Rail Transit Radiation Source Identification Method Based on Enhanced Diagonal Integral Bispectrum

• Published in: Chengshi guidao jiaotong yanjiu
• Main Authors: Haichuan LIU, Kexin ZHANG, Hui HUI, Lu WEN
• Format: Article
• Language: Chinese
• Published: Urban Mass Transit Magazine Press 2024-01-01
• Subjects: urban rail transit; radiation source identification; radio frequency fingerprint; integral bispectrum
• Online Access: https://umt1998.tongji.edu.cn/journal/paper/doi/10.16037/j.1007-869x.2024.01.004.html

[Objective] Numerous external interference signals exist in urban rail transit wireless communication system, posing a significant threat to operational safety. Targeting the issue of low identification accuracy due to the radiation source RF (radio frequency) characteristics susceptible to noise and interference, it is necessary to propose an individual identification method for communication radiation sources based on enhanced diagonal integral bispectrum. This method provides an effective new approach to ensuring the security of rail transit wireless communication systems. [Method] The data processing procedure and principles of DCLIB (diagonal-correlation local-integral bispectrum) are analyzed. The calculations for bispectrum transformation, enhanced diagonal integral bispectrum calculation, division of adaptive bispectrum integration interval, and radiation source identification method based on residual networks are explained. Simulation experiments are conducted using actual Wi-Fi (wireless fidelity) devices to analyze and compare the identification performance of the DCLIB method with that of other radiation source identification methods. [Result & Conclusion] The DCLIB method first estimates the bispectrum of communication radiation source signals and utilizes the autocorrelation characteristics of each parallel line on the sub-diagonals to form new spectral information for the enhancement of the signal subtle features. Subsequently, the method adaptively selects a reasonable spectral signal integration interval based on the spectral signal strength, reducing both noise impact and algorithm computational complexity. Thus, an enhanced diagonal integral bispectrum is obtained. The proposed DCLIB signal is then used as the RF fingerprint feature of the radiation source, and individual source identification is achieved using a deep residual network. Simulation identification experiments based on actual Wi-Fi devices demonstrate that the DCLIB method achieves the highest identification accuracy and exhibits excellent noise-resistance performance.