Deep Learning-Aided Pareto Front Prediction in Secure Noma Systems

The physical layer security of a non-orthogonal multiple access (NOMA) system is investigated. In order to maximize the security level of each NOMA user in the system, a multiobjective optimization (MOO) problem is proposed for handling the relationship among conflicting objectives. A deep neural network-based framework is designed for solving the associated MOO problem and for estimating the Pareto front. The framework shows that the estimated Pareto front is very close to the true one, thus allowing designers to select Pareto optimal solutions for striking the most appropriate compromise for all users, even on dynamically time-variant basis. Numerical results are provided for illustrating the associated trade-offs.