Spectrum markets for service provider spectrum trading with reinforcement learning

We present an auction-based spectrum market approach to service provider spectrum trading. Service providers buy and sell spectrum amongst one another in a spectrum market and simultaneously compete for customers from a common pool. Multi-agent reinforcement learning solutions are applied in both customer nodes and service providers to dynamically manage participation in the market. We outline four possible regulatory scenarios with varying degrees of flexibility and competition. Simulations demonstrate that the allocation of spectrum is efficient and fair. Customers and service providers of varying size are shown to benefit from this approach while the system spectrum efficiency is also significantly improved.