A Programmable Gain Calibration Method to Mitigate Skin Tone Bias in PPG Sensors

Figures

• 

  Figure 1. Schematic diagram of a reflective PPG sensor measurement setup. Light is emitted by an LED, penetrates the skin tissue, and the reflected light from blood vessels is captured by a photodetector (PD) [3].

• 

  Figure 2. Key features of a typical PPG waveform, including the systolic peak (main peak), dicrotic notch, pulse width, and pulse amplitude. These features provide insights into cardiovascular status [4].

• 

  Figure 3. Illustrative examples of different PPG pulse morphologies and their potential association with various cardiovascular conditions. For instance, a “small and weak” pulse can indicate different physiological states compared with a “normal” pulse [5,6].

• 

  Figure 4. TI AFE 4400 commercially available board to use as a pulse oximeter sensor [23].

• 

  Figure 5. Experimental setup showing the 3D-printed enclosure designed to minimize ambient light and motion artifacts. (Left) Exterior view of the enclosure. (Right) A participant’s hand is inserted into the enclosure, with their finger in the PPG clip sensor connected to the TI AFE4400 EVM.

• 

  Figure 6. Illustration of the skin pigmentation extraction process. An image of the participant’s hand is processed using a k-means clustering algorithm to identify the dominant RGB color value, which serves as a quantitative measure of skin tone.

• 

  Figure 7. Example of PPG signal segments from Subject 3 processed with different low-pass filter settings (e.g., 500 Hz and 1 kHz). The choice of filter parameters can impact the amplitude and morphology of the resulting waveform.

• 

  Figure 8. Python code snippet illustrating the process of reading the SQLite survey database (.db file) into a Pandas DataFrame for further manipulation and merging with physiological data.

• 

  Figure 9. Representation of the filtered Pandas DataFrame containing subject IDs and their corresponding mean RGB skin pigmentation values, extracted from the survey database and image processing pipeline.

• 

  Figure 10. Python code snippet demonstrating the directory traversal and file filtering logic used to locate and parse relevant CSV data files generated by the TI AFE4400 evaluation software.

• 

  Figure 11. Python code snippet illustrating the assembly of processed TI AFE4400 signal data (e.g., AC/DC components), calculation of SpO2 using the empirical formula, and subsequent merging with the survey-derived skin pigmentation data.

• 

  Figure 12. Overall system architecture for the PPG sensor with automated one-time gain calibration. The process is divided into an initial “set-and-hold” phase, where the gain is fixed based on skin tone, and a continuous measurement phase that uses the calibrated gain.