An adaptive echo canceller using parallel Kalman filters

This paper presents a new method for an adaptive echo canceller in a long distance transmission channel. Since the number of parameters of an echo path is usually greater than 100, it is desired to estimate them as efficiently as possible for real-time processing. We propose new type filters, called parallel Kalman filters, based on the divided estimation of impulse response of an echo path. If the parallel Kalman filters are used, then the number of parameters to be estimated by one processor becomes much smaller than that by the usual Kalman filter. Since complete parallel computations are possible, this algorithm is suitable for an adaptive echo canceller operating in real time. It is very important that the convergency of the estimates to true parameters is guaranteed in the proposed echo canceller under some assumptions.