Deep Reinforcement Learning-Based MEC Offloading and Resource Allocation in Uplink NOMA Heterogeneous Network

With the advancement of fifth generation(5G) technology, mobile edge computing (MEC) has been considered an effective solution to 5G technical problems. The applications of non-orthogonal multiple access (NOMA) in heterogeneous networks is gradually being considered as an method to increase network throughput and improve spectrum utilization. By assigning non-orthogonal communication resources to different users at the transmitting end, the utilization rate of the spectrum can be maximized. Based on these advantages, we analyze thoroughly the MEC based on NOMA in this paper. In the NOMA system, we focus on optimizing channel resources, user offloading pattern and transmit power. These all characteristics have major role in obtaining the optimized user energy consumption. In recent years, deep Q network (DQN) is considered to be an effective method to solve the model-free problems. Different from traditional heuristic algorithms, we design multi-agent DQN to solve resource allocation in NOMA system. Due to the strong coupling between multiple decisions and the large solution space in dynamic optimization, there are found great challenges to the optimization of resources allocations. According to the simulation results, we can see that the DQN method for multi-agents can allow each agent to find approximately the optimal solution.