Documente Academic
Documente Profesional
Documente Cultură
Abstract In this paper, we propose an efficient scheme using a orthogonality between sub-carriers. The lower is the sub-
new single-symbol preamble structure for OFDM data-aided carriers spacing, the more is the sensitivity to frequency offsets.
synchronization. The preamble symbol is constituted of two
successive equal-duration sub-symbols, the second one of which is The timing and frequency offsets deteriorate modulation
a differentially encoded version of the first one, using an performance. To face those limitations, many approaches have
adequately designed precoding sequence. This structure leads to been proposed in the literature to estimate the time and/or the
almost perfect channel autocorrelation profile with sharp in- frequency offsets. Some of them exploit the redundancy in the
phase peak and zero out-of-phase correlation. For comparison, CP, as the work of Van de Beek in [3], for blind
we review some previous methods also based on data-aided synchronization, which is more suited to continuous flow
synchronization for OFDM systems. We here explore the transmission, such as streaming video. Others use a preamble
performance of the new method in the case of the AWGN before OFDM symbols data transmission, as in [1] and [2].
channel. The performance is presented in terms of correct frame Those approaches, known as data-aided synchronization
start detection rate and estimators variance for time estimation. methods, cost in terms of bandwidth but are generally more
For fractional frequency offset estimation, the performance is efficient than blind ones, especially for bursty packet traffic. In
evaluated in terms of mean squared error. The obtained results between, we have semi-blind synchronization that uses the
prove that the accuracy of the frame start detection and the redundancy in the CP in addition to pilot symbols sent regularly
fractional frequency offset estimation are greatly enhanced, even for channel estimation, as presented in [4].
at very low SNRs. We show that compared to the best
benchmark, gains of about 5 dB and 20 dB are achieved for This paper addresses the problem of data-aided
timing precision of one sample and of one tenth of a sample, synchronization. It is organized as follows. Section II briefly
respectively. We also show that for a target correct detection rate defines the OFDM signal. In section III, we study some existing
of 95%, our approach provides more than 17 dB gain with synchronization methods presented in [1] and [2], with the
respect to the best benchmark. Although we concentrate here on intention of using them as benchmarks. Then, we describe our
OFDM, the proposed approach can be applied to other systems synchronization approach, in section IV. Simulations results
such as TDMA or CDMA. and discussions are given, in section V. Finally, our conclusions
and perspectives are provided in Section VI.
Keywords- OFDM; preamble-based synchronization; precoding;
preferred pair of m-sequence; time synchronization; frequency II. OFDM SIGNAL
synchronization; preamble sequence design
The OFDM signal is generated by modulating equally
I. INTRODUCTION spaced sub-carriers, using QAM or PSK complex data symbols.
Orthogonal Frequency Division Multiplexing (OFDM) These symbols denoted ck , n are indexed by the OFDM symbol
systems have improved both wired and wireless
number k and the sub-carrier number n (n = 0,1,..., N u 1).
communications thanks to their high data rate transmission and
robustness to channels imperfections. A misalignment between the transmitted baseband OFDM signal after pulse shape
the sent symbol start and the demodulated one (timing error) filtering is given by
can introduce Inter-Symbol Interference (ISI). The use of a Nu
Cyclic Prefix (CP) longer than the channel delay spread renders
s ( t ) = ck , n e , t [ kT , ( k + 1)T ],
j 2 f n ( t kT Tg )
the OFDM systems robust to timing offsets lower than the CP (1)
n =0
excess length. As the CP length is generally chosen to sensibly
equal the channel delay spread, the timing offset must be th
where N u is the number of sub-carriers. The n sub-carrier
accurately recovered to avoid the ISI. A frequency error results
in Inter-Carrier Interference (ICI) that destroys the frequency is denoted by f n = n / Tu , where Tu is the length of
j 2 ( l N g )
P1 (d ) = rd*+ m rd + m + L , (5)
N u 1 m =0
skN + l = c k ,n
e
Nu
, l [ 0, N 1] , k Z, (2)
and
n=0
L 1
where N = N u + N g . R1 ( d ) = rd + m + L .
2
(6)
m =1
Due to miss-synchronized transmitter and receiver
oscillators, the received signal is affected by a frequency offset, The main drawback of this approach is that the metric
the normalized version of which, with respect to sub-carriers exhibits a plateau which has a length equal to that of the guard
spacing, is denoted by . Moreover, the received signal has a interval minus the length of the channel impulse response. Due
time offset, the normalized version of which, with respect to the to noise, this plateau leads to some uncertainty as to the start of
sampling period is denoted by . The channel introduces a the frame and results in a large TO estimation variance. To
alleviate this drawback, Schmidl and Cox proposed an
zero-mean Additive White Gaussian Noise (AWGN) k averaging method, where two points to the left and right in the
(k Z). The received sample expression is then time domain, which are exactly within 90% of the maximum
value, are first found. Then, the timing estimate is taken as the
j 2
k
middle of those two points. In this way, the estimate is
sk + k .
Nu
rk = e (3) guaranteed to fall within the plateau.
In the following, we consider a class of preamble-based The metric in (5) is also used for frequency synchronization
synchronization methods that use a preamble, with specific to detect the FFO by
structure for synchronization aims. This preamble is sent either 1
periodically, to track the synchronization parameters, in the = (( P1 ( d opt )) (7)
case of continuous stream traffic or at the start of a Tu
transmission, to detect the beginning of emission, in the case of
bursty packet traffic. where ( denotes the phase operator.
III. SURVEY ON PREAMBLE-BASED SYNCHRONIZATION To further reduce the uncertainty due to the timing metric
METHODS plateau, two approaches have been proposed by Minn et al. in
[2]. These methods are presented hereafter.
This section introduces three methods applied for
preamble-based OFDM synchronization. The first method, B. Minns Sliding Window Method
proposed by Schmidl and Cox, suggests two estimators, for
This method uses the same preamble as the method
Time Offset (TO) and Fractional Frequency Offset (FFO) [1]. presented in subsection A. Two aspects have been changed:
The second, known as Minns Sliding Window Method, is a
modified version of the previous TO estimator [2]. Finally, the The first modification consists in computing the
third method, named Minns Training Symbol Method, energy R1 ( d ) over the whole symbol, rather than over
presents a different TO estimator using a different preamble the second half. This leads to
structure [2].
1
1N
r
u
A. Schmidl and Cox method R2 (d ) =
2
. (8)
d +m
Schmidl and Cox have presented in [1] a robust frequency 2 m =0
and timing synchronization algorithm. The preamble is
composed of two symbols. Only the first is used for TO and The second modification is averaging the metric over a
FFO estimation. This symbol is designed to be window of length N g + 1 instead of the 90% averaging
Illustrative curves representing the metrics of Schmidl and (whereq ` ). The concatenated preamble samples form the
Coxs method and Minns methods are shown in Fig. 1. The
Sliding Window Method of Minn is referenced as Method A, sequence s = [ s1 , s2 ] = [ s1 , s1 : p ] where : denotes the
and the Training Symbol Method of Minn is referenced as element-wise product operator. To observe the OFDM symbol
Method B. We here consider the case of an OFDM signal structure, a CP of length N gm = N g + N u N um is added to
with N u = 1024 sub-carriers, with a guard interval of
to keep the total OFDM symbol length of N .
N g = 102 samples, and a normalized carrier frequency offset
The choice of the sequences p and s1 is of utmost
= 12.4 sub-carriers spacing, and an ideal channel case (no
noise). importance basic step. These sequences should be meticulously
chosen to have an important correlation property, that will be
1 Proposed method exploited later when calculating the metric. M-sequences are
Schmidl & Cox
Minn Method A
well known for having specific correlation properties in terms
0.9
Minn Method B of minimizing the maximum value of the out-of-phase
0.8 autocorrelation. The smaller is the out-of-phase autocorrelation
0.7 of the chosen sequence, the more accurate is the timing offset
Metric Mi, i=1,2,3,4
0.6
estimation. Thus, they are adequate for synchronization. Hence,
in this work two m-sequences are applied to generate the time
0.5
domain preamble ( s1 and p ). For a judicious choice, we use a
0.4
preferred pair of m-sequences for which the above mentioned
0.3
correlation property is better verified than arbitrary chosen sub-
0.2 sequences.
0.1
B. Timing offset estimation
0
0 100 200 300 400 500 600 700 800 900 1000 To detect the frame start, the idea is to search for two
Time (in samples) strongly correlated symbol halves. This is realized by searching
Figure 1. Timing metrics under noiseless conditions. the maximum of metric
P4 = r *
r
d + m d + m + Lm
*
p m, (15) The timing estimator performance is evaluated in terms of
rate of correct detection and estimator variance. The five cases
m =1
presented in [2] (I-Schmidl and Cox, II-Schmidl and Cox using
and (8), III-Minns Sliding Window Method, IV-Minns Training
Lm 1 Symbol Method, and V-Minns Training Symbol Method using
r
2
R4 (d ) = . (16) (8)) are compared to the proposed method.
d + m + Lm
m =0 Fig. 2 shows the rate of correct detection for all methods. As
The correlation property of the chosen sequences ensures can be noticed, beyond an SNR of -5 dB, the proposed method
that the proposed timing metric has its peak value at the correct provides a correct start frame detection with rate 1. At very low
SNR (less than 0 dB) method III outperforms the others. At
symbol starting point. The frame start corresponds to the
higher SNR, the detection rates of methods IV and V are better
sample index d opt that maximizes the metric in (14). Fig .1 than III, as the metric peak can be more easily distinguished. In
shows that this metric exhibits an almost perfect channel all cases, methods I and II give the worst detection rates due to
autocorrelation profile with a sharp peak at the correct frame the plateau effect. A gain of about 15 dB is achieved by the
proposed method with respect to the best benchmark (method
start, using a system of N um = 1022 samples and a guard V) for a target correct detection rate of 90%. This gain goes to
interval of N gm = 104 samples. 17 dB for a target of 95%.
0.7
Schmidl & Cox (I)
1 Schmidl & Cox modified (II)
= ( ( P4 ( d opt )). (17)
0.6
Minn Method A (III)
0.5 Minn Method B (IV)
Minn Method
Minn B modified
B modified (V) (V)
This equation provides a detection range of 1 in terms of sub- 0.4 Proposed method
0
The performance of the proposed synchronization method -10 -5 0 5 10 15 20 25
and the methods presented in section III are investigated by SNR (dB)
method gives better results when the SNR is less than -5 dB.
Beyond this value, performance for true and estimated frame
MSE()
-3
10
starting point are similar. The FFO estimation is robust to
small timing offsets. For our method, this is obtained thanks to -4
10
its sharp metric around the true frame start allowing a correct
timing acquisition. For the method of Schmidl and Cox, the -5
FFO estimation is insensitive to timing errors whenever the 10
Schmidl and Coxs method, the proposed one gives lower SNR (dB)
MSE.
Figure 4. MSE of FFO for perfect and estimated symbol timing.
5
10 REFERENCES
[1] T.M. Schmidl, D. Cox, Robust frequency and timing synchronisation in
4
10 OFDM, IEEE Trans. on Comm., vol. 45, Dec. 1997, pp. 1613-1621.
[2] H. Minn, M. Zeng, V. K. Bhargava, On timing offset estimation for
3
10 OFDM systems, IEEE Comm. Letters, vol.4, no. 7, July 2000, pp. 242-
Schmidl & Cox (I)
244.
2
10
Schmidl & Cox modified (II) [3] J.-J. van de Beek, M. Sandell, M. Isaksson, and P. Brjesson, Low
Minn Method A (III) complex frame synchronization in OFDM systems, in Proc. ICUPC,
Minn Method B (IV) Nov. 610, 1995, pp. 982986.
var()
1
10 Minn Method B modified (V) [4] D. Landstrm, S. K. Wilson, J. J. van de Beek, P. dling, and P. O.
Proposed method Brjesson, Symbol time offset estimation in coherent OFDM systems,
in Proc. Int. Conf. on Communications, Vancouver, BC, Canada,
0
10
June1999, pp. 500505.
-1
10 [5] L. Atallah, M. Siala, A New scheme for preamble detection and
frequency acquisition in OFDM systems, in Proc. ICECS, Hammamet,
-2 Tunisia, December 2009.
10
-3
10
-10 -5 0 5 10 15 20 25
SNR (dB)
VI. CONCLUSION
A data-aided time and frequency synchronization method
for OFDM systems has been presented. It is based on a new
preamble structure that uses one symbol whose two halves are
related by a differential encoding. A judicious design of the
involved sequences guarantees a robust metric shape which
enhances the frame start detection capacity and the FFO
estimation accuracy.