Optimizing 5G Massive MIMO Systems Using DNN and DSAC-T: A Scalable Adaptive Beamforming Framework

Advanced Fifth Generation (5G) technologies play a crucial role in meeting the growing demand for high-capacity wireless communication, leading to the adoption of massive multiple-input multiple-output (MIMO) systems. This paper proposes an adaptive beamforming framework for 5G networks that integrates deep neural networks with the Distributional Soft Actor-Critic with Three Refinements (DSAC-T) algorithm. The framework optimizes beamforming vectors and transmission parameters in orthogonal frequency division multiplexing (OFDM)/MIMO systems while addressing challenges such as computational complexity, scalability, and system instability in dynamic channel conditions. DSAC-T improves learning stability through twin value distribution learning and critic gradient adjustments while mitigating policy overestimation using variance-based target return clipping. Simulation results demonstrate its superiority over traditional reinforcement learning methods, showing enhanced spectral efficiency, energy efficiency, latency reduction, and bit error rate performance. The framework remains robust across diverse channel conditions, making it well-suited for real-time deployment in 5G and beyond. This work provides a scalable and efficient approach to optimizing communication systems and lays a foundation for future research on reinforcement learning-based beamforming in advanced wireless networks.