| A new optimization solution is proposed for the design of time-domain equalizers (TEQ) for discrete multitone modulation (DMT) systems. It considers the mean square error, which measures the efficacy of the shortening of the channel, and the geometric signal-to-noise ratio, which indicates the performance associated with the effective bit rate through the combined channel-TEQ transmission system. This dissertation formulates a new optimization problem and outlines the iterative solution based on a design that iterates on the TEQ taps directly. The approach for the channel-shortening objective in this dissertation is to represent the DMT system as a least-squares-with-noise model and solving it using a combination of highly convergent inverse power method iteration and the conjugate gradient method. This dissertation then approaches the bit rate optimization problem by solving a linear quadratic dual problem in the time domain formulated by a target optimal solution to the geometric average in the frequency domain. |
