Detection of radar probing signals with intra-pulse modulation using second-order synchrosqueezed Fourier transform in cognitive electronic warfare systems

Problem Statement. Modern electronic warfare (EW) systems operate in multiple modes and employ adaptive algorithms, which significantly complicates their detection and, in some cases, makes them virtually undetectable. One of the primary functions of airborne EW systems designed for radio-frequency environment analysis is the early detection of hostile radar systems. Current trends in EW development are driving the adoption of cognitive approaches, particularly artificial intelligence (AI), to reduce the reliance on human operators in the detection and identification of radar probing signals. This transition enables the search process to be fully automated and allows signal analysis to be performed in real time. Combining artificial intelligence and machine learning with advanced EW technologies provides an opportunity to demonstrate the effectiveness of neural networks in detecting radar probing signals with intra-pulse modulation Conventional electromagnetic spectrum analysis methods based on windowed Fourier transforms are often unable to extract a sufficient number of discriminative features for reliable modulation identification. This reduces the probability of detecting emitters and increases data-processing time, making real-time response to an adversary difficult. Neural networks can effectively address the modulation recognition problem; however, their performance depends on the availability of informative features. Such features can be obtained not only through high-resolution time-frequency representations of signals but also through energy-based representations. The second-order synchrosqueezed Fourier transform (FSST2) is particularly well suited for this purpose.

Objective. The objective of this study is to develop an intra-pulse modulation recognition method based on the second-order synchrosqueezed Fourier transform (FSST2) for generating representations of the radio-frequency environment and subsequently detecting radar probing signals and identifying their intra-pulse modulation types using a convolutional neural network (CNN).

Results. The study compares the accuracy of intra-pulse modulation recognition for radar probing signals obtained using different signal-processing methods and neural network architectures in order to determine the most effective combination of processing techniques and deep-learning models.

Practical Significance. The application of the second-order synchrosqueezed Fourier transform to generate representations of received signals provides substantially higher energy concentration in the resulting images. This improves the recognition of radar probing signals, including under negative signal-to-noise ratio (SNR) conditions, thereby enabling the timely implementation of appropriate countermeasures.