Towards a principled approach to designing distributed MAC protocols

Emerging applications such as cyber-physical systems and the electric smart grid have placed a greater demand for distributed protocols that can scale and adapt without centralized control, while providing throughput and delay guarantees. This paper proposes a formal approach to designing a distributed scheduling protocol. Similar to previous queue-length based CSMA-type schedulers, the protocol proposed in this paper is a CSMA-type scheduler that is throughput-optimal, generates collision-free schedules, is fully distributed, and adapts to traffic demands without explicitly knowing the arrival rates. However, our design takes a more formal approach, where the multi-information (a natural extension of mutual information) between links is minimized, rather than maximizing the entropy, as was done by previous CSMA-type schedulers. Zero multi-information corresponds to a fully distributed scheduler. The formal approach in this paper also admits side information (e.g., a universal clock from GPS) into the framework. Simulations show our proposed protocol has better delay performance compared with previous CSMA-type schedulers.