The analysis and real-time implementation of full-band and sub-band acoustic echo cancellers

Includes bibliographical references (pages [76]-77).In teleconferencing systems, the most widely used technique to cancel acoustic echo makes use of an adaptive transversal filter and the least mean square algorithm. This research analyzes the performance of a finite length adaptive transversal filter using different input signals for both full-band and sub-band adaptive filtering techniques. The analysis shows that for a sinusoid input, the mean-square error reduces to zero if the filter order is greater than 2. This result can be extended for a multiple sinusoid signal, in which case the mean-square error reduces to zero if the filter order is greater than 2K, where K is the number of sinusoidal components. Also, for the same filter order, the speech signal showes higher echo return loss enhancement than that for white noise signal, because the voiced section of speech waveform can be represented as a combination of multiple sinusoids. The residual echo for sinusoid inputs is independent of the length of the echo-path impulse response; however, for white noise input, the residual echo is proportional to the length of the uncovered portion of the echo-path impulse response. The sub-band technique is explored and the analysis of the effect of finite filter length on the performance of the adaptive transversal filter is extended to the sub-band case. The analysis results are verified by computer simulation using both 8-channel and 16-channel sub-band structures. Real-time experiments with full-band and 8-channel sub-band acoustic echo cancellers are performed using speech signals. The results verify that in order to achieve the same performance, the order of the sub-band adaptive filter can be chosen much smaller than that of the full-band acoustic echo canceller.M.S. (Master of Science