Research Of Resource Allocation Method For Millimeter-wave Massive MIMO Systems Based On Deep Reinforcement Learning

With the commercialization of fifth-generation mobile communications and the development of sixth-generation mobile communications,in order to further meet the explosive growth requirements of the number of users and their corresponding wireless data traffic,millimeter wave massive multiple-input and multiple-output,(mm Wave massive MIMO)technology can provide greater bandwidth and higher spectral efficiency,and thus significantly improving the achievable sum-rate performance to multi-gigabits per second,which is considered to be one of the most promising technologies.As the number of base station antennas and users grows intensively,the radio resources of communication system become relatively limited,and the unreasonable allocation of resources have an impact on system performance,posing a challenge for mm Wave massive MIMO systems.Therefore,it is of great significance to study the resource allocation algorithm for mm Wave massive MIMO systems under the resource constraints.The objective of this thesis is to improve the system throughput under the constrained resources through exploiting deep reinforcement learning(DRL)to jointly allocate the sub-channel and power resources of the systems.First of all,in this thesis,we establish the mm Wave massive MIMO-NOMA system model by introducing the non-orthogonal multiple access(NOMA)technique into the mm Wave massive MIMO system.Moreover,in order to reduce the multi-user interference,the user grouping algorithm is designed.For the sub-channels and power resources of the system,we propose a joint sub-channels and power allocation model for the mm Wave massive MIMO-NOMA system under a variety of constraints,which portray the effects of different constraints on the achievable sum-rate performance.Furthermore,the resource allocation optimization problem is non-convex and Np-hard,we make full use of DRL technique with adaptive and autonomous decision-making capabilities to propose a joint user grouping,sub-channels and power allocation algorithm based on deep Q network(DQN)for mm Wave massive MIMO-NOMA systems.The simulation results indicate that our proposed DQN-based joint resource allocation algorithm has a good convergence speed and can achieve better achievable sum-rate performance of the mm Wave massive MIMO-NOMA system.Finally,since the quantized power can cause the system performance loss and have the problem of slow convergence in the traditional DQN algorithm,we design a new efficient joint user grouping,sub-channels and power allocation scheme by using the competitive network structure of Dueling DQN to accelerate the convergence speed and using the deep deterministic policy gradient(DDPG)to effectively handle the continuous action space problem and thus reducing the high dimension of the action space.The simulation results show that our proposed Dueling DQN-DDPG-based resource allocation scheme can accelerate the convergence speed and maximize the system achievable sum-rate.