Inter-Channel Decorrelation by Resampling via Time-Domain Interpolation Filters Derived from the Time-Shifting Property

ABSTRACTWe have shown previously that decorrelation by resampling (DBR)effectively alleviates the non-uniqueness problem caused by highlycorrelated reference signals during multi-channel acoustic echocancellation. We present in this paper a novel DBR procedure forinter-channel decorrelation by using the time-domain interpola-tion filters derived from the time-shifting property that appears infrequency-domain resampling (FDR) through the discrete Fouriertransform. By pre-computing a set of interpolation filters, DBR canbe performed efficiently in real time in the time domain. Moreover,the block-processing artifacts can be completely eliminated throughcenter filtering by the proposed method. Simulation results showthat the new approach has higher resampling accuracy compared tothe original FDR scheme. Index Terms — Resampling, interpolation filter, inter-channeldecorrelation, time shifting, multi-channel acoustic echo cancella-tion1. INTRODUCTIONWe proposed the decorrelation by resampling (DBR) technique in[1] that introduces time-varying delay across channels to effectivelyalleviate the non-uniqueness problem caused by highly correlatedreference signals during multi-channel acoustic echo cancellation(MCAEC) based on the least mean square (LMS) algorithm. As ex-tensionsofDBR,weproposedfrequency-domainresampling(FDR)[2] and sub-band resampling (SBR) [3] for computation and im-plementation flexibility. When applied along with our noise-robustfrequency-domain MCAEC system, DBR enables a faster echo pathtracking performance over other decorrelation procedures while in-troducing minimal distortion to signal quality and statistics [1,2].However,toimprovetheresamplingaccuracybyFDRorSBR,arelatively large zero-extension factor is required. The resulting com-putational complexity with the discrete Fourier transform (DFT) canbe rather high due to the large signal block size after zero padding.Although there exists a way to reduce the computational cost whenthe zero-padding technique is used with the fast Fourier transform(FFT) [4], the modified FFT algorithm may not be implementablefor all real-time applications.We present in this paper a novel approach to DBR via resam-pling by interpolation (RBI) by using the time-domain interpolationfilters determined from linking the time-shifting operation on sam-ples after resampling to the DFT utilized by FDR. RBI achieveshigher resampling accuracy compared to FDR. The disadvantage,however, is that the amount of computation required to calculate thefilterscanbemuchmorethansimplyapplyingFDR.Theresamplingratio

[1]  Ted S. Wada,et al.  Decorrelation by resampling in frequency domain for multi-channel acoustic echo cancellation based on residual echo enhancement , 2011, 2011 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics (WASPAA).

[2]  Ted S. Wada,et al.  Inter-channel decorrelation by sub-band resampling in frequency domain , 2012, 2012 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).

[3]  A. Loan,et al.  On Cooley-Tukey FFT method for zero padded signals , 2005, Proceedings of the IEEE Symposium on Emerging Technologies, 2005..