A min-sum approach for resource allocation in communication systems

This paper considers distributed protocol design for resource allocation (RA) problems. We propose a fully decentralized RA scheme based on the min-sum message passing (MP) approach in which each message is the solution of small distributed allocation problems. Due to the presence of cycles in the network graph, the MP routine may not converge to a fixed point. To this end, we introduce a reweighted MP (ReMP) algorithm that perturbs the ordinary min-sum algorithm by suitably re-weighting messages. ReMP distributes the computational effort of achieving an optimal RA among nodes. Such feature makes ReMP particularly attractive in wireless networks allowing the convergence to a fixed and provably optimum point without employing any central controller. Numerical results show that ReMP outperforms conventional MP-based algorithms for RA problems in terms of computation time.