Multi-agent Competitive Spectrum Handoff Based on Improved MADDPG Algorithm

Abstract

Spectrum handoff plays an important role in realizing dynamic spectrum management. When there are multiple competing secondary users (SUs) in a cognitive radio network, the dynamics of the environment will increase dramatically, and efficient spectrum handoffwill become difficult. In this paper, the competition of multiple SUs in spectrum handoff is transformed into competition between SU teams, and the Multi-agent deep deterministic policy gradient (MADDPG) method with auxiliary network is proposed to deal with the dynamics of the environment. By predicting the opponent’s actions, the SU can improve the accuracy of the handoff. In simulation experiments, the proposed method is proved to bring SU with higher Quality of Experience (QoE).