Acceleration without gravity

[espruino-discuss2]

Posted at 2023-05-13 by AccMagno

Hi,

Is there any good ways to get acceleration sensor data without gravity? Like it would use magnetometer to get the orientation and then remove gravity based on it? Some ready made examples.
I would like to calculate 3-4 seconds the speed the watch has and if possible, also a location/orientation.

[espruino-discuss2]

Posted at 2023-05-14 by HughB

location will be GPS lat/lon ? It will take you at least 20 seconds to power on the GPS and get a fix and that is assuming you have downloaded AGPS data - otherwise expect 12 minutes for the first fix.

---

Posted at 2023-05-14 by Mi

I guess you not think of GPS but double integration of acceleration.
Be warned, that is very difficult - basically an ill posed problem for several reasons!

Main thing is that every sensor is noisy. Integration amplifies this noise, so you will have wrong speed. This will change your location all the time without bounds, even if in reality you are standing still. So after double integration you will have very big error after very short amount of time.
The only way you have a chance to get reasonable location from accelerometer is to make use of some other constraints.

That could be the Magnetometer if you operate quite close to a strong magnet, but no way to count on earth magnetic field (lots of noise from all kind of metal and fields in the environment)

Depending on what you want to do, you might think of instead make use of the gravity, since this is something that can anker you. But of course only if you can discriminate gravity and acc through some other constraints or system model (not magnetometer unfortunately).

All this is not specific to bangle but true even for much more expensive sensors.

---

Posted at 2023-05-15 by @gfwilliams

Yes, I'm afraid it's an almost impossible problem with only the accelerometer... A Gyro would help no end but the Bangle doesn't have one (Puck.js does).

But, one hack that might help you is you say that when you're not accelerating, gravity should be 1, and the accelerometer will be pointing towards the earth. So when that happens you know kind-of where you are. For example:

var Vec3 = require("Vec3");
var G = new Vec3();

Bangle.on('accel',a=>{
  if (a.mag>0.98 && a.mag<1.02) {
    G = new Vec3(a);
  } else {
    var diff = G.sub(a);
    print(diff);
  }
});

---

Posted at 2023-05-17 by AccMagno

Yeah, I was wondering why bangle doesn't have gyro. I thought magnetometer is there for that.
Your idea is good. I'm drying to measure hand speed when you throw disk or ball, so there is time frame when there is no acceleration( I can get the gravity direction) and the throw will last 2s so I might get the right speed and maybe even hand location.

---

Posted at 2023-05-17 by @gfwilliams

That's interesting - you can put the accelerometer into a higher speed mode as well (check the acceleration recorder source). I would have thought for something like throwing you may well be able to at least get some values that are useful for comparison.

Because your arm will be rotating the actual velocity may be different, but it may be some more information can be gathered using the compass as well.

---

Posted at 2023-05-19 by AccMagno

I have been using "Acceleration Logger" app and it gets the data in 12.5Hz. What would be the max logger frequency (about a guess) and how I can get "higher speed mode" on?
I got some data, but it is not accurate enough or I have to analyze it more.

---

Posted at 2023-05-19 by @gfwilliams

Ahh, right - I was thinking of https://banglejs.com/apps/?id=accelrec but that's only available on Bangle.js 1 at the moment.

I've just fixed that and it's in the development app loader, so you can check out how it gets the higher frequency by poking the accelerometer itself - right now it runs at 100Hz but you might be able to push it higher.

---

Posted at 2023-05-22 by AccMagno

" I have just fixed that.." Where is that? I can't find any new commit on "Acceleration Recorder" or on the " Acceleration Logger" app in development branch :/

---

Posted at 2023-05-22 by @gfwilliams

It was here: espruino/BangleApps@796bfda

---

Posted at 2023-05-22 by AccMagno

I could swear that wasn't there for two days. I looked through all your commits and I didn't see that one. Thanks!

---

Posted at 2023-05-22 by AccMagno

It works, but it only records for 2s ( or 5 second as code says so). I need to record 30min. What if I increase sample size, like 1800 * Hz. Will my watch burn or something else? Is 256 kB ram enough?
Other option is to modify "Acceleration Logger" app and change setPollInterval to lower value. Maybe like 10 ms. Documentation mentions "max value" for this but not a minimum :D

---

Posted at 2023-05-24 by @gfwilliams

How's it gone from "I would like to calculate 3-4 seconds " to 30 minutes?

Increasing the buffer size is ok, but you have 6 bytes/sample - and lets say you have ~150k free RAM, that's 25000 samples, or 250 seconds at 100Hz.

If you want more, you're going to have to save to flash memory. Lots of info on how to do that at http://www.espruino.com/Data+Collection#flash-memory - I don't see a problem writing 100Hz, but you may struggle getting it higher.

You're not going to 'burn' your watch playing with settings, but I guess you might crash it and have to long-press to restart. Have a try and see what works.

---

Posted at 2023-05-25 by AccMagno

Good to know that I can't 'burn' it :D Then I will test different thinks.

"How's it gone from..." ah yeah, sorry for that. The idea is measure now the data and explore it and see how it behaves. Then later I will make an app that shows you live how many throws you have had and speed of the throws and so one. So I don't eventually need to measure 30 min, just now to collect the data from the game. In the future app I need to have live array or list of acceleration data and then update it by adding new one and removing oldest. Then when the app measures spike on the acceleration, it can calculate all the data from the array. But first I need to explore the data. For example I was expecting do see from data, which way the hand is moving (backwards or forwards), but for some reasons I cant see that. I only see spikes on x-axis but all the spikes are negatives, so I need to explored more why. Maybe the hand just turns when there is bull forward.

But now I think I will modify the "Acceleration Logger" app and add option to change the data collection frequency, so I can see what is good enough, so when I will make the app, I know what frequency to have and it doesn't use all battery in one call :D

Thanks for the "http://www.espruino.com/Data+Collection#­flash-memory " link. This was good!

I still feel I need to order second watch, just in case my one breaks down ^^ or is there there bangle.js 3 coming with gyro ;)?

---

Posted at 2023-05-25 by @gfwilliams

I still feel I need to order second watch, just in case my one breaks down

Well if you want to that's great, but I wouldn't do it because you'll be worried you'll break it - they're made for this. Look at it this way - we've sold around 10,000 Bangle.js 2s, mostly to folks who spend a lot of time messing around with all kinds of things on them, and I don't believe I've had a single one returned that was broken by something done in software.

is there there bangle.js 3 coming with gyro?

Not any time soon I'm afraid, I'll be sticking with Bangle.js 2 for quite a while!

---

Posted at 2023-05-27 by Philip

I'm [trying] to measure hand speed when you throw disk or ball,

If I am throwing a frisbee, the actual throw probably takes place over a quarter second or less. The acceleration due to the throw could be multiple orders of magnitude greater than gravity.

I have heard quotes that a frisbee throw can involve instantaneous force vectors of 150 times the force of gravity. If you just ignore any data point where the accelerometer is reading a vector with a magnitude less than 10 Gs, and then don't even bother to account for gravity, you will probably be able to end up with less than 5-10% error overall compared to if you did account for gravity (which also might not be perfectly accurate, but that's a different problem).

https://www.youtube.com/watch?v=x4wnbKYfPLs

On top of that, if you have a magnitude limit, you can throw for 30 minutes and end up with only like 10 seconds of actual data that you need to save.

---

Posted at 2023-05-31 by AccMagno

just a update:
tried simple trick by editing Accellog app by changing the "Bangle.setPollInterval(x)" value to 50 ms. First sample gets in around 50 ms, but rest of the values have around 100ms interval. I tried to disable layout updating, because maybe that takes too much time, but didn't help.
Maybe I need to go with accelrec app and modify it. Maybe it can't write to file so fast. I will continue to try different things.

you can throw for 30 minutes and end up with only like 10 seconds of actual data

True. Now I just want to collect data. If the writing to text file takes too much time, I can make it write only when there is a peak. Otherwise I will just update an array on the ram with new values. Lets see.

---

Posted at 2023-06-05 by @gfwilliams

The app does a bit more than changing just the poll interval.

The poll interval is the rate the accelerometer is checked for new data - but the accelerometer is only recording at 12.5Hz by fault anyway - it needs configuring to output data faster using accelWr like the app does.

---

Posted at 2023-06-06 by AccMagno

Yeah I was watching this from the accelrec, but I didn't find any documentation about the options and what to write.

  Bangle.accelWr(0x18,0b01110100); // off, +-8g
  Bangle.accelWr(0x1B,0x03 | 0x40); // 100hz output, ODR/2 filter
  Bangle.accelWr(0x18,0b11110100); // +-8g

What happens in here?
First you turn 8g off?
Set the update ratio to 100 hz with 0x03? How can I set it to 50 or 30 hz?
What is ODR/2 filter. Di didn't find anything about this topic.
And then in the end you turn 8g back?

update: reading from the datapage, I can see why you have to turn the acceleration on and off and what are the update options.
But still don't understand the '(0x1B,0x03 | 0x40)' part, what happens in here?

---

Posted at 2023-06-06 by @gfwilliams

0x03 | 0x40 just means an 'OR' - it's the same as writing 0x43. It's just configuring the sensor to output at 100hz with a filter on it to stop it being too noisy (and you have to do that while the sensor is turned off).

---

Posted at 2023-07-03 by AccMagno

Ok this project is on hold, because I can't connect to my watch anymore. Have to get new one. There is no possibility to connect it by cable, is there?

---

Posted at 2023-07-03 by @thyttan

Have you tried finding the watch with one of the 'nrf Connect' apps? https://www.espruino.com/Firmware+Update#android-ios-app

(Or toggled the different bluetooth settings, doing factory reset from the watch etc...?)

---

Posted at 2023-07-03 by @gfwilliams

You can't connect with the cable, no (apart from for low level debug access, which is not what you want).

But it's extremely likely you can get bluetooth to work again. Sometimes it's even just that your PC gets confused and your computer needs a restart before it'll see the Bangle again

---

Posted at 2023-07-03 by AccMagno

I did everything from other forum post: reset settings, factory reset, scan with different deices, all the bluetooth settings ok, no pass code, pinch from the corner... nothing for a 2-3 weeks. It is not the first time, I lost it, but now I think it really break.

---

Posted at 2023-08-09 by AccMagno

I got a new one. Testing is continued

---

Posted at 2024-04-21 by AccMagno

Now I'm back and started again. I was testing the Acceleration Recorder app and that is good one. It is recording everything with 100hz and it works. I tried even with 400hz, but then the file size got too large and it could not write the csv file ( is there way to see how much space I have?)
But then I was trying to get it work with 200 hz and it started to give me a error:

 Error: Invalid interval
 at line 1 col 203
...;Bangle.setPollInterval(5.0);setTimeout(function(){Bangle.on("accel"...
                                     ^

I did manage try it with 2.5 pollInterval, but suddenly I can't anymore try anything else that 10 ms pollinterval. Any ideas why? Here is the edited part of the code:

HZ = 200
  //Bangle.accelWr(0x1B,0x03 | 0x40); // 100hz output, ODR/2 filter
  Bangle.accelWr(0x1B,0x04 | 0x40); // 200hz output, ODR/2 filter
  Bangle.accelWr(0x18,0b11110100); // +-8g
  //Bangle.setPollInterval(10); // 100hz input
  Bangle.setPollInterval(5); // 00hz input
  setTimeout(function() {

---

Posted at 2024-04-21 by AccMagno

double message. skips this

---

Posted at 2024-04-22 by @gfwilliams

is there way to see how much space I have?

require("Storage").getStats()

Ideally if you're writing data that fast, you could look at writing binary data - see the second example on https://www.espruino.com/Data+Collection#flash-memory

It looks like we don't support a poll interval less than 10ms (100Hz): https://github.com/espruino/Espruino/blob/8f3a9cb52/libs/banglejs/jswrap_bangle.c#L2487

So you can't do above 100Hz. I'm not sure what happened with 400Hz but maybe it failed with out of memory before it got started.

The Accelerometer does have a built-in FIFO which we don't actually use at the moment.

It might be possible to change the Espruino firmware to use it, but also you could probably use accelRd to read the data directly from the FIFO at a much lower level, and then you don't need to change PollInterval at all.

---

Posted at 2024-04-22 by AccMagno

Ok. No worries.
Yeah, I was just wondering how the 400 hz first worked and then not :D
But the 100hz looks much better than the 12.5 hz. I was just wondering am I missing something with 100hz, what can I only see with 200 or 400hz.
But I will test with this and will report in here.
By the way Gordon, thanks for everything. Really nice to have this tool. And thanks for all the effort you put in <3

---

Posted at 2024-04-27 by AccMagno

I'm getting now some weird values. Image of values In some point, the x value reverses to positive value from negative. Any ideas why? Something to do with int16Array format?

And any idea why there is a scale of 5000 and that multiplied with two like in line 27:

accelx[i] = a.x*SCALE*2;

---

Posted at 2024-04-29 by @gfwilliams

The scale is there to ensure that the values which were fractional can be saved in the Uint16Array (which is much more efficient memory-wise). I have no idea why it's 5000 though.

The strange values will be because you're approaching such high G that the value is going above 32767 (the max range of uint16array) and wrapping over.

Maybe just try using a smaller number for SCALE and that'll fix it

---

Posted at 2024-05-13 by AccMagno

So I got more data.

1. The number two in the scale value is for 8g mode. If you don't add that, it the values will be wrong.
2. The scale value has nothing to do with the hitting the limit. When the sensor hits the 8g limit, it starts to flip from -8g to positive 8g value. Don't know why.
3. The timestep is not always 10ms. Sometimes it is 20 or something else, so I have to save also the timestamp.
4. If you take the magnitude value, its limit is something around 10,5 g. But if you get the x,y and z value and and calculate the value from that, you can go over that limit(if the x, y or z value don't hit the 8g limit). If the sensors magnitude value hits over 10,5 g, it goes to zero
5. Got some results and made a video.
   The results look good. I can get really good data.
   Now with this data I can start to build my own "throw" app. Lets see how that will go. I will post results in here.

---

Posted at 2024-09-18 by AccMagno

update:
Now I have two new puck.js sensors! Lets see what I will to with them.
I also made web interface for bangle.js based on this: https://github.com/AkosLukacs/PuckStreaming
My app works little different: every time watch detects throw, it send the record to the web interface. Next step is make same communication with puck.js, but it looks like the code for puck is little harder than bangle.
I already made a holder for buck, so I can measure forces with puck. (see the picture)
I also update the video to make it more clear, what happens.
https://www.youtube.com/watch?v=XkkYb0-aGRo

---

Attachments:

• Puck_kyykka_anturi.jpeg
  
• bangle_js_web_app.png

---

Posted at 2024-09-18 by @gfwilliams

That's great! Thanks for the update!

So the plan is to put the Puck into the things that are knocked over, rather than the thing that is thrown?

---

Posted at 2024-09-19 by AccMagno

Both. I'm also developing a sensor place in the bad also. But this is start.
Should I start a new thread as "project" ?
Need to see how can get puck.js connect to webpage same way as bangle.js and send the data it has recorded.

---

Posted at 2024-09-20 by @gfwilliams

It's up to you, but it seems as this thread is already started you might as well ask here as it'll be for very similar things.

---

Posted at 2024-12-18 by AccMagno

I use bluetooth command in bangle to send the data package, but in puck.js you can use bluetooth and nrf to send data. Witch one is better in this case? I will send data only when throw is done.
Wondering also can I use same method to record acceleration as in bangle.js ( add/replace value in array with new one) or should I use something else.

---

Posted at 2024-12-19 by @gfwilliams

but in puck.js you can use bluetooth and nrf to send data

I don't really understand what you mean here - communications options between Puck.js and Bangle.js should be basically the same.

The NRF.* methods are for bluetooth, and are available on both. Bluetooth.print/etc is still Bluetooth, but is Bluetooth UART.

The Bluetooth UART provides buffering and slightly faster transfer, so I'd say use that if you can

[pinq-]

So now we are here. Looks good.
Question: is there a good way to ensure, that bluetooth message has been received? I will try to send a package from puck.js or bangle.js to browser, but I want it to try send as long ( or try 10 times) as the message is received.

[gfwilliams]

Bluetooth itself should pretty much ensure data is reliably sent (I believe it retries if it fails) so the only issue would be if the connection disconnected while you were sending.

You could have the browser send a response message down bluetooth to show that data was received, but if you end up waiting for the response every time that could slow down transmission.

[pinq-]

My problem is that sometimes puck.js goes too far and then it looses the connection and then the data is lost.
I was wondering can I use BLE to advertise the data for 10 times or something and then continue. I doesn’t matter if it slows down after the data collection, if it waits until the data is sent. Maybe just check if bluetooth connection is on and if not then wait.

Also I noticed that if I try to run the acceleration sensor with 208 Hz, the max that I really can get is ~142 hz (~7 ms). Is there any way I can improve the speed?

[gfwilliams]

With advertising there's no two-way communications, but you could advertise the data plus the previous data, and then leave it to advertise a few times before you change it. But IIRC the max rate for advertising is 20ms, which is only 50hz even if you update it each time.

You should be able to get 208Hz I believe but the the speed is dependent on how much JS you're running for each data point you receive. I don't recall seeing your code but at minimum you could enable minification and pretokenisation in the IDE, and maybe you could look at whether you can make your code a bit faster