LAUSR

This article considers radio-frequency (RF) energy harvesting devices that use an irregular slotted Aloha (IRSA) channel access protocol to transmit their data to a hybrid access point (HAP). Specifically, it addresses the fundamental problem of optimizing the number of packet replicas transmitted by each device in each time frame. Unlike prior works, it considers a learning approach to optimize the number of replicas according to the energy level of devices. This article first uses a model-based Markov decision process (MDP) to study the problem at hand. Then, it proposes a model-free, centralized, and a distributed $Q$ -learning-based solution that aims to maximize the number of successful transmissions in each time frame. Our simulation results show that our centralized and distributed solutions, respectively, achieve up to 38% and 29% more successful transmissions than conventional Aloha.