Multirate acoustic echo cancellation: Which adaptive filters for which subbands?

Multirate signal processing is one of the best ways to minimise the computational complexity of Acoustic Echo Cancellation (AEC). The subband decomposition of the excitation signal (speech) results in different statistical properties in each subband. As there are a wide range of Adaptive Filters (AFs) with different performances and costs, a challenge is posed in how to match AFs to subbands so as to maximise performance while minimising computational cost. This paper discusses a hypothetical automated benchmarking methodology that addresses some of the issues raised by the attempt to optimise multirate AEC architectures as a function of desired performance. In this context, a performance comparison of the NLMS and RLS algorithms is conducted.