Adaptive Multichannel Time Delay Estimation Based on Blind System Identification for Acoustic Source Localization

Abstract

Time delay estimation is a difficult problem in a reverberant environment and the traditional generalized cross-correlation (GCC) methods do not perform well. Recently, a blind channel identification-based adaptive approach called the eigenvalue decomposition (ED) algorithm has been proposed to deal with room reverberation more effectively. The ED algorithm focuses on a system with two channels whose impulse responses can be blindly identified only if they do not share common zeros. The assumption often breaks down for acoustic channels whose impulse responses are long. In this chapter, the blind channel identification-based time delay estimation approach is generalized to multiple (more than 2) channel systems and a normalized multichannel frequency-domain LMS algorithm is proposed. The proposed algorithm is more practical since it is less likely for all channels to share a common zero when more channels are available. It is shown by using the data recorded in the Varechoic chamber at Bell Labs that the proposed method performs better than the ED and GCC algorithms.