Those devices are kinda noisy so you will get some of that. That said, ugh, this thing puts out analog that you lug over to the processor... Lots and lots of opportunity to intercept more noise.

Got a picture of your on-bike setup?

If you set the accelerometer on a piece of foam rubber and hold your breath what do you get?

Keith Cress
kcress - http://www.flaminsystems.com

Keith- Dizzy? :)

-AK2DM

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
"It's the questions that drive us"
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

I ran a desktop study. Excel plot attached.
Little noise when sitting still.
Much noise with even very small movement.

1. do I have the wrong accelerometer? It costs about $20 thru Adafruit. I can spend $7 or $100 on an accelerometer. Not sure how to differentiate based on data sheets. All appear to have capacitors on board and very little noise.

2. Is this level of noise - especially the slow movement in the 6 to 8s range - typical? Unfortunately, this doesn't work for me, as I'm trying to measure structural vibrations and the different between 0.05g and 0.08g is non-trivial.

3. Calibration drift. I manually calibrate the accelerometer "by hand" flipping the accelerometer 180 deg for each axis and viewing the output pins (if there's an easier way, I'd like to know). I did notice drift in a short time. A lot of drift even after 3 minutes. The accelerometer was getting warm after 10 minutes, so I disconnected. Is that normal?

What's the best path forward if I'm trying to measure +/- 2.0 g with 2% accuracy, and sampling systems with less than 10 Hz?

https://res.cloudinary.com/engineering-com/image/upload/v1616080672/tips/ACCDATA_desktop20210318_n6enea.pdf

Also, as requested, the precision testing apparatus in my shop (bike crank with accelerometer and bike on a stand)

I suspect the signal wires are too long. The sample rate is set by output side capacitors; it's possible that the length of wire is coupling too much capacitance for the sensor to drive. Putting the microprocessor on the pedal with much shorter connections may help.

See https://www.analog.com/media/en/technical-document...

Just a thought for a quick test- try removing the chain from the pedal sprocket and repeat the slow pedal movement.
Wonder if you are picking up extraneous noise from the sprockets/chain system?

-AK2DM

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
"It's the questions that drive us"
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

This looks very DIY and not work related; as such it should moved to the hobbies forum

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! https://www.youtube.com/watch?v=BKorP55Aqvg
FAQ731-376: Eng-Tips.com Forum Policies forum1529: Translation Assistance for Engineers Entire Forum list http://www.eng-tips.com/forumlist.cfm

IR - I appreciate your sentiment to keep this forum serious and professional.
This is actually serious and professional.
The bike crank in my garage was my attempt at trouble-shooting. To be clear, this is not about bike cranks but instead structural instrumentation on a shake table.
Attached is the shake table accel data, which is the source of my troubles. The bike crank test just provided a confirmation that noise was not just a function of the table.
Shake table input: f = 2.0Hz, A = 5mm. Nema 34 Servo (Teknic) has been auto-tuned already.
https://res.cloudinary.com/engineering-com/image/upload/v1616097386/tips/ACCDATA_2.0Hz5mm_post-AutoTune20210315_wwuhap.pdf

Top view of shake table, with screw and rails

I'm looking at the hand-job graph (snicker) but it's in Gs. What's the voltage that is relating to in your setup? We're trying to estimate what the electrical noise verses the signal magnitude is.

If you take a small loop of wire and hook it to a sensitive scope or meter just waving it around cuts DC magnetic field lines caused by things around you including the earth's own field. The more of the earth's field you encompass the more noise you 'create' with movement.

Typically you'd go to great pains to the shield your setup to prevent this. Then you'd do things like amplify the signal to 0-10V and send that down the long wires to the 'input'. Then any uV or mV noise is washed out.

Ideally you get your seismic system setup such that you send power out to the sensor it does all the G sensing right at that spot then pumps out data RS485 type data (numbers) that can't be bugged by wire runs at all. One normally wouldn't take the raw analog data very far and by far I mean more than 6 inches. If you want a noiseless system don't take it more than 2 inches and do that carefully.

Keith Cress
kcress - http://www.flaminsystems.com

3DDave and itsmoked: Thanks for the comments, even with the snickering.
I've mounted my accelerometer right next to my board with jumper wires about 4". Noise is significantly reduced to the equivalent of about 0.02g (which is good enough for my application of -1.0 to 1.0g range).
Rookie mistake. Next time I'll come to the forum with better questions.

No problem. My first lesson was setting up a very sensitive differential sensor at my then boss's house across from a mountain with an ICBM Air Force Over The Horizon Early Warning RADAR. Every 4 seconds my circuit went bonkers, settled down, went bonkers. Twenty feet of tinfoil helped and proved the problem but didn't fix it.

Keith Cress
kcress - http://www.flaminsystems.com