Content uploaded by Chetan B M
Author content
All content in this area was uploaded by Chetan B M on Oct 17, 2017
Content may be subject to copyright.
Fair Decentralized Data-Rate Congestion Control
for V2V Communications
Chetan Belagal Math∗, Hong Li†, Sonia Heemstra de Groot∗, Ignas Niemegeers∗
∗Electrical Engineering, Eindhoven University of Technology, Eindhoven, The Netherlands
†Car Infotainment & Driving Assistance, NXP Semiconductors, Eindhoven, The Netherlands
Email: c.belagal.math@tue.nl, hong.r.li@nxp.com, s.heemstradegroot@tue.nl, i.g.m.m.niemegeers@tue.nl
Abstract—Channel congestion is one of the most critical
issues in IEEE 802.11p-based vehicular ad hoc networks
because congestion may lead to unreliability of applications.
As a counter measure, the European Telecommunications
Standard Institute (ETSI), proposes a mandatory Decentralized
Congestion Control (DCC) framework to control the channel
load. DCC algorithms are proposed to tune parameters such as
message-rate, data-rate, etc. to avoid congestion. An important
requirement for DCC algorithms is fairness, which ensures
that vehicles experiencing similar channel loads are entitled
to similar transmission parameters, in particular, message-rate
and data-rate. Message-rate DCC (LIMERIC) ensures a fair
message-rate selection, while data-rate DCC (DR-DCC) might
end up with different data-rates, creating unfairness among
the vehicles: vehicles with lower data-rate have a larger
communication range than those using higher data-rates.
Therefore some vehicles are less visible than others, which is
detrimental to the reliability of the safety applications. To avoid
this, the paper defines a novel packet-count based decentralized
data-rate congestion control algorithm (PDR-DCC), which
enforces fairness and hence improves the application-reliability.
Simulation studies are performed to demonstrate that PDR-DCC
avoids congestion in a fair manner. We also show the
effect of fairness on the application-reliability by comparing
the performance of PDR-DCC with message-rate (LIMERIC)
and data-rate (DR-DCC) congestion control algorithms for a
stationary vehicle warning application in a synthetic highway
scenario and for various vehicular densities. We conclude that
PDR-DCC outperforms LIMERIC, and DR-DCC in terms of
application-reliability.
I. INTRODUCTION
Vehicle-to-vehicle (V2V) communication systems are
essential for increasing the safety and efficiency of
our transportation network. Safety applications rely on
beacon messages, exchanged periodically between neighbors.
These messages enable sharing of information such as
location, speed, braking action, road and intersection status,
thus detecting various dangerous situations and enabling
different safety applications [1]. The beacon messages are
referred to as Cooperative Awareness Messages (CAMs)
in Europe [2] and Basic Safety Message (BSMs) in the
U.S. [3]. The corresponding European Telecommunications
Standard Institute (ETSI) and Wireless Access for Vehicular
Environments (WAVE) standards implement the IEEE 802.11p
PHY and MAC layers.
Most V2V applications require a high penetration rate,
i.e., a high percentage of vehicles equipped with active
V2V communication modules. The drawback of exchanging
periodic beacon messages is that they generate a significant
channel load. Thus at high traffic density, this leads to
congestion, causing beacon message collisions and hence,
degradation of the safety application reliability. To mitigate
this problem, various Decentralized Congestion Control
(DCC) algorithms have been proposed. By controlling
parameters such as message-rate and transmission power,
they maintain the channel load below a threshold to avoid
congestion. This, however, may have a negative impact on
the application-reliability. Furthermore, Fair allocation of
resources such as message-rate, data-rate, etc. is an important
requirement for DCC algorithms. Fair allocation means that
vehicles experiencing similar channel loads, should be entitled
to similar transmission parameters. As unfair allocation of
resources may lead to variation in range, message-rate, etc.
affecting application-reliability.
To ensure fairness and improve application-reliability, we
propose an adaptive decentralized data-rate congestion control
algorithm (PDR-DCC), that uses a packet count to enforce
selection of the optimal data-rate by each vehicle. We show
that this has the required fairness property and hence also a
higher reliability performance compared to other congestion
control algorithms. This is done for various traffic densities
by means of simulations, using network simulator ns-3 in
combination with a traffic simulator Simulation of Urban
MObility (SUMO) [4] to simulate a realistic highway scenario.
Following this introduction, we describe in Section II
discuss related works. In Section III we discuss PDR-DCC
and explain the unfairness issue. In section IV we present
the packet count mechanism and how it mitigates the
unfairness problem. In Section V we evaluate the performance
of our solutions. First, we demonstrate that PDR-DCC
converges the channel load to the desired target while
ensuring a fair allocation of the data-rates. Next, we compare
its fairness performance with a data-rate DCC algorithm
(DR-DCC) for different vehicular densities. Finally, we
compare the application-reliability of PDR-DCC, with other
other congestion control algorithms (LIMERIC and DR-DCC)
for different vehicular densities. We summarize our results and
discuss open issues in Section VI.
II. RELATED WORKS
Several DCC algorithms have been proposed. Most of the
DCC algorithms rely on the Channel Busy Ratio (CBR)
measurements to represent channel load. The CBR represents
the fraction of time that the channel is sensed busy by a
vehicle over a period 𝜃, where “vehicle” refers to a vehicular
communication node. The CBR measurement is performed by
each vehicle independently. One can also use a global CBR,
derived by exchanging the local CBR measurements using
beacon packets.
Many research efforts such as [5]–[9] have been carried
out to study and improve the performance of V2V
applications in congested scenarios. DCC algorithms could
be broadly classified as message-rate, transmission power,
and, data-rate congestion control algorithms, as discussed
in [8]. Message-rate DCC algorithms such as LIMERIC
[5] and AIMD [6] tune message-rate based on CBR to
avoid congestion. Transmission power DCC algorithms such
as D-FPAV [7] and TP-DCC [8] maximize the minimum
transmission power for vehicles, while avoiding congestion.
Data-rate congestion control algorithms such as DR-DCC [8]
and PHY-DCC [9] are proposed to tune data-rate based on
CBR.
Among above discussed works, the most prominent
DCC approaches are message-rate and data-rate congestion
control. Typical representatives of message-rate and data-rate
congestion control algorithms are LIMERIC [5] and DR-DCC
[8] respectively. Both LIMERIC and DR-DCC aim at keeping
the CBR below a threshold, 𝐶𝐵𝑅𝑇, above which the channel
is considered to be congested. We discuss them in more detail
below.
A. LIMERIC
LIMERIC is a distributed linear adaptive algorithm that
forces the channel load to converge to 𝐶𝐵𝑅𝑇.𝐶𝐵𝑅𝑘(𝑡)is
the global channel load computed by vehicle 𝑘at time 𝑡,
by averaging the CBR measurements obtained from neighbor
vehicles within its communication range. The message-rate 𝑅𝑘
is adjusted every 𝜃seconds as follows:
𝑅𝑘(𝑡)=(1−𝛼)×𝑅𝑘(𝑡−𝜃)+𝛽×(𝐶𝐵𝑅𝑇−𝐶𝐵𝑅𝑘(𝑡−𝜃))
(1)
where the speed of convergence is determined by 𝛼, and 𝛽
ensures stability and steady state convergence. In steady state,
LIMERIC converges to the channel load 𝐶𝐵𝑅𝑐𝑜𝑛 given by:
𝐶𝐵𝑅𝑐𝑜𝑛 =𝑛×𝛽
20 ×𝛼+𝑛×𝛽×𝐶𝐵𝑅𝑇(2)
where 𝑛is the number of vehicles within the interference
range. It is evident that at larger vehicular densities the
difference between 𝐶𝐵𝑅𝑇and 𝐶𝐵𝑅𝑐𝑜𝑛 in steady state
becomes smaller.
B. DR-DCC
DR-DCC is a data-rate decentralized congestion control
algorithm. Each vehicle adjusts its data-rate, based on the CBR
it observes. If the CBR is greater than an upper threshold,
𝐶𝐵𝑅𝑇, it increases the data-rate, while, if it is less than
a lower threshold, 𝐶𝐵𝑅𝑚𝑖𝑛, it decreases the data-rate. It
thus avoids congestion by maintaining CBR below a threshold
𝐶𝐵𝑅𝑇. CBR measurements are subject to some randomness,
hence, to avoid unnecessary fluctuations of the data-rate it
utilizes a hysteresis as discussed in [8]. However, the reliance
of DR-DCC on CBR measurements alone can lead to an
unfair data-rate allocation among vehicles experiencing similar
channel load conditions. As CBR measurements alone cannot
provide sufficient information for fair allocation of data-rates.
i.e., two nearby vehicles experiencing a CBR of 0.6 may select
6 Mbps and 24 Mbps data-rates. This unfair allocation of
data-rates will cause different vehicles with similar channel
activity to have different transmission ranges and hence affects
application-reliability.
LIMERIC decreases the message-rate to reduce congestion
without increasing the channel capacity. On the contrary,
DR-DCC increases the data-rate to reduce congestion,
effectively making messages shorter in time, thus increasing
channel capacity and lowering the channel load for a
given message-rate. DR-DCC maintains a higher message-rate
even at high traffic densities compared to LIMERIC,
by increasing channel capacity enabling better application
reliability. However, LIMERIC ensures a fair message-rate
selection, while in DR-DCC different vehicles might end
up with a different data-rate, creating unfairness among the
vehicles, affecting application-reliability.
Algorithm 1 : PDR-DCC algorithm
1: while PDR-DCC active do
2: get 𝑃𝑐periodically for 𝜃seconds
3: if (𝑃𝐶<𝐶𝐵𝑅𝑇×𝜃
𝑇3)then
4: D = 3 Mbps
5: else if ((𝑃𝐶>𝐶𝐵𝑅𝑇×𝜃
𝑇3) and (𝑃𝐶<𝐶𝐵𝑅𝑇×𝜃
𝑇6)) then
6: D = 6 Mbps
7: else if ((𝑃𝐶>𝐶𝐵𝑅𝑇×𝜃
𝑇6) and (𝑃𝐶<𝐶𝐵𝑅𝑇×𝜃
𝑇9)) then
8: D = 9 Mbps
9: else if ((𝑃𝐶>𝐶𝐵𝑅𝑇×𝜃
𝑇9) and (𝑃𝐶<𝐶𝐵𝑅𝑇×𝜃
𝑇12 )) then
10: D = 12 Mbps
11: else if ((𝑃𝐶>𝐶𝐵𝑅𝑇×𝜃
𝑇12 ) and (𝑃𝐶<𝐶𝐵𝑅𝑇×𝜃
𝑇18 )) then
12: D = 18 Mbps
13: else
14: D = 24 Mbps
15: end if
16: end while
III. THE PDR-DCC ALGORITHM
To ensure fairness of data-rate congestion control and
improve application-reliability further, we propose PDR-DCC,
which uses packet count 𝑃𝐶together with CBR measurements
to adjust the data-rate. 𝑃𝐶represents the total number of
packets sensed by a vehicle over a period 𝜃. CBR depends
on various factors such as the number of transmitters, the
distribution of packet lengths, the distribution of packet
transmission time, etc. If we assume that all packets are sent
with the same data-rate 𝐷, then the total time 𝑇𝑇the channel
is sensed busy by the vehicle can be expressed as:
𝑇𝑇=𝐶𝐵𝑅 ×𝜃=𝑃𝐶×𝑇𝐷(3)
PDR-DCC is described in Algorithm 1. PDR-DCC adjusts
𝐷based on 𝑃𝐶, measured every 𝜃seconds. 𝐷is selected
such that CBR converges to the target 𝐶𝐵𝑅𝑇. Depending
on the value of 𝑃𝐶and 𝑇𝐷the transmission time of a
packet for different data rates 𝐷. According to the IEEE
802.11p standard, 𝐷can take on the values as 3, 6, 9,
12, 18, and 24 Mbps. Each vehicle independently selects
the lowest permissible data-rate, that avoids congestion and
provides the maximum possible range [8]. In PDR-DCC
vehicles experiencing similar channel loads have a similar
packet count and thus the same data-rate, ensuring fairness.
where 𝑇𝐷is the packet transmission time with data-rate 𝐷.
Since beacons have a fixed length, 𝑇𝑇only depends on the
data-rate. Thus, increasing the data-rate, decreases 𝑇𝑇, thus
decreasing CBR.
IV. PACKET COUNT
In this section, we motivate the advantages of using 𝑃𝐶.
We also discuss how 𝑃𝐶is estimated.
A. Why packet count
Most of the message-rate [5], [6] based congestion control
algorithms rely on CBR. Message-rate based congestion
control schemes try to maintain CBR below 𝐶𝐵𝑅𝑇by
decreasing the message-rate, thus reducing the load by
maintaining the same channel capacity. On the contrary,
the data-rate based approaches increase the channel capacity
by transmitting packets at higher data-rates and therefore
reducing the channel occupancy. For example: consider
packets transmitted at 6 Mbps resulting in a transmission time
of 540 𝜇s. In order to maintain a CBR of 0.7 over a period
of 1 s, the maximum number of packets that the channel can
support is 1296. Similarly, if 12 Mbps is used, the transmission
time is 360 𝜇s. To maintain a CBR of 0.7 over a period of
1 s, the maximum number of packets that can be transmitted
is increased to 1944. As seen, the CBR is a function of both
the total number of packets in the channel and the selected
data-rates. 𝑃𝐶also represents the channel load and can be used
to select a data-rate 𝐷that converges the CBR to 𝐶𝐵𝑅𝑇as
it is independent of the selected data-rates. The selected 𝐷is
the same for nearby vehicles, which experience similar packet
count.
B. Packet count estimation
The 𝑃𝐶is calculated by every vehicle, based on information
from the physical layer. There are three cases where the
channel is sensed busy: 1) the vehicle is transmitting a packet,
2) the vehicle is receiving a packet, 3) the vehicle is neither
transmitting nor receiving, however, the signal strength is
higher than the carrier sensing threshold, due to collisions,
interference, etc. The total number of transmitted packets 𝑃𝑇
by a vehicle and the total transmission time 𝑇𝑇𝑋, can be
obtained from the physical layer. The IEEE 802.11p packets
consist of preamble, signal field, and payload [10]. The
signal field consists of the data-rate and packet length, which
determines the transmission time. This information is used
to calculate 𝑃𝑅, the total number of packets sensed by the
receiver and 𝑇𝑅𝑋, the total time the channel is sensed busy
due to reception at the physical layer. In order to obtain 𝑃𝐵,
the total number of packets and 𝑇𝐵𝑈, the total time during
which the channel is busy while not receiving or transmitting,
we utilize CBR as in Eq.(4). 𝑇𝑇, the total time the channel
is sensed busy by a vehicle is obtained from Eq.(3). 𝑃𝐶is
obtained from 𝑃𝑇,𝑃𝑅, and 𝑃𝐵as in Eq.(6), where 𝑃𝐵is
obtained from linear estimation over time, 𝑇𝐵𝑈,asinEq.(5).
The 𝑃𝐶estimation is performed by each vehicle individually.
𝑇𝐵𝑈 =(𝐶𝐵𝑅 ×𝜃)−(𝑇𝑇𝑋 +𝑇𝑅𝑋 )(4)
𝑃𝐵=((𝑃𝑇+𝑃𝑅)×𝑇𝐵𝑈 )
(𝑇𝑇𝑋 +𝑇𝑅𝑋 )(5)
𝑃𝑐=𝑃𝑅+𝑃𝑇+𝑃𝐵(6)
V. P ERFORMANCE EVA L U AT I O N
We performed a simulation study using the network
simulator ns-3 in conjunction with the vehicle traffic
SUMO [4] to evaluate the capabilities of PDR-DCC and
compare with DR-DCC and LIMERIC. We compare the
application-reliability performance of the stationary-vehicle
warning application, which alerts approaching vehicles about
any vehicle being dangerously motionless on the road [11].
We consider a realistic highway scenario: a road section
of 3 km with 3 lanes in each direction (Figure 1). When a
vehicle reaches the end of the road it loops around and enters
the opposite direction. SUMO provides the mobility of the
vehicles to the network simulator.
3.5
m
3.5
m
3.5
m
3.5
m
3.5
m
3.5
m
5 m
2 km
1 km
OBSERVING
ZONE
OBSERVING
ZONE
0 km
3 km
Fig. 1: Simulation scenario
The simulation parameters are taken from [8] [12]. They
are listed in Table I. The parameters for the channel models
are obtained from the highway scenario channel specifications
in the ETSI standard [13]. We consider three vehicular
density scenarios: 200 (medium), 300 (dense) and 400
(extreme) vehicles/km. For LIMERIC the data-rate is fixed
to 6 Mbps (default data-rate as per standards [12]) and the
message-rate is varied. Similarly, for DR-DCC and PDR-DCC,
the message-rate is fixed at 10 𝐻𝑧, which is the maximum
permissible message-rate [12] and the data-rate is varied. The
initial values of the data-rates are drawn from a discrete
uniform distribution over all permissible data-rates: 3, 6, 9,
12, 18, and 24 Mbps. To eliminate the effects of the boundary
conditions of the network, we only observe the performance
of vehicles in the observing zone, a 1 km section in the middle
of the 3 km road section (Figure 1).
For all three algorithms, we set 𝐶𝐵𝑅𝑇=0.7and 𝜃=
200𝑚𝑠 as discussed in [14]. The other parameters are as listed
in Table II. The parameters for LIMERIC were taken from [5],
and for DR-DCC from [8]. For PDR-DCC the transmission
time 𝑇𝐷for different data-rates 𝐷depends on the payload,
preamble, and other headers. The decrease in transmission
time is not proportional to the increase in data-rate as the
preamble occupies a significant part of the transmission time
[10].
TABLE I: Simulation parameters
Vehicular density medium, dense, extreme
Carrier sensing
threshold -85 dBm
Transmit power 24 dBm
(Large scale fading)
Dual slope model [13]
Path loss exponent 1 (until 80 m) 1.9
Path loss exponent 2 (after 80 m) 3.8
(Small scale fading)
Nakagami m model [13]
Distance bin in m m
0-50 3
51-150 1.5
Above 150 1
Payload size 300 Bytes
Message rate 10 Hz
Simulation area 3000 m x 25 m
Simulation period 30 s
Mobility SUMO
Stationary vehicle
warning application
N 1
T(s) 1
TABLE II: Algorithm parameters
Algorithm Parameter Value
DR-DCC CBRmin 0.5
PDR-DCC
T31026 𝜇s
T6540 𝜇s
T9370 𝜇s
T12 290 𝜇s
T18 200 𝜇s
T24 170 𝜇s
LIMERIC 𝛼0.1
𝛽0.033
A. Evaluation Metrics
In this section, we discuss the metrics used to evaluate
channel load, fairness, and application-reliability of the
congestion control algorithms.
1) Channel load: The congestion control algorithm should
avoid congestion by keeping the channel load below desired
threshold 𝐶𝐵𝑅𝑇. The average CBR measurements of all
vehicles in the observing zone is used as the metric to represent
the channel load.
2) Fairness: Fairness in this context means equal
resource allocation (message-rate, data-rate, etc.) for vehicles
experiencing the same channel load. Unfair allocation of
resources may lead to variation in range, message-rate,
etc. among vehicles. This may affect application-reliability.
Fairness is quantified using Jain’s fairness index [15] over 𝑋,
where 𝑋represents the fraction of time spent by a vehicle for
beacon message transmission in the observing zone. The 𝑋
of a vehicle 𝑖is calculated as shown below:
𝑋𝑖=
𝑆
∑
𝑗=1
𝑇𝑗
𝛾(7)
where a vehicle 𝑖transmits 𝑆packets each with a 𝑗th packet
transmission time of 𝑇𝑗over a period 𝛾in the observing zone.
Due to mobility, a vehicle may stay in the observing zone
for 2 s and transmits 2 beacon messages at 6 Mbps or for 4 s
and transmits 4 beacon messages at 6 Mbps. The total beacon
transmission time, generally used to quantify fairness [12] for
both cases are different, as they depend on the amount of time
the vehicle stays in the observing zone. Hence, we make use
of X, a normalized metric over time, which is the same for
both cases. This makes the fairness measure independent of
the time a vehicle spent in the observing zone.
The Jain fairness index 𝐽𝐹of 𝑀vehicles passing through
the observing zone during the simulation is assessed as shown
in Eq.(8). It varies from 1
𝑀(worst case) to 1(best case).
The best case is when all vehicles have an equal share of
the channel for beacon message transmissions.
𝐽𝐹=(𝑀
∑
𝑖=1
𝑋𝑖)2
𝑀×
𝑀
∑
𝑖=1
𝑋2
𝑖
(8)
3) Application reliability: In order to support a safety
application reliably, a number of packets 𝑁over a tolerance
time 𝑇needs to be received from each neighbor within the
awareness range of the application. The awareness range is
defined as the minimum distance within which all vehicles that
could constitute a hazard for the vehicle running the particular
safety application, need to be detected.
In our study, we focus on the requirements of the stationary
vehicle warning application. The stationary vehicle warning
application alerts approaching vehicles about any vehicle
being dangerously motionless on the road [1]. The application
requires at least a single packet 𝑁=1within a 𝑇window
of 1 s [11]. The awareness range depends on the speed of the
vehicles, the reaction time of the driver, etc. In a worst case
scenario (speed of 120 km/h), a vehicle requires an awareness
range of 200 meters to avoid collision [11].
In [16] the authors propose the T-window application
reliability metric to quantify the application-reliability. It is
defined as the probability of successfully receiving 𝑁packets
over a tolerance time 𝑇for a particular application. Thus,
the metric indicates the probability that a safety application is
0 5 10 15 20 25 30
Time (sec)
0
0.2
0.4
0.6
0.8
1
Average CBR
CBRT
DR-DCC
PDR-DCC
(a) Medium
0 5 10 15 20 25 30
Time (sec)
0
0.2
0.4
0.6
0.8
1
Average CBR
CBRT
DR-DCC
PDR-DCC
(b) Dense
0 5 10 15 20 25 30
Time (sec)
0
0.2
0.4
0.6
0.8
1
Average CBR
CBRT
DR-DCC
PDR-DCC
(c) Extreme
Fig. 2: CBR vs Time for different vehicular densities
supported reliably by the underlying communication system.
T-window application reliability is utilized in this paper to
assess application-reliability of the stationary vehicle warning
application.
B. Channel load
The average CBR measurements at different densities
demonstrate the capability of PDR-DCC to converge the
channel load (average CBR) to the desired target 𝐶𝐵𝑅𝑇.
Figure 2 shows the average CBR over time of PDR-DCC
and DR-DCC, in the observing zone for medium (Figure 2a),
dense (Figure 2b), and extreme (Figure 2c), traffic densities.
As seen after the initial transition time of PDR-DCC to
select the optimal data-rate, the average CBR at different
vehicular densities converges to 𝐶𝐵𝑅𝑇, which is represented
by a horizontal line in Figure 2. The average CBR is closer
to 𝐶𝐵𝑅𝑇for PDR-DCC than for DR-DCC specifically at
medium and dense traffic densities indicating more efficient
utilization of the channel for PDR-DCC. The variation in
CBR is due to fading, discreet data-rates and vehicle density
variation due to mobility.
C. Fairness
In order to evaluate the capability of PDR-DCC to provide
fair allocation of data-rates, we study and compare the
performance of PDR-DCC and DR-DCC, using Jain’s fairness
index as suggested in the ETSI standard [12]. We perform
simulations at medium vehicular density with two different
setups for initial data-rates: a) the uniform distribution
presented before, and b) the case where all vehicles start with
a data-rate of 24 Mbps. For medium vehicular density, Jain’s
fairness index can vary from 0.005 ( 1
𝑀, worst case) to 1(best
case).
Figure 3 shows the results for the uniform distributed
data-rate initialization. Figure 3a shows Jain’s fairness index
for PDR-DCC and DR-DCC. As seen PDR-DCC shows better
fairness than DR-DCC by 25 %. Figure 3b and Figure 3c
represents the percentage of data-rates selected by the vehicles
in the observing zone over time by DR-DCC and PDR-DCC
algorithms respectively. In PDR-DCC after the initial transition
all the vehicles in the observing zone select 9 Mbps. However,
in DR-DCC vehicles select different data-rates to maintain the
CBR below threshold leading to unfairness.
Figure 4 depicts the results where vehicles are initialized
with 24 Mbps data-rate. Figure 4a shows the Jain fairness
index for PDR-DCC and DR-DCC. Similar to earlier results
PDR-DCC has better fairness than DR-DCC by 10%. Figure
4b and Figure 4c show the percentage of data rate selected by
the vehicles in the observing zone over time by DR-DCC and
PDR-DCC algorithms respectively. After the initial transition,
all the vehicles in the observing zone using PDR-DCC select
9 Mbps, whereas DR-DCC selects a fraction from 9, 12, 18
and 24 Mbps.
As seen in Figure 3 and Figure 4 the fairness and fraction
data-rate distribution in DR-DCC are dependent on the initial
conditions of data-rate distribution. From the study, it is
evident that DR-DCC may lead to unfair data-rate allocation
and decreased fairness. However, PDR-DCC implements
a mechanism to enforce homogeneous data-rate selection
amongst all vehicles to ensure fairness. For example: Let us
consider a group of vehicles selects 12 Mbps and has a CBR
0.6 and a new vehicle joins the group which has 6 Mbps
data-rate in the case of DR-DCC it still maintains 6 Mbps as
CBR measured is below 0.7 leading to unfairness. As CBR is
dependent on selected data-rates. However, in PDR-DCC the
new vehicle selects 12 Mbps depending on the packet count,
as it is independent of selected data-rates, thus leading to fair
allocation of data-rate.
D. Application reliability and awareness range
To benchmark application reliability, we compare the
stationary vehicle warning application reliability and the
awareness range performance of PDR-DCC with other widely
accepted (message-rate) LIMERIC and (data-rate) DR-DCC
algorithms at different vehicular densities.
As the awareness range depends on velocity, relative
velocity, etc., we assess the maximum possible range up to
which an application can be supported reliably. We consider
communication support to the application reliable, if the
underlying communication channel satisfies the requirements
PDR-DCC DR-DCC
0
0.2
0.4
0.6
0.8
1
Jain fairness index
(a) Jain fairness index
0 5 10 15 20 25 30
Time (sec)
0
10
20
30
40
50
Percentage of datarate
3 Mbps
6 Mbps
9 Mbps
12 Mbps
18 Mbps
24 Mbps
(b) DR-DCC
0 5 10 15 20 25 30
Time (sec)
0
20
40
60
80
100
Percentage of datarate
3 Mbps
6 Mbps
9 Mbps
12 Mbps
18 Mbps
24 Mbps
(c) PDR-DCC
Fig. 3: Uniform data-rate initialization
PDR-DCC DR-DCC
0
0.2
0.4
0.6
0.8
1
Jain fairness index
(a) Jain fairness index
0 5 10 15 20 25 30
Time (sec)
0
20
40
60
80
100
Percentage of datarate
3 Mbps
6 Mbps
9 Mbps
12 Mbps
18 Mbps
24 Mbps
(b) DR-DCC
0 5 10 15 20 25 30
Time (sec)
0
20
40
60
80
100
Percentage of datarate
3 Mbps
6 Mbps
9 Mbps
12 Mbps
18 Mbps
24 Mbps
(c) PDR-DCC
Fig. 4: 24 Mbps data-rate initialization
25 50 75 100 125 150 175 200 225 250 275 300
Distance (Meters)
0.8
0.85
0.9
0.95
1
T- window application reliability
PDR-DCC
DR-DCC
LIMERIC
(a) Medium
25 50 75 100 125 150 175 200 225 250
Distance (Meters)
0.8
0.85
0.9
0.95
1
T- window application reliability
PDR-DCC
DR-DCC
LIMERIC
(b) Dense
25 50 75 100 125 150 175 200
Distance
(
Meters
)
0.8
0.85
0.9
0.95
1
T- window application reliability
PDR-DCC
DR-DCC
LIMERIC
(c) Extreme
Fig. 5: Stationary vehicle warning application reliability for various vehicular densities
of the application with a probability of at least 0.99 [17].
Thus an application is considered to be reliable if T-window
is greater than 0.99. We measure the T-window application
reliability based on the distance between vehicles periodically,
with a period of 0.1 s, using the technique described in [16].
We calculate T-window application reliability of all links in
observing zone (where a link represents a sender-receiver
pair). We sort the links in bins based on the distance between
sender-receiver pair, among rings with a width of 25 m around
a vehicle. The T-window application reliability is measured
based on 80000 to 200000 links (depending on density and
algorithm) in the observing zone.
The T-window application reliability performance of
PDR-DCC, DR-DCC, and LIMERIC for different densities
is shown in Figure 5 for medium (Figure 5a), dense (Figure
5b) and extreme (Figure 5c) vehicular densities. For medium
density, LIMERIC is the best, while the reliability of DR-DCC
drops quickly with distance. The reliability of PDR-DCC
remains close to the one of LIMERIC up to 250 𝑚.For
medium and dense densities the reliability of PDR-DCC is
better than the one of DR-DCC. For dense and extreme
densities PDR-DCC is better than LIMERIC up to 175 𝑚.
The 0.99 application-reliability is indicated by horizontal
line in Figure 5. Table III represents the maximum awareness
range of the stationary vehicle warning application with
LIMERIC, PDR-DCC and DR-DCC algorithms at various
traffic densities. As seen the awareness range of PDR-DCC
and DR-DCC is better compared to LIMERIC specifically
at dense and extreme densities. As LIMERIC reduces
message-rate drastically at dense and extreme traffic densities
to avoid congestion leading to decreased application-reliability
and awareness range. It is evident that the awareness
range of PDR-DCC is better compared to DR-DCC. In
DR-DCC, few vehicles selects higher data-rates than the
optimal data-rate leading to decrease in application-reliability
and awareness range. However, PDR-DCC provides fair and
optimal allocation of data-rates among vehicles, results in an
improvement of application-reliability and awareness range.
TABLE III: Awareness range
Traffi c
Density
Maximum possible awareness range
PDR-DCC DR-DCC LIMERIC
Medium 250 175 250
Dense 200 175 175
Extreme 175 150 100
VI. CONCLUSIONS AND FUTURE WORK
Fairness is an important requirement of DCC algorithms.
We have seen that CBR-based data-rate congestion control
may lead to unfairness: vehicles experiencing the same
CBR, can end up with different data-rates and hence
different transmission ranges, to the detriment of the
application-reliability. To overcome this, we have proposed a
packet-count based decentralized data-rate congestion control
algorithm, which enforces a homogeneous data-rate selection
amongst vehicles to ensure fairness. We have discussed several
implementation aspects such as packet-count estimation and its
relation to CBR.
Simulation results for different traffic densities show
the capability of PDR-DCC to make the channel load
converge to the desired target for various vehicular
densities. We have shown that fairness is indeed ensured
by PDR-DCC. Furthermore, PDR-DCC outperforms other
data-rate (DR-DCC) and widely accepted message-rate
(LIMERIC) congestion control algorithms by providing better
fairness, greater application-reliability, and larger awareness
range. We conclude that the packet count based mechanism
of PDR-DCC guarantees a fair and optimal allocation
of data-rates. This results, for various vehicular densities,
in improved application reliability and awareness range
compared to DR-DCC and LIMERIC.
Research is needed to see how packet-count based
approaches along with adapting other parameters such as
message-rate, data-rate, power, etc. can be used to further
improve reliability. Although the current results are based on
the highway scenario, we believe that our conclusions also
apply to other scenarios such as rural and urban. Future work
should confirm this.
ACKNOWLEDGMENT
The work in this paper is supported by TU/e
Impuls program, a strategic cooperation between NXP
Semiconductors and Eindhoven University of Technology.
It was partially funded by the European ARTEMIS EMC2
subsidy project.
REFERENCES
[1] “Intelligent Transport Systems (ITS); Vehicular Communications; Basic
Set of Applications; Definitions,” ETSI ITS, Tech. Rep., ETSI TR 102
638, 2009.
[2] “Vehicular Communications; Basic Set of Applications; Part 2:
Specification of Cooperative Awareness Basic Service ,” ETSI ITS, Tech.
Rep., ETSI EN 302 637-2 V1.3.2, 2014.
[3] “Dedicated Short Range Communications (DSRC) Message Set
Dictionary,” SAE, International, Tech. Rep., SAE J2735, 2009.
[4] D. Krajzewicz, J. Erdmann, M. Behrisch, and L. Bieker, “Recent
Development and Applications of SUMO - Simulation of Urban
MObility,” International Journal On Advances in Systems and
Measurements, vol. 5, no. 3&4, pp. 128–138, December 2012.
[5] J. B. Kenney, G. Bansal, and C. E. Rohrs, “LIMERIC: A Linear Message
Rate Control Algorithm for Vehicular DSRC Systems,” in Proceedings of
the Eighth ACM International Workshop on Vehicular Inter-networking,
ser. VANET ’11. New York, NY, USA: ACM, 2011, pp. 21–30.
[6] T. Tielert, D. Jiang, Q. Chen, L. Delgrossi, and H. Hartenstein, “Design
methodology and evaluation of rate adaptation based congestion control
for vehicle safety communications,” in 2011 IEEE Vehicular Networking
Conference (VNC), Nov 2011, pp. 116–123.
[7] M. Torrent-Moreno, J. Mittag, P. Santi, and H. Hartenstein,
“Vehicle-to-vehicle communication: Fair transmit power control for
safety-critical information,” IEEE Transactions on Vehicular Technology,
vol. 58, no. 7, pp. 3684–3703, Sept 2009.
[8] C. Belagal Math, A. Ozgur, S. Heemstra de Groot, and H. Li, “Data
Rate based Congestion Control in V2V communication for traffic safety
applications,” in Communications and Vehicular Technology in the
Benelux (SCVT), 2015 IEEE Symposium on, Nov 2015.
[9] S. Yang, H. Kim, and S. Kuk, “Less is more: need to simplify etsi
distributed congestion control algorithm,” Electronics Letters, vol. 50,
no. 4, pp. 279–281, February 2014.
[10] J. A. Fernandez, D. D. Stancil, and F. Bai, “Dynamic Channel
Equalization for IEEE 802.11p Waveforms in the Vehicle-to-Vehicle
Channel,” in Communication, Control, and Computing (Allerton), 2010
48th Annual Allerton Conference on, Sept 2010, pp. 542–551.
[11] C. Belagal Math, S. Heemstra de Groot, and H. Li, “Risk assessment
for traffic safety applications with v2v communications,” in 2016 IEEE
84th Vehicular Technology Conference (VTC-Fall), Sept 2016, pp. 1–6.
[12] “Cross Layer DCC Management Entity for operation in the ITS G5A
and ITS G5B medium; Validation set-up and results,” ETSI ITS, Tech.
Rep., ETSI TR 101 613 V1.1.1, 2015.
[13] “Intelligent Transport Systems (ITS); STDMA Recommended
Parameters and Settings for Cooperative ITS; Access Layer Part,”
ETSI ITS, Tech. Rep., ETSI TR 102 861 V1.1.1, 2012.
[14] Y. Fallah, C.-L. Huang, R. Sengupta, and H. Krishnan, “Analysis
of Information Dissemination in Vehicular Ad-Hoc Networks With
Application to Cooperative Vehicle Safety Systems,” Vehicular
Technology, IEEE Transactions on, vol. 60, no. 1, Jan 2011.
[15] R. Jain, D.-M. Chiu, and W. R. Hawe, “A Quantitative Measure
of Fairness and Discrimination for Resource Allocation in Shared
Computer System,” 1984.
[16] F. Bai and H. Krishnan, “Reliability Analysis of DSRC Wireless
Communication for Vehicle Safety Applications,” in Intelligent
Transportation Systems Conference, 2006. ITSC ’06. IEEE, Sept 2006,
pp. 355–362.
[17] N. An, T. Gaugel, and H. Hartenstein, “VANET: Is 95% Probability of
Packet Reception Safe?” in ITS Telecommunications (ITST), 2011 11th
International Conference on, 2011.