Vous êtes sur la page 1sur 5

Uneven Pilot-Assisted Channel Estimation Method

for Mobile OFDM Systems


D.Li, Sl.Feng, W.Ye Hc.Zhuang
Dept. of Electronic Engineering Pre-research Dept.
South China Univ. of Technology Huawei Technologies CO. Ltd
Guangzhou, Guangdong, 510640, China Shenzhen, Guangdong, 518100, China
lidan_1121@163.com zhc@huawei.com

Abstract—This paper proposes an uneven pilot-assisted channel maximal vehicle speed, we use the practical linear fashion to
estimation (CE) technique for OFDM sysetms in mobile approximate CIR variations. We find that the theoretical
environments. We find that the prerequisites of most previous researches using the comb pilots and linear fashion to
studies are using an equal spaced pilots pattern and sample- approximate time domain CIR variations[9-11] are mostly
spaced channels. This is not the case in practical systems. based on requierments as follows: (1) Pilots must distributed
According to conventional pilot-assisted estimators, we develop evenly in whole frequency band; (2) No power leakage in
an improved channel estimation method based on the criteria of multipath fading channels. In practical systems, many factors
Lesat Square (LS) for IEEE802.16e OFDM systems. In this should be taken into consideration. Such as the uneven pilot
practical system, pilots are not evenly inserted in the frequency
distribution and error floor caused by CIR leakage due to the
band and the non-sample-spaced multipath delay leads to
necessary guard bands and the unsampled multipath delays [8].
channel impulse response (CIR) leakage in the time domain. The
proposed uneven pilot-assisted CE method is conducted in the Significant performance reductions [7] will be caused when
frequency domain and it uses a linear model to approximate the the requirements are not satisfied. Based on analysis above,
time-varying characteristic of CIR in the frequency domain we develop a feasible uneven pilot-assisted CE method, which
during each OFDM symbol. Analysis and simulations is conducted in the frequency domain using the linear model
demonstrate its low computational complexity, good to approximate frequency domain CIR variations.
performance and practicability in actual OFDM systems. In the rest of the paper, we first present the OFDM system
Keywords-channel estimation; uneven pilot; LS; linear model;
model of time-varying channles used in the paper. Besides, ICI
time-varying channel
analysis are given in Section II. In Section III, we address the
I. INTRODUCTION problems of conventional methods. Based on the analysis
Orthogonal frequency division multiplexing (OFDM) is above, an improved un-even pilot assisted CE method is
one of the most important physical layer transmission developed. Simulation results are presented in Section IV.
techniques in wireless communications. In OFDM, data are Finally, conclusions and future research works are stated.
transmitted in parallel over a large number of orthogonal
subcarriers. OFDM is robust to multi-path channels for its Basic notations: italic letters to denote variables, 䯴 .䯵 H for
lower symbol rate. However the increased symbol duration Hermitian transpose, 䯴 .䯵* for complex conjugat, ⊗ for
makes it more susceptible to time-variations of channels. In circular convolution.
high mobility environments, CIR varies severely in a certain
period of time. Inter-carrier interference (ICI) destroys the II. SYSTEM MODEL AND ICI ANALYSIS
orthogonality. Accordingly, OFDM systems encounter
performance degradations in such time-varying channels. As a A. System Model
result, CE techniques in time-varying scenarios are critical .
We consider an N-point OFDM system, Ts denotes one
There has been a large body of works devoted to channel symbol period, the Doppler frequency shift f d is normalized
estimation over time-variant channels to mitigate ICI in
OFDM systems. Some of them [1-3] are proposed only to as f D = f d Ts . Perfect timing and synchronization are
track time-varying channles during one OFDM frame assumed. The i-th OFDM symbol X (i ) (k ) is used to modulate
assuming that the channel is stationary over the period of one
OFDM symbol. Previous works of [4-6] have investigated the N- carriers as x (i) (n) = F H X (i ) (k ) , where F stands for the N
CE techniques when the channel fading envelope is varying point discrete Fourier transform (DFT) matrix with
during one OFDM symbol. [F] p , q = 1 N esp (− j2πpq/N ). x (i) (n) is next concatenated by
In this paper, we mainly specialize to channel estimations cyclic prefix (CP). After sending over the channel, the
using the stated comb pilots in IEEE802.16e OFDM Physical received signal can be expressed as
Layer (PHY) mode. According to system parameters and the
This research is jointly founded by The HK-Guangdong Scientific and
Technological Project in Key Areas under grant No:20060104-2 and the Pre- y (i ) (n) = x (i ) (n) ⊗ h(n, l ) + w (i ) (n) , (1)
research Dept.of Huawei Technologies.

1-4244-2424-5/08/$20.00 ©2008 IEEE 1076 ICCS 2008

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 21, 2009 at 08:20 from IEEE Xplore. Restrictions apply.
where w (i ) (n) represents the additive white Gaussian noise { }
E C (Δk , k )C (Δk , k ) ∗ = S ( f ) ⊗ V ( f ) |
f=
Δk ¦σ l
2
. (5)
(AWGN) and h(n, l ) denotes the sampled 2-demention time- N l

variant multipath propagation CIR. h(n, l ) can be formulated


S(f) denotes the Doppler spectrum and V(f) is the DFT of
L −1
the N-point rectangular window [9]. From (5), we can deduce
as h(n, l ) = ¦ h (n)δ (l − l (n)) ,
l =0
l where n is interpreted as that even a small normalized Doppler frequency shift, such as
f D =0.005, induces widespread ICI.
“time” index and l as the “lag” index, hl (n) represents the
gain, l (n) represents the delay of the l-th taps at time n and L is Fig. 1 plots the ICI impacts centered at the 128th subcarrier.
the maximum channel length. We assume that N=256 ,bandwith B=2MHz and Ts = 128us .
Widespread ICI effects introduced from adjacent Δk
Using H to denote the time-variant circular convolution subcarriers to the 128th subcarrier are clearly seen when
matrix such that H [n, l ] = h(n, n − l N ) . Then (1) can be assuming various f D values. It can be seen form Fig.1 that the
rewriten as y (i ) (n) = Hx (i ) (n) + w (i ) (n) . The DFT output at the value of f D is larger, the ICI effects is more obvious.
receiver can be expressed as Y (m) = FHF H X (k ) + W (m) .
H N×N
ª h(0,0) 0 " h(0, L − 1) " h(0,1) º
« »
« h(1 ,1) h (1 ,0) 0 " h (1, L − 1) " »
« # # # # # # »
=« »
«h( L − 1, L − 1) " h( L − 1,0) 0 " 0 »
« 0 h( L − 1, L − 1) " h( L − 1,0) 0 " »»
«
«¬ 0 " 0 h( N − 1, L − 1) " h( N − 1,0)»¼

We further express it as

N −1
Figure 1. Impacts of ICI versusd for various f D values
Y (i ) (m) = ¦ G(m, k ) X
k = 0,
(i)
(k ) + W (i ) (m),0 ≤ m ≤ N − 1. (2)

III. CHANNEL ESTIMATION METHOD


In (2). G(m,k) is evaluated as
From Jeon’s work [6], we know that when f D < 0.1 ,
N −1 L −1
1 j2πn( k − m) j2πkl variations of the CIR expression h(n, l ) , for all L taps, can be
G (m, k ) =
N ¦¦
n=0 l =0
h(n, l ) exp(
N
) exp(−
N
) .
approximated by straight lines with low slopes during a DFT
(3) block period. Therefore, the linear model with a constant slop
over an OFDM symbol are used in conventional estimators.
B. ICI Generation
When the channel is time-variant during a DFT block A. Conventional CE Method
period, G (m, k ) in (3) becomes a nondiagonal matrix. The even pilot-assisted CE methods proposed in [4][5] use
the channel characteristic of linearization in time domain to
Nondiagonal subcarrier coupling matrex G (m, k ) introduces
track channel variations. Equalization is conducted in the
ICI from the m-th subcarrier to the k-th subcarrier. To frequency domain by IDFT. They are implemented as follows:
demonstrate the effects of ICI, we will analyze the channel
frequency response matrix G (m, k ) . We define a new At first, the LS channel frequency response at equally
spaced pilots are obtained as:
variable Δk =m-k, then G (m, k ) can be rewrite as
G (m, k ) = C (Δk + k , k ) . C (Δk , k ) can be expressed as (4).
Ĝ ( i ) ( pD, pD) = Yi,pD X i,pD . (6)
N −1 L −1
1 2πΔkn 2πkl
C (Δk , k ) =
N ¦¦ h(n, l ) exp(− j
n =0 l =0
N
) exp(− j
N
). (4) In (6), i is the index of OFDM symbol, p is the index of
pilot subcarriers, N P is the number of pilots and the interval
between tow adjacent pilots is D = N N P . Then㧘through the
C (Δk , k ) can be interpreted as the influence of the
IDFT of length N P ,
frequency-domain responses at subcarriers k+ Δk , to a
frequency-domain response at the subcarrier k. The variance of
{
C (Δk , k ) can be expressed as E C (Δk , k )C (Δk , k ) ∗ , which } hˆave
(i )
(l ) = 1
N p −1

¦ Gˆ (i )
( pD, pD) exp
j2πpl
.l = 0, ! , N p − 1 . (7)
characterizes ICI energy distributions around the subcarrier k. N p=0 Np

1077

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 21, 2009 at 08:20 from IEEE Xplore. Restrictions apply.
pilot subcarriers. The index 0 is DC subcarrier. The indices of -
ave − h ( n, l ) } is minimized
(i )
For the l-th channel tap, E{ h (i)
128 to -101 and 101 to 127 are guard bands.
when n = N − 1 .Then the linear mode is applied and then
2
channel variations in time-domain can be obtained. Assuming
the constant slope of the l-th channel tap is αˆ (i ) (l ) , h(n, l ) can ...... ...... ...... ...... ...... ...... ......
be approximated as forllows: ⋅N
⋅⋅⋅⋅⋅ ⋅N
⋅⋅⋅⋅⋅

hˆ (i ) (n, l ) = hˆave
(i )
( )
+ αˆ (i ) (l ) n + 1 − N .n = 0, " , N − 1 .
2
(8)

Frequency domain relationship can be derived as Figure 2. OFDM frequency description

L −1 N −1
From Fig.2, we can see that pilots are actually not even
1 2π (k − m)n 2πlk over the whole frequency band for the guard band and DC
Gˆ ( i ) (m, k ) =
N ¦¦
l =0 n=0
h(n, l ) exp( j
N
) exp( − j
N
). (9)
subcarriers. Uneven pilots and CIR leakage prevent us from
using the trational estimation methods[4-6]. Therefore, we
When the slope in each symbol are divided into two rigions develop a modified method which is totally performed in the
frequency domain. Our method avoids the mathematical
obtained by utlizing adjacent symbols, Ĝ (i ) (n, m) can be processing in transform domain which requires even comb
further described in form of (10). pilots pattern. It is performed as followings:
First, estimates of the channel frequency response at pilot
Gˆ (i ) = Gˆ ave
(i)
+ C1Gˆ slope
(i ) 2 ˆ (i )
1 + C G slope 2 . (10) subcarriers can be acquierd as (12):

where, Gˆ slope = diag{DFT ([ a(0), a(1), " , a( N − 1)])} , C1 and Gˆ (i ) ( pD, pD) = Yi,p X i,p䯴 0 ≤ p ≤ N p 䯵 . (12)
C are N × N coefficient matrices.
2

Finally, according to LS alogorithm and the frequency Then, channel frequency responses at the data subcarriers
can be obtained by interpolation[3] according to the sampling
domain expressio: Y䯴 i ) (k ) = Gˆ (i) X (i)(k) + W (i ) (k ) , the estimated theorem. DFT interpolations [11] can not be applied due to the
value of Xˆ (i) (k ) can be derived as unequal-spaced pilot pattern. We use the cubic splines
interpolation, which may cause relatively large mean square
error at the edge subcarriers[11]. To get more accurate
i) H 䯴 i) −1 䯴 i) H estimations, channel frequency responses of the 2 OFDM
Xˆ (i) (k ) = (Gˆ䯴 Gˆ 䯵 Gˆ Y (i) (k ) . (11)
preambles are employed as a reference. We synthesize the
coarse estimation by pilots and preambles togehtoer as the
This conventional methods strictly take the advantage of expression (6) in [10] to get the ultimate estimations. Then,
the even allocation of pilots over total frequencies and the guard band and DC subcarriers are set at zeros. Now, G (k , k ) is
mathematic properties of even interpolation of DFT (as the
obtained according to (3) and (4).
second step).
In additon, non-sample-spaced multi-path delays and 1
N −1
guardband subcarriers may lead to CIR leakage[8]. In (7), L
taps of CIR in time domain are kept, valid taps of CIR leakage
C (0, k ) = G (k , k ) =
N ¦ H ( n, k ) = H
n=0
ave (k ) (13)

beyond the maximum multipath delay L are cut. As a result,


transforming to time domain as (7) discards some channel L −1
2πkl
information and thus a symbol error floor is formed. H (n, k ) = ¦ h(n, l ) exp(− j
l =0
N
). (14)

B. Proposed CE Method
As for a practical OFDM system, the conditions above can For the k-th subcarrier, E{ H ave (k ) − H (n, k ) } is
not be satisfied. Usually, guard band subcarriers are not used to
minimized when n = N − 1 . Therefore, we use H ave (k ) to
avoid aliasing problems and the DC subcarrier is not used to 2
avoid intermodulation effects and difficulties in D/A and A/D represent an estimate of the channel at discrete time instant of
conversion. In other words, pilots in practical are not evenly N − 1 . We will have H (k ) = H ( N − 1, k ) .
inseted and CIR leakage will occured. 2 ave 2
The pilot pattern for CE in the 802.16 OFDM system is According to IEEE802.16e protocol, we choose a set of
shown in Fig. 2. All subcarriers are numbered from -128 to system parameters (N=256, B=2MHz, Ts = 128us V=120km/h).
127. The indices of -88, -63, -38, -13, 13, 38, 63 and 88 are f D = V ⋅ f c ⋅ Ts c ≈ 0.05 < 0.1 , where f c represents the
carrier frequency, V is the maximum vehicle speed and c is the

1078

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 21, 2009 at 08:20 from IEEE Xplore. Restrictions apply.
velocity of light . Thus, the linear model can be applied here to matrix H (n,䰆) varies in a linear model with n on each specified
approximate H (n, k ) variations varying with n of H (n, k ) on subcaiier. The channel estimation and signal correction are
the k-th subcaiier. performed in the frequency domain avoding transform domain
processes. Therefore, CIR leakage cutting in time domain
Consider each three consecutive OFDM symbols, the index
process can be avoided.To demonstrate the effectiveness of
of which is i-1, i, i+1. As for the i-th OFDM symbol (the
the proposed CE approach for time-varying multipath
current symbol), the slopes of the variations can be obtained
channels, the fowlling simulations were performed.
through using the (i+1)-th OFDM symbol (the next OFDM
symbol) and the (i-1)-th OFDM symbol (the previous OFDM
symbol)[5]. It will cause a delay of one OFDM system. We IV. SIMULATIONS
assuming the constant slopes of H (n, k ) on the k-th subcarrier In this section, we will report simulation results
is α k1 and α k2 , which is depicted as Fig. 3. demonstrating the effectiveness of the proposed CE method.
Due to the limited number of pilots in 802.16 OFDM
system, a narrow bandwith of 2MHz is used for simulations.
H ( n, k )

Thus, the frequency fadings in the scope of adjacent pilots can


be regarded as flat. The carrier frequency is 3.5GHz. The
α 1
k
α 2
k
normalized CP is 1 16 . The modulation scheme for transmit
data is QPSK and for pilots is BPSK. To test the estimator’s
Symbol-Error-Rate (SER), no channel coding is employed.
The ITU VA model suggested as a special test channel model
in IEEE 802.16e was adopted to generate channels in
Figure 3. Schematic diagram of linear model and slops of channel in an
OFDM symbol.
simulations. Table. 1 provides key parameters of the ITU VA
channel models. The vehicular speed was set as 120km/h. The
Then H (n, k ) can be approximated as forllows: normalized maximum Doppler frequency offset f D is fixed at
0.05. It means that f d is 385Hz at the 33m/s vehicle speed in
a 3.5GHz channel.
­ Hˆ ave
°
(i )
+ αˆ k1 (n + 1 − N ).0 ≤ n ≤ N − 1
ˆ 2 2
H ( n, k ) = ®
䯴 i)
,
°¯ H ave + αˆ k (n + 1 − N 2 ). N 2 ≤ n ≤ N − 1
ˆ (i ) 2
TABLE I. ITU M.1225 VEHICULAR CHANNEL A
Taps 1 2 3 4 5 6
Relative delay(ns) 0 310 710 1090 1730 2510
( i −1)
Hˆ ave − Hˆ ave
(i )
Hˆ (i +1) − Hˆ ave
(i ) Average power(dB) 0 -1.0 -9 -10 -15 -20
where αˆ k1 = ; αˆ k2 = ave . (15)
N N
The linear model applied in this paper is simulated
According to (4), C (Δk , k ) can be expressed as the Fourier according to the assumed parameters in Table. 1. Variations of
the frequency domain CIR matrix H (n, k ) is depicted in Fig.4.
transform of H (n, k ) multiplied by a coefficient 1 :
N We can see that the linear fitting model is completely
consistent with the schematic diagram in Fig. 3.
N −1 L −1
1 2πΔkn 2πkl
C ( Δk , k ) =
N ¦¦ h(n, l ) exp(− j
n =0 l =0
N
) exp(− j
N
)

N −1
1 2πΔkn
=
N ¦ H (n, k ) exp(− j
n =0
N
). (16)

Then, the channel frequency response matrix G (m, k ) can


be deduced from the expression G (m, k ) = C (Δk + k , k ) by
using the cycle shift operator in each column.
Finally, the estimated value of Xˆ (i) (k ) can be derived from
~ ~ H H
(11). Xˆ (i) (k ) = GY (i) (k ) ,where G = (Gˆ䯴 i) Gˆ䯴 i)䯵 −1Gˆ䯴 i) . And
then, the equalization is performed by multiplying with an
~ Figure 4. Linear fitting in three consecutive OFDM symbols
channel estimated matrix G to cancel ICI effects.
Then, the Channel Mean Square Errors (CMSE) are
Our method uses the linear characteristic of CIR in the
calculated according to the expression (18), where S
frequency domain. The channel frequency impulse response
respresents the number of OFDM symbols used in simulations.

1079

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 21, 2009 at 08:20 from IEEE Xplore. Restrictions apply.
KS −1 2 V. CONCLUSION
1
CMSE =
NS ¦n=0
Hˆ (n, k ) − H (n, k ) . (18) In this paper, an uneven pilot-assisted CE method tracing
time-varying channels for mobile OFDM sysetms is proposed.
The method designed for the uneven comb pilot pattern, uses a
The validity of the linear model is verified by simulations. linear model to approximate the frequency domain CIR
In Fig. 5, we can see that when f D < 0.1 , variations of the variations during each OFDM symbol period. CE and
actual frequency domain CIR matrix H (n, k ) can be fitted equalizations are completely conducted in the frequency
well by the linear model during each OFDM symbol period. domain avoiding mathematical processing in the time domain.
Therefore, the linear model is proved advisable. Therefore, even comb pilot pattern is not necessary and the
cutting of the valid taps of CIR leakage can be avoided.
Simulation results shows the SER results of the proposed CE
method are satisfactory. When used in practical systems, the
algorithm is of low complexity and high efficiency. Howerver,
due to the fixed FFT size and limited quantity of pilots of
802.16 OFDM PHY mode, its ability to resist double selective
fading for a larger bandwith is relatively poor. The 802.16e
OFDMA PHY mode support FFT sizes of 128, 512, 1024 and
2048 to adapt to various bandwith requirements and the pilot
density increases correspondingly. We are currently
investigating the CE technique of double selective channels in
OFDMA PHY mode.

REFERENCES
Figure 5. CMSE of Linear model according to different f D values
[1] Linhai Li, Hong Li, Hongyi Yu and Hanying Hu, “A New Algorithm
for MIMO Channel Tracking Based on Kalman Filter,” Wireless
Communications and Networking Conference, 2007, pp.164 – 168.
[2] H.Li, Yong Wang and Mingyan Jiang, “Adaptive MIMO time-varying
channel equalization using particle filtering,” 2007 International
Symposium on Microwave, Antenna, Propagation and EMC
Technologies for Wireless Communications, 2007, pp.1084 – 1087.
[3] S.Coleri, M.Ergen, A.Puri, and A.Bahai, “Channel estimation
techniques based on pilot arrangement in OFDM systems,” IEEE
Transactions on Broadcasting, Vol. 48, Issue 3, 2002, pp.223 – 229.
[4] Xiaoguang Wu, Guixia Kang, Tian Tang, Ping Zhang and Mingyu
Zhou, “An pilot-assisted channel estimation method for OFDM system
in time-varying channels,” IEEE 18th International Symposium on
Personal, Indoor and Mobile Radio Communications, 2007, pp.1– 6.
[5] Y. Cox, D.C and A.Bahai, “ICI mitigation formobile OFDM receivers,”
IEEE International Conference on Communications, Vol.5
2003, pp.3351 – 3355.
[6] Won Gi Jeon, Kyung Hi Chang andYong Soo Cho, “An equalization
Figure 6. ER performance comparison technique for orthogonal frequency-division multiplexing systems in
time-variant multipath channels,” IEEE Transactions on
Communications, Vol.47, Issue 1, 1999, pp.27–32.
Fig. 6 shows the performance of the proposed CE method.
Curves “1” and “2” represent the SER preformances of the [7] Baxley R.J, Kleider, J.E and Zhou G.T, “Pilot Design for IEEE 802.16
OFDM and OFDMA,” IEEE International Conference on Acoustics,
method in [4]. The curve “1” uses the uneven pilot pattern Speech and Signal processing, April.2007, pp.321 –324.
described as Fig. 2. The curve “2” uses evenly inserted pilots [8] Jiang Jun, Tang Tian, Zhang Yongjing and Zhang Ping, “A Channel
as [4] described, where 64 of 256 subcarriers are for pilots. Estimation Algorithm for OFDM Based on PCC training Symbols and
The curve “3” shows the SER performance of the proposed Frequency Domain Windowing,” International Symposium on
method with uneven pilots pattern as Fig. 2 shows. It is clearly Communications and Information Technologies, 2006, pp.629 – 632.
seen that the SER performance of the proposed uneven pilot- [9] Schniter. P, “Low-complexity equalization of OFDM in doubly
assisted CE method is almost identical to the even pilot- selective channels,” IEEE Transactions on Signal Processing, Vol. 52,
2004, pp.1002 – 1011.
assisted method in [4]. When the uneven pilot is applied, the
[10] Xuetao Dong, Xianzhong Xie and Xin Chen, “Joint Channel Estimation
CE method in [4] suffers poor SER performances as the curve for Wimax by preamble and Uneven Pilot,” International Conference on
“1” shows. Due to the limited pilot number and the frequency Wireless Communications, Networking and Mobile Computing, 2007,
domain interpolations, our method causes relatively large pp.1104 – 1107.
mean squared errors, so SER floor exists. The curev “4” shows [11] Kan Zheng, Jian Su and Wenbo Wang, “DFT-bsaed channel estimation
BER performances of our method when channel coding (RS- in comb-type pilot-aided OFDM systems with virtual carriers,” IEEE
CC2/3) and interleaving are employed. All bit errors will be 18 th International Symposium on Personal, Indoor and Mobile Radio
Communications,Sep.2007, pp.1 – 5.
corrected when SNR ≥ 10dB. The maximu f d supported by
the proposed method can reach 600Hz.

1080

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 21, 2009 at 08:20 from IEEE Xplore. Restrictions apply.

Vous aimerez peut-être aussi