Multiband Radar Signal Fusion and Extrapolation Method Based on Transformer Model

The improvement of range resolution is crucial for enhancing the capability of imaging radars to detect small targets, such as drones. However, due to technical and cost limitations, current radars cannot indefinitely increase hardware bandwidth to enhance range resolution. To address this challenge, the multiband fusion and extrapolation technique has emerged. This method uses multiple discrete sub-band radar signals and fuses them to extrapolate the missing frequency bands, effectively achieving a wideband radar signal. Traditional approaches primarily rely on constructing signal models and estimating model parameters to accomplish fusion and extrapolation. This paper is the first to apply machine learning techniques to the field of multiband fusion and extrapolation, proposing a radar multiband signal fusion and extrapolation method based on a Transformer model. By training the Transformer network, the method enables accurate estimation of unknown frequency band signals. The effectiveness of the approach is validated through electromagnetic simulation data of drone echoes, demonstrating significantly improved estimation accuracy compared to traditional signal fusion and extrapolation methods.