A full-duplex fast training algorithm for simultaneously estimating echo and channel response

The authors present a new scheme that extends the correlation based fast training of echo cancellers introduced by Long and Ling (1990) to allow for ultra-high-speed initialization of a full-duplex data transmission system over two-wire lines. A fast training procedure is derived to coestimate the near echo and the far echo with bulk delay, and the channel responses. The effects of the noise and symbol rates' offset between two ends are examined in terms of both mean square error and SNR. Both the theoretical analysis and computer simulations show that there are three degrading factors due to symbol rates' offset. The new algorithm delivers fst convergence speed with the same computational requirements as those of stochastic-gradient algorithms and reduces the tap-setting time to half of that required by the traditional half-duplex fast training scheme.<
>