The Apple watch is a giant, heavy mass of a watch that moves with the inertia of its’ high center of gravity and fashionable, comfortable wrist straps. The ideal watchband for an accurate heart rate monitor should be flexible and stretchy. The flexibility and stretchiness of the watchband allows the sensor to follow the movement of the arm, expand and contract with your muscles under constant tension, and dampen vibration. The Apple watch does not do this, and how could they? It is impossible with a metal or rubber strap – they are simply not elastic. Along with the high center of gravity there’s too much movement on the skin for an accurate measurement. It looks pretty though. We spent a lot of time on this at Scosche to keep a low center of gravity and test the right materials to get an accurate, consistent signal.
Apple is correct about green light being absorbed by the hemoglobin in blood, however certain pigments in darker skin tones can impede green light. The Valencell PhD’s (whose Performtek sensor technology is in Rhythm+) discovered yellow light is less sensitive to skin pigment. The best optical sensor will have a combination Green + Yellow light for the most accurate signal on all skin tones.
It is now clear that the new Apple Watch uses dynamic algorithms to compensate for the design errors in points 2 and 3. A dynamic algorithm changes the colors, intensity, and sampling rate based on the signal integrity measured by the photosensor(s). The best algorithms in the world cannot make up for poor signal quality. A fixed algorithm on the other hand uses the optimal intensity, sampling rate, and a color usage and these parameters are locked from the moment the sensor is powered on. In my experience with Scosche’s testing, the engineers and other smart people in the room have all agreed using dynamic algorithms are a bad idea and prone to infinite loops like a dog chasing it’s tale. The human body can provide a dynamic response with endless parameters, which can change at the same time, all the time. There are simply too many possible scenarios for the algorithms to figure out what is going and to be accurate in all situations, especially taking in to account the first 3 hardware hurdles mentioned above.