A subband approach to adaptive acoustic echo cancellation

Abstract

In this paper a novel subband-based acoustic echo canceller (AEC) is proposed for teleconferencing applications. The proposed AEC consists of K non-overlapping main subbands and K auxiliary subbands. The auxiliary subbands are introduced in order to provide frequency coverage in the non-overlapping region between the neighboring main subbands. The main and auxiliary bands are constructed using the discrete Fourier transform (DFT) filter bank approach. Owing to the complex modulation feature implied in the DFT filter bank, the proposed structure consists of ( K 2 + 1) distinct main subbands and ( K 2 + 1) distinct auxiliary subbands. Central to computational efficiency of the proposed AEC is the down-sampling of the auxiliary bands by a factor of two compared to the main bands. The normalized mean squares algorithm (NLMS) is employed for updation of subband adaptive filter coefficients. The proposed subband AEC is implemented both in a real-time teleconferencing system employing a floating-point DSP chip and in simulation for the case of K = 16 i.e. nine distinct main bands and nine distinct auxiliary bands.