Moving window signal concatenation for spectral analysis of ECG waves

In this paper we postulate a replacement for the averaging of successive waves spectra, by the concatenation of selected type waves (P, QRS or T) in a moving time window. The window size is selected as a result of compromise between the spectrum resolution (high for long window) and temporal response (fast for short window). To avoid the influence of a border effect, the order of concatenated waves is optimized with regard to the maximum smoothness of resulting signal. The smoothness estimate is based on the similarity of the first derivative of the adjacent signals. Our results for a 10s window shows a considerable drop of estimated P-wave bandwidth from 151Hz to 107Hz (CSE Dataset3), which seems to be much more reasonable considering the physiological limitations of ECG signal generation. Similar results were observed for QRS and T waves.