Performance comparison of three methods for blind channel identification

We study the performance of the subspace (SS), cross-relation (CR) and two-step maximum likelihood (TSML) methods for estimating the impulse responses of multiple FIR channels driven by an unknown input sequence. Assuming a large data size or/and a high SNR, the estimation variances of the SS and CR methods are derived. The performances of the three methods are studied and compared against the Cramer-Rao bound (CRB). It is shown that the TSML method significantly outperforms the SS and CR methods and attains the CRB over a wide range of SNR. It is also shown that the SS method is more robust to ill channel conditions than the CR method, while the two methods show nearly identical performances for well-conditioned channels