Approaching Fair Collision-Free Channel Access with Slotted ALOHA Using Collaborative Policy-Based Reinforcement Learning

A novel expert-based approach to reinforcement learning is applied to slotted ALOHA in order to approach fair collision-free transmissions. Active nodes use known periodic transmission schedules as policies and assign weights to them based on the Quantitative Tree (QT) algorithm introduced in this paper. Nodes learn to transmit following the policies with the highest weights to minimize packet collisions. This results in two variants of slotted ALOHA, which we call ALOHA-QT and ALOHA-QTF, that converge to transmission schedules that are almost free of collisions within a short period of time, and that attain near perfect transmission throughput even when node additions and departures occur frequently. In addition, ALOHA-QTF attains very fair bandwidth distribution among nodes, where the nodes with bottom ten percent of bandwidth still fare reasonably well. ALOHA-QTF is shown to be better than slotted ALOHA with exponential back off and framed slotted ALOHA with Q learning (ALOHA-Q) in terms of throughput and fairness.