Fast maximum likelihood for blind identification of multiple FIR channels

We study the estimation of the impulse responses of multiple FIR channels which are driven by a virtually arbitrary unknown input sequence (including nonstationary, non-Gaussian, or/and colored sequences). The channel identifiability conditions are derived using a Fisher information matrix, which is in contrast to several existing approaches which are based on some specific estimation methods. A fast maximum likelihood method is developed, which is a two-step procedure where each step minimizes a quadratic function. The two-step maximum likelihood (TSML) method requires no initial estimates and is asymptotically (high SNR) optimum. Its implementation is discussed. An orthogonal complement system matrix is introduced which lays a foundation for this work