| 摘要 |
<p>A method for fast training of equalizers in a DMT system begins by normalizing the incoming receive signal (y) via steps (108-116). By normalizing the signal, the convergence rate of the training algorithm becomes relatively independent of channel line length so that long line lengths may converge to optimal equalizer coefficients in short time periods. The method also iteratively adjusts the filter coefficients w over time by using an adaptive gain vector mu that is updated on a component-by-component basis on each iteration via steps (114-146). By allowing each component of the vector mu to iteratively adapt independent of all other components in the vector mu based upon the binary sign bit of both real and imaginary components of frequency domain gradient vectors G, a convergence to optimal equalizer filter coefficients will occur in a short period of time. <IMAGE></p> |
