Automatic detection of signal spectrum modulation using dual convolution network

Automatic waveform recognition has been widely adopted in radar systems and spread spectrum communications due to its advantages, such as increased reliability and sensitivity. Identifying the modulation of received signals helps to recognize different invader transmitters. Using a variety of modulation types, robust features must be extracted from received signals using contemporary deep neural networks. The purpose of this paper is to present a dual network for automatic classification of modulations. The suggested approach comprises of two pipelines: one for contextual feature extraction from time-frequency representation and the other for feature extraction from time-domain signals. Choi-Williams representation is employed to describe time-frequency analysis, and a hybrid convolutional neural network is created to identify received signal modulation. An artificial picture dataset is produced to train visual context, and is utilized to evaluate the model. Experimental results demonstrate the advantage of the proposed method on accuracy that we will clearly show in this article.