On the Performance of LMS-Based Algorithms for the Identification of Low-Rank Systems

A recent approach has been proposed that exploits the nearest Kronecker product (NKP) decomposition in conjunction with low-rank approximation methods, for efficiently solving low-rank system identification problems. Following this idea, several adaptive filtering solutions were developed. In this paper, we focus on least-mean-square (LMS) algorithms based on NKP, which own the advantage of low computational complexity. The performance features of these LMS-based algorithms are briefly investigated and a variable step-size (VSS) version is derived. The proposed VSS algorithm achieves a better compromise in terms of the main performance criteria.