ADC Multi-sampling

[bill88t]

This PR introduces a way to read multiple ADC samples in one go and additionally moves ADC object initiation and calibration outside of get_value for Espressif. This vastly improves ADC performance on Espressif.

A new AnalogIn property has been added, sample_size which can be both read and written to setting the number of samples taken on every .value read.
Alongside the property, there is also the samples keyword that can be used to set sample_size during the AnalogIn object construction.

The default sample size is set to 2 on all ports. As it was, only espressif did take 2 samples, with all the rest of the ports taking one.

Task checklist:

• Implement on espressif.
• Test on espressif.
• Implement on raspberrypi.
• Test on raspberrypi.
• Implement on atmel-samd.
• Test on atmel-samd.
• Implement on cxd56.
• Test on cxd56.
• Implement on mimxrt10xx.
• Test on mimxrt10xx.
• Implement on nordic.
• Test on nordic.
• Implement on silabs.
• Test on silabs.
• Implement on stm.
• Test on stm.

As a note, I cannot test stm, silabs, mimxrt10xx, cxd56.
Those will have to be tested by someone else.

[dhalbert]

If you reload before .deinitializing an analogio object it cannot be remade until reset.

We can fix this by by making the AnalogIn object have a finalizer. It's on the list to do that anyway.

[bill88t]

Is there any desire to see this completed?

It makes sampling a lot faster.
It doesn't break the existing api.

[dhalbert]

I might have done this for #8690. If you want to complete it to check AnalogIn off in that issue, that would be great. Look for how finalizers are done in, say, #9165. Note that finalizer is spelled finaliser in function names due to the British/Australian spelling difference. The basic thing to do is to say it has a finalizer when you construct it, and to make __del__() call deinit(). You also need to remove any global init/reset.

[bill88t]

Implemented finalisers. Now I just need to implement the sampling logic on all ports.

[bill88t]

Updated the title an opening comment.

Also, @dhalbert turns out it's still possible to trigger a watchdog safemode with just CONFIG_FREERTOS_UNICORE.
You just have to be unlucky enough to .value during a dhcp leash renewal.
Well, with this PR this also goes away, since the object gets initialized only once.
EDIT: I was wrong, microcontroller.cpu.temperature can watchdog safemode.

Do you maybe want me to also include CONFIG_FREERTOS_UNICORE in this PR too?
Or are you going to PR it seperately?

[dhalbert]

EDIT: I was wrong, microcontroller.cpu.temperature can watchdog safemode.

That probably uses a an ADC.

Do you maybe want me to also include CONFIG_FREERTOS_UNICORE in this PR too?
Or are you going to PR it seperately?

No, I am not prepared to turn off the second core yet. I would rather debug the underlying problem.

[bill88t]

Note, the actions on 39d26d6 should not have succeeded.
Viewing them from:
https://github.com/bill88t/circuitpython/actions/runs/9453514997
they appear successful.

It didn't build any boards.

[bill88t]

While I'm at it, should I maybe add a property for configuring sample size after the object has been created?

As is, the sample size can only be set during object creation like a so:
adc = analogio.AnalogIn(board.ADC_PIN, samples=5000)

I kinda think like adding a adc.samples property would be helpful both for visibility and ease of use.

[bill88t]

This PR is pretty much ready, it just need testing.
Please refer to the opening comment.

I will test on nordic and atmel-samd when I have some free time.
EDIT: nordic tested.

[bill88t]

Forgot this existed, tested on samd51. Now I have tested every port I have.

There is nothing else to be done from my side.

[bill88t]

Fixed merge conflicts, please validate against RP2350 too.

adc.sample_size = 65535
adc.value

[bill88t]

I cannot get that one samd board to fit. I don't know what to do from here.

[bill88t]

The only thing I see that could possibly be disabled is CIRCUITPY_PS2IO..

[tannewt]

Honestly, I'm not sure I want to add sample number configuration as an API. I'd be ok if the ESP did multiple samples internally but adding an API impacts all ports (as you see) and all ADC drivers. It is backwards compatible but any code that starts using it won't work elsewhere that hasn't added it.

Multiple samples can be taken from Python too if you really want to too.

[bill88t]

Sure, it's new api, but something that arguably should have been there from the start.
If you need precision you kinda have to take multiple samples.
Taking them from python is fine, but python is notoriously slow on for loops.
This is something not so complicated, that passes on the burden to the underlying core.

but any code that starts using it won't work elsewhere that hasn't added it.

Yea, but how often is an application developed and it's expected for it to work on earlier version of CircuitPython?
Besides, it could trivially be tested with a hasattr, if being version agnostic is a must.

impacts all ports (as you see)

This is why I followed standard procedure and created a task list.
While I do not have every single port, I did test everything I had.
On those I did not have, I was extra cautious with the code as to do it with the least number of changes, to avoid unforeseen issues, skipping optimizations.

[tannewt]

Sure, it's new api, but something that arguably should have been there from the start.
If you need precision you kinda have to take multiple samples.

analogbufio can be used to take multiple samples at once.

Taking them from python is fine, but python is notoriously slow on for loops.
This is something not so complicated, that passes on the burden to the underlying core.

It is simple but likely to not be used. (We've gone 7 years without it.) Furthermore, it has the code size issue on SAMD. Newer APIs like analogbufio don't have this problem because they aren't on the small SAMD21s.

Yea, but how often is an application developed and it's expected for it to work on earlier version of CircuitPython?

I'm more concerned about other places that implement this API like drivers and Blinka.

Besides, it could trivially be tested with a hasattr, if being version agnostic is a must.

hasattr can't check for a kwarg.

I'd prefer not to change the AnalogIn API at all. Instead, I'd be ok adding internal multisampling for ESP.

[bill88t]

hasattr can't check for a kwarg.

It's an arg and an attribute, it can be set during creation or after.
adc.sample_size = 65535
So the check would be: hasattr(adc, "sample_size").

In either case, i'll close the PR. It could probably be optimized to fit, but if analogbufio can do it then it's redundant.