Wavelet Analysis and Self-Similarity of Photoplethysmography Signals for HRV Estimation and Quality Assessment

Figures

• 

  Figure 1. Examples of PPG signals: (a) a clean signal with clearly visible peaks; (b) a noisy signal where peaks associated to cardiac cycles still can be recognized; (c) a corrupted signal where no accurate peak detection is possible.

• 

  Figure 2. Mexican hat wavelet function y=ψ(t).

• 

  Figure 3. Algorithm workflow.

• 

  Figure 4. The PPG signal (top) and its corresponding STFT spectrogram (bottom), part of the PPG signal of subject 01 in the TROIKA dataset. Heart rate frequency is growing from 1.2 Hz to 1.9 Hz.

• 

  Figure 5. Continuity filter highlighting curves with bounded rates of change; the horizontal length is adjusted to correspond to 10 s of time and the vertical length corresponds to a change in frequency of 0.3 Hz.

• 

  Figure 6. Filtered spectrogram. Black lines show the curves consisting of local maxima in the spectrogram columns.

• 

  Figure 7. Scaled Mexican hat wavelet and its frequency spectrum for the smallest scale a1 and the largest scale a50 in the chosen scale range.

• 

  Figure 8. PPG signal, the corresponding scalogram computed with the Mexican hat wavelet for the scale range a1,…a50, and the ridge lines in the scalogram. This signal is a part of the red channel PPG from subject 01 in the Welltory-PPG-dataset. The R peaks in the signal are detected as top points of the ridge lines chosen by the filtration Algorithm 3.

• 

  Figure 9. Part of the S3 signal of the PPG-DaLiA dataset.

• 

  Figure 10. Filtration of the RR intervals of Figure 9. Green intervals are kept by the algorithm and red intervals are discarded.

• 

  Figure 11. PPG-DaLiA, Subject S3, interval 0 to 100 s.