On Improving the BER Performance of Rate-Adaptive Block Transceivers, with Applications to DMT

We propose two strategies for improving the (un-coded) bit error rate (BER) performance of practical rate-adaptive block-by-block communication schemes, such as discrete multitone modulation (DMT). Our strategies are inspired by some recent work which showed that for uniformly bit-loaded schemes, the transmission strategy which minimizes the BER for a linear receiver involves allocating power to the subchannels that are implicit in the block-by-block framework in a minimum mean square error (MMSE) fashion and linearly combining these subchannels using a normalized discrete Fourier transform (DFT) matrix. This combining equalizes the decision point signal-to-noise ratios (SNRs) of the subchannels. Given a non-uniformly bit-loaded scheme, our first design strategy simply performs a DFT-based linear combination within the groups of subchannels which share the same constellation. Our second strategy provides further reduction in the BER by re-allocating power within these groups in a MMSE fashion prior to DFT combining. Our examples indicate that our design strategies can provide significant reductions in the BER, and give rise to substantial SNR gains (of the order of several decibels).