Physical Layer Security Performance Prediction Based on Deep Learning

This paper presents an innovative approach to enhance communication security by accurately predicting eavesdropper channel strength through real-time variations based on legitimate channels. Utilizing Long Short-Term Memory (LSTM) neural networks, this method trains a model to predict real-time amplitude of the channel. To address the location information of known unlicensed users within a network, an LSTM model is established by collecting channel information from licensed users in a real wireless propagation environment. This model is utilized to predict the channel strength of the unlicensed user. Experimental results demonstrate that this approach accurately predicts the strength of eavesdropping channels, thereby preventing eavesdroppers from obtaining communication activity information and enhancing communication security.