Development of high quality acoustic subband echo canceller using dual-filter structure and fast recursive least squares algorithm

A high quality acoustic subband echo canceller is developed based on a dual-filter structure and the fast recursive least squares (FRLS) algorithm. Methods for overcoming the instability problem of the FRLS algorithm and implementing it using the 32-bit fixed-point arithmetic are presented. A new tap-weight transfer method, which assures double talk detection, is proposed. Computer simulations demonstrate that fast convergence and fast tracking are achieved in an acoustic subband echo canceller using 32 subbands and a total of 4000-tap FIR filter for a 250 ms echo path. The combined use of the FRLS algorithm in the lowest eight frequency bands and the normalized least mean square (NLMS) algorithm in the rest of the frequency bands increases convergence rate for speech signals by five to ten times, and the echo return loss enhancement (ERLE) by more than 10 dB over those of using only the NLMS algorithm.