A novel approach to fast initializing digital subscriber loop echo cancelers

The authors propose a novel approach to fast initializing DSL (digital subscriber line) echo cancelers using full-duplex transmitted training sequences. The method used in formulating the fast estimation algorithm is similar to those of F. Ling and G. Long (1990), but the basic concept is quite different from theirs. The important difference between the proposed method and the previous ones is that with the proposed method one simultaneously estimates the impulse responses of the echo paths of both ends, rather than separately. Therefore, the initialization time is greatly reduced. The authors have achieved a factor-of-two improvement in convergence speed over that of the traditional half-duplex fast training schemes with the same computational requirements as those of stochastic-gradient algorithms. The effects of the noise and symbol rates´ offset between two ends are examined in terms of signal to noise ratio, and the results show that the new algorithm is robust in the presence of noise and symbol rates´ offset