Variationally diagonalized multichannel state-space frequency-domain adaptive filtering for acoustic echo cancellation

In this contribution, we present a novel low-complexity state-space algorithm for multichannel acoustic echo cancellation. The reduction in complexity is brought about by means of top-down imposition of mutual independence on the respective acoustic echo paths within a variational Bayesian framework. This results in a fully diagonalized multichannel echo-path state estimator with a complexity that varies linearly with the channel order. The state estimator is augmented with learning rules for the model parameters that are optimal in the maximum-likelihood sense. We substantiate the efficacy of our formulation by means of simulation results in the presence of changes in the echo paths and continuous double-talk.