Documente Academic
Documente Profesional
Documente Cultură
7, JULY 2009
Abstract—Carrier frequency offset (CFO) estimation is a of the proposed method. Relations between the SPS method
critical problem in orthogonal frequency-division multiplexing and the Kurtosis, the VE methods are also provided. Finally,
(OFDM) systems. This letter proposes a blind CFO estimator simulation results demonstrate the validity of the proposed
based on smoothing the signal power spectrum. A closed-form
CFO estimate is also presented, which greatly reduces the estimator.
computational complexity of the proposed method. Analysis and
simulation results show that the proposed estimator is very II. T HE OFDM S YSTEM M ODEL
effective for OFDM systems.
Consider an OFDM system with N subcarriers. The
Index Terms—Carrier frequency offset (CFO), power spec- transmitted data in the ith block, defined as s(i) =
trum, orthogonal frequency-division multiplexing (OFDM). [s0 (i), s1 (i), · · · , sN −1 (i)]T , are supposed to be independent
and circularly symmetric with zero mean and variance σs2 .
I. I NTRODUCTION They are then modulated onto the subcarriers by inverse
discrete Fourier transform (IDFT). The CP is inserted to avoid
To overcome the above obstacles, a blind CFO estimator φδ (i)FΛ(δ)FH , φδ (i) ej N δ[i(N +P )+P ] ,
is derived for OFDM systems in this letter, by exploiting Λ(δ) diag(1, ej2πδ/N , · · · , ej2πδ(N −1)/N ),
1 2π −1
−j
the key idea of smoothing power spectrum (SPS). Similar F { √N e N mn
}Nm,n=0 denotes the DFT matrix,
criterion has been advocated in [8], which can estimate as and wk (i) ∼ CN 0, σn2 is the independent noise. The second
well as our SPS method, but it is only proposed for the term in (1) represents the intercarrier interference (ICI). If
CM constellations. Besides, the noise is not considered in the the CFO is estimated perfectly, (1) becomes
analysis. However, this letter shows that the SPS method can yk (i) = Hk (i)sk (i) + wk (i). (2)
also be used for the non-CM cases. Interestingly, it is indicated
that the proposed objective function has a cosine form in the
III. A C LOSED -F ORM CFO E STIMATOR
noise case. More importantly, a closed-form CFO estimate
is obtained to greatly reduce the computational complexity A. The Proposed SPS Method
Using exactly the ith OFDM block, the classical signal
Paper approved by C. Tepedelenlioglu, the Editor for Synchronization and power spectrum can be given by
Equalization of the IEEE Communications Society. Manuscript received May
24, 2007; revised January 3, 2008 and June 15, 2008. 1 k 2
This work was supported by the National Natural Science Foundation of
k
Pr,ν |y (i)| .
(i) = (3)
China under Grants 60675002 and 60496311, and funded by the Basic Re-
N ν
search Foundation of Tsinghua National Laboratory for Information Science Similarly, the classical channel power spectrum has the form
and Technology (TNList).
The authors are with the State Key Laboratory on Intelligent Technol- 1
Phk+1 (i) = |Hk+1 (i)|2
ogy and Systems, Tsinghua National Laboratory for Information Science N
and Technology, Department of Automation, Tsinghua University, Beijing
1
L L
100084, China (e-mail: wulu04@gmail.com; zxd-dau@mail.tsinghua.edu.cn; 2πk n−l
lipeishengabcd@163.com; suyt02@mails.tsinghua.edu.cn).
= hn (i)h∗l (i)e−j N (n−l) e−j2π N (4)
N n=0
Digital Object Identifier 10.1109/TCOMM.2009.07.070067 l=0
0090-6778/09$25.00
c 2009 IEEE
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 04:14 from IEEE Xplore. Restrictions apply.
WU et al.: BLIND CFO ESTIMATOR BASED ON SMOOTHING POWER SPECTRUM FOR OFDM SYSTEMS 1925
0.25
Channel indicates that δ = 0 is a global minimum when A < 0 and
Received signal: without CFO compensation a global maximum when A > 0. The following propositions
Received signal: perfect CFO compensation
0.2 show the sign of A (proof is given in the Appendix B).
9
Proposition 2: If ks < 1 + 3+π 2 , A < 0 holds.
Classical Power Spectrum
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 04:14 from IEEE Xplore. Restrictions apply.
1926 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 57, NO. 7, JULY 2009
0 0
10 10
Kurtosis(curve fitting): 16QAM, SNR=10dB
SPS(parameter estimation): 16QAM, SNR=10dB
Kurtosis(curve fitting): 16QAM, SNR=20dB
−1
10 SPS(parameter estimation): 16QAM, SNR=20dB
Kurtosis(curve fitting): QPSK, SNR=10dB
SPS(parameter estimation): QPSK, SNR=10dB −2
10
Kurtosis(curve fitting): QPSK, SNR=20dB
−2 SPS(parameter estimation): QPSK, SNR=20dB
10
MSE
MSE
−3 −4
10 10
M=1, CP
M=1, Kurtosis(curve fitting)
−4
10 M=1, VE(curve fitting)
M=1, SPS(parameter estimation)
−6 M=1, SPS(search)
10 M=1, Subspace(search)
M=10, CP
−5
10 M=10, Kurtosis(curve fitting)
M=10, VE(curve fitting)
M=10, SPS(parameter estimation)
M=10, SPS(search)
−6 −8 M=10, Subspace(search)
10 10
0 2 4 6 8 10 12 14 16 0 5 10 15 20 25 30
Channel Order SNR(dB)
Fig. 2. Robustness of the SPS method to the increased channel frequency Fig. 3. Performance of the SPS method varying with SNR: QPSK, = 0.1,
selectivity: = 0.1, N = 64, M = 10, P = 16, 5000 Monte Carlo trials. N = 64, P = L = 5, 5000 Monte Carlo trials.
1 k 2
N −1 −1 −1
k+1 2 2
N
N
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 04:14 from IEEE Xplore. Restrictions apply.
WU et al.: BLIND CFO ESTIMATOR BASED ON SMOOTHING POWER SPECTRUM FOR OFDM SYSTEMS 1927
⎛ ⎞
9
N −1 N −1
1 ⎝ 2σs4 G ks − 1 − 3+π 2
N −1
= 4 Fmmaa + Fmppm ⎠ + T (12) < ρm,m (17)
N m=0 a=0
p=m
N4 m=0
N −1 N −1 where G = [(3 + π 2 )N 2 − π 2 ]/3π . 2
where Fxyzt n,q,b,d=0 E[Hx Hy∗ Hz Ht∗ ]E[sx s∗y sz s∗t ] k=0
2π The expression of A can be rewritten as
ej N [(δ−k)(n−q+b−d)+(xn−yq+(z−1)b−(t−1)d)] , and T is a
N −1
constant independent of δ. Thus, after some tedious but 2σs4 (ks − 1)(N 2 − 1) 2 − ks
straightforward algebra, we can show that A= 4 − ρm,m
N m=0 3 sin2 Nπ
⎧
⎨ A1 cos(2πδ) + B1 , a=m 2 ⎫
|Hm |2 − |Ha |2 ρm,m ⎬
N−1 E
N −1
Fmmaa = A2 cos(2πδ) + B2 , a=m+1 (13) + −
⎩ sin2 aπ ⎭
A3 cos(2πδ) + B3 , else a=m+1 sin2 (a−m)π
N a=1 N
and 2σ 4 N−1
(k − 2)(N 2
− 1) (k − 2)
⎧ = 4 s
ρm,m
s
+
s
⎨ A4 cos(2πδ) + B4 , p=m−1 N m=0 3 sin2 Nπ
Fmppm = A4 cos(2πδ) + B5 , p=m+1 . (14) 2 ⎫
⎩ −1 E |Hm |2 − |Ha |2 ⎬
B6 , p = m−1, m, m + 1 N
+ (18)
2π
4N 2 γs4 e−j N ρm,m sin2 (a−m)π ⎭
1 2 2 4
where A1 = 2π , A2 = 3 N (N −1)σs ρm,m+1 , a=m+1 N
(1−e−j N )2
N 2 σs4 ρa,m 2N 2 σs4 ρm,p where the last equation utilizes the fact that
A3 = − 2 (a−m−1)π , A4 = , B1 , B2 , B3 , N −1 2
a=1 cot (aπ/N ) = (N − 1)(N − 2)/3[11].
sin j 2π −j 2π
N (1−e N )(1−e N )
B4 , B5 and B6 are constants independent of δ. Substituting Consequently, Proposition 2 and Proposition 3 can be easily
(14) into (12), we get deduced from (17) and (18).
−1
N
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 04:14 from IEEE Xplore. Restrictions apply.