The original plan has been to be transparent about our claims. Nobody tells of their methodology, so I’m just going to spill the beans as much as I can on ours. This has been said over and over at trade shows.
A bending bridge on the inside face of the arm isn’t +/- 2% but if you want to it means claiming three assumptions at least.
- The Cleat interface NEVER changes EVER (not true, sprinting tends to change… Garmin’s PCO offers you information on this on Vector)
- Someone cannot twist their ankle causing a torque along the axis of the crank arm (Again, use Vector and look at Pedal center offset, you’ll see that’s not true)
- The calibration occurs exactly where a user puts their feet. (Speed play has a shop kit to specifically change the width!)
So, since all of these are false, if you test against them you can literally get +/- 10% error over a 40 mm distance. I’ve ripped apart commercial products and tapped into their strain gage with professional instrumentation equipment. 10% is kind for the scientific side of things. The real world usage will be less but you’ll get more error in sprinting or out of seat hill climbs.
What’s the risk? Well, if you are training with power, changing your seat, pedal, shoes, insoles, cleats, cleat types… they can all cause a slight change causing this pedal offset to change or causing you to create ankle torque in different positions. This means you could lose 10% off your FTP with a single change of these interface parts, or you could gain. The reality is you didn’t, the reality is your meter couldn’t compensate.
Below are the results of a Cannondale SL crankset with our product on it that I calibrated for one of the managers bike that has an SRM on it. You can see the bike on my twitter feed. (Yes you can put a hollowgram on a specialized transition, don’t know why people say you can’t)
Offset (mm) Torque (N-M) Bend Only Bend + Secondary B (err) B+S (err)
0 0 1.599E-02 1.779E-02 N/A N/A
0 0 1.372E-02 1.684E-02 N/A N/A
0 0 1.655E-02 1.904E-02 N/A N/A
16 2.357E+01 2.180E+01 2.357E+01 -7.5% 0.0%
16 2.357E+01 2.179E+01 2.356E+01 -7.6% -0.1%
36 2.357E+01 2.349E+01 2.357E+01 -0.3% 0.0%
36 2.357E+01 2.350E+01 2.359E+01 -0.3% 0.1%
56 2.357E+01 2.515E+01 2.354E+01 6.7% -0.1%
56 2.357E+01 2.520E+01 2.358E+01 6.9% 0.1%
0 0 -3.890E-02 -4.481E-02 N/A N/A
0 0 -7.215E-03 -8.784E-03 N/A N/A
0 0 -1.415E-04 -7.544E-05 N/A N/A
It’s a little gross to read, but when we turn off our secondary gage we get effectively what is on the market by two other companies. One of them offers a high speed torque reading method to smart phone. Hang a weight at different offsets and see what you get, or better yet, put your pedal in backwards and see what you get. Our torque sensing is 0.1% on this particular unit. Some are better (I’ve seen 0.08% error over the calibration). The worse factory calibrated unit when tested with other weights (random sample points) gave about 0.3% error, but that might be me being conservative.
So what else? Rotational algorithim. We’re using our 32768hz crystal for timing, but our accel runs slower. It runs through an algorithim (designed by the inventor and father of all footpods) and as a result gives us 0.4% maximum error. That’s a three sigma outlier (three standard deviations away, for I believe if my stats knowledge doesn’t fail me, 99.9% of the time). We’re looking at doing a method that will improve that via the firmware update, but right now we’re happy with that.
The last puzzle piece, and it saddens me we didn’t have to implement this calibration out of the gate was the active thermal compensation. We have the physical hardware and with user installation the plan the idea was that the device would learn it’s thermal calibration over time via user zeroing. We wanted to bring it in house and pre-cal them, but there wasn’t time. It’s still a project for factory install, but the algorithim will still come in via OTA. However, everyone else got to claim 2% beforehand without active, so that’s where we are. That’ll be fixed in a few months though.
Okay, So how does everyone else measure their error… no seriously I’m asking, because NOBODY is talking. Ours is a summation of errors (torque and rotational) + buffer room for crystal drift or stupid things. Yes, I said we have buffer on that claim.
One says “scientifically proven”, several just seem to have went with “sure stain gages, 2%… ya whatever”, and I read once that one spider based unit in the wattage google group was the maximum error from calibrating a spider in 4 positions.
Their is only one crank arm unit on the market that straight up doesn’t need correction. We aren’t compatible with them – so you can figure out who that is.
