A learning algorithm for adaptive microphone arrays based on mathematical programming methods

This paper proposes a new learning algorithm for adaptive microphone arrays consisting of multi-channel adaptive linear filters. The algorithm is based on the interior point method, which is a very powerful mathematical programming method. In this algorithm, the learning problem of multi-channel adaptive linear filters is formulated as a quadratic programming (QP) problem with some constraints. The optimal filter coefficients are updated using the analytic center for a feasible region of this QP, and the feasible region is shrunk by a factor dependent on the error energy of learning. Using this strategy, a numerical stability of learning can be ensured. It is shown experimentally that this numerical stability can be always maintained, and that a high quality of target extraction can be achieved