Frequency Offset Estimation Based on Dual-Channel Spatio-Temporal Transformer and Adaptive Hierarchical Search for LEO Satellite Communications

The rapid deployment of large-scale low Earth orbit (LEO) satellite constellations, such as SpaceX's Starlink, is transforming global communications. However, the high-speed movement of LEO satellites induces significant Doppler effects, causing frequency offsets in Ka-band (30 GHz) communications that exceed conventional synchronization algorithms. These offsets degrade performance as traditional methods face large search spaces, high computational complexity, and inconsistent accuracy. This paper presents a novel frequency offset estimation approach using a spatio-temporal Transformer architecture. Firstly, the method employs a dual-channel deep learning framework combined with a spatial attention mechanism to model and extract the temporal dynamics of satellite positions and velocities, capturing complex relative motion patterns. Secondly, it integrates an adaptive search strategy based on high-precision trajectory predictions with a hierarchical estimation structure that combines coarse and fine estimations, optimizing the search space and ensuring high accuracy and real-time performance. Experimental results show the proposed method achieves a Root Mean Square Error (RMSE) of 1.868 kHz, improving accuracy by 33.5% over traditional Fast Fourier Transform (FFT) methods. Computational efficiency is enhanced with processing delays reduced to 9.07 ms, representing a 35.5% gain. The method maintains stable performance across a Signal-to-Noise Ratio (SNR) range of -30 dB to 5 dB, demonstrating robustness and suitability for practical LEO satellite communication systems.