Blind multichannel identification in the stationary colored noise case

We investigate a new approach to identify a multichannel blindly when the additive noise is no longer considered as white but only stationary. We consider the case when multiple FIR channels are obtained from an array of n antennas and from oversampling the received signal with a factor m. In this case the covariance matrix of the stationary noise is block Toeplitz with n×n blocks, whereas the covariance matrices of the signal part and the total received signal are block Toeplitz with mn×mn blocks. We mainly concentrate on a subspace method based on an appropriate displacement of the covariance matrix in which the noise contribution disappears. The technique developed appears to give acceptable performance for the channel estimate, compared to the Cramer-Rao bound. Furthermore, since the proposed method is based on a parameterization in terms of the channel impulse response, prior knowledge of the transmission filter can easily be incorporated. The method mentioned previously can be extended to exploit this knowledge and hence its performance can be improved. Simulations are presented to illustrate the performance of the method