Synchronization Algorithms for MIMO OFDM Systems

En Zhou, Xing Zhang, Hui Zhao, Wenbo Wang

Center for Wireless Communications,
Beijing University of Posts & Telecommunications, Beijing, China, 100876
Email: [email protected], [email protected]

Abstract-This paper proposes a time and frequency frame structure used for synchronization. The proposed
synchronization solution for MIMO OFDM systems. The synchronization algorithms are described in Section III, and
synchronization is achieved using one preamble which is our simulation results are discussed in Section IV. Finally,
simultaneously transmitted from all transmit antennas in the
Section V gives the conclusion.
same OFDM time instant. The synchronization is accomplished
sequentially by coarse time synchronization, fractional
frequency offset estimation, integral frequency offset estimation II. SYSTEM MODEL
and fine time synchronization. Simulation results demonstrate
A MIMO OFDM system with N t transmit and N r receive
that the proposed synchronization algorithms have a satisfactory
performance even at a low SNR in rich multipath environment. antennas is consider and denoted as a N t × N r system. We
And the synchronization performances in MIMO channel are
super to those in SISO channel for its larger diversity gain .
assume that the system is operating in a rich multipath
environment. The channel between the pth transmit and qth
I. INTRODUCTION receive antenna is modeled as a wide sense stationary,
uncorrelated scattering (WSSUS), Rayleigh fading channel
Orthogonal frequency division multiplexing (OFDM) has with L paths in the complex equivalent low-pass time-variant
gained increased interest in the last years for its advantages in impulse response
digital transmissions over frequency-selective fading channels L −1
[1]-[4]. Multiple-input multiple output (MIMO) wireless (
h ( p , q ) = ∑ α l( p , q )δ τ − τ l( p , q ) ) (1.)
systems with multiple antennas at both transmit and receive l =0
sides can achieve very high spectral efficiency in rich scatter where α l( p ,q ) is the lth path gain from the pth transmit antenna
environment. The combination of MIMO and OFDM is a to the qth receive antenna; τ l( p ,q ) is the propagation delay for
promising technique for the future broadband mobile
communication systems. the lth path from the pth transmit antenna to the qth receive
MIMO OFDM system also requires synchronization in both antenna. For simplicity, the fades between different transmit
the time and frequency [5]-[9]. Time synchronization involves and receive antennas are assumed to be independent and
finding the best possible time instant for the start of received identical distributed (i.i.d.). And we also assumed that all
OFDM frame. Frequency synchronization deals with finding transmit/receive branches of one MIMO transmitter/receiver
an estimate of the difference in the frequencies between the use the same oscillator as [5], which is a valid assumption if
transmitter and receiver local oscillators. Frequency offset the different transmit/receive branches are co-located.
estimation in OFDM is critical since any frequency offset The frame structure of the MIMO OFDM system is shown
causes a loss of subcarrier orthogonality which results in Inter in Fig.1, with one preamble in a frame. The preamble here is
Carrier Interference (ICI) and hence performance degradation. mainly designed for synchronization of MIMO OFDM system.
Synchronization schemes based on training symbols must be It can also be used for channel estimation coincidently as the
efficient in terms of performance and overhead [5][6]. orthogonal structure in [6], hence not increasing the overhead.
In this paper, a new synchronization algorithm for MIMO Let S ( p ) , R (q ) be the transmitted OFDM symbol of antenna p
OFDM system is presented, which only needs one OFDM and received demodulated OFDM symbol of antenna q, H ( p , q )
preamble transmitted in the same OFDM instant. The be the channel frequency response between the pth transmit
preamble can also contribute to the channel estimation and the qth receive antenna.
followed hence reducing the overhead. The algorithm The OFDM symbol transmitted by the pth transmit antenna
performs well in low SNR and in rich multipath environment. is given by
The rest of the paper is organized as follows: Section II 1 N −1
2πnk
gives a brief description of MIMO OFDM system and the
sn( p ) =
N
∑S
k =0
( p)
k exp( j
N
) 0 ≤ n ≤ N −1 (2.)

This work has been funded by Motorola Co. (U.K.) and China National 863
Project, No. 2003AA12331004.

IEEE Communications Society / WCNC 2005 18 0-7803-8966-2/05/$20.00 © 2005 IEEE

 N r M −1m ( N + G ) + C −1
( )
Antenna 1
Φd = ∑ ∑ ∑
*
rd( +q )k rd( +q )k + N (6.)
G N G N G N q =1 m = 0 k = m ( N + G )
N r M −1m ( N + G ) + C −1
Pd = ∑ ∑ ∑
2
rd( +q )k + N (7.)

……
Antenna Nt q =1 m = 0 k = m ( N + G )
Here Γd represents coarse time acquisition metric. G, C is the
G N G N G N
length of CP and correlation window respectively. And M is
G+N
the number of consecutive OFDM symbols for averaging. The
Nd(G+N)
tentative time offset d and the fractional frequency offset ε f
Preamble Data
Fig.1. Frame structure for MIMO OFDM system can be estimated jointly as
dˆ = arg max{Γd } (8.)
where {S k( p ) }kN=−01 is the transmitted data sequence from the pth d

1
transmit antenna, 1 ≤ p ≤ N t . The received sample sequence εˆ f = − ∠ Φ dˆ (9.)
2π
after the removal of the guard interval on receive antenna q is
N t L −1 where ∠ ⋅ denotes the argument of a complex number, d̂ is
rn
(q )
= ∑∑α ( p,q) ( p)
l s
( n −τ l( p ,q ) ) N
+w (q)
n (3.) the estimated tentative time offset and εˆ f is the estimated
p =1 l = 0 fractional frequency offset.
where wn(q ) is complex additive white Gaussian noise samples In multipath faded environment, the correlation between CP
with variance N 0 at receive antenna q and (k ) N denotes the and its correspond data part will be corrupted by multipath.
The correlation window length fixed as the CP length (G
residual of integer k module N, and the result is in the interval samples) will not be optimal in multipath environment. In
of [0, N − 1] . The received signal is demodulated using FFT as order to reach the optimum synchronization performance, the
Rk( q ) = FFT {r ( q ) }(k ) length of the correlation window should be adjusted
Nt
adaptively according to the channel maximum time delay
(4.)
= ∑ H k( p , q ) S k( p ) + Wk( q ) spread [4].
In OFDM system, the channel frequency response must be
p =1
estimated and compensated. We can perform an IFFT to
Note that, for a fair comparison, the total transmit power
transform the estimated frequency response into time domain
should be keep the same as in the single-input single-output
to estimate the channel impulse response which contains the
(SISO) case. Therefore, the power per transmit antenna should
information of the path time delays and the complex value
be scaled down by a factor N t .
channel gains for different paths of the channel. The
multipaths can be estimated by finding the magnitude of
III. SYNCHRONIZATION ALGORITHMS channel impulse response larger than a proper threshold η .
Synchronization for MIMO OFDM system contains time
and frequency synchronization. Time and frequency
{
ψ = [m]N | hˆ[m] |> η , m = 0,", N − 1 } (10.)
synchronization can be performed sequentially in the Here ĥ is the total impulse response of the MIMO channel
following steps. estimated by preamble. Defining the principle interval
A. Coarse time synchronization and fractional frequency as (− N / 2, N / 2] , [q]N denotes that extend q with period of N
offset estimation and choose the value fallen in the principle interval as the
result. Sort the elements in ψ ascendingly and denote the
Coarse time synchronization locates the start of OFDM
symbol over an approximate range of sample values. Due to sorted elements by λk ( λ0 > " > λLˆ −1 ), where L̂ is the number
the presence of cyclic prefix (CP) in OFDM system, coarse of multipaths estimated. The channel maximum time delay
time synchronization can be performed by correlating the spread can be expressed as:
received samples that are at a distance of the number of
subcarriers, N, over a length of C window.
τˆmax = λLˆ −1 − λ0 (11.)
2 Generally, the channel time delay spread is associated with
Φd
Γd = (5.) the scatter environment and varies very slowly. The length of
( Pd ) 2 the correlation window C can be adjusted according to the
where channel maximum time delay spread estimated in the fore

IEEE Communications Society / WCNC 2005 19 0-7803-8966-2/05/$20.00 © 2005 IEEE

frame, hence S 2 k = ck S 2 k +1 k = 0,1," , N / 2 − 1 (15.)
C = G − τˆmax (12.)
where ck is the kth chip of a pseudonoise sequence with length
Thus, the coarse synchronization algorithm can track the
variety of channel and make the length of correlation window of N/2.
After the fractional frequency offset is removed from the
adjusted as that of ISI-free CP part adaptively. The
synchronization performance can be improved greatly received sample stream by multiplying it with
compared with the conventional case with correlation window exp{− j 2πεˆ f n / N } per antenna, the received preamble signal is
length fixed by G. extracted using the coarse FFT start position estimated.
The ability of multipath estimator and the synchronization Denoting by rn( q ) the nth sample value of received preamble at
performance will be affected by threshold η . The threshold receive antenna q, the total received preamble signal is
should be designed with different noise level. Generally, the Nr
length of CP is designed larger than the maximum time delay rn = ∑ rn( q ) 0≤n< N (16.)
spread and about quarter of the number of subcarriers N. The q =1
residual timing offset after coarse timing synchronization is Convert the total received time domain preamble signal to
assumed within the interval of [− G / 2, G / 2] , thus within the frequency domain by FFT
interval of (3G / 2, N − G / 2 ) , the estimated channel impulse R k = FFT {r}(k ) (17.)
response only contains noise. The threshold can be designed The integral frequency offset can be calculated by finding ε i to
according to the maximum magnitude of the noise observed in
maximize
this interval as follows
{} N / 2 −1 2

hˆn = FFT Hˆ (n) ∑ R(ε i + 2 k ) N / 2 R(*ε i + 2k +1) N / 2 ck*

hmax { }
= max hˆ n n ∈ (3G / 2, N − G / 2) F (ε i ) = k =0
N / 2 −1 N / 2 −1
(18.)
∑ ∑
2 2
R 2 k ⋅ R 2 k +1
η′ = β ⋅ hmax k =0 k =0

η = min{η ′, hmax } (13.) εˆi = maxF (εi ) εi ∈ (− N / 4, N / 4] (19.)

εi
Ĥ is the estimated total frequency response of MIMO channel The range of integral frequency offset estimation algorithm
and will be described in detail in subsection B. And β is a can reach half of the number of subcarriers. It is much larger
constant larger than 1. The value of β selected will affect the than the algorithm proposed in [5] and enough for practical
applications.
synchronization performance. Large value of β will improve
After the estimation of integral frequency offset, the total
the ability of combating against noise while decrease the frequency response of MIMO channel can be calculated by
ability of multipath detection. So the value of β should be
Hˆ k = R ( k + εˆi ) N / Sk (20.)
carefully designed.
The start of FFT window κ̂ 0 estimated by the coarse time The total channel frequency response will be used for coarse
and fine time synchronization as described in subsection A and
synchronization algorithm is given as C.
κˆ0 = dˆ + C (14.)
C. Fine time synchronization
B. Integral frequency offset estimation Fine time acquisition locates the start of the useful portion
of the OFDM symbol accurately. Once the fractional and
The algorithm of fractional frequency offset estimation
integral frequency offset is removed, fine time
described above can only estimated the frequency offset range synchronization can be performed by finding the position of
of ± 1/ 2 subcarrier space for the period 2π of angle operation. the first pulse response of the total MIMO channel. The
The integer frequency offset will not cause inter-carrier multipath information can be obtained by finding the
interference (ICI) but only subcarrier cyclic shift. If not being magnitude of channel impulse response larger than a proper
estimated and removed, the integral frequency offset remained threshold η as described in subsection A. According to the
will corrupt the whole frame. The integral frequency offset can position λ0 of the first path, the residual time offset g can be
be estimated in frequency domain by the preamble. For facility,
estimated as
superscript ˙ is used to denote the preamble signal or
preamble signal related. The preamble is designed that the data gˆ = λ0 (21.)
modulated at even subcarrier index has a relationship with that In the fine symbol synchronization algorithm, the start of the
at odd subcarrier index as followed FFT window κ̂ is adjusted as

IEEE Communications Society / WCNC 2005 20 0-7803-8966-2/05/$20.00 © 2005 IEEE

 TABLE I 4
10
SIMULATION PARAMETERS NtXNr=1X1 Coarse
Bandwidth 80MHz Coarse+Fine

MSE of the time offset estimation

N=512
Number of subcarriers, N 512 3
10 G=128
fD=20Hz
Modulation QPSK
SNR=10dB
6.4usec+1.6usec (CP)
OFDM symbol duration
(512+128 samples) 2
10
FFT sampling speed 12.5nsec
Frame length 32 OFDM symbols
1
10
Averaging OFDM symbols for
31 OFDM symbols
coarse symbol synchronization
Maximum Doppler frequency 20 Hz 0
10
Channel model Exponential model
1.00 1.25 1.50 1.75 2.00 2.25 2.50
κˆ = κˆ0 + ĝ (22.) Coefficient β

The received signal is first performed by coarse time and

fractional frequency offset estimator. In the phase of coarse Fig.2. MSE vs. coefficient β at SNR=10dB
time synchronization, the length of correlation window is
initialized as CP length G. After the removal of fractional
frequency offset, the preamble is extracted using the FFT start

MSE of the coarse time synchronization

100
position estimated in the coarse time synchronization. Then
the integral frequency offset is performed in frequency domain
by preamble. After the integer frequency offset has been
removed from the preamble signal, the overall impulse
10
response of MIMO channel can be obtained. Then the channel
maximum time delay spread is estimated and fine time
Nt X Nr=1X1 N=512
synchronization is accomplished. G=128
Nt X Nr=2X2
fD=20Hz
Nt X Nr=3X3
1 β=1.5
IV. SIMULATION RESULTS
-5 0 5 10 15 20
The performance of the proposed synchronization SNR(dB)
algorithm for MIMO OFDM system is evaluated through
simulations. Table I gives the simulation parameters used. The Fig.3. MSE of the coarse time synchronization
multipath channel per transmit/receive antenna pair is
modeled as 12 Rayleigh faded paths with 1dB exponential 14

decay of average power level, having an equal interval of 12

MSE after fine time synchronization

N=512 Nt X Nr=1X1
0.1usec (8 samples) [10]. And low mobility is considered with G=128 Nt X Nr=2X2
10
the maximum Doppler frequency fD of 20Hz in our simulation. fD=20Hz
Nt X Nr=3X3
β=1.5
Constant β associated with the threshold will affect the 8

ability of combating against noise and multipath detection as 6

described above. In order to find the optimum value of β, we 4

select different values at a fixed SNR of 10dB and the
2
corresponding mean square error (MSE) of the time offset
estimation is shown in Fig.2. For simplicity, the SISO channel 0

is adopted. It is obvious that the MSE performance of the -2

timing synchronization reach its optimal when β=1.5. Hence, 0 5 10 15 20
SNR(dB)
β=1.5 is adapted for the following simulations.
Fig.3 to Fig.6 are the simulation results obtained. It is clear Fig.4. MSE after fine time synchronization
that the synchronization performance is not only improved
with the increasing of SNR but also with the increasing of the antennas, making the total MIMO channel more flat. This
number of transmit/receive antennas. The synchronization makes the estimation of the maximum time delay spread of the
performance in MIMO channel environment is much super to total MIMO channel more robust and accurate improving the
that in SISO channel environment. The number of total coarse time synchronization performance.
multipaths is increased with the increasing of transmit/receiver Fig.3 shows the MSE performance of the coarse time

IEEE Communications Society / WCNC 2005 21 0-7803-8966-2/05/$20.00 © 2005 IEEE

 TABLE II
1 ERROR PROBABILITY OF INTEGERAL FREQUENCY OFFSET
ESTIMATION
Error probability of time synchronization

SNR(dB)
Coarse -5 0 5
Nt × Nr
0.1
1X1 0.16475 6.25E-4 0
2X2 0.00517 0 0
Coarse+Fine 4X4 0 0 0
0.01

transmit/receive antennas or at very low SNR, such as -5dB,

N=512
and the error probability is decreased sharply with the
1E-3 Nt X Nr=1X1
G=128 increasing of the number of transmit/receive antennas.
Nt X Nr=2X2 fD=20Hz
Nt X Nr=4X4 β=1.5
1E-4
V. CONCLUSION
0 5 10 15 20
SNR(dB) This paper addressed the synchronization schemes for
MIMO OFDM system and proposed a novel solution for
Fig.5. Error probability of time synchronization MIMO OFDM synchronization. The synchronization
algorithm only needs one preamble transmitted in all transmit
1E-4
antennas in the same OFDM time instant. The preamble can
also be cooperated for channel estimation reducing the overall
MSE of the fractional frequency offset

Nt X Nr=1X1
Nt X Nr=2X2 overhead. The synchronization is accomplished sequentially
1E-5 Nt X Nr=4X4 by coarse time synchronization, fractional frequency offset
estimation, integral frequency offset estimation and fine time
synchronization. The synchronization algorithm proposed
1E-6
shows satisfactory performance even at a low SNR in rich
N=512 multipath environment. And the synchronization
1E-7
G=128 performances in MIMO channel are much super to those in
fD=20Hz
SISO channel for its larger diversity gain.
β=1.5

1E-8
-5 0 5 10 15 20
SNR(dB)
REFERENCES
[1] J.J. van de Beek, M. Sandell, and P.O. Borjesson, “ML estimation of
Fig.6. MSE of the fractional frequency offset estimation time and frequency offset in OFDM systems,” IEEE Transactions on
Signal Processing, vol.45, pp.1800–1805, July 1997.
[2] T.M Schmidl, D.C. Cox, “Robust frequency and timing synchronization
synchronization. The performance of the coarse time for OFDM,” IEEE Transactions on Communications,
synchronization is quite good for its ability of dodging ISI vol.45, pp.1613-1621, Dec. 1997.
introduced by multipath. The MSE after fine time [3] Yun Hee Kim, Young Kwon Hahm, Hye Jung Jung, etc. An efficient
frequency offset estimator for timing and frequency synchronization in
synchronization is further decreased as shown in Fig.4. Fig.5 OFDM systems. IEEE Pacific Rim Conference on Communications,
compared the error probability of time synchronization in Computers and Signal Processing, Aug. 1999, pp580–583.
different MIMO channel environments. It is very excited that [4] En Zhou, Maomao Chen, Wenbo Wang, “A Novel Timing Estimation
Method for OFDM Systems over Multipath Fading Channels,” ICCC
the error probability after fine time synchronization is much 2004, pp700-704. Sept. 2004.
lower than that of only coarse time synchronization. The error [5] A.N. Mody, G.L. Stuber, “Synchronization for MIMO OFDM systems,”
probability of 4×4 MIMO OFDM system reaches 10E-3 at IEEE GLOBECOM '01, vol.1, pp.509-513. Nov. 2001.
[6] A.N. Mody, G.L. Stuber, “Receiver implementation for a MIMO OFDM
SNR about 20dB. The good time synchronization performance system,” IEEE GLOBECOM '02, vol.1, pp.716-720, Nov. 2002.
is thanks for the high diversity effect in rich scatter [7] T.C.W. Schenk, A. van Zelst, “Frequency synchronization for MIMO
environment and the good multipath estimation performance. OFDM wireless LAN systems,” IEEE 58th VTC 2003-Fall, vol.2,
pp.781-785, Oct. 2003.
The richer scatter in the environment, the better [8] G.L. Stuber, J.R. Barry, S.W. McLaughlin, Ye Li, M.A. Ingram and
synchronization performance can be obtained. T.G. Pratt, “Broadband MIMO-OFDM wireless communications,”
The accuracy of the fractional frequency offset estimation is Proceedings of the IEEE, vol.92, no.2, pp.271-294, Feb. 2004.
[9] van Zelst, T.C.W. Schenk, “Implementation of a MIMO OFDM-based
also improved with SNR and the number of transmit/receive wireless LAN system,” IEEE Transactions on Signal Processing, vol.52,
antennas as shown in Fig.6. The error probabilities of integral no.2, pp.483-494, Feb. 2004.
frequency offset estimation in different channel environments [10] H. Atarashi, S. Abeta, and M. Sawahashi, “Pilot channel assisted MMSE
combining in forward link for broadband OFCDM packet wireless
are listed in Table II. The performances of integral frequency
access,” IEICE Trans. Fundamentals of Electronics, Communications
offset estimation are much satisfied. The estimation errors of and Computer Sciences, pp.1635-1647, July 2002.
integral frequency offset exist in the case of small number of

IEEE Communications Society / WCNC 2005 22 0-7803-8966-2/05/$20.00 © 2005 IEEE