Deep Learning Approach for Outage-Constrained Non-Orthogonal Random Access

Abstract

This letter presents deep neural network (DNN) approaches for non-orthogonal random access (NORA) systems where several devices are allowed to occupy the identical preamble. We desire to improve the reliability of the packet transmission of NORA devices with a careful management of multi-user interference. A novel transmit power control (TPC) mechanism is proposed which minimizes the maximum transmit power under constraints on link outage probabilities. The nonconvexity and unavailable outage formulations are addressed through DNNs. It is trained to yield feasible TPC solutions for outage constraints based on timing advance values. The viability of the proposed DNN approach is demonstrated with system-level simulations.