International Journal of Scientific & Engineering Research, Volume 6, Issue 4, April-2015 519

ISSN 2229-5518




1SRM University, Department of Electronics & Communication Engineering Kattankulathur, Chennai, India

2SRM University, Department of Electronics & Communication Engineering Kattankulathur, Chennai, India

ABSTRACT: A particle filter is proposed to perform joint estimiation of the carrier frequency offset (CFO) and the channel in multiple-input multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) wireless communication systems. It marginalizes out the channel parameters from the sampling space in sequential importance sampling (SIS), and propagates them with the Kalman filter. Then the importance weights of the CFO particles are evaluated according to the imaginary part of the error between measurement and estimation. The varieties of particles are maintained by sequential importance resampling (SIR). Simulation results demonstrate this algorithm can estimate the CFO and the channel parameters with high accuracy. At the same time, some robustness is kept when the channel model has small variations

KEYWORDS: Carrier Frequency Offset,Multiple-Input Multiple-Output (MIMO), Orthogonal Frequency Division

Multiplexing (OFDM), sequential Monte Carlo, Particle Filter.

—————————— ——————————


The combination of orthogonal frequency division multiplexing (OFDM) and multiple antennas at both the transmitter and the receiver, referred to as multiple input multiple-output (MIMO). In the MIMO-OFDM transmission system the carrier frequency offset (CFO) induced by the local oscillator and the Doppler shift results in the interference of sub-carriers and further incurs an error floor effect. It is required to estimate and compensate for the CFO at the receiver side. Moreover, the unknown channel parameters mixed in the transmission will deteriorate the acquisition condition in practice .In contrast to frequency synchronization schemes of single-input single-output (SISO)-OFDM systems this problem is extended to multi- dimensional parameter acquisition in MIMO systems. Hence, it incurs much difficulty for the coherent detection. Various methods have been investigated to estimate multiple time-invariant frequency offset and
channel parameters, such as maximum-likelihood
estimation correlation-based method, expectation maximization (EM) and iterative methods. They all use the Cramer Rao bound as the precision reference. In addition, an extended Kalman filter (EKF) is designed to track time-variant parameters.
Recently, a sequential Monte Carlo approach in the Bayesian framework has been widely investigated in the wireless communication systems. These particle filtering algorithms estimate the state information with a posterior probability. Moreover, they can achieve theoretical optima in the presence of nonlinear and non-Gaussian models. In comparison with other nonlinear estimation methods such as EKF and unscented Kalman filter (UKF), a particle filter will yield more accurate results along with the flexibility of the algorithm design .However, in terms of this high-dimensional parameter estimation, it is very hard to get appropriate particles with small distribution variance on each dimension by the general sequential importance sampling (SIS) method.

IJSER © 2015

International Journal of Scientific & Engineering Research, Volume 6, Issue 4, April-2015 520

ISSN 2229-5518

When the transmit/receive antennas in a
MIMO system share one oscillator, the difference of
CFOs among different antenna pairs can be
time-domain vector by an 𝑁-point inverse discrete
Fourier transform (IDFT):

𝑝 𝑝

negligible. Under such a scenario, the particle filter is
investigated to marginalize out the unknown time- variant Rayleigh channel from the sampling space. The channel is estimated by the Kalman filter on the assumption that its response model is known. The particle weights of CFO are evaluated according to the estimation error. Simulation results demonstrate the algorithm can estimate the CFO and the channel jointly with high accuracy, as well as its robustness to small variations of the channel model.


𝑑𝑘 (𝑛) = 𝑤𝐻 𝑑𝑘 (𝑛) (1)
Where is the IDFT matrix. To prevent inter-
symbol interference (ISI), a cyclic prefix (CP) of 𝑁𝑔
symbols is appended in front of each IDFT output
block. The resulting vector of length is digital-to- analog converted by a pulse-shaping filter with a
finite support on [0, 𝑇𝑑] where with 1/(𝑁𝑇𝑠) being
the subcarrier spacing. Finally, the analog signal from
the pulse-shaping filter is transmitted from the 𝑝-th
antenna over the channel.
That is, be the normalized carrier frequency offset with respect to (w.r.t.) the carrier spacing
1/(𝑁𝑇𝑠) between transmit antennas of the 𝑘-th user and the 𝑞-th receive antenna of the BS. In the
presence of CFOs, the received vector signal after
removing the guard interval becomes

𝑞 𝑝


𝑟𝑞 (𝑛) = ∑𝑘
∆(∈𝑘 (𝑛) ∑𝑚𝑡
𝐷𝑘 (𝑛)ℎ𝑘
(𝑛) +
𝑣𝑞(𝑛) 0
Figure1 Block diagram for OFDM SIGNAL

𝑘 𝑝 (𝑛)ℎ𝑝.𝑘 (𝑛) + 𝑣𝑞 (𝑛)

= ∑𝑘=1 𝐷𝑘 𝑘

2.1 MIMO OFDM System:

Consider an uplink OFDM system with 𝑁 subcarriers and 𝐾 active users. The base station (BS) and each active user are equipped with 𝑀𝑟 and 𝑀𝑡 antennas, respectively. Each user is assigned 𝑁𝑘
exclusive subcarriers, we denote the index set of
carriers assigned to the 𝑘-th user as where 1 ≤ 𝑖𝑙 ≤ 𝑁
for 𝑙 = 1, 2, 𝑁𝑘. The proposed algorithms are
applicable to any carrier assignment scheme (CAS).
Note that even though we are using orthogonal subcarriers, self-interference and MAI are inevitable
Recall that the CP length equals the channel length plus timing offset. Under such an , the timing errors do not explicitly appear in the received signal model. Thus, we have suppressed the timing errors in (2). Several approaches have been proposed to model the time-varying channels and frequency offsets in mobile environments. Since we assume a normalized
Doppler spread 𝑓𝐷𝑇𝑑 ≪ 1,we adopt the following
first-order autoregressive (AR) parametric model
widely used to characterize the time-varying frequency offset and channel responses.

𝑞 (𝑛) = 𝛼𝑞 𝑞 (𝑛 − 1)

due to CFOs Denote by the data symbols transmitted

𝑘 𝐸 𝑘

by the 𝑘-th user from the 𝑝- th transmit antenna over the 𝑛-th OFDMA block. For convenience, we assume
that the data symbols are taken from the same
+ 𝑤𝑞 (𝑛)

𝑞 (𝑛) = 𝛼𝑞 𝑞 (𝑛 − 1) + 𝑤𝑞 (𝑛) (3)

complex-valued finite alphabet and independently
identically distributed. The 𝑖-th entry of is non-zero
if and only if .next, is converted to the corresponding

𝑘 𝐸 𝑘

𝑘 ℎ

IJSER © 2015

International Journal of Scientific & Engineering Research, Volume 6, Issue 4, April-2015 521

ISSN 2229-5518

Furthermore, we assume that Note that we assume independent fading across transmit antennas and multi-paths. The time variation in CFO in (4) arises from (a) local oscillator instability due to temperature/voltage variations and (b) changes in relative platform Doppler velocity. In the next section, we propose a pilot-aided parallel Schmidt extended particle filter approach to estimate based on
the received signal (𝑛), assuming exact knowledge of
and in the BS.
The incoming input binary streams are first mapped into constellation points according to any of the digital modulation schemes such as QPSK/QAM. In QPSK (Quadrature Phase Shift Keying) modulation, the incoming binary bits are combined in the form of two bits and are mapped into constellation point. The N constellation points are modulated using N sub- carriers whose carrier frequencies are orthogonal in nature. The modulation is similar to taking inverse discrete/fast fourier transform (IDFT/IFFT) operation. The output of N point (IFFT) block is the OFDM signal. Now the N OFDM signal samples are combined and then transmitted i.e., the parallel samples are now
Ec is the energy of subcarrier, All the preamble based frequency offset estimation methods given in literature aims at accuracy and increasing the range of frequency offset estimation. The importance of frequency offset estimation in various high speed broadband wireless applications can be understood

2.2 CFO Estimation :

Particle filtering is a sequential sampling method built on the Bayesian paradigm. From the Bayesian theory, at sample k , the posterior distribution p(x0:k | r0:k) is the main entity of interest. However, due to the nonlinearity of the measurement equation, its analytical expression is not tractable. Alternatively, particle filtering can be applied to approximate this PDF by stochastic samples generated using a sequential importance sampling strategy. Particle filtering is an extension of the sequential methodology. It consists in recursively estimating the required posterior density function p(x0:k | r0:k) by a set of M random samples with associated weights, denoted by


converted into serial sequence and then it is

𝑗 �𝑤(𝑗) (6)

transmitted. The OFDM baseband signal at the transmitter is expressed as in

𝑝̂ �
� = � 𝛿�𝑥0:𝑘 − 𝑥0:𝑘 𝑘


1/√𝑛 ∑𝑁−1 𝑥(𝑘)𝑒𝑗2𝜋𝑛𝑘/𝑛
𝑥(𝑛) =
Where is drawn from the importance function is the
Dirac delta function and is the normalized importance

𝑘=0 . 0<n<N-1 (4)

weight associated with the j-th particle. The weights are updated according to the

(𝑚) ,𝛾 �𝑝�𝛾 𝑥 (𝑚) ,𝛾 �

n -time domain sample index
𝑤(𝑚) 𝛼

𝑝�𝛾 𝑘 𝑥0:𝑘−1

(𝑚) 𝑚


𝑤𝑚 (7)
X (k) -modulated QPSK data symbol on the kth


/𝑥0:𝑘−1 𝛾0:𝑘)

N -total number of subcarriers and x (n) -OFDM signal.
In order to maintain a signal to noise ratio (SNR) of
20 decibels or greater for the OFDM carriers, offset is limited to 4% or less than the inter carrier spacing which is simulated in the lower bound for the SNR at the output of the DFT for the OFDM carriers in a channel with AWGN and frequency offset is derived as and is given by

𝑆𝑁𝑅 ≥ �𝐸𝑐 � �𝑠𝑖𝑛𝜋𝑣 � /{10.594(𝐸𝑐 )}( {𝑠𝑖𝑛𝜋𝑣)2 } (5)
After a few iterations, particle filtering is known to suffer from degeneracy problems. So we integrate a resampling step to select particles for new generations in proportion to the importance weightsThis is due to the fact that the particle filter is then only used to estimate the nonlinear states, while the remaining conditional linear-Gaussian states are estimated using the closed-form particle filter. In our case, conditionally on the nonlinear state variables the DSS model contains a linear substructure on h,
subject to Gaussian noise. Using the Bayes' theorem,


𝜋𝑣 2


IJSER © 2015

International Journal of Scientific & Engineering Research, Volume 6, Issue 4, April-2015 522

ISSN 2229-5518

the posterior density function of interest can be
written as
𝐸𝑝 [𝐵(𝑥𝜖𝐴)] = � 𝑝(𝑥). 𝐵(𝑥𝜖𝐴)𝑑𝑥

𝑝 �𝑥0:𝑘� = 𝑝(
) (8)

𝑝 (𝑥)


. 𝜋(𝑥). 𝐵(𝑥𝜀𝐴)𝑑𝑥 (13)



where _ is a distribution for which we require that
Where is analytically tractable and can be obtained via a particle filter associated with each particle. Indeed, the j-th PDF is a multi dimensional gaussian
𝑝(𝑥) > 0, 𝜋(𝑥) > 0
Thus, we can define a weight w(x) a
probability density function. The mean and the
covariance can be obtained using the particle
𝑤(𝑥) =

filtering equations given by the time update equations
This weight w is used to account for the differences between p and the _. This leads to









∑ 𝑗𝑘


∑ 𝑗𝑘



Ep [B(xϵA)] = � p(x). w(x). B(xϵA)dx

2.3 Particle Filter:

A particle filter is a nonparametric implementation of the Bayes filter and is frequently used to estimate the state of a dynamic system. The key idea is to represent a posterior by a set of hypotheses. Each hypothesis represents one potential
state the system might be in. The state hypotheses are
= Eπ [w(x)B(xϵA)] (15)
Let us consider again the sample-based
representations and suppose the sample are drawn from. By counting all the particles that fall into the region A, we can compute the integral of over A by the sum over samples
represented by a set S of N weighted random samples
[∫ π(x)dx ≈ 1/n ∑n
B(s ∈ A)

i=1 ] (16)

𝑆 = {< 𝑆[𝑖] , 𝑊 [𝑖] > 𝑖 = 1, … … 𝑁} (10)
where s[i] is the state vector of the i-th sample and
w[i] the corresponding importance
weight. The weight is a non-zero value and the sum over all weights is 1. The sample
set represents the distribution



3.1 CFO Estimation on SNR:

The increment of SNR estimation accuracy of channel tends to be higher .It can also be observed that the number of the particles is another direction Factor of the estimation accuracy

𝑝(𝑥) = ∑𝑖=1 𝑤𝑖 𝛿𝑠 (𝑖)(𝑥)



We are faced with the problem of computing the
expectation that x ∈ A, where A
is a region. In general, the expectation Ep[f(x)] of a
function f is defined as
𝐸𝑝 [𝑓(𝑥)] = � 𝑝(𝑥). 𝑓(𝑥)𝑑𝑥 (12)
Let B be a function which returns 1 if its
argument is true and 0 otherwise. We can
express the expectation that x ∈ A by

IJSER © 2015

International Journal of Scientific & Engineering Research, Volume 6, Issue 4, April-2015 523

ISSN 2229-5518




CFO estimation SNR

snr 0dB snr 6dB snr 12dB



CFO estimation NP














0 5 10 15 20 25 30 35 40 45 50

Calculation steps

Figure2: CFO Estimation on SNR



0 5 10 15 20 25 30 35 40 45 50

Calculation steps

FIGURE 3: CFO Estimation on number of Particles





Calculation steps












Table1: Comparison for CFO Estimation SNR

3.3 CFO Estimation on number Of Particles:

As the of particles increase the estimation error will be lower. In addition the particle with large weights have large number than that of the particle with lower weight.

Table2: Comparison for CFO Estimation NP


Particle filtering algorithm for frequency synchronization in MIMO-OFDM systems. It is based on the sequential Monte Carlo approaches in the Bayesian framework, and simplified to a one- dimensional particle sampling in order to improve the computation efficiency. Numerical simulation shows that the algorithm can obtain high accuracy of CFO and estimation performance. Additionally, it can simultaneously estimate time-varying CFO and channel. It also has some robustness to the variation
of the channel model.

IJSER © 2015

International Journal of Scientific & Engineering Research, Volume 6, Issue 4, April-2015 524

ISSN 2229-5518


[1] ZHANG Jing J Shanghai Univ (Engl Ed), 2009,
13(6): 438–443 Particle filter for joint frequency offset and channel estimation in MIMO-OFDM systems
[2] Pham T H, Nallanathan A, Liang Y C. Joint channel and frequency offset estimation in distributed MIMO flat-fading channels [J]. IEEE Transaction on Wireless Communication, 2008, 7(2):
[3] Liang Y M, Luo HW, Huang J G. Extended Kalman filtering-based channel estimation for space- time coded MIMO-OFDM systems [J]. Journal of Shanghai University (English Edition), 2007, 11(5):
[4] Magnus S, Darren M, Steve P. Analysis of frequency-offset tracking in MIMO OFDM systems [J]. IEEE Transaction on Communication, 2006,
[5] Qian Y, Guaoan B, Chunru W. SMC-based blind detection for DS-CDMA systems over multipath fading channels [J]. IEEE Transaction on Communication,2006, 54(6): 971–974.
[6] Huang Y F, Zhang J Q, Luna I T, Djuric P M, Padiuo D P R. Adaptive blind multiuser detection over flat fast fading channels using particle filtering [J].EURASIP Journal on Wireless Communication and Networking, 2005(2): 130–140.
[7] Elena P, Arnaud D, William J F. Particle filtering for joint symbol and code delay estimation in DS spread spectrum systems in multipath environment [J]. EURASIP Journal on Applied Signal Processing,
2004(15): 2306–2314.
[8] S. Shahbazpanahi, A. B. Gershman, and G. B. Giannakis, “Joint blind channel and carrier
frequency offset estimation in orthogonally spacetime block coded MIMO systems,” in Proc. IEEE Workshop Signal Processing Advances Wireless
Commun., 2005, pp. 363–367.

Author Profile

1 K.CHAITHANYA NIKLESH is pursuing his master’s degree in communication systems from department of electronics and communication, SRM University, Kattankulathur. His research interests include wireless communication, multiple-input- multiple-output.(MIMO).

2 P.MALARVEZHI is working as assistant professor in department of electronics and communication, SRM University, Kattankulathur

IJSER © 2015

International Journal of Scientific & Engineering Research, Volume 6, Issue 4, April-2015 525

ISSN 2229-5518

IJSER © 2015