Fast frequency-domain acoustic channel estimation with interference cancellation

We consider a training based acoustic channel estimation technique for the purpose of audio dereverberation. A transmitting device that generates a low power (for minimal disturbance) controlled periodic pseudo-random binary probing signal is placed near the sound source, and the convolved probing and desired signals are collected. By a coherent integration of multiple periods of the received signal, a SINR enhanced training signal is obtained for channel estimation. Due to the periodic nature of the training signal, the channel can be efficiently estimated in the frequency-domain without any edge effects. The frequency-domain estimation dramatically reduces the computational load from matrix multiplications and inversions in the time-domain to that of scalar multiplications. Computer simulations and experimental results are presented to demonstrate the usefulness of this algorithm.