Espruino JavaScript
HRM unreliable with sport activities #7738
-
|
For a time I was wondering why my heart rate was always around 100 on the Bangle.js 2 during intensive sport activities. This month I double measured my heart rate with a Polar H10 device during intensive running. This were the results: So clearly there is something wrong with the HRM of the Bangle.js v2. I am using the Recorder and the Run+ app. |
Beta Was this translation helpful? Give feedback.
Replies: 14 comments 55 replies
-
|
There are a few discussions around this on the forum, but you are comparing an ECG-based heart rate monitor stapped 6 inches from your heart to the Bangle's light-based system which is strapped to your arm and waved backwards and forwards at high speed while you're doing exercise. It's not that there's something wrong with the HRM (you could however check you removed the protective tape from it - that doesn't help), it's just that there are serious limits to what can be measured with a PPG system like the Bangle's during heavy exercise. If you look at the raw data that comes out of it when exercising it's effectively random noise - it's very hard to dig useful information out of it - but that'll be the same with pretty much any device using light to measure HRM on your wrist. You're welcome to experiment with different algorithms to try and make sense of the raw data (I did have a push a while back to try and get data to work with but there was very little interest). I've been through several iterations and have ended up settling on the manufacturer-provided algorithm, which seems to be a decent compromise for most situations. There is however an app which allows your Bangle to connect your Polar H10 and then record the Polar's HRM data instead of the inbuilt one, so that could be an option if you want super accurate heart rate without needing your phone. |
Beta Was this translation helpful? Give feedback.
-
|
This is interesting. While it is really tricky and mostly it indeed doesn't work very well once the PPG sensor is moving rapidly, I am still surprised the manufacturer provided algorithm gives such bad result for you. I have one SMA Q3 watch still with original firmware and I did use it while running many times and it was mostly reliable in the sense that it went to 160-180 rates and I could somehow use that to slow down or speed up to keep the HR in some interval. And it should be hopefully the same algorithm binary used there. This is much older revision of the watch with no barometer hole and older VC31 HR sensor. It would be interesting to test the current SMA provided watch firmware for current Bangle 2 watch model if the issue is there too. Which reminds me that I wanted to test how to restore SMA firmware and get back to Espruino just via DFU over bluetooth. If it works then more people could test it if interested. And BTW @lunctis-viribus it may help to keep it more tight so the watch vs your skin moves less. While even the rapid movement of watch tightly strapped to the hand is not ideal for PPG, it is much worse if the watch is loose and moves on your hand, then it is really just a random noise. |
Beta Was this translation helpful? Give feedback.
-
|
@lunctis-viribus We actually just published data examining the validity of PPG signal with the Vcare algorithm vs the Polar H10 (and also the step counting algorithm within the Espruino firmware). These data reiterate the answer by @gfwilliams. If you want high accuracy HR at higher intensities, especially when you have high arm movements - best to pair a HR strap. This is true for other wrist based PPG sensors; although their motion artifact detection is getting progressively better as technology matures. |
Beta Was this translation helpful? Give feedback.
-
|
Great paper! It indeed confirms that when moving, the Bangle.js2 has a significantly lower bpm than the Polar H10, especially when bpm is 100+. This explains the low CCC of 0.54 for Steps > 0. Thanks for mentioning and publishing this paper @nravanelli ! @fanoush next time I will tighten the watch strap, thanks for the suggestion. |
Beta Was this translation helpful? Give feedback.
-
Beware of this though: https://github.com/orgs/espruino/discussions/7319 I woudn't worry if you use a silicone strap since it's so flexible. I use a nylon fabric strap that doesn't flex and if tightening it too much the lugs are at higher risk of breaking. |
Beta Was this translation helpful? Give feedback.
-
Thanks. Actually, this is my second Bangle.js2 because a broken strap connection is exactly what happened with my previous one. And I am using a silicone strap band currently. |
Beta Was this translation helpful? Give feedback.
-
|
ok, it is an old thread, but it fits for my question: I observed some strange behavior: When I start the HRM and do not move my arm the messurement is quite good and keeps being good. But when I move the arm, the watch ore even have the watch not on the arm during the first seconds then the internal HRM goes to something around 100 bpm. And it stays at this number for very long, even for an hour in one testing. At this time the BT-HRM was about 50-60 all the time. I need more investigation with this with some sports in the following days. I will compare it with another optical HRM too (Sigma id.free). This device sometimes goes mad also... @gfwilliams you posted that you tried some other algorithm. Is there a way to test these other code? Or was it just an idea and nobody tryed it in the bangle? Would be great if we could improve the HRM a little bit... /Hendrik/ |
Beta Was this translation helpful? Give feedback.
-
Out of how many BPM? Someone (@fanoush?) did spot that it was reading ~10% low but that was over a year ago and I thought that problem had gone away. The original code attempted to 'calibrate' the HRM speed and we don't do that on Bangle.js, and I wonder whether that could cause some issues.
Yes - and then they'll continue to take effect between apps until you next do a complete restart of the Bangle. |
Beta Was this translation helpful? Give feedback.
-
|
I will test the different modes. And I have to do some more investigation. |
Beta Was this translation helpful? Give feedback.
-
|
here another one. HRM Motion Artefacts removal" switched to on Bangle.js Track 20250622 difference HRM walking HMAR on.xlsx
In my opinion this is quite good... |
Beta Was this translation helpful? Give feedback.
-
It did not go away, I just don't use HR on bangle anyway so did not care too much, and yes it was about 10% less when compared to the older open source algorithm and to other devices I have. But if others did not see it, it could be just with the older VC31 sensor version or something. This is on the old pre-kickstarter version without hole for the pressure sensor. Anyway, I will check again if is is still the case. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks! it'd be really good to know. I guess it's possible that the binary HRM algo expects data every 100ms and throws away readings outside that, so depending on how the readings/sec is calibrated it might end up +10%,+0% or -10% of where it should be if that's the case? Hard to be sure though as we don't have docs on how it works |
Beta Was this translation helpful? Give feedback.
-
|
ok, I have another one. Switched the recorder interval to 1s. I have to double check if this makes a difference...
|
Beta Was this translation helpful? Give feedback.
-
|
The interval itself shouldn't make a difference, but it'll record more data and because there could be a 1 sec interval without a heart rate some of the readings will be blank. As you say it seems strange it doesn't seem to have got anywhere close. What sport/etc were you doing when you recorded this? Just checking, but you definitely removed the protective film from your HRM? |
Beta Was this translation helpful? Give feedback.
-
|
Yes the protective film is removed. This sport was cycling. So no wild arm movements, only vibraitons... This is the kurve for the "walking sport" from 22.6.2025. It was the same bangle... And here it was good... I will do further testings. I still think that the kind of starting of the measurement makes ad difference...
|
Beta Was this translation helpful? Give feedback.
-
What do you mean here? BPM readings from the HRM are 'pushed' to the software on the watch (rather than requested) so the recording interval shouldn't really have an effect |
Beta Was this translation helpful? Give feedback.
-
|
I think if the measurement is good at the starting it keeps good. And if the BPM reading at the start ist way off it often keeps to be way off. |
Beta Was this translation helpful? Give feedback.
-
|
Ahh, I don't know how their algorithm works but I guess it's possible it's happier when it knows a heart rate and can then just 'follow' it. But also if it's better one time than another it could be related to how tight(or not) the Bangle is on your wrist. It would be interesting to look at the 'Confidence' reading you get for the internal HRM over that time - whether the algorithm thinks it's actually got your pulse or not |
Beta Was this translation helpful? Give feedback.
-
|
I updatet my post with a new graph... |
Beta Was this translation helpful? Give feedback.
-
|
Thanks - it seems lot more confident than I would expect! What I find interesting is that the HRM values do go up and down at about the right times - if you took the blue line and just multiplied the values by 1.35 it'd actually be quite close to what the other one reports! |
Beta Was this translation helpful? Give feedback.
-
|
Good morning, I did two other testings with my two bangles. They were both disappointing. Left factory reset one and right my standard watch with 50 apps... IMG_E4263_crop.MOVPerhaps we have a performance problem or just one app what drains too much cpu... |
Beta Was this translation helpful? Give feedback.
-
|
Hey we are getting closer to the solution. and then (drum roll...)
and next: Test which app is draining the power here... (ios integration? weather? or just the Circles Clock with all the clockinfos?) |
Beta Was this translation helpful? Give feedback.
-
|
I found out: The app "HRM Motion Artifacts removal" causes the stutters being visible in the "Heart rate monitor". Switching it off in the settings causes in a much better HRM. Perhaps this is different if we are running with the watch... |
Beta Was this translation helpful? Give feedback.
-
|
yes it looks so. I can reproduce it with switching the algorithm on and off in the settings. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks - I'll add a warning to the app. I'd suggest you uninstall it and don't use it. It was something that was contributed but I have no real control over how well these apps work in reality. The HRM manufacturer would say to use the output of their algorithm directly without passing it though other filters (like that app). |
Beta Was this translation helpful? Give feedback.
-
|
@gfwilliams I want to test the different modes, but I get this answer in web IDE:
is there a tutorial how to interact with the bangle with webIDE? |
Beta Was this translation helpful? Give feedback.
-
|
sorry... writing the command in at the right side of webIDE and then sending it to RAM is working... |
Beta Was this translation helpful? Give feedback.
-
You should just put it in the left-hand side, yes. The fact it writes If you ran |
Beta Was this translation helpful? Give feedback.
-
|
Hi folks, I think we must definitely have a bug here. In the Bangle or in the HRM firmware.
After this procedure the bangle will continue to display something around 100bpm and will perhaps never catch up with the correct heart rate. In my case this is easy to reproduce because I have a resting heart rate of about 50-60bpm. Here is an example with two bangles. Left side: correct hrm, right side: crazy hrm, and you can see that they display the same HRM raw data graph...
Sometimes I even got this bug to show with only a piece of cardboard or paper. So the bangle shows a bpm reading without any skin around. Can I do further testings to clarify this? Both bangles had the "HRM MA removal" off. Left bangle is with very little apps installed, right side is packed with 50 apps. But it can be reproduced on the other bangle. |
Beta Was this translation helpful? Give feedback.
-
|
(( I feel y'all will already be dealing to this, but in case it helps )):: The only times I've found the HR unreliable (iirc like twice as high; can't recall if it always returned the same fixed high value), I gave the sensor a wipe (with my T-shirt of course! 😅) These days I just try and give the sensor an occasional wipe when I have to take the Bangle2 off. Sun & skin creams are in use. I have isopropyl alcohol and lint-free wipes, but 🤷♀️🤷♂️ T-shirt has continued to be effective so far. |
Beta Was this translation helpful? Give feedback.
-
|
@gfwilliams Sorry!! I missed your post here in this place... This multi-threaded view is very irritating for me... The newest post can be anywhere... |
Beta Was this translation helpful? Give feedback.
-
|
wow, my account was blocked the second time now after posting the youtube link. |
Beta Was this translation helpful? Give feedback.
-
|
We've had quite a few spam posts that I've had to report recently so maybe GitHub is on 'high alert'. One thing I've noticed is your links seem to be messed up. I think you click the If you just want to put a link like https://youtu.be/-LkQbmKFZks - just paste the link in amongst the text as-is without clicking the 'add link' button and it'll be fine. Thanks for the video - looking at it, it feels like the algorithm isn't really 'locking up' - around 2 mins, it is actually changing its readings between 116-120, so it feels like it's consistently reading double the correct HRM. The 'confidence' value is very low too, so in the Health app where it has a bit of time to get the correct HRM reading, it'll be ignoring anything with a confidence below 90 anyway. I'd expect you can fool a heart rate by rubbing your finger on the HRM sensor, and then when you keep your finger on it's thinking it still has the same rate but it's just missing every other beat. Obviously you'd hope it would figure it out eventually - the 'open' heart rate algorithm we had before wouldn't have done this but then it was worse in other cases. I feel like if you placed the watch on your wrist where there's a bit more bloodflow, but might be able to detect your correct pulse a lot better? The end of your finger's not going to be great for it as the skin is quite thick there. I think in this case (if that's what the problem is - reading double the rate) I'm not sure there's much that can be done. It's a tricky position for me because I'm trying to get this as good as possible, but I basically just get complaints regardless of what I do. It's just with the current algorithm I get quite a lot less complaints than we had before. |
Beta Was this translation helpful? Give feedback.
-
|
@gfwilliams thanks for explaining the problem with the link. In every other forum I know I have to use the Link-Button... I am quite sure that the sensor just is not able to do it better. We can see it at the "heart rate monitor": I am cycling and turn my wrist to look at the watch and the turning is enough to make it necessary to readjust amplification. So too much movement at the starting sequence lets it take the "wrong path"... I would say we leave it like it is... It is just a optical sensor... Lets concentrate on other issues :) thanks |
Beta Was this translation helpful? Give feedback.
-
|
ok, let me add some positive results:
|
Beta Was this translation helpful? Give feedback.
-
|
I found a perhaps interesting article: https://www.mdpi.com/1424-8220/25/9/2926 I tried to downgrade to 2.19 but the bangle is not booting with a firmware younger then v2.26 Perhaps the situation is different when we wear the bangle all the day and the heart rate is measured every 10min. If we start the sensor and don't move in the first 20sek the sensor mostly don't hook on the wrong reading. I now started recording the HRM with the health app. Is there a possibility to download the data recorded by "health"? |
Beta Was this translation helpful? Give feedback.
-
I've shipped about 10,000 bangle.js 2 now I think, and while one or two people complain, if the sensor was completely useless there'd be an awful lot more complaints. We've been running with the binary heart rate algorithm for the past 2 years and this is the first time anyone has mentioned having your problem.
Yes - you just go to the health app in the app loader (same as you might for recorder) and you can download the info for each month as a CSV file. Definitely if it's just turning on and off every 3/10 minutes you're not going to have that issue. |
Beta Was this translation helpful? Give feedback.
-
|
The VC31B sensor is used in many lowcost smartwatches. Would be interesting to see if they have the same behavior... |
Beta Was this translation helpful? Give feedback.
-
Well you're welcome to try - but as I say I can't reproduce it. At the end of the day, I feel like the 'health' app will probably work fine for you day-day, and if you're really wanting to record accurate live heart rate during exercise you've got the ECG which works with the Bangle so I'd use that.
Do you have a link or some way to find it on that page? But those posts are from 2 years ago and I have made some tweaks since then |
Beta Was this translation helpful? Give feedback.
-
|
search for "Posted at 2023-12-30 by Graham_Jones" Sorry for my complaining. Perhaps all optical sensors have similar problems and all other 10000 Bangle users are not looking so close to this issue. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks! Just looked at those posts - I see he mentions 100bpm there. I'm pretty sure that what he's mentioning with I2C errors and a reset isn't your problem though as we can actually see your heartbeats in the screenshot (the red graph at the bottom). He may also have the older VC31 sensor (not VC31B). It's hard to know what to do really - the downside of having the proprietary heart rate algorithm is that it is just a 'black box' - and as far as I can see we're feeding it with the correct information (eg you can see a heart rate in the red graph) so if it stops working it's hard to know what to do. The only thing we feed it (in addition to the ppg/env values) is whether the LED intensity was adjusted. You could look at the I guess one option is we could log all your raw HRM data to a file, and then I could see if i can reproduce it with the algorithm here? If I can detect when it's got broken we could reset it in software |
Beta Was this translation helpful? Give feedback.
-
|
perhaps you are right and we cant do anything.
sorry, I don't know how to do this. I think at this point I can not help anymore. But perhaps there could be a workaround: For me it seems to be a problem with some kind of calibration when it starts the measurement. We could restart the algorithm every couple of seconds. Just like lifting the skin from the sensor for 2" because then it starts again and perhaps don't take the "wrong path". Perhaps I could try to modify a app with this cyclic command... (one more motivation to go through the tutorials :-) ) |
Beta Was this translation helpful? Give feedback.
-
|
(now after 20min the Sigma iD.Run switched to the correct reading. Just as I have seen with the bangle a few times) |
Beta Was this translation helpful? Give feedback.
-
|
Has the discussion here been cleaned up? I cannot see a lot of relevant recent posts of the last days by user @eisingpferd and the last one by @m3iow3r (possibly an AI bot) |
Beta Was this translation helpful? Give feedback.
-
|
I think they've been re-added now - sounds like @eisengpferd's account got blocked for some reason |
Beta Was this translation helpful? Give feedback.
-
|
my account was unblocked one hour ago... |
Beta Was this translation helpful? Give feedback.
-
|
Hej I just want to share my log files. I testet a couple of times with the bangle2 and my BT wristbelt (Polar H10). The bangle was quite tight at the arm and it was exactly the same tightness all the time. 
|
Beta Was this translation helpful? Give feedback.
-
|
ok, here is a graph with the confidence for one of the examples. 
|
Beta Was this translation helpful? Give feedback.
-
|
Thanks! It feels like the confidence there is almost useless. What I find strange is that the HRM readings often go up and down in the roughly the same places, even though they're 2/3 of the correct value. I know you were experimenting with forcing the sport mode. Are you still doing that, or is the sport mode automatic? I'm just wondering whether it's switching in and out of sport mode, and when in sport mode it's reading the wrong HRM? |
Beta Was this translation helpful? Give feedback.
-
|
my last tests were all with default settings. And you told me that it is the "normal" mode without trying to compensate any movement. |
Beta Was this translation helpful? Give feedback.
-
|
With the default settings, Bangle.js swaps automatically to sport mode when it detects a lot of movement, and I thought that might maybe be what's causing those periods of 2/3 heartrate - but if you've tried it specifically with It's frustrating as it's obviously reading heart rate fine sometimes, and even when it's not there is some correlation, just ~2/3 out. I mentioned in another post but we did have the open heart rate algorithm, but users were complaining about it and there wasn't much I could do to improve it with just the data I had collected, so we swapped to the binary algorithm. But the flipside is while it's better, there's not much we can really do to debug/change/improve it when it's not working. I'm sorry it's not working reliably for you, but in fairness I don't and have never advertised the Bangle as a fitness watch. I'm one developer, and I just don't have the time or resources to ensure the HRM works reliably in all situations. |
Beta Was this translation helpful? Give feedback.
-
|
it is ok Gordon. As I writed before even two other HR-watches i tested have the same problems and one of them (Sigma id.run HR) showes the same behavour with sticking at a wrong value for longer. The Bangle.js2 is great because it can be conected with a BT wrist belt and even with cycling sensors. So it can be used as a real sports tool. cheers /Hendrik/ |
Beta Was this translation helpful? Give feedback.
There are a few discussions around this on the forum, but you are comparing an ECG-based heart rate monitor stapped 6 inches from your heart to the Bangle's light-based system which is strapped to your arm and waved backwards and forwards at high speed while you're doing exercise.
It's not that there's something wrong with the HRM (you could however check you removed the protective tape from it - that doesn't help), it's just that there are serious limits to what can be measured with a PPG system like the Bangle's during heavy exercise. If you look at the raw data that comes out of it when exercising it's effectively random noise - it's very hard to dig useful information out of it - but …