Optimal transmit for packet-based data transmission on dispersive channels with application to the FIR MMSE-DFE

Optimal transmit filters for packet-based data transmission on dispersive Gaussian-noise linear time-invariant channels are derived by maximizing the channel throughput, subject to a fixed input energy budget. A quasi-stationary approximation to the optimal nonstationary input covariance process is derived and shown to exhibit negligible throughput loss compared with the optimal case, for situations of most practical interest. This accurate approximation results in efficiently computed lattice or pole-zero transmit filters. By considering the finite-impulse-response minimum-mean-square error decision feedback equalizer (FIR MMSE-DFE) as a receiver structure, it is shown that transmitter optimization results in an appreciable improvement in the decision point SNR when the output block length (N) is comparable with the channel memory. As N becomes infinite, the optimum finite-dimensional nonstationary input covariance process converges to a stationary process whose power spectrum obeys the water-pour distribution.<>