I observed that while the HR drops steeply, the values shown on the PM3 monitor could differ substantially from those on my Polar M460, although both devices use the signal from the Polar H10 chest sensor. The M460 values dropped more steeply than those on the PM3. This suggests that the receiving device adds a dynamic of its own. In stationary situations the two devices agree within 1 bpm.
A non-optical heart rate sensor, e.g. the Polar H10, measures the time intervals between the R-R peaks in the electro-cardiogram. From these intervals, the HR is calculated. This so-called beat-to-beat HR varies much more than what is usually shown on the display of the receiver.
The figure below shows a typical track of R-R intervals from a person in rest. There are long-term and short-term variations. This pattern is called Heart Rate Variability. HRV is an important metric of the cardiovascular condition of an individual : lager variations → higher fitness.
The short term amplitude in this graph amounts to about 100 msec. The average R-R interval is about 1200 msec (i.e. HR=50 bpm). Hence the short term intervals vary, roughly, between 1150 ms (=52 bpm) and 1250 ms (=48 bpm). Such fast HR-variations of some 4 beats are annoying when shown on a display. Most users prefer to see stable values. Hence all devices use some form of smoothing by time averaging: some more - some less. The difference in smoothing probably explains the discrepancy between the PM3 and the M460 when the HR changes quickly. Polar sells a few professional watches that enable to export the raw, unfiltered R-R intervals, but my M460 does not have this feature. However, time-averaging is done over a window of a few seconds and is unlikely to cause the 10 sec delay in the start of the recovery.
See https://support.polar.com/en/support/ti ... s_Filtered for more details about applied filtering.
All my recovery curves start with a horizontal section. I searched the sports science literature for confirmation and, possibly, explanations. I found a very interesting paper that contains a figure from unfiltered 'beat-to-beat' data showing the same features.
The paper deals with modelling the dynamics of the recovery. It seems that the major factor controlling the HR is the lactate content of the blood. The rise in HR during the exercise reflects the increase in lactate. After stopping the exercise, it takes several seconds before the lactate content starts to drop. Similarly, after the beginning of an exercise from rest there is a lag time before the HR starts to rise. Using blood lactate kinetics, the paper could accurately fit the observed HR recovery curve.
It is known that training can reduce the speed with which lactate is removed from the blood. This should result in a faster HR recovery. Hence HR-recovery reflects not only exercise intensity, but also effects of cardiovascular training.