Feasibility of Radar Vital Sign Monitoring Using Multiple Range Bin Selection

Figures

• 

  Figure 1. Schematic representation of the radar range bins’ distribution along the human body. The radar device is located at the foot end of the bed on a tripod. Range bin borders are represented with green lines. The distance between adjacent range bin borders ΔR is 5 cm (figure not to scale). This cartoon example shows how signals can be recorded separately from sections of the legs (bins #1–#3), the abdomen (bin #4) and the thorax (bin #5).

• 

  Figure 2. A 15 s time window showing the displacement in multiple range bins, estimated from the radar phase signal. The displacement values are represented by the coloring (see colorbar). One row of the figure corresponds to one range bin. The red framing shows a group of range bins which coherently display a clear breathing motion. The breathing motion can be seen by the periodic change of light and dark colors, which correspond to high and low displacement values, respectively. To make the breathing motion visible in the lower amplitude range bins as well, the displacement values are cut off above 0.1 mm and below −0.1 mm.

• 

  Figure 3. Process of going from time series (A) to persistence diagram (C) using Vietoris–Rips filtration. Before the filtration can be applied, the in-between step of time delay embedding is performed using embedding parameters τ = 10 and m = 3 (B). The embedding shows a loop which is characteristic for periodic signals. The persistence diagram computed with Vietoris–Rips filtration has a single one-dimensional feature with high lifetime, corresponding to the loop of the embedding.

• 

  Figure 4. Example for sublevel set filtration. (A) shows the input time series. For the sake of a clear visualization, a band-pass filter has been applied on the signal. By computing the sublevel set filtration, we obtain the persistence diagram presented in (B). The grid lines in the plots show which parts of the time series are represented by the points of the persistence diagram.

• 

  Figure 5. Flowchart of the breathing bin selection algorithm. The steps for the Vietoris–Rips filtration-based processing are highlighted with green, while the steps for the sublevel set filtration-based processing are highlighted with red.

• 

  Figure 6. Flowchart of the heartbeat bin selection algorithm. The steps for the sublevel set filtration-based processing are highlighted with red.

• 

  Figure 7. Range profile of one participant for breathing and heartbeat activity. The horizontal axis indicates the time from recording start, while the vertical axis shows the distance from the radar device (bottom is closer, top is further away). Yellow color indicates selected range bins for breathing (top panel) and heartbeat (bottom panel).

• 

  Figure 8. Breathing rate of one participant throughout the night; comparison between radar and PSG. The (top) panel shows the computed breathing rates. The (middle) panel shows the difference between the two sensors’ rates throughout the night and a summary statistic with the mean absolute percent error (MAPE). The (bottom) panel is the range profile as already presented in Figure 7. The x-axis is shared between the three panels; therefore, to reduce visual clutter, the labels are only displayed for the (bottom) panel.

• 

  Figure 9. Bland–Altman plot comparing the breathing rates computed using PSG and radar data from all six participants. The horizontal axis shows the average of the two rates, while the vertical axis shows their difference, both in units of breaths per minute (BPM).

• 

  Figure 10. Heart rate of one participant (same participant as in Figure 8) throughout the night; comparison between radar and PSG. The (top) panel shows the computed heart rates. The (middle) panel shows the difference between the two sensors’ rates throughout the night and a summary statistic with the mean absolute percent error (MAPE). The (bottom) panel is the range profile as already presented in Figure 7. The x-axis is shared between the three panels; therefore to reduce visual clutter, the labels are only displayed for the (bottom) panel.

• 

  Figure 11. Bland–Altman plot comparing heart rates computed using PSG and radar data from all six participants. The horizontal axis shows the average of the two rates, while the vertical axis shows their difference, both in units of beats per minute (BPM).