Whether a sensor can provide reliable data depends on the target information, the algorithm deployed, and the specific details of a given use case. Optical biosensing has become a mainstream feature in smart watches and fitness wearables. This article will provide an overview of the interaction around photoplethysmography (PPG) data collected by optical biosensors.
PPG measures the reflected (back-scattered) or transmitted light through tissue to investigate the variations in blood volume which occur with each heartbeat. Researchers and physicians have derived information including heart rate, respiratory rate, oxygen saturation, blood vessel viscosity, venous reflux, cold sensitivity, blood pressure, and cardiac output by processing PPG data with different algorithms. Whether the captured PPG data can provide all of these insights reliably depends on many factors.
In a previous article on optical biosensors, we presented the following equation:
where SNR is the signal-to-noise ratio of the PPG data, ILED is the LED current, CTR is the transfer ratio between LED current and photodetector (PD) current, PI is the perfusion index of the subject, and NTX, NPD, and NRX are the noises in the transmitter (LED driver), PD, and receiver (analog-to-digital conversion), respectively.
When we take SNR as a measure of PPG data reliability, we see that the measurement environment in a specific end-use and the underlying PPG system design play fundamental roles in assuring reliability. Consider these parameters:
Whereas poor SNR represents a lack of data reliability in general, other factors such as motion artifacts may affect the reliability of different PPG data-derived information to different extents. To explain this, we shall look at a few examples of information that could be embedded in the PPG data. As the discussion below illustrates, it is useful to examine PPG data in the frequency domain.
The most prevalent application of PPG is for monitoring heart rate, which is also the most reliably obtainable data (Figure 1 conveys average target heart rates by age). The human heart rate is below 3Hz. That it is largely tonal means that it is a narrowband signal and, therefore, relatively straight-forward to recover via algorithms applied to most PPG data from most background noises. In use cases where motion is expected, higher bandwidths may be used to allow algorithms to differentiate true PPG signals from in-band motion artifacts using the features of a true PPG wave shape.
Target Heart Rate
|Age in Years
|Average Maximum Heart Rate in Beats Per Minute
|Target Heart Rate Range in Beats Per Minute
|90 to 153
|88 to 149
|85 to 145
|83 to 140
|80 to 136
|78 to 132
|75 to 128
Because PPG measurements are low cost and non-invasive, many researchers have focused on using PPG data, taken under clinical settings, to diagnose peripheral vascular diseases such as atherosclerosis, where fatty deposits reduce the diameter of an artery.
A common way to process PPG data uses its second derivative, and the resulting waveform is often called an APG (see Figure 2). Over the course of many studies, researchers have categorized the shape of the APG waveform to help diagnose a subject's health and risk of coronary disease. Techniques in these studies include information at frequencies, often from 0.05Hz to 20Hz, which are significantly higher than fundamental heart rate. And because higher frequency components are much weaker, these PPG measurements require much higher SNR with less reliance on low-pass filters for signal conditioning. As examples, simple analyses of APG features include:
PRV, measured using PPG, is similar to heart-rate variability (HRV) and is used to gauge the autonomic nervous system by analyzing the inter-beat interval generated from heart activity. Because inter-beat intervals vary slowly, the reliability challenge for wearable PPG data lies in maintaining good sensor-to-tissue contact over a long period. Furthermore, the sampling rate used to acquire PPG data must be high enough to avoid time-domain quantization, which affects the accuracy of the PRV variable analyzed.
Pulse oximetry, or SpO2, is more complex than the PPG measurements listed above because it involves lights of two different wavelengths. Researchers have proposed many different approaches to remove the effects of noise and motion artifacts, but the problem persists in wearable applications. This is due to the complexity of motion artifacts.
Movements can result in a change in the tissue-to-sensor coupling. It is a fundamental assumption that these two wavelengths share essentially the same optical path. Data becomes unreliable when movements of the subject causes momentary differences between the paths traveled by each light source. Furthermore, motion can also affect the movement of venous blood and other non-pulsatile components along the optical path, thereby altering the characteristics of the tissue under observation.
Here, data reliability requires an algorithm to understand whether the PPG data is clean or corrupted by motion. So, the bandwidth need for reliable data relates more to supporting this classification and what other sensor information may be available besides PPG, rather than on heart rate or the corrupting motion.
Figure 3 is the power spectral density derived from a set of PPG data captured over fewer than 40s at 100 samples per second on an normally ambulatory subject. It has been truncated to show only the spectrum from 0 to 15Hz. The first prominent peak (below 3Hz) corresponds to heart rate. The higher frequency (>3 to 8Hz) spectral energies are dominated by motion artifacts. This illustrates the difficulties of deriving continuously usable data for higher order analyses such as using APG wave shapes. Algorithms must be capable of differentiating clean, usable data from data corrupted by motion or other artifacts.
This article looked at the contributors to PPG data reliability from both a general signal-to-noise perspective and from how useful information may lie in the frequency domain. Designers can improve data reliability by:
With reliable PPG data collected by optical sensors designed into wearable devices, those wearing these devices can get a more insightful picture of their health and well-being.
These articles provided background for this blog post: