Random Forest Bayesian Optimization Deep Transformer-Based Heterogeneous Spatiotemporal Graph Neural Network for Identifying Parameters in QKD

This study presents a comprehensive evaluation of the Random Forest Bayesian Optimization Deep Transformer-Based Heterogeneous Spatiotemporal Graph Neural Network (RFBO-THSGNN) algorithm, a proposed deep learning-based method for channel estimation in MIMO systems. The results demonstrate the superiority of RFBO-THSGNN in terms of accuracy, precision, Secure Key Rate(SKR) generation, transmission distance, and computational time. In channel estimation, RFBO-THSGNN achieves the highest accuracy, ranging from 88% at 100 meters to 99% at 500 meters. It also outperforms other algorithms in precision, secure key rate generation, and transmission distance. Notably, RFBO-THSGNN exhibits a significant improvement in accuracy and precision as the distance increases, making it more robust to channel distortion and noise. Furthermore, it achieves a higher secure key rate and transmission distance compared to other methods. In terms of computational time, RFBO-THSGNN demonstrates a significant advantage, requiring less than 130 seconds across all distances. Overall, the results suggest that RFBO-THSGNN is a promising solution for MIMO channel estimation, offering a superior balance of accuracy, precision, security, and computational efficiency.