An efficient structured estimator of sparse channels for OFDM communications

In the frame of block pilot channel estimation, an ameliorated version of the Least Squares (LS) algorithm is here proposed for OFDM systems. Former studies use the fact that the cyclic prefix (CP) duration upper bounds the length of the channel impulse response (CIR) to cancel the estimation noise lying outside the CIR first taps, equal in number to the CP samples. However, if the channel is sparse, the number of non zero coefficients is a small fraction of the channel length. This results in an important residual estimation noise and a further enhancement is possible. We propose to first derive the CIR structure by the effective taps localization. Then, an adequately chosen LS criterion is optimized to estimate the CIR. The CIR structure is derived by a computationally efficient probabilistic framework estimator (PFE). The PFE operates a non iterative thresholding on the LS estimate of the CIR to classify the taps into effective taps and zero taps. The thresholds are chosen such that the sum of false detection probability and of missing probability is minimized. Compared to former algorithms, the proposed method achieves better performance in terms of channel frequency response (CFR) estimate mean square error (MSE). It also leads to symbol error rates equivalent to those obtained with perfect channel knowledge.