January 15, 2020
| By: Ian Chen
Executive Director, Industrial & Healthcare Business Unit, Maxim Integrated
Getting reliable sensor data is complex. For example, an optical heart-rate sensor is actually sensing the changes in an electric current. Heartbeats cause the volume of arterial blood to change synchronously with each pulse. The change in volume changes the amount of light absorbed and reflected as it travels through live tissue. When that light exits the tissue and enters a photodetector, it changes the output current. By creating a sensing system that carefully sets up a proper series of “dominos,” a current sensor becomes a heart-rate sensor.
This is how most sensors work. Sensors generally make esoteric electrical measurements (capacitance, impedance, current, voltage). But through an elaborate system construct, a physical event of interest (acceleration, pressure, footsteps, distance) is made to change that measurement. Knowing the system construct, we could then interpret the change into a physical parameter, all the while assuming everything else in the sensing system stays constant or at least is well controlled.
But what if the dominos aren’t all under the designer’s control?
This article will illustrate the challenges of sensor data reliability using an optical heart-rate sensor as the example. However, the complex nature of sensor quality applies to most sensors, not just optical ones.
Understanding the Optical Path
The optical path is the path light travels from the source (emitter) to the detector (receiver). The path spans across one or more media and any change in those media could interact and affect the light characteristics at the detector. As such, the received light encapsulates all changes in the media along the entire optical path.
For an optical heart-rate sensor, light comes from one or more LEDs and is aimed at live tissue. Before arriving at the tissue, the light has to travel through air and sometimes a layer of cover glass. Air, cover glass, and the surface of the tissue are three different optical media. Similarly, the tissue is not homogenous and could be modeled as successive layers of optical media with different refractive indices.
At the interface of any two media with different optical properties, light could be absorbed (attenuated), reflected (scattered) into the first medium, or transmitted into the second medium. Figure 1 simplistically shows the many paths light can take after leaving the LED as it make its way through different media.
Figure 1. Example of the paths that light can take after leaving the LED and making its way through different media.
We are not concerned with all the possible optical paths; only those optical paths that end at the photodetector are relevant to optical sensing. Heart-rate sensing works because one of these media, the capillaries, changes volume over time synchronous to heart rate. The change affects the amount of light absorbed and reflected. Optical system design must make sure that the path most light takes to go from the LED to the photodetector also interacts with the capillaries.
Data reliability is compromised when light can reach the photodetector without interacting with the capillaries or when something along the optical path changes unexpectedly. Notably, every media above the cover glass is outside the control of the designer, and some optical properties could even change over time. Interpreting any change in the photodetector current as a change in heart rate would be a simplification. The following section takes a look at changes in the media traversed by the optical path and how they could affect the photodetector current.
Impact of Dirt on the Cover Glass and Other Attenuators
Dirt and grime on the cover glass may be unavoidable in practical applications. They largely serve to attenuate the photodetector current by reducing both the LED light that may reach the tissue and the light received by the detector. For heart-rate sensing, the important information is carried in the periodicity and not the overall amplitude of the signal. So, as long as the emitter is strong enough, some attenuation should not result in any loss of information. However, if the sensing configuration uses more than one LED or multiple wavelengths, it is possible that light intensity for each LED and/or wavelength would not be affected by the same proportion.
This discussion can be expanded to cover other light-attenuating factors outside of the designer’s control. These factors include hair, skin pigment, and color changes of the cover glass. Each medium serves to attenuate light passing through it, in both directions, and each may affect some wavelength of light more than others.
Changing Air Gap or Path Length
Figure 2 depicts an optical heart-rate sensor. The distance between the skin surface and the optical components (light source and photodetector) is often called the air gap.
Light could enter the photodetector after being scattered by the skin surface (Ir) or after it has traversed through some layers of tissue (I). Only the optical path of I could have been affected by changes in the tissue and could, therefore, contain useful information. Consequently, a sound heart-rate sensor must minimize the magnitude of Ir by tending to the separation between the light source and the photodetector and their associated mechanical housing design. In doing so, the designers must make an assumption on the size of the air gap, which in practice they cannot fully control. When the air gap becomes larger, both I and Ir will be weaker and harder to detect because the photodetector is now farther away. At the same time, proportionately more light directly reflected off the skin surface can now enter the photodetector. Both factors degrade the signal-to-noise ratio of the sensor data, making any derived information less reliable.
Figure 2. In the optical path of a heart-rate sensor, an air gap can degrade signal-to-noise, making it harder to attain reliable sensor data.
Furthermore, unlike dirt and grime, the air gap can change periodically as well; for example, when subjects are exercising vigorously, the mechanical coupling between the sensors and the target tissues could change with rhythmic motion. This will introduce a different periodic change in the photodetector current not controlled by capillary pulsation. As a result, heart-rate detection algorithms may become confused.
More than Air in the “Air Gap”
In many wearable applications, water (in the form of, say, sweat or rain) could be present in the “air gap.” The resulting combinations and variations are numerous, but we could consider some generalities. When the sensing target is live tissue, which is mostly comprised of water, having water in the air gap actually narrows the difference in refractive index between the air gap and the target. This should allow a proportionately greater amount of light to be transmitted into the tissue, strengthening the sensing mechanism.
Addressing Biological Change in the Subject
An inherent truth in long-term biosensing and monitoring is the fact that the target (living tissue) will grow and change. For example, increasing pressure on the tissue could pinch off blood flow and diminish or otherwise compromise the detected signal at the photodetector. Likewise, inflammation and swelling on the tissue changes the optical path of the sensor.
Often, these changes are not challenges, but, instead, the objectives behind long-term optical biosensing. Being able to capture biological changes by monitoring changes to light through the tissue is what makes optical biosensors a useful tool not just for heart-rate monitoring, but as the foundation for many different types of non-invasive health and wellness monitoring. Nevertheless, when monitoring for one set of biological change, designers have to remain cognizant of other potential biological changes and how they may interact with the optical path to provide false signals and make the sensing data less reliable.
Maximizing Signal Path Performance
With everything that could affect the optical signal, it becomes increasingly important that the part of the optical data path that is under the designer’s control gives the best signal-to-noise performance. A high-performance design makes discerning the reliability of the sensor data easier.
For example, any light which makes its way to the photodetector but did not enter the target tissue, not just the light emitted from the LED source, adds to the noise of the biosensing signal. Some integrated analog front-end (AFE) devices, like the MAX86140 and the MAX86171, sample any ambient current on the photodetector asynchronous to the LED light source and subtract them from the photodetector current. Indeed these AFEs even anticipate how ambient light conditions could change in typical use cases so that designers can trust that their effects will have little impact on the biosensing signals.
Know When Sensor Data Is Not Reliable
With many things along the optical path potentially changing, designers may include other mechanisms to detect potential changes to ensure the reliability of sensor data.
One coping strategy is to use sensors unaffected by optical path changes to monitor when such changes might occur. For example, accelerometers could notice moving targets and pressure sensors could sense increased compression. Because these sensors use a different modality than optical sensors, they can at least warn designers that sensor data may be compromised and, in some cases, could even be used to help mitigate against using data with different optical paths, making the result more reliable.
Another strategy is to use more than one light frequency as lights of different colors are attenuated differently by each optical medium. A change in the optical path would, therefore, attenuate or scatter each colored light with different proportions. By comparing the spectral composition of the emitted and the received light, designers can get information on how the optical path has changed.
Sensor data are inputs to algorithms which interpret the data and translate them into meaningful information. Algorithms could use known physical models or the context of the use case and historical sensor data to determine if new data has become unreliable. In a future article, I’ll take a more quantitative look at the interactions of sensor data and algorithms to provide reliable and actionable information to device end users.