tomato coupe

If you routinely ride on MUPs that twist and turn as much as a mountain bike trail, then you should do like mountain bikers and use a wheel sensor with your GPS.
You clearly have no idea what multipath errors are.

From Garmin:
"GPS drift, is the difference between your actual location and the location recorded by a GPS receiver. Consumer grade GPS receivers (such as the ones Garmin makes) are not 100% accurate, this will usually cause a difference between your actual location and your recorded location. The GPS location accuracy of Garmin fitness devices is around 3 meters (10 feet), 95% of the time. This means, at any given time, your device will save your location within 3 meters of your actual location."

I think they're a little optimistic. You missed the most important part anyway: "Actually, the correct number is about half of that, at 5m. But, this is for absolute accuracy, which is not the important figure for a bike computer."

There's no filtering going on. The point-to-point scatter is much lower than the (3m or 5m) absolute accuracy of the system because the dominant sources of error in GPS are highly correlated for adjacent readings.

This is the absolute silliest statement in this entire thread. I'll leave it at that ...

wphamilton

I would agree with that. except that the errors that I'm referring to are the path deviations from a smooth path, which the GPS doesn't measure at all. Those will accumulate linearly.

The other type of GPS error, the jagged path, also accumulates linearly. But I intermittently see that, so I don't describe it as characteristic.

This is what I mean. Exaggerated obviously, but when the bike's actual movement has a series of curves, and the GPS takes discrete measurements at periodic points along the curves. The distances are different, even though the end points are the same.

tomato coupe

Okay.

pdlamb

Yes, but I'm not sure that self-referential statement is germane to the topic.

tomato coupe

Sorry, but your previous statement indicates a profound lack of understanding of the concept of measurement error.

aplcr0331

Cool.

I was booking down the highway in the little ring, as I do from time to time, and a trucker I was drafting kept putting fingers out his window to let me know my speed.

Later me and my buddy Kelly L. tore up a bowling alley.

wphamilton

You're talking about 2 different things. He means (multipath) errors due to the different paths a GPS signal might take from satellites. There are various smoothing algorithms to deal with those kinds of outliers, but the result is always an approximation of course. You're talking about errors in the track of the bike's path. Your observation is more important than his, to the question of measuring the path length of the bike's travel btw.

The distances interpolated by GPS are also reduced by cutting corners due to the sampling frequency (best case for Garmin is 1 per second I believe).

If anyone is turned off by all the theory and math, it's pretty easy to compare precisions for yourself. Ride the same route a number of times and see how the GPS distance estimate varies, and compare that with how much the wheel sensor distance varies. Look at the measured track if it's available, and how much it varies from your actual path.

tomato coupe

Here you go ... 10 rides on the same route, logged on a Garmin Edge 520 without a wheel sensor. The starting point probably varied by about 20 feet, and the finishing point by about 5 feet. If you dig into the files, you can extract a couple more digits for the distance, and compute some basic statistics:

Mean distance = 29.794 ± 0.017 mi. (Actually computed from 20 rides)

So, the relative uncertainty is at the 0.06% level.


-----------------------

-----------------------

-----------------------

-----------------------

-----------------------

-----------------------

-----------------------

-----------------------

-----------------------

pdlamb

Where multipath causes problems is (mainly) in cities, where you'll also get some signals blocked by the buildings in addition to buildings reflecting the signal. Say you've got 7 satellites visible, three pretty close to the direction of 7th St. (where you're riding), two to the north, and two to the south. The block-long tall building on the north side blocks the signals from the satellites to the north, and a shorter building blocks line of sight to the southern satellites. The three in your LOS let your GPS know you're between Walnut and Maple Streets. They can't triangulate well from north to south. But signals from the southern satellites bounce off the building on the north side of the street before getting to your GPS (multipath), indicating you're farther away from the satellites, and therefore farther north. It's possible your GPS will conclude you crossed Walnut, went up an alley to 6th or even to 5th St., and then sprinted back to 7th when you get to the end of the block. You speed demon cyclist, you just rode three blocks while your wheel sensor went one block!

pdlamb

Don't worry, Rust Bucket, I caught the ad hominem attack the first time. I didn't remark on it because it wasn't remarkable. But it did entertain me.

(That is what you mean by your user name, isn't it? A coupe is an old car, and a car the color of an old tomato is going to be rusty, so your user name implies an old, rusty car -- what we often call a "rust bucket.")

Anyway, back on topic:
When I calibrate something, I compare the measurement of the device (the wheel sensor plus whatever's counting wheel revolutions) to a known standard (like a surveyed distance). If I ride a surveyed distance of 5 miles, and my device reads 4.97 miles, its accuracy is -0.6%. 0.6% => magnitude, negative => it measures less than the standard.

Now perhaps I don't understand what you mean by "measurement error." But I really don't want to get into an "Alice in Wonderland" kind of situation where I have to worry about what Humpty Dumpty meant every time I read someone else's words.

philbob57

Several years ago, with a Samsung Galaxy S5, 2 rides on sunny, windy days on tree-lined streets: Sigma wired computer read something like 5.01 and 5.05 miles. The Samsung read 4.5 and 4.9 miles. Strava said maybe 5.02 miles.

The GPS was 2 for 2, and in one case it was wildly wrong. I don't know for a fact that it would have failed again, but it wasn't worth the effort to run the test. Besides, even if it started measuring correctly, how would I know? In any case, it's known that trees with leaves can interfere with GPS signals. Anyone who rides under trees and wants an accurate odometer is more likely to be satisfied with a sensor than with a GPS alone.

And that means: if you want an accurate odometer and can't be sure of where you're riding, you're more likely to be satisfied with a sensor than with a GPS. Of course, you've got to remain aware that there will always be some fuzziness in what the sensor reports, too - it's just likely to be less than what a consumer-level GPS reports.

tomato coupe

Yeah, cell phones aren't very good bike computers. No news there.

tomato coupe

Anytime you calibrate a rollout, there is an error associated with that calibration. Later, when you use that rollout to measure distance travelled, there will be a systematic error associated with the rollout calibration and an error associated with the new measurement(s). By nature, you do not know the exact magnitude and sign of these errors – you can only make statistical statements about the errors.

Calibrating your rollout by riding a well known long distance is a good thing, because it can (potentially) reduce the calibration error. It cannot, however, completely eliminate calibration error.

In your example, you've simply determined that the new calibration differs from the previous calibration. The new calibration may or may not be more accurate – it depends on the details of how the two calibrations were made.

tomato coupe

In two days you've gone from not knowing what multipath errors are, to dreaming up incredible scenarios where multipath signals generate extreme errors, the magnitude of which are pulled from your •••. Sorry, science doesn't work that way.

vane171

I read above here that GPS has 10-15 feet (3-5 meters) accuracy. But when I zoom into the track that GPS draws on the map, it shows me consistently on the right side of the road, only some places it may occasionally show me as having gone in the ditch or something like that. I think it is quite precise, within maybe a meter or perhaps even a bit less (3 feet or less).

But the GPS measured speed, as when you look at your phone (I don't use cyclocomputer but phone that I sometimes mount on the bars) as on a speedometer, it is or it seems to be laggy. I never get a feeling from it that it is accurate. it probably shows some average speed I had some ten or fifteen seconds earlier.

So now I bought computer that uses those strap on sensors for speed and cadence (not a sensor coupled to a magnet). The 'speed' sensor goes on the wheel hub. I am wondering if it will work in vertical position when it is strapped on the wheel spokes right next to the hub, say hidden behind cassette (I suppose normally it would be on the front wheel?). For one thing, my hubs are conical in shape and also the sensor would look neater if strapped on the spokes.

I guess it depends what's inside the sensor, how it detects the wheel turns. It is probably some mechanical system, like a metal ball enclosed in the sensor body that 'falls' and closes a contact every time it is 'upside down' (or it may be a mercury contact)? This repeated opening and closing of an electric circuit is then broadcasted via BTooth to the computer. However this quote talks something about 'error rate from sensor' which goes against my logic where the computer simply counts wheel turns, each turn one signal 'click'.

pdlamb

I think you're confusing precision with accuracy.

tomato coupe

Not at all. My post is a direct response to wphamilton's suggestion to compare multiple results from the same route. I make no mention of precision nor accuracy.

philbob57

Funny ... I thought phones ARE supposed to be pretty good consumer-level GPSes.

I apologize, though. I mistook you for someone who wanted to cast light on the subject of this thread.

tomato coupe

Sorry, I meant that cell phones make fantastic bike computers. (Is that what you want to hear?)

philbob57

Millions - maybe 10s, maybe 100s of millions - people rely on their phones' GPSes every day. I expect there are far more phone GPSes around than car, truck, boat, hike, and bike GPSes combined. My bet is that phone GPSes are what most people think of when they read or hear 'GPS'. And here you are dissing phone GPSes on one hand and arguing that GPSes are more accurate than sensors on the other. That's a contradiction that I really didn't expect.

You can have the last word.

I really hate over-generalizations.

philbob57

A bike forum? You didn't really think anyone needed the reminder, did you?

The topic is GPS accuracy in general, and one of the sub-topics is GPS alone vs sensor-based speed and distance accuracy. My own few observations tell me sensor-based reporting is more accurate than GPS alone. That's based on 3 GPSes - 2 phones and one Lezyne GPS unit. I didn't do extensive testing, because the first few tests with each were so telling.

One poster states categorically that GPSes are more accurate. I have no doubt that it's true under certain circumstances, but it's just not true under all conditions, so the statement is false. It's an over-generalization.

A GPS may sometimes be more accurate than a sensor-based system. Sometimes it's less accurate. If you know the conditions in which you ride, you have a good chance of picking the method that will be more accurate. If you don't have that knowledge, you'll probably get more accuracy from a wired, sensor-based system than from a wireless sensor or a naked GPS.

It matters not what package the GPS comes in ... but don't forget that many cyclists rely on their phones for GPS services.

Notso_fastLane

Also note that there are power saving settings in phones that allow for less accurate GPS readings to save power. Some phones, this is automatic depending on if the phone is plugged in, below a certain battery percent, or possibly other factors, so that's one thing to check as well.

Notso_fastLane

The faster one goes (and more importantly, the faster one accelerates/decelerates) the more pronounced this effect becomes.

My favorite speed run around is about 1/2 mile straight with a 50 mph speed limit that I always exceed. I typically hit speeds in the mid 60s as indicated by my wired wheel sensor, whereas the GPS (on a phone no less) rarely goes above 49-52. The wierd thing is, I can hold 65-68 mph for at least a couple yards of that run, but the GPS still lags enough that it 'smooths' out that top speed quite a bit. On my more normal routes, the smoothing is much less noticeable, usually within 3-4 mph peak.

wphamilton

Not to imply anything about TC's data, but it reminds me that having suggested it I'm obligated to say something about methodology.

• Make sure you're not selecting among the different rides. Use all of them that are applicable.
• It's good to select from a given time frame, and not from a given distance, type of ride etc. Same reason as above, and to eliminate variables.
• If you have a "snap to road" option in your software, turn it off for the checks. Snap to road will basically report the map distance, not the bike's true distance.

wphamilton

GPS software performs smoothing on the raw speed data to mitigate suspicious jumps. Smoothing is basically a weighted average over a period of several seconds. Probably over a longer period if the sampling rate is longer - the Garmin "accurate" mode is once per second. Android GPS will vary, but the objective is to have the slowest sampling rate that has acceptable performance because high sampling rates cost battery power.

Short duration speed peaks will disappear due to the smoothing. Wheel sensor devices also do that, because the magnetic switches are noisier than you'd think with some random variation on exactly when in the cycle they activate. But they have a much higher sampling rate than do GPS applications, which is the main reason why that setup is inherently more precise than GPS.