Connected GPS Pace and Distance is Inaccurate

[DanielMena9]

Went on a run today with my Charge 2 on one wrist and Charge HR on my other wrist. Also had RunKeeper to measure my distance. I started my run with about the same number of steps on both FIT devices. 

After the run, my Charge HR had almost 2K more steps than my Charge 2. In addition, my GPS map on FitBit matched my Runkeeper map which is about a 3.5 mile run BUT FitBit pegged the same route as 1.73 miles >.> 

Why doesn't this device work as advertised? I really like the design and feel of Charge 2 but if it can't do its job accurately, all is for not and I will be shorting FIT as it appears to be a widespread issue (I have maps and photos if someone from FIT wants the data).

Moderator Edit: Edited thread title.

[Glennbest]

This just illustrates either how dodgy the blaze and charge 2 are with distances or how ill-informed the support staff are.  Why on earth would the tracker take distances calculated via stride length over distances calculated via gps?  I find the statement "...tracker data will always override your GPS data as your tracker contains more reliable data than the distance provided by your
GPS" to be an astonishing statement.  If that is the case that is crazy.  All the running apps (strava, mapmyrun, myfitnesspal...) use gps to calculate distance with very few problems, why would or should blaze and charge 2 do it differently by using stride lengths which is inherently inaccurate.  If I run fast or down hill my stride length is going to be different to if I run slow or uphill.

The help article below clearly states the if an exercise is tracked using gps it is the gps distance that is used.  I 'believe' the trackers are only 'supposed to' estimate distance when automatically detecting exercise, but based on the confusion coming from fitbit support who knows?

https://help.fitbit.com/articles/en_US/Help_article/1142/?q=gps&l=en_US&fs=Search&pn=2

We want accurate distances as described in fitbit documentation, not an estimated ones, why won't fitbit listen to their customers?

[hawkeye1952]

Glennbest (Base Runner) What it sounds lke to me. Is that they are making things up as they go along. The program that I use. Tells be when I lose the gps and when the gps is restored.


In reality they do not have a why solve the [problem. So they give people a line of BS on how to solve it. When this one dies. I will not buy another one.


George Howell Jr

[Rollerskier]

Hi all,

I have analysed some more the GPS points from my iphone6+ and iphone7.

It looks like Fitbit sample GPS position every second.

There is an average deviation of the GPS position reported by ipphone7 and iphone6+ at the same time ( at the same second) of 4,35 meters..

This is based on comparing about 2700 GPS points during a 45 min walk.

Iphone 7 reports I been walking 19,8% longer in the latitude direction (east-west) and 21,9% in the longitude (north-south) in average every second  during my Fitbit run.

Fitbit reported I had a 3,22 km walk according to iphone6+ and 3,87 km according to iphone7  That is 20,2% longer for iphone7.  So the error is about the same in all directions (where I live).

[Rollerskier]

Hi all,

We are investigating Fitbit Charge 2 connected GPS accuracy in different parts of the earth. This is based on downloading the TCX file for a particular run from the Fitbit dashboard.  So far there are reported specific deviations in east-west direction but not in the north-south direction from Chile and Australia.  This issue was not found on my test files in Norway. The theory is that there could be a calculation error in Fitbit when converting from latitude/longitude position in degrees to distance in meters.  To investigate this further it would be great if some of you following this thread that are using Charge 2 could participate by sending over a TCX-files with GPS positions.  The run must be based on starting from Fitbit Charge 2 not the App itself.

Do we have somebody from North America, Asia or Africa (south of equator) reading this post that would like to participate with a TCX-file?  If you do please post a confirmation on the community and we will contact you for further information.

[Kobalto]

Following @Rollerskier post, these graphs are the results of analyzing TCX files from Australia, Norway and Chile. First, there are three lines in each graph:

- Blue: Pace according to the Charge 2. 

- Orange: Pace calculated using the time and GPS coordinates from the TCX file

- Gray: Pace calculated using an incorrect method, which is basically ignores that the distance between longitud coordinates varies depending on the latitude .

So, in theory,  the Charge 2 blue line should always be very close to the Orange line. But these are the results:

ChileAustraliaNorway

Conculsions:

- In Norway, the blue line and the orange one are close. The Connected GPS works fine.

- In Chile and Australia, the Charge 2 line (blue) is not close to the orange line. Instead, it is close to the gray line (i.e. the incorrect method), thus giving incorrect pace and distance readings.

I was able to replicate very accurately the incorrect Pace shown by the Charge 2, and I did it using only the GPS information of the TCX file, without using the stride length information. Therefore, this is not a problem of poor GPS signal (or bluetooth), and it isn't using the stride length to calculate pace.

The problem is that the Charge 2 is using incorrectly the GPS coordinates: The algorithm to translate coordinates into distance contains an error. Moreover, while in some locations the algorithm works well, in others it doesn't.

Anyone wanting to submit a TCX file is wellcomed, just drop me a message and I'll give my personal email. Thanks @Rollerskier and @Glennbest for submitting theirs.

@DerrickS, I hope this info helps the team to correct the error in future firmwares. 

[Annabel_b]

@Kobalto I'm pretty sure no one from Fitbit actually reads any of these community forum posts. It may be worth emailing this info to Fitbit Support as they seem to be particularly clueless about fixing this problem and could definitely do with the help.

[Kobalto]

It's a good suggestion @Annabel_b, I will wait files from a couple of additional locations and send the data too support too.

[hawkeye1952]

All the support does. Is to read what someone else has written to solve the problem. They have no idea of what is going on. The sad past id I sent a email to the CEO several weeks ago. Never got a response back.

[Kobalto]

Thanks to @Kenny54, we have now a 8k TCX file from Ireland. In this file, the blue line of the Charge 2 is close to the orange line, which means that the Charge 2 is correctly translating coordinates into distance. 

So far we've analyzed 4 locations. The two from the southern hemisphere (Australia and Chile) show the error while the two in the northern hemisphere (Norway and Ireland) seem to be  right.

There were some reports from the US with inaccurate readings with connected GPS, could someone from there send me TCX file?

[Rollerskier]

Looks we are starting to see a pattern here..

Great if we finally can break the inaccurate distance code..

I think we need another southern hemisphere point too..

Anybody from Africa reading this? South Africa, Morocco, Kenya, Tanzania...you have some great runners..are you out there using Fitbit?

Another northern hemisphere point from Asia too. 

Anybody from Singapore, Japan, HongKong, China, India, Thailand, Russia or other countries reading this?

[HNYBDGR]

Rollerskier (and all the other frustrated users out there): I think I have a simple, although not super-elegant, solution to the accuracy problem, if you are willing to use another fitness app to view exercise.

As people have noted, the Fitbit records a GPS route (using the phone's GPS), which is accurate, but it also records a separate distance based on stride length, which is often inaccurate. In the .tcx data file that you can export (Thanks, Fitbit!), a typical entry looks as follows:

<Trackpoint>
<Time>2017-03-09T17:08:13.000-06:00</Time>
<Position>
<LatitudeDegrees>40.08627474308014</LatitudeDegrees>
<LongitudeDegrees>-88.2899899482727</LongitudeDegrees>
</Position>
<AltitudeMeters>220.1</AltitudeMeters>
<DistanceMeters>9798.65</DistanceMeters>
<HeartRateBpm>
<Value>79</Value>
</HeartRateBpm>
</Trackpoint>

Note that each trackpoint has GPS latitude/longitude (which is accurate), heart rate (so-so accurate, but that's for another post), and "DistanceMeters", which is the inaccurate distance calculated from strides.

So really, what you want is to recalculate the crappy "DistanceMeters" using the accurate GPS data in each trackpoint.

Luckily, there's a web app that does it!

1) Export your Fitbit exercise as a .tcx file, like 31765273.tcx (click the ... at the top right dashboard)

2) Go to http://tcxtools.com/tcx/distance/

3) Choose your file, and Methodology "Recalculate all the distances in the TCX file"; leave the "Factor"=1

4) Submit, then on the next page "Export TCX"

5) This will download a file like "out.tcx.xml"

This new file now contains all the same GPS, corrected distance, and same heart rate data. For example, the above point will be turned into

<Trackpoint>
<Time>2017-03-09T17:08:13.000-06:00</Time>
<Position>
<LatitudeDegrees>40.08627474308014</LatitudeDegrees>
<LongitudeDegrees>-88.2899899482727</LongitudeDegrees>
</Position>
<AltitudeMeters>220.1</AltitudeMeters>
<DistanceMeters>51562.4184</DistanceMeters>
<HeartRateBpm>
<Value>79</Value>
</HeartRateBpm>
</Trackpoint>

Note the "DistanceMeters" is now 51 kilometers, instead of 9.7 kilometers, the stride-derived distance.

You can then read this file into any fitness app (I like to use RideWithGPS) and display correct heart rate, distance, and time.

Unfortunately, Fitbit does not currently allow import of TCX files. It would be nice if they added that feature - and it would be nice if their programmers, as suggested by many in this forum, simply coded the wristband so it uses strides only when no phone with GPS is tethered, and GPS otherwise. As has been remarked, strides is really never going to be more accurate than modern GPS, although when well-calibrated, it has its uses for indoor activities where GPS may not work.

Cheers to all.

[Glennbest]

Hi,

I'm fairly certain the distancemeters field is not calculated using stride length, I've tried setting different stride lengths and it made no difference to the distancemeters field. The distancemeters field is accurate when calculated by the app but inaccurate when calculated by the charge 2, in the order of 0.3m per second in my experience, regardless of pace.

The charge 2 appears to be mishandling the gps co-ords in the Southern Hemisphere, although there are reports its also doing it in the northern hemisphere.

Honestly if you want accuracy I recommend a unit with gps built in, Fitbit have been unable to fix this issue since it first began occurring with the blaze, over a year ago. It shouldn't take that long to fix a flaw, unless you either can't or can't be bothered.

[Annabel_b]

@Kobalto @Rollerskier I think the analysis you are doing is really helpful. I live in Australia and my experience exactly reflects what you are finding.

An option to get someone from Fitbit to actually read this information is to start a new thread in the Community Forum. I suspect that because this thread is marked as solved, it is considered to be closed by Fitbit, regardless of how much activity it has. Kobalto, it may be worth using one of your posts that analyses the tcx files to start a new thread. At least that way a moderator will be assigned to it and it won't be marked as solved.

[Rollerskier]

Thanks for the workaround.  I am familiar with this as a temporary solution when new IT-systems at work do not have the functionality yet or there is a bug to fix.

However, we would like Fitbit to fix this issue with conversion from degrees between 2 points to distance.

I find it strange if the stride length ends up in this file when there are GPS coordinates available?

I thought stride length would override if you lost the GPS connection during your run?  Since it is one sample every second there should be some seconds without any coordinates in the extracted file if this is the case or perhaps the seconds are missing?   

Interesting location on your example. If you, HNYBDGR, are situated in China please provide us with a TCX-file for our global investigation.

[Kobalto]

@Rollerskier, I've seen in tcx files that through several samples, coordinates are exactly the same. This probably means lost GPS connection. Nevertheless, the distance continues to increase in each sample, so fitbit probably uses the stride length to estimate the distance. But ineed, as you say, fitbit is not using stride length when GPS connected is enabled and working.

@Annabel_b I will do so, and start a new post with the info I've collected.

Here is a new graph, now from @hawkeye1952 in Nebraska. Again, northern hemisphere has a correct use of coordinates:

He is nevertheless showing inaccuracies at some points, but not sure of the reason. But in the end, the total distance and average pace are within a 1% error (orange vs blue line).

[HNYBDGR]

Rollerskier, nothing as exotic as China 🙂  Just a data point on one of my run loops in Champaign, Illinois.  The -88° refers to longitude west (same as 360-88° = 272° E). Fitbit uses the GPS from the phone, when available, to create the map (GPS longitude/latitude pair), which is as accurate as phone GPS typically is. It utilizes stride length in some fashion (exact algorithm unknown) to do DistanceMeters. Thus DistanceMeters will advance even when there is a GPS dropout. 

Fitbit is off the worst when I get on the bike and then "Run." Since my body is not properly activating the 3-axis accelerometer on the smoother bike ride, the distances are usually vastly shorter (say, 6 vs. 30 miles) than the real thing. Even in that extreme case, the coordinate recalculation of DistanceMeters using GPS gives the correct result, as the GPS phone data is always OK.

An yes, I totally agree: when presenting final data online, Fitbit should use GPS data whenever that's accurately available, and limit interpolation with step count/stride length to lengthy GPS dropouts (running through a tunnel, indoors). Cheers.

[Kenny54]

Hi

HNYBDGR

, I tried the tcxtool site this morning several times over 4 hours but keep getting network error when I try to export the processed file, I also tried using the contact link but that failed to send  also? are you sure the site is still up and running?

Kenny

[hawkeye1952]

I took my morning walk. I use my main program and fitbit. My main program showed 1.17 miles and fitbit charge2 1.10 miles. way off. I also notice that it did not map it correctly either

[hawkeye1952]

This is fitbit map

Tjis is the endomondo

[Rollerskier]

Dashboard and Export TCX works for me with Macbook/Safari.
Took a test just now.