This study introduces a Double Deep Q-Network (DDQN) optimization framework to improve massive MIMO-OFDM systems via reinforcement learning-driven adaptive parameter selection. It utilizes a dual network architecture to mitigate overestimation bias and incorporates dynamic optimization for power allocation, subcarrier fraction distribution, and modulation scheme selection across QAM-16, QAM-64, and QAM-128 configurations. Extensive simulations performed across Signal-to-Noise Ratio ranges from -5 to 35 dBm reveal substantial performance enhancements, with DDQN-augmented systems attaining 5-6 dB SNR savings for equivalent SE, a 50% increase in EE reaching 15.5-16 Gbps/W compared to conventional 10.5-11 Gbps/W implementations, and a 2.5 dB SNR reduction for a BER performance of 10⁻⁵. The optimization framework ensures uniform parameter selection across diverse SNR conditions, facilitating a 40-50% increase in coverage through enhanced low-SNR performance while delivering a 5 dB SNR improvement in low-power operating scenarios. The study establishes a basis for intelligent communication systems that can autonomously adapt to 6G wireless networks, supporting ultra-reliable low-power communications and mobile edge computing applications.
Ayad Atiyah Abdulkafi (Mon,) studied this question.