Modulation Order Identification of CPM Signals Based on Deep Learning

The modulation order is a crucial parameter of continuous phase modulation (CPM) signals. Its accurate identification is a fundamental prerequisite for achieving blind demodulation of CPM signals in non-collaborative communication. By exploring the inherent characteristics of CPM signals and the correlations among their parameters, this paper proposes a deep learning approach for modulation order identification of CPM signals based on exponential adjustment. Specifically, the instantaneous frequency derived from signal preprocessing serves as the input to the deep learning model, and a modulation index estimator is employed to approximate the modulation index of the signal. Subsequently, the estimated modulation index is utilized to adjust the instantaneous frequency of the signal, which is then fed into a recognition neural network for feature extraction and classification. The experimental results demonstrate that the method is applicable to a wide range of CPM signals, with superior recognition accuracy and noise immunity compared to traditional algorithms.