Deep Reinforcement Learning for Aerial Data Collection in Hybrid-Powered NOMA-IoT Networks

With the help of unmanned aerial vehicle (UAV), remote terminals that out of wireless coverage can be connected to the Internet of Things (IoT) networks. Currently, the IoT relies on a large number of low-cost wireless sensors with limited energy supply to realize ubiquitous monitoring and intelligent control. The hybrid-powered networks composed of wireless-powered communication (WPC) terminal and solar-powered UAV can solve the energy supply problem of the IoT networks, and the nonorthogonal multiple access (NOMA) technique can solve the massive access problem of IoT terminals. Exploiting these benefits, we investigate joint UAV 3-D trajectory design and time allocation for aerial data collection in hybrid-powered NOMA-IoT networks. To maximize the total fair network throughput, we jointly consider energy limitation, Quality of Service (QoS) requirements, and flight conditions. The problem is nonconvex and time-dimension coupled which is intractable to solve by traditional optimization methods. Therefore, we develop a deep reinforcement learning (DRL) algorithm called fair communication is accomplished by trajectory design and time allocation (FC-TDTA), which uses the deep deterministic policy gradient (DDPG) as its basis. Simulation results show that our proposed algorithm performs better than benchmarks in fair throughput maximization. The proposed FC-TDTA algorithm can make the UAV: 1) fly in appropriate direction and speed, so that the UAV can arrive at the charging station before the energy runs out and 2) conduct WPC energy transmission and data collection to achieve fair communication.