Distributed optimization of coupled systems with applications to network utility maximization

In Network Utility Maximization (NUM) problems, it is generally assumed that user utilities are uncoupled, i.e., each utility depends only on local variables. Then the coupling in constraint functions among users sharing common resources can be decoupled by standard methods such as dual decomposition. However, in problems where cooperation or competition is modeled through the objective function, such as rate allocation in clustered system and power control in interference limited system, each utility may depend not only on its local variables but also on the local variables of other utilities. Applications of this coupled utility model include wireless power control and DSL spectrum management, where the utilities are functions of the Signal-to-Interference Ratios (SIR) that depend on the transmit powers of other users. We present a systematic approach of consistency pricing to decouple NUM problems with coupled utilities, obtaining distributed algorithms that efficiently handle couplings in utilities with two alternative timescales, as well as a method to reduce message passing overhead in the case of interference-based coupling.