ADC value 20% too low.

RE: ADC value 20% too low.

maarten v — Wed, 07 May 2025 12:51:17 GMT

The project is finished. It was decided to use an external ADC.

For another project using an nrf52840 we experience the same issue. It looks like the nrf9160 has the same ADC issues as the nrf52840.

Turned out that the ADC is just not that accurate, and:
- the ADC calibration is only valid for the exact same settings (gain, resolution), which is not documented.

- the gain, especially the high gain (6x) is not llnear in the high value of the range (also not documented).

- even with 12 bit resultution, the actual accuracy is only 8 bit (this is documented).

For the project with the nrf52840 we now do a 'calibration' by measuring the 'zero' value configuring the disconnect for the input and a pull-down in the resistor ladder for each required configuration.

We also avoid the high input values by configuring a lower gain than possible..

Now we get about 0.5% to 1% accuracy for our measurements, which is good enough, and use oversampling to remove noise.

Hope this helps. For me, the ticket can ble closed.

RE: ADC value 20% too low.

Edvin — Wed, 07 May 2025 07:31:43 GMT

Hello,

What behavior do you observe? Is it a constant offset, or is it proportional to the voltage you are measuring?

Best regards,

Edvin

RE: ADC value 20% too low.

BartT — Tue, 06 May 2025 14:01:57 GMT

Hi, Is there any update on the issue yet? We have a similar issue. ADC works fine on VDD_IO 3V, but not on VDD_IO 1,8V.

RE: ADC value 20% too low.

maarten v — Mon, 19 Dec 2022 08:12:08 GMT

Thanks, i am looking forward to know the actual cause of this.

RE: ADC value 20% too low.

Edvin — Fri, 16 Dec 2022 11:14:01 GMT

Hello,

Karl asked if I could take over this ticket, because he will be out of office for a while.

I have seen something like this before, but that DevZone ticket "died". I have reported this internally, and I will keep you posted when I receive a reply.

Best regards,

Edvin

RE: ADC value 20% too low.

maarten v — Tue, 06 Dec 2022 07:46:12 GMT

I finally got time to take a further look at this. It looks like the measement with the nrf9160dk is off, but only when the vdd_io is set to 1.8V.

I modified the test program to show this better:

#include <zephyr.h>
#include <sys/printk.h>
#include <drivers/gpio.h>
#include <drivers/adc.h>
#include <drivers/sensor.h>
#include <nrfx_saadc.h>
#include <string.h>

#include <logging/log.h>


LOG_MODULE_REGISTER(main,LOG_LEVEL_DBG);

#define MY_ADC DT_NODELABEL(adc)
#define ADC_PIN NRF_SAADC_INPUT_AIN3 // P0.16
#define ADC_RESOLUTION 14

const char* err_str(int err)
{
	switch (-err)
	{
		case 0: return "OK";
		case EINVAL: return "EINVAL";
		case ENOMEM: return "ENOMEM";
		case ENOTSUP: return "ENOTSUP";
		case EBUSY: return "EBUSY";
		default: return "?";
	}
}

int read_adc(const char *tag, 
		const struct device *device, 
		uint8_t input,
		enum adc_reference ref,
		bool calibrate )
{
	struct adc_channel_cfg cfg = 
	{
		.gain = ADC_GAIN_1_6, // enough to measure VDD referenced to VDD_1_4 at vdd_io set to 3V
		.reference = ref, 
		.acquisition_time = ADC_ACQ_TIME(ADC_ACQ_TIME_MICROSECONDS, 10),
		.input_positive = input,

		.input_negative = NRF_SAADC_INPUT_DISABLED,
		.differential = 0,
		.channel_id = 0, // just use the first logical channel
	};

	// choosing the correct resolution for the channel is important for accuracy!
	// but changing the resolution between readings can cause offset deviations!
	int32_t err = adc_channel_setup(device, &cfg);
	if( err != 0 )
	{
		LOG_ERR("ADC channel setup failed with %d: %s.\n", err, err_str(err));
		return -1;
	}

	int16_t sample_buffer[1] = {0};
	struct adc_sequence sequence = 
	{
		.options = NULL,
		.channels = 0b00000001, // only measuring cfg.channel_id = 0
		.buffer = sample_buffer,
		.buffer_size = sizeof(sample_buffer),
		.resolution = ADC_RESOLUTION,
		.oversampling = NRF_SAADC_OVERSAMPLE_256X,
		.calibrate = calibrate
	};

	if( calibrate )
	{
		LOG_DBG("calibrate ADC");
	}

	// Read sequence of channels (fails if not supported by MCU)
	err = adc_read(device, &sequence);
	if (err != 0) 
	{
		LOG_ERR("ADC reading failed with error %d: %s.", err, err_str(err));
		return -2; 
	}

	volatile float part_from_range = sample_buffer[0] / (float)(1<<ADC_RESOLUTION);
	#define VDD_IO_mV 3000
	volatile int32_t ref_mv = ref == ADC_REF_VDD_1_4 ? VDD_IO_mV/4 : adc_ref_internal(device);

	int v_meas_mv = sample_buffer[0];
	__unused int result = adc_raw_to_millivolts(ref_mv, cfg.gain, ADC_RESOLUTION, &v_meas_mv);
	__ASSERT( result == 0, "could not convert to millivolts" );

	if (tag)
	{
		LOG_INF("%s:{adc:%hd, part_range_%%%%:%d, ref_mv:%d, measured_mv:%d}", 
			tag, sample_buffer[0], 
			(int)(1000*part_from_range), ref_mv, v_meas_mv);
	}

	return sample_buffer[0];
}


void main(void)
{
	char hw_rev[8];
	memcpy( hw_rev, (char*)&NRF_FICR->INFO.VARIANT, 4);
	hw_rev[4] = 0;
	char hw_type[8];
	memcpy( hw_type, (char*)&NRF_FICR->INFO.DEVICETYPE, 4);
	hw_type[4] = 0;

	LOG_INF("nrf91 FICR HW revision: %s, type: %s", hw_rev, hw_type);

	// see: https://developer.nordicsemi.com/nRF_Connect_SDK/doc/2.0.0/zephyr/build/dts/howtos.html
	const struct device *dev_adc = DEVICE_DT_GET(MY_ADC);
	// or runtime
	//const struct device *dev_adc = device_get_binding("ADC_0");

	if (!device_is_ready(dev_adc)) 
	{
		LOG_ERR("ADC device not found\n");
		return;
	}

	// calibrate once:
	read_adc(NULL, dev_adc, NRF_SAADC_INPUT_VDD, ADC_REF_INTERNAL, true);

	while (1) 
	{
		LOG_INF("ADC_reading:");
		LOG_INF("{");
		LOG_INF("  full_range:%d,\n", 0x01<<ADC_RESOLUTION);
		
		volatile __unused int vdd_ref_vdd = read_adc("  vdd_ref_vdd", dev_adc, NRF_SAADC_INPUT_VDD, ADC_REF_VDD_1_4, false);
		volatile __unused int vdd_ref_int = read_adc(", vdd_ref_int", dev_adc, NRF_SAADC_INPUT_VDD, ADC_REF_INTERNAL, false);

		// measure 0.5 Vdd at pin p0.16 (using 2 same resistors in sweries between GND and VDD)
		volatile __unused int p016_ref_vdd = read_adc(", p0_16_ref_vdd", dev_adc, ADC_PIN, ADC_REF_VDD_1_4, false);
		volatile __unused int p016_ref_int = read_adc(", p0_16_ref_int", dev_adc, ADC_PIN, ADC_REF_INTERNAL, false);

		LOG_INF("},");

		k_sleep(K_MSEC(300));
	}
}

With the board set to Vdd_io=3V this logs:

*** Booting Zephyr OS build v3.1.99-ncs1  ***

[00:00:03.074,737] <inf> main: nrf91 FICR HW revision: 0BAA, type: 
[00:00:03.074,768] <dbg> main: read_adc: calibrate ADC
[00:00:03.095,123] <inf> main: ADC_reading:
[00:00:03.095,153] <inf> main: {
[00:00:03.095,184] <inf> main:   full_range:16384,

[00:00:03.098,144] <inf> main:   vdd_ref_vdd:{adc:10957, part_range_%%:668, ref_mv:750, measured_mv:3009}
[00:00:03.101,104] <inf> main: , vdd_ref_int:{adc:13626, part_range_%%:831, ref_mv:600, measured_mv:2993}
[00:00:03.104,095] <inf> main: , p0_16_ref_vdd:{adc:5451, part_range_%%:332, ref_mv:750, measured_mv:1497}
[00:00:03.107,055] <inf> main: , p0_16_ref_int:{adc:6782, part_range_%%:413, ref_mv:600, measured_mv:1490}
[00:00:03.107,086] <inf> main: },

With the board set to Vdd_io=1.8V this logs:

*** Booting Zephyr OS build v3.1.99-ncs1  ***

[00:00:00.959,930] <inf> main: nrf91 FICR HW revision: 0BAA, type: 
[00:00:00.959,930] <dbg> main: read_adc: calibrate ADC
[00:00:00.980,316] <inf> main: ADC_reading:
[00:00:00.980,346] <inf> main: {
[00:00:00.980,346] <inf> main:   full_range:16384,

[00:00:00.983,337] <inf> main:   vdd_ref_vdd:{adc:8885, part_range_%%:542, ref_mv:450, measured_mv:1464}
[00:00:00.986,297] <inf> main: , vdd_ref_int:{adc:7597, part_range_%%:463, ref_mv:600, measured_mv:1669}
[00:00:00.989,257] <inf> main: , p0_16_ref_vdd:{adc:4439, part_range_%%:270, ref_mv:450, measured_mv:731}
[00:00:00.992,218] <inf> main: , p0_16_ref_int:{adc:3799, part_range_%%:231, ref_mv:600, measured_mv:834}
[00:00:00.992,248] <inf> main: },

To me it looks like something is wrong with the nrf9160DK.

Our own PCBA works with 3 vdd_io (no defined input here) so executing the above gives:

*** Booting Zephyr OS build v3.1.99-ncs1  ***

[00:00:02.464,172] <inf> main: nrf91 FICR HW revision: 0BAA, type: 
[00:00:02.464,172] <dbg> main: read_adc: calibrate ADC
[00:00:02.484,008] <inf> main: ADC_reading:
[00:00:02.484,039] <inf> main: {
[00:00:02.484,069] <inf> main:   full_range:16384,

[00:00:02.486,968] <inf> main:   vdd_ref_vdd:{adc:10924, part_range_%%:666, ref_mv:750, measured_mv:3000}
[00:00:02.489,837] <inf> main: , vdd_ref_int:{adc:13865, part_range_%%:846, ref_mv:600, measured_mv:3046}

Conclusion: no problem with the code but with the nrf9160DK PCBA working on vdd_io=1.8V

RE: ADC value 20% too low.

Karl Ylvisaker — Fri, 14 Oct 2022 19:50:00 GMT

Hello again,

[quote user="maarten v"]

We have a simple work-around: just measure the vdd and don't expect it to be full-range. Then use that to calculate a correction for actual measured value.

I'll come back later on this topic when i have more time.

[/quote]

While this might be a pragmatic solution - and I completely understand the need to be pragmatic about things like this sometimes - I would want to understand the root cause of this, in order to guarantee that you wont run into any other unexpected issues in the future.

Please let me know if you get a chance to take a look at the questions in my previous comment, or if you have any updates on this issue.

Best regards,
Karl

RE: ADC value 20% too low.

maarten v — Wed, 28 Sep 2022 10:00:55 GMT

We have a simple work-around: just measure the vdd and don't expect it to be full-range. Then use that to calculate a correction for actual measured value.

I'll come back later on this topic when i have more time.

thanks

RE: ADC value 20% too low.

Karl Ylvisaker — Wed, 24 Aug 2022 06:20:20 GMT

Hello again,

Thank you for your extreme patience with this.

Thank you for confirming regarding the calibration.
This sounds very strange indeed - I too would expect your raw ADC value to come out at the top of the resolution range, in this case.

Have you had a chance to scope the VDD, using an oscilloscope?
Could you try to short the pin used for measuring to GND directly, and see what the sample reads then? For good measure, could you also check what the PSELP and CONFIG register of the SAADC is set to, right before the sampling is triggered?

Best regards,
Karl

RE: ADC value 20% too low.

maarten v — Mon, 01 Aug 2022 11:20:11 GMT

I am sorry to respond so slow: i am only working 2 days a week on this project.

[quote userid="87869" url="~/f/nordic-q-a/90251/adc-value-20-too-low/379005"]

maarten v said:
I am not sure what you mean: before ADC (offset) calibration the offset error is different but the gain error is not.

What do you mean by this, could you provide some examples of what you are seeing with and without the offset calibration at the beginning?

[/quote]

I don't a an exact output right now. The offset calibration is of no consequence for this mentioned problem: it only differs a few millivolts before and after calibration.

The gain error i mention was that i assumed that the full ADC range would be from gnd to vdd_io (1.8V) with the chosen reference, but i observe something completely different (as mentioned earlier).

So when measuring 0.5Vdd_io with 12 bit resolution, i would expect an ADC value of about 2048 but i get something like 1415 (as mentioned in the first post). ADC calibration does not change that significantly.

RE: ADC value 20% too low.

Karl Ylvisaker — Thu, 28 Jul 2022 02:43:43 GMT

[quote user="maarten v"]nrfcnonnect SDK version 2.0.0[/quote]

Thank you for confirming this.

[quote user="maarten v"]I am not sure what you mean: before ADC (offset) calibration the offset error is different but the gain error is not.[/quote]

What do you mean by this, could you provide some examples of what you are seeing with and without the offset calibration at the beginning?

Best regards,
Karl

RE: ADC value 20% too low.

maarten v — Wed, 27 Jul 2022 07:47:28 GMT

> Could you confirm that you saw the same results with the added offset calibration at the beginning?

I am not sure what you mean: before ADC (offset) calibration the offset error is different but the gain error is not.

> Could you also let me know which nRF Connect SDK version that this code is made for?

nrfcnonnect SDK version 2.0.0

RE: ADC value 20% too low.

Karl Ylvisaker — Wed, 27 Jul 2022 01:34:02 GMT

Thank you for providing the code.

Could you confirm that you saw the same results with the added offset calibration at the beginning?
Could you also let me know which nRF Connect SDK version that this code is made for?

Best regards,
Karl

RE: ADC value 20% too low.

maarten v — Tue, 26 Jul 2022 06:54:24 GMT

Attached is the code.

I experimented a bit more the the ADC so the code differs a bit from the code above, but the result is the same: consistent incorrect measurements.devzone.nordicsemi.com/.../adc_5F00_tryout.zip

RE: ADC value 20% too low.

Karl Ylvisaker — Tue, 26 Jul 2022 00:30:50 GMT

Hello again,

Thank you for your patience with this. The summer holiday season has begun here in Norway, and so the DevZone forum is operating at reduced capacity for the next couple of weeks.

Thank you for providing the additional information as well. Did you try to explicitly trigger an offset calibration before you begin the sampling, to see if this had any effect?
Do you have this as a minimal project you could send me, so that I could reproduce this on my end and take a closer look?

Best regards,
Karl

RE: ADC value 20% too low.

maarten v — Fri, 22 Jul 2022 07:59:19 GMT

The board power is just via the usb debug connection J4.

Measuring the vdd gives 1.8V and measuring then 5V gives indeed 5V.

The board version: PCA10090 1.1.0.

I tried this on two nrf9160dk boards with the same result.

Edit: i use a resistor ladder of 2x30k to divide the vdd by 2 to measure the voltage with the board analog input..

RE: ADC value 20% too low.

Karl Ylvisaker — Fri, 22 Jul 2022 06:41:38 GMT

Hello,

That looks strange indeed - would you mind detailing how you are powering the nRF9160 DK?
Could you also explicitly perform an offset calibration, to see if this has any influence on the measurements you're seeing?

Best regards,
Karl