基于深度强化学习的水声网络公平跨层MAC协议OA北大核心CSTPCD

In view of the unfair channel allocation and energy-constrained nodes in underwater acoustic communication heterogeneous networks,a proportional-fair optimized cross-layer medium access control(POCL-MAC)protocol is proposed based on deep reinforcement learning(DRL).It strives to optimize fair channel access and power control by cross-layer joint.By feedback ACK packets,the channel conflict outcomes and receiver-side signal-to-noise ratio(SNR)under delay status are obtained.Autonomous learning is carried out based on the state,action and reward sequence of DRL to adjust the access slot and transmission power of cognitive users.Fairness function is employed to achieve proportional fairness between cognitive users and primary users' throughput performance in heterogeneous networks.A joint state sequence and independent reward function are designed to enhance the accuracy of sub-action decisions in cross-layer joint optimization without increasing neural network complexity.Simulation results demonstrate that the proposed algorithm can achieve near-optimal fairness throughput performance while exhibiting better energy utilization efficiency in comparison with the traditional DRL-based algorithms.