A Deep Reinforcement Learning Scheme for SCMA-Based Edge Computing in IoT Networks

The application of sparse code multiple access (SCMA) to multi-access edge computing (MEC) networks can provide massive connections as well as timely and efficient computation services for resource-constrained Internet of Things (IoT) devices. This paper investigates the maximization of computation rate in SCMA-MEC networks under a dynamic environment. We first formulate an initial optimization problem to maximize the long-term computation rate of IoT devices under task delay constraints. Then, a joint computation offloading and SCMA resource allocation algorithm based on long short-term memory (LSTM) network and dueling deep Q network (DQN) is proposed. Specifically, each IoT device acts as an agent in the algorithm. Since each device can only observe part of the environment state, the LSTM network is used to predict the states of other devices. The computation rate of devices is taken as a reward to conduct action exploration in dueling DQN, and then the near-optimal computation offloading decision, SCMA codebook allocation, and power distribution of IoT users are obtained after training. Numerical simulation results demonstrate that the proposed algorithm can achieve higher computation rate compared with other baseline schemes.