KAN-Transformer Based Carrier Frequency Offset Estimation Method for LEO Satellite OFDM Communication Systems

The substantial Doppler shift inherent in Low-Earth Orbit (LEO) satellite communications presents a major challenge to Orthogonal Frequency Division Multiplexing (OFDM) systems, making accurate Carrier Frequency Offset (CFO) estimation essential. Although deep learning (DL)-based methods have recently shown potential as successors to conventional model-based estimators, they are often constrained by the limited adaptability and nonlinear modeling capabilities of standard networks like Multilayer Perceptrons (MLPs) and Convolutional Neural Networks (CNNs). This paper proposes a hybrid deep architecture, termed KAN-TransformerNet, which integrates the recently proposed Kolmogorov-Arnold Networks (KAN) with a Transformer architecture for enhanced CFO estimation. The model leverages KAN’s superior adaptability and nonlinear representation capabilities through its learnable activation functions to model intricate signal distortions, while the Transformer’s self-attention mechanism effectively captures long-range temporal dependencies in the received signal sequence. Extensive simulations under typical LEO satellite channel conditions demonstrate that the proposed KAN-TransformerNet achieves superior estimation accuracy and robustness compared to conventional Cyclic Prefix (CP) based and standard MLP-based methods across a wide range of signal-to-noise ratios (SNR) conditions. The presented results affirm the effectiveness of our method, demonstrating its potential as an attractive, high-performance option for reliable frequency synchronization under adverse wireless conditions.