Linear-combined multichannel adaptive digital filters

The fundamental structure of linear‐combined multichannel adaptive digital filters which can process a multiple‐input single‐output time series, and their tapcoefficient correction algorithms with fast convergence are described. These are fundamental circuits required for multichannel echo‐cancellers, and echo‐cancellers with noise‐cancelling facilities. The structure of linear‐combined multichannel adaptive digital filters using the learning identification (LI) was introduced, and their characteristics were analyzed. Adaptive algorithms using the per‐channel normalization of power were proposed to improve the convergence speed. With a minor increase in the amount of computations, this method is a significant improvement of the convergence speed which, inherently, is degraded by the differences of average powers among channels in a linear‐combined multichannel adaptive filter. For further improvement in the convergence speed, another tap‐coefficient correction equation (the projection method) also was introduced. In the tap‐coefficient equation using the projection algorithms, the convergence speed can also be improved significantly by applying the per‐channel power normalization.