Band Preference Design Algorithms for Improved Iterative Water-Filling

Band preference (BP) attempts to improve the performance of iterative water-filling (IWF) when applied to the multiuser interference channel. With important applications relating to digital subscriber line (DSL) access networks such as those based on current-day ADSL2 and VDSL standards, we develop novel techniques for determining suitable band levels that lead to significant performance gains in those circumstances where IWF would otherwise perform far from the optimum. Our techniques are based on a novel optimization framework that avoids inherent difficulties in analysis that have, until now, obscured an elegant design path. Our developments are easily customized to any clustering band-pattern, particularizing to the spectrum balancing result in the special-case of unit band-sizes. A practically palatable implementation is induced by a two-level primal decomposition whereby a master problem is responsible for band management, based on the result from independent slaves each undertaking a relaxed water-filling (WF) operation. Numerical results show that 2-3 bands are sufficient for a downstream near-far scenario comprising two user-groups.