Feature Extraction and Classification of Unknown Types of Communication Emitter

In the field of cognitive radio, communication emitter recognition plays an important role. The traditional methods need to determine expert feature in advance, which leads to inefficiency in some case. One is that the emitters may not be observed in the training and we need to classify much larger new classes examples without knowledge of the label. Therefore, this paper proposes a semi-supervised method for extracting RF fingerprint based on supervised Convolutional Neural Networks and unsupervised clustering. Experimental results demonstrate that the features extracted from CNN can distinguish emitters and the proposed method has a great performance in classifying unknown types of emitters.