Learning-Based Latency and Energy Optimization in SCMA-Enhanced UAV-MEC Networks

Sparse code multiple access (SCMA) technology can provide massive connectivity and timely computing in unmanned aerial vehicle (UAV) assisted mobile edge computing (MEC) networks. In this paper, the long-term task latency and energy consumption minimization problem in SCMA-enhanced UAV-MEC networks is investigated. We first formulate it as a Markov decision process (MDP) under a dynamic environment and then propose a joint computation offloading, SCMA resource allocation, and UAV trajectory algorithm based on convolutional neural network (CNN) and deep deterministic policy gradient (DDPG). Specifically, the UAV is taken as an agent to extract channel and task features of multiple devices through CNN for action exploration. The weighted sum of task latency and energy consumption is used as a reward with a penalty for failing to complete the task within the deadline or when the UAV battery runs out. Eventually, the near-optimal strategy is obtained through experience training and interaction with the environment. Simulation results illustrate that the proposed algorithm can achieve convergence and has greater advantages over other benchmark algorithms.