Wireless Pers Commun (2015) 81:225238

DOI 10.1007/s11277-014-2125-0

Novel Time and Frequency Synchronization Techniques

for OFDM Systems in Double Selective Fading Channel
Hossam Asran Ehab F. Badran Amira I. Zaki

Published online: 18 October 2014

Springer Science+Business Media New York 2014

Abstract OFDM receiver performance is affected by the time offset and the carrier frequency offset, as these two parameters have sever effect on the signal reception quality. In
this paper, two novel schemes for time estimation (TE) and frequency estimation (FE) are
proposed to overcome the time and carrier frequency offset problem, and therefore improve
the performance of the OFDM reception. The proposed techniques are based on the fact that,
using the correlation of constant amplitude zero auto correlation sequence with a unity peak
to average power ratio gives a sharp time metric peak. The proposed timing symbol based
on this fact introduces a TE technique that has a sharp peak at the correct time instant and
no side lobes. In addition, the proposed FE technique is based on the same training symbol
and it takes place in two steps. The first step is the fine FE technique, which depends on
calculating the phase difference between the cyclic prefix and the preamble tail. It is referred
to as fractional FE. The second step is the coarse offset FE and it is referred to as integer FE.
The coarse offset FE is based on FFT and it has less mean square error than other methods.
Keywords

OFDM Frame synchronization Frequency offset compensation

1 Introduction
The developing of new services with promising high data rate and quality of service (QoS),
like the 4G wireless networks, increased the demand on free frequency bands. Due to the

H. Asran E. F. Badran (B) A. I. Zaki

Department of Electronics and Communication Engineering, Arab Academy
for Science, Technology and Maritime Transport, Alexandria,
Egypt
e-mail: ebadran@aast.edu
H. Asran
e-mail: hossamasran@gmail.com
A. I. Zaki
e-mail: amzak10@aast.edu

123

226

H. Asran et al.

scarcity of free bands in the spectrum, researches introduce advanced techniques to overcome
the problem of the spectrum scarcity, increase the system capacity and at the same time
provide reasonable QoS. One of the most important techniques is the orthogonal frequency
division multiplexing (OFDM) technique.
The OFDM has been widely used in different systems like digital audio/video broadcasting
(DA/VB), it has been selected due to its high spectral efficiency and its ability to overcome
the multipath fading effect [1]. This made the OFDM the downlink technique used in the
mobile communication system long-term evolution (LTE) system.
The OFDM system transmits data as a set of low parallel data rate streams over orthogonal
narrowband subcarriers (100 Hz50 KHz) [2,3]. The OFDM transmission can be simply
generated using inverse fast Fourier transform (IFFT) at the transmitter side, and received
using fast Fourier transform (FFT) at the receiver side. The OFDM can provide very high
data rate by using large number of carriers. The orthogonality of the OFDM subcarriers
eliminates the possibility of the inter carrier interference (ICI) and inter symbol interference
(ISI). The cyclic prefix insertion in the OFDM symbols, which implies that the last part of
the OFDM symbol is copied and inserted at the beginning of the OFDM symbol, preserves
the subcarrier orthogonality in the case of a time dispersive channel, as long as the channel
dispersion time is shorter than the cyclic prefix length [2].
The synchronization between the transmitter and the receiver is another important factor
that affects the system performance. However, loss of synchronization leads to time and
frequency shifts that will cause ISI and ICI. Thus, the loss of synchronization will degrade
the performance of the system [2,3].
The OFDM frequency and time synchronization is divided into two main categories the
data-aided and non-data-aided categories ([410] and the references therein). The data-aided
method, which depends on a training sequence or pilot symbol for the estimation of both
of the time offset and the frequency offset. This method has the advantage of using simple
calculations and providing high accuracy, but it has the disadvantage of wasting the bandwidth
and the data transmission rate.
The non-data-aided methods usually use the cyclic prefix for estimation. These methods
do not waste the bandwidth or data speed, but the estimation range is limited.
In this paper, some of the previously mentioned time and frequency synchronization
schemes are discussed and compered to the two proposed techniques. The rest of the paper
is organized as follows. Section 2 describes the OFDM system mathematical model. Section
3 presents and compares the previous time synchronization schemes. Section 4 presents the
proposed time offset estimation algorithm. Section 5 covers the proposed fine and coarse
frequency synchronization method. The computer simulation and results are carried out in
Sect. 6. Finally, Sect. 7 presents the conclusions.
2 OFDM System Model
The OFDM system is shown in Fig. 1. The OFDM data can be represented mathematical by
x (n) =

M1


S(k) e

j2 kn
N

N = 0, 1, 2, . . . , N 1

(1)

k=0

where S(k) is the complex modulated symbol modulating the kth subcarrier. The frequency
of the kth sub-carrier is given by f k = kn/N and n is the sample number. The transmitted
signal will be affected by multipath Rayleigh fading wireless channel which has an impulse
response that is given by

123

Double Selective Fading Channel

227

Fig. 1 OFDM system block

diagram

s0
s1

s0, s1, sN-1

S/P

IFFT

sN-1
Channel

~s s~ s~
0, 1, N-1

P/S

FFT

N-1

Fig. 2 Schmidl training symbol

N/4
B

I (n) =

L1


h (l) (n l ),

Time &
Freq.
Sync.

N/4
B

(2)

l=0

where L is the number of multipath component, l is the time delay and h (l) is the gain of
the lth multipath component, which must not exceed the length of CP. Thus, the received
signal can be expressed as
r (n) =

L1


h (l) x(n l ) e j2 n/N ,

(3)

l=0

where is the carrier frequency offset due to the receiver oscillator mismatch and the Doppler
shift of the moving mobile user.

3 OFDM Synchronization Schemes

This section introduces a brief description of the research work previously introduced for
OFDM system synchronization. The time synchronization is essential in OFDM systems to
keep the orthogonality of the sub-carriers and to avoid the ICI and ISI. The target in time
synchronization is to estimate the time offset and to find the beginning of the OFDM frame.
3.1 Schmidl and Cox Method
In Schmidl and Cox [4] introduce a synchronization technique based on generating two
identical parts of training symbols in the time domain each of length N/2 as shown in Fig. 2
(where B an m-sequence).

123

228

H. Asran et al.

schmidl

Timing Metric

1
0.8
0.6
0.4
0.2
0

50

100

150

200

250

300

350

400

Time (sample)
Fig. 3 Schmidl and Cox method timing metric in an ideal channel

The time metric introduced by Schmidl [4] can be computed by the correlation of these
two parts. In this technique the beginning of the symbol duration can be estimated at the
maximum point of the time metric as follows [4]
M (m) =

|P (m)|2
,
(R(m))2

(4)

where
N

P (m) =

2 1


n=0



N
,
r (m + n) r m + n +
2

N

2 1  


N 2

R (m) =
r m + n + 2  ,

(5)

(6)

n=0

where r () is training symbol. Thus the time offset will be as follows [4]
= arg max (M (m)) ,

(7)

and the frequency offset will be

angle(P( ))
.
(8)
f =

However this time metric suffer from plateaus which mean uncertain estimation of the
beginning of the frame especially at low SNR. Figure 3 shows the timing metric in an
ideal channel (no noise) and Fig. 4 shows the timing metric in an AWGN channel with low
SNR = 1 dB.
3.2 Minns Method
Minns proposed a different training symbol in [5] as a modification to the Schmidl and Cox
Method [4]. Minns training symbol can be designed in time domain as shown in Fig. 5.
As shown in the Figure, Minns symbol has four equal length parts of PN sequence each of

123

Double Selective Fading Channel

229
schmidl algorithm
SNR=1dB

Timing Metric

1
0.8
0.6
0.4
0.2
0

50

100

150

200

250

300

350

400

Time (sample)

Fig. 4 Schmidl and Cox method timing metric in an AWGN channel with SNR = 1 dB

Fig. 5 Minns training sysmbol

Fig. 6 Minns training symbol

N/4

N/4

N/4

N/4

-B

-B

N/4

N/4

N/4

N/4

length N/4. The first two parts are identical and the other two marts are the negative version of
the first two parts. This negative sign is used to overcome the time metric plateau of Schmidl
algorithm, and gives a peak at the starting of the OFDM symbol.
Thus, the time metric can be computed using Eq. (4), Where
N

P (m) =

1 
4 1

l=0 n=0

 


N
N
N
r m +n +l
+
r m + n + l
2
2
4

N

1 
4 1  


N
N 2

R (m) =
r m + n + l 2 + 4  .

(9)

(10)

l=0 n=0

Although, Minns method [5] overcomes the plateau problem it still suffers from large Mean
square error (MSE) due to the large variance of the time metric loops, which makes it
unreliable to be used especially in a multipath Rayleigh fading environment.
3.3 Parks Method
In order to enhance the time offset estimation, Park [6] introduces a new symbol with four
portions. Every two closed symbols are of great difference. In Parkers method, the OFDM
training symbol can be generated in time domain as shown in Fig. 6.
where A is an m-sequence with a length of N/4 passed throw IFFT, and B is the symmetric
time version of the conjugate of A. The time metric can be computed by Eq. (4), where

123

230

H. Asran et al.

Park

Timing Metric

0.8
0.6
0.4
0.2
0

50

100

150

200

250

300

350

400

Time (sample)
Fig. 7 Parks method timing metric in an ideal channel
Fig. 8 Chois training symbol

N/2

N/2

D
Chois training symbol

P (m) =

4 1


n=0

 


N
N
n r m +n +
1
r m +
2
2

(11)

R (m) =

4 1


|r (m + n 1)|2 ,

(12)

n=0

and the time offset will be calculated using Eq. (7). Due to the difference between the two
adjacent portions of the symbol, Park method gives a sharp peak at the correct time sample
as shown in Fig. 7, which illustrates the timing metric in an ideal channel. This peak has very
small variance, which means a better estimation in the ISI environment when compared to
both methods of Minn and Schmidl.
Although the sharp peak of Parks scheme, the method suffers from large side lobes at
N/4 samples around the main lobe due to the repetition of portion one in portion three, which
will degrade the scheme performance at low SNR with frequency selective channel [6].
3.4 Chois Method
Choi has introduced a time domain training symbol using constant amplitude zero autocorrelation (CAZAC) sequence in [7]. Chois training symbol is generated in time domain as
shown in Fig. 8.
where C is the IFFT output of N/2 (CAZAC) sequence and D is the complex conjugate of
the time reversed version of C. Again time metric is defined using Eq. (4), using the following
P(m) and R(m) functions [7]
N

P (m) =

2 1


n=0

123

r (m n) r (m + n + 1)

(13)

Double Selective Fading Channel

231

N
2
2 1  


1 
r m + n N  ,
R (m) =

2
2 

(14)

n=0

The time offset will be calculated using Eq. (7). The Chois time metric gives a very sharp
peak at the correct time symbol. However, a small minor loop which affects the estimation
in the very low SNR. This can be concluded from Fig. 9, which illustrate timing metric of
Chois scheme in an ideal channel.

4 Proposed Time Synchronization Scheme

Although the time synchronization scheme introduced by Choi [7] has a sharp peak at the
correct time sample, it has a minor lobe which will affect the estimation performance at low
SNR. As a modification of Chois training symbol, this paper proposes a new training symbol
which eliminates the minor lobes occurred in Chois symbol. The new symbol is generated
by multiplying the whole symbol of Choi with an m-sequence of length N. So that it can be
constructed as illustrated in Fig. 10 and will have the form shown in Fig. 11 in time domain.
c (k) = e

jrk 2
N
2

, k = 1, 2, 3, . . .

N
,
2

(15)

where CC is the IFFT of c (k) multiplied by the first N2 part of the m-sequence, and DD is
complex conjugate of the time reversed version of c (k) after the IFFT multiplied by the last
Fig. 9 Chois method timing
metric in an ideal channel

Choi

Timing Metric

1
0.8
0.6
0.4
0.2
0

50

100

150

200

250

300

350

400

Time (sample)
Fig. 10 Proposed time offset
estimation method

Fig. 11 The proposed training

symbol

N/2

N/2

CC

DD

123

232

H. Asran et al.

Proposed

Sharp peak at the

beginning of the frame

Timing Metric

0.8
0.6
0.4
0.2
0
0

50

100

150

200

250

300

350

400

Time (sample)

Fig. 12 The timing metric of the proposed time offset technique

N
2

part of the m-sequence. The time metric is defined using Eq. (4), using the following P(m)
and R(m) functions


P (m) =

2 1


n=0

rr




N
N
n rr
+n+1
2
2

N
2

2 1 


1 
rr m + n N 
R (m) =

2
2 

(16)

(17)

n=0

Thus the time offset will be

= arg max(M (m))
The proposed time metric outperforms the previous methods as it gives a very sharp peak
at the starting instant of the frame, and at the same time eliminates the side lobe appeared
in Parks and Chois methods. The timing metric of the proposed time offset estimation
technique is shown in Fig. 12.

5 Proposed Frequency Synchronization Scheme

The carrier frequency offset (CFO) estimation is an important process in the OFDM system
as the time offset estimation, to ensure the high performance of the OFDM system. Carrier
frequency errors, which are created due to different factors such as differences in sampling
clock frequencies of the transmitter and the receiver and clock jitter result in a shift of the
received signal spectrum in the frequency domain. If the frequency error is an integer multiple
of the subcarrier spacing, then the subcarriers are still mutually orthogonal, but the received
data symbols, which are mapped to the OFDM spectrum, are in the wrong position in the
demodulated spectrum, resulting in large bit error rate (BER). If the CFO is not an integer
multiple of the subcarrier spacing, then energy spills over between the subcarriers, resulting
in loss of their orthogonality causing ICI.
This section presents two techniques to perform the CFO estimation. The first one for the
fractional part as in [4] in which the offset will be less than one cycle and the other is the

123

Double Selective Fading Channel

233

( + )

FFT

G(k)

( )
{

( )}

Max

Fig. 13 Proposed coarse frequency offset estimation method

proposed technique for the integer part which can estimate frequency offset as wide as the
whole OFDM symbol.
5.1 Schmidl and Cox Method for Fine CFO Estimation
In fractional/fine CFO the frequency mismatch is less than one subcarrier spacing. This
small frequency error happens manly in slow varying environment, usually indoor. It that
can be easily compensated by using non-data-aided methods in order not to lose the spectral
efficiency. As the cyclic prefix contains redundant information that can be used to estimate
the fine frequency offset using:




Ng

angle
k=1 r k r k + N
fi =
(18)
2
To improve the overall performance, this method can be carried out on the next data OFDM
frames then the average value is taken.
5.2 Proposed Coarse/Integer Frequency Estimation
After the fine frequency offset has been compensated, integer frequency estimation takes
place. The proposed system is shown in Fig. 13 and uses the
symbol as the training
 proposed

symbol. Consider the received fine compensated symbol r + k and k = 0, 1, 2, . . . , N 1
then multiply it by PN sequence pn (k)


(19)
r (k) = r + k pn (k) ,
As the transmitted training symbol is known to the receiver then
G (k) = r (k) Con j{Seq (k)}

(20)

where Seq(k) is the transmitted training symbol and G(k) will contain the phase difference.
The difference between the transmitted training symbol Seq (k) and the received symbol is
located only in the phase so if G(k) is passed through FFT that means convolution between
G (k)and Con j{Seq (k)} in the frequency domain, then the output will gives a peak at the
correct shift. This shift happens due to the offset in the other domain. Then the Integer
frequency offset can be calculated by finding the maximum of the output of the FFT as
follows
fc = Max {F F T {G (k)}}

(21)

The output of the proposed coarse CFO estimation is shown in Fig. 14. It is concluded
from Fig. 14 that the proposed technique succeeded to define the correct CFO.

123

234

H. Asran et al.

Fig. 14 Proposed coarse CFO estimation at 100.35 Hz

Table 1 Simulation parameters
No. of subcarriers

N = 256

Length of CP

N g = N /4 = 32

Modulation type

Mod = QPSK

Channel model

AWGN + Rayleigh fading

Channel delays

[0

Channel gains

[1 0.8913 0.3548 0.3162

1 2 3 4] sample time
0.1000]

It is also worth to compare the proposed technique to the technique previously introduced
by Wan [8] from the complexity side. The complexity of the CFO estimator is most probably
due to the computational complexity carried out by the calculations in the FFT phase. As
shown in Fig. 13, the proposed method uses only one FFT calculation per estimation and
gives a sharp peak at the correct integer frequency shift. On the other hand, the method
proposed in [8] uses N number of FFT at least per estimation, which causes much higher
computational load and hardware power.

6 Simulation and Results

In this section, simulation results are carried out using MATLAB. The OFDM system parameters are defined in Table 1.
The comparison between traditional estimation algorithms presented in Sect. 3 shows that
Chois technique [7] outperforms the other techniques. The comparison shows the plateau of
Schmidl time metric [4] with width equal to the CP length. Park [5] and Choi [6] methods
gives a peak at the end of the CP. Moreover, Park and Choi methods give side lobes in the
timing metric pattern.
In this section, simulation of the proposed time estimation technique takes place and
compared to the Chois method one in a multipath Rayleigh fading channel. The simulation
results are demonstrated in Fig. 15. By comparing the results shown in Fig. 15, it is concluded
that, the proposed scheme has a low variance and a sharp peak at the correct sample while
Choi [7] has wider variance of the main lobe and a plateaus due to the channel dispersion. It
is also observed, that the time metric of the proposed technique suffers no side lobes.
This section also presents the simulation results of the proposed CFO estimation algorithm.
Figures 16 and 17 compares the performance of the proposed Fine CFO estimation method
with Chois method [7] in both AWGN and Rayleigh fading channel respectively. Figure 16
shows that the proposed Fine estimation has less mean square error in AWGN channel then

123

Double Selective Fading Channel

235
proposed

Timing Metric

0.4
0.3
0.2
0.1
0
290

295

300

305

310

315

320

310

315

320

Time (sample)
Choi

Timing Metric

0.4
0.3
0.2
0.1
0

290

295

300

305

Time (sample)

Fig. 15 Proposed and Choi [7] time metric at Rayleigh fading channel

10

-2

Mean Square Error

Proposed

10
10
10
10
10

Choi
Seung

-3

-4

-5

-6

-7

10

15

20

25

30

Fig. 16 Fine CFO estimation in a AWGN channel

10

-2

Mean Square Error

Proposed

Choi
10

10

10

10

-3

-4

-5

-6

10

15

20

25

30

SNR (db)

Fig. 17 Fine CFO estimation in a multipath Rayleigh fading channel

123

236

H. Asran et al.
300

FFT(G(k))

Proposed coarse CFO Estimation

200
100
0
0

50

100

150

200

250

300

Frequency Sample

FFT(G(k))

0.2

[8] Method

0.15
0.1
0.05

0
0

50

100

150

200

250

300

Frequency Sample
Fig. 18 Coarse CFO estimation at SNR = 15 dB over AWGN channel

FFT(G(k))

200
Proposed coarse CFO Estimation

150
100
50
0
0

50

100

150

200

250

300

Frequency Sample
FFT(G(k))

0.8
[8] Method

0.6
0.4
0.2
0

50

100

150

200

250

300

Frequency Sample

Fig. 19 Coarse CFO estimation at SNR = 0 dB over Rayleigh fading channel

the Chois method in AWGN channel for different SNR. On the other hand, Fig. 17 indicates
that proposed estimation algorithm has better mean square error performance in a multipath
Rayleigh fading channel for SNR up to 15 dB. This is in addition to the less complexity of
the proposed technique when compared to the traditional ones as stated before.
Figure 18 shows that the proposed coarse estimation method has a sharp peak at the correct
frequency shift in AWGN channel. Although the method proposed in [8] also have a correct
peak, it has large ground noise which lead to miss estimation in the low SNR and in large
offsets. In Fig. 19, the signal is passing through a Rayleigh fading channel which leads to
side lobes in both methods. However, the proposed side lobes are small compared to the main
one.

123

Double Selective Fading Channel

237

7 Conclusions
In order to improve the performance of the OFDM Synchronization method, this paper
proposed time and CFO Synchronization schemes with a CAZAC sequence. The proposed
schemes use the property that the IFFT CAZAC sequence is a CAZAC. Several time and CFO
Synchronization schemes [48] have been presented and explained. Computer simulation
results show performance comparisons and show that the proposed methods achieve superior
performance with simple and accurate computational load compared to the existing methods
in both AWGN and Rayleigh fading channels.

References
1. Schulze, H., & Lueders, C. (2005). Theory and applications of OFDM and CDMA: Wideband wireless
communications. New York: Wiley.
2. Ratasuk, R., Tolli, D., & Ghosh, A. (2010). Carrier aggregation in LTE-advanced. In Proceedings of IEEE
71st vehicular technology conference VTC-2010 (pp. 15). Taipei.
3. Dahlman, E., Parkvall, S., & Skld, J. (2011). 4G LTE/LTE-advanced for mobile broadband. Amsterdam:
Elsevier Ltd.
4. Schmidl, T., & Cox, D. (1997). Robust frequency and timing synchronization for OFDM. IEEE Transactions on Communications, 45(12), 16131621.
5. Minn, H., Zeng, M., & Bhargava, V. K. (2000). On timing offset estimation for OFDM systems. IEEE
Communication Letters, 4(7), 242244.
6. Park, B., Cheon, H., & Kang, C. (2003). A novel timing estimation method for OFDM systems. IEEE
Communications Letters, 7(5), 239241.
7. Choi, S. D., Choi, J., M., & Lee, J. H. (2006). An initial timing offset estimation method for OFDM systems
in Rayleigh fading channel. Proceedings of IEEE 64th vehicular technology conference VTC-2006.
8. Wang, H., et al. (2012). A novel synchronization algorithm for OFDM systems with weighted CAZAC
sequence. Journal of Computational Information Systems, 8(6), 22752283.
9. Boshehba, S. A., Badran E. F., & Mahmoud, M. (2013). A modified blind deterministic carrier frequency
offset estimator for OFDM systems. Proceedings of the 2013 JapanEgypt conference on electronics,
communications and computers (JEC-ECC2013) (pp. 1822). Cairo, Egypt, 1719.
10. Badran, E. F., Samara M., & Aly, M. H. (2013). A novel frame synchronization scheme via wavelet
packet transform for OFDM systems. Proceedings of the 2013 JapanEgypt conference on electronics,
communications and computers (JEC-ECC2013) (pp. 1217). Cairo, Egypt, 1719.

Hossam Asran received the B.Sc. degree (Honors) in electronics

and communications engineering from Arab Academy for Science and
Technology and Maritime Transport university, Alexandria, Egypt, in
August 2008. Received the M.Sc. degree in August 2013.

123

238

H. Asran et al.

Ehab F. Badran received the B.Sc. degree with honors and M.Sc.
degree in electrical engineering from Assiut university, Assiut, Egypt,
in May 1995 and March 1998, respectively, and the M.Sc. and Ph.D.
degrees in electrical engineering from Louisiana State University
(LSU), Baton Rouge, USA, in May 2001 and May 2002, respectively.
From 1995 to 1998, he was an instructor with the department of electrical engineering, Assiut university, where in May 1998, he was promoted to assistant lecturer. From January 2000 to May 2002, he was
a teaching and research assistant with the department of electrical
and computer engineering, Louisiana State University during his Ph.D.
studies. From September 2002 to August 2003, he was an assistant professor with the department of electrical engineering, Assiut University.
From September 2003 to May 2007, he worked as an assistant professor in the department of electronics and communication engineering,
Arab Academy for Science and Technology and Maritime Transport,
Alexandria, Egypt. From June 2007 to May 2011, he was an associate
professor in the same department. In June 2011, he was promoted to be a professor. His research and teaching
interests are in wireless communications, signal processing, MIMO systems, and image signal processing.
Prof. Dr. Ehab Farouk Badran was the vice dean of college of engineering and technology for educational
affairs from 12 January 2012 until 14 March 2014.
Amira I. Zaki received the B.Sc. degree (Honors) in electrical engineering from Arab Academy for Science and Technology and Maritime Transport university, Alexandria, Egypt, in August 2000. Worked
as a GTA from 2000 till 2005 in the same university. Received the
M.Sc. degree in June 2004. She was promoted in 2004 and worked as a
teacher assistant in the department of electronics and communications
in the Arab Academy for Science and Technology and Maritime Transport university. She received the ph.D. from the faculty of engineering,
Alexandria University, Alexandria, Egypt in April 2004 and promoted
to work as a teacher in the Arab Academy for Science and Technology
and Maritime Transport, Alexandria, Egypt. Her current research areas
of interest include wave propagation in different media, smart antenna
arrays, spacetime coding, communication systems, fractal antenna,
wireless communications, signal processing, MIMO systems and UWB
system.

123