Enhanced Spring Clustering in VANETs with

Obstruction Considerations
Leandros A. Maglaras, Student Member, IEEE, Dimitrios Katsaros
Department of Computer & Communication Engineering, University of Thessaly, Volos, Greece

Abstract—Vehicular networks have a diverse range of applica- position and their relative mobility. The clusterhead change of
tions that varies from safety applications to comfort applications. Spring Clustering is relatively low and the overall performance
Clustering in VANETs is of crucial importance for addressing the of the method is stable to different topologies and transmission
scalability problems of VANETs. The performance of communi-
cation protocols is greatly influenced by the existence of vehicles ranges.
in the neighborhood; vehicles acting as obstacles change the Though, that protocol suffered from the fact that it treated
behavior of protocols when different density, speed and car sizes vehicles only as senders, relays and receivers and not as part of
scenarios are investigated since reliable communication range the environmental obstacles. A number of V2V measurements
among vehicles varies. The Enhanced Spring Clustering is a new have been performed to study the statistical properties of V2V
distributed clustering protocol, which forms stable clusters based
on vehicle dimensions. An investigation of the performance of the propagation channels [8], [9]. In [10] it is observed that in
Enhanced Spring Clustering in realistic environments is presented rush hours the received signal strength gets worse compared
confirming its superiority over the examined, competing cluster- to no traffic hours for the same part of an open road. These
ing protocol. observed differences can only be related to other vehicles
obstructing Line-of-Sight (LOS), since the system parameters
I. I NTRODUCTION remained the same during the measurements. Recent work
To enhance the safety of drivers, to provide a comfort- reported in [11], [12] showed that vehicles as obstacles have
able driving environment and to contribute to fuel economy, an important influence on the behavior of a VANET system.
messages for different purposes need to be sent to vehicles Highly realistic channel models [13] gives results that
through intervehicle communications. An important issue for are in very good agreement with the real world. However,
a VANET system is the broadcast storm problem that is present these models are computationally too expensive making them
in high density environments. Of the solutions proposed for impractical for extensive simulation studies. In [11] it is shown
scaling down networks with large numbers of nodes, network that the vehicles as obstacle have a significant impact on LOS
clustering is among the most investigated for mobile ad hoc obstruction in both the dense and sparse vehicular networks,
networks [1], [2], for sensor ad hoc networks [3], [4], and for therefore, shadow fading effects due to other vehicles are very
vehicular ad hoc networks [5], [6], [7]. important to be included in channel models.
The basic idea is that of grouping network nodes that are in
physical proximity. The subsequent backbone uses the induced
hierarchy to form a communication infrastructure that is func-
tional in providing desirable properties such as minimizing
communication overhead, choosing data aggregation points,
increasing the probability of aggregating redundant data, and
so on.
In Figure 1 an accident happens in a highway at a point
in time where traffic is intense, and we also suppose that
the vehicles approaching the place of accident are able to
“detect” the accident. The accident results in the highway Fig. 1. Illustration of an example where vehicle clustering is important.
being blocked. In such a situation density of vehicles increases
dramatically and a clustering method is necessary for the
proper dissemination of safety messages. A. Contributions
Spring Clustering which was introduced [5] forms stable The present work presents a new clustering protocol for
clusters based on force directed algorithms. The proposed VANETs, namely the Enhanced Spring Clustering protocol
method uses a mobility metric based on “forces” applied with incorporates vehicles Several scenarios are investigated
between nodes according to their current and their future in a highway environment where traces are created by out mo-
bility model. Mobility traces can also be loaded by SUMO [14]
Research supported by the project “REDUCTION: Reducing Environmental in order to investigate more complex scenarios. The reliable
Footprint based on Multi-Modal Fleet management System for Eco-Routing
and Driver Behaviour Adaptation”, funded by the EU.ICT program, Challenge communication range that is produced by simple diffraction
ICT-2011.7. models, depends on cars density, velocity and car’s dimensions

978-1-4673-6337-2/13/$31.00 ©2013 IEEE

changing the overall performance of communication protocols. them. The correct choice of the clusterhead is very important
The article makes the following contributions: for the stability of the method, the cluster lifetime and the
• Proposes the Enhanced Spring Clustering protocol by overhead involved in forming and maintaining these clusters.
incorporating vehicle heights in clusterhead election. In Sp-Cl parameter qi [5] is used to favour vehicles (according
• Evaluates the performance of Enhanced Spring Clustering to their paths, driver habits, etc.) to become clusterheads,
under different network characteristics (density, velocity, strengthening positive forces applied to these nodes and weak-
car dimensions, placement of antenna). ening negative ones.
• Incorporates vehicles as obstacles in a VANET simulator.
• Calculates reliable communication range for every pair of
nodes at every instance based on the diffraction caused
by obstructing vehicles.

II. S PRING C LUSTERING

Spring Clustering is based on force-directed algorithms. The
force-directed assign forces among the set of edges and the
set of nodes in a network. The most straightforward method Fig. 3. Reliable communication range is significantly larger for tall vehicles
is to assign forces as if the edges were springs and the nodes
were electrically charged particles. The entire graph is then In Figure 3 when a tall vehicle is the transmitter the
simulated as if it were a physical system. The forces are reliable communication range (RCR) is almost the same to
applied to the nodes, pulling them closer together or pushing the ideal communication range (CR) when vehicle obstruction
them further apart. is neglected (a). When a car being a transmitter is obstructed
Vehicles periodically broadcast beacon messages. The bea- by a tall vehicle the reliable communication range is strongly
con message consists of node Identifier (ID), node location, affected in the area around the obstacle (b). Taking this
speed vector, total force F , state and timestamp. Node location characteristics in mind we incorporate height as a criterion
is used in order to calculate the distance between the nodes. for the clusterhead election. When a node i finds itself to be
Each node i using the information of the beacon messages among the tallest in its one-hop neighborhood then parameter
calculates the pairwise relative force Frelij for every neighbor qi is used to favor it become a clusterhead. In order for this
j using the Coulomb law. new criterion to be incorporated to our method an extra byte
is added to the beacon message. The beacon message now
except from node Identifier (ID), node location, speed vector,
total force F , state and timestamp, also includes vehicle height
m or, for simplicity, a vehicle category in terms of tall or short.
k
Frelk Frelm The new method is called Enhanced Spring Clustering (ESC).
i
III. N ETWORK MODEL
Freln A. VANET diffraction models
n Several propagation models applied in VANET research
can be used to quantify the impact of vehicles as obstacles
Fig. 2. Relative forces applied to vehicle i. on the electromagnetic wave propagation. Geometry based
deterministic models are used to analyze particular situations.
Vehicles that move to the same direction or towards each A highly realistic model, based on optical ray tracing was
other apply positive forces while vehicles moving away apply proposed in [13]. The accuracy of the model is achieved at
negative forces. The node with the highest positive force, in the expense of high computational complexity and location-
its one hop neighborhood, is elected clusterhead. specific modelling. There are simplified geometry based de-
terministic models [11], [15]. In particular, the research work
A. Special role of vehicles – Enhanced Spring Clustering proposed by Boban et al. in [11] derive a simplified geometry-
In the Spring Clustering method, vehicles that follow pre- based deterministic propagation model, in which the effect of
defined routes or keep a relative constant velocity are favored vehicles as obstacles on signal/wave propagation is isolated
to become clusterheads due to being more stable in terms and quantified while the effect of other static obstacles (i.e.,
of mobility. Except from being best candidates based on buildings, overpasses, etc.) is not considered. The research
mobility criteria, tall vehicles like trucks and busses suffer work in [11] focuses on vehicles as obstacles by systematically
less from vehicle obstruction. The maximum distance over quantifying their impact on line of sight and consequently on
which communication is still possible is significantly larger the received signal power.
than when neither the transmitter nor the receiver is a tall For the received power level, the impact of obstacles can be
vehicle. Selecting tall vehicles as clusterheads increases the represented by signal attenuation. This increase in attenuation
probability that members stay for a longer time connected to is due to the diffraction of the electromagnetic waves. To
model vehicles obstructing the line of sight, we use the knife- their antennas. If the line intersects with one rectangle then
edge attenuation model. the single knife model is used to calculate the additional loss.
If only one obstacle is located between T x and Rx, then In case when the line crosses more than one rectangle then
the single knife-edge model described in ITU-R recommen- the Epstein-Peterson model is used.
dation [16] is used. If the direct line-of-sight is obstructed In our simulator the vehicles are produced by our generator.
by a single knife-edge type of obstacle, we use the following In case where the vehicles are coming from a mobility
diffraction parameter v: simulator like SUMO the same procedure is used. When the
 cars are blocked by a building then we have a NLOS situation.
2 1 1
ν =h∗ ∗( + ) (1) In this case it is impossible to make a straight line between the
λ d1 d2 two vehicle positions without being obstructed by a building
where h is the height of the top of the obstacle above the and the transmission range is usually limited to a few meters.
straight line joining the two ends of the path. If the height is In the scenarios investigated in this research we are focused to
below this line, h is negative. d1 and d2 are distances of the find the impact of vehicles in reliable transmission range and
two ends of the path from the top of the obstacle. for that reason when we have NLOS situation we consider the
The diffraction loss can be closely approximated by range to be zero.
⎧ When the line connecting two vehicles is obstructed by
⎨ 6.02 + 9.00ν + 1.65ν 2 if − 0.8 ≤ ν < 0
Ad = 6.02 + 9.11ν − 1.27ν 2 if 0 ≤ ν < 2.4 another car then we say that we have a OLOS situation
⎩ (Obstructed line of sight). According to the number of ob-
12.953 + 20 log ν if ν ≥ 2.4
structing vehicles the corresponding diffraction method is
When two or more vehicles exist between the transmitter used to calculate the power loss and determine the reliable
and receiver horizons then diffraction loss is calculated using communication range among the nodes. OLOS situations
the Epstein-Peterson method [17]. The diffraction loss is taken are very important since vehicles, due to diffraction, have
as the sum of the individual knife edges according to Figure 4. different communication ranges with their neighbors according
to their environment. The implementation of the simulator that
incorporates the vehicles as obstacles is presented in simple
steps in the next section.
IV. VANET IMPLEMENTATION
• Every vehicle and building is modelled as a rectangle
with proportional dimensions.
• For every instance and pair of nodes a straight line is
Fig. 4. Epstein-Peterson method.
drawn from antenna position of each TX vehicle to the
antenna position of each RX Vehicle.
• If the line does not touch any other rectangle, TX/RX has
B. Placement of antennas LOS.
The antennas in vehicles are usually mounted either on • If the line passes through one rectangle at least , LOS is
the roof or inside the vehicle (e.g., under the windshield, in obstructed by a vehicle or by a building, the two cases
rear view mirror, on a seat or near a dashboard). The effect can easily be distinguished by using the geographical
of the antenna placement in vehicles is significant [12]. The information available in simulator or by the sizes of the
differences in the cumulative link packet error rates are until obstacles.
25 − 30 % depending on the antenna locations under LOS • Once the propagation condition is identified, the simulator
conditions. This is due to the fact that the height of the antenna can simply use the relevant model to calculate the power
affects the attenuation caused by obstructing vehicles. In order loss.
to make the environment more realistic we incorporated the • According to power loss the reliable communication
placement of the antenna in our simulation environment. The range (RCR) between each pair of nodes for every time
model used has cars that either have the antenna mounted on instance is calculated.
the roof at a height of 10cm above the car or under windshield • Enhanced Spring Clustering performance is evaluated
at 50cm below the height of the car. according to reliable communication ranges.
• Enhanced Spring Clustering is evaluated according to
C. Obstructed Line Of Sight proper clusterhead election.
To separate the Line Of Sight from OLOS (Obstructed Line
Of Sight) and NLOS (No Line Of Sight) cases at every time V. S IMULATION AND PERFORMANCE EVALUATION
instance we use a simple algorithm to detect intersections A simulation study was conducted to evaluate the perfor-
between line segments. Every node in the simulator is rep- mance of our protocol using a custom simulator. The competi-
resented as a rectangle occupying an area of the simulation tor is the Spring Clustering method [5], since it is shown there
space. For every pair of nodes we draw a line connecting that it is superior to many current high-performance VANET
 Lane Speed km/h
clustering protocols. In our simulation, we consider various
1 80
road traffic and network data parameters. The simulation 2 100
environment (Figure 5) is a two direction, 3-lane per direction 3 120
highway with a turn in order to evaluate the performance
of the scheme. All nodes are equipped with GPS receivers TABLE II
S PEED PER LANE FOR BOTH DIRECTIONS .
and On Board Units (OBU). Location information of all
vehicles/nodes, needed for the clustering algorithm is collected
with the help of GPS receivers. The only communications
paths available are via the ad-hoc network and there is no The density of the vehicles depends on parameter λ. The
other communication infrastructure. The power of the antenna number of vehicles per lane is between (2 -15 v/km/Lane)
is Ptx = 18dBm and the communication frequency f is 5.9 depending on the speed being used and the value of parameter
Ghz. λ according to Table III.
Parameter λ ν/km/lane
3 8-15
5 5-9
7 3-6

TABLE III
D ENSITY PER LANE .

Fig. 5. Simulation environment.

B. Evaluation criteria
The communication range of the vehicles is calculated In order to evaluate the effect of vehicles as obstacles in
according to Table I. In our simulations, we use a minimum the performance of Enhanced Spring Clustering method we
sensitivity (Pth ) of -69 dBm which gives a transmission range conducted several simulation scenarios. The vehicles that enter
of 130 for LOS. the simulation are of 25 % trucks with average height of 2.5
meters, when the rest of the cars have height of 1.5 meters.
Data Rate (Mb/sec) Minimum Sensitivity(dBm) The placement of the antenna is in the roof of the car at a
3 -85 height of 10 decimeters above the car or inside the car at a
4.5 -84
6 -82 height of 0.5 meter below the roof of the vehicle(e.g., 2 meters
9 -80 for trucks and 1 meter for cars).
12 -77 We measured how density of the cars affects the perfor-
18 -70
mance of our method when cars are treated as obstacles. We
24 -69
27 -67 observed that the placement of the antenna inside the car gives
a big difference in reliable communication range and to the
TABLE I overall performance of the clustering.
M INIMUM SENSITIVITY IN RECEIVER ANTENNA ACCORDING TO DATA We also measured how the ratio of tall vehicles affect the re-
RATE .
liable communication range. We investigated how favoring tall
vehicles to become clusterheads (Enhanced Spring Clustering)
In a situation of OLOS the reliable communication range in a realistic scenario, where reliable communication range is
(RCR) is calculated according to: computed among each pair of nodes, makes the method more
stable.
Ptx − F P LS − Ad < Pth (2) Clustering performance. In order to evaluate the effect
of OLOS in stability of our clustering method we created
where the value of d is the biggest one that makes Equation 2 scenarios of different car densities. The results clearly indicate
true and FPLS is the loss in signal strength of an electro- that vehicles as obstacles have a significant impact on the
magnetic wave that would result from a line-of-sight path formation of clusters in a typical VANET clustering method
through free space (usually air), with no obstacles nearby to compared to those acquired when we neglect this phenomenon.
cause reflection or diffraction. FPLS is calculated according This is due to the fact that the medium contention is overes-
to Equation 3 timated in models that do not include vehicles as obstacles
F P LS = 92.45 + 20 ∗ Log(RCR) + 20 ∗ Log(f ). (3) in the calculation and that the reliable communication range
is considered to be significantly bigger than in the real world
A. The mobility model (Figure 6).
The arrival rate of the vehicles follows the Poison process Also, the number of mean cluster change (Figure 7) and the
with parameter λ. The speed assigned to the vehicles is average lifetime (Figure 8) of clusters are also affected when
according to the lane it chooses to follow according to Table II. the cars are treated as obstacles in the simulation.
 160 Antenna placement. The placement of the antenna inside
RCR Sp-Cl
Sp-Cl the car or at the roof of it, plays a significant role at the
140
attenuation caused by the obstructing cars. Table IV shows
120 mean reliable communication range for different antenna
Number of Clusters

placements. When the antenna in mounted inside the car it

100
is under windshield at 50cm below the height of the car and
80 the cars enter the simulation with parameter λ = 5.

60 placement RCR Ad
roof 80 10
40
inside 45 25
3 5 7
density TABLE IV
A NTENNA PLACEMENT.

Fig. 6. Impact of OLOS in average number of clusters

Tall vehicles. When the percentage of the tall vehicles
(trucks) increases the mean reliable communication range is
RCR Sp-Cl
14 Sp-Cl strongly affected. This is due to the fact that the attenuation
caused by this vehicles when in the middle of the communica-
12
Average Cluster changes

tion range of other cars is bigger. The reliable communication

10 range, when there is an obstacle between transmitter and
receiver, drops to almost half of the static communication
8
range when 50 % of the cars are tall vehicles and increases
6 again as this percentage increases.
4
80 30
RCR
2 Ad 28
3 5 7
26
Reliable communication Range

density
75 24

Mean diffraction loss

22

Fig. 7. OLOS influence average cluster changes / vehicle 20

70
18
16
70
RCR Sp-Cl 65 14
Sp-Cl
12
65
10
60
60 15 25 50 75 85
Cluster lifetime

Percentage of tall vehicles

55

50 Fig. 9. Number of Trucks (tall vehicles) influence reliable communication

range
45

40
Special role of vehicles. In order to evaluate how favoring
3 5 7
tall vehicles to become clusterheads affects Enhanced Spring
density
Clustering, we created scenarios of different vehicle distri-
butions. The average density is of λ = 5, the percentage
Fig. 8. Average cluster lifetime under OLOS of tall vehicles is 15% and all vehicles move to the same
direction. The simulations showed that favoring tall vehicles
in a realistic scenario, where reliable communication range is
The obstructing vehicles decrease reliable communication computed among each pair of nodes, makes Enhanced Spring
range which leads to what is called a lost link event, causing Clustering more stable (Figure 10).
the presence of often transition events to vehicles. Nodes that This is due to the fact that tall vehicles have biggest average
loose communication with their neighbors, due to attenuation reliable communication range (RCR) with their neighbors
by obstructing vehicles, leave their clusters, join other nearby compared to short vehicles, since the latter are more vulnerable
ones or form new clusters degrading the overall performance to diffraction losses. This phenomenon doesn’t show up when
of the clustering method. obstructing vehicles are neglected, since in that situation all
 95 pairs of vehicles. The incorporation of heights in clusterhead
RCR tall vehicle
90 RCR Short Vehicle election makes the method more stable. More sophisticated
85 approaches where the relative heights of vehicles according to
Mean lifetime of Clusters

80 the heights of all its neighbors are under investigation in order

75
to further stabilize Spring Clustering.
70
65 R EFERENCES
60
[1] A. Amis, R. Prakash, T. Vuong, and H. D.T., “Max-min d-cluster
55
formation in wireless ad hoc networks.” in Proceedings of the IEEE
50 Conference on Computer Communications (INFOCOM), 2000, pp. 32–
45 41.
40
[2] C. Tselikis, S. Mitropoulos, N. Komninos, and C. Douligeris, “Degree-
0 2 4 6 8 10 12 based clustering algorithms for wireless ad hoc networks under attack,”
Simulation IEEE Communications Letters, vol. 16, no. 5, pp. 619–621, 2012.
[3] N. Dimokas, D. Katsaros, and Y. Manolopoulos, “Energy-efficient
distributed clustering in wireless sensor networks,” Journal of Parallel
Fig. 10. Tall vehicles play significant role in Spring Clustering. and Distributed Computing, vol. 70, no. 4, pp. 371–383, 2010.
[4] O. Younis, M. Krunz, and S. Ramasubramanian, “Node clustering
in wireless sensor networks: Recent developments and deployment
challenges,” IEEE Network magazine, vol. 20, no. 3, pp. 20–25, 2006.
vehicles have static communication ranges (CR) with all their [5] L. Maglaras and D. Katsaros, “Distributed clustering in vehicular
networks,” in Proceedings of the International Workshop on Vehicular
neighbors, independently of their height or other obstructing Communications and Networking (VECON), 2012.
cars (Figure 11). [6] M. Ni, Z. Zhong, and D. Zhao, “MPBC: A mobility prediction-based
clustering scheme for ad hoc networks,” IEEE Transactions on Vehicular
Technology, vol. 60, no. 9, pp. 4549–4559, 2011.
120 [7] S. Vodopivec, J. Bester, and A. Kos, “A survey on clustering algorithms
CR tall vehicle
CR Short Vehicle for vehicular ad-hoc networks,” in Proceedings of the International
110 Conference on Telecommunications and Signal Processing (TSP), 2012,
pp. 52–56.
Mean lifetime of Clusters

[8] I. Sen and D. W. Matolak, “Vehicle channel models for the 5-GHz band,”
100
IEEE Transactions on Intelligent Transportation Systems, vol. 9, no. 2,
pp. 235–245, 2008.
90 [9] A. Paier, J. Karedal, N. Czink, C. Dumard, T. Zemen, F. Tufvesson,
A. F. Molisch, and C. F. Mecklenbräuker, “Characterization of vehicle-
80 to-vehicle radio channels from measurements at 5.2GHz,” Wireless
Personal Communications, vol. 50, no. 1, pp. 19–32, 2009.
70 [10] J. S. Otto, F. E. Bustamante, and R. A. Berry, “Down the block and
around the corner: The impact of radio propagation on inter-vehicle
60
wireless communication,” in Proceedings of the IEEE International
0 2 4 6 8 10 12 Conference on Distributed Computing Systems (ICDCS), 2009, pp. 605–
Simulation 614.
[11] M. Boban, T. T. V. Vinhoza, M. Ferreira, J. Barros, and O. K. Tonguz,
“Impact of vehicles as obstacles in vehicular ad hoc networks,” IEEE
Journal on Selected Areas in Communications, vol. 29, no. 1, pp. 15–28,
Fig. 11. Height of clusterhead does not affect Spring Clustering’s perfor-
2011.
mance, when static communication range is used
[12] S. Kaul, K. Ramachandran, P. Shankar, S. Oh, M. Gruteser, I. Seskar,
and T. Nadeem, “Effect of antenna placement and diversity on vehicular
Average number of clusters and mean cluster transitions per network communications,” in Proceedings of the IEEE Conference on
Sensor, Mesh and Ad Hoc Communications and Networks (SECON),
vehicles also follow the same patterns, showing that Enhanced 2007, pp. 112–121.
Spring Clustering which assigns special roles to tall vehicles [13] J. Maurer, T. Fugen, T. Schafer, and W. Wiesbeck, “A new inter-vehicle
is more stable than the initial one. communications (IVC) channel model,” in Proceedings of the IEEE
Vehicular Technology Conference – Fall, vol. 1, 2004, pp. 9–13.
[14] D. Krajzewicz, G. Hertkorn, C. Rossel, and P. Wagner, “Sumo (simula-
VI. C ONCLUSIONS tion of urban mobility); an open-source traffic simulation,” 4th Middle
East Symposium on Simulation and Modelling (MESM), pp. 183–187,
Treating vehicles as obstacles, has a significant impact on 2002.
the reliable communication range. Reliable communication [15] J. B. H. Ledy, P. A. M., H. B., and V. R., “A semi-deterministic channel
range that is calculated for every pair of nodes for every time model for VANETs simulations,” International Journal of Vehicular
Technology, 2011.
instance according to the attenuation caused by obstructing [16] D. Krajzewicz, G. Hertkorn, C. Rossel, and P. Wagner, “Propagation by
vehicles is exploited for the design of a new VANET clustering diffraction.” International Telecommunication Union Radiocommunica-
protocol, namely the Enhanced Spring Clustering algorithm. It tion Sector, Geneva, Recommendation P.526, 2007.
[17] J. Epstein and P. D. W., “An experimental study of wave propagation at
is shown that significant benefits are observed when the height 850M C,” Proceedings of the IRE, vol. 41, no. 5, pp. 595–611, 1953.
of vehicles is taken into account when electing clusterheads;
this feature is in the heart of the Enhanced Spring Clustering.
This behavior is observed only in a realistic scenario where
reliable communication range is computed among each pair
of nodes instead of using a static communication range for all