Multi-agent learning for multi-channel wireless sensor networks

An increased bandwidth demand and the problem of interference have resulted in the advent of multi-channel protocols for Wireless Sensor Networks. In this paper, we propose a distributed contention-free multi-channel access scheme. This scheme is based on the parallel rendez-vous principle, which exploits the possibility of concurrent transmissions on different channels in the same collision domain. We describe a multi-agent learning algorithm that resolves all contention in a traffic adaptive manner. Moreover, the medium access resolution is combined with route selection in order to increase the number of parallel transmissions. The results of simulation experiments show that the proposed protocol can outperform McMAC, a state-of-the-art parallel rendez-vous protocol, in terms of throughput and latency.