Adaptive IIR filtering and system identification via rational subspace methods

A recently proposed adaptive orthogonal subspace filter is specialized to the two-channel case, resulting in a subspace system identification algorithm. In addition to the standard benefits of orthogonal-based adaptive IIR filters-inherent stability of the filter structure even with time variation in the parameters, and robust performance in finite precision environments-the proposed method allows the identification of noncausal components using a causal filtering method. Moreover, the estimation algorithm is unbiased with respect to white noise in both the input and output sequences of the system to be identified, unlike previous approaches which are unbiased with respect to output noise alone. An a posteriori based algorithm is derived, which enjoys superior convergence properties over a previous a priori based algorithm