Sie sind auf Seite 1von 5

2436 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO.

7, JULY 2008

ML Joint CFO and Channel Estimation in


OFDM Systems with Timing Ambiguity
Jianwu Chen, Yik-Chung Wu, Shaodan Ma, and Tung-Sang Ng, Fellow, IEEE

Abstract—This letter addresses the problem of joint estimation null-space based joint CFO and channel estimator is proposed
of carrier frequency offset (CFO) and channel for OFDM systems by Zeng et al. [6] (referred as ZJCCE in the following).
in the presence of timing ambiguity. Based on two signal models It is noticed that perfect time synchronization is always
for quasi-synchronized OFDM systems, two joint CFO and
channel estimators are derived from the Maximum Likelihood assumed in these joint CFO and channel estimation schemes
(ML) criterion. The first estimator obtained is a joint estimator [4]-[6], but such an assumption is too restrictive in practice.
for all unknown parameters, which needs a multi-dimensional Since the metrics for many existing timing synchronization
search. The second estimator has a reduced complexity, but algorithms have a plateau in the presence of frequency selec-
its performance slightly degrades compared with the first one. tive fading channel ([7], [8] and [9]), perfect time synchro-
Through MSE analyses, we find that when the number of
subcarriers is large, the two estimators are equivalent. nization is difficult to achieve. Up to now, how the existing
joint CFO and channel estimators behave in the presence of
Index Terms—Carrier frequency offset (CFO), channel estima- timing ambiguity is still unclear. As an initial work, in [10],
tion, Cramér-Rao bound, maximum likelihood (ML), orthogonal
frequency-division multiplexing (OFDM). an improved timing robust version of the null-space based
estimation scheme in [6] is recently proposed using an ad
hoc argument. Unfortunately, that work provides little insights
I. I NTRODUCTION into the effects of timing ambiguity on a general joint CFO
and channel estimation case. Motivated by this, the letter
O RTHOGONAL frequency division multiplexing
(OFDM) systems are sensitive to the carrier frequency
offset (CFO) caused by mismatch of the local oscillator
first introduces two signal models with timing ambiguity:
one uses an explicit matrix to represent the effects of timing
offset and the other one embeds the timing offset in the
in transceivers and Doppler shifts due to mobility. The
channel. Then two ML estimators for joint CFO and channel
inter-carrier interference (ICI) induced by CFO destroys
estimation are derived and analyzed based on the two signal
the orthogonality between subcarriers and causes significant
models respectively. For the first estimator, a two dimensional
performance degradation, so it is critical to estimate the
search is needed to obtain the optimal estimates since the
CFO at the receiver and compensate its effects. Moreover, in
timing offset and CFO are coupled with each other. For
order to achieve simple coherent equalization, which is one
the second signal model, two dimensional search is avoided,
of the dominant advantages of OFDM technique, accurate
however, as more unknown parameters need to be estimated,
estimation of the channel is usually required.
slight performance degradation occurs. Further mathematical
Generally, the CFO and channel estimation are treated analyses show that the degradation vanishes as the number of
separately with the aid of different pilots [1]-[2]. In order to subcarriers increases. Note that the seemingly related scheme
save bandwidth, blind and semiblind approaches exploiting proposed in [11] for DS-CDMA is not applicable to the
the inherent structure of the received OFDM signal gain problem under consideration since it is based on an assumption
some interests [3]. However, since the blind schemes typically that the effects of CFO within one block of data can be simply
require observation over a large number of OFDM symbols, approximated as a scalar, which is not valid in OFDM systems.
they are not suitable for packet-based transmission where CFO
is changing packet by packet. As an efficient method with
reduced training overhead, joint CFO and channel estimation II. S YSTEM D ESCRIPTION
schemes based on common pilots are highlighted in [4]-[6]. A packet-based OFDM system with N subcarriers is con-
For systems with embedded training, an iterative algorithm is sidered. For each data packet, it is preceded by some training
proposed to jointly estimate the CFO and channel based on the blocks as shown in Fig. 1. It is assumed that timing synchro-
pilot tones in [4]. For block training, the optimal pilot designs nization will be completed by exploiting the training blocks at
for both CFO and channel estimation are investigated in [5] the beginning of the preamble (Part 1). The CFO and channel
using the worst-case asymptotic Cramér-Rao bound (CRB). estimation tasks will be carried out on the second portion of
Based on one kind of the optimal pilots presented in [5], a the preamble which consists of only one OFDM block (Part 2).
Similar training structures have been proposed in many OFDM
Manuscript received January 10, 2007; revised May 23, 2007; accepted based standards, such as IEEE802.11a and HiperLAN2.
August 30, 2007. The associate editor coordinating the review of this letter
and approving it for publication was G. Vitetta. This work was supported by
At the transmitter, the frequency domain signal in one
grant HKU7168/03E from the Research Grants Council of Hong Kong. OFDM block d = [d(0), d(1), . . . , d(N − 1)]T is first mod-
The authors are with the Department of Electrical and Electronic Engi- ulated onto different subcarriers. The corresponding time-
neering, The University of Hong Kong, Hong Kong (e-mail: {jwchen, ycwu,
sdma, tsng}@eee.hku.hk).
domain vector s is given by s = FH d, where F is the
j2πkl
Digital Object Identifier 10.1109/TWC.2008.070026. FFT matrix with F(k, l) = √1N e− N . A cyclic prefix
1536-1276/07$25.00 
c 2008 IEEE

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 09:09 from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 7, JULY 2008 2437

in the estimation. This assumption is not restrictive since many


practical sequences possess these properties.

A. ML Estimator for Signal Model (1)


Based on the signal model in (1), the ML estimates of
parameters {h, ωo , θo } is given by minimizing
Fig. 1. The packet structure for OFDM systems.
Λ1 (x; h̃, ω̃, θ̃) = [x − Γ(ω̃)A(θ̃)h̃]H [x − Γ(ω̃)A(θ̃)h̃] (3)
where h̃, ω̃ and θ̃ are trial values of h, ωo and θo . Due to
(CP) of length Lcp is inserted ahead of s to cope with the
the linearity of parameter h in (1), the least square estimate
inter-symbol interference (ISI) caused by multipath fading.
(also ML estimate in this case) for h (when ω̃ and θ̃ are fixed)
The channel impulse response (including all transmit/receive
is ĥ = (FH H
L D DFL )
−1
(Γ(ω̃)A(θ̃))H x = (Γ(ω̃)A(θ̃))H x.
filtering effects) is denoted as h = [h0 , . . . , hL−1 ]T , and
Substituting this result into (3), and dropping the irrelevant
is quasi-static over one data packet. The CP length Lcp is
terms, the ML estimator for ωo and θo is given by
assumed to be larger than the channel order L. The normalized
CFO between the transmitter and receiver is assumed to be [ω̂, θ̂] = arg max{J1 (ω̃, θ̃)  AH (θ̃)ΓH (ω̃)x2 }. (4)
εo . At the receiver, a correlation-based timing synchronization ω̃,θ̃
scheme is used (e.g. [7] [8]), such that the ISI-free region for
In the above estimation scheme, the channel h, CFO ωo
the subsequent OFDM symbols can be located. Defining the
and timing offset θo need to be jointly estimated, so in the
sample indexes of a perfectly synchronized OFDM block as
following, we will refer this estimator as joint timing, CFO and
[−Lcp , . . . , 0, . . . , N −1], the estimated starting position of the
channel estimator (JTCCE). Notice that the ZJCCE scheme in
FFT window can be regarded in the range [−(Lcp − L), 0]. [6] can be regarded as a special case of the JTCCE derived
In the following, we denote the timing offset between the
here when θo = 0 (i.e., no timing offset).
estimated starting position of the FFT window and the perfect
timing point as θo . After CP removal, the received signal
vector x, which consists of N consecutive samples, is given B. ML Estimator for Signal Model (2)
by With the signal model in (2), the estimates of parameters
x = Γ(ωo )A(θo )h + v (1) {ξ, ωo } is given by minimizing
where ωo  2πεo /N ; Γ(ωo )  diag(1, . . . , e(j(N −1)ωo ) ); Λ2 (x; ξ̃, ω̃) = [x − Γ(ω̃)Bξ̃]H [x − Γ(ω̃)Bξ̃] (5)
A(θo )  T(θo )FH DFL with FL = F(:, 0 : L − 1);
T(θo )  [eθo +1 , . . . , eN , e1 , . . . , eθo ] with ei = IN ×N (: where ξ̃ is the trial value of ξ. With DH D = IN ×N , it is easy
, i); D  diag(d(0), d(1), . . . , d(N − 1)); and vector v  to check that BH B = ILcp ×Lcp . Then the estimate for ξ can
[v0 , v1 , . . . , vN −1 ]T denotes the complex white Gaussian noise be obtained as ξ̂ = (BH B)−1 (Γ(ω̃)B)H x = BH ΓH (ω̃)x.
with zero mean and covariance matrix Cv = E{vvH } = Substituting this result into (5), the ML estimator for ωo is
σ 2 IN ×N . MATLAB notations are used throughout the paper.
Equivalently, we can rewrite the above system model as ω̂ = arg max{J2 (ω̃)  BH ΓH (ω̃)x2 }. (6)
ω̃
[11]:
x = Γ(ωo )Bξ + v (2) In the above estimation scheme, the timing offset is embed-
ded in the channel, so two dimensional search scheme like that
where B  FH DFLcp ; ξ  [0Tθo ×1 hT 0T(Lcp −θo −L)×1 ]T . of (4) is avoided. Since the timing ambiguity θ̃ is an integer
The difference between (1) and (2) lies in the way how the in the region [−(Lcp − L), 0], it is clear that the complexity
timing ambiguity is presented. In the first model the timing for this estimator is Lcp − L times lower than that of JTCCE.
ambiguity θo is shown in A(θo ) explicitly, while in the second In the following, we will refer this estimator as joint CFO and
one the timing ambiguity is embedded in the channel response channel estimator (JCCE). It is also noticed that (6) coincides
ξ. with the ad hoc based timing robust CFO estimator in [10].

III. ML E STIMATORS W ITH O PTIMAL T RAINING


C. Relationship Between JTCCE and JCCE
With the signal models (1) and (2) in Section II, two ML
estimation schemes are derived in this section. To achieve the Denoting C(θ̃) = diag(1, . . . , e(j2π(N −1)θ̃/N ) ), it is easy
best estimation performance, the optimal training sequence to check that FTH (θ̃) = C(θ̃)F. Putting this result into (4),
is considered. As pointed out by [5], the optimal training we have
sequence d for joint CFO and channel estimation should be an J1 (ω̃, θ̃) = FH H H H
L D FT (θ̃)Γ (ω̃)x
2
orthogonal sequence and also meet FH H
L D DFL = Es IL×L , = FH H H
L C(θ̃)D FΓ (ω̃)x
2
where Es is the power of training symbols. Without loss of
generality, we assume the pilot has unit power (Es = 1). It is = FH (:, θ̃ : θ̃ + L − 1)DH FΓH (ω̃)x2 . (7)
noticed that the orthogonal sequence with constant-modulus On the other hand, from (6), we can obtain
(i.e., DH D = IN ×N ) is one subclass of the above optimal
training sequence. Hereafter, this kind of sequence is assumed J2 (ω̃) = FH (:, 0 : Lcp − 1)DH FΓH (ω̃)x2 . (8)

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 09:09 from IEEE Xplore. Restrictions apply.
2438 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 7, JULY 2008

where F⊥ L = F(:, L : N − 1) is the null space of


FL . Substituting the signal model (1) in (10), we have
f1 (ω̃, θ̃) = (F⊥ H H H 2
L ) D FT (θ̃)Γ(Δ)A(θo )h + n1  , where
Δ = ωo − ω̃ is the error of CFO estimation; and n1 =
(F⊥ H H H
L ) D FT (θ̃)Γ(ω̃)v is a noise vector having the co-
variance matrix Cn1 = E{n1 nH 2
1 } = σ I(N −L)×(N −L) .
Since generally Δ is small, with the Taylor series ex-
pansion, Γ(Δ) can be approximated as Γ(Δ) ≈ IN ×N +
jMΔ with M = diag(0, . . . , N − 1). Using this re-
sult, we have f1 (ω̃, θ̃) ≈ G1 (θ̃)h + K1 (θ̃)hΔ + n1 2 ,
where G1 (θ̃) = (F⊥ H H H
L ) D FT (θ̃)A(θo ) and K1 (θ̃) =
⊥ H H H
j(FL ) D FT (θ̃)MA(θo ). For a given estimate of timing
θ̃ = θ̂ and a realization of h, the estimation problem in (10)
Fig. 2. Physical meanings for the two proposed estimators JTCCE (maxi-
mizing (7)) and JCCE (maximizing (8)). is equivalent to Δ = arg min{G1 (θ̂)h + K1 (θ̂)hΔ + n1 2 }.
Δ
The solution to this equivalent problem can be obtained by
differentiating G1 (θ̂)h + K1 (θ̂)hΔ + n1 2 with respect to
Denoting h̄ = FH DH FΓH (ω̃)x and assuming the esti- Δ and setting the result to zero (details omitted). Using the
mated CFO is perfect (ω̃ = ωo ), we have the following resultant Δ, the MSE can be approximated as
equation by using the signal model (1):
N2
h̄ = FH DH FΓH (ωo )x MSE1 (h, θ̂) = · En1 {Δ2 }
4π 2
= FH C(θo )DH FFH DFL h + FH DH FΓH (ωo )v N2 
≈ · 2hH KH 2
1 (θ̂)K1 (θ̂)hσ +
= [0Tθo ×1 hT 0T(N −θo −L)×1 ]T + FH DH FΓH (ωo )v. (9) 4π 2 · 4(hH KH 1 (θ̂)K1 (θ̂)h)
2

It is noticed that in the absence of CFO, h̄ can be regarded (hH KH1 ( θ̂)G 1 (θ̂)h + hH H
G 1 (θ̂)K 1 (θ̂)h)2

as a shifted time domain channel vector. Graphically, h̄ (11)


consists of three parts as shown in Fig. 2. The two separate We will show by simulation in the next section that when
shaded areas consist of noise samples only and the blank block SNR is high enough the proposed JTCCE can always achieve
is the shifted time domain channel plus noise. the perfect timing estimation (θ̂ = θo ), so the MSE perfor-
Comparing h̄ in (9) with (7) and (8), it is clear that mance of CFO estimation in this scheme can be approximated
maximizing (7) is equivalent to locating the position of a by MSE1 (h, θo ). For ZJCCE scheme, since perfect timing is
window of length L (shown as the trial window for JTCCE always assumed (θ̂ = 0), the performance can be numerically
in Fig. 2) and finding an ω̃ such that the energy within the calculated by MSE1 (h, 0).
window is maximized; while maximizing (8) is to find an ω̃ Due to the highly non-linear nature of the parameter θo in
such that the energy within the window from 0 to Lcp (shown the signal model (1), it is very difficult to derive the exact CRB
as the trial window for JCCE in Fig. 2) is maximized. In some for the joint estimation in (4). To give a lower bound of the
sense, the JCCE is an approximation to JTCCE by using a CFO estimation, the conditional Cramér-Rao bound (CCRB)
larger window during frequency estimation. Of course, using a with perfectly known θo is given by [12]
window larger than L would cause degradation in performance
since more noise is included in the observation. However, we N2
CCRB1 (h, θo ) = ·
will show in the next section that the performance loss is 4π 2
negligible when the number of subcarriers N is large. σ2
.
Remark: In the proposed schemes, by using a pseudorandom 2(hH AH (θo )M(I − A(θo )AH (θo ))MA(θo )h)
training sequence, the acquisition range of CFO estimation (12)
is [−N/2, N/2]. This range is considered large compared to For JTCCE, it is possible that the timing offset estimation
other existing schemes such as [1] and [7]. is not perfect, then the performance of CFO estimation will
degrade since some extra noise areas will be included in the
IV. P ERFORMANCE A NALYSES CFO estimation as depicted in Fig. 2. However, the above
In the following, the performances of the above two ML CCRB is still a valid lower bound.
estimators are analyzed. Since the performance of the con-
sidered system is mainly determined by the accuracy of CFO B. Performance of the JCCE
estimation, only the performance of the CFO estimation is The equivalent estimator of (6) can be written as
analyzed.
ω̂ = arg min{f2 (ω̃)  (F⊥ H H H 2
Lcp ) D FΓ (ω̃)x } (13)
ω̃
A. Performance of the JTCCE where F⊥Lcp = F(:, Lcp : N − 1) is the null space of FLcp .
The equivalent estimator of (4) can be written as By following a similar procedure for deriving (11), the MSE
[ω̂, θ̂] = performance of the estimator in (13) can be approximated as
arg min{f1 (ω̃, θ̃)  (F⊥ H H H H 2 N2 N2 σ2
L ) D FT (θ̃)Γ (ω̃)x } (10) MSE2 (h) = · E{Δ2 } ≈ · H H (14)
ω̃,θ̃ 4π 2 2
4π 2h K2 K2 h

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 09:09 from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 7, JULY 2008 2439

1
ZJCCE (simulated)
−2 ZJCCE (theoretical)
10
f(δ)

0.5 SNR = 20 dB JTCCE (theoretical)


JTCCE CCRB
−3
JTCCE (simulated)
0 10

MSE of CFO estimation


−7 −6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 7 JCCE (theoretical)
JCCE CRB
1
JCCE (simulated)
−4
10
f(δ)

0.5 SNR = 30 dB

−5
0 10
−7 −6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 7
1
−6
10
f(δ)

0.5 SNR = 40 dB
−7
10
0 5 10 15 20 25 30 35 40
−7 −6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 7 SNR(dB)
δ(samples)
Fig. 4. The MSE performance of the CFO estimation with timing error.
Fig. 3. The pdf for the error of timing estimation in JTCCE.

parameters: N = 64, Lcp = 16, which are consistent with the


where K2 = j(F⊥ H H
Lcp ) D FMA(θo ). WLAN standard. The second part of the preamble section is
With the signal model in (2), the CRB for JCCE is given constructed by a Chu-sequence [13]. 5 data OFDM blocks are
by [12] sent after the preamble in each packet with 16QAM modula-
N2 σ2 tion. A multipath Rayleigh fading channel with eight sample-
CRB2 (ξ) = · . (15) spaced taps and exponential power delay profile (normalized
4π 2 2(ξH BH M(I − BBH )MBξ)
to unit power) is considered. Without loss of generality, the
C. Comparison normalized CFO for each packet εo is generated as a random
variable uniformly distributed in [−0.5, 0.5] and it is assumed
Though it is shown graphically in Section III that the JCCE
that the timing estimate is uniformly distributed in the range
is an approximation of JTCCE, it is more meaningful to
[−(Lcp − L), 0]. All results are averaged over 10000 Monte
quantitatively compare the performance of the two estimators.
Carlo runs.
Denoting the ratio of the MSE performance between JCCE
In Fig. 3, the probability density function (pdf) of the
and JTCCE as r, and applying the result in (11) and (14), the
following equation is obtained: estimation error of time offset in JTCCE is shown, where
δ = θo − θ̂ is measured in simulation. It can be seen that
MSE2 (h) at different SNRs of interest, the JTCCE has a very large
r=
MSE1 (h, θo ) possibility of identifying the correct timing point. Therefore,
hH KH
1 (θo )K1 (θo )h we can set θ̂ = θo for MSE1 (h, θ̂) to obtain the approximated
= MSE performance for JTCCE.
hH KH 2 K2 h
FH (:, L : Lcp − 1)DH FMFH DFL h2 In Fig. 4, the MSE performances of the CFO estimation
=1+ . (16) in JTCCE, JCCE and ZJCCE are plotted versus SNR. The
FH (:, Lcp : N − 1)DH FMFH DFL h2
theoretical curves and lower bounds are obtained by numer-
In the above equation, the derivation from the first step to the ically averaging the results in (11), (14), (12) and (15) over
second step is based on the fact that hH KH 1 (θo )G1 (θo )h + channel realizations. It is apparent that the results of analyses
hH GH 1 (θo )K1 (θo )h = 0. It is noticed that r in the above match the simulations very well. Furthermore, the JCCE and
equation is only determined by a power ratio between two JTCCE can approach their lower bounds respectively and only
projections of a common vector onto different subspaces of a marginal performance gap between the two schemes can
F. For practical systems with a large number of subcarriers, be observed. It is also noticed that in the considered system,
the dimension of F(:, L : Lcp − 1) is much smaller than the ZJCCE scheme degrades a lot compared with JCCE and
that of F(:, Lcp : N − 1), so for this case the second term JTCCE, especially in high SNR range where a MSE floor
in (16) approaches zero. Then, the following result can be occurs.
N →∞
obtained: r −−−−→ 1. It means that the JCCE is asymptotically Next, in Fig. 5, the number of subcarriers is varied from
equivalent to the JTCCE for systems with a large number of 32 to 1024 (SNR=20dB) to show the effects of the training
subcarriers. OFDM block length on the proposed CFO estimators. It is
found that the performance gap between JCCE and JTCCE
V. S IMULATION R ESULTS AND D ISCUSSIONS decreases when the number of subcarrier increases. When the
In this section, simulation results are presented to demon- number of subcarriers is larger than 128, there is no noticeable
strate the effectiveness of the proposed schemes. In all sim- difference between the two estimators, thus corroborating our
ulations, the considered OFDM system has the following theoretical result that the two estimators are asymptotically

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 09:09 from IEEE Xplore. Restrictions apply.
2440 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 7, NO. 7, JULY 2008

−2
10
ZJCCE
Furthermore, although theoretically, JCCE suffers loss respect
JCCE to JTCCE, but in terms of SER performance, JCCE performs
JTCCE the same as JTCCE. This makes JCCE very attractive as there
CCRB for JTCCE
−3
10 is no need to deal with timing ambiguity explicitly. For the
ZJCCE scheme, a poorer SER performance compared with
MSE of CFO estimation

JTCCE and JCCE over the whole SNR range is observed.


−4
VI. C ONCLUSION
10
In this letter, based on two different signal models, two
ML estimators for joint CFO and channel estimation un-
−5
der timing ambiguity have been derived. The first estimator
10
jointly estimates the timing offset, CFO and channel, but
the complexity is high due to the need of two dimensional
search. The second one has a lower complexity compared
−6
10 with the first one, but requires more unknown parameters
to be estimated resulting in slight performance degradation.
1 2 3 4
10 10 10 10
Number of Subcarriers N
Through mathematical analyses and comparisons, it has been
Fig. 5. The MSE performance of the CFO estimation vs. the number of shown that the two estimators are asymptotically equivalent
subcarriers. when the number of subcarriers is large. This makes the
0
second estimator very attractive in practice.
10
ZJCCE
JCCE
JTCCE
R EFERENCES
Perfect synchronization
10
−1 [1] P. H. Moose “A technique for orthogonal frequency division multiplexing
Ideal channel
frequency offset correction,” IEEE Trans. Commun., vol 42, no. 10, pp.
2908–2914, 1994.
[2] M. Morelli and U. Mengali, “A comparison of pilot-aided channel
estimation methods for OFDM systems,” IEEE Trans. Signal Processing,
SER

10
−2
vol. 49, no. 12, pp. 3065–3073, 2001.
[3] X. Ma, C. Tepedelenlioglu, G. B. Giannakis, and S. Barbarossa, “Non-
data-aided carrier offset estimators for OFDM with null subcarriers: Iden-
tifiability, algorithms, and performance,” IEEE J. Select. Areas Commun.,
−3
vol. 19, no. 12, pp. 2504–2515, 2001.
10 [4] H. Jin and J. Moon, “Joint CFO, data symbol and channel response
estimation in OFDM systems," in Proc. IEEE Int. Conf. Commun. (ICC),
pp. 2127–2133, 2005.
[5] P. Stoica and O. Besson, “Training sequence design for frequency offset
10
−4 and frequency-selective channel estimation,” IEEE Trans. Commun., vol.
0 5 10 15 20 25 30 35 40 51, no. 11, pp. 1910–1917, 2003.
SNR(dB) [6] Y. Zeng, W. Leon, Y. C. Liang, and A. R. Leyman, “A new method for
frequency offset and channel estimation in OFDM," in Proc. IEEE Int.
Fig. 6. The SER for the proposed ML estimation schemes. Conf. Commun. (ICC), pp. 4606–4611, 2006.
[7] T. M. Schmidl and D. C. Cox, “Robust frequency and timing synchroniza-
tion for OFDM,” IEEE Trans. Commun., vol. 45, no. 12, pp. 1613–1621,
1997.
equivalent. For the ZJCCE scheme, the performance is far [8] J. J. van de Beek, M. Sandell, and P. O. Borjesson, “ML estimation
worse than the two proposed ML estimators. of time and frequency offset in OFDM systems,” IEEE Trans. Signal
To show the effects of the proposed schemes on the whole Processing, vol. 45, no. 7, pp. 1800–1805, 1997.
[9] H. Minn, M. Zeng, and V. K. Bhargava, “On timing offset estimation for
system performance, the curves of symbol error rate (SER) for OFDM systems,” IEEE Commun. Lett., vol. 4, pp. 242–244, July 2000.
different schemes are plotted in Fig. 6. The SER performance [10] J. Chen, Y.-C. Wu, and T. S. Ng, “Timing robust joint carrier frequency
of the perfectly synchronized case (channel still needs to be es- offset and channel estimation for OFDM systems," in Proc. IEEE Wireless
Communications and Networking Conf. (WCNC), pp. 1025–1029, 2007.
timated from training) and ideal case (perfectly synchronized [11] A. A. D’Amico and M. Morelli, “Frequency estimation and timing
and channel perfectly known) are also shown as comparisons. acquisition in the uplink of a DS-CDMA system,” IEEE Trans. Commun.,
It can be seen that both the proposed ML estimators have the vol. 52, no. 10, pp. 1809–1819, Oct. 2004.
[12] M. Morelli and U. Mengali, “Carrier-frequency estimation for transmis-
same SER performance as the perfectly synchronized case and sions over selective channels,” IEEE Trans. Commun., vol 48, no. 9, pp.
only a small gap from the ideal case. This means that with 1580–1589, 2000.
the proposed two frequency estimators, the bottleneck of the [13] D. C. Chu, “Polyphase codes with good periodic correlation properties,”
IEEE Trans. Inform. Theory, vol. 18, pp. 531–532, July 1972.
system performance no longer lies in frequency estimation.

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 09:09 from IEEE Xplore. Restrictions apply.

Das könnte Ihnen auch gefallen