MMU-STCNN-BDQ: a deep reinforcement learning framework for secure and energy-efficient beamforming in 6G mMIMO networks.

To achieve energy-efficient and secure beamforming in 6G massive MIMO networks, especially in millimeter-wave (mmWave) bands. This research work introduces a hybrid framework utilizing the Massive Multi-User Spatiotemporal Convolutional Neural Network (MMU-STCNN) and Bi-Gated Deep Q-Learning (BDQ). The MMU-STCNN extracts secure beamforming prediction vectors from channel state information (CSI) using spatiotemporal and pooling layers enhanced by Multi-User Attention (MUA) mechanisms. These predicted vectors are further optimized through a dynamic bi-gated deep Q-reinforcement learning (BDQ) to enhance both security and energy efficiency. The proposed hybrid architecture addresses the critical challenges in energy efficiency and adversarial robustness, outperforming baseline beamforming methods such as Deep Neural Network (DNN), Reinforcement Deep-Q-Network (DQN), and Adversarial Robust Beamforming (ARBF) across multiple metrics. Experimental results show that the proposed method outperforms DNN, DQN, and ARBF. The proposed method reduces MSE, BER, and increased throughput for 10, 20, and 50 users at SNR levels from 10 to 20 dB. This method is effective for large-scale multi-user communication environments due to its high spectral and energy efficiency. This framework is the key solution for future deployments in 6G mMIMO networks in terms of intelligence, security, and energy efficiency.