Spectrum analysis of non-uniformly sampled signals

The spectral analysis of non-uniformly sampled biomedical signals presents significant challenges due to technical and physiological constraints that limit uniform sampling. This study compares two approaches for spectral estimation: the Non-Uniform Fourier Transform (NUFT) and the Fast Fourier Transform (FFT) applied to interpolated data. The methodology evaluates the accuracy and stability of the spectral features derived from both methods using respiratory flow signals. The respiratory pattern is characterized through the following time series: expiratory time (T_E), inspiratory time (T_I), breathing duration (T_Tot), and tidal volume (V_T), and frequency-tidal volume ratio (f/V_T) where f is respiratory rate. The interpolated methods: linear, spline, pchip, and makima are analyzed. The Results show that NUFT preserves spectral integrity more effectively by avoiding artifacts introduced by interpolation. These findings support the use of NUFT in biomedical signal processing, particularly in the development of robust machine learning models for clinical decision-making.Clinical Relevance-Reliable spectral features are essential for classification systems used in clinical settings. This study emphasizes the importance of preprocessing in preserving those features and demonstrates how NUFT can support early disease detection and personalized patient monitoring by improving the spectral analysis of irregularly sampled physiological signals.