Communication Radio Signals Fingerprint Feature Extraction And Recognition Based On Manifold Learning

Communication signals' fingerprint feature comes from emitter error based on designation, manufacturing technology and uncontrollable factors. The controllable errors are consciously removed for protecting the information. Differ radio individual generates signals difference, meanwhile, the very signal marks the particular radio emitter.Signals fingerprint feature extraction is based on multiple transformations of the signals.The cumbersome high dimensional data would cause the curse in the process afterwards,which may lead to error. Manifold learning procedures could realize the visualization by embedding the high-dimension data into 2 or 3dimensions while preserving as much as possible the metric in the natural feature space, which makes observation and analysis easier.On account of the analysis above, the fingerprint feature extraction based on manifold learning algorithm is studied. The main research contents are presented as following.Modulation types of communication radio signals were analyzed. Common domain transformation methods were simulated to extract the signals feature.Manifold learning algorithms were studied. Focused on the research of 2 manifold learning methods, simulation was executed to find out the optional parameter of the algorithm to minimize the loss during the embedding. Furthermore,a feature fusion method based on manifold learning was presented.Communication signals modulation identification system is established. Short-time Fourier transformation or wavelet transformation was used as pretreatment for different kinds of signals. Then extracted the signals feature combined with manifold learning. Furthermore,an effective modulation type recognition system is designed with classifier.Individual radio transmitter identification is achieved. The mechanism of AM, FM, FSK radio fingerprint feature was analyzed separately,different feature extraction methods were provided for the specific radio. Manifold learning based feature fusion algorithm was used to extract low-dimension feature. Simulation indicated the method has good performance.