Enhancing Spectrum Sharing Efficiency in Large-Scale MIMO Systems over Integration of Cognitive Radio and Reinforcement Learning

Cognitive Radio Networks (CRNs) aim to optimize the limited frequency spectrum by enabling spectrum sharing among different networks and making use of
unoccupied frequency bands. The combination of massive multiple-input multiple-output (mMIMO) and CRNs has the potential to greatly improve the efficiency of upcoming
wireless communication networks. In our research, we introduce an innovative approach to spectrum sharing in cognitive radio, utilizing 3D spatial data acquisition and Deep
Learning for learning and decision-making. We incorporate mMIMO structures into cognitive radio base stations (CRBS) to extract angular information from user equipment
(UE) and estimate Direction of Arrival (DoA) using Iterative Hard Thresholding (IHT). Our method involves deploying two base stations per cell for comprehensive 3D spatial
spectrum coverage during spectrum prediction. We employ advanced Deep Learning techniques for spectrum sensing instead of reinforcement learning, enhancing CRN
performance. Our approach includes a two-fold spectrum scheduling strategy, one focused on maximizing CR coverage and the other on optimizing transmission rates in
CRN mMIMO scenarios. By fine-tuning Deep Learning parameters, our model achieves significantly higher Average Aggregate Sum Rate (AASR) compared to previous
CRN spectrum sharing methods, without relying on reinforcement learning for spectrum sensing. Our research underscores the effectiveness of integrating Deep Learning
into cognitive radio networks, offering the potential for enhanced spectrum utilization and network performance. Additionally, we address energy efficiency using the Nakagami
fading channel model and evaluate key metrics, including channel occupancy and energy efficiency, through experimental analysis.