Using QR Factorization in Subspace Based Blind Channel Identification

Abstract

Subspace factorization based methods are usually used to blindly identify a SIMO system. To divide signal subspace and noise subspace, correlation matrix of the receiving data should be estimated. A new Subspace factorization method is proposed in this paper. It uses a QR factorization of samples matrix instead of SVD of correlation matrix, so it can be viewed as a square-root version of the latter, which renders it more robust against ill-conditioned channels and naturally able to identify channels driven by color signal. Without the requirement of estimating statistics, this algorithm works with very short observation sequence. Compared to SVD, the QR factorization method requires fewer computations, so it is computationally more efficient.