Effective Resource Management for Hybrid NOMA-OMA Scheme in Satellite Networks With Statistical Delay QoS Constraints

To provide services with various statistical delay quality-of-service (QoS) constraints and minimum performance limitations in future satellite networks, this paper proposes a hybrid access scheme for the satellite networks, where the satellite can transmit with users either in non-orthogonal multiple access (NOMA) or orthogonal multiple access (OMA) mode. Taking user’s statistical delay QoS constraints into consideration, the effective capacities of two modes are investigated and a maximization problem is formulated by jointly considering multiple access selection and power allocation strategy. Specifically, this jointly problem is decomposed into three steps: Firstly, with K-means based Gaussian mixture model algorithm, users are classified into different clusters according to their delay exponents. Based on which, a roughly multiple access suggestion and three user pairing strategies are proposed, by taking the achieved performance with OMA scheme as constraints. Then, users, who are potentially served in NOMA mode, are carefully decided whether the NOMA model is suitable by a deep neural network. Finally, to find the near optimal power allocation factor for each NOMA group, a deep reinforcement learning (DRL) with varying bounds is proposed, with which the lower and upper action space bounds can be flexible varied according to reward, to optimize the power allocation scheme in the considered system. Numerical simulations are provided to show the impacts of statistical delay QoS requirements on users’ achievable effective capacities, the minimum performance on the superiority of pure NOMA scheme, the user pairing strategy on achieved performance, and the proposed method on further enhancing convergence rate as well as achievable performance in satellite networks.