Subband echo canceler structure

A Subband echo canceler structure is proposed which enables the use of large decimation factors with a relatively short transmission delay, while keeping the near-end signal undistorted. With the subband echo canceler structure we propose a time varying step size LMS algorithm, which reduces distortion during double-talk without increasing the computational cost nor decreasing the convergence rate of the conventional normalized LMS algorithm. Simulation results using white noise and actual speech signals show that by using relatively large decimation factors, in addition to reducing the computational load, convergence rates becomes almost independent of the statistics of the input signals. Computer simulations also show that proposed adaptation algorithm greatly reduces distortion during double talk periods.