You are on page 1of 5

Cross-Validation of DSRC Radio Testbed and NS-2

Simulation Platform for Vehicular Safety Communications


Gaurav Bansal, John B. Kenney Aaron Weinfield
Toyota InfoTechnology Center, USA Denso International America Inc., USA
[gbansal,jkenney]@us.toyota-itc.com aaron_weinfield@denso-diam.com

Abstract—One of the most important challenges for DSRC using 30 radios with each transmitting at 50 Hz, using a
deployment is maintaining high performance under heavy technique described in Section II. The tests were conducted for
channel load. The study of congestion control mechanisms will be various data rates, message sizes, Enhanced Distributed
impractical if it requires hundreds of physical radios deployed on Channel Access (EDCA) parameters, channel bandwidths,
vehicles. In this paper we demonstrate two important results that
transmit powers, and numbers of emulated vehicles.
together suggest a strategy for investigating congestion control.
First, we show that the NS-2 simulator accurately models the
MAC and PHY associated with an IEEE 802.11 hardware Network Simulator 2 (NS-2) is a discrete event network
implementation. This opens the door to using simulation as the simulator, which is widely cited in wireless communication and
primary investigative tool. Second, we show that a technique in ad-hoc networking research. Version NS-2.33, released in 2008,
which one radio emulates N vehicles can produce results quite added improvements to make the simulator compatible with the
similar to the case in which a larger number of distinct vehicles IEEE 802.11p standard MAC and PHY protocols for Wireless
exist. This will allow selective validation tests with tens of physical Access in Vehicular Environments (WAVE) [10]. The key
radios emulating hundreds of vehicles. The means by which we modifications include cumulative SINR computation, capture,
demonstrate these results is a cross-validation technique. We
and more accurate modeling of the IEEE 802.11 CSMA/CA
show that three test approaches produce consistently similar
results: physical radios with emulation, NS-2 with emulation, and mechanism. In [15] the authors identify certain weaknesses of
NS-2 without emulation. The close agreement among these results NS-2, and they suggest some configuration options that might
validates both the radio emulation technique and the NS-2 allow it to perform on par with commercial simulator packages
implementation of the 802.11 MAC and PHY protocols. We use for the special case of fixed length packets. However, we are
this cross-validation technique to compare results for a wide not aware of any study comparing NS-2 with real DSRC radios.
variety of transmit power, message rate, data rate, channel
bandwidth, message size and EDCA settings. Testing with physical radios and testing with simulated radios
each have advantages compared to the other. For example,
Keywords-DSRC, NS-2, IEEE 802.11, WAVE, Radio Testbed
physical radios provide a more realistic measure of performance
with respect to channel fading, frame capture, and vehicle
I. INTRODUCTION mobility in real time. On the other hand, simulations are much
The US Federal Communication Commission (FCC) has more efficient when testing large networks with many system
allocated 75 MHz of spectrum in the 5.9 GHz band for variables. We believe there is a role for each in developing
Dedicated Short Range Communication (DSRC) among solutions for DSRC channel congestion, with the bulk of testing
vehicles, and between vehicles and roadside infrastructure [1, 2]. carried out via simulation and with selective use of radio testbed
The primary purpose is to enable safety applications that can experimentation.
prevent accidents. DSRC radio technology is standardized in
IEEE 802.11p [3, 4], IEEE 1609.3 [5], and IEEE 1609.4 [ 6]. We recreate a radio testbed topography in the NS-2 simulator.
When modeling M radios we use two techniques: a) M distinct
In [7, 8, 9] techniques have been proposed to perform simulated nodes, and b) 30 distinct nodes each of which
congestion control in the DSRC safety channel. However the emulates N = M/30 vehicles by transmitting at N times the
existing work does not include radio tests to demonstrate the vehicle message rate. We then address two issues. First, we use
severity of DSRC congestion. To study the congestion the simulator without emulation to validate the accuracy of the
performance, we configured a testbed with DSRC radios in a radio emulation technique in the testbed. Second, we use the
laboratory setting. The purpose of the tests was to model the radio tests to validate the accuracy of the NS-2 MAC and PHY
channel loading in a stationary environment in which a large model. In this way, we use the radios and simulator to
number of vehicles (around 200) are transmitting at a nominal cross-validate each other.
rate (10 Hz). Further, as it is difficult to assemble 200 actual
radios for experiments, we developed an emulation technique In this paper we show that performance results obtained by
where each physical radio is used to emulate N vehicles by NS-2 simulation, both with and without emulation, tend to
transmitting at a higher rate. For example, channel loading follow the radio results quite closely and in a consistent manner.
equivalent to 150 vehicles transmitting at 10 Hz was created by This implies that the goal of cross-validation is achieved, i.e. the

978-1-4244-8327-3/11/$26.00 ©2011 IEEE


NS-2 simulator gives a good estimate of the performance results AC1, AC2, AC3), channel bandwidth (10 MHz channel 172, 20
of actual radios and the physical radio emulation technique MHz channel 175), transmit power (10 dBm, 20 dBm), number
provides an accurate model of the congestion performance that of emulated vehicles (30, 60, 90, 120, 150, 180, 210, 240, 270,
can be expected from a large number of vehicles. These results 300, 330, 360) and emulated vehicle message rate (5 Hz, 10 Hz).
show that testing of congestion scenarios can proceed on two
complimentary fronts: a) reasonably scaled tests using tens of III. NS-2 SIMULATOR CONFIGURATION
physical radios can emulate the performance of hundreds of We attempted to configure the NS-2 simulator to model the
radios, and b) simulations can be used to assess various radio testbed environment as accurately as possible. Our
algorithms, protocols, and parameters without incurring the emphasis was on MAC collision modeling in the congested
complexity of a physical radio test each time. network regime. For the PHY, we found that the simple Rician
fading model works quite well. It has a dominant line-of-sight
The paper is organized as follows. In Section II, we describe the (LOS) component. The nodes in NS-2 were placed in the same
laboratory environment of the tests conducted with actual radios. physical topology as in the radio testbed. We implemented the
The NS-2 configuration is described in Section III. We present attenuation of radios at the transmitter side by decreasing the
comparative results of NS-2 simulations and radio tests in transmit power and at the receiver side by increasing the values
Section IV. Finally the paper is concluded in Section V. of carrier sense threshold and reception sensitivity threshold.
We implemented simulations with and without emulation. In an
II. LABORATORY TESTS CONDUCTED WITH FIELD RADIOS emulated scenario we had 30 nodes transmitting. For example,
The radio testbed was constructed by placing DSRC radios if each of the 30 nodes emulated five vehicles transmitting at 10
physically close to each other in an indoor office environment, Hz, the emulated node transmitted at 50 Hz. The corresponding
with 15 dual-radio DENSO wireless safety unit (WSU) systems non-emulated scenario used 150 nodes transmitting at 10 Hz,
on the periphery of a rectangular area roughly 20 feet by 10 feet, with five nodes co-located at each physical radio location. The
as shown in Fig. 1 [11]. While these laboratory tests were carrier sense threshold value for 6 Mbps data rate was set to -92
conducted in indoor environments, similar performance has dBm, and for 12 Mbps data rate it was set to -89 dBm. The
been reported for outdoor environments as well [12]. To better reception sensitivity threshold values are taken to be same as
model the normal vehicle density, each radio antenna was carrier sense threshold values. These values were all determined
configured with an attenuator between 0 dB and 50 dB with a by calibration tests with the radio testbed. The noise floor was
goal of distributing the Receive Signal Strength (RSS) over a set to -99 dBm in 10 MHz channel and -96 dBm in 20 MHz
range of -30 to -95 dBm at each receiver. Attenuating a radio channel. The preamble capture feature has been enabled with a
antenna affects both transmit and receive paths. Therefore, a SINR capture threshold of 4 dB.
few radio pairs with high amounts of attenuation were
completely out of range (i.e. “hidden”) from each other, but still Some of the differences between the radio testbed and the NS-2
able to communicate with radios having lower attenuation. In simulator were as follows:
each dual-radio WSU both radios were configured to transmit 1) Generation Time: This subsection describes how
and only one radio was configured to receive, so there were a messages are generated in the testbed and in NS-2. In NS-2
total of 28 x 15 = 420 transmitter-receiver pairs, or links without vehicle emulation, when a vehicle generates a message
(excluding the radios in the same WSU from the calculation). at time t it also schedules its next message at time t+T. If the
Of these, 8 links were classified as out of range. A histogram of transmit rate is X Hz, then T is a uniform random variable over
the link RSS values is shown in [11]. the interval [1/2X, 3/2X]. In NS-2 with vehicle emulation, 30
simulated nodes each emulate N vehicles transmitting at X Hz.
WSU 7 WSU 8 When a node generates a message at time t it also schedules its
next message at time t+T, where T is a uniform random variable
WSU 4 WSU 5 WSU 6 WSU 9 WSU 10 WSU 11
~ 8 feet
over the smaller interval [1/2NX, 3/2NX]. This is illustrated in
the upper portion of Fig. 2. In the testbed a radio always
WSU 3
~ 10 feet WSU 12 emulates N vehicles at X Hz. Each radio generates N messages
WSU 2 WSU 13
in each fixed interval of duration 1/X seconds, and each message
~ 20 feet
generation time is chosen independently and uniformly over the
WSU 1 WSU 15 WSU 14 entire 1/X interval. This is illustrated in the lower portion of Fig.
2. Note that when using the emulation technique it is not
possible for two messages from the same node to collide. The
Fig 1: Physical WSU Layout in the Laboratory Testbed, each results in Section IV, however, show this to be a minor factor.
WSU has two DSRC radios 2) EDCA: The EDCA mechanism is supported in 802.11
chips in the test radios, but is not fully implemented in NS-2.
A total of 101 tests were conducted for various combinations of We simulated different EDCA parameters, but restricted each
the following parameters: data rate (6 Mbps, 12 Mbps), message in a given simulation to use the same access category
over-the-air (OTA) message size (300 bytes, 378 bytes, 464 (AC). We set the NS-2 parameter CWMin to the desired CWMin
bytes), EDCA parameters (both IEEE 1609.4 trial-use [13] and value for the AC. Similarly, we set the NS-2 DIFS parameter to
IEEE 802.11p [4] defaults for access category indices AC0,
the desired AIFSN value. We observe in the simulations that TABLE I. DEFAULT PARAMETERS FOR DSRC RADIO TESTS

our results tend to match radio results more tightly in the cases
of IEEE 802.11p default AC2 (CWMin = 3, AIFSN = 3) and AC3 Parameter Value
(CWMin = 3, AIFSN = 2). The match between simulations and
Power 20 dBm
radio tests is still good, but somewhat looser in the cases of AC0
(CWMin = 7, AIFSN = 6) and AC1 (CWMin = 15, AIFSN = 9). It EDCA mechanism AC0 IEEE 1609.4 trial version
may be that our modelling of EDCA by configuring CWMin and OTA message size 378 Bytes
DIFS is a better approximation for AC2 and AC3 than it is AC0
and AC1. We expect that when EDCA is fully implemented in Data rate 6 Mbps
NS-2 (rather than just configuring the DIFS and CWMin Channel 172 (10 MHz channel)
parameters), the results will match more tightly for all EDCA
Message rate 10 Hz
mechanisms.

Fig. 3 illustrates the basic results using the default parameters in


NS-2 Emulation: Divide 100 msec into N sub-intervals; one packet Table I. Results are shown for two RSS bins: -45<RSS<=40
per sub-interval 100 msec Message Interval dBm, and -85<RSS<=80 dBm. As expected, due to the fading
effect PER is higher for the low power bin (-85 to -80 dBm) and
lower for the high power bin (-45 to -40 dBm). Within each
power bin, PER increases with the number of vehicles
100/N msec subinterval
(emulated or individually simulated). This is consistent with the
Each packet could collide with any of 29*N packets from other higher frame collision probability that is associated with higher
radios in 100 msec
channel load [11].
Radio Testbed Emulation: Send N packets per 100 msec. The most important observation from Fig. 3 is that the three
Each packet time uniform over 100 msec, independent of others
100 msec Message Interval
curves associated with a given parameter combination and a
Do not enter back-off until prior packet sent; no self-collision given power bin tend to agree with each other quite closely.
This implies two things. First, it implies that the NS-2 simulator
platform accurately models the MAC and PHY of actual radios.
Each packet could collide with any of 29 packets from other radios in
Second, it implies that the emulation technique used in the radio
100/N sub-interval testbed leads to approximately the same frame collision
probability as would be observed without emulation.
Fig 2: Emulation Techniques employed in the testbed and NS-2,
assuming 30 radios and each emulating N=6 radios. Generally, the NS-2 results without emulation show slightly
higher PER than the NS-2 results with emulation. This is
IV. RESULTS intuitive; when a node emulates multiple vehicles, the frames
transmitted by those vehicles cannot collide with each other, so
The default parameters for the radio and simulation tests are
shown in Table I. These are representative of defaults used in the emulation technique leads to lower PER. The NS-2 results
without emulation might also be expected to show higher PER
industry testing [14].
than the radio testbed results (with emulation). In our tests,
however, the radio testbed PER is slightly but consistently
For each combination of parameters, the number of emulated
vehicles per node was varied from N = 1 to 6. For the 5 Hz higher than the simulation PER. This might reflect a small
unmodeled source of error in the radio tests, e.g. multipath
message rate tests the number of emulated vehicles was
fading slightly worse than the Rician model, or sensitivity
extended to include N = 7 to 12. In general the tests consisted of
slightly worse than the modeled value. Even with this caveat,
varying one variable at a time from the default set. The primary
performance metric is packet error rate (PER), defined as the however, the agreement among the three test results for a given
power bin is quite high, indicating that our goal of
ratio of packets not received correctly to total packets
cross-validation is achieved in the default case.
transmitted. PER can be measured over one or more links and
over a variety of time intervals. In these tests PER is reported for
the set of links with RSS within a 5 dB range, e.g. -80 to -85 Figs. 4-9 demonstrate the same general trends and relationships,
reflecting a series of tests in which one default parameter is
dBm, and the first 30 seconds of each test are excluded from the
modified. In all cases the agreement among the three test
results to avoid transient conditions. Graphical results are
approaches remains quite high. Fig. 4 represents the case where
presented in the Appendix. Each graph includes PER results for
one or more 5 dB RSS bins. Within each bin the graph transmit power is reduced from 20 dBm to 10 dBm, which is
often considered a lower bound for DSRC safety
compares three tests with identical parameters: radio testbed
broadcasts. Fig. 4 also includes an extra bin (-65<RSS<=60
(with emulation), NS-2 simulation with emulation, and NS-2
dBm) to illustrate performance at medium power levels. Figs. 5
simulation without emulation.
and 6 show the effect of modifying the OTA message size from
the default 378 bytes, to 300 and 464 bytes, respectively. DSRC
safety message sizes vary depending on options related to REFERENCES
security and message content; the sizes tested reflect the [1] US Federal Communications Commission, R&O FCC 03-324,
approximate minimum and maximum expected in “Dedicated Short Range Communications Report and Order”, Dec. 17,
deployment. Fig. 7 illustrates a reduction in the message rate to 2003.
5 Hz per vehicle. Note that in the NS-2 simulation with [2] D. Jiang, V. Taliwal, A. Meier, W. Holfelder, and R. Herrtwich, “Design
of 5.9 Ghz DSRC-based Vehicular Safety Communication”, in IEEE
emulated vehicles, a test using 10 Hz message rate and M Wireless Communications, vol. 13, no. 5, pp. 36-43, Oct. 2006.
vehicles is identical to a test using 5 Hz message rate and 2M [3] IEEE Std. 802.11-2007, “Wireless LAN MAC and PHY Specification”.
vehicles. For NS-2 simulations without emulation, there are [4] IEEE 802.11p, July 2010, “Amendement to Wireless LAN MAC and
small differences between comparable 10 Hz and 5 Hz tests; this PHY Specification”.
is true of the radio tests as well. Fig. 8 shows test results [5] IEEE Std. for Wireless Access in Vehicular Environemnts-Networking
obtained using a 20 MHz channel with 12 Mbps data rate (i.e. Services, IEEE Std 1609.3-2010, Dec. 2010.
the modulation and coding rate remain QPSK and ½, [6] IEEE Std. for Wireless Access in Vehicular Environments-Multi-Channel
respectively). Finally, Fig. 9 shows the effect of using the Operations, IEEE Std 1609.4-2010, Dec. 2010.
higher priority EDCA access category AC3. We tested every [7] C.-L. Huang, Y. P. Fallah, R. Sengupta, and H. Krishnan, “Adaptive
Intervehicle Communication Control for Cooperative Safety Systems”, in
access category in trial use 1609.4 (now obsolete) and IEEE Network, vol. 24, no. 1, pp. 6-13, Jan.-Feb. 2010.
802.11p. The results for AC0 (Fig. 3) and AC3 (Fig. 9) are [8] M. Torrent-Moreno, J. Mittag, P. Santi, and H. Hartenstein,
representative. “Vehicle-to-Vehicle Communication: Fair Transmit Power Control for
Safety-Critical Information”, in IEEE Transactions on Vehicular
Technology, vol. 58, no. 7, pp. 3684-3703, Sep. 2009.
In every case there is close agreement among the results of the
[9] A. Weinfied, J. Kenney, and G. Bansal, “An Adaptive DSRC Message
three test approaches, suggesting we have achieved our Transmission Interval Control Algorithm”, ITS World Congress 2011,
cross-validation goal for these non-default cases as well. The Oct. 2011, Accepted for Publication.
trends in this group of figures are mostly as expected, e.g. all [10] Q. Chen, F. Schmidt-Eisenlohr, D. Jiang, M. Torrent-Moreno, L.
other things being equal, PER increases with decreasing Delgrossi, and H. Hartenstein, “Overhaul of IEEE 802.11 modeling and
transmit power, with increasing message size, with increasing simulation in ns-2”, Proceedings of the 10th ACM Symposium on
Modeling, analysis, and simulation of wireless and mobile systems,
message rate, and with decreasing channel bandwidth. The October 2007.
latter result might not be obvious, since the advantage of shorter [11] A. Weinfield, “Methods to Reduce DSRC Channel Congestion and
packet transmission times is partly offset by the higher noise Improve V2V Communication Reliability” in ITS World Congress,
floor associated with the 20 MHz channel. Since most losses in Busan, Korea, 2010.
these tests are due to collisions, the shorter packet time appears [12] US Department of Transportation, National Highway Traffic Safety
Administration, “Vehicle Safety Communications-Applications VSC-A,
to be a more important factor. The performance comparison Draft Final Report, Appendix I – Multiple OBE Scalability Testing
between AC0 and AC3 is also not obvious. AC3 is a higher Results”.
priority category which uses a shorter contention window that [13] IEEE Std 1609.4-2006, “IEEE Trial-Use Standard for Wireless Access in
might lead to more collisions. We might expect AC3 to perform Vehicular Environments (WAVE)-Multi-channel Operation”.
worse, but in fact AC3 has lower PER. This is a complex [14] US Department of Transportation, National Highway Traffic Safety
question to analyze and we are investigating it as future Administration, Rep. DOT HS 811 073, “Vehicle Safety
Communications-Applications VSC-A, First Annual Report”, Jan. 2009.
research.
[15] M. Takai, Y. Owada, and K. Seki, “A Comparative Study on Network
V. CONCLUSIONS Simulators for ITS Simulation IEEE802.11 Medium Access Control
(MAC) Models” in ITS World Congress, Stockholm, Sweden, 2009.
In this paper we demonstrate two important results. First, we
show that for an indoor office environment where LOS is APPENDIX
available, the NS-2 simulator accurately models the MAC and
PHY associated with an actual 802.11 implementation. Second,
we show that a technique in which one radio emulates N
vehicles can produce results quite similar to the case in which a
larger number of distinct vehicles exist. Collectively, these
results suggest a two prong strategy for investigating methods
of controlling DSRC channel congestion: 1) use simulation as
the primary tool for testing a wide variety of algorithms and
parameters, and 2) use the emulation technique in radios so that
tens of nodes can model hundreds of vehicles, and perform
selective tests to validate the effectiveness of the simulated
approaches. Among many potential extensions of this work,
we consider a study of EDCA effects, based on an accurate
NS-2 EDCA implementation, to be the most
important. Outdoor testing is also important, though the close
agreement between our indoor radio tests and prior outdoor
tests is encouraging.
Fig 3: PER vs emulated vehicles for default parameters.
Fig 4: PER vs emulated vehicles when transmit power is 10 dBm Fig 7: PER vs emulated vehicles when message rate is 5 Hz

Fig 8: PER vs emulated vehicles while using channel 175 (20 MHz bandwidth)
Fig 5: PER vs emulated vehicles when OTA message size is 300 Bytes

Fig 6: PER vs emulated vehicles when OTA message size is 464 Bytes Fig 9: PER vs emulated vehicles when EDCA mechanism is AC3 (trial version)

You might also like