A robust echo canceler for acoustic environments

This paper presents a novel acoustic echo canceler for teleconference and speakerphone systems based upon input orthogonalization and spread spectrum communication techniques. The new echo canceler represents a significant departure from traditional methods of removing acoustical echo and solves the open problem of echo cancellation during double-talk transmission. A decorrelation based adaptive filtering algorithm is introduced and its performance studied. A contraction mapping principle is applied to obtain a tight bound on the step size for the convergence of the adaptive algorithm. In conjunction, a direct sequence spread spectrum correlation technique is used to estimate the room impulse response during double-talk. Both components of the echo canceler rely on an innocuous wide-band training signal which is transmitted into the room along with the speech. The new echo canceler was implemented on a TMS320C30 DSP processor and experimental results are presented. The echo canceler was found to give excellent performance for all operating conditions, including during double-talk transmission.