Multi-agent reinfocement learning for stochastic power management in cognitive radio network

Frequency spectra are nowadays getting overcrowded because of increasing cell phone users. Cognitive radio network offers an alternative modality to utilize unused spectra efficiently among unlicensed users. This paper attempts to allocate transmission power among cognitive users in an efficient way without creating interference to the licensed users. We here adopt multi-agent reinforcement learning for cooperative power allocation in cognitive radio network. Multi-agent learning is here used to handle stochastic behavior of the environment. We use three mixed strategies (Correlated equilibrium) to control transmission power in multi-agent learning. After the learning algorithm converges, we obtain the optimum power level under different situations for subsequent use in power utilization during communication. Experimental results indicate that the proposed algorithm outperforms its classical counterparts by a significant margin.