This paper proposes respective policies for uplink power control and downlink power allocation in cell-free wireless networks. Both policies rely only on large-scale quantities and are expressed in closed form, being therefore scalable. The uplink policy, which generalizes the fractional power control employed extensively in cellular networks, features a single parameter; by adjusting this parameter, the SIR distribution experienced by the users can be compressed or expanded, trading average performance for fairness. The downlink policy dualizes the uplink solution, featuring two parameters that again allow effecting a tradeoff between average performance and fairness.

This paper proposes a power control policy for the uplink of cell-free wireless networks. Such policy, which generalizes the fractional power control used extensively in cellular networks, relies only on large-scale quantities, is fully distributed, and features a single control parameter. By adjusting this parameter, the SIR distribution experienced by the users can be compressed or expanded, effecting a tradeoff between average performance and fairness.

This paper applies a feedforward neural network trained in an unsupervised fashion to the problem of optimizing the transmit powers in centralized radio access networks operating on a cell-free basis. Both uplink and downlink are considered. Various objectives are entertained, some leading to convex formulations and some that do not. In all cases, the performance of the proposed procedure is very satisfactory and, in terms of computational cost, the scalability is manifestly superior to that of convex solvers. Moreover, the optimization relies on directly measurable channel gains, with no need for user location information.

This paper studies the viability of feedforward neural networks (NNs) for centralized power control in the uplink of cell-free wireless systems with matched-filter reception. The formulation relies only on large-scale channel behaviors as inputs, without the need for user location information, and on unsupervised learning, to avoid the onerous precomputation of training data that supervised learning would necessitate for every system or environment modification. Two different power control objectives are entertained, and for both of them the NN closely approximates the optimum solutions produced by convex solvers while vastly reducing the complexity, thereby opening the door to power control implementations for very large systems.

This letter proposes the unsupervised training of a feedforward neural network to solve parametric optimization problems involving large numbers of parameters. Such unsupervised training, which consists in repeatedly sampling parameter values and performing stochastic gradient descent, foregoes the taxing precomputation of labeled training data that supervised learning necessitates. As an example of application, we put this technique to use on a rather general constrained quadratic program. Follow-up letters subsequently apply it to more specialized wireless communication problems, some of them nonconvex in nature. In all cases, the performance of the proposed procedure is very satisfactory and, in terms of computational cost, its scalability with the problem dimensionality is superior to that of convex solvers.