A fully distributed learning algorithm for power allocation in heterogeneous networks

In this work, we present a fully distributed Learning algorithm for power allocation in HetNets, referred to as the FLAPH, that reaches the global optimum given by the total social welfare. Using a mix of macro and femto base stations, we discuss opportunities to maximize users global throughput. We prove the convergence of the algorithm and compare its performance with the well-established Gibbs and Max-logit algorithms which ensure convergence to the global optimum. Algorithms are compared in terms of computational complexity, memory space, and time convergence.