Intelligent separation method for instantaneous narrowband time-frequency aliasing signals based on diffusion generation

The rapid development of electronic warfare at present has led to an increasingly complex electromagnetic environment, which presents characteristics such as density, transience and high dynamics. The aliasing phenomena of various types of modulated signals and interferences are increasing day by day. Electronic reconnaissance systems are facing the severe challenge of resolution failure, and it is urgent to establish a theory of the separability of aliased signals in the entire domain and efficient separation methods. The existing research on the separation of time-frequency aliasing signals has problems such as abundant prior information and poor separation effect.

This paper proposes an intelligent signal separation method based on diffusion generation. Firstly, in the time-frequency domain, the semantic segmentation of the aliased signal is carried out based on the UNet network to obtain each signal region corresponding to the time-frequency non-overlapping part and form the signal mask. Furthermore, the time-frequency graph of the single-component signal is obtained based on the mask, and the time-frequency inverse transformation is performed to obtain the single-component signal with missing parts. Finally, based on the latent diffusion model, the training module of latent variables was removed. By improving the network parameters and adapting to the time-domain signal processing, the missing single-component signals were used as conditions and concatenated with noise as the model input. A time-frequency consistency loss was designed to complete the model training and achieve the reconstruction of each signal component.

When the signal-to-noise ratio is greater than -10 decibels, high-fidelity separation and reconstruction of frequency-modulated aliasing signals can be achieved. When the signal-to-noise ratio is equal to 10dB, the correlation coefficient between the reconstructed signal and the original signal is higher than 0.98. All experimental indicators are higher than those of the existing models.

This paper aims at the problems such as excessive prior information and poor separation effect in the research of time-frequency aliasing signal separation in complex electromagnetic environments, and proposes a method of segmentation first and then reconstruction. By performing semantic segmentation on the time-frequency graph of the aliased signal, the component signals are separated in the time-frequency domain, and then the original signal is reconstructed based on the conditional diffusion model. The simulation results show that in the scenario where the aliasing of linear frequency modulation signals and nonlinear frequency modulation signals is relatively complex, the signal reconstruction effect of this algorithm is better than that of the existing models.