Channel Estimation Algorithms for OFDM Systems

Prasanta Kumar Pradhan , Oliver Faust , Sarat Kumar Patra and Beng Koon Chua
of Electrons & Comm. Engg. National Institute of Technology, Rourkela, India 769008 Email: pradhanp, [email protected] DSP Center, School of Engineering Ngee Ann Polytechnic, Singapore 599489 Email: fol2, [email protected]
Dept.

AbstractIn this work we have compared different type of channel estimation algorithm for Orthogonal Frequency Division Multiplexing (OFDM) systems. The result of the Adaptive Boosting (AdaBoost) algorithm was compared with other algorithms such Least Square (LS), Best Linear Unbiased Estimator (BLUE) and Minimum Mean Square Error (MMSE). Index TermsOFDM, Channel Estimation, LS, MMSE.

I. I NTRODUCTION In communication system many techniques, like Frequency Division Multiplexing Access (FDMA), Time Division multiplexing Access (TDMA) and Code Division Multiplexing Access (CDMA), are used for transmission of signal. Where FDMA has very bad spectrum usage and TDMA performance degrades by multipath delay spread causing Inter Symbol Interference (ISI). In contrast OFDM enables high bit-rate wireless applications in a multipath radio environment the need for complex receivers. OFDM is a multi-channel modulation system employing Frequency Division Multiplexing (FDM) of orthogonal sub-carriers, each modulating a low bit-rate digital stream. OFDM uses N overlapping (but orthogonal) sub bands, each carrying a baud rate of 1/T and they are spaced 1/T apart. Because of the selected frequency spacing, the sub-carriers are all mathematically orthogonal to each other. This permits proper demodulation of symbol streams without the requirement of non overlapping spectra. Currently, there is increasing interest in OFDM as it combines the advantages of high data rates and easy implementation. This is reected by the many standards that considered and adopted OFDM those for Digital Audio Broadcast (DAB) and Digital video Broadcast (DVB), high speed modems over digital subscriber lines, and Wireless Local Area Network (WLAN) broadband systems as of ieee 802.11a, 802.11b and 802.11g. The accuracy of channel state information greatly inuences the overall system performance [1]. The main challenges associated with OFDM systems today are- channel identication and tracking, channel coding and equalization. In wideband mobile channels, pilot-based signal correction schemes are feasible method for OFDM systems [2]. Most channel estimation methods for OFDM transmission systems have been developed under the assumption of a slow fading channel, where the channel transfer function is assumed stationary within one OFDM data block. In addition, the channel transfer

function for the previous OFDM data block is used as the transfer function for the present data block. In practice, the channel transfer function of a wideband radio channel may have signicant changes even within one OFDM data block. Therefore, it is preferable to estimate channel characteristic based on the pilot signals in each individual OFDM data block [3]. Recently, an elegant channel estimation method for OFDM mobile communication systems has been proposed by [4] in which a semi-blind low complexity frequency domain based channel estimation algorithm for multi-access OFDM systems was proposed [4]. Many researchers have pursued channel estimation in the time domain. A joint carrier frequency synchronization and channel estimation scheme, using the expectation-maximization (EM) approach, is presented in [5] while [6] used subspace tracking. In [7], a joint and data estimation algorithm is presented which makes a collective use of data and channel constraints. A joint frequency-offset and channel estimation technique for multi-symbol encapsulated (MSE) OFDM system is proposed in [8], while the authors of [9] presented a sequential method based on carrier frequency offset and symbol timing estimation. The authors of [10] estimated the channel based on power spectral density (PSD) and least squares (LS) estimation for OFDM systems with timing offsets. A pilot aided channel estimation algorithm in the presence of synchronous noise by exploiting the a priori available information about the interference structure was presented in [11] while [12] used implicit pilots for joint detection and channel estimation. A joint time domain tracking of channel frequency offset for OFDM systems is suggested in [13] while a time domain carrier frequency offset (CFO) tracking method based on Particle ltering is presented in [14]. Radial Basis Function (RBF) network based channel estimation has been investigated in [15]. This is based on using the radial basis function (RBF) network to model the dynamics of the fading process. In one-dimensional channel estimation, only the time-correlation of the fading channel is exploited, whereas in two-dimensional channel estimation, both the time and the frequency correlations of the fading channels are exploited. RBF networks are essentially nonlinear interpolators of the pilot channels. AdaBoost based channel estimation is proposed in [16]. AdaBoost not only increases the performance of the OFDM systems it also renders the

Bit Stream Mapping

Serial To Parallel

Pilot Insert

IFFT

Add Cyclic Prefix

Parallel To Serial Channel

Pilot Carrier Data Carrier

Pilot Carrier Data Carrier

AWGN
Desired Bit stream

Demap

Parallel To Serial

FFT

Pilot Extract

Remove Cyclic Prefix

Serial To Parallel
Frequency
Time

(a) Block Type Fig. 2.

Frequency

Time

(b) Comb Type Pilot Arrangements

Channel Estimate

Fig. 1.

Block Diagram of the OFDM systems

QAM mapping obsolete and thereby reducing the complexity of receiver designs. The rest of the article is arranged as follows: Section. 2 describes the mathematical model of OFDM systems and its channel model along with two types of pilot arrangement. The implementation of AdaBoost and other algorithms (like LS, MMSE, and BLUE) is discussed in Section. 3 . Section. 4 describes the performance analysis of different algorithms and nally Section. 5 draws up the conclusions. II. S YSTEM D ESCRIPTION The block diagram of an OFDM system is given in Figure.1. The binary information data are grouped and mapped into multi-amplitude-multi-phase signals. In this paper, we consider the 16-QAM modulation. After pilot insertion, the modulated data X (k ) are sent to an IDFT and are transformed and multiplexed into x(n) as x(n) = IF F T {X (k )} =
N 1 k=0

Where r the total number of propagation paths, hi is the complex impulse response of the ith path, fDi is the ith paths Doppler frequency shift which causes Inter channel Interference (ICI) of the received signals, is the delay spread index, and i is the ith path delay time normalized by the sampling time. After removing the guard interval from yg (n), the received samples y (n) are sent to a Fast Fourier Transform (FFT) block to demultiplex the multi-carrier signals. Y (k ) = F F T {y (n)} =
1 N N 1 n=0

y (n)ej 2kn/N

(5)

f or k = 0, 1, . . . , N 1 If we assume that the guard interval is longer than the length of channe1 impulse response- there is no inter-symbol interference between OFDM symbols- the demultiplexed samples Y (k ) can be represented by Y (k ) = X (k )H (k ) + W (k ),
j 2f T

k = 0, 1, . . . , N 1
2i

(6)

X (k )ej 2kn/N

(1)

f or n = 0, 1, ..., N 1 where N is the number of subcarriers. The guard interval Ng is inserted to prevent possible inter-symbol interference in OFDM systems, and the resultant samples xg (n) are xg (n) = x(N + n) n = Ng , Ng 1, . . . , 1 x(n) n = 0, 1, . . . , N 1 (2)

Di Where H (k ) = hi e ej N k and W (k ) is the Fourier transform of the AWGN w(k ). After that, the received pilot signals {Yp (k )} are extracted from {Y (k )}, the channel transfer function {H (k )} can be obtained from the information carried by {Hp (k )}. With the knowledge of the channel responses {H (k )}, the transmitted data samples {X (k )} can be recovered by simply dividing the received signal by the channel response:

sin(fDi T ) fDi T

x (k ) =

y (k ) (k ) H

(7)

where Ng is the number of samples in the guard interval. The transmitted signal is then sent to a frequency selective multipath fading channel. The received signal can be represented by yg (n) = xg (n) h(n) + w(n) (3) Where h(n) is the channel impulse response (CIR) and w(n) is the Additive White Gaussian Noise (AWGN) and is the circular convolution. The channel impulse response h(n) can be expressed as [17]
r 1

h(n) =
i=0

hi ej 2fDi T N ( i )

(4)

(k ) is an estimate of H (k ) . After signal demapping, where H the source binary information data are reconstructed at the receiver output. The OFDM transmission scheme makes it easy to assign pilots in both time and frequency domain. Figure.2 shows two major types of pilot arrangement. The rst kind of pilot arrangement shown in Figure.2a is denoted as block-type pilot arrangement. The pilot signal is assigned to a particular OFDM block, which is sent periodically in time domain. This type of pilot arrangement is especially suitable for slow-fading radio channels. The estimation of channel response is usually obtained by either LS or MMSE estimates of training pilots [5]. The second kind of pilot arrangement, shown in Figure. 2b, is denoted as comb-type pilot arrangement. The pilot signals

are uniformly distributed within each OFDM block. Assuming that the payloads of pilot signals of the two arrangements are the same, the comb-type pilot assignment has a higher retransmission rate. Thus, the comb-type pilot arrangement system is provides better resistance to fast-fading channels. Since only some sub-carriers contain the pilot signal, the channel response of nonpilot subcarriers will be estimated by interpolating neighboring pilot sub-channels. Thus, the combtype pilot arrangement is sensitive to frequency selectivity when comparing to the block-type pilot arrangement system. That is, the pilot spacing (f )p , must be much smaller than the coherence bandwidth of the channel (f )c [3]. III. C HANNEL E STIMATION For comb-type pilot sub-carrier arrangement, the Np pilot signals Xp (m) , m = 0, 1, ..., Np 1, are uniformly inserted into X (k ). That is, the total N sub-carriers are divided into Np groups, each with L = N/Np adjacent sub-carriers. In each group, the rst sub-carrier is used to transmit pilot signal. The OFDM signal modulated on the k th sub-carrier can be expressed as X (k ) = X (mL + l) Xp (m), = inf ormation data,

systems which are based on comb-type pilot arrangement is highly dependent on the rigorousness of estimate of pilot signals. Thus, a estimate better than the LS estimate is required. The MMSE estimate has been shown to be better than the LS estimate for channel estimation in OFDM systems based on block-type pilot arrangement. The Mean Square Error (MSE) estimation, given in [17], shows that MMSE estimate has about 10-15 dB gain in SNR over the LS estimate for the same MSE values. The major drawback of the MMSE estimate is its high complexity, which grows exponentially with the observation samples. B. MMSE Estimation The MMSE estimator employs the second order statistics of the channel conditions in order to minimizes Mean Square Error (MSE). Let Rhh , RHH , and RY Y be the autocovariance matrix of h, H , and Y , respectively and RhY be the cross 2 covariance matrix between h and Y . Also w denotes the 2 AWGN variance E {|w |}. Assume the channel vector h, and the AWGN w are uncorrelated, it can be derived that [18] RHH = E {HH H } = E {(F h)(F h)H } = F Rhh F H (13) where

l = 0, l = 1, 2, . . . , L 1

Let HP = [Hp (0) Hp (1) . . . Hp (Np 1)]T = [H (0) H (L 1) . . . H ((Np 1)(L 1))]T

(8)

00 T WN . . F = . (N 1)0 T WN
k

... .. . ...

T WN . . .

0(N 1)

(14)

(N 1)(N 1) T WN

be the channel response of pilot carriers, and Yp = [Yp (0) Yp (1) . . . Yp (Np 1)]

(9)

(10)

i,k = 1 and T WN ej 2i N is called as the twidle factor N 2 matrix. Assuming Rhh (thus RHH ) and w are known at the receiver in advance, the MMSE estimator of the h is given by M M SE = R R Y Y H . At last it can be estimated that h hY Y Y

be a vector of received pilot signals. The received pilot signal vector Yp can be expressed as Yp = Xp Hp + Wp where Xp (0) . Xp = . . 0 ... 0 . . . (11)

Xp (1) . . . 0 0 . . . Xp (Np 1) where Wp is the vector of Gaussian noise in pilot sub-carriers. A. LS Estimation In conventional comb-type pilot based channel estimation methods, the estimation of pilot signals, is based on the LS method is given by p,ls H = [Hp,ls (0) Hp,ls (1) . . . Hp,ls (Np 1)] 1 = Xp Yp Y (Np 1) Yp (0) Yp (1) ] = [ Xp (0) Xp (1) . . . Xp p (Np 1) (12)

M M SE = R [R + 2 (XX Ht )1 ]1 H M M SE = F h LS H w HH HH (15) MMSE employs the second order statistics of the channel for estimation. Some times the channel statistics are not available, so it is difcult to estimate the channel in this situation. However, in OFDM systems the signal can be available at the receiver by means of pilot carriers. C. BLUE If we restrict the estimator to be linear in data and and nd the linear estimator that is unbiased and minimum variance then the estimator is called Best Linear Unbiased Estimator (BLUE). BLUE can be determined with knowledge of only the rst and second moment of the PDF. Since complete knowledge of the PDF is not necessary, the BLUE is more suitable for practical implementation [19]. Gauss-Markov Theorem 1. If the data can be modeled in the following linear form Y = XH + W (16)

The LS estimate of Hp is susceptible to AWGN and InterCarrier Interference (ICI). Because the channel responses of data subcarriers are obtained by interpolating the channel characterstics of pilot subcarriers, the performance of OFDM

where X is a known N p matrix and H is a p 1 vector of parameters to be estimated, and W is a N 1 noise vector

with zero mean and covariance C (the PDF of W is otherwise arbitrary ), then BLUE of H is given by HBLU E = (H Ht C 1 H )1 H Ht C 1 X (17)

TABLE I OFDM PARAMETERS No of Subcarriers No. of Pilot Carriers Guard Interval Guard Type 256 32 64 Cyclic Extension

where C = (X E {x})(X E {x})Ht and Ht is the conjugate transpose or Hermitian Transpose. and the minimum i is variance of H i ) = [(H Ht C 1 H )1 ]ii var(H D. AdaBoost Boosting is a general method for improving the accuracy of any given learning algorithm. AdaBoost was originally dened for two class problems but it can be extended boosting to multi-class and regression problems. The AdaBoost algorithm, introduced in 1995 by Freund and Schapire, solved many of the practical problems of the earlier boosting algorithms [20]. The AdaBoost algorithm for multiclass problem is described as below. Suppose we are given a set of training data (x1 , c1 ), (x2 , c2 ), . . . , (xn , cn ), where the input (prediction variable) xi Rp and the output (response variable) ci is quantitative values in anite set, e.g. 1, 2, . . . , K . Where K is the number of classes. Usually it is assumed that the training data are composed from independently and identically distributed (iid) samples from a unknown probability distribution P rob(X, C ). The gaol is to nd out a classication rule C (x) from the training data, so that when given a new input x, we can assign a class label c from 1, 2, . . . , K . The misclassication error rate of a classier K C (x) is given by 1 k=1 EX [IC (X )=k P rob(C = k |X )]. So from [20] C (x) = arg maxk P rob(C = k |X = x) (19) will minimize this quantity with the misclassication error rate equal to 1 EX maxk P rob(C = k |X ). This classier is known as Bayes Classier and its rate is Bayes Error Rate. The multiclass AdaBoost algorithm tries to approximate the Bayes Classier C (x) by combining many weak classiers. Starting with an unweighted training sample, the AdaBoost builds a classier that produces class labels. If a training data point is misclassied, the weight of that training data point is increased (boosted). A second classier is built using these new weights, which are no longer equal. Again, misclassied training data have their weights boosted and the procedure is repeated. Typically, one may build 500 or 1000 classiers this way. A score is assigned to each classier, and the nal classier is dened as the score weighted linear combination of the classiers from each stage. Specically, let T (x) denote a weak multi-class classier that assigns a class label to x, then the AdaBoost algorithm proceeds as follows: [21] Multiclass AdaBoost Algorithm: 1. The Algorithm is as follows: 1) Initialize the observation weights wi = 1/n, i = 1, 2, . . . , n. 2) For m = 1 to M: (18)

Fig. 3.

Performance Comparison of Different Algorithms

a) Fit a classier T (m) (x) to the training data using weights wi . b) Compute
n n

err(m) =
i=1

wi I(ci = T (m) (xi ))/

wi (20) (21)

c) Compute (m) = log d) Set 1 err(m) + log(K 1) err(m)

wi wi exp((m) . I(ci = T (m) (xi ))) (22) for i=1, 2, . . . , n. e) Re-normalize wi . 3) Output
M

C (x) = arg maxk

(m) I (T (m) (x) = k ) (23)

IV. S IMULATION R ESULTS The OFDM system channel estimation was simulated with LS, MMSE, BLUE and AdaBoost methods. In all simulations we have used QAM16 as the modulation scheme for the individual carriers. Other parameters of the simulation are given in the Table.I. Figure.3 shows comparison of Bit Error Rate (BER) for LS, MMSE, BLUE and AdaBoost. Figure.3 shows that

the AdaBoost algorithm improves the performance specially in low SNR values. However, at high SNR both MMSE and AdaBoost show a similar performance. Furtheremore, AdaBoost gives better performance when compared to LS and BLUE. Nevertheless, BLUE and LS are performing same or we can tell BLUE gives very marginal improvement to LS. The reason for this performance increase is because of the covariance matrix used in the BLUE. As our noise is ASWGN and it has variance of 1 so the BLUEs performance is all most that of LS algorithm. V. C ONCLUSSION The computational complexity of the MMSE increases exponentially as the number of carrier increases. Whereas the computatinal complexity is not exponential in the case of AdaBoost. Hence, AdaBoost can be employed when a high number of carriers is required. Furthermore, as it is a classication algorithm so in the receiver side we will require a separate demapper (or decoder) to get the desired data bits. AdaBoost not only increases the performance of the OFDM systems it also renders the QAM mapping obsolete and thereby reducing the complexity of receiver designs. R EFERENCES
[1] Y. Li and G. Stber, Orthogonal Frequency Division Mutiplexing for Wireless Communication, Y. Li and G. Stber, Eds. Springer, 2006. [2] F. Tufvesson and T. Masseng, Pilot aided channel estimation for ofdm in mobile cellular system, in in Proceeding IEEE 47th Vehicular Technology Conference. Phoneix, USA: IEEE, May 1997, pp. 11 639 1643. [3] M.-H. Hsieh and C.-H. Wei, Channel estimation for ofdm systems based on comb type pilot arrangement in frequency selective fading channels, IEEE Transactions on Consumer Electronics, vol. 44, no. 1, pp. 217225, February 1998. [4] M. Sohail and T.Y.Al-Naffouri, An em based frequency domain channel estimation algorithm for multi access ofdm systems, Elsevier Journal of Signal Processing, vol. 10, pp. 15621572, 2010. [5] J. Lee, J. Han, and S. Kim, Joint carrier frequency synchronization and channel estimation for ofdm systems via em algorithm, IEEE Transactions on Vehicular Technology, vol. 55, no. 1, pp. 167172, 2006. [6] X. Hou, Z. Zhang, and H. Kayama, Time domain mmse channel estimation based on sbspace tracking for ofdm systems, in IEEE 63rd Vehicular Technology Conference, Melbourne, Victoria, May 2006, pp. 15651569. [7] T. Al-Naffouri, An em based forward backward kalman lter for estimation time variant channels in ofdm, IEEE Transaction on Signal Processing, vol. 1, no. 11, pp. 17, 2007. [8] X. Wang, Y. Wu, J. Chouinard, and H. Wu, On the design and performance analysis of multisymbol encapsulated ofdm systems, IEEE Transactions on Vehicular Technology, vol. 55, no. 3, pp. 9901002, 2006. [9] J. Liu and J. Li, Parameter estimation and error reduction for ofdm based wlans, IEEE Transactions on Mobile Computing, vol. 3, no. 2, pp. 152163, 2004. [10] S. Qin, P. Liu, L. Zheng, and D. Wang, Channel estimation with timing offset based on psd & ls estimation for wireless ofdm systems, in International Symposium on Intelligent Signal Processing and Communication Systems, November-December 2007, pp. 248251. [11] A. Jeremic, T. Thomas, and A. Nehorai, Ofdm channel estimation in the presence of interference, IEEE Transactions on Signal Processing, vol. 52, no. 12, pp. 34293439, 2004. [12] R.Dinis, N.Souto, J.Silva, A.Kumar, and A.Correia, Joint detection and channel estimation for ofdm signals with implicit pilots, in IEEE Mobile and Wireless Communications Summit, July 2007, pp. 15. [13] T.Roman, M.Enescu, and V.Koivunen, Joint time-domain tracking of channel and frequency offsets for mimo ofdm systems, in Wireless Personal Communications, vol. 31, no. 3-4, 2004, pp. 181200.

[14] T.Nyblom, T.Roman, M.Enescu, and V.Koivunen, Time-varying carrier offset tracking in ofdm systems using particle ltering, in Proceedings of the Fourth IEEE International Symposium on Signal Processing and Information Technology, 2004, pp. 217220. [15] X. Zhou and X. Wang, Channel estimation for ofdm systems using adaptive radial basis function networks, IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, vol. 52, no. 1, pp. 4859, January 2003. [16] P. K. Pradhan, O. Faust, S. K. Patra, and B. K. Chua, Channel estimation of ofdm system using adaptive boosting algorithm, in Submitted to ICC 2011. [17] J. V. de Beek, O. Edfors, M. Sandel, S. Wilson, and P. Borjesson, On channel estimation in ofdm systems, in IEEE 45th Vehicular Technology Conference, 1995. [18] Y. Shen and E. Martinez, Channel estimation in ofdm systems, in Frescale Semiconductor Application Note, 2006. [19] S. M. Kay, Fundamentals of Statistical Signal Processing- Estimation Theory. Prentice Hall PTR, 1993, vol. 1. [20] Y. Freund and R. E. Scapire, A short introduction to boosting, Journal of Japaneese Society for Articial Intelligence, vol. 14, no. 5, pp. 771 780, 1999. [21] J. Zhu, H. Z. S. Rosset, and T. Hastie, Multiclass adaboost, Journal of Statistics and its Interface, vol. 2, pp. 349360, 2009.