Accurate running speed - gps v footpod

Hi,

I recently got a banglejs2 (love it, btw), and I am in the process of replacing my garmin fr245 with it, specifically for running. My main question is about feasability of getting accurate running speed, but I'll give some context.

I have now been using both watches for a couple of week on all my runs. I have found that the gps and heartrate monitor are not as accurate on the bangle as on the garmin, which I guess is no surprie, given garmin's reputation, and the price difference. When comparing gpx tracks on the laptop using a map, the bangle gives a route more or less conforming to the one I did, but the garmin route seems to conform much better.

On one hand, the heart rate monitor seems to give very inaccurate values. I should be able to connect a polar h10 chest strap to the bangle without problem, right? As far as I can tell, people have already done this, and this should give better accuracy than any wrist based optical heart rate monitor.

On the other hand, I would be super happy to have an accurate reading of speed when running. Although the bangle and garmin give me more or less the same reading on total distance, the number that bangle gives me during the run seems to be pretty inconsistent. Although I have no clear way of measuring this at the moment, it seems to me that the speed is often underreported, and occasionally overreported. This is in comparison with the garmin, but also supported by my "feeling" when looking at the two numbers (which is weak as evidence, but ok). I have tried configuring the recorder app to use 10, 3 and 1 second intervals, but the resuls are similar.

Idea 1: Is there any software solution to get a speed on the gps that fluctuates less? In principle, running in a mostly straight line should probably give an accurate reading on speed, even if the points are shifted. I'll admit that I have no idea about that, but it's worth asking.

Idea 2: There is a company out there, called stryd, which makes footpods specifically designed to measure a host of running data, and as far as I know, it is very accurate (but expensive). I might post something on their formus to see what they say, but does anyone know if it is possible to connect a stryd to a banglejs2, and in that case, what data could be logged down on the watch? For me, speed would be top concern, but I have been logging down cadence and ground contact time on my garmin (using a garmin running dynamics pod) and continuing that would be great. Their webpage has a list of "compatible" devices, and it does include many brands, but it doesn't mention being able to connect to some device and spit data at it. I'd love to try it and see what happens, but these devices are expensive.

Idea 3: The most fun project would be to replicate the functionality of commercially available footpods using a puckjs (or any other accelerometer). In principle, I understand that intergrating acceleration gives speed, and then distance. But I have no engineering experience, so I have no clue about noise and errors. So the question is, basically:

Is a puckjs accurate enough to measure the length of a step while running?

Does anyone have experience with a project like this?

Would it involve programming more complicated than summing up the stream of numbers to integrate?

Out there, footpods are reported to give a much more responsive reading of speed than gps. People swear by it, especially for sprints. Therefore, this could be a fantastic project. I currently don't have a puckjs, but I should probably get one, even if just to play with, since one stryd (mentioned in the previous question) is about ten puckjs, in terms of money.

Hi! It's worth checking the Bangle had a minute or two to get enough GPS satellites for tracking - if you turn it on and start running immediately, the first few minutes it won't have many satellites, and as it gets more and gets more accurate the position it reports will fluctuate.

I should be able to connect a polar h10 chest strap to the bangle without problem, right?

Yes, absolutely. You just need the bthrm app

Is there any software solution to get a speed on the gps that fluctuates less?

You could for instance store the last 10 GPS values (which occur every second) and then divide the distance between the current position and the one 10 seconds ago by the timestamp between them... Or either average the 10 instantaneous 'speed' values or do a median filter?

There is something called 'kalman filter' that can be used to filter GPS coordinates/speed, and there was a running app that did use it for a while but unfortunately it got broken and wasn't maintained so it's not there any more - but if someone could figure out a way to re-add the code, it might be a really cool addition.

If you seriously want to dig into this, it might be worth setting the recorder to record at 1s intervals (in which case it'll record everything), and then you could go back over it on your PC and see if different algorithms provide better values.

footpods specifically designed to measure a host of running data

I think @Fteacher may have been asking about this before, but I'm not sure anything came of it. I think for some of these they basically just stream the accelerometer data to a phone and get the phone to do the analysis, so it may not work to connect them to a Bangle.

Is a puckjs accurate enough to measure the length of a step while running?

In terms of accuracy it'll be fine - it's probably got the same exact sensor as the Sryd. It's just what you do with that data that matters :)

For instance if the Puck is worn on your ankle then it might twist slightly as you start and stop, the stride, which will mess with the readings - and to work around that you'd have to do proper sensor fusion to use the gyro to also work out its angle so it can apply the acceleration in the right direction - and things get a lot more tricky.

If you can be reasonably sure the sensor is only accelerating and de-accelerating in the same direction each time (which is perpendicular to the earth so you don't have to worry about gravity) then it shouldn't be too hard though.

Potentially even if it isn't 100% accurate, when calibrated against your actual speed from GPS when you're going at a steady speed, it could allow you to get far more responsive values.

If all you care about is the accelerometer, you could actually just use your Bangle on your foot as well :) Although I admit that's not idea for also allowing you to look at speed.

I used running footpods from the days before 😱 wrist based gps and have spent way too long thinking about the subject and looking into it on various projects.
(including one of my more recent projects building a GPS tracker where I spent some time looking at the distance calculation from GPS data)
Basically you're right with your ideas about smoothing and responsiveness of stride monitors. There are actually cheaper footpods than the stryd that can be used too (from both garmin and suunto) although I think those are ANT+ only which isn't (natively) supported by the bangle firmware.
If you look at the footpod profiles for ant (and bluetooth) it's pretty simple protocol and is just basically calculating and giving a "stride" count/length based on a fixed calibration. (the stryd has power and other metrics in addition) Earlier models of garmins etc used to have a calibration step you needed to define this where you'd run for a fixed distance and it would set a ratio. Although I believe there were some auto-calibration options in later garmin versions where it tried to calculate this value in reverse off the GPS data (I assume it detected a straight line for a period with quality GPS)
Interestingly the earlier Garmin e.g. 310XT seemed to actually use that stride data to improve and smooth the GPS tracks it recorded (fellrunr did some tests showing this, also has good commentary and reviews about themin general - https://fellrnr.com/wiki/Footpod), but that seemed to be left to history as the watches improved and I expect they found other tricks to achieve that.

The other problem with the GPS data was the "wobble" this has improved considerably in recent years with the different multi-band chipsets and the overall improvements in accuracy and consistency but can still be an issue where if you're measuring "second to second" whilst you'll get a smooth trace the actual points are actually slightly off by a small amount each second, so for the overall trace it's fine and smooth, but if you actually take the instant data and use that to calculate pace then it jumps back and forth due to the small distances on the calculations.
You can smooth the GPS data to get a more accurate speed data but you need to take into account, as you say, whether the path is straight or "bendy" but it's basic trigonometry, the downside then is the lag you introduce into the returned data.
(also where I was calculating distance taking into altitude and curvature of the earth I did worry about the extra calculations I was doing on battery life!)

The lag is something you definitely notice on a Garmin (without footpod) so I agree re footpods giving more accurate speed data, I've always used them for instant pace data or sprint data, so yes I reckon you definitely could get a puck to calculate this.
I'll try to see if I can find some of my old code to calculate this (albeit that was for TI340 based chip) it was just detecting the "hit" of the foot from the G sensor, then getting the max of the acceleration between that and the next "hit" and from that estimating the speed, would definitely be a fun puck project.

Thanks for the responses.

I decied to go for a Polar H10, actually got in with a discount for some reason. It arrived just now, and I can confirm that bthrm and the hrv app work. Until the two puckjs which I ordered arrive, I might play with accelerometer data coming from it, if I can figure out how.

When the puckjs arrive, I'll see if I can get it to trace out my steps to measure speed and distance, it would be super cool.

if you turn it on and start running immediately, the first few minutes it won't have many satellites, and as it gets more and gets more accurate the position it reports will fluctuate.

I always wait for the "GPS" in the lower left corner of the run+ app to turn green before starting a run (although once I ran out of patience, but usually it doesn't take long). I had a look just now at a gpx file though, and the first turn is clearly way off, but the rest is mostly not bad. There are some squiggles where I go under bridges though, which is another point for footpods, if they work.

There is something called 'kalman filter' that can be used to filter GPS coordinates/speed

Uh-oh, I do know some mathematics, but not so much statistics. I might dive into this to see if I can make it work, but I won't promise anything soon. But thanks for the tip, I'll see if I get anywhere.

they basically just stream the accelerometer data to a phone and get the phone to do the analysis

Do you think the banglejs is powerful enough for this sort of analysis? As a pure mathematician, I don't have a good sense for how much computation power is enough. My sense is that, theoretically, I should be able to use acceleration and gyro data to draw the curve of my foot for each step, using calculus. But wheather the watch (or puckjs) can do the computation in real time, I have no idea.

In terms of accuracy it'll be fine - it's probably got the same exact sensor as the Sryd

I can absolutely believe that :D

I used running footpods from the days before 😱 wrist based gps

I used to use a casio with a stopwatch back in the day. I would prepare and memorize an 8k/10k/14k route on google maps, and then write down my time in a little book :D

But yes, basically any footpod would do, but it seems to me that some of them use ant+, and not bluetooth. Banglejs doesn't support ant+, does it? A quick look at the webpages of some of the companies that sell footpods indicate that they only work with their watches. Even if it is maybe possible to get it to work with other devices, I would hesitate to buy one, if I don't know whether it will give any useful data.

But it seems to me that for such a device, there would be two distinct categories. One which calculates speed (as well as cadence, etc), and then sends that data over. It seems to me that this is what stryd does. The other category would be a device which send over accelerometer data, and then the receiving device is left in charge of computing the data to be displayed and logged. It seems to me that some of the garmin devices are in this category.

The former category would certainly be a nice, quick and easy solution. Even if it wasn't open source, having a black box that gives you this data would be great tool. If the puckjs can be programmed to serve such a function, this would be the best.

As for the second category, if it works, it would also be great. But if I have to program the banglejs, then I would see no reason not to use a puckjs, as long as it is accurate.

One of the difficulties seems to be simply distinguishing these cases, since the promotional material from any company just says it works well with their watch, and you're left guessing.

whilst you'll get a smooth trace the actual points are actually slightly off by a small amount each second, so for the overall trace it's fine and smooth, but if you actually take the instant data and use that to calculate pace then it jumps back and forth due to the small distances on the calculations

This makes a lot of sense, but I am surprised to mostly see banglejs underreport speed compared to garmin. If the points are jumping to the left or right, it should think it was going faster. I think this actually happened at some point when I was going under a bridge, but only for a moment.

Any code would be welcomed, to say the least. I hope I can put a bit of work into this project, and get at least an accurate cadence. It might take some time though.

Speaking of cadence, I'll just mention one last thing. On the wrist, banglejs gives a much lower value than the garmin, and I'm pretty sure the garmin is correct, since I used to actually follow a metronome function on the garmin, to get 180 steps per minute. Also, that data is coming from the garmin run dynamics pod, which clips on the waist, and should be pretty accurate. Unfortunatly, though, it is only ant+ (I think).

I am surprised to mostly see banglejs underreport speed compared to garmin

We report the speed we get from the GPS chip. Inside the chip there's all kinds of insane maths going on so I'm not really sure what happens there, but it's likely a lot more convoluted than just taking the distance between the last and current points :)

Kalman

This was the typescript code for the Kalman filter: https://github.com/espruino/BangleApps/blob/2e8b9ac2cbd0ea70248bbd2ece53ae37203eab7e/apps/banglerun/src/gps.ts#L94

However I'm not entirely sure how well it worked, or whether it worked at all in that snapshot. It could be a good starting point.

My vague memory was it depended on the dop value from the GPS, which is a bit of a 'finger in the air' accuracy figure - but at some point Bangle.js reported that differently and that may have broken things.

Do you think the banglejs is powerful enough for this sort of analysis?

Totally depends on the analysis - if you're looking at doing something simple at 20Hz then totally (and depending on the code you can JIT things or even use C code to get them running faster). If you're doing a bunch of matrix math then it could get tricky though