Time Delay Estimation Based on Mutual Information Estimation

A novel time delay estimation (TDE) method is proposed based on the concept of mutual information (MI). The relative delay between two sensor output signals is estimated by maximizing the estimated MI of the two signals. In estimating the MI, it does not require any information about prior distribution of the source signal. Instead, the Parzen window estimator is used to estimate the density function of the source signal from two sensor output signals. Moreover, based on the Parzen window estimator, the quadratic mutual information (QMI) is incorporated to effectively and efficiently estimate the MI. A set of TDE experiments have been conducted for speech signals to evaluate the performance of the proposed approach for different data block size and reverberation times compared to the generalized cross-correlation approach.