Symmetric minimum noise subspace for multi-input multi-output system identification

This contribution deals with a particular family of blind system identification methods, referred to as Minimum Noise Subspace (MNS). MNS method is a computationally fast version of subspace method. Here, we develop a symmetric version of MNS method (referred to as SMNS) that has the advantage of better robustness and estimation accuracy at the cost of a slight increase of the computational cost in comparison with original MNS. In the same time, we present and compare different algorithms for the block implementation of the SMNS. Finally, we show that under certain additional assumptions on the channel transfer functions, we can go beyond the minimum in the sense that less than ( being the number of outputs and the number of inputs) noise vectors are sufficient for unique identification of channel parameters.