You are on page 1of 14

Digital Communications and Networks 4 (2018) 13–26

Contents lists available at ScienceDirect

Digital Communications and Networks


journal homepage: www.keaipublishing.com/en/journals/digital-communications-and-networks/

Trade-off between accuracy, cost, and QoS using a beacon-on-demand


strategy and Kalman filtering over a VANET
Zineb Squalli Houssaini a, *, Imane Zaimi b, Maroua Drissi b, Mohammed Oumsis b, c,
Saïd El Alaoui Ouatik a
a
IT Laboratory and Modelling (ILM), Faculty of Science Dhar El Mahraz (FSDM), Sidi Mohammed Ben Abdellah University (USMBA), Fez, Morocco
b
LRIT, Associated Unit to CNRST (URAC 29), Faculty of Science, Mohammed V University, Rabat, Morocco
c
High School of Technology, Mohammed V University, Sale, Morocco

A R T I C L E I N F O A B S T R A C T

Keywords: Vehicular Ad-hoc Networks (VANETs) have been suggested as an active and powerful field of research to mitigate
VANET environmental problems and challenges. The main challenge in a VANET is to ensure routing with a good Quality
Kalman filter of Service (QoS). The Greedy Perimeter Stateless Routing (GPSR) protocol is one of the most promising position-
GPSR based routing mechanisms used to overcome this challenge. Its effectiveness depends entirely on the information
Beacon on demand on a node's mobility and the precision of this information. By broadcasting periodic beaconing within trans-
Estimation
mission boundary ranges, GPSR can manage neighbors' mobility information and maintain up-to-date lists of
Routing
QOS
neighbours. Nevertheless, information on the position of a neighboring vehicle quickly becomes outdated, which
negatively influences the efficiency of the routing. In order to monitor information mobility and to increase the
QoS in this challenging area, position estimation needs to be considered.
Thus, in this study, we examine the position estimation problem, and propose an improvement to the GPSR
protocol, named KF-GPSR, where each vehicle estimates in real time the position of its neighbors using the
Kalman filter algorithm. Indeed, by employing this strong estimation technique, it is possible to reduce consid-
erably the frequency of exchanged beacon packets, while maintaining high position accuracy. For greater reli-
ability, we also propose an extension to KF-GPSR, called BOD-KF-GPSR, that uses the “beacon-on-demand”
process only if a node needs to rediscover its neighborhood. Simulation experiments using the network simulator
NS-2 are presented to demonstrate the ability and usefulness of our two proposals. Here, we compare the pro-
posed protocols against diverse common protocols: GPSR, AODV, DSR, and ZRP. The results show that BOD-KF-
GPSR achieves a significant enhancement in terms of its packet delivery ratio, routing cost, normalized routing
load, end-to-end delay, and throughput.

1. Introduction VANET-based applications are expected to provide methodical and


proven solutions for innovative, resource-efficient, and trustworthy
Current transport systems are pivotal to achieving sustainable future intelligent transport systems. These applications cover many
development in a rapidly changing world. However, there are many fields, such as safety services (i.e., collision avoidance and security
intractable challenges directly related to this sector, such as pollution distance), and services providing road information, advertisements,
caused by carbon emissions, traffic accidents, resource depletion, and the and entertainment for drivers and passengers (i.e., driver assistance,
difficulty of ensuring a pleasant driving experience. To this end, the automatic parking, and internet access) [2]. A VANET is a network in
innovation and improvement of existing systems are necessary to over- which communication between nearby smart vehicles (inter-vehicle
come these problems. communication, V2V), or between vehicles and roadside infrastructures
In this context, a Vehicular Ad-hoc Network (VANET) is a promising (vehicle-to-roadside communication, V2I) is possible without any fixed
technology for emerging Intelligent Transport Systems (ITS) [1]. infrastructure. VANET are considered inheritors of Mobile Ad-hoc

* Corresponding author.
E-mail addresses: zinebsqualli@gmail.com (Z.S. Houssaini), imanzaimi@gmail.com (I. Zaimi), drissimaroua@gmail.com (M. Drissi), mohamed.oumsis@um5.ac.ma (M. Oumsis), s_
ouatik@yahoo.com (S.E.A. Ouatik).

https://doi.org/10.1016/j.dcan.2017.09.001
Received 21 May 2017; Accepted 8 September 2017
Available online 14 September 2017
2352-8648/© 2017 Chongqing University of Posts and Telecommunications. Production and hosting by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://
creativecommons.org/licenses/by-nc-nd/4.0/).
Z.S. Houssaini et al. Digital Communications and Networks 4 (2018) 13–26

Networks (MANETs) which benefit from technologies, such as short information may not be exact, and can become outdated well before a
radio transmission ranges, low bandwidth, self-organization, and data exchange takes place. This may adversely affect the functionality of
self-management [3]. Nonetheless, a VANET is still unique in that it the protocols. Therefore, it is strongly recommended that neighbors'
behaves distinctly differently to other infra-structureless networks. In positions are updated as often as possible. The naive solution to this issue
fact, it exhibits four specific characteristics that offer opportunities to is to reduce time interval between two successive beacon packets, which
increase network performance, but at the same time that present will increase the frequency of their dissemination. However, this
considerable challenges [4]. These characteristics are as follows: approach leads to many packets being exchanged, increasing the over-
head of the network and route costs, and causing a high channel occu-
 High dynamic topology: this manifests in frequent changes of the pancy and many collisions.
network topology. Indeed, a node can move quickly in and then out of Thus, the real challenge is to seek a trade-off between accuracy, cost,
range of another node owing to its high-speed environment. and other Quality of Service (QoS) metrics. Our proposed solution
 Intermittent network connection: the communication link between comprises two steps:
two vehicles may be disconnected, especially in the case of a low
density or high mobility of vehicles. The frequent cut-offs can 1. Integrate and adapt the powerful Kalman filter algorithm to not only
considerably increase end-to-end delays and packet loss. estimate the current positions of neighbors, but also to predict breaks
 Mobility modeling and prediction: the mobility patterns of vehicles in links while effectively decreasing the frequency of beacon packet
depend on the traffic environment which tends to be limited by pre- dissemination, which differs from previous studies (e.g., Refs.
built highways, roads, and streets as well as the speed limit. Thus, the [26,27]).
future position of a vehicle can be anticipated based on its speed and 2. Use the two parameters (i.e., the link state and number of neighbors)
the street map, although this is made difficult by the vehicle's to determine if it is necessary to update the neighbors list by
movement, which depends on the driver's behavior and road requesting hello packets (beacon packet-on-demand approach).
situations.
 Energy is not a significant issue because the batteries in the vehicles Through the “prediction and correction” methods used in the Kalman
are self-chargeable; therefore, there is no power constraint in such a filter algorithm and the proposed “beacon-on-demand” approach, we can
network. exchange periodic beacon packets at a far lower frequency (i.e., widening
beacon period) while offering more accurate neighbor position infor-
Furthermore, the routing process is strongly influenced by some of the mation. Hence, the load of communications is greatly reduced and the
above-mentioned characteristics and by the use of ineffective routing QoS performance is improved. We check the effectiveness of this solution
protocols originally conceived for a MANET [5]. Indeed, there are several by implementing the Greedy Perimeter Stateless Routing (GPSR) proto-
MANET routing algorithms, but only a few can be employed in a VANET. col [16], which is one of the typical and most promising examples of
In addition, the literature attests that their performance in a VANET is geographical protocols. Thus, we first propose the KF-GPSR protocol,
unsatisfactory owing to the substantial differences between the two which combines the Kalman filter algorithm and the GPSR protocol.
networks [6]. Therefore, there is a need to design routing protocols Then, we extend the KF-GPSR protocol to the BOD-KF-GPSR protocol
specific to vehicles. Accordingly, the data routing issue is a major design (KF-GPSR with the proposed “beacon-on-demand” approach). Finally,
goal of a VANET. The routing protocols are grouped using diverse through simulation, we provide a thorough and complete performance
methods and aspects [7–12]. In this study, we consider the well-known evaluation to compare the KF-GPSR and BOD-KF-GPSR protocols with
and widely used classification that divides routing protocols into two various well-known protocols: AODV (Ad hoc On-demand Distance
classes: topological routing protocols [13] and position-based routing Vector) [17], DSR (Dynamic Source Routing) [18], GPSR, ZRP (Zone
protocols (geographical routing protocols) [14]. Topology-based routing Routing Protocol) [19], and our previous proposal, GPSR þ Predict. The
protocols use link information in routing tables to achieve packet for- simulation results show the benefits of the proposed variants of GPSR
warding, while geographical routing protocols use the current under different network conditions.
geographic location of nodes to determine the optimal route as packets The remainder of this paper is organized as follows. Section 2 con-
are forwarded. The latter is becoming increasingly popular, and is tains a review of relevant works in the domain of VANET routing in
considered the most attractive category for highly dynamic environ- addition to existing solutions related to our protocol. In Section 3, we
ments, such as VANETs. This is because it benefits from the progress and introduce the GPSR mechanism, its mobility management, and provide
accessibility of localization systems and electronic maps. However, its an overview of the background theory of the Kalman filter. Sections 4 and
usefulness depends on two essential tasks: the routing, which defines the 5 describe, KF-GPSR and BOD-KF-GPSR protocols, respectively. Our re-
strategies and the rules followed by nodes when choosing the best sults, an analysis, and a general discussion are presented in Section 6.
neighbor to send data packets to destination. The second one is the Finally, we conclude the paper in Section 7.
monitoring, which defines how nodes manage the mobility information
of their neighbors. A common, monitoring strategy is to use a beaconing 2. Related work
procedure. Here, each node shares its geographic location with its im-
mediate neighborhood by periodically sending a simple hello packet (or In the literature, most works focus on the routing mechanism. In order
periodic beacon packet). Since a GPS1 device is available in most existing to preserve network performance in critical circumstances, as in a
vehicles and will be in all future vehicles, accurate knowledge of the VANET, geographical routing protocols are the most appropriate. Thus,
real-time positions of vehicles is achievable. Most forceful geographical we give special attention to one of the best protocols belonging to this
protocols, whether designed or recommended for a VANET, need category: the well-known GPSR protocol which still attracts the interest
extremely precise location data for realistic and efficient routing services of researchers, who continue to propose improvements.
which greatly influence the effectiveness of a routing decision. Moreover, Various proposals have been designed to improve GPSR and to
owing to the dynamic nature of vehicles, a neighbor's position address diverse issues. For example, the authors of [20] presented a new
GPSR routing protocol, based on the node's buffer that controls network
congestion. The geographical locations of nodes and the lengths of their
1
residual buffers are also taken into account. The authors of [21] proposed
There are many localization mechanisms than using to standard GPS, such as dead
reckoning, cellular localization, and image/video localization [15]. However, their use is
a novel GPSR-based routing protocol. They considered four metrics,
infrequent, especially in a VANET. Since GPS receivers can be installed easily in vehicles, a namely, distance to destination, vehicle density, trajectory, and the
number of vehicles already include this technology. available bandwidth to make forwarding decisions, thereby, minimizing

14
Z.S. Houssaini et al. Digital Communications and Networks 4 (2018) 13–26

packet loss and packet delay. Then, in Ref. [22], the authors considered iteratively. In [30], the authors proposed a Mobility Assisted Location
just three metrics: location information, direction information, and link Management protocol, called MALM, in which a vehicle calculates the
quality to produce the routes. In Ref. [23], GPSR was enhanced for very current location of neighboring vehicles using a Kalman filter estimation.
high mobility. Indeed, a new concept based on vehicles MOvement They demonstrated that the algorithm can provide accurate location in-
PRediction, defined as MOPR, was conceived in order to perform the formation availability. The authors of [32] affirmed that a Kalman fil-
routing process. MOPR estimates the stability of each communication ter-based estimation improves the performance of beacon safety message
link of the network in terms of communication lifetime. Then, it selects dissemination in vehicular networks. The authors of of [33] presented a
the most stable route using the most stable intermediate links from the method that uses a Kalman filter to predict the residual lifetime of a
source to the destination. communication link. Their method relies on predicting the position of a
Vehicle mobility management (monitoring) and the accuracy of ve- vehicle and then using these estimates to find the residual lifetime of the
hicles' geographical position information have also driven many research link. Then, the authors of [34] were interested in predicting the location
efforts. The authors of [24] studied the impact of mobility metrics (i.e., of a vehicle accurately. Their proposed prediction method relies on a
the beacon interval and the node's speed) on the inaccuracy of the Kalman filter method that enables both a technical and a fundamental
location information in a GPSR protocol over a MANET. Through this analysis of location information. In addition, their extended experiments
study, the problem of a Neighbor Break Link (NBL) was identified. showed a higher degree of location performance.
Therefore, the authors proposed prediction schemes to solve this draw- However, there are many other uses of Kalman filters in terms of
back and to avoid forwarding data packets to nodes outside the trans- routing mechanisms. Kalman filtering enhances routing protocols by
mission range. In Ref. [25], the authors showed that the information exploiting its capabilities in terms of reliable performance in predicting.
stored in a neighbor's list is often outdated, and no longer reflects the real In reference [35], the authors conceived a novel solution for Cognitive
topology of the network, which leads to performance degradation. Radio Vehicular Ad hoc Networks (CR-VANETs) using a Kalman filter
Therefore, they proposed several simple alternatives that adjust a bea- algorithm. They select the best relay node by combining cognitive
con's interval in order to improve the accuracy of neighborhood infor- capability and a prediction technique to overcome the problems of
mation. Additionally, the authors of [26] analyzed the impact of position spectrum scarcity and high latency in a VANET. The main goal of their
information inaccuracy on GPSR, caused by both the Beacon Packet In- work is to establish a stable routing path between the source and the
terval Time (BPIT) and the Node Moving Speed (NMS). To improve the destination, and for the communication to be entirely V2V in a highway
reliability of a neighbor's list, they proposed a Fuzzy Logic Dynamic scenario, without using any fixed units. Then, the authors of Ref. [36]
Beaconing (FLDB) strategy based on the correlation between the NMS, proposed a scalable Mobility-Adaptive Location Service (MALS) with
Number of Neighboring Nodes (NoNNs), and BPIT, using a Fuzzy Logic Kalman-based prediction to handle the location updates sent by nodes
Control (FLC) mechanism. In the same context, the authors of [27] and to resolve their queries. Additionally, a hierarchy geographic clus-
confirmed through simulations that the position information of a tering structure has been used to address the scalability issue. In [37], a
neighbor vehicle fast becomes outdated in a VANET. To overcome this hybrid routing algorithm KPHR (Kalman Prediction Hybrid Routing) is
challenge, they proposed adding to periodic messages the information proposed. KPHR makes full use of the store-carry-forward mechanism, as
that will be used to predict the nodes' positions in the near future. Thus, well as the Kalman algorithm to predict the real-time location of a
whenever a vehicle requires the positions of its neighbors, the predicted vehicle. The results can be regarded as the weight adjustment values
positions are utilized. Note that we have also previously proposed an between the new measurement values of the observation model and the
enhancement to the GPSR protocol. In fact, the strong relation between predictions based on all previous measurements. Similar to the reported
inaccurate information in neighbors' lists, reliability, and the required work, each of them takes a quite different approach to modeling and
QoS over a VANET, motivated us to propose a new concept named solving the challenge of network connectivity and the relationship be-
GPSR þ Predict to improve and adapt the GPSR protocol to this envi- tween a vehicle's localization and the connectivity of a network. We
ronment [28,29]. This proposed extension was based on a neighbor's consider that using a stochastic mobility prediction model in an urban
movement prediction, where each vehicle estimates its location ahead for scenario is less interesting than highway or freeway scenarios. The po-
the next interval, and disseminates these data along with the current sitional variation of the vehicles is constrained by the road infrastructure
position, hence, making smart routing decisions. Simulation results and the higher mobility compared to the other topologies, so basic GPSR
showed the effectiveness of the improvements. can successfully manage the location of vehicles in advance in these types
From the aforementioned studies and their proposed solutions, we of scenarios. Here, we investigate an appropriate model for the position
found that the routing mechanism of GPSR depends fully on the infor- accuracy in a VANET under a realistic highway topology.
mation on a node's mobility and its accuracy. However, the great To the best of our knowledge, no other studies offer a high degree of
mobility required in a VANET exceeds the default GPSR monitoring neighbor position accuracy using the Kalman filter, or decrease the
capability. While mobility prediction in VANETs has presented potential communication load and the overhead on the network by reducing the
advantages most current research papers use the deterministic mobility beacon's broadcast frequency. The proposed model comprises two basic
prediction model [24,25,27,29]. This model represents a fast and simple steps. First, the algorithm adapts a stochastic Kalman filter model and
method, but is inaccurate because of noise and interference that may implements it with the GPSR protocol in order to have more accurate
occur. As a result it is very limited compared to the other current information on a neighbor's position. Consequently, it is possible to
methods. The authors of provided a complete review of prediction maintain updates of their movement, predict broken links with neigh-
models. Their survey illustrates previous successful attempts aimed at bors, and eliminate neighbors that are outside the forwarder's radio
improving mobility in a wireless network, as well as adapting these range, all without needing to disseminate beacon packets in shorter pe-
techniques to mobile ad hoc networks. They also described several suc- riods. Second, the required broadcast of a hello packet request is initiated
cessful applications to routing and location management in ad hoc net- by a node in the case of a loss of about 50% of its neighbors. Hence, we
works. Among these prediction methods, the Kalman filter is a popular ensure an improvement on GPSR protocol performance and all
algorithm. On the one hand, it predicts the state of the system, using a QoS metrics.
stochastic approach, and then corrects the estimation using data pro-
vided by the measurements. In the literature, there are many successful 3. Background
uses of Kalman filtering methods in the context of VANETs. In Ref. [31],
the researchers proposed an automatic trust prediction mechanism, 3.1. Greedy Perimeter Stateless Routing (GPSR) protocol overview
based on a Kalman filter for trust computation. The method fuses existing
trust with the verified behavior of nodes to obtain an updated trust value In this section, we first introduce the geographical routing protocol

15
Z.S. Houssaini et al. Digital Communications and Networks 4 (2018) 13–26

GPSR [16]. Then, we describe its two main strategies. The first is called a
greedy forwarding strategy, where an intermediate node receives a
packet and forwards it to a neighbor node closest to the destination. The
second, defined as a perimeter strategy, is invoked when the greedy
process fails.

3.1.1. GPSR: strategies and process


GPSR is one of the most promising routing protocols in ad-hoc
wireless networks. It has received much attention owing to its advan-
tages as a geographical protocol over other proactive or reactive pro-
tocols, such as AODV, DSR, and ZRP. The most important benefit of GPSR
is that it eliminates some limitations of topology-based routing using
precise movement information (i.e., geographic position) with the aid of
a positioning system such as a GPS device.
To forward packets using GPSR, a node makes a decision on routing
based only on three factors: its geographical position, the positions of its Fig. 2. Greedy forwarding strategy.
(one-hop) neighbors, and the position of the final destination node. As
stated at the beginning of this section, GPSR takes into account two
processes: greedy forwarding, used whenever possible, and perimeter Table 1
List of neighbors of the source node S.
forwarding when greedy forwarding is not suitable (see Fig. 1).
Node ID neighbori(xi,yi)

3.1.1.1. Greedy forwarding strategy. The greedy forwarding strategy is n1 n1(x1,y1)


used to select the next hop forwarder. The idea is that a sending node S n2 n2(x2,y2)
n3 n3(x3,y3)
forwards a packet to its neighbor that is closest to the destination.
n4 n4(x4,y4)
Additionally, the packet is advanced to the final destination at every hop. n5 n5(x5,y5)
As a first step, node S compares the distance Distance(ni, dest) between all
its neighbors ni, including itself, and the final destination node dest. Then,
it chooses the neighbor nmin with the minimum distance. In the other
terms, when a source node needs to send packet, it searches for the
optimal candidate within its radio range in its list of neighbors. There-
fore, each vehicle participating in the routing mechanism selects the best
candidate neighbor as a next-hop forwarder using the following Equa-
tion (1):

nmin ¼ argmin Distanceðni ; destÞ (1)


This strategy is adopted throughout the forwarding process, from
node to node, until the packet reaches the final destination. Fig. 2 clearly
illustrates this strategy. Here, node S needs to send a packet to D, where
S's radio range is denoted by the dotted circle that surrounds nodes n1, n2,
n3, n4, and n5. The node n1 is listed in the neighbors list of S (see Table 1).
This node is closer to D than is S or any other neighbor. Therefore, the
packet moves from n1, n1 to n11, and n11 to D.
Fig. 3. Greedy strategy failure.
This strategy is powerful and efficient while there is a close neighbor
to the destination, but does not work in some situations. In fact, it not
performs well when the forwarder node, which is outside the scope, is that when arriving at node n11, the next traversed edge is the one
itself nearest to the destination (see Fig. 3). sequentially counterclockwise about n11, as shown in Fig. 4. Around the
Here, n11 is the closest node to D and is selected as the next hop. void region, the greedy transmission takes over.
However, being out of the radio range of n11, D becomes inaccessible in
one hop. Hence, the greedy forwarding strategy no longer works. In order 3.1.2. GPSR: mobility management
to solve this problem, the node switches to the perimeter forwarding Position accuracy is very important for any position-based routing
strategy to transfer the packet around the void region. protocol. It influences greatly the effectiveness of routing decisions. As
such, each node must manage the mobility of its neighbors by main-
taining updates of their positions. In the GPSR protocol, this procedure is
3.1.1.2. Perimeter forwarding strategy. In the perimeter forwarding pro-
assured by broadcasting periodically, within a time interval, a typical
cess, the protocol draws a planar graph of a node's neighbors and applies
simple beacon packet that provides the position of all nodes and those of
the right-hand rule to avoid the void region. The right-hand rule states
their direct neighbors (their one-hop neighbors). Each node transmits a
beacon packet, within its own radio range, that comprises the node
identifier (e.g., IP address) and the coordinates of its geographical posi-
tion (x, y). Simultaneously, it creates or updates its neighbors' lists ac-
cording to the information it receives. The beacon packets play a
significant role by ensuring the dissemination of information between
vehicles. Indeed, sending periodic beacon packets spreads the informa-
tion over the topology in a one-hop manner, which minimizes the
network overhead. However, it has several drawbacks, such as inaccur-
acies in the node positions, especially in a VANET. Owing to the intense
Fig. 1. Switch between the two strategies of GPSR.

16
Z.S. Houssaini et al. Digital Communications and Networks 4 (2018) 13–26

is an optimal controller, in the stochastic sense. The Kalman filter aims at


estimating the state of a system, which changes over time in a linear
manner, from a series of noisy, indirect, inaccurate, and uncertain ob-
servations. “The Kalman filter is a set of mathematical equations that
provides an efficient computational (recursive) means to estimate the
state of a process, in a way that minimizes the mean of the squared error.
The filter is very powerful in several aspects: it supports estimations of
past, present, and even future states, and it can do so even when the
precise nature of the modeled system is unknown” [40]. In other words,
the robustness of the Kalman filter is manifested in its ability to predict
and correct errors, not only in sensors but also in the model itself. It
considers two principal vectors: a state vector, which represents the
estimated state, and a measurement vector, which reflects the observed
state. The latter is used to correct the estimation. The state of the system
is represented as a vector (Xt) that represents the minimal set of data
needed to describe the dynamic behavior of the system. The Kalman filter
assumes that the current state (at time t) has evolved from the previous
state (at time t  1), in which cases, the current state is predicted
Fig. 4. Perimeter forwarding strategy: right-hand rule. considering a stochastic process according to the following Equation (2):

speed mobility, opposite directions, and inaccuracies in predicting a Xt ¼ AXt1 þ wt (2)


driver's direction at intersections, the precision of the neighbor's location
decreases and can become erroneous between two successive trans- where A is the state transition matrix. It attaches the current state to the
missions of beacon packets. This failing can negatively affect the choice previous state. The indices t and (t  1) signify, respectively, the current
of the next hop, thereby decreasing the routing protocol performance. state and the previous state.
Thus, it is recommended that information on locations be accessed as The observation vector Y attaches the measurement received from
often as possible. GPS to the state. It is derived from the current node and is calculated
Here, we find that a reliable prediction of the neighbor's next move as follows:
resolves the issue. Using a prediction models based on current position
Yt ¼ HXt þ ut (3)
and velocity, each vehicle can anticipate the change of all direct neigh-
bors, and then estimate their future position. This linear statistical The Kalman filter equations are grouped into estimation and correc-
location prediction scheme is typically easy, optimal, and deterministic tion equations.
in the sense that the prediction error variance is minimal. Most of the
current research on mobility predictions is based on piecewise linear 3.2.1. Estimation equations
statistical node motion between successive beacon packet updates. Thus, The one-step-ahead estimation is expressed by:
they do not consider random phenomena. Nonetheless, according to the
moving nature of nodes in a VANET scenario, their positions change Xt∕t1 ¼ AXt1 þ wt (4)
constantly and several random phenomena may occur, which decrease and the one-step-ahead process covariance matrix that estimates the
the reliability of predictions obtained using this type of model. In addi- error covariance is formulated as follows:
tion, for accuracy and a reliable estimation, it is possible to apply a
Kalman filter algorithm that incorporates all provided information and Pt∕t1 ¼ APt1 þ MQt1 M T (5)
takes into account the random phenomena that occur. As mentioned in
Ref. [38], “It processes all available measurement, regardless of their 3.2.2. Correction equations
precisions, to estimate the current value of the variable of interest, with The change or residual measurement represents the difference be-
use of: (1) knowledge of the system and measurement device dynamics, tween what is actually observed (Equation (3)) and the predicted values
(2) the statistical description of the system noises, measurement errors (C*Xt ∕t1). For that, we use Equation (6):
and uncertainly in the dynamics models, and (3) any available infor-
mation about initial conditions of the variable of interest.” In ¼ Yt  ðHXt∕t1 Þ (6)
Furthermore, the Kalman gain plays a very important role in the
3.2. Overview of the Kalman filter
correction step. It helps to determine the confidence level that we can use
in the model. In fact, the Kalman gain is always between 0 and 1. For
The Kalman filter is a prediction algorithm used extensively in various
example, if there is no process noise and Gt is equal to 0, then the Kalman
fields related to signal processing, radar, and image processing [38,39]. It

Fig. 5. Kalman Filter overview.

17
Z.S. Houssaini et al. Digital Communications and Networks 4 (2018) 13–26

Table 2 Once the updated estimation is determined, the update process


Summary of notation. covariance matrix, which represents the error in the estimation, is given
Notation Definition by the following equation:
Xtjt1 Priori state vector estimated using the previous state vector
Xtjt Posteriori estimate of the state Xt
Pt∕t ¼ ð1  Xt HÞPt∕t1 (9)
Ptjt1 The covariance matrix of the priori error
Fig. 5 provides a general overview of the Kalman filter and the
Ptjt The covariance matrix of the posteriori error
wt The system error noise at t aforementioned equations.
ut The measurement noise on the GPS at t In this study, the values of the process noise covariance Q and the
A The state transition matrix measurement noise covariance R are constructed through empirical
In Kalman change
analysis [41]. Superior performance of the filter can be achieved by
Q The covariance matrix of the model noise
R The covariance matrix of measurement noise
tuning the estimates of the error covariance matrices. The remaining
M The total number of predicted values notation used in Kalman filter is given in Table 2.
H The observation matrix : The output matrix relating the node's state to the
measurement Yt 4. KF-GPSR: the adapted Kalman filter algorithm applied to the
G The Kalman gain
GPSR protocol
I The identity matrix
t The state iteration cycle
t1 The previous time In this section, we propose an adapted Kalman filter algorithm for the
T The transpose operator GPSR, protocol defined as KF-GPSR. The description of the proposed
stochastic prediction process is divided into three tasks, namely, i)
creating and initializing the neighbors lists and the state vector of nodes,
Table 3 ii) maintaining updates of the lists of neighbors, and iii) observing, cor-
Format of a beacon packet. recting, and updating estimates.
ID node Geo-coordinates of node Velocity of neighbor nodes In GPSR, an individual node has no prior knowledge about the ve-
Id X, Y Vel.x, Vel.y locity of its neighbors. Thus, in order to enhance the proposed prediction,
we need additional fields:

Table 4
● On the beacon packet: we need a vehicle identifier, its x, y co-
Format of neighbors list. ordinates, and Vel.x, Vel.y which indicate the velocity of neighbor
nodes (see Table 3). The speed and direction are sent as a velocity
ID Geo-coordinates of Velocity of Kalman Filter
neighbors neighbors neighbors State vector, computed using the following equations (10) and (11):
X ¼ (x, y, vx,
Id X, Y Vel.x, Vel.y
Vel:x ¼ dx*speed (10)
vy)T

Vel:y ¼ dy*speed (11)


filter will have greater confidence and will attach less importance to the
measurements. This is expressed as follows: where dx and dy are unit vectors that specify the direction of the mo-
bile node.
ð1Þ
Gt ¼ Pt∕t1 H T ðHPt∕t1 H T þ RÞ (7)
● On each node in neighbors list: we need an identifier, the x, y co-
As soon as the change and the Kalman gain are calculated, the
ordinates, velocity, and Kalman state vector. The neighbors list
updated state estimate derived as:
format is shown in Table 4.
Xt∕t ¼ Xt∕t1 þ Gt In (8)

Fig. 6. Illustration of the general case of our proposed model, and the proposed Kalman filter mechanism under the GPSR protocol.

18
Z.S. Houssaini et al. Digital Communications and Networks 4 (2018) 13–26

4.1. Initialization of neighbor state vector cycle starts again.

The first step executed in the KF-GPSR protocol is an exchange of


information about the state of the network. Using periodic beacon sig-
nals, each vehicle informs other vehicles (immediate neighbors) about its
localization and velocity at the current time. Upon receiving a beacon
packet, each vehicle creates a new entry (new neighbor) in its list of
neighbors, enters the information received, and then initializes the state
vector the neighbor using Equation (2). Ultimately, each vehicle must
possess its neighbors list, filled with exchanged information.

4.2. Neighbors mobility monitoring

Before receiving the next periodic beacon packets, the neighbors' in-
formation received previously may have changed and is almost certainly
outdated. This issue is aggravated in a VANET because the location of a
neighbor may change very fast. Then, with the help of the estimation
process used in Kalman filter algorithm (see Equation (4)), we can antic-
ipate these changes recursively by predicting the new state for neighbors
Xtjt1 at each time t. The estimation is based on previous state Xt1,
available at (t  1). Hence, the neighbors lists are kept up-to-date each
time t. Thereafter, we can estimate the error covariance matrix using
Equation (5). Through these predictions, each vehicle can follow and
reconstruct in real time the state of its neighborhood. In other words, at
each iteration cycle t, the forwarder node estimates the current state of its
neighbors using the previous estimated state. Consequently, when it wants
to forward packets, it takes the new available positions into consideration
instead of the previous information. In this way, the issue of outdated
neighbors' information is addressed.

4.3. Correction and update estimations

Note that convergence is guaranteed when the system is observable.


In fact, based on the received observation and the estimation of the
current state, the Kalman filter optimizes the weight of the correction in
order to obtain more accurate estimations. To this end, we use the
following procedure:

 At the same time that the node receives the next measures after period
Δt (next periodic beacon packet), it predicts the current state from the
previously predicted state (at t  1) using Equation (4).
 Since the current predicted state is already calculated, and the new
measurement of the current state is collected from the beacon packet, 4.4. Impact of beacon period on KF-GPSR
it is possible to calculate the innovation using Equation (6).
 Calculate the Kalman gain Gt using Equation (7). In our KF-GPSR protocol, whenever a vehicle needs the location in-
 Finally, using the values of the weighted measurement data and the formation of its neighbors, it will use the estimated state values instead of
Kalman gain, we update the state estimate and the error covariance those from the packet beacon. Our aim is to improve the precision of the
using equations (8) and (9), respectively. localization of neighboring vehicles, while expanding the interval of
beacon dissemination, thus reducing the number of packets that need to
The change values are weighted by the Kalman gain Gt. The latter is exchanged. Using simulation, we evaluate the effectiveness and useful-
used to determine the information that must be added to the measure- ness of our protocol in terms of the trade-off between the accuracy of the
ment information to update the state estimate. That means, it is used to positions and cost, while still considering other QoS metrics.
optimally weight the state vector correction according to the uncertainty Note that the parameter that profoundly affects the performance of
of the current state estimate and how noisy the measurements are. Al- our protocol is the frequency at which packet beacons are sent, also
gorithm 1 describes the two phases “estimation and correction” consid- known as the Beacon Period (BP). Since we mainly want to examine the
ered in the adapted Kalman filter algorithm for monitoring one neighbor. trade-off between accuracy and cost, our evaluation is based on a vari-
As shown in Fig. 6, when t ¼ 0, the first beacon's packet is broadcasted ation of the BP. Additionally, another important parameter is the period
by a neighbor. Here, we initialize the state vector using the measure- β. GPSR uses β to manage the neighbors list. If the packet beacon is not
ments attached to the beacon. In a second step, where t ¼ 1, the estimate received during this period, then the entry in the neighbor list may be
model uses the previous state Xt0 to estimate the current state Xt1. This deleted, assuming that it is out of reach. The value chosen for β is related
approach is repeated recursively at each iteration cycle t until the next to the BP. In this work, we round this to between two and three times the
packet beacon arrives (in case of t ¼ 5). At this moment, the correction BP. In Fig. 7, five key performance metrics are compared between our
model is applied to determine the error between the predicted state and proposed KF-GPSR and two other protocols, namely, the default GPSR
the current state. Therefore, the estimated state is updated (Xt5) and the and our previous GPSR þ Predict protocol. We validate the influence of

19
Z.S. Houssaini et al. Digital Communications and Networks 4 (2018) 13–26

BP changes on the performance metrics and confirm the effectiveness of


KF-GPSR, despite a large decrease in the number of packet beacons. The
simulation parameters are shown in Table 5.
Fig. 7 a, b, and c describe, respectively, the Packet Delivery Ratio
(PDR), Routing Cost (Cost), and Normalized Routing Load (NRL) for
different values of BP. The figures illustrate the trade-off between these
metrics. Under normal circumstances, if we widen the interval of beacon
exchanging, we would expect to see a degradation in PDR due to the
outdated and inaccurate localization information, and a decrease in Cost
and NRL due to the reduction in the number of packet beacons. However,
our protocol contradicts this and guarantees the best results in terms of
PDR, Cost, and NRL in all BP variations. Indeed, it provides more than
75% of the PDR, despite the BP being equal to 9(s). Similarly, for the
throughput and the end-to-end delay metrics illustrated in Fig. 7 d and e
respectively, the KF-GPSR outperforms the other protocols, thus ensuring
the best performance.
In summary, KF-GPSR presents very satisfactory results, and achieves
a compromise between accuracy and routing cost. However, increasing
the BP may cause difficulties, mainly as a result of the loss of neighbors
that are no longer in the transmission range of vehicles, and the infre-
quent updates to find new neighbors. Therefore, to handle this problem,
it is crucial to update the lists of neighbors. In the following section, we
suggest a solution to beacon-on-demand this issue.

5. BOD-KF-GPSR: beacon-on-demand process applied to KF-GPSR

In this section, we address the issue of how to update the neighbors


list of each vehicle a large proportion of neighbors move out of radio
range and the neighbor list becomes empty. Here, if required, we send a
request for beacon packets in contrast to the periodic beacon packets,
which are sent in each BP interval, the “beacon packet request” is only
exchanged if needed by a particular node. Each vehicle estimates its need
to rediscover its neighborhood, and then broadcasts a beacon request
packet. Upon receipt of the request, the neighbors respond by trans-
mitting a simple beacon package. Finally, each vehicle fills its neighbor
list by adding any new neighbors. However, in order to use the “beacon-
on-request” process, each vehicle must be based on two parameters: link
breakage and number of neighbors. Using the Kalman filter prediction
technique, each vehicle is able to determine these two parameters rela-
tively easily.

5.1. Links breakage parameter

A link breakage occurs when the distance between two nodes be-
comes larger than the transmission range. Therefore, communication
between them becomes impossible. Using the position estimation tech-
nique in the Kalman filter algorithm, we can predict a link breakage. At
the same time that a node estimates the positions of its neighbors, it
verifies, based on the new estimated values and on equation (12), which
now falls outside its transmission zone. If there are any, there will
certainly be a break in the link between them. As a result, the node
removes the latter from its list of neighbors. The equation used to
examine whether a node is outside the transmission range of another
node is given as follows:

DistanceðV; ni Þ > R (12)


where the distance is given by the following equation:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
DistanceðV; ni Þ ¼ ðVðxÞ  ni ðxÞÞ2 þ ðVðyÞ  ni ðyÞÞ2 (13)
Fig. 7. Impact of BP on KF-GPSR, GPSR þ Predict and GPSR performance.
In Fig. 8, at t ¼ n, vehicle V1 has n1, n2, n3, n4, and n5 as neighbors.
Using the Kalman filter estimation, V1 predicts that the neighbor n2 will
be outside its radio range at t ¼ n þ 1. Therefore, V1 eliminates n2 from its

20
Z.S. Houssaini et al. Digital Communications and Networks 4 (2018) 13–26

Table 5
Parameters used for simulation.

Parameter Values Unit

Ns-2 Version 2.33 –


PHY-MAC protocol 802.11p –
Transmission Range 532 m
Dimension of Area 700*5000 m2
Type of traffic CBR –
Number of nodes 100 vehicles
Routing Protocols KF-GPSR, GPSR þ Predict, GPSR –
Packet length 512 bytes
Packet rate 1 packets/s
Simulation Time 1000 s
BP 1 to15 s
β 3 to 30 s

neighbor list.

Fig. 9. Illustration of the process of sending a request for a beacon exchange and the
5.2. Number of neighbors parameter reply process.

Each vehicle cleans its list of neighbors recursively excluding those


who are no longer in its neighborhood. Accordingly, the number of neighbors list. In this case, it transmits a beacon request message to these
neighbors decreases progressively. The strategy of our proposed protocol neighbors, inviting them to exchange their information. Each vehicle
is founded on a long interval between packet beacons, in which case, the receiving a BRP from another vehicle must send a reply through a simple
discovery of the neighborhood will be a little late. Hence, we expect that beacon packet.
the vehicle will not find the transmitter choice that will guarantee the Fig. 9 illustrates the process of sending a request for a beacon ex-
best quality of services. Even worse, the vehicle may not have nodes in its change and the reply process. When the vehicle loses half its former
neighbors list. Consequently, once a vehicle loses half its neighbors, it is neighbors n1, n2, and n5, it sends a beacon request within its transmission
imperative that is rediscover its neighborhood and update its list of boundary ranges inviting neighbors to send beacon packets. Nodes n3, n4,
neighbors. Therefore, we propose the “beacon-on-demand” process. n6, and n7 indicate their presence by sending to V1 a simple unicast
Here, each node considers its need to update its neighbor list, it makes a beacon packet. Consequently, V1 updates its neighbors list, adding any
call to discover its neighborhood using a Beacon Request Packet (BRP) new neighbors.
and waits for the reply. If a node loses half its neighbors (nbb), it fills its

Fig. 8. Illustration of a link breakage between vehicles and the update to the list of neighbors.

21
Z.S. Houssaini et al. Digital Communications and Networks 4 (2018) 13–26

Before a node sends a BRP, it must meet the conditions specified in


Algorithm 2:

Fig. 10. Illustration of freeway vehicular mobility scenario used for our simulation.

BOD-KF-GPSR. The simulations are carried out using the IEEE 802.11p2
standard on the physical and MAC layers. We have chosen to apply our
simulation to freeway scenarios. The topology comprises vehicles that
are distributed uniformly on the lanes and that circulate with random
speeds between 90 and 120 km/h. We opted for an area of 700 * 5000 m2
composed of straight roads in both directions and the same number of
lanes in each direction. Fig. 10 shows an example of the freeway model
used in our simulations.
GPSR þ Predict is our previous proposed approach, based on a
neighbor's movement prediction, that improves on GPSR. As explained in
Section 2, GPSR þ Predict only uses a deterministic prediction to monitor
neighbors' positions. We analyze this version of GPSR to show the dif-
ference between the two methods: deterministic and stochastic.
In order to evaluate the performance of the suggested protocols, we
analyze the impact of density on five standardized QoS metrics, namely,
the Packet Delivery Ratio (PDR), End-to End delay (E2E), throughput,
Normalized Routing Load (NRL), and routing cost (Cost). In addition, to
study the impact of density, we vary the number of vehicles from 50 (low
density) to 350 (high density) and keep the CBR data traffic at 15 con-
When a vehicle loses half its neighbors, it demands that its neigh-
nections. We consider two sizes of CBR data packets: medium and vol-
borhood send a beacon packet, and waits for feedback. Upon receiving,
uminous data. The general parameters considered in this simulation are
the BRP will proceed as follows:
shown in Table 6.
1. The neighbor verifies the time of the request; if it is very close to the
time of the periodic beacon packet, then the request is rejected; if not, 6.2. Experimental results considering QoS
it continues the process.
2. The neighbor responds to the requesting node through a unicast The main purpose of all simulations is to analyze the impact of vehicle
beacon packet reply. density on the performance of our proposed protocols. We analyze the
3. The neighbor checks whether this node exists in its neighbor list; if so, performance of BOD-KF-GPSR and KF-GPSR, and then compare them
it updates the node's information; if not, it creates a new entry and with our previous model GPSR þ Predict, as well as with common pro-
adds the attached information. tocols such as AODV, DSR, GPSR, and ZRP.
4. The neighbor increments by 1, the number of its own neighbors.
6.2.1. Packet delivery ratio measurement
6. Performance evaluation In this section, we study the measure of the packet delivery rate
defined as the ratio between the amount of received data and the amount
6.1. Simulation setting of data sent (see equation PDR). This metric shows the effectiveness of
the protocol in terms of successfully received packet rates.
To evaluate our proposed protocols, we use NS-2 as the network The result of the PDR for the difference densities is practically the
simulator [42] because it offers an environment with high fidelity (i.e. to same, regardless of the size of the data packets routed.
represent the propagation phenomenon and the physical/network First, we discuss the results of the well-known routing protocols:
layers). For the mobility model, VanetMobiSim [43] is used to generate AODV, DSR, GPSR, and ZRP. As shown in Fig. 11a and b, the best per-
realistic mobility models and realistic traffic environments. VanetMobi- centage is that of the GPSR protocol, because it guarantees a relatively
Sim is considered the most efficient mobility modeling framework. It good PDR in both cases of data size. GPSR is followed by the AODV
provides several models, including the Intelligent Driver Model (IDM), protocol, which offers an acceptable PDR, while those of the ZRP and
Intelligent Driver Model with Lane Changes (IDM-LC), Intelligent Driver DSR protocols are disastrous and nearly equal to zero, especially in the
Model with Intersection Management (IDM-IM), and the Fluid Traffic case of medium to high density. This shows that these two protocols do
Model (FTM) [44,45]. We focus on the IDM-LC owing to the presence of
intersections, traffic lights, vehicle speeds, and lane changing. The
2
mobility traces generated by VanetMobiSim are used directly in NS-2. The IEEE 802.11p or WAVE (Wireless Access in Vehicular Environments), also known
as Dedicated Short-Range Communication (DSRC), intended for VANETs. It is a standard
The simulation considers various commonly used protocols: AODV, that uses short-range to medium-range communication systems (typically between 250
DSR, GPSR, and ZRP. These are compared with our previous and 300 m). In 1999, the United States Federal Communications Commission (FCC)
GPSR þ Predict protocol and the two new approaches, KF-GPSR and allocated 75 MHz of its spectrum at the 5.9-GHz band [46].

22
Z.S. Houssaini et al. Digital Communications and Networks 4 (2018) 13–26

Table 6 stochastic model compared to those of the simple deterministic model.


General simulation parameters. Finally, in terms of the two proposed models (i.e, the KF-GPSR pro-
Parameter Values Unit tocol, and its extension BOD-KF-GPSR), it can be seen that the latter
Ns-2 Version 2.33 – exhibits better performance. It outperforms the first proposal and all
PHY-MAC protocol 802.11p(802.11Ext) – other tested routing protocols as well. This is because BOD-KF-GPSR
Transmission range 532 m contributes greater accuracy and reliability, and eliminates the risk of
Antenna type Omni-directional – not having sufficient neighbors or of an empty neighbors list. This
Propagation model Two Ray Ground –
Dimension of area 700*5000 m2
advantage is the result of the implementation of “beacon-on-de-
Number of nodes 50, 100, 150, 200, 250, 300, and 350 vehicles mand” technique.
Routing protocols BOD-KF-GPSR, KF-GPSR, –
GPSR þ Predict GPSR,AODV, DSR, ZRP 6.2.2. End-to-end delay measurement
BP/β of GPSR and 3/10 s
As already mentioned, the geographical routing protocol uses the
GPSR þ Predict
BP/β of BOD-KF-GPSR 10/10 s current geographic location of nodes to determine the optimal route as
and KF-GPSR the packets are forwarded. Since the delay represents the average time
Type of traffic CBR – that a data packet takes from source to destination, GPSR and all its
Packet length 512, 2048 bytes/packet proposed improvements belonging to this category of protocols are
Packet rate 1 packets/s
Number of data 15 source,
distinguished by the lowest transmission delay. These delays are shown
traffics destination in Fig. 12a and b. However, the topology-based routing protocol uses link
Model mobility IDM-LC – information of routing tables, which offers more options to achieve
Number of lanes/ 3*2 lanes*direction packet forwarding. Consequently, whatever the packet's size, the pro-
direction
tocols based on the AODV, DSR, and ZRP topologies suffer from
Maximum traffic 0 traffic lights
lights extremely high delays.
Minimum vehicle 90 km/h With regard to which of the protocols GPSR þ Predict, KF-GPSR, and
movement speed BOD-KF-GPSR have the shortest delays, the reduction of the end-to-end
Maximum vehicle 120 km/h delay is quite impressive, giving very satisfactory performance. In
movement speed
Traffic duration 300 s
contrast, BOD-KF-GPSR obtains the lowest transmission packet delay,
Number of 20 – followed by KF-GPSR and GPSR þ Predict. This proportional decrease in
simulations the delay can be justified by the fact that the geographic location preci-
sion attained is very high, and the neighbor mobility information is well
monitored. Effectively, BOD-KF-GPSR purges more outdated position
not manage the high mobility of nodes well. This was expected because
information entries for neighbors, thus offering the optimal time.
these common protocols were, initially developed for a MANET, rather
than a VANET.
With regard to the results of the deterministic and stochastic methods,
we can clearly see that KF-GPSR works better than GPSR þ Predict,
which proves the robustness of the prediction technique using the Kal-
man filter algorithm. This confirms the efficiency and usefulness of the

Fig. 12. The impact of node density on the behavior of the protocols BOD-KF-GPSR, KF-
Fig. 11. The impact of node density on the behavior of the protocols BOD-KF-GPSR, KF- GPSR, GPSR þ Predict, GPSR, AODV, DSR, and ZRP in terms of their end-to-end delays are
GPSR, GPSR þ Predict, GPSR, AODV, DSR, and ZRP in terms of their packet delivery ratios presented in (a) a medium data scenario (512 bytes/packet), and (b) a large data scenario
are presented in (a) a medium data scenario (512 bytes/packet), and (b) a large data (2048 bytes/packet).
scenario (2048 bytes/packet).

23
Z.S. Houssaini et al. Digital Communications and Networks 4 (2018) 13–26

6.2.3. Throughput measurement


Throughput can be defined as the flux or the quantity of data traveling
over a communication channel per unit time, usually expressed in kilo-
bits per second (kbps) [47]. Its levels rely on three conditions: the size of
the data packets, the number of traffic sources, and the effectiveness of
the protocol in terms of PDR.
The results shown in Fig. 13a and b affirm this hypothesis in terms of
the size and PDR. Effectively, the throughput attains 250 kbps when the
size of a packet is large, while in the case of a medium packet size, the
throughput is just 65 kbps. Additionally, the throughput obtained for
each protocol is proportional to the PDR (11a and 11b): the protocols
with high PDR have higher throughput. In contrast, protocols with low
PDR have lower throughput.
With regard to the proposed protocols, BOD-KF-GPSR again exhibits
better performance in both cases. In fact, it increases the throughput
of the original GPSR by up to around 50 kbps, which is a good
enhancement.

6.2.4. Routing cost measurement


An interesting observation on the performance of our protocol comes
from the analysis of the results plotted in Fig. 14a and b. These figures
show the average route-cost for each routing protocol evaluated. When
the number of vehicles increases, the routing costs of all protocols in-
creases as well. This is because the increase in the number of vehicles
directly increases the number of hops, as well as the beacon packets
number, thus increasing congestion in the network.
In both cases of data packet size, the common protocols AODV, DSR,
and ZRP perform worst and must be avoided for VANETs, especially in Fig. 14. The impact of node density on the behavior of the protocols BOD-KF-GPSR, KF-
freeway settings. This may be due to the recurrent topology changes that GPSR, GPSR þ Predict, GPSR, AODV, DSR, and ZRP in terms of their routing costs are
presented in (a) a medium data scenario (512 bytes/packet), and (b) a large data scenario
result in link breakages and the frequent need to discover the route.
(2048 bytes/packet).
However, KF-GPSR and basic GPSR have almost the same route-cost,
which can be explained by the fact that they consider the same BP.
Therefore, practically the same number of beacon packets are exchanged.
Furthermore, the proposed enhancements show a considerable reduction compared to the initial GPSR. Despite that, the BP used in the GPSR is
smaller than the BP used in the KF-GPSR and BOD-KF-GPSR protocols. In
contrast, KF-GPSR manages to incur a slightly lower cost than BOD-KF-
GPSR does, which is expected because it does not use the “beacon-on-
demand” mechanism.

6.2.5. Normalized routing load measurement


The NRL of a protocol is defined as the number of packets exchanged
per data packet received. It depends on three processes, namely, the high
mobility, which implicates position accuracy and, therefore, much more
re-forwarding [48], the density of the vehicle, and the frequency of
beacon packet dissemination. The bigger the density or the frequency,
the bigger is the overload.
From the simulation results illustrated in Fig. 15a and b, the NRL is
not affected by the modification to the data packet size. In contrast, it
increases linearly with the density. However, by comparing the NRL of all
tested protocols, it is obvious how many extra beacon packets are
exchanged by using common BP, which illustrates why we use the
“beacon-on-demand” strategy and a large BP. Hence, BOD-KF-GPSR has
much lower NRL than other protocols.
For the ZRP protocol, its hybrid nature is the main cause of the drastic
increase in its NRL. In fact, the ZRP behavior is unstable vis-a-vis high
mobility, and consequently, uncontrolled flooding and re-
forwarding occurred.

6.2.6. Discussion
Note that our proposal focuses mainly on VANETs, where we try to
increase the routing mechanism in order to use the various applications
with a high QoS.
Fig. 13. The impact of node density on the behavior of the protocols BOD-KF-GPSR, KF-
GPSR, GPSR þ Predict, GPSR, AODV, DSR, and ZRP in terms of their throughput are
Using a Kalman filter prediction mechanism with GPSR (KF-GPSR)
presented in (a) a medium data scenario (512 bytes/packet), and (b) large data scenario has shown gains in terms of QoS. Indeed, it increases all evaluated
(2048 bytes/packet). metrics even though we considered a large interval between consecutive

24
Z.S. Houssaini et al. Digital Communications and Networks 4 (2018) 13–26

However, whenever a vehicle requires information about the positions of


its neighbors, it uses the new estimated position rather than the old po-
sition. For better reliability and effectiveness, we propose an extension of
KF-GPSR, called BOD-KF-GPSR, which uses a “beacon-on-demand” tech-
nique that allows a vehicle to send neighbors queries only when it needs
to discover its neighborhood.
We realized our experimental simulations using Network Simulator
NS2 and VanetMobiSim. We quantitatively compared our two proposed
approaches BOD-KF-GPSR and KF-GPSR with our previous proposal
GPSR þ Predict, as well as with the classical GPSR, AODV, DSR, and ZRP
protocols. Through extensive simulations, we have shown that using a
Kalman filter prediction model with GPSR (KF-GPSR) improves the
routing performance, whatever is the beacon period. In addition, we have
shown that its extension BOD-KF-GPSR, which is based mainly on the
“beacon-on-demand” technique achieves much better performance.
Thus, we have accomplished our objective to achieve the compromise
between accuracy, cost, and the other QoS(i.e., PDR, end-to-end delay,
throughput, and NRL).

References

[1] Y. Wang, F. Li, Vehicular ad hoc networks. Guide to Wireless Ad Hoc Networks,
Springer, 2009, pp. 503–525.
[2] K.N. Qureshi, A.H. Abdullah, Localization-based system challenges in vehicular ad
hoc networks: Survey, Smart CR 4 (6) (2014) 515–528.
[3] A. Singh, M. Kumar, R. Rishi, D. Madan, A Relative Study of Manet and Vanet: its
Applications, Broadcasting Approaches and Challenging Issues, Advances in
Networks and Communications, 2011, pp. 627–632.
Fig. 15. The impact of node density on the behavior of the protocols BOD-KF-GPSR, KF- [4] W. Liang, Z. Li, H. Zhang, S. Wang, R. Bie, Vehicular ad hoc networks: architectures,
GPSR, GPSR þ Predict, GPSR, AODV, DSR, and ZRP in terms of the normalized routing research issues, methodologies, challenges, and trends, Int. J. Distributed Sens.
load are presented in (a) a medium data scenario (512 bytes/packet), and (b) a large data Netw. 11 (8) (2015), 745303.
scenario (2048 bytes/packet). [5] S. Singh, S. Agrawal, Vanet routing protocols: issues and challenges, in: Engineering
and Computational Sciences (RAECS), 2014 Recent Advances in, IEEE, 2014,
pp. 1–5.
[6] H. Maowad, E. Shaaban, Efficient routing protocol for vehicular ad hoc networks,
beacons. However, the extension BOD-KF-GPSR, which uses the “beacon- in: Networking, Sensing and Control (ICNSC), 2012 9th IEEE International
Conference on, IEEE, 2012, pp. 209–215.
on-demand” process, is more advantageous. It increases the performance [7] N.H. Saeed, M.F. Abbod, H.S. Al-Raweshidy, Manet routing protocols taxonomy, in:
significantly. Effectively, this solution handles the trade-off between cost, Future Communication Networks (ICFCN), 2012 International Conference on, IEEE,
accuracy, and QoS. However, the only disadvantage that we find in our 2012, pp. 123–128.
[8] Q. Wu Wu, W. Wen, Q. Liu, Comparative study of vanet routing protocols, in:
proposals is that they use a complex prediction model (Kalman filter Cyberspace Technology (CCT 2014), International Conference on, IEEE, 2014,
algorithm), requiring higher processing power and memory. However, pp. 1–5.
this is not a serious problem since in VANET (distinct to WSN or MANET), [9] G. Kavitha, A study on transmission strategies in vanet, Int. J. Innovative Res.
Comput. Commun. Eng. 3 (5) (2015).
vehicles are powered by a self-chargeable battery and contain memory.
[10] S. Allal, S. Boudjit, Geocast routing protocols for vanets: survey and guidelines, in:
Therefore, there are no battery constraint or memory. Innovative Mobile and Internet Services in Ubiquitous Computing (IMIS), 2012
With regard to the well-known AODV, DSR, GPSR, and ZRP routing Sixth International Conference on, 2012, pp. 323–328.
protocols, the performance measures obtained are mediocre. These re- [11] L. Hanzo, R. Tafazolli, A survey of qos routing solutions for mobile ad hoc networks,
IEEE Commun. Surv. Tutorials 9 (2) (2007) 50–70.
sults are attributed to the fact that high mobility surpasses these pro- [12] M. Altayeb, I. Mahgoub, A survey of vehicular ad hoc networks routing protocols,
tocols capacities. In other words, they have serious problems in managing Int. J. Innovation Appl. Stud. 3 (3) (2013) 829–846.
information about neighbor's mobility in a VANET. Accordingly, they [13] K. Bayad, E.H. Bourhim, M. Rziza, M. Oumsis, Comparative study of topology-based
routing protocols in vehicular ad hoc network using ieee802. 11p, in: Electrical and
must be avoided for such an environment. Information Technologies (ICEIT), 2016 International Conference on, IEEE, 2016,
pp. 526–530.
7. Conclusion [14] J. Liu, J. Wan, Q. Wang, P. Deng, K. Zhou, Y. Qiao, A survey on position-based
routing for vehicular ad hoc networks, Telecommun. Syst. 62 (1) (2016) 15–30.
[15] A. Boukerche, H.A. Oliveira, E.F. Nakamura, A.A. Loureiro, Vehicular ad hoc
The precise localization of vehicles, from the perspective of a neigh- networks: a new challenge for localization-based systems, Comput. Commun. 31
boring node, is very important for the usefulness of geographic routing (12) (2008) 2838–2849.
[16] B. Karp, H.-T. Kung, Gpsr: greedy perimeter stateless routing for wireless networks,
protocols. However, owing to the dynamic nature of vehicles in a VANET, in: Proceedings of the 6th Annual International Conference on Mobile Computing
this information obtained through periodic beacon packets changes and Networking, ACM, 2000, pp. 243–254.
rapidly and becomes outdated. This can influence the routing negatively [17] C.E. Perkins, E.M. Royer, Ad-hoc on-demand distance vector routing, in: Mobile
Computing Systems and Applications, 1999. Proceedings. WMCSA ’99. Second IEEE
in terms of QoS. Thus, the issue is how to ensure accurate neighbors
Workshop on, IEEE, 1999, pp. 90–100, https://doi.org/10.1109/
locations without causing an overhead on the network and an increase of MCSA.1999.749281.
route costs. In order to handle the trade-off between accuracy, routing [18] D.B. Johnson, D.A. Maltz, Dynamic source routing in ad hoc wireless networks, in:
cost, and QoS, a reliable prediction of vehicles' next moves can be Mobile Computing, Kluwer Academic Publishers, 1996, pp. 153–181.
[19] Z. J. Haas, M. R. Pearlman, P. Samar, The Zone Routing Protocol (Zrp) for Ad Hoc
significantly beneficial since it is possible to monitor the mobility of Networks.
neighbors and at the same time, to consider a large interval between [20] T. Hu, M. Liwang, L. Huang, Y. Tang, An enhanced gpsr routing protocol based on
consecutive beacon packets. the buffer length of nodes for the congestion problem in vanets, in: Computer
Science and Education (ICCSE), 2015 10th International Conference on, IEEE, 2015,
In this paper, we suggest an enhancement to the well-known geographic pp. 416–419.
routing protocol GPSR, KF-GPSR, where each vehicle is able to estimate [21] C. Tripp-Barba, L. Urquiza-Aguiar, M.A. Igartua, D. Rebollo-Monedero, L.J. de la
the current neighbor's position by exploiting the robust Kalman filter al- Cruz Llopis, A.M. Mezher, J.A. Aguilar-Calder on, A multimetric, map-aware routing
protocol for vanets in urban areas, Sensors 14 (2) (2014) 2199–2224.
gorithm. Certainly, using this estimation model, the accuracy is improved.

25
Z.S. Houssaini et al. Digital Communications and Networks 4 (2018) 13–26

[22] C. Bouras, V. Kapoulas, E. Tsanai, A gpsr enhancement mechanism for routing in Computational Systems (RAICS), 2015, pp. 268–273, https://doi.org/10.1109/
vanets, in: International Conference on Wired/Wireless Internet Communication, RAICS.2015.7488426.
Springer, 2015, pp. 94–107. [34] H. Feng, C. Liu, Y. Shu, O.W. Yang, Location prediction of vehicles in vanets using a
[23] H. Menouar, M. Lenardi, F. Filali, Movement prediction-based routing (mopr) kalman filter, Wirel. Personal. Commun. 80 (2) (2015) 543–559.
concept for position-based routing in vehicular networks, in: 2007 IEEE 66th [35] H. Ghafoor, I. Koo, Spectrum-aware geographic routing in cognitive vehicular ad
Vehicular Technology Conference, 2007, pp. 2101–2105, https://doi.org/10.1109/ hoc network using a kalman filter, J. Sensors 2016 (2016) 10. Article ID: 8572601.
VETECF.2007.441. [36] E. Amar, S. Boumerdassi, A scalable mobility-adaptive location service with
[24] R.A. Saqour, M. Shanudin, M. Ismail, Prediction schemes to enhance the routing kalman-based prediction, in: Wireless Communications and Networking Conference
process in geographical GPSR ad hoc protocol, Mob. Inf. Syst. 3 (3–4) (2007) (WCNC), 2011 IEEE, IEEE, 2011, pp. 593–598.
203–220, https://doi.org/10.1155/2007/364101. [37] Z.Z.-H. Wang Guang-Yu, Liu Chun-Feng, S. Yan-Tai, Hybrid routing algorithm in
[25] M. Heissenbüttel, T. Braun, M. W€alchli, T. Bernoulli, Evaluating the limitations of vehicular ad hoc network based on kalman prediction, Comput. Eng. 40 (8) (2014)
and alternatives in beaconing, Ad Hoc Netw. 5 (5) (2007) 558–578. 91–95.
[26] R. Alsaqour, M. Abdelhaq, R. Saeed, M. Uddin, O. Alsukour, M. Al-Hubaishi, [38] P.S. Maybeck, Stochastic Models, Estimation, and Control. Vol. 2, Vol. 3, Academic
T. Alahdal, Dynamic packet beaconing for gpsr mobile ad hoc position-based press, New York, N.Y., London, Paris, 1982.
routing protocol using fuzzy logic, J. Netw. Comput. Appl. 47 (2015) 32–46. [39] G. Welch, G. Bishop, An Introduction to the Kalman Filter.
[27] A. Boukerche, C. Rezende, R.W. Pazzi, Improving neighbor localization in vehicular [40] W. Greg, G. Bishop, An Introduction to the Kalman Filter. Updated: 5.4, Department
ad hoc networks to avoid overhead from periodic messages, in: Global of Computer Science, University of North Carolina at Chapel Hill, Chapel Hill, NC,
Telecommunications Conference, 2009. GLOBECOM 2009. IEEE, IEEE, 2009, 2004.
pp. 1–6. [41] J. Yim, J. Joo, C. Park, A kalman filter updating method for the indoor moving
[28] Z.S. Houssaini, I. Zaimi, M. Oumsis, S.E.A. Ouatik, Improvement of gpsr protocol by object database, Expert Syst. Appl. 38 (12) (2011) 15075–15083.
using future position estimation of participating nodes in vehicular ad-hoc [42] NS-2, the Network Simulator Ns-2, 2010. URL, http://www.isi.edu/nsnam/ns/.
networks, in: 2016 International Conference on Wireless Networks and Mobile [43] VanetMobiSim, Canu Mobility Simulation Environment, February 2007. URL,
Communications (WINCOM), 2016, pp. 87–94, https://doi.org/10.1109/ http://vanet.eurecom.fr.
WINCOM.2016.7777196. [44] J. H€arri, F. Filali, C. Bonnet, M. Fiore, Vanetmobisim: generating realistic mobility
[29] Z.S. Houssaini, I. Zaimi, M. Oumsis, S. El, A. Ouatik, GPSR þ predict : an patterns for vanets, in: Proceedings of the 3rd International Workshop on Vehicular
enhancement for GPSR to make smart routing decision by anticipating movement of Ad Hoc Networks, ACM, 2006, pp. 96–97.
vehicles in VANETs, Adv. Sci. Technol. Eng. Syst. J. 2 (3) (2017) 137–146. [45] J. Harri, M. Fiore, Vanetmobisim–vehicular Ad Hoc Network Mobility Extension to
[30] Jerome Haerri, Christian Bonnet, Fethi Filali, The Challenges of Predicting Mobility, the Canumobisim Framework, Institut Eurecom Department of Mobile Commu
08, 2006. EURECOMþ2240, http://www.eurecom.fr/publication/2240. 6904, 2006, pp. 1–19.
[31] A. Bhargava, S. Verma, B.K. Chaurasia, Kalman filter for trust estimation in vanets, [46] Y.L. Morgan, Notes on dsrc & wave standards suite: its architecture, design, and
in: Electrical Computer and Electronics (UPCON), 2015 IEEE up Section Conference characteristics, IEEE Commun. Surv. Tutorials 12 (4) (2010) 504–518.
on, IEEE, 2015, pp. 1–6. [47] S. Zafar, H. Tariq, K. Manzoor, Throughput and delay analysis of aodv, dsdv and dsr
[32] M. Armaghan, M. Fathy, S. Yousefi, Improving the performance of beacon safety routing protocols in mobile ad hoc networks, Int. J. Comput. Netw. Appl. (IJCNA) 3
message dissemination in vehicular networks using kalman filter estimation, (2) (2016) 1–7.
Commun. Netw. (2009) 74–82. [48] Q. Chen, S.S. Kanhere, M. Hassan, Adaptive position update for geographic routing
[33] S. Shelly, A.V. Babu, Prediction of link residual lifetime using kalman filter in in mobile ad hoc networks, IEEE Trans. Mob. Comput. 12 (3) (2013) 489–501.
vehicular ad hoc networks, in: 2015 IEEE Recent Advances in Intelligent

26

You might also like