This paper proposes a novel adaptive algorithm for an echo canceler. In thi
s algorithm, the number of operations and memory capacity are equivalent to
those of the conventional NLMS algorithm but the convergence speed is twic
e that using the conventional algorithm. This adaptive algorithm is referre
d to as subband ES (exponentially weighted stepsize). In the algorithm, the
frequency bands of the received input signal and echo signal are divided i
nto multiple subbands, and echo is independently canceled in each subband.
Each adaptive filter in each subband has independent coefficients with an i
ndependent stepsize. The stepsize is time-independent and its weight is exp
onentially proportional to the change of the impulse response within the fr
equency region, such as the expected value of the difference between the wa
veforms of two impulse responses. As a result, the characteristic of the ac
oustic echo path in each frequency band is analyzed using the adaptive algo
rithm to improve the convergence characteristic. Using the results of compu
ter simulation and experimental results obtained via an experimental setup
with DSP, it is shown that the convergence speed with respect to input voic
e signal can be about 4 times faster when using echo cancellation based on
the new algorithm than in conventional full-band echo cancellation based on
the NLMS algorithm. (C) 1998 Scripta Technica, Electron Comm Jpn Pt 3, 82(
3): 49-57, 1999.