Filtered output approximation perspective on blind deconvolution

Yu Wang

doi:10.1016/j.ymssp.2026.114633

Filtered output approximation perspective on blind deconvolution

2026-07-01

1

PRE

AI

DOI：10.1016/j.ymssp.2026.114633

Original

Original request for help

Save

Abstract

En 中文

Blind deconvolution (BD) methods, such as minimum entropy deconvolution (MED) and maximum second-order cyclostationarity blind deconvolution (CYCBD), are widely adopted in rotating machinery condition monitoring to extract defect-induced repetitive transients under noise and interference. However, the implicit filtered output preference induced by BD remains insufficiently understood, and resulting optimal outputs may deviate from intended fault-related components. This article establishes a filtered output approximation (FOA) framework by introducing a relaxed optimization problem to explicitly interpret the preferred structure of optimal filtered outputs. Under this perspective, MED is proven to drive optimal filtered outputs to a single impulse, and optimal solutions of CYCBD may not correspond to genuine cyclostationary filtered outputs. To mitigate such ambiguity, an improved method, termed uniformity-regularized CYCBD (UR-CYCBD), is proposed by incorporating a uniformity regularizer, which suppresses non-cyclostationary patterns and enables more reliable identification of fault-related components. Moreover, a numerical example and a bearing run-to-failure case study demonstrate the effectiveness of UR-CYCBD.

Keywords:

Blind deconvolution

Filtered output approximation

Minimum entropy deconvolution

Maximum second-order cyclostationarity blind deconvolution

Filtered output approximation perspective on blind deconvolution

Abstract

Journal

Organization

Cited Papers