Normalized Multichannel Frequency-Domain LMS Filter With Nearest Kronecker Product Decomposition for Blind Identification of Low-Rank Acoustic Systems

This paper proposes a multichannel frequency-domain adaptive filtering algorithm to blindly identify low-rank acoustic systems. The model filters of the multichannel acoustic impulse responses are decomposed into two sets of short sub-filters through the nearest Kronecker product (NKP). An extended multichannel frequency-domain signal model and its associated cost function are established by using these short sub-filters. The normalized multichannel frequency-domain least-mean-square (NMCFLMS) algorithm based on NKP is subsequently derived according to the Newton's iteration criterion. Simulations show that the proposed algorithm is computationally more efficient and has a better convergence behavior for blindly identifying multichannel acoustic systems than the conventional NMCFLMS adaptive algorithm, regardless of whether the excitation is a white sequence or a speech signal.