Optimal AI-Enabled Secured NOMA Among Untrusted Users

To develop a cyber-physical artificial intelligence enabled wireless network, it is essential to support unprecedented high throughput and efficient spectrum utilization in a practically unknown channel. In this regard, we need to investigate the design aspects of the network exploiting deep learning-based non-orthogonal multiple access (NOMA) for a model-free environment. In this work, a model-free deep learning algorithm based on deep deterministic policy gradient is proposed that provides a continuous course of actions under the optimal policy for an untrusted NOMA network. Utilizing the concept of physical layer security, we focus on maximizing the sum secrecy rate of the system in terms of decoding order and transmitting power allocation to users under the limited energy constraint at the base station. Via extensive simulations, while training, we measure the performance of the deep learning algorithm in terms of cumulative sum secrecy rate, convergence rate and stability. Also, after the training, we obtain various insights on the performance of the obtained optimal policy by varying the independent system parameters and compare the algorithm against a benchmark that provides the improvement of nearly 55% in the noisy channel.