STM32 ADC: Support differential mode

[mrodgers-witekio]

Is your feature request related to a problem? Please describe.

Differential mode is currently not supported in the adc_stm32.c driver. In adc_stm32_channel_setup:

	if (channel_cfg->differential) {
		LOG_ERR("Differential channels are not supported");
		return -EINVAL;
	}

Describe the solution you'd like

Support for differential mode specified in the usual devicetree bindings.

I had a quick look at what it would take to implement this, and there seems to be some added complication around differential mode (at least in the F3 part I'm looking at the reference manual for, not sure if the same applies to all STM32 series):

1. Differential/single ended mode selection can only be performed while the ADC is disabled
2. Calibration is performed separately for differential/single ended modes, so the driver needs to know during initialisation if any channels are using differential mode so that it can perform the correct calibration routine(s).

I think the simplest way to implement this would be to set the channel mode to differential during initialisation for any channels which have the zephyr,differential property in the devicetree. Then the driver knows during initialisation whether it should perform a differential/single ended calibration (or both). If the user calls channel_setup with a different single-ended/differential setting to the one configured during initialisation, it could just return an error like it does for differential mode currently.

Even better would be if it's possible to dynamically change between differential and single ended mode for a channel, but I don't really have a need for this and I'm not sure whether the hardware would support this well (ie. is it possible to disable the ADC, change the channel config and re-enable it without performing a full re-initialisation?)

I'm happy to give this a try and open a PR, if the proposed simpler option above sounds OK. If anyone knows whether the same constraints apply to differential mode on all STM32 series (that support differential mode) without me having to check all of the reference manuals, that would also be much appreciated!

Describe alternatives you've considered

see above

[gautierg-st]

Hello,

Thank you for considering adding this feature.

1. Differential/single ended mode selection can only be performed while the ADC is disabled

This isn't a problem to disable the ADC. Many other configurations require it already, and in a typical application, single input and differential channels won't change types.

2. Calibration is performed separately for differential/single ended modes, so the driver needs to know during initialisation if any channels are using differential mode so that it can perform the correct calibration routine(s).

I'd say we keep the single ended calibration always on, and if it's possible to detect that there are at least one differential input to perform the differential calibration then that's perfect. Otherwise we can consider having a Kconfig for that.

I think the simplest way to implement this would be to set the channel mode to differential during initialisation for any channels which have the zephyr,differential property in the devicetree. Then the driver knows during initialisation whether it should perform a differential/single ended calibration (or both). If the user calls channel_setup with a different single-ended/differential setting to the one configured during initialisation, it could just return an error like it does for differential mode currently.

Even better would be if it's possible to dynamically change between differential and single ended mode for a channel, but I don't really have a need for this and I'm not sure whether the hardware would support this well (ie. is it possible to disable the ADC, change the channel config and re-enable it without performing a full re-initialisation?)

I agree that there are minimal needs to change it dynamically, but I still think it's simpler to configure the DIFSEL register in the channel_setup function. All we need to do is check whether the channel is already configured or not. If it is, no need to disable the ADC and waste time to reenable it, if not, then we disable the ADC and configure it. I don't see any reason why switching between the two modes wouldn't work (at a software level at least). This mechanism is already used for resolution or oversampling.

I'm happy to give this a try and open a PR, if the proposed simpler option above sounds OK. If anyone knows whether the same constraints apply to differential mode on all STM32 series (that support differential mode) without me having to check all of the reference manuals, that would also be much appreciated!

I've had a quick look at some RefMan and configuring the differential channels seem relatively easy, just a call to LL_ADC_SetChannelSingleDiff (series with this function are F3, G4, H5, H7, H7RS, L4, L5, N6, U5, WB).

The calibration may be the most tricky part (I'm thinking especially of STM32N6).

If you feel that you can pull this off, go for it. And if you have questions regarding the ADC driver, don't hesitate to ask me.