Residual Bi-RNN driven adaptive deep learning-optimization for the joint power allocation and signal detection in a secure MIMO-NOMA system

In Sixth Generation (6G) mobile networks, the Non-Orthogonal Multiple Access (NOMA) serves as a qualified member, which has been obtaining relatively high research interests because of its massive connectivity as well as high spectral efficiency. Different from the existing OMA, like Time Division Multiple Access (TDMA), the NOMA utilizes the power sector to serve distinct users continuously. The previous works on NOMA have highly concentrated on the spectral efficiency improvement. Meanwhile, this experiment concentrates on improving the Secrecy Sum Rate (SSR) in MIMO-NOMA systems by addressing the underlying nonlinear characteristics of both power allocation and signal detection tasks. In the MIMO-NOMA system, the Singular Value Decomposition (SVD) technique is employed to break down the channels in the network. The complex and nonlinear nature of the communication environment, most importantly under secrecy constraints and the multi-user interference, demands a hybrid optimization and a learning-aided solution. To this end, the designed framework leverages the Fitness-based Random Number Swarm Bipolar Algorithm (FitRand SBA) to fine tune the power allocation for near and far users, maximizing the SSR in a nonlinear multi-dimensional search space. Simultaneously, a deep learning-assisted signal detection mechanism utilizing an Adaptive Residual Bi-directional Recurrent Neural Network (AR-BiRNN) is designed to handle the nonlinear temporal dependencies and inter-reference in the signal decoding. The FitRand SBA is employed to optimize the AR-BiRNN parameters, adapting to the wireless channel’s nonlinear behavior. The nonlinear modeling, optimization, and detection methods collectively improve the communication functionality of the MIMO-NOMA system. The simulation results on a 2x2 MIMO-NOMA setup demonstrate that the suggested model attains 28.94 % SSR improvement over conventional optimization algorithms and achieves 95.16 % accuracy contrasted to the existing models. The analytical insights supported by the symbolic computation estimate the optimization process and the detection mechanism. The joint method provides a robust and effective solution for secure, high-performance MIMO-NOMA communications.