated 12: for j = 1; j 1= RankedRoads.size; j ++ do
based on neighbours mobility information (i.e. speed, 13:    TP = RankedRoad[j]:CTI:center
direction and latest position), which is obtained through 14:    PFG [j ] sameRoad = {n: n ! N / n.RSSI # RSSI threshold /
beacons. 15: n.predictedPosition in CF.coverage / n.predictedPosition closer to TP than CF /
■■ Neighbour’s predicted position is closer to TP than CF.
16: n is in the road RankedRoads [j ]}
17:    if PFG [j ] sameRoad .size! = 0 then
18:       Set NF1 = Null
Neighbour_RSSI $ RSSI threshold 19:       Set NF2 = Null

where RSSI threshold = 0.6 ) RSSI max (1) 20:       Set Z = CF
21:       for x = 1; x 1= PFG [j ] sameRoad .size; x ++ do
Afterwards, the CF vehicle forms a sub set of the PFG 22:          candidateNeighbour = PFG [j ] sameRoad [x]
23:          if candidateNeighbour closer to TP than Z then
which includes the neighbours that belong to the same 24: //Giving priority to the neighbour with the most updated mobility information
road as CF; i.e., PFG sameRoad (line 10, Algorithm 1). Next, if 25:             if CF received candidateNeighbour’s beacon in the last beacon
the PFG sameRoad set is not an empty set, then the CF apply interval then
Case 1 to select the NF; otherwise, Case 2 is applied. How- 26:                NF1 = candidateNeighbour
ever, in case of failing to find a neighbour in Case 2, then 27:                Z = NF1
28:             else if CF received atleast one candidateNeighbour’s beacon
the NF is selected based on Case 3. in the last two beacon intervals then
Case 1 (lines 12–18, Algorithm 1): In this case, the CF 29:                NF2 = candidateNeighbour
still has some neighbours on the same road and located be- 30:                Z = NF2
tween itself and TP. Figure 1 demonstrates this case where 31:             Endif
TI (1) is the TP. Therefore, the CF inspects the neighbours 32:          Endif
33:       Endfor
included in the PFG sameRoad group, and then the closest 34:       if NF1! = Null then
neighbour to the TP is selected as NF. 35:          NF = NF1
Case 2 (lines 20–41, Algorithm 1): In this case, the CF 36:          break
does not have a neighbour located on the same road and 37:       else if NF2! = Null then
closer to TP than itself. This may occur when the CF is 38:          NF = NF2
39:          break
approaching the TI area, as shown in Figure 2. Therefore, 40:       Endif
the CF inspects the PFG group and chooses the closest 41:    Endif
neighbour to TP, which at least one beacon message was 42: Endfor
received from it in the last beacon interval (lines 26–28 43: if NF == Null then //Select a neighbour in the CI area
and 35–36, Algorithm 1). However, if such a neighbour 44:    PFG CI = {n: n ! N / n.RSSI # RSSI threshold /
45: n.predictedPosition in CF.coverage / n.predictedPosition closer to CI.
is not available, then the CF selects from PFG the closest center than CF }
neighbour to TP which at least one beacon message was 46:    NF = x: x ! PFG CI / x is the closest to CI.center
received from it in the last two beacon intervals (lines 29– 47: Endif
31 and 37–38, Algorithm 1). This is to give priority for the 48: if NF == Null then
neighbours to be chosen as next forwarders, whose their 49:    CF carries the data packet
50: Endif
mobility information was received recently. Thereby, the
probability of forwarding data packets to unreachable
neigbours is reduced. to TP to be the NF (lines 42–48, Algorithm 1). However,
Case 3 (lines 42–51, Algorithm 1): When neighbours if such a neighbour is not available, then the CF keeps
with recent mobility information are not available in PFG, ca r rying the packet until it finds a suitable NF (lines
the CF inspects PFG and chooses the closest neighbour 49–51, Algorithm 1).

IEEE Intelligent transportation systems magazine • 65 • spring 2018


In fact, beacon messages carry instantaneous mobility warding data packets to unreachable neighbours, thereby
information of neighbours, which might cause prediction improving the reliability of data packet routing.
errors, especially in VANETs dynamic environment. In ad-
dition, neighbours positions prediction is more error-prone B. The Intersection Area Reliable Routing (IARR)
when vehicles are nearby or in intersection area, as vehicles Once a data packet reaches an intersection area (TI (1) in
might change their directions by turning into different roads Figure 2), the CF forwards the data packet directly if the
suddenly. Accordingly, as the CF in Case 2 is approaching an destination vehicle is its direct neighbour. However, if the
intersection, and the selected NF neighbour is not located on destination vehicle is unreachable (i.e., not a direct neigh-
the same road, thus, the selected neighbour might turn into bour), RTAR selects the best road to forward the packet
a different road soon and become unreachable. Therefore, through, by executing Algorithm 2 (IARR). References to
selecting the neighbour with the recently updated mobil- Algorithm 2 line numbers are used in this section to ex-
ity information to be the NF, reduces the probability of for- plain the algorithm workflow.

Destination
C Vehicle
G

D H

T1(1) B F
T1(2)

Next E
A Forwarder

Source
Vehicle

Fig 1 Data packet forwarding within a road based on RARR algorithm (case1).

Destination
C Vehicle
G

D H

T1(1) B F
T1(2)

Next
Forwarder A E

Source
Vehicle

Fig 2 Data packet forwarding towards intersection based on RARR algorithm (case2).

IEEE Intelligent transportation systems magazine • 66 • spring 2018


The main concept of IARR is to forward data packets as the closest neighbour to CTI that its beacon message
through the best adjacent road and reliably choose a NF was received in the last beacon interval (lines 23–27, Al-
from the selected road. To choose a neighbour in a reliable gorithm 2). In case such a neighbour is not available, the
way, the neighbour’s predicted position and average RSSI NF is selected as the closest neighbour to CTI, which at
value are considered as well as the recentness of the re- least one of its beacon messages was received in the last
ceived neighbour’s mobility information (i.e., the reception two beacon intervals (lines 28–31, Algorithm 2). However,
time of the recently received beacon message). This is to if the considered PFG sameRoad does not include a neighbour
give priority to the neighbours that sent their mobility in- which fulfils any of the two conditions mentioned previ-
formation recently, thereby reducing the probability of for- ously, the PFG sameRoad of the road with the second highest
warding packets to unreachable neighbours. For adjacent score is formed and inspected. This process is repeated
roads evaluation, the RTNSM process, introduced in [29], is until the CF selects a NF from one of the adjacent roads,
employed to provide the lightweight road evaluation (LRE) otherwise, the closest neighbour to the CI in the PFG CI set
results for each of the adjacent roads. As mentioned previ- is selected as NF (lines 43–47, Algorithm 2). The PFGCI set
ously, the LRE result represents the road status in terms includes the neighbours that satisfy the conditions men-
of vehicular traffic density, inter-vehicle connectivity and tioned in Section III-A, whereas CI is the target point (i.e.,
network load. TP). In the case of not finding any neighbour that fulfils
Initially, to select the best road, the CF evaluates each any of the aforementioned selection criteria, the CF keeps
adjacent road, which it does not have its valid LRE results, carrying the data packet until it finds a suitable NF (lines
except the road where the packet came from (lines 4–7, Al- 48–50, Algorithm 2).
gorithm 2). After receiving the roads’ LRE results, the CF Accordingly, IARR forwards packets through the best
calculates the roads scores based on Equation 2 (line 8-ten, road in terms of road’s progressive distance towards the
Algorithm 2), whereas b 1 and b 2 are the weighting factors destination and its traffic and network status. As RTNSM
for the two routing metrics, namely the road’s LRE result and provides accurate real-time road evaluation results, IARR
the progressive distance ^PDh towards destination. The routing decision is highly adapted to changes in traffic
weighting factors represent each of the two routing met- and network conditions. In addition, packet forward-
rics significance in calculating the roads scores, whereas ing reliability is improved by considering the neighbours
b 1 + b 2 = 1 . The sensitivity analysis of the weighting fac- predicted position, received RSSI values, and mobility in-
tors values is discussed in Section III-C. The calculated formation recentness. Moreover, since RTNSM produces
road score value is in the range [0.0, 1.0] as both PD and low network overhead, network congestion and transmis-
LRE result have values in the same range. sion collisions are reduced, which leads to more reliable
packet forwarding.
RoadScore = b 1 ·LRE result + b 2 ·PD (2)
C. Sensitivity analysis of weighting factors
The PD metric measures the distance progress that the This section presents the analysis of the weighting factors
selected road contributes in the process of delivering data values incorporated in the road score calculation process.
packets to the destination. Thus, roads which make data Equation 2 formulates the road score calculation, which
packet closer to the destination are given higher scores. The combines the values of LRE results and PD of the evalu-
PD is calculated using Equation 3, where Distance fromCTI ated road based on the weights of b 1 and b 2 . Basically, the
and Distance fromCI are the distance from Candidate Target weighting factors are used to determine the percentage of
Intersection (CTI) to destination and distance from Cur- contribution for each component in calculating the road
rent Intersection (CI) to destination, respectively. CTI is score. Accordingly, the summation of the percentages of
where the data packet should reach in case it is forwarded contributions that b 1 and b 2 represent must equal to 100%
through the associated road. ^i.e., b 1 + b 2 = 1 h . For instance, considering the values of
b 1 = 0.5 and b 2 = 0.5 means that each of LRE results and
Distance fromCTI
PD = 1.0 - (3) PD contributed by 50% in the calculation of the road score.
Distance fromCI
In general, there are no optimal ratios to be assigned to
Afterwards, the CF sort adjacent roads based on their b 1 and b 2 which give the best routing performance in all
calculated scores (line 11, Algorithm 2). Then, the road scenarios. Therefore, the sensitivity analysis for b 1 and b 2
with the highest score is considered first. Subsequently, ratios is conducted to determine the most suitable values
the PFG sameRoad of the considered road is formed by ap- for making routing decisions.
plying the same conditions introduced in Section III-A, Three different weighting factors configurations are
whereas the CTI of the considered road is the target point evaluated to analyse the effect of b 1 and b 2 values on rout-
(i.e., TP) (lines 12–16, Algorithm 2). Through inspecting ing performance, where RTAR (0.2,0.8) corresponds to
the PFG sameRoad of the considered road, the NF is selected the configuration of b 1 = 0.2 and b 2 = 0.8, RTAR (0.5,0.5)

IEEE Intelligent transportation systems magazine • 67 • spring 2018


refers to the configuration of b 1 = 0.5 and b 2 = 0.5, and packet delivery ratio and end-to-end delay in high and low
RTAR (0.8,0.2) represents the configuration of b 1 = 0.8 and vehicular density scenarios.
b 2 = 0.2. These configuration are adopted from a previous Figure 3 shows the packet delivery ratio of the three
sensitivity analysis introduced in [19]. In particular, RTAR considered configuration in high and low vehicular den-
(0.2,0.8) favors progressive distance and RTAR (0.8,0.2) sity scenarios. In addition, Figure 4 depicts the end-to-end
favors LRE results value while calculating the road score delay of packet delivery for each of the evaluated configu-
value. The three considered configuration RTAR (0.2,0.8), rations. Accordingly, configuration RTAR (0.2,0.8) results
RTAR (0.8,0.2) and RTAR (0.5,0.5) are evaluated in terms of in better performance in terms of packet delivery ratio

100 100

98 98

96 96
Packet Delivery Ratio (%)

Packet Delivery Ratio (%)


94 94

92 92

90 90

88 88

86 86
RTAR (0.5,0.5)
84 84 RTAR (0.2,0.8)
82 82 RTAR (0.8,0.2)
80 80
2 4 6 8 10 12 14 16 2 4 6 8 10 12 14 16
Number of CBR Connections Number of CBR Connections
High Vehicular Density High Vehicular Density

Fig 3 Packet delivery ratio with various number of CBR connections under different RTAR configurations.

0.14 0.12
RTAR (0.5,0.5)
0.12 RTAR (0.2,0.8)
0.1
RTAR (0.8,0.2)
0.1
0.08
End-to-End Delay (s)

End-to-End Delay (s)

0.08
0.06
0.06

0.04
0.04

0.02 0.02

0 0
2 4 6 8 10 12 14 16 2 4 6 8 10 12 14 16
Number of CBR Connections Number of CBR Connections
High Vehicular Density High Vehicular Density

Fig 4 End-to-end delay for different number of CBR connections under different RTAR configurations.

IEEE Intelligent transportation systems magazine • 68 • spring 2018


and end-to-end delay in high ve-
hicular density scenario. This is
because in high vehicular den-
sity scenario the majority of roads Basically, to make routing decisions at intersections and
have high evaluations and provide forward data packets through the most reliable roads,
good connectivity, which reduces
the significance of selecting the RTAR exploits lightweight and accurate roads’ traffic and
road with highest road evaluation network status measurements.
result. In such a case, favouring
distance reduces the delivery delay
resulting in higher packet delivery
level. On the other hand, in low ve-
hicular density scenario, RTAR (0.8,0.2) produced the best performance is studied using 16 Constant Bit Rate (CBR)
performance in both packet delivery ratio and end-to-end connections, while varying data sending rate from 1.0 to
delay. This is due to giving higher preference to roads with 0.1 seconds. A CBR connection is established when a source
better traffic and network conditions even though such vehicle starts sending data packets (in constant bit rate) to a
roads may increase the routing distance, which is more destination vehicle through multi-hop communication. In
suitable for the highly dynamic VANETs environment es- some protocols the location services are merged with rout-
pecially in low vehicular density situation. ing process [32]–[34]. However, in this work it is assumed
Based on the results of Figure 3 and 4, it is obvious that that the destination vehicles’ position can be obtained by
each weighting factor has a positive and negative impact utilizing a location service system such as HLS or GLS [35].
on the performance of RTAR in different vehicular density Variety of scenarios are considered to evaluate the perfor-
scenarios. However, giving equal values to both weighting mance of RTAR in high, average and low vehicular den-
factors results in a more balanced influence of LRE results sities. In fact, high vehicular density reduces the vehicles
and PD, which contributes towards optimizing routing mobility speed, which results in more accurate predictions
decisions at intersections under different vehicular den- for vehicles positions and longer inter-vehicle link lifetime.
sity conditions. Therefore, b 1 and b 2 are assigned equal In addition, high vehicular density provides more alterna-
weights ^ b 1 = b 2 = 0.5 h, which gives equal importance tives while choosing the next hop. On the other hand, low
for both roads evaluations and progressive distance while vehicular density allows vehicles to move faster, which
making routing decisions in any vehicular density scenar-
io. In all subsequent simulations b 1 and b 2 are assigned
Table I. Simulation parameters.
uniform values which are equal to 0.5.

Parameters Values
IV. RTAR Performance Evaluation
In this section, RTAR routing performance is evaluated Simulation time 400 seconds
against the Traffic Flow-Oriented Routing (TFOR) proto- Mobility model Car following model
col [24] and Intersection-based Connectivity Aware Rout-
Vehicular Density High density (40-50 vehicle/km/lane)
ing (ICAR) protocol [22], in terms of packet delivery ratio Average density (13-16 vehicle/km/lane)
(PDR) and end-to-end delay. The implementation and eval- Low density (6-8 vehicle/km/lane)
uation phases of this research involved the usage of OM-
Maximum vehicle speed 60 km/h
NET++ 4.6 [30] as a network simulation environment along
with SUMO [31] for urban traffic mobility simulation. The Beacon interval 1 seconds
protocols evaluation is carried out based on the simulation N con 12 vehicles
parameters in Table I. As RTAR, TFOR, and ICAR are all
Payload size 512 bytes
intersection-based routing protocols, part of Manhattan
map (latitude: 39.1912 to 39.1839 and longitude: –96.5737 MAC layer protocol IEEE 802.11p
to –96.5629) is used to emulate vehicles movement in ur- Transport layer protocol UDP
ban areas. The map data and structure is obtained from Network layer protocol IPv6
OpenStreetMap contributions. The area of simulation map
Propagation model Two Ray Ground reflection
is approximately 2000 m # 2000 m with 112 bidirection-
al roads and 64 intersections. In addition, the real traffic Transmission range 300 m
regulations (e.g., traffic lights, speed limits and traffic pri- Channel capacity 18 Mbps
orities), which are applied in that part of Manhattan city,
Transmission Power 20mW
are also considered in the vehicular traffic simulation. The

IEEE Intelligent transportation systems magazine • 69 • spring 2018


results in short inter-vehicle links lifetime and unreliable effect on RTAR performance in comparison with existing
packet forwarding. Each value in the results graphs is the schemes. This is due to using the lightweight RTNSM pro-
average of 10 values recorded through 10 simulation repe- cess for evaluating roads, which reduces the roads evalu-
titions. In addition, the confidence intervals are calculated ation overhead.
and included in all graphs, whereas 95% confidence level Considering the average vehicular density scenario,
is considered. RTAR, ICAR and TFOR packet delivery ratio is presented as
a function of packet sending rate in Figure 6. By analysing
A. Packet Delivery Ratio the graphs of Figure 6, it can be deduced that in compari-
For the high vehicular density scenario, Figure 5 illus- son to the existing schemes RTAR has achieved the highest
trates the behaviour of the three simulated protocols in packet delivery ratio, which scored an average improve-
high ­vehicular density, while varying the sending rate from ment of 11%. In particular, the performance of ICAR and
0.1 to 1.0 seconds. It is obvious that RTAR has the highest TFOR degrades with faster data packet sending rates (i.e.,
PDR, as an average of 29% improvement is achieved in de- from 0.4 to 0.1 seconds), while the performance of RTAR
livering data packets; while considering the fastest send- remains in a relatively stable situation. This is because of
ing rate. Moreover, Figure 5 shows that faster data sending avoiding to choose unreachable neighbours and reducing
rates (e.g., sending data packet every 0.1 or 0.2) has less the number of trials while sending a data packets, which
loosen network load impact.
Figure 7 depicts the behaviour of RTAR, ICAR and TFOR
95 in the scenario of low vehicular density; while considering
90 different packet sending rates. Based on the presented re-
sults, it can be inferred that RTAR has superior performance
Packet Delivery Ratio (%)

85
in comparison to the existing schemes as it attained 15%
80
average improvement. In low vehicular density scenarios,
75 the network is less congested and there is lower contention
70 on transmission channels as fewer vehicles are exchanging
beacon messages. Accordingly, the main cause for packet
65
loss in low vehicular density is the rapid movement of vehi-
60 TFOR cles. Therefore, with the developed algorithms of IARR and
ICAR
55 RARR, RTAR maintained a PDR around 90%, while ICAR
RTAR
50
and TFOR had an average PDR of 80%. This is due to apply-
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 ing the multi criteria selection while choosing the next hop,
Data Packet Sending Rate (s) which increases the forwarding process reliability.
Based on the provided simulation results in this sec-
Fig 5 Packet delivery ratio in high vehicular density scenario. tion, it can be concluded that RTAR achieved a distinguished

100 100

95 95
Packet Delivery Ratio (%)

Packet Delivery Ratio (%)

90
90
85
85
80

80 RTAR TFOR
75 ICAR
ICAR
TFOR RTAR
75 70
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Data Packet Sending Rate (s) Data Packet Sending Rate (s)

Fig 6 Packet delivery ratio while varying data packet sending rate in Fig 7 Packet delivery ratio while varying data packet sending rate in low
average vehicular density scenario. vehicular density scenario.

IEEE Intelligent transportation systems magazine • 70 • spring 2018


­ erformance in terms of PDR as compared to existing
p gation, the performance of RTAR still needs to be tested
work, especially in the low vehicular density scenario. This while considering a more realistic propagation model.
is due to employing the IARR and RARR algorithms, which
reliably choose a next-forwarder by considering multiple B. End-to-End Delay
criteria. As a result, more priority is given to neighbours For the high vehicular density scenario, RTAR, ICAR and
with high RSSI and more updated mobility information; TFOR are compared in terms of their average consumed
while considering the difference between roads and in- end-to-end delay in Figure 8. RTAR achieved 70% aver-
tersections areas routing. In addition, IARR utilized the age reduction of end-to-end delay at sending rate equals
LRE results as roads status evaluations, which are used to 0.1 seconds. This is due to the reliable routing decision
as traffic awareness metric while making routing deci- of RTAR that exploits the road evaluations provided by
sions at intersection areas. Accordingly, RTAR forwards RTNSM, which produces low routing overhead. As a result,
data through more reliable roads, which have the high- the number of attempts to transmit a data packet is reduced
est average inter-vehicle link life time, lowest network as well.
load and highest vehicular density moving in the packet Figure 9 shows the average end-to-end delay for RTAR,
forwarding direction. In high vehicular density scenario ICAR and TFOR of the average vehicular density scenario
the RTNSM process generates higher network load than simulations. Based on Figure 9, it is obvious that RTAR
low or average vehicular density scenario [29]. Therefore, has a lower end-to-end delay in comparison to existing
the PDR value in high density scenario is lower than that schemes, especially when network load is increased. More
achieved in low or average density scenarios. Moreover, precisely, in comparison with the existing schemes, RTAR
as in low vehicular density scenario the network is not has achieved an average of 11% reduction of end-to-end
congested, the only cause for packet loss is the poor received delay with a sending rate of 0.1 seconds. Actually, in aver-
signals or out of date neighbours mobility information. age vehicular density there are less vehicles to forward the
Nevertheless, RTAR succeeded in reducing the effects of packets and their movement is faster than the high density
these two factors through introducing IARR and RARR al- scenario. However, as RTAR employs a multi criteria selec-
gorithms and improved the PDR in low vehicular density tion for the next-hop it is more adaptable to such dynamic
scenario remarkably. environment than existing schemes. Moreover, consider-
As can be observed from the simulation parameters in ing the recentness of the neighbours mobility information
Table I, the used propagation model is considered simple prevents using outdated mobility information, which reduces
and less realistic. The reasons for choosing such propaga- errors in neighbours positions predictions.
tion model are: 1) to avoid loosing packets due to signal Considering the low vehicular density scenario, Fig-
propagation issues; and 2) to focus on addressing the prob- ure 10 describes the performance of RTAR, ICAR and TFOR
lems neighbours mobility information freshness and the in terms of the average end-to-end delay for delivering a
position of the next forwarder (in street or intersection). packet. It is obvious that RTAR has slightly higher end-
Therefore, to be able to analyse the effects of signal propa- to-end delay than ICAR and TFOR, which reached an ­average

0.6 0.1
TFOR 0.09 TFOR
0.5 ICAR ICAR
RTAR 0.08 RTAR
End-to-End Delay (s)

End-to-End Delay (s)

0.07
0.4
0.06
0.3 0.05
0.04
0.2
0.03
0.02
0.1
0.01
0 0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Data Packet Sending Rate (s) Data Packet Sending Rate (s)

Fig 8 End-to-end delay for various data packet sending rates in high Fig 9 End-to-end delay for various data packet sending rates in average
vehicular density scenario. vehicular density scenario.

IEEE Intelligent transportation systems magazine • 71 • spring 2018


the highly dynamic networks of vehicles. Therefore, this
0.1 research work addressed the issue of greedy forwarding
TFOR and improved the reliability of traffic aware routing by in-
0.09 ICAR troducing the RTAR protocol. Basically, to make routing de-
RTAR cisions at intersections and forward data packets through
End-to-End Delay (s)

0.08 the most reliable roads, RTAR exploits lightweight and ac-
curate roads’ traffic and network status measurements.
0.07
In addition, the two algorithms RARR and IARR are intro-
0.06 duced, which perform the next-hop selection at roads and
intersection areas respectively. Moreover, RTAR selects for-
0.05 warders based on multiple criteria, which considers neigh-
bours predicted positions and their RSSI values as well as
0.04 the recentness of the received mobility information from
neighbours. Consequently, forwarding data to unreachable
0.03
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 neighbours is eliminated which leads to more reliable data
Data Packet Sending Rate (s) packet forwarding and enhanced routing performance.
For future work, this protocol needs to be examined while
Fig 10 End-to-end delay for different for various data packet sending considering a more realistic evaluation environment with
rates in low vehicular density scenario. a more complicated signal propagation models. In addition,
RTAR can be improved and tested to access infrastructure
networks over multi-hop communications.
i­ncrement of 17%. This is because RTAR attempts to for-
ward data packets through reliable roads even though Acknowledgment
such roads might be longer in distance, which increases This work has been done under the Post-doctoral Fellow-
the routing distance and end-to-end delay. Moreover, RTAR ship Scheme of Universiti Teknologi Malaysia for the proj-
aims to increase the next forwarder selection reliability ect:” Big data analysis on vehicular networks for intelligent
which may increase the number of hops. However, this transportation systems”
slight increment in end-to-end delay comes as a price for
the significant improvements that RTAR exhibited in terms About the Authors
of packet delivery ratio. Tasneem S. J. Darwish is a researcher
By studying the end-to-end delay graphs of this section, of Universiti Teknologi Malaysia under
it can be concluded that RTAR succeeded in minimizing the Post-doctoral Fellowship Scheme for
the delivery delay, especially in the high vehicular density the project:” Big data analysis on ve-
scenario. This is achieved through the reliable routing of hicular networks for intelligent trans-
RTAR, which considers neighbours predicted positions, the portation systems”. She is a member
received RSSI values, the recentness of the mobility infor- of the Pervasive Computing research
mation received from neighbours, and the current position group. She received her B.Sc. degree in computer engineer-
of CF whether it is in a road or intersection area. However, ing from Islamic University of Gazah in 2005 and M.Sc. de-
in the low vehicular density scenario, optimum roads for gree in electronics and electrical engineering from the
reliably delivering data packets might not be the shortest. University of Glasgow, United Kingdom, in 2007. She re-
Therefore, increasing the reliability of data packets deliv- ceived her PhD degree in computer science in 2017 from
ery slightly increases the end-to-end delay in low vehicu- Universiti Teknologi Malaysia. From 2008 to 2013 she was
lar density cases. The performance evaluation of RTAR was working as a research and lecturer assistant at the Faculty
performed over a relatively small area of Manhattan city. of applied engineering, University of Palestine. Her current
However, in our future work a larger area is going to be research focuses on vehicular communications, wireless
considered to evaluate the scalability of RTAR. ad hoc networks, and mobile computing.

V. Conclusion Kamalrulnizam Abu Baker is a profes-


As TAR protocols utilize traffic status while selecting roads sor in Computer Science at Universiti
for packet routing, they are considered a promising solu- Teknologi Malaysia (UTM), Malaysia,
tion for urban VANETs environment. However, most of and a member of the Pervasive Com­­
the existing TAR protocols adopted the greedy forwarding puting research group. He received his
mechanism, which selects next forwarders based on dis- B.Sc. degree in computer science from
tance. As a result, such protocols have low performance in the Universiti Teknologi Malaysia (UTM),

IEEE Intelligent transportation systems magazine • 72 • spring 2018


Malaysia, in 1996, M.Sc. degree in computer communica- [13] C. C. Lo and Y. H. Kuo, “Traffic-aware data delivery strategy for ve-
hicular ad hoc networks,” in Proc. 17th Int. Conf. Advanced Communi-
tions and Networks, from Leeds Metropolitan University, cation Technology, July 2015, pp. 18–22.
United Kingdom, in 1998, and his Ph.D. degree in computer [14] T. Darwish, K. A. Bakar, and A. Hashim, “Green geographical rout-
ing in vehicular ad hoc networks: Advances and challenges,” Comput.
science from Aston University, United Kingdom, in 2004. Elect. Eng., 2016.
His research interest includes mobile and wireless com- [15] V. Naumov and T. R. Gross, “Connectivity-aware routing (car) in ve-
hicular ad-hoc networks,” in Proc. 26th Int. IEEE Conf. Computing
puting, ad hoc and sensor networks, information security Communications, May 2007, pp. 1919–1927.
and grid computing. He is a member of ACM and Interna- [16] A. Benslimane, S. Barghi, and C. Assi, “An efficient routing protocol
tional Association of Engineering. He involves in many re- for connecting vehicular networks to the internet,” Pervasive Mobile
Comput., vol. 7, no. 1, pp. 98–113, 2011.
search projects and is a referee for several scientific journals [17] H. Hashemi and S. Khorsandi, “Load balanced vanet routing in city
and conferences. environments,” in Proc. 75th IEEE Vehicular. Technology Conf., May
2012, pp. 1–6.
[18] Y. Xiang, Z. Liu, R. Liu, W. Sun, and W. Wang, “GEOSVR: A map-based
Khalid Haseeb (received his M.Sc. de- stateless {VANET} routing,” Ad Hoc Netw.(Special Issue on Theory,
­Algorithms and Applications of Wireless Networked Robotics and Re-
gree in information technology from In- cent Advances in Vehicular Communications and Networking), vol. 11,
stitute of Management Sciences, Pakistan. no. 7, pp. 2125–2135, 2013.
[19] M. Jerbi, S.-M. Senouci, T. Rasheed, and Y. Ghamri-Doudane, “To-
He completed his Ph.D in computer sci- wards efficient geographic routing in urban vehicular networks,”
ence from Faculty of Computing at Uni- IEEE Trans. Veh. Technol., vol. 58, no. 9, pp. 5048–5059, 2009.
versiti Teknologi Malaysia in June 2016. [20] S. Bilal, S. A. Madani,, “Enhanced junction selection mechanism for
routing protocol in vanets,” Int. Arab. J. Inform. Technol., vol. 8, no. 4,
During Ph.D, he was a member of the pp. 422–429, 2011.
Pervasive Computing research group. Since 2008, he works at [21] N. Alsharif, S. Céspedes, and X. Shen, “ICAR: Intersection-based con-
nectivity aware routing in vehicular ad hoc networks,” in Proc. IEEE
Computer Science Department, Islamia College Peshawar, Int. Conf. Communications, June 2013, pp. 1736–1741.
Pakistan. His research areas include sensors and ad-hoc net- [22] N. Alsharif and X. Shen, “ICARII: Intersection-based connectivity
aware routing in vehicular networks,” in Proc. IEEE Int. Conf. Com-
works, network security, cloud computing and mobile com- munications, June 2014, pp. 2731–2735.
munication. He is a reviewer for many reputed international [23] Y. He, C. Li, X. Han, and Q. Lin, “A link state aware hierarchical road
routing protocol for 3D scenario in vanets,” in Internet Vehicles Tech-
journals and conferences. nologies Services, Series Lecture Notes in Computer Science, vol. 8662,
R.-H. Hsu and S. Wang, Eds. New York: Springer International, 2014,
pp. 11–20.
References [24] I. A. Abbasi, B. Nazir, A. Abbasi, S. M. Bilal, and S. A. Madani, “A traffic
[1] F. A. Silva, A. Boukerche, T. R. M. B. Silva, L. B. Ruiz, E. Cerqueira, and
A. A. F. Loureiro, “Vehicular networks: A new challenge for content-deliv- flow-oriented routing protocol for vanets,” EURASIP J. Wireless Com-
ery-based applications,” ACM Comput. Surv., vol. 49, no. 1, pp. 11:1–11:29, mun. Netw., vol. 2014, no. 1, pp. 1–14, 2014.
June 2016. [25] C. Li, C. Zhao, L. Zhu, H. Lin, and J. Li, “Geographic routing protocol
[2] D. Naboulsi and M. Fiore, “Characterizing the instantaneous connec- for vehicular ad hoc networks in city scenarios: A proposal and analy-
tivity of large-scale urban vehicular networks,” IEEE Trans. Mobile sis,” Int. J. Commun. Syst., vol. 27, no. 12, pp. 4126–4143, 2014.
Computing, no. 99, pp. 1–1, 2016. [26] Y. Sun, S. Luo, Q. Dai, and Y. Ji, “An adaptive routing protocol based on
[3] G. Yan and S. Olariu, “A probabilistic analysis of link duration in ve- QOS and vehicular density in urban vanets,” Int. J. Distrib. Sens. Netw.,
hicular ad hoc networks,” IEEE Trans. Intell. Transport. Syst., vol. 12, vol. 2015, pp. 5:5–5:5, Jan. 2015.
no. 4, pp. 1227–1236, Dec. 2011. [27] R. Akamatsu, K. Obara, and H. Shigeno, “Road-oriented geographic
[4] R. Du, C. Chen, B. Yang, N. Lu, X. Guan, and X. Shen, “Effective urban routing protocol for urban vehicular ad hoc networks,” in Proc. 29th
traffic monitoring by vehicular sensor networks,” IEEE Trans. Veh. Int. IEEE Conf. Advanced Information Networking Applications Work-
Technol., vol. 64, no. 1, pp. 273–286, Jan. 2015. shops, Mar. 2015, pp. 721–726.
[5] S. Zeadally, R. Hunt, Y.-S. Chen, A. Irwin, and A. Hassan, “Vehicular ad [28] T. Darwish and K. A. Bakar, “Traffic density estimation in vehicular ad
hoc networks (vanets): Status, results, and challenges,” Telecommun. hoc networks: A review,” Ad Hoc Netw., vol. 24, Part A, pp. 337–351, Jan.
Syst., vol. 50, no. 4, pp. 217–241, 2012. 2015.
[6] F. Malandrino, C. Casetti, C. F. Chiasserini, C. Sommer, and F. [29] T. Darwish and K. Abu Bakar, “Lightweight intersection-based traffic
Dressler, “The role of parked cars in content downloading for vehicu- aware routing in urban vehicular networks,” Comput. Commun., vol. 87,
lar networks,” IEEE Trans. Veh. Technol., vol. 63, no. 9, pp. 4606–4617, pp. 60–75, 2016.
Nov. 2014. [30] A. Varga. (2009). Omnet++. [Online]. Available: http://www.omnetpp.org/
[7] Y. A. Daraghmi, C. W. Yi, and I. Stojmenovic, “Forwarding methods [31] D. Krajzewicz, J. Erdmann, M. Behrisch, and L. Bieker, “Recent devel-
in data dissemination and routing protocols for vehicular ad hoc net- opment and applications of SUMO: Simulation of Urban MObility,” Int.
works,” IEEE Network, vol. 27, no. 6, pp. 74–79, Nov. 2013. J. Adv. Syst. Meas., vol. 5, no. 3&4, pp. 128–138, Dec. 2012.
[8] S. M. Bilal, C. J. Bernardos, and C. Guerrero, “Position-based routing [32] R. Khadim, A. Ennaciri, M. Erritali, and A. Maaden, “Coupling of
in vehicular networks: A survey,” J. Netw. Comput. Appl., vol. 36, no. 2, geographic location-based service and routing for wireless sensor
pp. 685–697, 2013. networks,” in Proc. Int. Arab Conf. Information Technology, 2016,
[9] B. Karp and H. T. Kung, “GPSR: Greedy perimeter stateless routing for pp. 1–7.
wireless networks,” in Proc. 6th Annu. ACM Int. Conf. Mobile Comput- [33] M. Ayaida, M. Barhoumi, H. Fouchal, Y. Ghamri-Doudane, and L.
ing Networking, New York, 2000, pp. 243–254. Afilal, “Joint routing and location-based service in vanets,” J. Parallel
[10] T. Darwish and K. Abu Bakar, “Traffic aware routing in vehicular ad Distrib. Comput., vol. 74, no. 2, pp. 2077–2087, 2014.
hoc networks: Characteristics and challenges,” Telecommun. Syst., [34] S. M. Das, H. Pucha, and Y. C. Hu, “Performance comparison of scal-
vol. 61, no. 3, pp. 489–513, 2016. able location services for geographic ad hoc routing,” in Proc. 24th
[11] C. Zhao, C. Li, L. Zhu, H. Lin, and J. Li, “A vehicle density and load Annu. IEEE Joint Conf. Computer Communications Societies, 2005, vol. 2,
aware routing protocol for vanets in city scenarios,” in Proc. Wireless pp. 1228–1239.
Communications Signal Processing Int. Conf., Oct. 2012, pp. 1–6. [35] M. Ayaida, H. Fouchal, L. Afilal, and Y. Ghamri-Doudane, “A comparison
[12] G. Li, L. Boukhatem, and S. Martin, “An intersection-based QOS rout- of reactive, grid and hierarchical location-based services for vanets,” in
ing in vehicular ad hoc networks,” Mobile Netw. Appl., vol. 20, no. 2, Proc. IEEE Vehicular Technology Conf., 2012, pp. 1–5.
pp. 268–284, 2015. 

IEEE Intelligent transportation systems magazine • 73 • spring 2018

Vous aimerez peut-être aussi