Testing your power meter

How do you know if your power meter is right? The SRM and Power Tap can be checked for accuracy by hanging a known weight from the crank and looking at the recorded torque. Unfortunately, this same static procedure will not work with the Polar S710 since its power measurements are based on the vibrational frequency of a moving chain; in order to check the accuracy of the Polar a dynamic procedure must be used.

Why it's important to check power meter accuracy

During the last part of 2000 and the first part of 2001, a large batch of SRM's appeared to lose their "calibration" when a new type of adhesive was used that required a longer drying time than the previous type (this problem has been reported to be fixed). In addition, although properly-installed Polar S710's appear to give consistent results compared to the SRM and Power Tap (see a three-way comparison here), a poorly-performing S710 can be very bad indeed. This is not meant as criticism of either the SRM or the Polar--there are also oddities with the Power Tap--but it does point out the importance of testing one's power meter.

How bad can these errors be? Here is a graph based on data taken from a gear test of a Polar S710 on a fluid-trainer where the rider held constant speed across all gears. Each yellow bar shows the spread of power readings for each gear (these bars are known as boxplots). The left panel of the plot shows the boxplots for the small chainring, the right panel for the large chainring. In a properly working power meter, the yellow bars should all be about level. For this particular setup, the power meter showed huge variation (from a low around 120W up to a high of almost 220W) among gears. In case you're wondering if the variability in power readings was due to an inability to hold the speed constant, I have labeled each boxplot with the average speed in red--the rider was trying to hold 26kph in all gears.

I believe this to be an unusally aberrant case, but it does point out that installation of the Polar S710 is not as simple as getting the LEDs to light up and that one should test the installation to ensure that it is working. See Sandiway Fong's S710 page for a less severe example of odd S710 readings, and what he did to improve them.

I believe that the pattern of errors displayed in Sandiway Fong's installation has a different cause than the pattern displayed above. In the case shown here, notice that the cassette pattern is (roughly) similar across the two chainrings. This means that it's not the chain angle, or even exactly the distance from the sensor to the chain. Since the wheel speed was approximately constant in this test, what is it that varies with cog but not gear? I think it's chain speed. The Polar uses a chain speed sensor in addition to a chain tension sensor, and I think either it or its cable is picking up interference. As a first step and working hypothesis, if one observes this pattern of variation across gears with a Polar S710, I would re-install the chain speed sensor and re-route the cable to see if it makes a difference.

I have exchanged messages with two of the developers of the technology underlying the Polar S710. One believes that raising the chain tension sensor so that it is as close as possible to the chain will alleviate this problem. The other believes that this is a phenomenon that only occurs on trainers, and points to the two on-road comparisons here and here as evidence.

How to test a Polar S710

As mentioned above, all meters should be checked. However, a Polar S710 cannot be checked by hanging a weight from the crank as can a SRM or a Power Tap--it must be tested by getting on the bike, riding it, and comparing it under controlled conditions with other information. In addition, unlike the Power Tap and SRM (which measure torque in one spot, either the rear hub or the crank), you may be wondering if the Polar's power readings will be sensitive to different gear combinations. Although test procedures for a Power Tap or SRM are already available, I have been unable to find a description of a test procedure for the Polar. Accordingly, I present a test protocol here, of which there are two parts: testing for consistency across gears, and testing for overall accuracy. Both are tedious, though if one has an ordinary trainer (fan, fluid, or magnetic) available, the consistency check may be simpler and faster; therefore, I recommend that you do this step first. If you don't have a trainer you will obviously have to go directly to the on-the-road testing.

Checking for consistency with a trainer. This test will show you if your power output is consistent across all gears. If you have a fan- or mag-trainer handy, this test will be simpler and faster than an on-road test. It has recently been suggested to me that a constant-speed trainer test may not provide definitive answers with the Polar Power unit. I am currently investigating this.

• Place the bike on a regular trainer. If you have a Computrainer or a Tacx Grand Excel or some similar type of ergometer, I wouldn't advise using it in ergometer mode this time--it is just one more variable that you'll have to control for. Likewise, for this particular test since you are looking for consistency across the gears and not particularly consistency across tests you needn't worry particularly about tire pressure or skewer tension, though it may be good to warm up the tires thoroughly.
• For this test, pick a speed in the range of 25-30kph (15-18mph) that you can hold across all of your gears. Shift through all the gears holding your target speed for about 60-90 seconds per gear. It may be hard to maintain a steady cadence in the lowest and highest gears, but do your best. Because the Polar records at 5-second intervals, you will need at least 60 seconds per gear to get enough observations, especially if it takes you a few seconds to steady your cadence after changing gears. There is diminishing benefit of additional observations beyond about 90 seconds.
• Examine a graph of your power against time or power against speed/cadence (Speed/cadence is equivalent to gear ratio times a constant). You ought to be able to tell immediately if power varies a lot across gears, or if it is fairly consistent. If you get good results and all you care about is consistency, you needn't do the second part of the test.

Here is a graph of such a test. The rider maintained a constant speed of 30kph in each gear for about 2 minutes per gear. The top panel shows his gearing by time as he shifted from a 39x13 down to a 39x21, then switched over to the big ring and shifted from 53x21 back up to 53x12 (he skipped the 39x12 combination). The red dots show the measurements for the small chainring, the black dots for the large chainring. In the bottom panel you can see the recorded power by time, on the same time scale as the upper panel so you can tell what gear the rider was in when he recorded each dot.

For this rider's setup, note that the power was more even across the cassette for the small chainring than for the large. However, also note that one can still discern a small curved pattern of power across the cassette for both chainrings. Here is a plot that combines the information of the two panels above which may make this point a bit more clearly:

There are two preliminary conclusions from this test on this particular setup: first, this rider needed to check all of his gears because the error appears larger over one chainring than the other. In fact, the error across the small chainring might be acceptable because most of the observations lie within a 5% band of its mean; however, the error across the large chainring is about twice that size. Second, the variation across the cassette was similar for each of the chainrings, and does not appear to depend on the gear ratio but rather the particular cog. This is a symptom that the chainspeed signal is either weak or getting interference. In addition, that the error is greater for the big chainring suggests that the chain tension signal was also weak (relative to the small chainring signal). I advised this rider to raise his chain tension sensor as high as possible, and to re-install his chainspeed sensor and re-route the cable to minimize interference.

Checking for accuracy with a hillclimb. If you didn't get good results from the previous part, or if you want to check the overall accuracy of your Polar S710, you'll have to do a road test.

• Locate an appropriate hill. Find a hill of perhaps 3 to 4% that is about 60 to 90 seconds long (perhaps 500m or a third-of-a-mile), protected from the wind and of even grade. The reason why a hill is good is because you'll go at lower speeds so aero drag which varies nonlinearly with speed will have less of a distorting effect.
• Even though you'll want to do the test at fixed speed (or nearly so), pre-calculate the cadence in each gear for that speed and ride to cadence rather than speed--this will also give you an idea of the range of gearing you'll be able to use. Riding to cadence is better than speed both because it has slightly finer resolution, and because I think one's legs recognize (and can hold) a particular cadence more easily than they recognize a particular speed. Otherwise, you'll have to keep your eye on the speed and I think that's distracting, especially when there's lag in the reporting.
• It isn't necessary to do this at high wattages. Depending on the grade of the hill, you can use analyticcyling.com to make a rough estimate of the power at some reasonably low speed, then convert that to cadence-by-gear.
• Examine the graph of power against gear, as described above. Compare the power both across each gear combination, and with the prediction made at www.analyticcycling.com.

Installing a Polar S710

Some hints and photos on installing the power sensor for the Polar S710 can be found here.