Achieving Clinical-Grade PPG via Wearables
Author : Andrew Burt, Analog Devices
10 August 2023
Figure 1: Measuring HR and SpO2 using a wrist-worn device
Heart rate (HR) and blood oxygen saturation (SpO2) are quickly moving from being ‘desirable’ and going firmly into the ‘expected’ feature sets of health and fitness wearables.
A consequence of this transition has, however, been a reduction in the quality of readings derived. This comes from some sensor manufacturers making questionable claims about the accuracy of their products in the rush to meet the market demand for these features. While the accuracy of readings may not be critical in everyday wearables, measurement integrity must be beyond doubt in clinical-grade wearables.
A key challenge for designers is how to make high-quality HR and SpO2 measurements in a way that doesn’t place a heavy drain on the battery reserves of the device. Conventional methods of taking optical readings will waste precious power. Consequently, a new approach is needed.
Figure 2: PPG measurement using an LED and a photodiode
Understanding PPG
HR and SpO2 are measured using an optical technique called photoplethysmography (PPG). A PPG signal is obtained by illuminating the skin using an LED. Changes in the intensity of light reflected from blood vessels below the skin surface are subsequently detected (as shown in Figure 2). This is done using a photodiode that generates a current proportional to the amount of received light.
The current signal is conditioned by the PPG analogue front-end (AFE) before being passed through an analogue-to-digital converter (ADC), then processed by an optical algorithm running on the system’s microcontroller. In principle, a single LED-photodiode pairing is sufficient to make a PPG measurement, and this architecture is common in equipment currently used in clinical settings. However, these devices operate in conditions very different to those encountered in everyday life. Firstly, the patient is relatively immobile, and the measurement is performed using a sensor that is securely fastened to a fingertip. Lighting conditions are relatively constant, which simplifies light detection for the photodiode. Furthermore, power consumption is not a concern, since these devices are usually mains powered.
Figure 3: SpO2 and HR measurement in a clinical environment
By contrast, a wearable device is typically wrist-worn - meaning the level of skin contact is constantly varying, due to the motion of the wearer. There are also personal preferences (like strap tightness) that need to be taken into account. Lighting conditions can vary considerably, depending on location and time of day. Also, since these devices are battery-powered, it is important for the current consumption of the sensor to be as low as possible. This is made even more challenging by the variety of skin tones of different wearers. Darker skin is described as having a lower perfusion index than lighter skin, meaning it requires greater illumination (thereby causing more power to be needed in order for measurements to be made).
Figure 4: Using 2 LEDs to achieve better skin illumination
PPG AFE with a single ADC channel
Increasing LED current, or using 2 LEDs, presents an intuitive way to achieve a higher degree of skin illumination (see Figure 4) since it can illuminate a greater skin area. However, this is a power-hungry arrangement, because LED current accounts for at least 50% of the power consumed in a PPG system - which can be a figure reaching 1mW (depending on the skin perfusion index of the wearer). Overall, this approach is inefficient and detrimental to battery life.
PPG AFE with dual ADC channels
A better way to increase skin illumination is to have a single LED with 2 photodiodes. This can be used to detect a greater amount of reflected light (as described in Figure 5). The advantage here is that the standard 20mA LED current can be reduced to 10mA, while achieving the same level of total photodiode current when compared to using a single photodiode device. In challenging operating conditions (like low skin perfusion and/or when the wearer is moving) where the system algorithm determines that a higher LED current is required, a proportional increase in system sensitivity can be attained. For example, applying the same LED current as in the previous arrangement delivers a 100% increase in photodiode current, thus delivering higher overall sensitivity, albeit at the cost of increased power consumption.
Figure 5: Using 1 LED with 2 PDs for better light detection
PPG AFE with quad ADC channels
Using 4 photodiodes to detect reflected light (as outlined in Figure 6) saves yet more power budget, as the LED needs to draw less current. This architecture requires support of a quad-channel ADC. It delivers higher quality readings, because blood vessels and bones are distributed asymmetrically in the wrist, and having 4 light sensing devices helps mitigate the effects of motion or strap tightness. The probability of detecting light reflected from illuminated blood vessels is also increased. Figure 7 shows HR measured using 4 photodiodes (configured as 2 independent pairs: LEDC1 and LEDC2) with respect to a reference measurement (polar). The wearable needs to ensure that good skin contact is maintained while measurements are being taken. Initially, the wearer is at rest, then after 300s (5 minutes) they begin exercising, thus causing their HR to increase. It is clear that the signals on LEDC1 and LEDC2 deviate from the reference measurement and the benefit of using 2 pairs of photodiodes to capture and combine all of these deviations is apparent.
Table 1 summarises the relative power consumption of each of the architectures previously considered, assuming a 1.6V typical supply voltage.
Figure 6: PPG using 1 LED and 4 photodiodes
Practical quad ADC solution
Addressing clinical-grade (as well as general purpose) wearable devices, Analog Devices’ MAX86177 is an ultra-low power quad-channel optical data acquisition system with both transmit and receive channels. On the transmitter side, it has 2 high-current 8-bit programmable LED drivers that support up to 6 LEDs. On the receiver side, it has 4 low-noise charge integrating front ends that each include independent 20-bit ADCs capable of multiplexing input signals from 8 photodiodes (configured as 4 independent pairs). It achieves a 118dB dynamic range and provides ambient light cancellation (ALC) up to 90dB at 120Hz. Lab-tested samples of this AFE have exhibited an overall root-mean-square error for hypoxia measurement of 3.12%, well inside the 3.5% FDA limit for clinical-grade monitoring.
Conclusion
A major headache for designers of clinical-grade wearables is how to make optical PPG measurements of HR and SpO2 without depleting the device’s battery. A 4-channel ADC architecture, as discussed above, can deliver power savings of up to 60% when compared to the basic single LED/photodiode architecture.
Figure 7: HR readings using 2 pairs of independent photodiodes
Contact Details and Archive...
