Wireless Devices "Fingerprint" Feature Extraction Based On Signal Analysis

Wireless device identification is a very complex and difficult problem. How to make use of various methods in the field of modern signal processing and pattern recognition is the key point. In addition to routine analysis for communication signals, analyzing the subtle features of radio signals would be an excellent method in the communication signals processing field.In the paper we design, implement, and evaluate a technique to identify GSM signal transmitters. This technique leverages the Higher Order Spectrum Analysis (HOSA) to eliminate additive Gaussian noise in GSM signals, the Principle Component Analysis (PCA) to extract signal’s features and the BP Neural Network to classify the provided signals. We measured GSM signals from6different mobile phones using the same SIM card. Then we implemented the up-following3methods to identify the6different mobile phones. We experimentally demonstrate the identification accuracys of this technology in differentiating between300signal samples isup to90%. Our results also show that GSM signal identification is feasible using the mash-up of traditional data processing and pattern recognition techniques.