Deep Automatic Modulation Classification Using Deformation-Insensitive Color Constellation

Automatic modulation recognition(AMC) is an intermediate process between signal detection and signal demodulation, which is an important technology in wireless communication systems. Its main purpose is to determine the modulation mode of the received wireless communication signal, thereby realizing demodulation and subsequent processing of the signal. AMC is considered as the promising methods to improve the quality of the service in cognitive radio(CR). However, AMC suffers from the phase offset of the signal and low recognition accuracy. Therefore, we proposed deformation-insensitive color constellation(DICC) to improve the recognition accuracy in AMC. In this paper, DICC is insensitive to the deformation of the constellation caused by the phase offset and able to represent the density information of points in the constellation diagram. Firstly, we use the method of phase difference to prevent the phase offset. Particularly, we use different colors to match with density information of constellation diagrams, and use deep learning models, VGG-19 and GoogleNet for classification. The results show that for the received signal constellation with carrier phase offset, it still has a high recognition accuracy and the classification accuracy is 3%-4% higher than previous methods.