Leak noise filtering method for water supply pipeline based on main frequency-band matching and VSS-LMS

A four-part diagram illustrating the leak noise filtering method. The top left shows the raw input signal and its spectrogram. An arrow points down to the MFB matching process, displaying the power spectral density and cumulative energy graphs. An arrow points right to the desired signal generated via band-pass filtering. A final arrow points up to the clean output signal and spectrogram processed through the VSS-LMS adaptive filter. Leak noise in water supply pipelines is a complex random signal, hindering the application of adaptive filters like least mean square (LMS) which typically require a known desired signal. This paper proposes a novel filtering method based on main frequency-band (MFB) matching and a variable step-size (VSS)-LMS algorithm. The method utilizes power spectral density (PSD) to adaptively identify the leak frequency band and generate the requisite desired signal. Additionally, VSS-LMS balances convergence speed with steady-state error through dynamic step-size adjustment. Simulations and field experiments demonstrate that the proposed method outperforms wavelet denoising, empirical mode decomposition, and singular value decomposition in noise suppression and convergence speed. It effectively extracts leak noise under strong environmental interference and varying signal-to-noise ratios. This method enhances leak detection accuracy and offers significant practical value for smart water distribution networks.