In the evolving landscape of wireless communications for 5G, 6G, and beyond, the deployment of unmanned aerial vehicles (UAVs) has emerged as a groundbreaking strategy to expand coverage areas due to their flexibility and ease of deployment. Simultaneously, reflecting intelligent surfaces (RISs) have introduced a transformative paradigm aimed at improving key performance metrics, such as average sum-rate and energy efficiency (EE). The seamless integration of advanced technologies, including UAVs, RIS, and nonorthogonal multiple access (NOMA), presents a promising avenue for significantly boosting the performance and efficiency of next-generation communication systems. This study investigates EE maximization for two scenarios in a NOMA-enabled mmWave network: 1) multi-UAV-mounted base stations (BSs) and 2) multi-UAV-mounted distributed RIS. In both cases, each UAV serves a NOMA cluster with imperfect successive interference cancellation (SIC), capturing the impact of hardware impairments in real-world NOMA systems. For each scenario, an optimization problem is formulated to maximize EE by jointly optimizing the beamforming matrix, phase shift matrix, NOMA power allocation, and UAV 3-D placement. The nonconvex problems are tackled using both model-based and model-free deep reinforcement learning (DRL) algorithms under constraints, such as minimum Quality of Service (QoS), beamforming and phase shift limits, and UAV trajectory constraints. The simulation results demonstrate that the proposed DRL algorithms significantly enhance spectral efficiency (SE) and EE, showcasing their suitability for 6G communication systems. Furthermore, a comparative analysis with orthogonal multiple access (OMA) and spatial-division multiple access (SDMA) confirms that NOMA outperforms both techniques, achieving substantial gains in efficiency and performance.

In this paper, a novel beyond-diagonal RIS (BD-RIS) is suggested an architecture to improve the spectral efficiency (SE) of wireless communication systems. We use deep reinforcement learning (DRL) to solve the joint design problem of the RIS phase shift matrix and BS beamforming to maximize the SE. In addition to the phase of each element, we also optimize the position of non-diagonal elements in the RIS phase shift matrix. Simulation results show that the proposed BD- RIS architecture with DRL outperforms the conventional diagonal RIS (D-RIS) architecture with DRL in terms of SE. We also investigate the effect of the number of quantization bits on the performance of the DRL algorithm. We show that there is a trade-off between accuracy and complexity.

In this paper, a cost-effective and high-accuracy precoding technique based on$\epsilon $-Fuzzy pareto active learning (FPAL) is proposed for millimeter wave (mmWave) massive MIMO communications. The proposed method achieves a low iteration convergence with low complexity. Two practical structures, namely fully-connected and partially-connected structures are considered for hybrid precoding. Furthermore, the effect of high and low-resolution quantization in the digital-to-analog converter, phase shifter, and imperfect channel state information are discussed. The performance results of the proposed technique and alternating minimization methods beside fully digital techniques are compared and discussed in the terms of the spectral efficiency (SE), bit error rate (BER), normalized mean square error (NMSE), energy efficiency (EE) for phase-shifter (PS) with low bit quantization, and imperfect channel state information (CSI). To address the applicability of the proposed method, experimental results are obtained from a real mmWave hardware setup compliant with 3GPP standards, and verify the simulated ones for the proposed$\epsilon $-FPAL hybrid precoding scheme. The reduced complexity, higher performance and EE make the proposed method suitable for 5G and beyond communication systems.