Abstract

The evolution of wireless communication toward 6G has intensified the demand for MIMO-OFDM systems that deliver higher spectral efficiency, robust interference handling, and adaptive performance under dynamic conditions. The rapid fluctuation of traffic patterns and channel states has created persistent challenges for conventional optimization strategies. These limitations have often constrained throughput, increased latency, and reduced the reliability of high-density 6G environments. The present study addressed this gap by integrating a reinforcement learning framework that has enabled autonomous decision processes for power allocation, beamforming, and subcarrier mapping. An agent interacted with the channel environment, evaluated the reward based on instantaneous signal quality, and adapted the transmission strategy through repeated exploration that has refined action selection. The method employed a deep Q-learning architecture that extracted real-time channel features and updated the policy based on cumulative rewards. The system used an iterative training cycle that improved convergence stability and prevented premature policy saturation. Experimental simulations using an 8×8 MIMO configuration with 256-QAM modulation and 128–256 subcarriers show that the reinforcement learning-enhanced system achieves spectral efficiency of 6.8–7.2 bps/Hz and reduces the bit-error rate to 1.8–1.5×10-3. Total power consumption decreases to 3.32–3.30 W, while the learning agent converges within 700–780 episodes. Compared to PSO-based resource allocation, DNN channel estimation, and DQL power allocation, the proposed method improves spectral efficiency by up to 1.0 bps/Hz, reduces BER by up to 25%, and lowers power consumption by 2–3%. These results indicate that the proposed framework delivers adaptive, high-throughput, and energy-efficient performance under dynamic 6G channel conditions.