Research On Communication Resource Allocation Mechanism Of D2D-NOMA

The continuous and rapid development of wireless communication technology,as well as the continuous emergence of various new communication business scenarios,is prompting the comprehensive interconnection of Man-Machine-ObjectSpace.Accompanied by continuous growth of the number of network terminals,the contradiction between the scarcity of wireless spectrum resources and the spectrum utilization is becoming progressively apparent,which aggravates the burden of the whole network.Next-generation wireless communication strategic technologies,such as D2 D communication and NOMA technology,can effectively alleviate wireless spectrum resource pressure.The collaboration of the two techniques will increase the system capacity and carrying capacity,and improve the spectrum utilization and system throughput,which has very important application value and research significance.The interference problem is one of the key issues common to both D2 D communication and NOMA technology,and interference will be effectively suppressed through efficient resource allocation strategies.Based on this,this thesis is intended to investigate the resource allocation technique under a D2D-NOMA communication network scenario.In the ”one-to-two” D2D-NOMA communication cluster scenario formed by the combination of D2 D communication and NOMA technology,in this thesis,two algorithms for resource allocation based on the deep reinforcement learning framework are proposed for the two links of resource allocation.Aiming at the channel selection and power control problems of the D2D-NOMA communication cluster,a resource allocation algorithm based on the combination of a Gate Recurrent Unit(GRU)neural network and Double Deep Q Network(DDQN)is proposed to convert the subchannel allocation and power control into a reinforcement learning problem.Viewing the network as an optimal scheme is a multi-agent system consisting of a single intelligent optimal resource allocation policy,taking advantage of the gate structure of the GRU network to control inputs and outputs,determining the degree of retention of input data,helping D2 D users to predict the actions of other D2 D users with historical information,and then entering the data output from the GRU layer into DDQN.Taking advantage of the scalability of DDQN,the channel choice and power control actions of the D2 D transmitter are set as the union of resource blocks and discrete power ranks to update the resource allocation strategy to maximize throughput.Simulation experiments show that the GRU-DDQN algorithm outperforms other schemes such as DDQN,DQN,and random access in terms of throughput performance in a larger number of accessed users,more distant communication distances,and poorer channel conditions,and it also demonstrates the superiority of combining D2 D with NOMA.Aiming at the power allocation problem in the D2D-NOMA communication cluster,a power allocation algorithm based on Deep Deterministic Policy Gradient(DDPG)is proposed.Since errors can occur when the NOMA receive side is decoded leading to failure,in this thesis,the performance of the proposed algorithm under different SIC parameters is analyzed considering the NOMA imperfect SIC decoding scenario.The simulation results show that the throughput performance of this scheme is effectively improved compared with that of the PPO algorithm,random algorithm,and average power allocation scheme,meanwhile,the power allocation fairness of the system is analyzed and the importance of successful SIC decoding in D2 DNOMA communication system is demonstrated.