IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO.
4, MAY 2014
the WiFi-offloading ratio on the per-user throughput. The proposed
network-assisted user-centric WiFi-offloading model utilizes the net-
work information while offloading more traffic into WiFi networks
and achieves the maximum per-user throughput. The proposed model
outperforms the other existing models with 20% higher per-user and
network throughput in a dense traffic environment.
For further work, we will investigate the impact of power con-
sumption and mobility on WiFi-offloading models in a heterogeneous
network.
R EFERENCES
[1] Cisco, Cisco Visual Networking Index: Global Mobile Data Traffic Fore-
cast Update, White Paper, Feb. 2012.
[2] CTIA, The Wireless Association Semi-Annual Survey Shows Significant
Demand by Americans for Wireless Broadband, Apr. 2012. [Online].
Available: http://www.fiercewireless.com/press-releases/ctia-wireless-
association-semi-annual-survey-shows-significant-demand-ameri
[3] N. Ristanovic, J. L. Boudec, A. Chaintreau, and V. Erramilli, “Energy
efficient offloading of 3G networks,” in Proc. 8th Int. IEEE Conf. MASS,
Oct. 2011, pp. 202–211.
[4] A. Aijaz, O. Holland, P. Pangalos, and H. Aghvami, “Energy savings for
cellular access network through Wi-Fi offloading,” in Proc. IEEE ICC,
Jun. 2012, pp. 4369–4373.
[5] K. Lee, I. Rhee, J. Lee, S. Chong, and Y. Yi, “Mobile data offloading:
How much can WiFi deliver?” in Proc. ACM SIGCOMM, Nov. 2010,
pp. 425–426.
[6] X. Zhuo, W. Gao, G. Cao, and S. Hua, “An incentive framework for
cellular traffic offloading,” IEEE Trans. Mobile Comput., accepted for
publication.
[7] S. Dimatteo, P. Hui, B. Han, and V. Li, “Cellular traffic offloading through
WiFi networks,” in Proc. IEEE MASS, Oct. 2011, pp. 192–201.
[8] D. S. Deif, H. El-Badawy, and H. El-Hennawy, “Topology based modeling
and simulation of UMTS-WLAN wireless heterogeneous network,” in
Proc. IEEE Wireless Opt. Comm. Netw., Sep. 2010, pp. 1–5.
[9] A. Balasubramanian, R. Mahajan, and A. Venkataramani, “Augmenting
mobile 3G using WiFi,” Proc. MobiSys, pp. 209–222, Jun. 2010.
[10] J. Kim, N. Song, B. H. Jung, H. Leem, and D. K. Sung, “Placement of
WiFi access points for efficient WiFi offloading in an overlay network,”
in Proc. IEEE PIMRC, Sep. 2013, pp. 3081–3085.
[11] A. B. Pontes, D. P. Silva, J. Jailton, Jr., O. Rodrigues, Jr., and
K. L. Dias, “Handover management in integrated WLAN and mobile
WiMAX networks,” IEEE Trans. Wireless Commun., vol. 15, no. 5,
pp. 86–95, Oct. 2008.
[12] “Access Network Discovery and Selection Function (ANDSF) Manage-
ment Object,” 3GPP, Sophia-Antipolis, France, Tech. Rep. TS 24.312,
Dec. 2012.
[13] M. Ismail and W. Zhuang, “A distributed multi-service resource allocation
algorithm in heterogeneous wireless access medium,” IEEE J. Sel. Areas
Commun., vol. 30, no. 2, pp. 425–432, Feb. 2012.
[14] K. Duffy, D. Malone, and D. Leith, “Modeling the 802.11 distributed
coordination function in non-saturated conditions,” IEEE Commun. Lett.,
vol. 9, no. 8, pp. 715–717, Aug. 2005.
[15] P. Chatzimisios, V. Vitsas, and A. C. Boucouvalas, “Throughput and delay
analysis of IEEE 802.11 protocol,” in Proc. IEEE Int. Workshop Netw.
Appliances, Oct. 2002, pp. 168–174.
[16] J. Z. Zhang and C. X. Xu, “Trust region dogleg path algorithms for
unconstrained minimization,” Annals Oper. Res., vol. 87, no. 10, pp. 407–
418, Apr. 1999.
[17] X. Yang, “Designing traffic profiles for bursty Internet traffic,” in Proc.
IEEE GLOBECOM, Nov. 2002, pp. 2149–2154.
0018-9545 © 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1945
An Efficient Blind Estimation of Carrier
Frequency Offset in OFDM Systems
Said Lmai, Student Member, IEEE, Arnaud Bourré,
Christophe Laot, Senior Member, IEEE, and
Sebastien Houcke, Member, IEEE
Abstract—In this paper, we propose a low-complexity blind car-
rier frequency offset (CFO) estimation scheme for constant mod-
ulus (CM)-signaling-based orthogonal frequency-division multiplexing
(OFDM) systems. Provided that the channel can be assumed to be slowly
time-varying, subcarriers having the same indexes in two consecutive
OFDM symbols will experience nearly the same channel effect. This
assumption enables us to derive a cost function that is determined by the
sum of the products of the signal amplitudes on each pair of equivalent sub-
carriers from two successive OFDM symbols. The maximization process
of this cost function makes it possible to find an appropriate estimate of
the CFO. Over frequency-selective Rayleigh fading channels, the proposed
CFO estimation method provides improved performance over existing
techniques. Moreover, in the context of narrow-band noise and signal gain
variations, the simulations demonstrate the robustness and immunity of
our scheme.
Index Terms—Blind estimation, carrier frequency offset (CFO), con-
stant modulus (CM), narrow-band noise, orthogonal frequency-division
multiplexing (OFDM).
I. I NTRODUCTION
The multicarrier modulation technique in the form of orthogo-
nal frequency-division multiplexing (OFDM) is an efficient scheme
adopted in several standards for systems demanding a high data rate.
In addition to its simple implementation when using a fast Fourier
transform algorithm for discrete Fourier transform (DFT), the OFDM
system is also known for its relatively low sensitivity to time synchro-
nization errors. However, in the presence of carrier frequency offset
(CFO), the orthogonality between subcarriers is quickly destroyed,
and intercarrier interference (ICI) arises, resulting in bit-error-rate
performance loss. CFO is mostly due to transmitter–receiver relative
movement, channel variability, and the imperfect alignment between
local transmitter and receiver oscillators.
Several CFO estimation techniques have been proposed in the
literature. They are either data aided [1]–[4] or simply blind [5]–[10].
This second category, which is also referred to as non-data aided,
may rely on inserting null subcarriers (e.g., [5] and references therein)
and is then power efficient or can be bandwidth efficient in the case
when the OFDM system is fully loaded. Additionally, in [11]–[15],
estimators operating over a post-DFT signal have been proposed. In
[12], it is assumed that the channel response on two adjacent sub-
carriers remains practically unchanged. Therefore, over each OFDM
symbol, the authors proceed by using the power difference minimiza-
tion between two neighboring subcarriers to develop an estimator
adapted to constant modulus (CM) signaling. In [13], another estimator
is introduced, which exploits the slow channel changes in the time
Manuscript received March 29, 2013; revised July 14, 2013; accepted
September 18, 2013 Date of publication October 11, 2013; date of current
version May 8, 2014. This work was supported in part by Fonds Unique
Interministériel, France, through the COMET Project. The review of this paper
was coordinated by Dr. A. J. Al-Dweik.
The authors are with the Telecom Bretagne, Institut Mines-Télécom, Brest
29238, France (e-mail: said.lmai@telecom-bretagne.eu).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TVT.2013.2285479
1946 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 4, MAY 2014

domain over two consecutive OFDM symbols. The differential OFDM where W is the N ×√N normalized I-DFT matrix, whose ele-
system was the context of use discussed in [13]. The same estimator is ments are Wm,n = (1/ N ) exp(j2πmn/N ) by the mth row and
proposed in [15] with further developments and without the constraint the nth column along with m, n = 0, 1, . . . , N − 1. Afterward,
of differential modulation. As reported in [16], let a power difference the signal so far constructed passes through the multipath fad-
estimator in the frequency domain (PDE-F) and its analog in the time ing channel, having the corresponding frequency response Hl =
domain (PDE-T) denote the CFO estimators proposed in [12] and diag([Hl0 , Hl1 , . . . , HlN −1 ]). Let  denote the introduced CFO,
[15], respectively. By minimizing the power difference between all which is normalized with respect to the subcarrier spacing (in this
pairs of subcarriers, on one hand adjacent subcarriers in each OFDM case, (1/Tu )). Thus,  ∈ (−0.5, 0.5), and the resulting accumulated
symbol (PDE-F), or having the same indexes in two successive OFDM phase shift on the OFDM symbol time-domain samples is depicted
  
symbols (PDE-T) on the other hand, estimation functions are derived. by C() = diag([ej2π N ×0 , ej2π N ×1 , . . . , ej2π N ×(N −1) ]). Remov-
In this paper, we propose a novel and efficient blind CFO estimator ing the CP is the first operation performed at reception, and the
for fully loaded CM-OFDM systems. The basic assumption used to retained signal is yl = [yl0 , yl1 , . . . , ylN −1 ]T . This received OFDM
derive the cost function is always that the channel taps slowly change symbol might be described as
in the time domain. Provided that the channel response remains almost

constant over two successive OFDM symbols, we initially assume yl = ej2π N (l−1)(N +Ncp ) C()WHl dl + vl (3)
the minimization of the difference in post-DFT signal amplitudes
between all pairs of subcarriers having the same indexes. Analogous to where vl = [vl0 , vl1 , . . . , vlN −1 ]T is the vector of white Gaussian
PDE-F and PDE-T, let us use ADE-T to refer to the proposed estimator, noise with zero mean and variance σv2 . Note that the resulting
although it stands for an amplitude difference estimator in the time common phase shift relative to the lth OFDM symbol is given by
domain. Using the exhaustive search method, we show that PDE-F, ej2π(/N )(l−1)(N +Ncp ) . In the case of perfect time synchronization,
PDE-T, and ADE-T performance achievements are almost identical. CFO is estimated and then compensated before conducting the DFT
Likewise, the cost function obtained for ADE-T has quasi-regular operation. In other words,  is first estimated by ˆ and after that,
shape and is simple to implement. It can be closely approximated yl is multiplied by C(ˆ )∗ for CFO compensation and by WH for
by a sinusoid with a global maximum corresponding to the desired the DFT process. As a result, we get vector sl = [s0l , s1l , . . . , slN −1 ]T
CFO estimate value. The low-complexity curve-fitting method, as defined by
introduced in [11] and used in [12] and [15], can be efficiently
employed. Compared with PDE-F and PDE-T, ADE-T performance is  ) ∗ yl
sl = WH C(ˆ (4)
identified over frequency-selective fading channels. The realized gain
indicates the reliability of the estimate while applying an appropriate
and the kth element is
tradeoff between performance and complexity. Moreover, ADE-T is
less sensitive to noise and insensitive to gain fluctuations, as discussed
1  m −j2π m (ˆ+k)
N −1
in the Section III. skl = √ yl e N , k = 0, 1, . . . , N − 1 (5)
The remainder of the paper is structured as follows. Section II is N m=0
devoted to describing the system model, including the CFO problem
formulation. In Section III, we present the proposed CFO estima- where
tion technique. Simulation results are given in Section IV. Finally,
ej2π N (l−1)(N +Ncp )  n n j2π n+ m
Section V concludes this paper. Throughout this paper, we denote  N −1

(·)T for complex transposition, (·)∗ for complex conjugation, (·)H for ylm = √ d l Hl e N + vlm ,
complex conjugate transposition, | · | for complex modulus or absolute N n=0
value (of a real), IN for the N × N identity matrix, diag(·) for the m = 0, 1, . . . , N − 1. (6)
diagonal matrix whose entries are the elements of (·), and := for equal
by definition.
As in [12] and [15], developments hereinafter are made under the
noise-free assumption.
II. S YSTEM M ODEL

Let us consider a conventional OFDM transmission system with III. P ROPOSED C ARRIER F REQUENCY
N orthogonal subcarriers. The lth OFDM symbol carries off N O FFSET E STIMATION S CHEME
information symbols gathered as follows dl = [d0l , d1l , . . . , dlN −1 ]T ,
At reception, a proper CFO compensation allows us to have an
after being drawn uniformly from a CM constellation. Then, the
ICI-free signal after the DFT operation. In terms of literary formu-
modulation of vector dl is performed, and the inverse DFT (I-DFT)
)∗ C() = C()∗ C() = WH W = IN ,
lation, if ˆ =  and since C(ˆ
operation output is ul = [u0l , u1l , . . . , ulN −1 ]T . By inserting the Tcp
the resulting sequence after the DFT process is
(Ncp ) duration (number of samples) cyclic prefix (CP), the total dura-
tion of the obtained OFDM symbol for transmission is Tt = Tu + Tcp sl[ˆ=] = Hl dl . (7)
(Nt = N + Ncp ) for a useful part of Tu length and a sampling interval
of Ts = (Tu /N ). Thus, the N useful samples of the lth baseband In the case of CM signaling, i.e., |dkl | = 1, and by trying to focus on
OFDM symbol are the amplitudes, the elements of vector sl will be expressed as
 k  k
1  n j2π n k
N −1
 s l  =  Hl  , k = 0, 1, . . . , N − 1. (8)
ukl = √ dl e N , k = 0, 1, . . . , N − 1 (1)
N n=0
If we assume that the channel response is slowly time-varying and
obtained from therefore |Hlk | and |Hl+1
k
| are almost equal, i.e.,
 k   
ul = Wdl , l = 1, 2, . . . (2) sl−1  ≈ skl [ˆ=] (9)
=]
[ˆ
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 4, MAY 2014 1947

then, to estimate the value of , we initially propose the following cost

function:
 
M N −1
 k   k 2
Ji (˜
) = sl  − sl−1  . (10)
l=1 k=0

CFO is assumed to remain constant over M contiguous OFDM

symbols, and Ji (˜ ) should be minimized with respect to parameter ˜,
which is the trial value of . Thereby

ˆ = arg min Ji (˜
). (11)
˜∈(−0.5, 0.5)


N −1 N −1
However, the two terms k=0
|skl |2 and k=0
|skl−1 |2 are inde-
pendent of ˜ and  (see Appendix A for the proof). Consequently,
minimizing (10) is reduced to maximizing the following final cost
function with respect to ˜:
 
M N −1
 k  k 
Jf (˜
) = sl  sl−1  (12) Fig. 1. Jf versus ζ in the case of N = 64.
l=1 k=0
If we take a = −1/4, b = 1/4, and c = 0 in particular, (15) becomes
so that we can deduce the CFO estimate value ˆ.
simpler.
ItNis−1worth noting that, if we assume the OFDM symbol energy In comparison with [14], for implementation, we use the curve-
( k=0 |skl |2 ) to be almost insensitive to the Doppler effect, maximiz-
fitting method, which is simple without any parameters to set, unlike
ing (12) could be seen as the maximization of the SINR, as introduced
the early–late-gate recovery loop implemented in [14].
in [14]. Indeed, to estimate the CFO, it is proposed that a frequency-
The PDE (PDE-F and PDE-T) cost functions consist of terms where
oriented SINR estimate should be maximized. However, our proposed
the signal amplitude is raised to the fourth power. Therefore, the
scheme is based on a new approach exploiting channel coherence
impact of narrow-band noise on one subcarrier, or even more, would
in time, appropriately. The optimization operation of (12) might be
have a significant paralyzing effect. Concerned subcarriers’ terms will
realized using several techniques. As the use of the gradient descent
drastically reduce the contribution of the other subcarriers’ terms in the
is not possible because the function is not differentiable, exhaustive
sum. Thus, the value obtained by the estimation process will be con-
search on a discrete set of values can be employed. However, it should
siderably different from the real value. In contrast, (12) will experience
be noted that the exhaustive search would require more computation
much less degradation in case of narrow-band noise. This is due to the
time and complexity, thus making it impractical. Therefore, while
low power of all the elements of the sum. Furthermore, the gain control
seeking simplicity of implementation and using the basic assumption
in digital communication systems is an inevitable process. Inappropri-
(9), we come up with a quasi-regular function, which is a good
ate increase or decrease in gain over one OFDM symbol period could
approximation of Jf (˜ ), namely
often occur. In such a case, the ADE-T performance isN −1
not affected.
) ≈ α · cos [2π( − ˜)] + β Indeed, the desired parameter when maximizing |skl skl−1 |
Jf (˜ (13) N −1 k k k=0
(12) or k=0 ρ|sl sl−1 | is exactly the same, where ρ is a (nonzero)
where α and β are constants and, above all, independent of  and ˜. positive real number that corresponds to the gain variation on the
They have positive real values and depend only on the frequency lth OFDM symbol, and k = 0, 1, . . . , N − 1. Thus, the cost func-
channel response and the sequence of information symbols. Through tion used is gain variation independent, unlike the function em-
simulations, we always obtain a quasi-sinusoidal shape for Jf (ζ), ployed in [15].
where ζ =  − ˜. The channel coefficients used are independent com-
plex Gaussian random variables with zero mean and any power delay
profile. In Fig. 1, we have considered Ts -spaced ten-discrete-path IV. S IMULATION R ESULTS
fading channel, which has normally distributed random taps with a
We consider a conventional quadrature phase-shift keying OFDM
decaying power delay profile. Furthermore, in Appendix B, we prove
system having N = 64 subcarriers with a CP length of Ncp = 16.
that the function Jf is 1-periodic and also even. Aiming to perform
Monte Carlo simulations were conducted on a multipath Rayleigh
straightforward maximization operation with less computing time,
fading channel using 106 OFDM symbols on each simulation run.
we decided to make use of the curve-fitting process [11]. Similar
The channel varies every sampling period Ts = (Tu /N ), and Jakes’
to the method description in [15], we first evaluate function (12) at
model [17] is used to generate the fading channel coefficients.  is
three distinctive points: ˜ = a, b, and c. Then, we compute the two
assumed to be uniformly distributed in the range (− 0.5, 0.5) and
intermediate parameters γ and δ defined as
⎧ is kept constant over two consecutive OFDM symbols (M = 1). It
⎪ γ := [Jf (˜ = a) − Jf (˜
 = c)] · [sin(2πb) − sin(2πc)] changes randomly from one pair of OFDM symbols to the next. The
⎪
⎪
⎨ + [Jf (˜ = c) − Jf (˜
 = b)] · [sin(2πa) − sin(2πc)] channel model used is identified by five path delays [0 Ts 2 · Ts 6 ·
(14) Ts 11 · Ts ] with average power gains [0.34 0.28 0.23 0.11 0.04].
⎪
⎪ δ := [J (˜
 = b) − J (˜
 = c)] · [cos(2πa) − cos(2πc)]
⎪
⎩
f f The normalized Doppler frequency fd Tt is set to different values,
 = c) − Jf (˜
+ [Jf (˜  = a)] · [cos(2πb) − cos(2πc)] . where fd designates the Doppler frequency. The underlying OFDM
system operates at a central carrier frequency of 2.6 GHz, whereas
Finally, the CFO estimate is given by the subcarriers’ data rate is 16.68 kb/s. The regular MSE is used to
1 assess the ADE-T performance with respect to that of PDE-F and
ˆ = · arg (γ + j · δ). (15) PDE-T.
2π
1948
Fig. 2. MSE versus SNR using the exhaustive search method over a
frequency-selective channel for fd Tt = 0.030 and 0.042.
Fig. 3. MSE versus SNR using the curve-fitting method over a frequency-
selective channel for fd Tt = 0.030 and 0.042.
A. General Estimator Performance Analysis
First, we perform the maximization process using the exhaustive
search method with a precision value of 10−3 by testing 1000 equally
spaced ˜ candidates over the range (−0.5, 0.5). We are interested
in this optimization method to get an idea about achievement limits
as a benchmark for the three schemes in question. Fig. 2 shows
the results from simulations conducted over a frequency-selective
Rayleigh fading channel with fd Tt = 0.030 and 0.042. These values
correspond to vehicle speeds greater than 100 km/h. One can notice
that the three estimators display very similar levels of performance.
However, the exhaustive search method requires important memory
resources and is very time-consuming to be practical for real-time
wireless communication applications.
Being a good tradeoff between performance and complexity, the
curve-fitting method described in (14) and (15) is employed since
the three functions targeted for optimization could be approximated
by classic sine curves. Note that the exhaustive search method has
involved at least 333 times more operations than curve fitting. Fig. 3
shows the simulation results when the curve-fitting method is used
for the three schemes. The first general observation to make is that,
compared with the exhaustive search results, there is a performance
loss of approximately one decade in terms of MSE. Nevertheless,
ADE-T outperforms PDE-F and PDE-T for both values of fd Tt and
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 4, MAY 2014
Fig. 4. MSE versus SNBNR over a frequency-selective channel for fd Tt =
0.024 and 0.042.
over the entire range of SNRs. The gain achieved with respect to
PDE-T is less significant than that with respect to PDE-F. In the
following, the curve-fitting method will be used.
B. Estimator Performance Exploration in Particular
Scenarios of Interest
Let us now consider the potential contribution of the proposed
method in another important context. Narrow-band noise is an inter-
esting case from a practical perspective (cognitive radio, underwater
acoustic communications, etc.). The phenomenon is even more severe
when the operating frequencies of interferers are constantly changing.
We evaluate the performance of the three schemes over a multipath
fading channel introducing narrow-band noise. Results are given using
MSE versus the signal-to-narrow-band-noise ratio (SNBNR), which
is the mean power of emerging narrow-band interferer signal at the
receiver. Two values of the normalized Doppler frequency are used:
fd Tt = 0.024 and 0.042. Performance levels obtained are shown in
Fig. 4, and we can see clearly that the ADE-T achievement is much
greater than the other estimators.
The third scenario investigated implies operations of driving the
automatic gain control, where the amplitude of the received signal
varies over the duration of one OFDM symbol. As indicated earlier,
one can identify the proposed estimator insentivity to gain changes,
which is not the case for PDE-T. To pinpoint this issue, we have
conducted simulations wherein we vary the gain in the range (−3,
+3 dB). To compare the performance in such situations, the metric
used is MSE versus the gain variation occurred in decibels. As demon-
strated in Fig. 5, where SNR = 15 dB, PDE-F and ADE-T are com-
pletely independent of the gain variation, unlike PDE-T, which suffers
from penalizing sensitivity. Additionally, it is obvious that ADE-T
outperforms PDE-F even if it is insensitive to gain fluctuations. For the
same normalized Doppler frequencies fd Tt = 0.030 and 0.042, and in
the case of SNR = 20 dB, the MSE performance presented in Fig. 6
confirms the behaviors of the estimators.
V. C ONCLUSION
In this paper, we have presented a new method for blind CFO
estimation in fully loaded CM-signaling-based OFDM systems. It has
been proven through numerical simulation results that the proposed
low-complexity technique outperforms prominent existing estimators
in various situations. In harsh time-varying propagation environments,
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 4, MAY 2014 1949

where p = m1 − m2 . Knowing that


N −1 
k N, if p = 0
e−j2π N p = (18)
0, otherwise
k=0

(16) becomes
  2
N −1 
N −1 
N −1
skl  = 1 šn 1 n2 ∗
l šl
2π
ej N mq (19)
N
k=0 n1 ,n2 =0 m=0

N −1
where q = n1 − n2 and m1 = m2 = m. Accordingly, k=0
|skl |2 is
completely independent of ˜ and .

A PPENDIX B

From (16), we have

 −1 N −1 
Fig. 5. MSE versus gain variation occurred on an OFDM symbol of two  k  N  

consecutive ones with SNR = 15 dB and for fd Tt = 0.030 and 0.042.  
N sl =  n
šl j2π m
e N (n+ζ−k)
. (20)
 n=0 m=0 

Therefore
 N −1 N −1
      N −1  m1
N 2 skl  sk∗  
l−1 =  šn 1 n2 ∗
l šl−1
2π
ej N (n1 +ζ−k)

n1 =0 n2 =0 m1 =0


N −1  m2 
−j 2π 
× e N (n 2 +ζ−k)
 . (21)
m =0

2

Now, since
 
N −1
2π
m1 1
ej N (n1 +ζ−k) = (−1)n1 −k eπζ (1− N )
m1 =0

sin(πζ)
π 
1
× ejπ(n1 −k)(1− N ) (22)
sin N (n1 + ζ − k)

Fig. 6. MSE versus gain variation occurred on an OFDM symbol of two and ejπ(n1 −n2 ) = ejπ(n1 +n2 ) = (−1)n1 +n2 = (−1)n1 −n2 , (21) can
consecutive ones with SNR = 20 dB and for fd Tt = 0.030 and 0.042. be expressed as

modeled using a frequency-selective Rayleigh fading channel with  k   k∗   N  
−1 N −1
n1 −n2
N sl  sl−1  = 
2
šn 1 n2 ∗ jπ
l šl−1 e
N
Doppler shift, in a narrow-band noise context and in a situation n =0 n =0
involving fluctuating signal gain, the estimator introduced achieves 1 2

very solid levels of performance. The efficiency carried out is due to 
  .
sin2 (πζ)
the judicious manipulation of the received signal amplitudes to derive × π  π (23)
sin N (n1 + ζ − k) sin N (n2 + ζ − k) 
the appropriate cost function.
Hence, for M = 1, (12) might be formulated as
A PPENDIX A  −1 N −1
−1  N
  
N −1
 1 
N
   n1 n2 ∗
Jf (ζ) = sl  sl−1  =
k k
 šl šl−1
In the noise-free case, substituting (6) in (5) gives N 2

k=0 k=0 n1 =0 n2 =0

ej2π N (l−1)(N +Ncp )  n  j2π m (n+ζ−k)
N −1 N −1



  .
skl = šl e N (16) n1 −n2 sin2 (πζ)
N × ejπ N π  π (24)
n=0 m=0 sin N (n1 + ζ −k) sin N (n2 +ζ − k) 
where ζ =  − ˜, and šn n n
l := dl Hl . Since we are interested in the Now, let us show that Jf (ζ) is periodic with period 1. From (24),
signal amplitude, we obtain we have
  2
N −1 
N −1 
N−1  N−1 N−1
skl  = 1 šn 1 n2 ∗ 1     n1 n2 ∗ jπ n1−n2
2 l šl Jf (ζ +1) = 2  šl šl−1 e N
N N 
k=0 n1 ,n2 =0
k=0 n1 =0 n2 =0


N −1 
N −1 
  .
ζ 2π (m n −m n ) k
−j2π N sin2 (πζ +π)
× e j2π N p j
e N 1 1 2 2
e p
(17) × π  π (25)
m1 ,m2 =0 k=0
sin N
(n1 +ζ −(k−1)) sin N (n2 +ζ −(k−1)) 
1950 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 4, MAY 2014


which can be expressed otherwise, when k = k − 1, as N
   n1 n2 ∗ j Nπ (n1 −n2 +N N +1 )
−1 N −1

= šl šl−1 e
Jf (ζ +1)

n1 =0 n2 =0
 
N −2  N −1 N −1 
1     n1 n2 ∗ jπ n1 −n2 π
sin2 (πζ)
 π  
= 2  šl šl−1 e N × (31)
N   sin N (n1 + ζ − k) sin N (n2 + ζ − k) 
k =0 n1 =0 n2 =0

 where n1 − n2 + N N +1 ≈ N N +1 and |ejπN
N +1 N
| = |ejπN | = 1.
 
sin2 (πζ)
× π  π We can conclude that Jf (−ζ) ≈ Jf (ζ).
sin N (n1 +ζ −k ) sin N (n2 + ζ − k ) 
 N −1 N −1
  R EFERENCES
1  n −n2
n2 ∗ jπ 1
+ 2  l × šl−1 e
šn 1 N
[1] T. Schmidl and D. Cox, “Robust frequency and timing synchronization
N  for OFDM,” IEEE Trans. Commun., vol. 45, no. 12, pp. 1613–1621,
n1 =0 n2 =0
 Dec. 1997.
 [2] M. Morelli and U. Mengali, “An improved frequency offset estimator
  . (26)
sin2 (πζ)
× π  π for OFDM applications,” IEEE Commun. Lett., vol. 3, no. 3, pp. 75–77,
sin N (n1 + ζ + 1) sin N (n2 + ζ + 1)  Mar. 1999.
[3] H. Minn, V. Bhargava, and K. Letaief, “A robust timing and frequency
synchronization for OFDM systems,” IEEE Trans. Wireless Commun.,
Since vol. 2, no. 4, pp. 822–839, Jul. 2003.
π  π  [4] G. Ren, Y. Chang, H. Zhang, and H. Zhang, “An efficient frequency offset
sin (n1 + ζ + 1) sin (n2 + ζ + 1) estimation method with a large range for wireless OFDM systems,” IEEE
N N Trans. Veh. Technol., vol. 56, no. 4, pp. 1892–1895, Jul. 2007.
π  [5] S. Attallah, Y. Wu, and J. Bergmans, “Low complexity blind estimation
= sin (n1 + ζ − (N − 1)) of residual carrier offset in orthogonal frequency division multiplexing
N based wireless local area network systems,” IET Commun., vol. 1, no. 4,
π  pp. 604–611, Aug. 2007.
× sin (n2 + ζ − (N − 1)) . (27) [6] B. Park, H. Cheon, E. Ko, C. Kang, and D. Hong, “A blind OFDM syn-
N chronization algorithm based on cyclic correlation,” IEEE Signal Process.
Lett., vol. 11, no. 2, pp. 83–85, Feb. 2004.
We obtain [7] M. Shi, Y. Bar-Ness, and W. Su, “Revisiting the timing and frequency
N −1  N −1 N −1
 offset estimation based on cyclostationarity with new improved method,”
1     n1 n2 ∗ jπ n1 −n2 IEEE Commun. Lett., vol. 13, no. 7, pp. 537–539, Jul. 2009.
Jf (ζ + 1) = 2  šl šl−1 e N [8] M. Tanda, “Blind symbol-timing and frequency-offset estimation in
N   OFDM systems with real data symbols,” IEEE Trans. Commun., vol. 52,
k =0 n1 =0 n2 =0
no. 10, pp. 1609–1612, Oct. 2004.

 [9] W.-L. Chin, “ML estimation of timing and frequency offsets using
 
sin2 (πζ)
× π  π (28) distinctive correlation characteristics of OFDM signals over dispersive
sin N (n1 + ζ − k ) sin N (n2 + ζ − k )  fading channels,” IEEE Trans. Veh. Technol., vol. 60, no. 2, pp. 444–456,
Feb. 2011.
[10] H. T. Hsieh and W. R. Wu, “Maximum likelihood timing and carrier
and therefore, Jf (ζ + 1) = Jf (ζ). Moreover, let us prove that Jf (ζ) frequency offset estimation for OFDM systems with periodic pream-
is even. Still from (24), we get bles,” IEEE Trans. Veh. Technol., vol. 58, no. 8, pp. 4224–4237,
Oct. 2009.

N −1  N −1 N −1 [11] Y. Yao and G. Giannakis, “Blind carrier frequency offset estimation in
1     n1 n2 ∗ jπ n1 −n2 SISO, MIMO, and multiuser OFDM systems,” IEEE Trans. Commun.,
Jf (−ζ) = 2  šl šl−1 e N
N  k=0 n1 =0 n2 =0
vol. 53, no. 1, pp. 173–183, Jan. 2005.
[12] X. Zeng and A. Ghrayeb, “A blind carrier frequency offset estimation
 scheme for OFDM systems with constant modulus signaling,” IEEE

 
sin2 (πζ) Trans. Commun., vol. 56, no. 7, pp. 1032–1037, Jul. 2008.
× π  π (29) [13] X. Zeng and A. Ghrayeb, “CFO estimation schemes for differential
sin N (−n1 + ζ + k) sin N (−n2 + ζ + k)  OFDM systems,” IEEE Trans. Wireless Commun., vol. 8, no. 1, pp. 124–
129, Jan. 2009.
knowing that [14] W. Chin and S. Chen, “A blind synchronizer for OFDM systems based
on SINR maximization in multipath fading channels,” IEEE Trans. Veh.
π  π  Technol., vol. 58, no. 2, pp. 625–635, Feb. 2009.
sin (−n1 + ζ + k) sin (−n2 + ζ + k) [15] A. Al-Dweik, A. Hazmi, S. Younis, B. Sharif, and C. Tsimenidis,
N N “Carrier frequency offset estimation for OFDM systems over mobile
π  π  radio channels,” IEEE Trans. Veh. Technol., vol. 59, no. 2, pp. 974–979,
= sin (n1 + ζ − k ) sin (n2 + ζ − k ) (30) Feb. 2010.
N N [16] S. Younis, A. Al-Dweik, C. Tsimenidis, B. Sharif, and A. Hazmi, “The
effect of timing errors on frequency offset estimation in OFDM systems,”
where n1 = N − 1 − n1 , n2 = N − 1 − n2 , and k = N − 1 − k. in Proc. 11th IEEE Int. Symp. Signal Process. Inf. Technol., Bilbao, Spain,
The term ejπ((n1 −n2 )/N ) in (24) and (29) does not have any significant Dec. 2011, pp. 202–206.
weight at all since we can, for instance, take in the following equalities: [17] T. S. Rappaport, Wireless Communications: Principles and Practice.
Englewood Cliffs, NJ, USA: Prentice-Hall, 2002.
 N −1 N −1
 
 1 n2 ∗ jπ
n1 −n2
 šn
l šl−1 e
N

n1 =0 n2 =0


 
sin2 (πζ)
× π  π
sin N (n1 + ζ − k) sin N (n2 + ζ − k)