Acoustic Echo Cancellation Using Subband Adaptive Filtering

Adaptive ltering has been successfully used in canceling line echos in the telephone network since the 1960's [179, 443, 387]. In these cases, the adaptive lter is used to model the unknown echo path in a communications channel and then cancel the echo signals from the conversation. An adaptive lter is required because the echo path is not only unknown, but also because it is di erent for each network path. However, the acoustic echo cancellation problem is not as easily solved with conventional adaptive ltering systems due to the requirements for a much longer adaptive impulse response and for faster convergence since the acoustic echo path is time-varying during the communication transmission. In this chapter, we summarize key results from adaptive lter theory, and discuss the least mean square (LMS) and normalized least mean square (NLMS) algorithms used to adjust an adaptive lter. We then examine adaptive ltering in subbands as one method for acoustic echo cancellation. Simulations (using a measured room response with speech signals) demonstrate that the oversampled subband adaptive system provides improved performance for acoustic echo cancellation applications.