Title :
A multi-processor structure for signal processing application to acoustic echo cancellation
Author :
Degryse, D. ; Druilhe, F. ; Gilloire, A.
Author_Institution :
OROS Grenoble, Fr.
Abstract :
Existing echo cancellers are inadequate for use in the quality improvement of audioconference and videoconference services. To be useful for this purpose, a bandpass of up to seven kHz is required, which would multiply by four the volume of computations in the case of the classical LMS algorithm ; larger word sizes are also required to handle data coded with more than twelve bits per sample and to get better estimates of the echo path. For the purpose of testing LMS-type echo cancellation algorithms in real time, a modular structure for signal processing has been designed, with its related software. This structure is built around a synchronous bus allowing 5 millions 24 bits transfers per second, the bus is based on a multi-master concept and includes hard-wired synchronization primitives. This structure allows the implementation in real time of a LMS adaptive filter 3000 coefficients and a 16 kHz sampling rate, which corresponds to almost 100 millions multiplications and the same number of additions per second.
Keywords :
Acoustic applications; Acoustic signal processing; Adaptive signal processing; Echo cancellers; Least squares approximation; Signal processing algorithms; Software algorithms; Software testing; Video signal processing; Videoconference;
Conference_Titel :
Acoustics, Speech, and Signal Processing, IEEE International Conference on ICASSP '85.
DOI :
10.1109/ICASSP.1985.1168202