Double DQN-based secrecy energy efficiency and fairness performance in IRS-assisted NOMA systems with friendly...

Faced with the challenge of increasing energy consumption and the need for sustainability in 6th generation (6G) wireless communication networks, this paper investigates the secrecy energy efficiency (SEE) and user fairness performance in an intelligent reflecting surface (IRS)-assisted non-orthogonal multiple access (NOMA) system in the presence of a friendly jammer and a passive eavesdropper over THz-Rician channels. We formulate both optimization objectives: maximizing the system's total SEE and maximizing the maximum-min SEE to guarantee fairness for the worst-case user. Additionally, Jain's fairness index is used to quantitatively evaluate the SEE balance among users. To solve these problems, we apply a Double Deep Q-Network (Double DQN)-based SEE of our proposed system model to jointly optimize power allocation and IRS phase shifts. This proposed approach enables efficient learning of optimal policies in dynamic environments without requiring explicit knowledge of the channel distribution. Furthermore, conventional Deep Q-Netowrk (DQN) and random allocation strategies are also implemented for comparison with Double DQN. Simulation results are presented for both IRS-assisted NOMA and OMA (orthogonal multiple access) systems to highlight the advantages of NOMA in terms of the secrecy-energy trade-off and spectral efficiency. Finally, the effects of system essential parameters, such as transmitted power, the number of IRS elements, atmospheric absorption coefficients, and the passive eavesdropper's position, are examined. These simulation results show that the proposed Double DQN-based scheme significantly outperforms DQN and random methods in both SEE maximization and fairness enhancement, and that the NOMA system outperforms the OMA system. These findings confirm that the proposed model provides a basis for deploying a secure, energy-efficient, and sustainable wireless communication for future 6G networks.