Data-selective uniform probability density function for adaptive filtering

• Published in: Digital signal processing Vol. 158; p. 104948
• Main Authors: Wang, Qizhen, Wang, Gang
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.03.2025
• Subjects: Adaptive filtering; Data-selective; Least mean square; Recursive least squares; Uniform; Adaptive filtering; Uniform; Recursive least squares; Data-selective; Least mean square
• ISSN: 1051-2004
• DOI: 10.1016/j.dsp.2024.104948

Data-selective (DS) adaptive filtering algorithms based on the Gaussian probability density function (PDF) can reduce the computation burden while preserving the estimation accuracy at a suitable data update rate. When the data update rate is relatively low, its performance may degrade. In this paper, a new discovery about DS strategies is presented. We demonstrate that under the uniform PDF, the DS strategy not only lessens computational burden, but also reduces the steady-state mean square deviation (SS-MSD) of adaptive filtering algorithms, even at substantially low update rates. A novel probabilistic model to describe the actual PDF faced by uniform-based DS adaptive filtering algorithms is proposed, termed the DS-uniform PDF (DS-UPDF), along with its mathematical property analysis. Furthermore, corresponding data-selective uniform least mean square (DS-ULMS) and data-selective uniform recursive least squares (DS-URLS) algorithms are developed. Utilizing properties of the DS-UPDF, respective analyses of mean value and mean square stability are also provided in detail. Additional tests are conducted using data-selective uniform least mean fourth (DS-ULMF) and data-selective uniform generalized maximum correntropy criterion (DS-UGMCC), confirming the universality of performance enhancement phenomena in DS adaptive filtering under uniform distributions. All empirical simulations align well with theoretical predictions.