Professional Documents
Culture Documents
Abstract—The IEEE 802.11p standard (or Dedicated Short • Multipath components are considerably longer in
Range Communication, DSRC) [1] has been proposed to be the outdoor channels than they are in indoor channels. This
standard for vehicle-to-vehicle (V2V) and vehicle-to- increases the likelihood of inter-symbol interference
infrastructure (V2I) communications. Since the IEEE 802.11a- (ISI);
based DSRC standard is not fully customized for outdoor, • The V2V channel is highly dynamic. There are moving
highly mobile channels, the performance of DSRC standard
might degrade in a challenging V2V channel. In this paper, we
reflectors (such as other cars and buses) in addition to
develop several equalization schemes that are able to closely large stationary reflectors (such as buildings, hills, trees
track the V2V channel dynamics and thus improve and signs);
performance at the physical layer. Through a set of empirical • Both the transmitter vehicle and the receiver vehicle can
experiments, we show that the performance (in terms of Packet be moving with respect to the environment and each
Error Rate) can be significantly improved from 41% (using a other. These motions significantly affect the Doppler
simple Least Square Estimator) to 19% (using a Spectral spectrum, and hence the fading properties of the
Temporal Averaging Estimator). channel.
Because of these factors, it has been shown by a large body
I. INTRODUCTION of research literature that the V2V channel changes
543
demodulation) while only changing the actual channel
estimation/equalizer blocks. This enables a direct (1)
comparison of different equalization schemes. Fig. 1
illustrates our software-based receiver. and
. (2)
Because the bits in the two training symbols are equal, their
N-point DFTs are equal, i.e., X1(k) = X2(k) = X(k). The LS
.
2
(3)
Fig. 1. Our software-based receiver for testing equalization schemes with
actual recorded waveforms. The data in the packet is then equalized using this channel
estimate. For a given received symbol sR[n], the symbol is
IV. CHANNEL ESTIMATION AND EQUALIZATION
first demodulated using the DFT:
.
TECHNIQUES
The IEEE 802.11p standard requires four evenly spaced (4)
pilot subcarriers that are used to track residual frequency
offset between the transmitter and receiver. These pilot The received DFT vector is then equalized such that the
!
estimate of the transmitted data is
" .
subcarriers are unable to adequately sample the dynamic
channel variation in the V2V setting. In the IEEE 802.11p (5)
standard, the channel estimation is accomplished by sending
two identical training symbols at the beginning of a packet. Note that this is a simple, one-tap equalizer for each
The drawback of this method is that the channel will change subcarrier in the frequency domain. This procedure is
considerably over the duration of a packet, thus making the repeated for all symbols in the packet. It is clear that if the
initial channel estimate obsolete. Therefore, an appropriate channel changes significantly over the duration of a packet,
method for tracking the channel and updating the channel the channel estimate H(k) no longer accurately represents
estimate is desired. the channel, and equalization could actually distort the
received signal rather than correct it. Thus, an accurate and
A. Least Squares Estimator
efficient means of tracking the channel is crucial.
As a benchmark, we use a simple Least Squares (LS)
estimator [14] that only uses the preamble at the beginning
of a packet to estimate the channel and subsequently
equalize future symbols. This basic equalizer also uses the
pilots in the packet to correct for residual frequency offset,
which is a side effect of non-perfect frequency offset
estimation from the training preamble. This equalizer
represents the most basic equalization possible, which would Fig. 2. The packet structure of IEEE 802.11p [1].
be typical of a commercial 802.11a card. The benefit of this
technique is that it is straightforward and easy to implement B. Comb Pilot Interpolation
in hardware. Therefore, it is used as a benchmark The IEEE 802.11p standard uses subcarriers numbered
equalization technique. from -26 to 26. The subcarrier at zero frequency (DC) is not
The packet structure of 802.11p is shown in Fig. 2. Here, used. The standard allocates the use of four pilot subcarriers
the first ten short symbols (t1 through t10) are used for (located at subcarriers -21, -7, 7, and 21), which are to be
synchronization, and are each 1.6 µs long. The two 6.4 µs used for frequency offset tracking. This means that there are
symbols that follow this (T1 and T2) are identical and are 48 subcarriers used for data transmissions, with the unused
used for estimating the channel. Because T1 and T2 are subcarriers at -32 through -27 and 27 through 31 used as a
identical, they make use of a longer guard interval (GI2) of guard band.
length 3.2 µs. After the training section of the packet, the In some OFDM schemes, a grid of pilots spaced in both
data symbols are sent. Each symbol contains a guard interval time and frequency allow the channel estimator to obtain
(GI) of length 1.6 µs and a data portion of length 6.4 µs. The feedback from the channel so as to equalize the signal.
first data symbol contains the SIGNAL field, which contains However, to capture the variation of the channel in both time
information about the modulation rate and the packet length. and frequency, the pilots must be spaced such that they
fulfill the Nyquist criterion for sampling the channel [15]. In
For LS channel estimation, first the time domain symbols the 802.11p standard, the pilots used for frequency offset
T1[n] and T2[n] are extracted. Then their N-point Discrete correction are spaced by 14 subcarriers (2.1875 MHz) in the
Fourier Transforms (DFT) are computed as follows. frequency domain, which is inadequate for channel
estimation because the V2V channel’s 90% coherence
544
bandwidth can be smaller than 410 KHz 10% of the time [4]. subcarriers. Estimating the channel using this method is
However, some information about the channel is better than inherently less reliable because such estimation assumes that
no information. In this case, it is better to update the channel the data was demodulated correctly. However, we can
estimate with limited feedback rather than to rely on the design our receiver to use several approaches to combat this
channel estimate from the packet preamble that may be unreliability:
outdated. • Redundancy: we use several data subcarriers to make a
In comb pilot interpolation, each symbol is demodulated as single measurement at a given subcarrier. This increases
in (4). Then, the received values in the frequency domain at the probability that correct channel feedback is at least
each of the pilot subcarriers are extracted. These values are partially obtained.
designated by the 4-element vector Yp. The known DFT • Averaging: By averaging multiple measurements, we
values of the sent pilots at these subcarriers are specified by combine them into a single channel measurement,
the standard, and are given by the 4 element vector Xp. The reduce the effects of measurement noise, and reduce the
#
LS estimate at each pilot subcarrier is then formed as
# .
effects of erroneous channel measurements.
#
(6) • Conservative update rate: If we update our channel
estimate slowly, we put more weight on previous
This gives a four element vector that represents evenly measurements and do not make drastic changes to the
spaced estimates of the channel. To interpolate these channel estimate that we use for symbol equalization.
estimates, first endpoints are appended to the vector to This combats erroneous channel estimates, because an
! ",) ,)
After the channel is estimated at a given symbol, the updated), giving
.
)
overall channel estimate is appropriately updated so as to
1 1
(10)
track the channel. The new estimate at time t is given by
) *1 , . )
/#01)2 .
- -
(8) Once this is done, then the channel estimate at a subcarrier k
,)
may be formed from the bit decision at this subcarrier, or
,
Here, α is a memory parameter. A larger α implies longer
memory. Note that H0 is the initial channel estimate given (11)
by (3). Once the channel is estimated, the symbol at time t is where H(k) is the channel estimate at a data subcarrier, and
,)
!
equalized
",) .
X(k) is the decided symbol value, i.e., the complex value in
)
(9) the constellation plot corresponding to the decided bit(s).
In general, the copilot channel estimate is formed as a
This procedure continues until all symbols in the packet linear combination of channel estimates of the data
have been equalized. subcarriers in its vicinity. The copilot channel estimate at
6
subcarrier λ is formed as
C. Comb Co-Pilot Interpolation
545
a copilot needs to be formed outside the subcarrier range, a waveform with a DFT, the resulting complex numbers at
copilot at -26 or at 26 is used instead at this location (this is each subcarrier position map directly to the points in a
to maintain equal distances between pilots and copilots, constellation pattern. A sample constellation pattern is
which is necessary for the interpolation). Finally, if a copilot shown in Fig. 4. In (a), all of the demodulated data points for
needs to be formulated at the zero subcarrier location, the a packet with no errors are shown. In this case, all of the data
Fig. 4. The BPSK constellations in the complex plane for (a) a packet with
no errors, and (b) a packet with errors. Here, a 1 maps to a binary 1 and a -1
maps to a binary 0.
we increase the weights for copilot terms that include pilot (16)
subcarriers. < <
1
where the mean mP of the constellation is given by
%M ;< .
D. Constellation-Aware Data Equalization (CADE)
< <
subcarriers with a low probability of being correct can be
1 , %O ,
(18)
neglected while forming the channel estimate. This method
can be used to formulate a more robust channel estimate.
where mD is the mean of the data points D in the symbol of
KGDN 1 , 0 1.
The foundation of this method relies on the use of BPSK interest. In theory, mD is equal to 0, and
on all data subcarriers, but could be extended to other (19)
constellation schemes. When demodulating the OFDM
546
However, because the sample size is only n=48 points, the Based on these PDFs, thresholds ±q can be developed with
variance of D for a given symbol may not be equal to 1. an associated confidence C that serves as a measure of the
This is because the variance of D is only equal to 1 if there likelihood that the point of interest is correctly demodulated
are equal numbers of 1s and 0s transmitted in a given (from region P0) rather than incorrectly demodulated (from
;
symbol. If this is not the case, then the term E(D) = mD will region P1). Here, the confidence parameter is defined as
[ .
; ;
1
deviate from 0. The mean of D is given by
%O >< .
(26)
],1
(20)
^
It can be shown that
<
Z
[ \], Z .
],1 ]1
Because Di is modeled as a Bernoulli random variable, the
^ ^
Z Z
distribution of mD is a Binomial. Since Di takes on values -1 (27)
and +1 with probability 0.5, we can show that
; *%O . P Q 0.5 ,
Thus, given a desired confidence value C' and the calculated
(21) σN, one can solve for the threshold q. This is done
(28)
* . 0.5 , S 0
is found to be
2S
It should be noted that q may be solved offline using a given
/2 J, (22)
; R%O T U
value of C' and incremental values of σN2. Then, lookup
2* . , S 1,2, … 24
tables are generated and are embedded in the receiver.
/2 , S
Therefore, this algorithm maintains a low computational
complexity, as the only parameter that needs to be calculated
where c takes on values from the set {0, 1, … , 24}. during demodulation is σN2.
Next, the variance of the constellation points can be When a packet is received, it is demodulated and
KGDL KGDY N KGDY KGDN.
written as equalized with the current channel estimate. Like the other
(23) schemes previously discussed, each symbol is used to update
KGDY Z KGDL , KGDN.
Solving for σN2, we obtain the channel estimate. An initial version of this channel
(24) estimate update is determined by using the comb pilot
1
Now we may substitute (16) and (18) to obtain interpolation. The variance of the noise (σN2) of the
Z ;< , %M , 1 %O .
547
and the scheme reduces to comb pilot interpolation. These averaged CADE approach the PER performance of the fixed
results can be seen in Fig. 6. threshold (q=0) schemes as the confidence parameter
decreases. These observations led to the conclusion that the
best approach is to use all data subcarriers to formulate a
channel estimate at each symbol, and then to perform
averaging in both time and frequency to improve
performance. This method is formalized in the next section.
E. Spectral Temporal Averaging
The so called Spectral Temporal Averaging (STA)
estimation can be formalized as follows. First, the initial
channel estimate is obtained from the training preamble as in
(3). This initial estimate is applied to the first symbol in the
packet. Once this symbol is demodulated, a channel estimate
,<
is formed:
< ,
<
Fig. 6. Packet error rate results for CADE at different moving average (29)
parameter values.
where Xi is the decided constellation at symbol i, SR,i is the
demodulated subcarrier values at symbol i, and Hi is the
estimate formed at symbol i.
This estimate is then averaged in frequency. The average
is constructed as a simple moving average, such that the
6
estimate at subcarrier λ is formed as
∑86
6 5 1. It should be noted that nonexistent
affect how many terms are included in the average, and
Fig. 7. A comparison of CADE and averaged CADE. These plots were subcarriers are excluded from this averaging operation. For
generated by using the best performing moving average parameter (in a example, for β=3 and subcarrier 26, only subcarriers 23, 24,
PER sense) at a given confidence parameter. 25, and 26 are used in the average, and the weights are
adjusted accordingly. Additionally, the value of Hi(0) is
Another problem with CADE is that making an error in replaced with an average of the subcarrier -1 (Hi(-1)) and the
estimating the channel at a data subcarrier location is very subcarrier 1 (Hi(1)) as there is no data transmitted on the
costly. If an error is made, it can propagate in the channel zero subcarrier.
estimate for the duration of the packet. To avoid this, the use After the frequency averaging is computed for all 52
1 1
of averaging in frequency was explored. Averaging adjacent subcarriers, the channel estimate is updated using
e"f,) *1 , . e"f,)
/#01)2 ,
- -
subcarrier estimates helps to reduce the effect of errors. The
intuition behind this is to have multiple measurements of the (31)
channel at a given subcarrier location. When more data where α is a moving average parameter in time. Note that
subcarriers are used to form the average, the accuracy HSTA,0 is the initial channel estimate obtained from the
increases, because the probability of making multiple errors preamble estimation. The estimate in (31) is then applied to
over a block of subcarriers is smaller than the probability of the next symbol’s equalization, and the process is repeated
making an error at a single subcarrier. However, if too many until the packet is completely demodulated.
data subcarriers are used to form the average, then the
resulting estimate will not be as accurate, because the V. RESULTS
coherence bandwidth of the channel may be less than the
range of average subcarriers. We now validate the schemes described in section IV
Fig. 7 shows the result of frequency averaging the CADE using the methodology in section III. We do not show results
scheme. First, CADE is plotted with no averaging versus for CADE, because STA was developed by optimizing
CADE with averaging. For the CADE scheme, there is a CADE.
performance spike (valley in PER) at the optimal confidence First, we present the overall results from the highway and
parameter. At this same optimal confidence parameter, rural environments in Table 1. Here, we show the results for
performance of the averaged CADE scheme is actually the best performing schemes after testing various scheme
degraded by an increase in PER; in fact, the averaged CADE parameters (i.e. α, β, and L as applicable).
scheme performs better when using a lower value for C' (and Next, we analyze the PER against different environmental
therefore, more data subcarriers). When plotted, CADE and variables. For these plots, only packets with valid GPS data
are used. Points are binned with each bin containing 229
548
points. For highway, this results in 17 bins of 229 points and ground reflection between the two vehicles seen in the
one bin of 227 points. For the rural environment, this results highway environment. The STA scheme has significantly
in 6 bins of 229 points and one bin of 43 points. We first better performance here (28% PER) as opposed to the LS
show the signal to noise ratio (SNR) vs. PER for both the scheme (66% PER). As expected, performance suffers with
highway and rural environments in Fig. 8 and Fig. 9, increased distance. Unlike the highway environment, we see
respectively. For PER = 10%, we see that STA achieves the a linear relationship between distance and PER for the rural
same performance as the LS scheme at an SNR that is 7.6 environment.
dB less and 7.5 dB less for the highway and rural cases,
respectively.
TABLE I
OVERALL PACKET ERROR RATE RESULTS
Highway Rural
Scheme (4,120 packets) (1,417 packets)
LS 41.21% 41.92%
Comb Pilot 25.53% 26.53%
α=12 α=6
Comb Copilot 22.48% 21.38%
L=3, α=4 L=3, α=4
STA 19.64% 19.62%
β= 2, α= 2 β= 2, α= 2
i
j k\k l \ Δ\m\
j
bandwidth. The coherence bandwidth is computed as
gΔ\ ,
i
j k\k l \m\
j (32)
549
have been developed clearly help to track the channel better,
and as a result have smaller increases in BER over the
duration of the packet. The data aided techniques (comb
copilot and STA) exhibit a linear increase in BER. Comb
copilot interpolation eventually has a higher BER than comb
pilot interpolation. The better PER performance of comb
copilot interpolation can be explained by the fact that the
first half of the packet experiences a lower BER than the
second half of the packet. This behavior is more easily
corrected than a flatter error rate curve (comb pilot
interpolation) by leveraging 802.11p’s interleaving, coding,
and scrambling mechanisms. The understanding and
improvement of these error correction mechanisms with
different BER distributions is a topic for future research.
Clearly, spectral temporal averaging outperforms the other
schemes.
Fig. 12. 90% Coherence Bandwidth vs. PER for highway case.
Fig. 14. BER plotted against subcarrier number for the different schemes for
Fig. 13. 90% Coherence Bandwidth vs. PER for rural case. the highway scenario.
550
is insufficient for combating the V2V channel effects. Communication (DSRC) Frequency Band,” IEEE Journal on Selected
Areas in Communications, vol.25, no.8, pp.1501-1516, Oct. 2007.
Because the standard does not provide a sufficient means of
[14] Engels, M., Wireless OFDM Systems - How to Make Them Work?,
pilot signal feedback, data must be used to update the Springer-Verlag, 2002.
channel estimate. We have developed several enhanced [15] Van Nee, R.; Prasad, R., OFDM Wireless Multimedia
equalization schemes using the existing pilot subcarriers or Communications, Boston: Artech House, 2000.
using the pilot subcarriers in combination with data
subcarriers. We have tested these schemes on real V2V
waveforms and have shown that adding more reliance on
data to aid in channel estimation improves the packet error
rate, assuming the scheme is well designed (i.e., employs
frequency averaging with reasonable averaging parameters).
The spectral temporal averaging outperforms our other
schemes, and decreases PER significantly. Additionally,
these equalization methods are receiver-centric and will not
require a change to the 802.11p standard.
REFERENCES
[1] "IEEE Draft Standard for Information Technology, IEEE Unapproved
Draft Std P802.11p /D7.0, May 2009,
URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=5174144
&isnumber=5174143, Accessed Nov. 2009.
[2] Oteri, O.; Wang, X.; Mujtaba, S.A.; and Paulraj, A., “Time and
Frequency Equalization in 802.11a/g,” Conference Record of the
Thirty-Seventh Asilomar Conference on Signals, Systems and
Computers, vol.1, pp. 698-702, Nov. 9-12, 2003.
[3] Alexander, P.; Haley, D.; and Grant, A., “Outdoor Mobile Broadband
Access with 802.11,” IEEE Communications Magazine, vol. 45, no.
11, pp.108-114, Nov. 2007.
[4] Cheng, L.; Henty, B.E; Cooper, R.; Stancil, D.D.; Fan Bai, “A
Measurement Study of Time-Scaled 802.11a Waveforms Over the
Mobile-to-Mobile Vehicular Channel at 5.9 GHz,” IEEE
Communications Magazine, vol. 46, no. 5, pp. 84-91, May 2008.
[5] Belotserkovsky, M., “An Equalizer Initialization Algorithm for
IEEE802.11a and HIPERLAN/2 Receivers,” IEEE Transactions on
Consumer Electronics, vol.48, no.4, pp. 1051-1055, Nov 2002.
[6] Kella, T., “Decision-Directed Channel Estimation for Supporting
Higher Terminal Velocities in OFDM Based WLANs,” Global
Telecommunications Conference, 2003. GLOBECOM '03. IEEE ,
vol.3, pp. 1306-1310, Dec. 1-5, 2003.
[7] Yong-Hua Cheng; Yi-Hung Lu; and Chia-Ling Liu, “Adaptive
Channel Equalizer for Wireless Access in Vehicular Environments,”
6th International Conference on ITS Telecommunications
Proceedings, pp.1102-1105, June 2006.
[8] Lee, S.; Kao, Y.; and Chen, H. “Performance of a Robust Inner
Receiver with Frequency Domain LMS Equalizer for DSRC systems,”
Proceedings of the 2006 International Conference on Wireless
Communications and Mobile Computing, pp. 985-990, July 3-6, 2006.
[9] Abdulhamid, H.; Abdel-Raheem, E.; Tepe, K.E., “Channel Tracking
Techniques for OFDM Systems in Wireless Access Vehicular
Environments,” 9th International Symposium on Signal Processing
and Its Applications, pp.1-4, Feb. 12-15 2007.
[10] Zhang, Y.; Tan, I. L.; Chun, C.; Laberteaux, K.; and Bahai, A., “A
Differential OFDM Approach to Coherence Time Mitigation in
DSRC,” Proceedings of the Fifth ACM International Workshop on
Vehicular inter-Networking, pp. 1-6, Sept. 15, 2008.
[11] Woong Cho; Sang In Kim; Hyun kyun Choi; Hyun Seo Oh; Dong
Yong Kwak, “Performance Evaluation of V2V/V2I Communications:
The Effect of Midamble Insertion,” 1st International Conference on
Wireless Communication, Vehicular Technology, Information Theory
and Aerospace & Electronic Systems Technology (VITAE), pp.793-
797, May 17-20, 2009.
[12] Sibecas, S.; Corral, C.A.; Emami, S.; Stratis, G.; Rasor, G., “Pseudo-
Pilot OFDM Scheme for 802.11a and R/A in DSRC Applications,"
IEEE 58th Vehicular Technology Conference (VTC), Fall 2003, vol.2,
pp. 1234- 1237, Oct. 6-9, 2003.
[13] Cheng, L.; Henty, B.E.; Stancil, D.D.; Fan Bai; and Mudalige, P.,
“Mobile Vehicle-to-Vehicle Narrow-Band Channel Measurement and
Characterization of the 5.9 GHz Dedicated Short Range
551