Multichannel Relay assisted NOMA-ALOHA with Reinforcement Learning based Random Access

We investigate multichannel relay assisted non-orthogonal multiple access (NOMA) in slotted ALOHA systems, where each user randomly accesses one of different channel slots and different transmit power for uplink transmissions over two-hop links, to and from the relay. By using multi-agent reinforcement learning, we propose greedy and non-greedy random access methods so that each user can learn its best strategies of random access over multiple relay slots. Random collisions and fading over the relay slots are both considered. The behaviors of relay-aided NOMA-ALOHA strategies are evaluated with the simulation. It is shown that the greedy method outperforms the non-greedy method in terms of average success rate. For deployment of relay, the greedy method benefits in improving transmission reliability under the symmetric relay channels (between the two-hop links) compared to asymmetric channels. Thus, it is interpreted that the proposed greedy method is more promising to the NOMA-ALOHA systems under a symmetric multichannel relay.