An electromagnetic leakage signal characterization method for multi-feature selection and fusion

Unintentional electromagnetic leakage is generated during the use of various components of computer equipment. These leaked electromagnetic signals contain a significant amount of useful information, which can be captured and recovered, potentially leading to security problems. However, characterizing these signals is challenging due to substantial environmental noise, making it impossible to extract effective information using a priori knowledge. To address this issue, this paper proposes a multi-feature filtering and fusion-oriented method for characterizing electromagnetic leakage information. This method calculates ten types of feature curves, such as peak-to-peak, variance, and mean, to highlight effective signals in the data. Additionally, it introduces the index based on Wasserstein distance for screening to retain effective features and eliminate irrelevant and redundant ones. A multi-feature fusion model, called the EFT model, is designed using the encoder of the transformer architecture to fuse and process complementary information from the features. Experimental results demonstrate that the proposed method effectively improves data characterization, fully exploits the complementarity between multiple features, and provides more accurate vector representations for tasks such as electromagnetic leakage source localization and leakage source analysis, thereby enhancing the final detection performance.