You are on page 1of 6

The 21st International Symposium on Wireless Personal Multimedia Communications (WPMC- 2018)

Analysis of Timing Synchronization Techniques in


OFDM for SDR waveform: Performance
Comparison
Muhammad Ali Mumtaz, Muhammad Zeeshan, Qasim Umar Khan Abdullah Hasan
Department of Electrical Engineering Department of Electrical Engineering
College of Electrical & Mechanical Engineering School of Engineering & Applied Sciences
National University of Science & Technology Isra University
Islamabad, Pakistan Islamabad, Pakistan
alimumtaz91@gmail.com, ranazeeshan@ceme.nust.edu.pk abdullahhsn8@gmail.com

Abstract—Software Defined Radio (SDR) is the future of software methodologies like virtual radio machines are used
wireless communication technology as it provides a flexible to modulate and demodulate the signal. It is capable of trans-
platform capable of providing various functionalities through mitting and receiving a signal but not simultaneously. SDR
software modifications. Orthogonal Frequency Division Multi-
plexing (OFDM) is an efficient communication technology for the is a good platform to implement OFDM concepts and verify
design of wide-band waveform of SDR. Timing synchronization is different synchronization algorithms in various propagation
an integral part for the OFDM waveform of SDR. For burst mode environments.
of communication, it includes the correct detection of the start Orthogonal frequency division multiplexing (OFDM) sys-
of burst. Several algorithms are present in literature for burst tem are very prone to Synchronization errors which include
detection in OFDM. Since SDR faces resources scarcity prob-
lem, the selection of appropriate algorithm which can provide timing and frequency synchronization errors. Many techniques
desired detection performance with acceptable computational have been proposed using preambles in the start of the data
complexity is a major challenge. In this paper four algorithms stream to overcome these errors. Timing offset errors can
including Minn’s algorithm, S&C’s algorithm, Park’s algorithm, disturb the orthogonality of the sub carriers resulting in inter
and Leila’s algorithm have been investigated based on their detec- carrier interference (ICI) and inter symbol interference (ISI)
tion performance and computational complexity. The presented
analysis is helpful for appropriate selection of burst detection thus corrupting the whole data stream. The main theme of
algorithm for SDR wide band waveform based on OFDM. The OFDM is the orthogonality of sub carriers which allows to
results show that Leila’s and Park’s algorithms give sharp timing achieve high data rates thus it should be maintained. ISI in
metric without any signi?cant distortion compared to Minn’s and OFDM can be removed by the insertion of cyclic prefix (CP)
S&C’s algorithm. The Probability of correct detection (PD) plots which has the length of 25% of the total number of sub
show that S&C’s algorithm gives the worst detection performance
as it is unable to reach the probability of detection equal to carriers. The OFDM synchronization has two types one is data-
1 because of a large plateau. Other three algorithms achieve aided and other is non-data-aided. Training symbol sequence is
probability of detection equal to 1 but Minns algorithm is inferior used for estimation in data-aided. It has more accurate results
to the other two algorithm in terms of detection performance. and less computations, but it has shortcoming like bandwidth
On the other hand, S&C’s algorithm has the least computation losses and reduction in data rate. On other hand the cyclic
complexity as compared to the other three algorithm.
Index Terms—OFDM, SDR, timing synchronization, m-
prefix correlation is used in Non data-aided. And no data rate
sequence, Golay complementary sequence transmission decreases and neither use large bandwidth, but
it has very small estimation range. Only Data-aided methods
are discussed in this paper and analysis of the synchronization
I. I NTRODUCTION
algorithms are discussed.
Software-defined radio has become a universal platform for As preamble base algorithm, S&C’s [1] suggested a tech-
implementing any type of waveform without any need for nique for synchronization for time and frequency. In the
the change of hardware. A single SDR hardware can process beginning of the frame two symbols for training are engaged.
all the signal processing functions. It is the collection of First symbol has equal halves in time domain, timing metric
hardware and software with reconfigurable architecture. SDR at receiver side can be made by correlating between these two
consist of smart RF antenna which minimize noise,multipath halves. Though, the metric give a plateau shaped graph which
and interference.At receiver end the signal is digitized into has some ambiguity in detecting the beginning of the frame. To
IF band using ADC while at the transmitter IF frequency remove the ambiguity generated by plateau of timing metric,
is converted to analog signal. The processing of the signal Minn’s [4] improved S&C’s algorithm and recommended
is performed FPGA,DSP controllers or ASIC’s. Different two new algorithms. In primary algorithm timing metric half

978-1-5386-5757-7/18/$31.00 ©2018 IEEE 606


The 21st International Symposium on Wireless Personal Multimedia Communications (WPMC- 2018)

symbol energy can be calculated by all samples over one


symbol period (eliminating CP). The secondary algorithm X
N −1

working with the training symbol which have four parts of x(n) = Ce(j(2pi/N )kn) (2)
k=0
equal length however last two have different sign. We use
second algorithm of Minn’s for this paper. To get more good h(m) is the Impulse response of the signal and L represent
result of timing synchronization. In [9] other Minn’s method the length of the sample. x(n) is the OFDM signal in time
mean square error is achieved by auto-correlation between domain. Number of sub-carriers are represented by N . To
the training symbol. Side lobes are still present their in the observe the synchronization parameters we sent the training
results thus reducing the estimation efficiency of the both symbol periodically in the transmitted data by appending it at
methods in [4] [9] in the timing-metric. Park’s [2] suggested the start of the data, in the case of burst packet transmission.
a better timing synchronization technique for OFDM systems.
It shows sharper timing metric than above two algorithms, but III. S YNCHRONIZATION T ECHNIQUES
the timing performance may get disturbed by two side-lobes. Timing synchronization is an essential part of OFDM
In [6] author presented the new method for more accuracy receiver as it is necessary to find exact starting point of
in symbol detection and avoid the noise which are in shape each OFDM symbol to avoid loss of data. In this section
of side-lobes in Park’s method. He suggested that almost the four methods applied for OFDM timing synchronization are
same method as suggested by Park’s but slightly different discussed.
training symbol. In [10], [11] a new timing metric is proposed
which is based on cross-correlation of the training symbol and A. Schmidl and Cox Method
signal received. Performance of this method is very poor as the In [1] this technique timing synchronization is attained by
number of terms used for correlation to estimate timing metric using a training symbol which is divided in two equal half in
are limited to N . Leila’s [3] suggested another technique for the time domain. Purpose of this technique is that the symbol
more accurate symbol detection, in his method he used m- timing errors will have tiny effect on the signal itself. For
sequences along with precoding matrix for training symbol. making these halves identical we send a PN sequence on the
The m-sequences have very good correlation properties which even frequencies and zeros on odd frequencies. According to
are used to estimate the timing synchronization metric. This IFFT property, taking the IFFT of this sequence make them
training symbol can be obtained by linking together the two identical. S = [A, A] where A denote the number of complex
sub-sequences s1 and s2 of equal lengths. The sub sequence sample of length N/2 in one OFDM symbol.
s1 and precoding matrix, p are two different m-sequences,
whereas s2 can be calculated by multiplying s1 with p. X
L−1
P (d) = r∗ (d + m).r(d + m + L) (3)
m=0
II. OFDM SIGNAL MODEL
X
L−1
Orthogonal frequency division multiplexing is a scheme E(d) = |r(d + m + L)|2 (4)
which is designed to achieve high data rates in various propa- m=0
gation environments. Single carrier is converted into multiple
sub carriers and these sub carriers are transmitted into parallel |P (d)|2
M (d) = (5)
streams which can be easily equalized at the receiver [8]. E(d)
Cyclic prefix in OFDM is used to enable one tap equalization P (d) is the cross-correlation between the two halves in
at the receiver as it tends to make the linear convolution with frame and R(d) shows the auto-correlation of the second half.
channel response as circular convolution. All sub carrier needs L shows the number of complex samples and received signal
to be orthogonal with each other so that data does not gets is represent by r.
corrupted. In OFDM firstly the data is modulated using BPSK
or QAM and then IFFT operation is carried out to make sub B. Minn Method
carriers orthogonal to each other. After IFFT, CP is inserted in
In [4] Minn’s algorithm introduced for improving S&C’s
the signal and it is transmitted over a channel. At the receiver,
method. The timing metric of S&C’s method gives a plateau
CP is removed and FFT of the received signal is taken which is
shaped graph which shows some ambiguity associated with
followed by demodulation of the signal to recover the original
the initial point of the symbol. Peak of Minn’s method at the
data bits. Block diagram of the OFDM system is shown in
accurate starting point for the symbol. Peak of interest is the
figure 1 and shaded block is the scope of this paper.
middle peak. But, there are extra peaks with approximately
The standard complex-valued base-band equivalent signal the same size as the middle peak. However, this might not
model used in OFDM. The nth received sample has the affect the performance. If we take the energy of the symbol
standard form: at the peak point. In our simulations, we expected that the
central peak is not disturbed by the both side peaks. S =
X
L−1
[T, −T, T, −T ] Where T are Golay complementary sequence
y(n) = h(m).x(n − m) (1)
m=0
length N/4.

607
The 21st International Symposium on Wireless Personal Multimedia Communications (WPMC- 2018)

Add cyclic
Bit Stream Modulator S/P IFFT P/S
prefix

Insert pilot
Symbol
Channel

Remove Burst
Bit stream Demodulator P/S FFT S/P
cyclic prefix Detection

Remove
Pilot Symbol

Fig. 1. OFDM Block Diagram

D. Leila’s Method
X
L−2 X
L−1 In this method [3] training symbol is designed using differ-
P (d) = b(k) r∗ (d + m + kM ).r(d + (k + 1)M + m) ential coding techniques. To construct a time domain training
k=0 m=0 symbol two sub sequences s1 and s2 of same length are
(6)
concatenated. By multiplying sub sequence s1 with precoding
X
M X
−1 L−1 sequence p, second sub sequence s2 is obtained. Length of
E(d) = |r(d + i + kM )|2 (7) these sequences s1 , s2 , p are L2q−1
m .
i=0 k=0 Choice of sequence s1 and precoding sequence p is the
most important step of this algorithm. These sequences must
|P (d)|2
M (d) = (8) have a correlation property so that it can be utilized while
E(d) calculating the timing metric. M-sequences have good cor-
C. Park’s Method relation properties as the maximum value of out of phase
Minn’s algorithm decreases the timing metric created in auto correlation can be minimized to get more accurate timing
S&C’s method, but the MSE is still present mainly in ISI metric. in this technique pair of m-sequences is generated to
channels. Minn’s and Park’s have same timing metric values utilize correlation property effectively.
at correct timing points but Park [2] algorithm proposed
to enhance the difference between the peak timing metric LX
m −1

values with respect to others. This algorithms training symbol P (d) = r∗ (d + m).r(d + m + Lm ).Pm

(12)
consist of four parts. First two are symmetric and last two m=0
are conjugate of first two respectively. The result shows the LX
m −1

sharper timing metric. And it also has significant smaller E(d) = |r(d + m + L)|2 (13)
MSE.S = [C, C ∗ , C, C ∗ ] Where C denote the samples length m=0
of N/4 and C ∗ is the symmetric conjugate of C.However,
|P (d)|2
in timing metric of this method has large side lobes at the M (d) = (14)
positions with N/4 samples. The accuracy of timing offset E(d)2
estimation will be effected. The correlation property of the chosen sequences ensures
that the proposed timing metric has its peak value at the correct
X
N/2−1
symbol starting point.
P (d) = r(m + k).r(m + N − k) (9)
k=0 IV. S IMULATION R ESULTS AND C OMPLEXITY
C OMPARISON
X
N/2−1
E(d) = |r(m + k)|2 (10) A. Simulation Results
k=0 The table I shows the simulations parameters used for
|P (d)|2 analysis in this paper. Figure 2 shows the timing metric under
M (d) = (11) AWGN channel for these methods. The correct timing point is
E(d)2

608
The 21st International Symposium on Wireless Personal Multimedia Communications (WPMC- 2018)

TABLE I
1
S IMULATION PARAMETERS
0.9 PFA =2.17x10-4

Parameters Values 0.8


PFA =5.13x10-6

Probablity of correct Detection


FFT Size 64 PFA =5.88x10-8
0.7
No of Sub-carriers 64
Signal Constellation QP SK 0.6
Channel Model AW GN 0.5
Cyclic Prefix Length 16
0.4

0.3

1 0.2

0.9 0.1
S&C
0.8 Hminn 0
-35 -30 -25 -20 -15 -10 -5 0
Lella
0.7 Eb/N0 (dB)
Park
Timing Metric

0.6
Fig. 4. Rate of correct detection of symbol for Minn.
0.5

0.4

0.3 1

0.2 0.9 PFA =2.17x10-4

PFA =5.13x10-6
0.1 0.8

Probablity of correct Detection


PFA =5.88x10-8
0 0.7
0 50 81 150 200 250 300
Time (sample) 0.6

0.5
Fig. 2. Comparison of Timing Metric. 0.4

0.3

0.2
indexed 81 in the figure. If we compare these timing metrics
0.1
we clearly see this S&C’s create plateau for entire cyclic prefix
0
length. Minn’s shows peak with some disturbance at both sides -35 -30 -25 -20 -15 -10 -5 0
Eb/N0
of the main peak, Park’s gives us a sharp peak and Leila’s gives
us a impulse shape peak which is more accurate in symbol Fig. 5. Rate of correct detection of symbol for Park.
detection.
Figure 3 show the probability of detection of S&C’s algo-
rithm at different values of PF A . It can be observed that as of Park’s algorithm. It shows that the correct start of symbol
we increase the threshold level the value of PF A decreases. is detected at SNR −16dB and figure 6 shows the correct
In S&C’s by varying threshold levels the correct starting point detection rate of Leila algorithm which approaches to 1 also
of symbol is never detected. at SNR −16dB.
Figure 4 show the probability of detection of Minn’s algo- Figure 7 shows the probability of detection for all these
rithm at different values of PF A by varying threshold levels. methods with SNR range −30dB to 0dB at threshold level 0.5
Correct point of the symbol is detected at SNR −13dB on all and PF A is 5.13×10−6 . It can be noticed that at SNR −16dB
values of PF A . Figure 5 shows the probability of detection the method of Park’s and Leila’s outperforms Minn’s and

1 1

0.9 PFA =2.17x10-4 0.9 PFA =2.17x19-4

PFA =5.13x10-6 PFA =5.1x10-6


0.8 0.8
Probablity of correct Detection

Probablity of correct Detection

PFA =5.88x10-8 PFA =5.98x10-8


0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
-35 -30 -25 -20 -15 -10 -5 0 -35 -30 -25 -20 -15 -10 -5 0
Eb/N0 Eb/N0 (dB)

Fig. 3. Rate of correct detection of symbol for S&C. Fig. 6. Rate of correct detection of symbol for Leila.

609
The 21st International Symposium on Wireless Personal Multimedia Communications (WPMC- 2018)

Probability of detection vs SNR for white noise signal TABLE III


1
C OMPUTATIONAL C OMPLEXITY C OMPARISON N UMERICAL VALUES
Lella
0.9
Hmin
0.8 Park Synchronization Techniques Complex Multiplier Complex Adder
Probablity of correct Detection

S&C Schmidl&Cox 32 31
0.7
Haling Minn 48 47
0.6 Park 48 47
0.5
Lela Nasraou 64 63
0.4

0.3
B. Computational Complexity Comparison
0.2

0.1
Computational Complexity Comparison is discussed under
0
this section. The trends of Computational Complexity Com-
-35 -30 -25 -20 -15 -10 -5 0
Eb/N0
parison are shown in table II the table is composed of complex
adder and complex multiplier for training symbol size=N .
Fig. 7. Rate of correct detection of symbol between three algorithms. These formulas are used to calculate the numerical values
for the algorithms shown in table III with the restriction of
N = 64. These numerical values clearly shows that which
Probability of detection vs SNR for white noise signal algorithm has high computational complexity in shape of
1
Lella complex adder and multiplier. Although the Leila’s algorithm
0.9
Hmin
Park
has high number of computations as compared to S&C’s,
0.8
Probablity of correct Detection

S&C Minn’s and Park’s algorithms but it has better result for symbol
0.7
detection at correct timing and its timing metric is sharp as
0.6
compared to S&C’s, Minn and Park. Performance of the timing
0.5
estimator can be evaluated in term of probability of correct
0.4
detection.
0.3

0.2 V. CONCLUSION
0.1 In this paper different timing synchronization techniques
0
-35 -30 -25 -20 -15 -10 -5 0
have been implemented for OFDM based SDR. The main aim
Eb/N0 was to find suitable tradeoff between detection performance
and computational complexity for the ease of implementation
Fig. 8. Rate of correct detection of symbol between three algorithms.
on SDR platform. The simulations shows that Leila and Park’s
methods gives best results for the correct detection of the
symbol but at cost of high computational complexity. Minn’s
S&C’s methods with 100% detection rate. As SNR increases method also shows good results but with side peaks and S&C’s
the detection rate of Minn’s method also improve to 100% at method results degrade because of a plateau for entire length
SNR −13dB while S&C’s gives the worst performance as as of CP. Correct detection rate of all four method discussed in
the rate of detection very low. In Leila’s and Park’s algorithms this paper. Minns algorithm is inferior as compared to Leila’s
there is detection performance improvement of about 3dB and Park’s which achieved detection probability of 1. As
compared to Minn’s algorithm. compared to other methods S&C’s does not achieve probability
of detection equal to 1. Leila’s timing synchronization method
Figure 8 also shows the probability of detection for all makes it possible to estimate symbol timing offset with a
these methods with SNR range −30dB to 0dB at threshold much smaller MSE and has impulse type peak and correct
level 0.6 and PF A is 5.88 × 10−8 .It can be noticed that now timing point for symbol detection but this algorithm has high
at SNR−15dB the method of Park and Leila outperforms computational complexity as compared to other algorithms.
Minn’s and S&C methods with 100% detection rate.In this
case detection performance improvement about 2dB compared R EFERENCES
to Minn’s method. [1] T. M. Schmidl and D. C. Cox, ”Robust Frequency and Timing Syn-
chronization for OFDM”, IEEE Trans. Commun., vol. 45, no. 6, pp.
16131621, 1997.
[2] B. Park, H. Cheon, C. Kang, and D. Hong, ”A novel timing estimation
method for OFDM systems”, IEEE Commun. Lett., vol. 7, no. 5, pp.
TABLE II 239241, 2003.
C OMPUTATIONAL C OMPLEXITY C OMPARISON [3] L. Nasraoui, L. N. Atallah, and M. Siala, ”A very efficient time and
frequency synchronization method for OFDM systems operating in
Synchronization Techniques Complex Multiplier Complex Adder
AWGN channels”, Commun. Netw. (ComNet), 2010 Second Int. Conf.,
Schmidl&Cox N/2 N/2 − 1
no. 1, pp. 15, 2010.
Haling Minn 3N/4 3N/4 − 1 [4] H. Minn, V. K. Bhargava and K. Ben Letaief, ”A robust timing and
Park 3N/4 3N/4 − 1 frequency synchronization for OFDM systems”, IEEE Trans. Wireless
Lela Nasraou N N −1 Commun., Vol. 2, No. 4, pp. 822-839, July 2003

610
The 21st International Symposium on Wireless Personal Multimedia Communications (WPMC- 2018)

[5] A. Nesbitt, ”A Comparison of Timing Methods in Orthogonal Frequency


Division Multiplexing (Ofdm) Systems”, no. March, 2004.
[6] W. Jibin and Z. Jinkang, ”A Novel Time and Frequency Synchronization
Approach for OFDM Systems”, Information Technology and Applica-
tions (ICITA)., Third Int. Conf., vol. 2, pp. 311315, 2005.
[7] M. Young, The Technical Writer’s Handbook. Mill Valley, CA: Univer-
sity Science, 1989.
[8] Andreas F. Molisch, Wireless Communications. Second Edition 2015, p.
417.
[9] K.Shiand E.Serpedin, ”Coarse frame and carrier synchronization of
OFDM systems: a new metric and comparison”, IEEE Transactions on
Wireless Communications., vol.3,no.4,pp.1271 1284,2004.
[10] Y. Kang, S. Kim, D. Ahn, and H. Lee, ”Timing estimation for OFDM
systems by using a correlation sequence of preamble”, IEEE Transac-
tions on Consumer Electronics., vol. 54, no. 4, pp. 16001608,2008.
[11] F. Yang and X. Zhang, ”Robust time-domain fine symbol
synchronization for OFDM-based packet transmission using
CAZAC preamble”, Proceedings of the IEEE Military
Communications Conference(MILCOM13).,pp.436440,IEEE,San
Diego,Calif,USA,November2013.

611

You might also like