Table I. Filter-Based Techniques for Extraction of Respiratory Signals.

Bandpass filter to eliminate frequencies outside the range of plausible respiratory frequencies [132]. Use (ensemble) empirical mode decomposition to extract a respiratory signal as either one particular oscillation mode (intrinsic mode function, IMF) or the sum of the IMFs indicative of respiration [170], [193]. Decompose signal using the discrete wavelet transform to reconstruct the detail signal at a predefined decomposition scale [59], optionally with automated selection of the mother wavelet [77]. Extract respiratory oscillation using principal component analysis (PCA) [144] or singular spectrum analysis [134] after identifying the periodicity using singular value decomposition. The use of PCA has been refined using multiscale PCA [137], and modified multiscale PCA [140], in which wavelet decomposition was combined with PCA. PCA has also been applied to intrinsic mode functions extracted using ensemble empirical mode decomposition [157], [158]. Extract the instantaneous amplitudes or frequencies of cardiac modulation using the continuous wavelet transform [41], the Teager-Kaiser energy operator [204], variable frequency complex demodulation [28], [67], the Hilbert transform [129], or the synchrosqueezing transform [79]. Filter using the centered correntropy function [90]. Decimate by detrending the signal, low-pass filtering to eliminate frequencies higher than respiration, and resampling at a reduced sampling frequency [85] of 1-2 Hz [10]. Extract an electromyogram signal from the high-frequency content (> 250 Hz) of the ECG caused by the activation of the diaphragm and intercostal muscles during respiration [95].