Robust blind identification of SIMO channels: a support vector regression approach

A novel technique for blind identification of multichannel FIR systems is derived from the learning paradigm of support vector machines (SVMs). Specifically, blind identification is formulated as a support vector regression problem and an iterative procedure, which avoids a trivial solution, is proposed to solve it. The SVM-based approach can be viewed as a regularized version of the least squares method for blind identification. We show that minimizing the complexity of the solution, as suggested by the structural risk minimization (SRM) principle, increases the robustness of the proposed SVM-based technique to channel order overestimation as well as to poor diversity channels (i.e., when a pair of subchannels have close zeros). The performance of the method is demonstrated through some simulation examples.