Cognitive automatic classification of radar signals using convolutional neural networks

Problem statement. Continuous improvement of radar stations (radars), which are used to control the electromagnetic spectrum, makes it necessary to respond symmetrically to the calls of the electronic warfare (EW) system. Radar signals are becoming more complex and sophisticated. EW systems are required to reduce the time for recognizing the structure of the signal and its classification, preferably in real time. Such tasks can only be solved using cognitive technologies. Cognitive electronic warfare (CEW), using artificial intelligence algorithms, allows faster and more accurate recognition and classification of electromagnetic pulses (EMP) of enemy radars, in contrast to existing electronic warfare systems. A suitable tool for solving the problem of reducing the processing time of an electromagnetic signal and its most accurate recognition is a convolutional neural network (CNN), which is effective in processing digital images. It has also shown good results in an automatic system for extracting the parameters of radio signals. It is the best model with a high Success Rate (HSR) with high noise background energy.

Objective. The purpose of the article is to demonstrate by examples how the modulation of various sounding signals of the enemy's radar is determined using the CNN.

Results. The CNN architecture for the system under study is determined. In particular: how many convolutional layers are needed, what number of hidden units is suitable, and what is the best pooling strategy.

Practical significance. The article demonstrates the results of radio signal classification by the method of computational experiment. The simulation results show that the total HSR is 93.7% with a signal-to-noise ratio (SNR) of -2 dB.