CNN-Based Synergetic Beamforming for Symbiotic Secure Transmissions in Integrated Satellite-Terrestrial Network

The integrated satellite-terrestrial network has received keen attention in the 6G wireless communication network, as it can significantly improve network coverage and system transmission efficiency. However, it is susceptible to eavesdropping threats. Physical layer security serves as an endogenous means to counter eavesdropping, but due to constraints such as communication link similarity and resource limitations, it fails to substantially enhance secrecy rates. The flourishing development of new technologies like machine learning has introduced novel opportunities for enhancing physical layer security. In the paper, we propose a synergetic beamforming scheme based on convolutional neural network (CNN) to achieve symbiotic security among heterogeneous downlinks in integrated satellite-terrestrial network. This method divides the acquired channel state information into real and imaginary parts, serving as inputs to the network, and outputs beamforming vectors for both the base station and satellite. During the training process, we incorporate constraint conditions as penalty terms into the loss function and correlate training parameters with iteration count, leading to improved training performance. Lastly, extensive simulation verification is conducted to demonstrate the effectiveness of this method in enhancing secrecy rate.