Deep Reinforcement Learning for Spectrum Sharing in Future Mobile Communication System

In recent years, the rapid growth of mobile communication services makes spectrum resources become increasingly scarce. This paper considers the multi-dimensional resource allocation problem in unlicensed spectrum communication system. A training method based on deep reinforcement learning is proposed to generate a spectrum sharing and power control strategy for secondary users in the communication system. Deep Q-Network and Deep Recurrent Q-Network are chosen as the structure of neural network. Experiments are conducted to investigate the effectiveness of the algorithm. The results show that collision rate decreases in training while average reward rises.