Labeling of Radio Signal by Jointly Using Feature Engineering and Unsupervised Learning

For the purpose of radio transmitter identification with deep learning under non-cooperative conditions, this paper proposes a radio frequency(RF) signal sample labeling method based on unsupervised learning and radio frequency fingerprints. Firstly, extracted the carrier component fingerprint information of the radio signal to form a characteristic vector of representative individual information. Then utilized an unsupervised learning method which is improved by canopy algorithm to realize sample labeling of individual transmitter. The experimental process is based on the ACARS signal, used six types of ACARS signals with different IDs to verify the performance of the method. The results revealed that the accuracy of the method for the overall identification and labeling of radio frequency signals was over 96%.