How to design an optical heart rate sensor into a wearable device’s wristband
Techniques for isolating the signal and minimizing the noise
The basic problem to solve in a PPG on a wristband is to maintain the integrity of the desired signal while minimizing the effect of the various noise sources.
The most important method for reducing the amplitude of motion noise is mechanical: the sensor needs to maintain an unchanged position on the skin, as even small movements in relation to the skin generate a large motion signal. The band should be worn with a comfortable tight fit. Close attention also needs to be taken to the placement of the sensor on the wrist: it should be approximately two fingers’ width from the wrist joint. When on or close to the wrist joint, the sensor generates a lot of motion noise. In addition, this area is characterized by low perfusion, rendering the PPG signal especially weak.
The mechanical design must of course take account of the vast variations in the human body. Wrist diameter, wrist curvature, perfusion depth, wrist hair density and color, and skin color all vary and affect the optical signal differently. Even tattoos interfere with the reflection of the sensor’s light. A general recommendation is that a small device will fit a small and a large wrist, but a large device will not fit onto a small wrist. Beyond this, each OEM must experiment with its own choice of wristband material, size, profile, and shape, and expect to spend much time balancing the competing demands of the PPG sensor’s performance on the one hand, and the dictates of style, fashion, and aesthetics on the other.
Wristband designers will also find that motion noise is by no means the only noise source interfering with the PPG signal – the effect of optical cross-talk, for instance, must also be managed. When the user has dark skin, the sensor automatically increases the brightness of its green LED, since dark skin attenuates green light much more than light skin does. At high brightness, however, there is a risk that cross-talk could saturate the sensor. Cross-talk occurs when LED light reflected from the inner and outer surfaces of the sensor’s (glass or plastic) cover reaches the photodiodes, without having passed through the user’s skin.
Cross-talk cannot be entirely eliminated, but the wristband may be designed to keep it below safe levels. Users of the ams AS7000 optical biosensor system-on-chip can take advantage of optical simulations (ray-traces) performed by ams, in order to generate a model of their intended design’s optical performance (Figure 1).
Figure 1 A rendering of an HRM sensor’s mechanical design (left) and its simulated ray-trace (right) is shown.
To simplify the effect of the model, Figure 2 shows only those rays which actually reach the sensor after reflection from the cover glass.
Figure 2 Simulated ray-traces show interference in a sensor design with a large air gap and thick glass (left) and a narrow air gap and thin glass (right).
These models show that the designer of the wristband has broadly two options for keeping optical cross-talk down to safe levels: either to make a PPG sensor with a very small airgap; or add an optical mask.
The experience of ams’s customers is that the more effective option is the first, keeping the size of the air gap to a minimum. This is because an optical mask reduces signal strength, making the sensor more vulnerable to other sources of noise, such as motion noise, and also increases the LED’s power consumption.
Cross-talk is not the only source of optical noise: bright sunlight also has the potential to reach the sensor’s photodiodes by passing through the skin. In its AS7000 biosensor, ams uses an integrated optical filter to reject most of the non-green portion of sunlight. But since the sensor’s LED is green, this filter allows green light, including the green portion of sunlight, to pass through.
To combat the risk of interference from (the green part of) sunlight, the AS7000 modulates the LED’s light emissions, with matching demodulation at the photodiodes. Digital circuitry in the sensor can then cancel out the non-modulated optical noise caused by sunlight.
Light modulation also enhances the system’s electrical performance: the noise performance of the SSoC’s operational amplifier only needs to be optimized for the modulation frequency. This means that 1/f noise can be ignored, as it contributes noise mostly at frequencies below the modulation frequency.
In the presence of the very small signals produced by a PPG, it is also essential that the electrical design be carefully tuned for the extraction of signals in the frequency band of interest (typically 0.5–4Hz, representing a heart rate of 30–240bpm). At the same time, noise from peripheral components must be minimized.
In an AS7000 device, a software algorithm converts the PPG signals into a heart rate measurement. In addition, it cancels out motion-induced signals identified by an external accelerometer: the accelerometer provides only motion signals, while the PPG consists of both motion and heart-rate signals. This allows the motion-induced portion of the PPG signal to be subtracted, leaving only the heart-rate portion.
The operation of this ams algorithm is more complex than this description would suggest, since it has to allow for the full range of operating conditions. For instance, it is more common than might be imagined for the motion signal to be an exact harmonic of the heart rate signal. It seems that some people walk in time to their heart beat! The ams algorithms run on the AS7000, allowing the host processor to sleep in the intervals between read-outs.