Reinforcement Learning for Dynamic Spectrum Management in WCDMA

Low use of licensed spectrum imposes a need for the advanced spectrum management for wise spectrum usage with the release of unneeded frequency bands for the secondary markets and opportunistic access. In this paper we present the possibilities to apply reinforcement learning in WCDMA to enable the autonomous decision on spectrum repartition among cells and release of frequency bands for possible secondary usage. The proposed solution increases spectrum efficiency while ensuring maximum outage probability constraints in WCDMA uplink. We give two possible approaches to implement reinforcement learning in this problem area and compare their behavior. Simulations demonstrate the capability of two methods to successfully achieve desired goals.