Waypoint averaging - how to

[+user13371]

If you collect a "track" while stationary, what's the correct way to average the waypoint? A simple average of all the coordinates, or do you need other adjustments to the data?

[+dfx]

the correct way would be to do weighted averaging, taking the momentary EPE/PDOP of each sample point into account. a simpler method would be to simply drop all data points with an EPE/PDOP above a certain threshold value.

 

of course a regular tracklog doesn't necessarily give you that information. also, depending on what you use to create that tracklog, you may not get all sample points, as it may filter out samples that aren't at least a certain distance from the last collected sample.

[+ecanderson]

If you collect a "track" while stationary, what's the correct way to average the waypoint? A simple average of all the coordinates, or do you need other adjustments to the data?

Under normal circumstances, I'd probably be expressing it in terms of standard deviation. Problem is, if your sky is any good, your drift isn't going to be enough to get much variation. I've had my units out in the front yard and (happily) been unable to get more than 0.001 drift back and forth in a single direction (not +/-, just 0.001 N or S, and 0.001 E or W). Doesn't make for much of a sample <g>.

 

I like DFX's idea of catching the EPE and PDOP data. If you capture the Garmin 0183 data, that will provide the EPE (in $PGRME) along with coordinate data at each point. Wouldn't change how I did my averaging, but would certainly add an interesting component to the study.

[+dfx]

I like DFX's idea of catching the EPE and PDOP data. If you capture the Garmin 0183 data, that will provide the EPE (in $PGRME) along with coordinate data at each point.

it's not my idea

when i was still using my PDA for caching, i used to use beelineGPS for averaging. it has a simple but configurable (P)DOP filter.

i'm pretty sure that at least the newer garmins take the EPE into account somehow when averaging, as the "confidence" indicator is slower to grow under bad reception conditions.

for any non-garmin receiver, you'd use the DOP values for weighted averaging.

Edited November 15, 2010 by dfx

[+ecanderson]

i'm pretty sure that at least the newer garmins take the EPE into account somehow when averaging, as the "confidence" indicator is slower to grow under bad reception conditions.

for any non-garmin receiver, you'd use the DOP values for weighted averaging.

I suspect the process is even more entertaining than that, and isn't based upon DOP, but actual coordinate calculation. The reality of DOP would, of course, explain variability in the calculation of current position, but I'm betting it's the calculated position data itself that is driving the "gauge".

 

I've seen the confidence indicator reverse any number of times - often a function of my own position around the GPS, possibly blocking signal from one direction or another. I suspect that my 450 is using a recognition of the standard deviation of the samples of calculated position to determine confidence, and only while that value remains small will the confidence indicator increase. Conversely, I suspect that if the standard deviation begins to increase, that's when the confidence indicator begins to reverse to a lower level.

 

One thing I find fascinating is the final description of what appears to be some level of correction based upon the result. Sometimes as low as 1 foot, sometimes a fair bit more. If I recall correctly, the page calls it "Distance Adjusted". I'm still unsure as to the meaning of that.

[+g-o-cashers]

One thing I find fascinating is the final description of what appears to be some level of correction based upon the result. Sometimes as low as 1 foot, sometimes a fair bit more. If I recall correctly, the page calls it "Distance Adjusted". I'm still unsure as to the meaning of that.

 

Garmin stores the number of samples taken for an averaged waypoint in the waypoint gpx file. Every time you take an new reading on that waypoint it uses the newest sample to adjust the coordinates based on the distance between the new sample and the location of the waypoint. The amount of the adjustment gets reduced proportional to the of samples already taken.

[+dfx]

One thing I find fascinating is the final description of what appears to be some level of correction based upon the result. Sometimes as low as 1 foot, sometimes a fair bit more. If I recall correctly, the page calls it "Distance Adjusted". I'm still unsure as to the meaning of that.

 

Garmin stores the number of samples taken for an averaged waypoint in the waypoint gpx file. Every time you take an new reading on that waypoint it uses the newest sample to adjust the coordinates based on the distance between the new sample and the location of the waypoint. The amount of the adjustment gets reduced proportional to the of samples already taken.

but what if it's a new waypoint? i go to averaging, tell it to make a new waypoint and at the end it tells me a "distance adjusted". adjusted from what?

[+ecanderson]

but what if it's a new waypoint? i go to averaging, tell it to make a new waypoint and at the end it tells me a "distance adjusted". adjusted from what?

Precisely my question. It shows up the first time you create an averaged waypoint. Good to know I'm not the only one scratching my head out here. I did try to research it before asking, and came up empty handed.

 

Will be curious to see if anyone comes up with an answer. I'd like to think that numbers such as "1 foot" are great, but I'm pretty sure that isn't necessarily the case... I've had that with truly rotten EPE figures. Although you can also have rotten EPE and very little coordinate drift. Hmmm... still pondering.

[+dfx]

Will be curious to see if anyone comes up with an answer. I'd like to think that numbers such as "1 foot" are great, but I'm pretty sure that isn't necessarily the case... I've had that with truly rotten EPE figures. Although you can also have rotten EPE and very little coordinate drift. Hmmm... still pondering.

the lowest i've ever seen was 0 meters. yep, zero. if that tells you anything. probably doesn't, same as me.

 

but maybe it works like this when you create a new waypoint: first it takes a snapshot reading of the current coords and then starts the averging. when it's done with the averaging, it shows you the distance between the averaged coords and the original snapshot coords. it would be a completely meaningless number, but at least there's something to show. makes sense? no? i didn't think so.

Edited November 16, 2010 by dfx

[+ecanderson]

but maybe it works like this when you create a new waypoint: first it takes a snapshot reading of the current coords and then starts the averging. when it's done with the averaging, it shows you the distance between the averaged coords and the original snapshot coords. it would be a completely meaningless number, but at least there's something to show. makes sense? no? i didn't think so.

That was my first guess when I saw the number when I first started using this feature, but after doing it a few times when placing caches or double checking another CO's peculiar coordinates for my find logs, I realized that the coordinates shown at the start of the averaging process (the current averaged coordinates are shown during the process on my Dakota and Oregon) weren't the same as the coordinates it came up with at the end, and tossed that meaning for the number out as a non-starter. I agree, it seemed like a great idea at the time. In fact, I still like it ... except it doesn't square with what the unit is telling me about the coordinates.

 

When running an averaged location, so you sometimes see some big league backtracking of the confidence percentage, a regroup, and sometimes bouncing back and forth between high and low confidence as the unit tries to come up with a set of numbers it likes? If so, were you able to associate this with anything in particular? I've always suspected it was the fact that I wasn't standing still, was too close to the GPS, and kept giving it a new set of solutions when I'd block out one or another of the satellites it was working with.

[+dfx]

When running an averaged location, so you sometimes see some big league backtracking of the confidence percentage, a regroup, and sometimes bouncing back and forth between high and low confidence as the unit tries to come up with a set of numbers it likes? If so, were you able to associate this with anything in particular? I've always suspected it was the fact that I wasn't standing still, was too close to the GPS, and kept giving it a new set of solutions when I'd block out one or another of the satellites it was working with.

i've seen the "progress bar" make leaps backwards occasionally, but i don't know about going back and forth several times, as i usually don't watch it closely while it's doing its thing. i've always interpreted it as the software figuring out that the variance and/or reception quality was worse than it had originally thought, which means it requires more samples than first anticipated. but it's really impossible to tell what's going on behind the scenes there...

[+g-o-cashers]

I just ran a couple of tests and I think its pretty much what dfx was saying (I did this on my 62s).

 

1) Goto Waypoint Averaging select "Create Waypoint".

 

Takes a snapshot of your position which you never see as your first "saved location" and then starts displaying the "averaged location" for the first sample.

 

2) Hit Done.

 

You see a "saved Location" which is the weighted average that factors in the sample count with the previous saved location and the most recent sample. Distance adjusted is the distance between the averaged location of the last sample and new saved location.

 

3). Goto Waypoint Averaging and select an existing waypoint

 

You will see the sample count, saved location and the distance away based on the current instantaneous location.

 

4). Hit Start

 

Averaged location for this sample starts to record

 

5) Goto 2)

 

When I ran my test I took 6 samples at the same location and the distance adjusted got down to 1-2' (and the saved location didn't change because it was beyond the precision of the coord format I was using). When I moved 100' away on the 7th sample I saw the new averaged location (far off from the saved location) and when it was done averaging the 7th sample the distance adjusted was around 13' (expected because of the weighting) and the new saved location was moved about that far based on my rough calculation.

 

As for the backtracks I usually see them if the averaged location is moving (either because I moved the unit or because variations in position calculated by the GPSr).

 

The bigger question is why do I have to go to a special menu to do this -- why can't it be part of the "Mark Waypoint" function?

 

GO$Rs

[+ecanderson]

I just ran a couple of tests and I think its pretty much what dfx was saying (I did this on my 62s).

 

1) Goto Waypoint Averaging select "Create Waypoint".

 

Takes a snapshot of your position which you never see as your first "saved location" and then starts displaying the "averaged location" for the first sample.

 

2) Hit Done.

 

You see a "saved Location" which is the weighted average that factors in the sample count with the previous saved location and the most recent sample. Distance adjusted is the distance between the averaged location of the last sample and new saved location.

OK - let's stop there for a moment, since that's often where one stops anyway (i.e., not taking yet more samples against a completed ("Done") waypoint.

 

In #2, you refer to a "previous saved location". By that, do you mean some snapshot that was taken when the button was clicked to "Create Waypoint"? If you look at the coordinate data immediately after making that selection, you can see the coordinate information at the start of the process.

 

For example, let's say that the initial values shown are 40º01.000 105º01.000 and the final values when the confidence level reaches 100% and the point is first saved is 40º01.002 105º01.002. If the assumption about what is displayed at the end is the distance between the initial and final values, I would not expect to see 5 feet, but that does happen. 0.002 errors in both the latitude and longitude add up to more than 5 feet.

 

MOST of the time, I could explain what I'm seeing by the first/last explanation, but not all of the time, which is what put me off that idea.

[+g-o-cashers]

In the few tests I did that seemed to be what was happening. Create waypoint takes a reading of your location and then the averaging starts. I guess there is no guarantee that the first "averaged location" you see is the same as the snapshot especially if the GPS hasn't settled or is moving around when you start the averaging. You may be right, there could be something else going on here, I'm just going based on what I saw and it seemed to make sense given the theory above.

 

Second time around (and thereafter) its pretty easy to see that this is what's going on, especially if you move around a bit to introduce some error.

[+swope2bc]

Ok, I see everything that has been said, but I have a couple of questions... When I do an average it always only claims 2 samples. Why is that? What do I need to do in order to display more samples in order to get a more accurate average?

 

Also, why does it tell me to wait 90min between samples?

 

Thanks for helping.

[+dfx]

When I do an average it always only claims 2 samples. Why is that? What do I need to do in order to display more samples in order to get a more accurate average?

think of the "sample" it's talking about as an averaging session, i.e. everything from the start of the averaging process until you hit "save" counts as one "sample". if you created the waypoint through the "mark waypoint" button, then this will take a single coord reading to create the waypoint and this will also count as one "sample".

 

if you want to create a new waypoint at an averaged location, you should probably skip the "mark waypoint" step.

 

Also, why does it tell me to wait 90min between samples?

this is explained here: http://garmin.blogs.com/softwareupdates/20...-averaging.html

in short, they believe it works better that way because it gives the satellites enough time to reposition.

[+swope2bc]

When I do an average it always only claims 2 samples. Why is that? What do I need to do in order to display more samples in order to get a more accurate average?

think of the "sample" it's talking about as an averaging session, i.e. everything from the start of the averaging process until you hit "save" counts as one "sample". if you created the waypoint through the "mark waypoint" button, then this will take a single coord reading to create the waypoint and this will also count as one "sample".

 

if you want to create a new waypoint at an averaged location, you should probably skip the "mark waypoint" step.

 

Also, why does it tell me to wait 90min between samples?

this is explained here: http://garmin.blogs.com/softwareupdates/20...-averaging.html

in short, they believe it works better that way because it gives the satellites enough time to reposition.

 

 

Thank you, does it pay to take an averaged waypoint and average it again?

[+dfx]

Thank you, does it pay to take an averaged waypoint and average it again?

it will definitely increase the accuracy, but so far i've found that a single averaging session easily gives coords that are good enough for any geocaching purposes.