Security authentication based on multimodal feature fusion and BO-BiLSTM
Yitao Wang, You Wu · 2024
Traditional physical layer security authentication methods often face unstable performance and malicious attacks when dealing with complex dynamic environments, resulting in high detection errors. To address this issue, this paper proposes a secure authentication method based on multimodal feature fusion and bidirectional long short-term memory. When establishing communication between the receiver and sender, obtain channel state information and extract amplitude and phase information to construct a two-dimensional channel feature vector; Perform discrete wavelet analysis (DWT) and Exponential Moving Average (EMA) on the two-dimensional channel feature sequence, optimize the model using Bayesian optimization algorithm, and construct a BiLSTM authentication model to obtain the predicted values of the fused feature sequence, and determine whether the information at the next moment comes from a legitimate user. Experiments have shown that this authentication method has lower detection errors than other traditional authentication methods.