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

摘要

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 x_t into a DFT-domain multichannel signal [X_τ,1, . . . X_τ,p]. The posterior of unknown states (e.g., mean Ŵ_τ and covariance P_τ of the echo path W_τ and the mean â_τ and covariance Q_τ of the nonlinear coefficients a_τ; or channel-wise mean Ŵ_τ,i and multichannel covariance P_τ of a compound quantity formed by merging together the echo path W_τ and the i^th nonlinear coefficient a_τ,i) and model parameters θ_τ are estimated; and Kalman gain factor(s) K_τ are computed for optimal adaptation of the posterior of unknown states. An echo signal Ŷ_τ is estimated using the multichannel input signal [X_τ,1, . . . X_τ,p] and the adapted posterior; and an error signal E_τ is generated. Residual echo is suppressed by post-filtering the error signal E_τ with a weighting function ψ_τ which depends on the adapted posterior, and the filtered error signal ŝ′_t is then transmitted to a far-end.