Multichannel time delay estimation for acoustic source localization via robust adaptive blind system identification

In the problem of acoustic source localization, time difference of arrival (TDOA) among multiple sensors is needed, which is often obtained through time delay estimation (TDE) techniques. Among the multiple TDE methods developed in the literature, the normalized multichannel frequency-domain least-mean-square (NMCFLMS) algorithm is shown robust to reverberation. The performance of this algorithm, however, deteriorates in non-Gaussian and low signal-tonoise ratio (SNR) Gaussian noise environments. In this paper, we re-derive a robust normalized multichannel frequency-domain leastmean- M-estimate (RNMCFLMM) algorithm to estimate TDOAs for acoustic source localization. The proposed algorithm exploits the non-sensitivity of an M-estimator to non-Gaussian noise and makes a tradeoff between the least-squares and least-absolute criteria to improve the robustness of TDE with respect to non-Gaussian and Gaussian noises. The effectiveness of the proposed algorithm is demonstrated in real acoustic environments.