Correction algorithms for extended range echo cancellers

Adaptive finite impulse response filters (AFIRF) for echo cancellation in 4- wire telephone lines and/or acoustic feedback in teleconference setups require 250 to 1000 coefficients and rapid convergence with highly correlated speech signals. This paper discusses several theoretical and design problems and possible solutions. The signals entering the correction processor are decorrelated by two pseudo-random binary sequences. A multiplicative algorithm is used to accelerate the adaptive process, and adaptive loop stability is enhanced by averaging the error signal over several samples. The computational speed of the digital convolution processor is increased by using a pseudo-logarithmic format.