Near-field AoA estimation with Complex Convolutional Kolmogorov-Arnold Network

Abstract

The development of 5G and beyond puts forward higher requirements for indoor positioning. However, conventional angle of arrival (AoA) estimation algorithms still rely on the far-field assumption, which is not applicable and may lead to a loss of accuracy. In this paper, we investigate the near-field AoA estimation problem and propose the complex convolutional Kolmogorov-Arnold network (CCKAN). With complex convolution, our method enhances the intrinsic relationship of signal amplitude and phase by simulating complex-valued operation in the convolution. We also introduce Kolmogorov-Arnold network (KAN), a deep learning model with better interpretability, as a feature extraction block. Results show that CCKAN achieves higher accuracy than other baseline methods in the near-field AoA estimation task. Furthermore, the model also exhibits remarkable reliability in both near-field and far-field cases.