Denoising Biomedical Signals with Wave-Shape Functions

The adaptive non-harmonic (ANH) model considers amplitude and frequency-modulated (AM-FM) signals as a superimposition of non-sinusoidal oscillations. This oscillatory pattern is described by a wave-shape function (WSF). This model is particularly suitable for the task of processing biomedical signals, as they are generally composed of various complex waveforms. The WSF is described by its Fourier expansion and the optimal number of harmonics is chosen adaptively for each signal. We show the utility of the ANH model by denoising simulated pulsewave signals from different sections of the circulatory system. Our signal denoising procedure outperforms denoising strategies based on thresholding the short-time Fourier transform (STFT) for all considered input noise levels.