Towards Applicable Unsupervised Signal Denoising via Subsequence Splitting and Blind Spot Network

Abstract

Denoising is a significant preprocessing process, garnering substantial attention across various signal-processing domains. Many traditional denoising methods assume signal stationary and adherence of noise to Gaussian distribution, thereby limiting their practical applicability. Despite significant advancements in machine learning and deep learning methods, machine learning-based (ML-based) approaches still require manual feature engineering and intricate parameter tuning, and deep learning-based (DL-based) methods, remain largely constrained by supervised denoising techniques. In this paper, we propose an unsupervised denoising approach that addresses the shortcomings of previous methods. Our proposed method uses subsequence splitting and blind spot network to adaptively learn the signal characteristics in different scenarios, so as to achieve the purpose of denoising. The experimental results show that our method performs satisfactorily on both single-sensor and array signal denoising problems under Gaussian white noise and Impulsive noise. Moreover, our method is also verified to be effective on some array signal processing problems of Direction of Arrival (DOA) estimation, Estimated Number of Sources, and Spatial Spectrum estimation. Finally, in the discussion experiments and generalization experiments, we demonstrate that our method performs well across a wide variety of array forms and degrees of signal correlation, and has good generalization. Our code will be released after possible acceptance.