Acoustic echo canceller with high speech quality

Acoustic echo cancellers utilizing adaptive finite impulse response (FIR) filters conventionally incorporate learning identification (LI) also called the normalized least mean square (NLMS) adaptation algorithm. This echo cancellers have a serious problem in that colored signal input and incremental tap length degrade their echo cancellation characteristics. In this paper, a new adaptation algorithm combining reductive correlation and variable loop gain methods is proposed. The proposed algorithm is shown in the simulation to have good convergence characteristics compared to conventional methods. Furthermore, a sub-band echo canceller is proposed in which the input signal is separated into several sub-band signals in the frequency domain. Using this echo canceller, the computation quantity per sampling period in a processing module can be greatly reduced. Finally, the relationship between the overall echo cancellation performance and the characteristics of the sub-band filters is clarified.