Is this a sensible limit? Well I tried various values, and 48 seems about the best. Here's teeth vs speed for a 4.25 hour 125km ride. After I removed all the data points with either zero speed or zero cadence, I still had over 13,500 data points:
So, 48 it is.
Next question, how does the actual data compare to modelled data? I created a matrix of teeth vs cadence, where the intersection is speed (in kph):
Now, if I scale and overlay them (I've drawn in lines to represent what my cassette offers):
Another thing you might notice is that that the lines of cadence are quite distinct, which we see zooming in:
What we're seeing here is that the GSC-10 only records integer values for cadence.
The cloud we see is thus where two imprecisions from sampling frequency and data recording surface:
- cadence to 1 rpm only
- speed and cadence recording at 1Hz intervals
If, in the example here, the number of teeth boils down to 17 ± .25 then that works out to be about 98% accuracy.