That's the point I'm making though. Temperature changes of that magnitude are quite normal. Such as doing a 1-2hr climb in the mountains, where you are unlikely to stop pedaling and zero offset. That's precisely why most major PM manufs these days implement some form of temperature compensation. So you're saying it's acceptable that the SRM user is at fault for not stopping mid-way through a climb to


3) I'm finding it funny though that you're referring to units/tests from 6-12 years ago, for companies that have advanced beyond various faults. A Quarq unit today is dramatically different than one from 2013 or 2009. In fact, in my Stages review, I actually demonstrated exactly this - where a Quarq from 2008 showed the drift in a 20*F swing that I noted above during a simply climb in the mountains in California. In their specific case, they noted that was due to water ingestion in the unit through the battery compartment. But in this specific example, Stages (ignoring leg bias issues for a moment) actually did compensation for temperature correctly.

If I had more space, I'd actually build some sort of walk-in fridge/shack that I could shift the temperature automatically from something like 60*F down to 40*F, or from 80*F down to 50*F. Typical mountain climb decreases. Alas, I luck such space to do that.

---

-
My tiny little slice of the internets: dcrainmaker.com

Not in a matter of minutes, they're not. IOW, you're talking about 0% error at the bottom of the mountain, 5% error at the top 1-2 h later, and very slow drift in between, and that's only for those who 1) live near such mountains, and 2) are too lazy, don't know, or don't care enough about the quality of their data to bother to stop and reset the zero offset.

IOW, I think you're making a mountain out of a molehill (pun intended), a conclusion supported by comparison of measured vs. estimated power of pro cyclists racing (thus not having the opportunity to stop pedaling) up such extended climbs by those who think that such data can tell you who does or doesn't dope. Specifically, although there is the occasional "clinker" the results generally agree to within a few percent. This would not be true if temperature-induced drift were as serious an issue as you make it out to be.


3) I'm finding it funny though that you're referring to units/tests from 6-12 years ago[/quote]
The point isn't to "dis" other brands, but to point out the importance of being able to accurately measure power under non-steady-state conditions, something that remains a significant challenge to this very day.

Ok, so 5% error is OK on an SRM power meter when riding up in weather changing conditions and you don't want to stop to continually calibrate? Got it.

My point here wasn't to set out de-anything SRM. But when folks like yourself start trying to say other power meters aren't accurate because they're not SRM, you just proved my point, and all future power meter accuracy points regarding why SRM isn't the golden standard in one single tidy post.

---

-
My tiny little slice of the internets: dcrainmaker.com

To be fair - it was atleast two decades ago, I've had my SRM for nearly a decade and it was already a 4th or 5th generation SRM.

Speaking of which I really should get a new battery in the Power control, so I feel motivated to use that road bike.

That's true. I think the IV (1993) is the first one that kind of settled on most of the design elements that we still have today?

Yep.

Just for reference, I (and I others I'm sure) have suggested Ray add peak F-V testing into his future power meter test protocols. Ray has responded positively to the suggestion. I guess that power meter accuracy validation by researchers since the last lot was done perhaps a decade or more ago isn't going to enter many published journals soon.

For everyone else...

I suggest people post some peak force-velocity data (correctly acquired) from all the various power meters and let's examine the reliability of their F-V curve. I think makes for a great sniff test, which is why I suggest it be included in any first pass power meter data validity check/test, especially since we have known science wrt to the nature of human performance in such a scenario with which to compare.

The problem many will find when they attempt this is many meters are not capable of reliably capturing such data in the first place, either because of a lack of temporal resolution, or because they do not respond quickly enough to record such data, or the manner in which they measure crank velocity is not up to the task.

As for validity and factors to be considered for static testing (for benefit of Trail), at least the static method for establishing the slope of an SRM is published science and has been demonstrated to be very reliable:
https://www.researchgate.net/publication/227000666_A_static_method_for_obtaining_a_calibration_factor_for_SRM_bicycle_power_cranks


When you examine dynamic testing methods, they are not without issues and I wouldn't exactly be hanging my hat on dynamic calibration as being the be all and end all of power meter accuracy testing. With static calibration measurement error is minimised, it's cheap, easy to do and reliable. At least it's easy to do for meters that permit one to do it. Not all do though (some have one of Robert Chung's favourite features, the missing "9" key). Jim Martin has also mentioned the work done to validate actual field test data collected with (SRM) power meters with wind tunnel data.


Accuracy (and all the things that go along with it) is a variable and what level of accuracy one demands depends somewhat on what applications you intend to use a power meter for. If one is simply using data in the manner they might have used HR for in the past, well you can probably buy whatever meter you like. But if you want to use the data for something more in depth, then choice of meter and the data it provides will matter more. And some meters are better in some scenarios than others.


Andy Coggan's point about the influence of erroneous peak power data might be subtle and many may not consider it immediately relevant to triathletes but as we evolve into a newer paradigm of understanding power meter data beyond what we've been familiar with for the past decade and what more it can reveal about us and to help guide our training (especially individualisation) then accuracy in measuring across the power-duration spectrum matters, even for triathletes.

---

http://www.cyclecoach.com
http://www.aerocoach.com.au

It's not intrinsically difficult -- it's just that it's not the sort of thing that most people have a need to do so I've never set up my analytical routines to automate it. I do it as a one-off analysis. I'd be willing to help you set it up next semester but in the fall semester I'm pretty busy with work (as should be obvious from my meager participation in this thread).

I don't think that any single kind of analysis captures everything you need, which is why testing for accuracy is a pain in the butt. PF-PV (or crank torque - cadence) plots don't, though they came in handy when I identified those anomalies with the first Stages analysis. Basically, there are several ways that a power meter can fail and there are different kinds of analyses that will zoom in on a particular failure mode. I think you're right about failures: nowadays, screwing up torque or force measurement isn't as common as once it was. As torque or force measurement became more reliable, other sources of error began to take their turn as the limit to accuracy. I think the problem with that initial batch of Stages was the cadence measurement from the accelerometers. I've suspected for a long time that the problems the Brim Brothers and the pedal-based PMs were having were less about force and more about (pedal) speed and direction.

Anyway, VE is one example (there are others) of an omnibus test that can help identify deviations from "expected" behavior -- then, once an anomaly is spotted you have to dig into the data to figure out what, why, how, and whether it's something you can do anything about. That is, VE (and some other omnibus tests) tell you when something is off but whether that was due to mis-calibration, or sloppy zeroing, or noisy cadence, or algorithmic deficiencies, or temperature change, or you had to use your brakes to avoid hitting a squirrel, is beyond its abilities. That's the real reason why this stuff isn't automated.

I see the largest barrier to this kind of detailed analysis is the limitations of the ANT+ protocol, that is getting the data out of the units. Stages and pioneer have high speed data that can be obtained through stages software and the pioneer head unit. Both resort to custom wireless protocols to do this. From previous discussions the Quarq units could also output some type of high speed data but AFAIK they dont have a wireless protocol to get the data. SRM has semi-high speed data but only on track units and only using their head units ( Do newish SRMs have enough reed sensors or accelerometer to know position accurately?).

The ant+ protocol could use an update to include a high speed data transmission method to at least 1/16s.

False claims don't flatter you. Temperature doesn't instantly change when riding up a mountain compared to yanking a bike out of a cold chamber, and as others (e.g., Josh L) have reported on the wattage list, many (newer?) SRMs are more stable than the older model testing in that study.


And again, you're drawing conclusions about SRMs without (as far as I can tell) ever subjecting one to the same fairly rudimentary tests you use with other powermeters.

It has been interesting to watch this thread.
I know I am somewhere in that mix of crowd and yet it is hard for me to pinpoint exactly where I stand on what I find as acceptable. For instance I was not too keen on the power difference between the 1st Gen Kickr and the Quarqs that I use. Can I verify the absolute accuracy of my meters at home with my crude methods? No, but of two Quarqs and a PT G3 are all within 4 watts of each other in a static torque test I am okay with them, but the Kickr was 20 watts consistently different and that was too much for me to be comfortable mixing that data in with the power meter data in a training software. Yet, how much difference am I am willing to accept? I don't know to be honest.

I wonder how many power meter owners fall in near to how I feel about the power meter accuracy?

Another view that I have that kind of offsets my thoughts on owning something that is near "gold standard" is that my own training is not gold standard. I have enough flaws in my training routine because what I would like to do and what life has me doing instead of a near perfect lab setting for training. I first have to be honest about my own genetics, training methods and honest effort before I look at the accuracy. In other words I feel this opinion goes along with what I highlighted from your post.

I do like this discussion though as it may help me and others for future use and consideration.

The power meter I am curious about in terms of comparing to SRM would be the Verve Infocrank, but I haven't seen much on it since it came to the market.

As a decent former competitive lifter and thinking about absolute accuracy you can go into almost any gym and randomly pick out a standard cast steel plate that has 45 lbs stamped as number on the plate and then weigh it on legally calibrated scale and it most likely not weigh 45 lbs. The last one I picked out to use for the static torque test weighed 44.3 lbs on a state certified scale in our lab. I trained and won competitions out of many years of using just regular cast steel plates. There was a lot more to training and winning than the few grams of difference. I was fortunate to train in a gym that had certified Olympic plates and bars, but only the women's Olympic team and some of the world class powerlifters were allowed to use that equipment. The rest of us had to use the stuff that the owner bought in bulk and it wasn't dead on the money accurate. Yet good enough for folk like me to use and compete successfully. When we went to decent level competitions the good equipment would be available. If one were at the highest levels and within seconds of winning at a high level then I can see the greater value to a gold standard meter. To people like me that need about a 50 watt bump in FTP not so much. Frankly I need a redo in parents if endurance sports is my desire :-)

Anyway what you posted has me looking at how I view what I own and use and if I am comfortable using that level of equipment. Yep I am.

There are lots of things you can do with a power meter, and general FTP training is one of the least demanding. Riders have been able to train effectively (or sometimes not) with HRMs and before that with wristwatches and a quiet road or hill, so if you're using your power meter like a HRM or a stopwatch you actually aren't asking much of it. I suspect that most riders fall into this category.

A more demanding application is when you're trying to train to do something specific, like maybe sprints on the track. Another is when you're trying to use a mathematical or statistical model of some kind either to fill in information or to predict or estimate something. An example is trying to fill in a power-duration curve, or you're trying to understand expenditure and recovery, or you're trying to estimate aero or rolling drag. Those are applications that require pretty high data fidelity not just at FTP but at all levels of power output. Consistency alone isn't enough. I suspect that riders who say "consistency is all that matters" either aren't interested in these things so can't imagine that anyone else needs to be, or else haven't yet figured out the link between those things and high data quality.

People seem to think that if they're not optimizing everything else then power meter accuracy isn't important. Sometimes this is true, but only because there's little correlation between how well we optimize things and what our needs are.

OH SHIT!! I just love the academics-grade subtle backstabbing. This is like a junior faculty smackdown session.

Do you think that is an unfair characterization of Ray's routine testing procedures? Alex, Robert, and tetonrider seem to agree with the assessment.

Ah, a meaningful contribution, finally! And AlexS.

Yes, I went cheap on the track bike (Stages), and that was a mistake. I was trying to use it to "fine tune" my pursuit standing start, e.g. figure out the balance between getting up to speed as quickly as possible vs. incurring too big a physiological bill that comes due 2 minutes later. Stages just doesn't get the job done, particularly with the ultra-low cadence over the first ~5 seconds.

For track FTP session the Stages was probably acceptable.

If not the least.

Do you think that is an unfair characterization of Ray's routine testing procedures? Alex, Robert, and tetonrider seem to agree with the assessment.[/quote]
No, that's a fair assessment. And Ray would probably agree. Just plotting raw data and providing interpretations involves little skill.

But I still find your tone a little unnecessarily condescending. I'd prefer to see you treat Ray as a peer, rather than a scientist looking down his nose at a "tech."

this thread does rais some interesting points inbetween all the posturing,

in its basic format measuring power should not be dificult, torque x rpm /5250....

one thing that springs to mind is how often is the cadance measured on these systems? I have not delved into them in detail so dont know, but in the case of the quarq, we have the (6?) reed switches, but what about if the accelorometer is being used for cadance, does the measument fall to 2 measurments per rotation?

it seems to me that this could have a fair contribution to inaccuracies especially during sprints, somthing suggested by AC as an issue. certainly on automotive ecu applications somthing that is popular is a toothed wheel, normaly 36 teeth , with a missing tooth to allow the ecu to locate where it is, or even 64 teeth, allowing much more accurate timinng etc.

on the temerature front, turning the climb on its head, its not out of the reals of posibility to have a fast temp change on a descent, especially into a valley/ wodded area.

I think that different people have different goals and needs. Ray's goals and needs aren't the same as yours and mine, so it's kinda unfair to judge him on a standard he's not interested in simply because we are. As my mother-out-law used to say, no matter how thin the pancake it still has two sides. If I can help Ray to improve his protocols and automate his analysis in ways that fit within his goals and needs (and within my own time constraints) then I'm happy to do that. That would be a win-win.

Accelerometers are usually sampled at >100Hz in order to track the gravity vector, a very reliable signal. And a set of filters would generally be used to model the rotation of the crank and predict its velocity and position using pretty reliable "tracking" algorithms.

However accelerometers can be susceptible to noise caused by high-frequency vibration, or can be saturated with shock. Hence Ray's comments that there are failure conditions in certain road conditions that can sometimes be seen. Also on the track with extreme lean, the gravity vector can go way off-axis from the crank. Per RChung's comments, these are usually spurious and quickly recovered from. So would probably have minimal effect on an FTP test. But could absolutely mess up measurement of a 10s jump.

I'm sorry you made that mistake but glad you shared that. Most reviews of power meters say something like, "I compared its average with the average from my other power meter and they're within a couple of watts." That's a horrendously low bar. What's important isn't how close they are on average, it's knowing the conditions when they're not, and by how much.

Well there you go. And yet, on the basis of just this unskilled (<- your characterization) analysis, Ray is adamant that an SRM is not better than other offerings, and hence not worth the extra cost. See the problem?

I agree with all that you stated. I was also in agreement with what Alex posted. The highlighted portion especially.

Apparently you missed this post:

http://forum.slowtwitch.com/...ost=6125595#p6125595

Although there are multiple reed switches in a Quarq, as I recall, they are only used for direction of rotation determination. The cadence signal is still just a once per rev determination. The same with the accel implementation (once per rev - looking at G vector peak) although the inherent noise of the accel implementation makes the cadence estimation slightly less accurate. This is why Quarq recommends using the magnet for best power accuracy.

---

http://bikeblather.blogspot.com/