تشخيص چند شكل موج مدوله شده فركانسي همپوشان با استفاده از يادگيري عميق

An important issue in electronic support systems is the automatic detection and classification of modulation. In such systems, by sampling the signal within a specific spectral range, necessary processing is carried out, and the existing signals along with their modulation types are identified. Extensive research has been conducted on the classification of radar signal modulation. Conventional methods attempt to classify signals using various features present in the signal. Feature-based approaches heavily rely on specialized knowledge, which may only perform well in specific cases and can generally be weak or suffer from high complexity and slow performance. To address this issue, machine learning and deep learning-based classifiers can be utilized, which is the focus of this research. One of the commonly used waveforms in radar systems is frequency modulated (FM) waveforms. On the other hand, in certain scenarios, such as coastal areas where there are numerous radars related to stations or vessels, or in military combat scenes where there are many radar equipment, the likelihood of overlapping radar signals in a given observation range is relatively high. Additionally, sometimes jammers in the environment utilize linear frequency modulated (LFM) waveforms. Therefore, a plausible scenario when detecting radar signal modulation involves the presence of multiple overlapping frequency-modulated signals, and for this reason, this topic is examined in detail in this thesis. In this scenario, methods based on instantaneous frequency detection do not perform adequately. As an alternative method, this research presents approaches based on time-frequency transforms and the detection of linear or nonlinear patterns in the time-frequency domain. To this end, we classify linear and nonlinear frequency-modulated signals using deep learning and attempt to estimate the parameters of multiple LFM signals when multiple LFM signals are present. The performance of the proposed method has been eva‎luated through simulation compared to existing methods, yielding a significant improvement.

5

دكتر محمد كاظمي