Blind Dereverberation Wideband Multi-Source 2-D DOA Estimation Method Based on Single Channel Nonnegative Matrix Factorization

Abstract

There are many algorithms for DOA estimation based on blind source separation (BSS), which assumes that each sensor perceives a linear instantaneous mixed narrowband signal with known number of sources in noise-free or low noise environments. However, in practical applications, the number of mixed signal sources is unknown, and the perceived signals are often acquired in strong noise and reverberant environments. This paper proposes a novel noise reduction single-channel nonnegative matrix factorization deconvolution (NRSNMFD) wideband multi-source 2-D DOA estimation algorithm. Firstly, LMS adaptive filtering is used to remove the noise of the observation signal. Secondly, the single channel observation signal is decomposed into multi-channel signals composed of multiple intrinsic mode functions (IMFs) by empirical mode decomposition (EMD). The number of signal sources is estimated and the determined signals are reconstructed. Thirdly, the dereverberation of the signals are realized by NMFD method. The GCC-PHAT method is used to estimate the delay of the signal source. Finally, signal source location is realized based on the delay. The simulation and actual test results show that the proposed algorithm can estimate 2-D DOA with high accuracy under noise and reverberation environment, providing a new method for engineering application.