[dexkid]

Having worked on a GPS automotive product, I can only say that the Gps has limited accuracy. the gps used in cars has an accuracy of around 50 m. That leads to the delta with the speedo.

Also my advice to people using phone gps to measure speed is please don't do so. Phone gps may be even less accurate. Also on phones like samsung wave multiple tasks are running on phone. You never know which one may interfere causing wrong calculation of speed.

Better go by the speedo! My 2 cents.

[Dr.AD]

Originally Posted by dexkid

Having worked on a GPS automotive product, I can only say that the Gps has limited accuracy. the gps used in cars has an accuracy of around 50 m. That leads to the delta with the speedo.

Also my advice to people using phone gps to measure speed is please don't do so. Phone gps may be even less accurate. Also on phones like samsung wave multiple tasks are running on phone. You never know which one may interfere causing wrong calculation of speed.

Thanks for this information. Yup, phone GPS may have even less accuracy. But I believe the dedicated GPD hardware (such as VX 140) for car is reasonably accurate.

I agree with you that the GPS used in cars has inaccuracy of about 50m. But why would this convert into errors in speed? Speed is calculated based on differential calculations, and as long as the error is 50m in both reference position and change in the position (delta_P) over time (delta_T), the speed (delta_P/delta_T) should be fairly accurate even if the positions themselves are inaccurate by about 50m.

Can you please explain why the inaccuracy in speed measurements? Thanks again!

Better go by the speedo! My 2 cents.

Well, regardless of the GPS discussion, I would still not believe speedo. Speedos are known to be inaccurate in many cars and as I mentioned, several road tests give a table of speedo errors and often these errors are non-negligible. I remember instances where my car speed was measured by various techniques (not GPS based) and every time it came different than the speedo numbers, even in some high-end cars.

[NetfreakBombay]

Originally Posted by Dr.AD

I agree with you that the GPS used in cars has inaccuracy of about 50m. But why would this convert into errors in speed? Speed is calculated based on differential calculations, and as long as the error is 50m in both reference position and change in the position (delta_P) over time (delta_T), the speed (delta_P/delta_T) should be fairly accurate even if the positions themselves are inaccurate by about 50m.

Error is not constant. It could be even be +50 on point 1 and -50 on point 2 (Extreme case). However, if you have a clear view of sky, measurement should be reasonably correct.

[Dr.AD]

Originally Posted by NetfreakBombay

Error is not constant. It could be even be +50 on point 1 and -50 on point 2 (Extreme case). However, if you have a clear view of sky, measurement should be reasonably correct.

That would be the case if the errors were to be random errors. In reality, the GPS errors are not random and they are systematic errors. Therefore, it is very unlikely that the errors fluctuate like you have said, at least over a short period of time. Over longer time horizons, they will fluctuate but they could safely be assumed to be constant over shorter intervals when the overall conditions do not change dramatically. Therefore, the speeds measured by GPS are fairy accurate even if the positions are not. Maybe some GPS expert can pitch in and comment on this ;-)

Thanks,
-AD

[DerAlte]

Originally Posted by dexkid

Having worked on a GPS automotive product, I can only say that the Gps has limited accuracy. the gps used in cars has an accuracy of around 50 m. That leads to the delta with the speedo. ...

Originally Posted by Dr.AD

... I agree with you that the GPS used in cars has inaccuracy of about 50m. But why would this convert into errors in speed? ...
I would still not believe speedo. Speedos are known to be inaccurate in many cars ...

Originally Posted by Dr.AD

... In reality, the GPS errors are not random and they are systematic errors. ...

50m inaccuracy??? That would be the difference between Jaynagar and JP Nagar, or GK2 and CR Park!!!

The accuracy permitted by US DoD for GPS for civilian apps (the satellites still belong to US DoD) was originally 10m / 30' or so - in the 80's! Over the years since then, they have permitted GPS developers to improve that to what it is today - 1m (about 3'). That has permitted positioning accurate enough to judge which lane on a road the vehicle is on.

The problem is map accuracy and resolution! The location that the mapping software shows location on a map may be inaccurate by +/- 3m, basically because the map returns the objects in its storage nearest to the GPS coordinates. Due to extrapolation errors, this can seem a lot. And the blessed system has hysteresis to boot - approach the same point from opposite directions, and the difference may be +/- 3m!!!

Yes, the delta with the speedo has nothing to do with that (as @Dr.AD is pointing out with his math logic ). In fact, speed calculated from the GPS fixes *is more accurate* than speedo reading.

The difference between displayed speed in the speedo and actual is about +4 Kmph (speedo shows higher speed). Modern speedos (which work on a pulse train from the wheel / gearbox mounted sonsor) may not have any other inaccuracies such as non-linearity etc. There is a design reason for this excess reading, though the exact reason slips from my geriatric mind. Anyhow, that hasn't troubled anyone - that is good enough for drivers in US and EU stay within speed limits without problems. To anyone whose mind thinks numerical enough to feel the loss of 4Kmph, that is usually the difference between being safe or being in an accident.

[tsk1979]

Even I am finding the number of 50m, esp when it comes from somebody saying they have worked with GPS chips.
From what I know, horizontal accuracy is dependent upon HDOP.
With WAAS enabled, and a HDOP of around 1.0, you can get around 5-7m accuracy.
With no WAAS, with a HDOP of 1-1.5 you can easily expect around 15m of accuracy.

As for speed, after you stop accelerating, GPS speed readout is very very accurate.

[NetfreakBombay]

Originally Posted by Dr.AD

That would be the case if the errors were to be random errors. In reality, the GPS errors are not random and they are systematic errors. Therefore, it is very unlikely that the errors fluctuate like you have said, at least over a short period of time.

It depends on the terrain and factors like cloud cover. These things can distort signals in matter of meters.

I am in Manhattan at the moment, and Garmin, Nokia 5230 and iPhone all fluctuate a lot in terms of position accuracy due to "urban canyon".

[Dr.AD]

Originally Posted by DerAlte

50m inaccuracy??? That would be the difference between Jaynagar and JP Nagar, or GK2 and CR Park!!!

DerAlte, thanks for the nice explanation!

Yes, thinking again, I agree with you that 50m sounds too much.

Anyway, regardless of the absolute errors in position coordinates, I reiterate that this error does not translate into speed errors and the speeds by GPS are fairly accurate.

Regarding the error in position, here are some empirical measurements: When I drive with my GPS, and whenever there is a precisely identifiable object in the map (such as a railway crossing, a small bridge etc.) I try to see what the map shows when I am exactly at that point (i.e. exactly on the railway lines crossing it, or exactly on a small bridge). My observation is that 90% of times, the map shows my car position as exactly on that object. Only about 10% of times (mostly in areas away from cities and towns) I have observed that the map shows me a few meters away from that spot. However, given the small size of the map and the lack of other distance markings, it is not easy to estimate the error in distance units in those instances.

Overall, yes, it is pretty accurate, and similar empirical data suggests that even the speed measurements are accurate.

Originally Posted by tsk1979

As for speed, after you stop accelerating, GPS speed readout is very very accurate.

Yes, exactly! This is my observation too.

Thanks,
-AD

[NetfreakBombay]

Originally Posted by Dr.AD

Overall, yes, it is pretty accurate, and similar empirical data suggests that even the speed measurements are accurate.

GPS chips have built-in logic to contentiously estimate this error. You can keep "satellite status" or similar screen open on the GPS. This error will almost always be higher than 20m. This is that screen:



Accuracy is dependent on:

1. Number of satellites that unit are connected to
2. Signal strength of each

[Dr.AD]

Originally Posted by NetfreakBombay

GPS chips have built-in logic to contentiously estimate this error. You can keep "satellite status" or similar screen open on the GPS. This error will almost always be higher than 20m. This is that screen:

Cool! Thanks for sharing the picture! Unfortunately, my device shows the signal strengths but does not show accuracy information.

By the way, you say that "This error will almost always be higher than 20m", but in the screen-shot you have given, the accuracy is 19 feet which is just about 6m. What am I missing here?

Also, when you move on a highway, do you see this error fluctuate randomly within a fraction of a time? And if yes, is that the limitation of that particular device or is it for all devices?

Although my device does not show the accuracy numbers, the signal strengths and various visible satellite information does not change so randomly within a fraction of a second.

Also, the "urban canyon" situation you mentioned may be an exception rather than the norm. On open highways, why would the GPS accuracy fluctuate so much?

Originally Posted by NetfreakBombay

Accuracy is dependent on:

1. Number of satellites that unit are connected to
2. Signal strength of each

I don't understand this too. As long as you have connections to a minimum of three satellites, your position can be estimated. Having connection to any more satellites should not change that. For example, if I have connection to 8 satellites and then that changes to 9 satellites because I got one more connection, why would the error change? Won't the three satellites with the best strengths be already estimating my position to the best possible accuracy (whatever that may be)?

Also, why does the signal strength matter? Isn't it a somewhat binary situation that either I have a connection to a satellite and related data to be used in the calculation or I don't have that. Why would the error in calculations change based on signal strength? The error should not be proportional to signal strength.

Thanks,
-AD

[NetfreakBombay]

Which device do you have? In Garmin, you need to touch the bar that sows GPS signal strength for 5 - 7 seconds to bring this up.

This picture has 8 satellites. If GPS receiver can see 8 satellites then accuracy would indeed be pretty good. This is required for lane guidance feature. In this case even altitude would be fairly accurate.

Usually, a lock on 8 satellites would be tough to obtain unless sky is clear and device has clear view horizon to horizon.

[DerAlte]

Originally Posted by NetfreakBombay

... built-in logic to contentiously estimate this error. ... This error will almost always be higher than 20m. This is that screen: ...

Did you mean "continuously"? There is no contention there, I guess. If at all, each 'strongest signal fix' (fix obtained from 3 strongest signals) may be compared against those from other 'fixes' obtained. This is called 'removing contention' (discarding invalid signals), I think, same as in most critical instrumentation with redundant signal sources, like in aircraft instrumentation.

20m? 19' = ~6m+, no?

Originally Posted by NetfreakBombay

... Accuracy is dependent on: ...

No, accuracy is independent of signal strength. As long as one is able to read the 'time stamps' in any satellite signal, it is just fine. The satellite information is effectively 'digital'.

Number of satellites should be 4 min., to 'remove contention' (as above).

[dexkid]

By 50m accuracy, I meant +- 25m.

This means that at time t1 when you are supposed to be at exactly spot s1, then raw gps coordinate data is g1 = s1 +- 25m. So at time t2 when the vehicle has moved to spot s2, the raw gps coordinate data is g2 = s2 +- 25m.

Actual velocity is av = (s2 - s1)/(t2 - t1).

(worst case)
If suppose g2 = s2 + 25 and g1 = s1 - 25
then gps velocity gv
= (g2 - g1)/(t2 - t1)
= (s2 + 25 - s1 + 25) / (t2 -t1)
= (s2 - s1 + 50) / (t2 - t1).

If sampling is done every 5 secs then
at av = 40 kmph = 11.11 m/s
worst case gv = (55.55m + 50) / 5 = 105.55 /5 = 21.11 m/s = ~76 kmph

at av = 80 kmph = 22.22 m/s
worst case gv = (111.11 + 50) / 5 = 32.22 m/s = ~116 kmph

Inferences -
1> Position does matter when calculating velocity.
2> At lower velocity you will get more inaccurate results.
3> Longer the duration of sampling at constant speed, the more accurate velocity you will get. But it is stale velocity data. So approximations are applied to gps positions in software and hence you achieve the accuracy that everyone here is talking about. But the core hardware will still give you +-25m accuracy.

To my knowledge, only defence forces have actual +- 1m accurate GPS devices.

I agree that I don't have complete info. If someone has the correct information about GPS working, then please let me know and help me correct my misconceptions.

[tsk1979]

dexkid, accuracy depends upon
1. Number of satellites - This int turn determines HDOP. For example, if your GPS is connected to 10 sats, you can easily get an error of only 5 or 6 meters or even lower in open spaces(no reflective issues).
2. WAAS - If you are getting WAAS signal, error is typically 2-3m only.

[DerAlte]

@dexkid, have you at least used a GPS unit practically, or are you indulging in an intellectual exercise with a mathematical hypotheses? You equations may be correct, your assumptions are definitely not. No 'self-respecting' GPS unit will give such wild readings, especially when there are real world references / alternates (like a speedometer / odometer) for comparison!

"Stale velocity data"? Would one really be interested in the fact that the speed of their vehicle has changed from 11.37 m/s to 11.49 m/s in the last 200ms - 'fresh' enough? And how did you come to the definitive conclusion that "core hardware will still give you +-25m accuracy" without knowing how the core hardware works?

Accuracy of measuring instruments is serious business, and manufacturers usually test accuracy of a model at least once in its design lifecycle ('type testing'). And the reference for comparison is usually a cross mark on a rather large granite blocks, which have been in existence for conventional cartography long before GPS was conceived.

There was a time when commercial GPS manufacturers were forbidden from achieving accuracy better than +/- 3m, similar to 128-bit encryption being forbidden. 'Achieving', because - as @tsk1979 has tried to explain - accuracy is dependent on hardware + software techniques which are not simple to explain, and have significant cost associated. One is allowed to achieve +/- 1m now for commercial devices. Whether devices with +/- 1m accuracy are made available to public is dependent on economic sense (price, sales numbers), and not technical idealism.

If we are talking accuracy w.r.t. a point being shown by software on a map display, which is not been calibrated in physical units (like "center-line of 2nd Cross is 41.5m from the center-line of 1st Cross") generated by the same software, we must be indulging in what in North India known as "chandu-khane ki gup-shup" !!!