Modulation Recognition Algorithm based on Digital Communication Signal Time-Frequency Image

Signal automatic modulation recognition stands for the modulation modes that can classify and recognize various communication signals automatically. Modulation recognition, as the midway between the detection and demodulation for signals, is critical in both military and civilian applications. Electronic countermeasures and software radio, for example. Although modulation recognition technique has advanced significantly in the past few years, there are still various challenges to be overcome as the complex engineering environment and wireless communication channel requirements become more sophisticated. This paper proposes an approach based on the Convolutional Neural Network (CNN) feature fusion to overcome the problem of numerous modulation modes failing to examine the correlation in distinct characters. Moreover, a pseudo Wigner-Ville distribution is appeared in this paper to turn a one-dimensional signal into an image, the image features are extracted by a kind of Convolutional Neural Network model, and the features extracted by CNN are combined with artificial features. The modulation recognition performance is further enhanced by the fusion feature. Simulation findings suggest the technique proposed in this paper enhances the ability especially when the SNR is low, implying this mode of building fusion features is more successful.

I. Introduction

Deep Learning (DL), which has made great advancements in computer vision [1], [2] and other fields, has been highly concerned by experts and researchers at home and abroad [3]–[6] in the past few years due to the quick expansion of large data and rapid popularization of high-performance computing equipment. As a result, it has been introduced into the field of modulation recognition by several researchers. Huang Gao [7] devised a DL-based automatic modulation recognition technique. A spectral correlation function is utilized to convert the signal into an image, and the characteristics of the complex signals from the signal picture is extracted by a deep confidence network. Ali [8] suggested an autoencoder-based automatic modulation recognition algorithm. It may also investigate the hidden properties of data by bringing in non-negative restrictions during the training process of autoencoders to know the input data’s sparse representation in order to promote the classification capacity of the signal.