Performance analysis of the subspace method for blind channel identification

We study the performance of the subspace (SS) method 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 variance of the SS method is derived. Its performance is studied and compared with that of the cross-relation (CR) method and the Cramer-Rao bound (CRB). It is shown that the SS method is more robust to ill channel conditions than the CR method, but the CR method is generally more efficient in computation. The two methods show nearly identical performances for well-conditioned channels.