Considering a low power wide area random access system (like LoRaWAN) where the individual transmitters are mobile, we 1) propose a way of quantifying the performance of tracking of the mobile devices, and 2) design a distributed algorithm to achieve a target tracking performance. The insights gained are then used to provide an analysis of a family of target-achieving reinforcement-learning algorithms used in the literature to learn the optimal (Nash Equilibrium) random access probabilities. By mapping the payoff function in the equivalent game to the performance metric, we establish that, under some general conditions of inter-node parameter separability, the algorithm convergence is independent of the payoff function used. The mapping from the desired performance metric to the success probability in random access can be used in the algorithm to achieve a target success probability.
Read full abstract