Using Reinforcement Learning in Slotted Aloha for Ad-Hoc Networks

Abstract

Slotted ALOHA is known to have poor channel utilization (a maximum of 37% when average offered load is one packet per time slot). Reinforcement learning has recently been proposed as a technique that allows nodes to learn to coordinate their transmissions in order to attain much higher network utilization. All reinforcementlearning schemes proposed to date assume immediate feedback on the outcome of a packet transmission. We introduce ALOHA-dQT, a reinforcement-learning protocol that achieves high utilization by having nodes broadcast short summaries of the channel history as known to them along with their packets. Our simulation results show that ALOHA-dQT leads to network utilization above 75%, with fair bandwidth allocation among nodes. ALOHA-dQT is the first reinforcement-learning approach applied to slotted ALOHA suitable for ad-hoc networks without centralized repeaters. CCS CONCEPTS• Networks → Network protocols; Network performance modeling; • Theory of computation → Reinforcement learning.

Survey of Reinforcement-Learning-Based MAC Protocols for Wireless Ad Hoc Networks with a MAC Reference Model

“…Many RL solutions proposed so far assume real-time feedback of the results of packet transmission. In [83], a novel RL-based MAC protocol ALOHA-dQT was proposed, which improved the channel utilization by letting nodes periodically broadcast short summaries of their known channel history. The channel history for the last N slot states was stored by each node, and it iteratively merged its information with the channel history based on the node broadcast.…”