Recursive Bayesian controllers for non-linear acoustic echo cancellation and suppression systems

Both a cascade and a multichannel joint Bayesian estimator are provided for suppressing acoustic echo. An expansion basis (Power/Fourier series) is selected to convert a sample-based input signal xt into a DFT-domain multichannel signal [Xτ,1, . . . Xτ,p]. The posterior of unknown states (e.g., mean τ and covariance Pτ of the echo path Wτ and the mean âτ and covariance Qτ of the nonlinear coefficients aτ; or channel-wise mean τ,i and multichannel covariance Pτ of a compound quantity formed by merging together the echo path Wτ and the ith nonlinear coefficient aτ,i) and model parameters τ are estimated; and Kalman gain factor(s) Kτ are computed for optimal adaptation of the posterior of unknown states. An echo signal τ is estimated using the multichannel input signal [Xτ,1, . . . Xτ,p] and the adapted posterior; and an error signal Eτ is generated. Residual echo is suppressed by post-filtering the error signal Eτ with a weighting function ψτ which depends on the adapted posterior, and the filtered error signal ′t is then transmitted to a far-end.