Table III. Techniques for Fusion of Respiratory Signals.

Spectral averaging: calculate the individual power spectra of multiple respiratory signals, and then find the average spectrum [110], [114], [196]. Peak-conditioned spectral averaging [110], [114]: only sufficiently peaked spectra are included in calculation of a peak-conditioned average power spectrum. To qualify, a spectrum must have a certain proportion of its power within an interval surrounding the frequency corresponding to the maximum power [51] or the previous BR estimate [114]. Cross power spectral analysis: calculate the individual power spectra of multiple respiratory signals, and then multiply the spectra [117]. Cross time-frequency analysis [168]: use the smoothed pseudo Wigner–Ville distribution to estimate time–frequency spectra between two signals. Time–frequency coherence [168]: used to measure the degree of coupling between two signals. Vector autoregressive (AR) modeling [138]: the poles of multiple AR models (one for each respiratory signal) are calculated. Only those poles that are common to both models, and which fall within the range of plausible respiratory frequencies, are used to extract a respiratory signal. Point-by-point multiplication of signals [109]. Use of a neural network with multiple input signals to identify periods of inhalation and exhalation [32], [44].