Real-time signal processing by adaptive repeated median filters

摘要

In intensive care, a basic goal is to extract the signals from very noisy time series in real time. We propose a robust online filter with an adaptive window width, which yields a smooth representation of the denoised data in stable periods and which is also able to trace typical patterns such as level shifts or trend changes with small time delay. Several versions of this method are evaluated and compared with a simulation study and on real data.