Precision-efficient use of block adaptive algorithms in data-driven echo cancellers

Abstract

Data-driven echo cancellers are used in full-duplex voiceband data transmission at rates of 4800 b/s and above, and in digital subscriber loops for ISDN (integrated services digital network) at rates of 160 kb/s (or above). It is well-known that the required 60 dB of cancellation mandates high coefficient and processor precision if the stochastic-gradient (least-mean-squares (LMS)) adaptive algorithm for the echo canceller is implemented in fixed-point arithmetic. However, the use of block LMS or block adaptive algorithms significantly improves this situation without increasing computation. The improvement in numerical accuracy is traded (almost linearly) for a decrease in tracking speed of the echo canceller, which, however, is often an acceptable tradeoff as the echo channel does not exhibit significant time variation. A drawback of the block adaptive algorithms, their requirement of more temporary internal variables, is also discussed. >