Fine Residual Carrier Frequency and Sampling Frequency Estimation in Wireless OFDM Systems
This paper presents a novel algorithm for residual phase estimation in wireless OFDM systems, including the carrier frequency offset (CFO) and the sampling frequency offset (SFO). The subcarriers are partitioned into several regions which exhibit pairwise correlations. The phase increment between successive OFDM blocks is exploited which can be estimated by two estimators with different computational loads. Numerical results of estimation variance are presented. Simulations indicate performance improvement of the proposed technique over several conventional schemes in a multipath channel.
Although the Orthogonal-Frequency-Division-Multiplexing (OFDM) technique significantly enhances the system performance under frequency-selective fading channels, it is vulnerable to synchronization non-idealities, including the symbol timing offset (STO), carrier frequency offset (CFO), and sampling frequency offset (SFO).
The previous works including [1, 2, 3] deal with the coarse STO and CFO estimation in time domain before Fast Fourier Transform (FFT). However, due to the imperfections of compensation, after FFT, the residual part of CFO remains to be corrected. Also, at this stage, SFO should be estimated and removed; otherwise, it would lead to a phase rotation not only proportional to the tone index within one OFDM block (inter-block increment), but also grows linearly for successive OFDM blocks (intra-block increment) .
In literature, several schemes are proposed to estimate or track the residual CFO and SFO in frequency domain with the assistance from pilot subcarriers [5, 6, 7]. In , Speth et al. utilize the symmetric locations of pilots to estimate CFO and SFO jointly. However, its performance degrades in the multipath channels.  suggests three estimators with the help of the least square estimation (LSE). An improved weighted LSE variant is proposed by Tsai et al. in  which requires the second-order statistics of the channel state information (CSI). In general, these schemes mainly rely on the linearly growing inter-block increment.
This paper proposes a novel technique to make use of the intra-block increment spanning a number of OFDM blocks. By dividing the subcarrier index into several regions, the method is capable of exploiting the pairwise correlation which leads to accurate results after applying least square fitting. Two variants differing in computation complexity are presented with their numerical variances derived.
Ii OFDM Signal Model with CFO and SFO
We consider an OFDM system where the transmitted data is modulated by an -point Inverse FFT (IFFT). Assuming a total of OFDM blocks to be delivered and each block consists of data samples (), the complex baseband signal is described by
where are the locations of the data subcarriers; for , is either pilot or null subcarrier. is the length of the guard interval, the total length of an entire OFDM block given by , and the sampling interval. is the rectangular function defined as
The multipath channel is
where is the total number of taps, the independent and Rayleigh distributed complex channel gains, the timing delay of each path, and the delta function. Here, we assume that .
Up-converting to a carrier frequency , the post-channel equivalent signal takes the form
where the notation stands for linear convolution, and the complex, identically independently distributed (i.i.d.), additive white Gaussian noise (AWGN) with zero mean and variance ; also, it is wide sense stationary (WSS), with independent real and imaginary part, and equal variance in both parts (). Now, assuming a CFO and a SFO given as
where is the deviated carrier frequency and the deviated sampling interval at the receiver. The received -th sample in the -th OFDM block is
After discarding the samples in the guard interval, the complex data for the -th block and on the -th subcarrier is 
where is the channel transfer function (CTF) in frequency domain; and the normalized CFO to the subcarrier spacing; is the amplitude attenuation approaching unity and can be safely neglected; the inter-carrier interference (ICI) due to distorted orthogonality of subcarriers; is the WSS i.i.d. Gaussian noise in frequency domain with independent real and imaginary parts. Without loss of generality, and can be regarded as the residual part of CFO and SFO after coarse synchronization or imperfect channel estimation and equalization.
Iii Proposed Technique
Define the full set of subcarrier index as , which can be further divided into equally-spaced regions (assuming even and , and is divisible by ), denoted as where
Here, denotes integers. Ignoring disturbances of ICI, using equation (II), for the -th segment in the -th OFDM block, the pairwise correlation is
and . Clearly, the extra phase rotation of the useful part in (13) is irrelevant to subcarrier index and ; it is only pertinent to the OFDM block index and segment index . The cross terms are the main disturbance in estimation. In practice, the contribution of signal and channel () should be replaced by
The notation is the full set of pilots and the full set of null subcarriers. is the estimated , obtained by the decision feedback device. is the estimated CTF. could be combined using a certain weight given as
where . See Appendix A for the details of such selection for the weighted . Computation of weighted requires the second-order statistics of signal, channel and noise, avoided by the simplified scheme. For constant-modulus modulation, the weighted reduces to
where . To obtain estimation of , we coherently stack by
where denotes the conjugation of its argument and
is the left half of while the right one; is the absolute complement of given by ; in general, the two-dimensional set . can be estimated by
The vector can be linearized into
where is the observation matrix expressed by
The vectors take the form
where ; is the associated estimation error vector. By least square fitting, is given by
Note that both and give estimation of ; denotes the -th entry of a vector/matrix. Here, we choose and arrange all into the vector which leads to
where , is the error vector, and
is the observation matrix. Another least square fitting yields
The estimated and are and respectively. A simple sketch with is drawn in Fig. 1.
Assuming correctness in tackling the phase ambiguity in the linearization process, derived in Appendix A, estimation using either the weighted or simplified is unbiased. On the other hand, the numerical variances of and are
Appendix A validates and thus , . The equality establishes if and only if (iff)
the channel experiences flat fading ()
constant modulus modulation ()
Otherwise, weighted estimation always outperforms simplified estimation. To achieve the best performance, further assuming
equal and maximal cardinality of each set , denoted by
The variances under conditions A1A3 are
where ; .
i) Apparently, an immediate way to enhance the performance is to raise , which leads to asymptotically decreasing variances in cubic scale. Nevertheless, if the application is real-time oriented rather than quality preferred, where and should be tracked in the fastest manner, and should be replaced by their minimums as .
ii) If is used for estimation, (39) and (40) are rewritten into
which are significantly higher than (39) and (40) increasing squarely with . Therefore, it is reasonable to use .
iii) According to (10), statistically, the proposed technique only relies on the independency and equal variance of the real and imaginary parts, and WSS assumptions of noise; it does not require the power spectrum density (PSD) of noise to be strictly flat (white), since the expectation of the cross terms is zero.
In this section, we consider a wireless OFDM system with FFT size . The guard interval is . Thus, the length of an entire OFDM block is . The total number of OFDM blocks is , and the total number of segments is . The carrier frequency is set at GHz. The sampling period is . For brevity and to exploit the best performance of the proposed estimators as well as other conventional pilot-assisted schemes, all subcarriers are regarded as pilots; otherwise, notations in (17) must be used which varies with the accuracy of decision feedback device. The signal is modulated from 16-PSK constellation. The channel consists of Rayleigh taps, which are statistically independent distributed with a power delay profile decaying exponentially:
For the proposed estimators and the scheme in , CTF is assumed to be known perfectly as well as 111For OFDM systems containing null subcarriers, could be estimated, which is omitted in this paper., unless otherwise mentioned. Mean squared error (MSE) results are used to benchmark the performance, defined as and where denotes the expectation of its argument.
Iv-a Comparison of
Fig. 3 highlights the comparison of among the proposed estimators with other schemes in [5, 6, 7]. Numerical result of in (39) is drawn by assuming A1A3. In the multipath channel, both of the weighted and simplified estimator achieve the best performances. In the flat fading scenario, the weighted estimator reduces to the simplified one. provides a tight bound in moderate .
Iv-B Comparison of
Iv-C with a Varying
Fig. 5 shows the deviation of when changes under dB in multipath channel. For the proposed estimators, is asymmetric for negative and positive due to the presence of a positive ; for , the performance degrades gradually with a higher since the ICI is increasing simultaneously. For a major part,  and the simplified estimator entangle with each other.
Iv-D with a Varying
Iv-E with Different Estimation Accuracy
where represents the estimation accuracy, and the additional complex noise with zero mean and variance in its real and imaginary part respectively, independent from . The weighted estimator degrades significantly when severe inaccuracy occurs, which hinders the performance improvements especially in moderate to high region. For moderate to high , the weighted estimator outperforms the simplified one under different . Similar conclusion can be drawn for .
Iv-F and under Mobility
Fig. 7 exhibits the and in presence of terminal mobility with dB under multipath channel. Merely the CSI pertinent to the first moment of the Rayleigh fading channel in simulation (=0) is assumed to be known a priori; for the ensuing frames, the same CSI is used which entails a loss in channel estimation accuracy. The Doppler bandwidths with respect to terminal speed of are Hz. In general, the performance deviation is insignificant if not imperceptible even the terminal velocity reaches , since the maximal value of the product between the Doppler bandwidth and the duration of an OFDM block is , a relatively small value. Thus, the CSI is sound enough to secure an excellent estimation.
V Concluding Remarks
In this paper, we propose a joint estimation technique to deal with residual CFO and SFO estimation. By dividing the subcarrier index into a number of regions and exploiting the pairwise correlation, we estimate the phase increment between adjacent OFDM blocks, which yields accurate estimation after times of least square fitting. Extensive simulations indicate better performance over several conventional pilot-assisted schemes.
Appendix A Bias and Variance of Estimation
If is in vicinity of , using the approximation for small enough, we may write
where and represent the real and imaginary part of the arguments. Expectation of in (A.46) is
which holds if . In fact, each component in contains either , , or (see (10)) and therefore, , which finally leads to the unbiasedness of in (31) since only linear intermediate operations are involved.
For the numerical variance of , we could use
if . Standard calculations yield
Note that, for visual clearance, we abbreviate the notation with . Thus,
To prove , we invoke the Cauchy-Schwarz-Inequality . The essential steps are listed below.
is equivalent to prove
Now, using and , the Cauchy-Schwarz-Inequality gives
and we have the conditions A1, A2 222Actually, it is merely a sufficient condition of (A.55). Specifically, consider that there is a sage at the transmitter side who could render where C is a constant, then (A.55) also establishes. However, it is not insightful to be pursued.. Therefore, we verify that .
-  J. van de Beek, M. Sandell, and P. Borjesson, “ML estimation of time and frequency offset in OFDM systems,” IEEE Trans. Signal Process., vol. 45, no. 7, pp. 1800 –1805, jul 1997.
-  P. Moose, “A technique for orthogonal frequency division multiplexing frequency offset correction,” IEEE Trans. Commun., vol. 42, no. 10, pp. 2908 –2914, oct 1994.
-  T. Schmidl and D. Cox, “Robust frequency and timing synchronization for OFDM,” IEEE Trans. Commun., vol. 45, no. 12, pp. 1613 –1621, dec 1997.
-  T. Pollet and M. Peeters, “Synchronization with DMT modulation,” IEEE Commun. Mag., vol. 37, no. 4, pp. 80 –86, apr 1999.
-  M. Speth, S. Fechtel, G. Fock, and H. Meyr, “Optimum receiver design for OFDM-based broadband transmission II: A case study,” IEEE Trans. Commun., vol. 49, no. 4, pp. 571 –578, apr 2001.
-  S.-Y. Liu and J.-W. Chong, “A study of joint tracking algorithms of carrier frequency offset and sampling clock offset for OFDM-based WLANs,” in Communications, Circuits and Systems and West Sino Expositions, IEEE 2002 International Conference on, vol. 1, june-1 july 2002, pp. 109 – 113 vol.1.
-  P.-Y. Tsai, H.-Y. Kang, and T.-D. Chiueh, “Joint weighted least-squares estimation of carrier-frequency offset and timing offset for OFDM systems over multipath fading channels,” IEEE Trans. Veh. Technol., vol. 54, no. 1, pp. 211 – 223, jan. 2005.
-  M. Speth, S. Fechtel, G. Fock, and H. Meyr, “Optimum receiver design for wireless broad-band systems using OFDM—Part I,” IEEE Trans. Commun., vol. 47, no. 11, pp. 1668 –1677, nov 1999.
-  K. S. Ahn, R. Heath, and H. K. Baik, “Shannon capacity and symbol error rate of space-time block codes in MIMO Rayleigh channels with channel estimation error,” IEEE Trans. Wireless Commun., vol. 7, no. 1, pp. 324 –333, jan. 2008.
-  D. Poole, Linear Algebra: A Modern Introduction. Brooks Cole, 2010.