Calibrating STM32 ADC (VREFINT)

5
votes

I'm trying to read VDDA on an STM32F042 microcontroller. I'm getting unexpected results with VDD at 3.29V. I must be missing something fundamental.

output:

VREFINT=1917; VREFINT_CAL=1524; VDDA=2623 mV
VREFINT=1885; VREFINT_CAL=1524; VDDA=2668 mV
VREFINT=1913; VREFINT_CAL=1524; VDDA=2628 mV
VREFINT=1917; VREFINT_CAL=1524; VDDA=2623 mV
VREFINT=1917; VREFINT_CAL=1524; VDDA=2623 mV

adc_test.c:

#include <stdio.h>
#include "stm32f0xx.h"

#define VREFINT_CAL_ADDR                0x1FFFF7BA  /* datasheet p. 19 */
#define VREFINT_CAL ((uint16_t*) VREFINT_CAL_ADDR)

extern void initialise_monitor_handles(void);

int main(void)
{
    RCC->APB2ENR |= RCC_APB2ENR_ADC1EN;     /* enable ADC peripheral clock */
    RCC->CR2 |= RCC_CR2_HSI14ON;            /* start ADC HSI */
    while (!(RCC->CR2 & RCC_CR2_HSI14RDY)); /* wait for completion */
    /* calibration */
    ADC1->CR |= ADC_CR_ADCAL;               /* start ADc CALibration */
    while (ADC1->CR & ADC_CR_ADCAL);        /* wait for completion */
    ADC1->CR |= ADC_CR_ADEN;                /* ADc ENable */
    while (!(ADC1->ISR & ADC_ISR_ADRDY));   /* wait for completion */
    ADC1->SMPR |= ADC_SMPR1_SMPR_0 |        /* sampling mode: longest */
      ADC_SMPR1_SMPR_1 |
      ADC_SMPR1_SMPR_2;
    /* VDD reference */
    ADC->CCR |= ADC_CCR_VREFEN;             /* VREF Enable */
    ADC1->CHSELR = ADC_CHSELR_CHSEL17;      /* CH17 = VREFINT */

    initialise_monitor_handles();           /* enable semihosting */

    while (1) {
        ADC1->CR |= ADC_CR_ADSTART;             /* start ADC conversion */
        while (!(ADC1->ISR & ADC_ISR_EOC));     /* wait for completion */
        uint32_t vdda = 3300UL * *VREFINT_CAL / ADC1->DR; /* ref. manual p. 252; constant and result in millivolts */
        printf("VREFINT=%lu; VREFINT_CAL=%lu; VDDA=%lu mV\n",
                (unsigned long)ADC1->DR,
                (unsigned long)*VREFINT_CAL,
                (unsigned long)vdda);
    }
}

Screenshot from Datasheet:

enter image description here

Screenshot from Reference Manual

note this refers to .3V, but I believe this to be a typo, as the datasheet above and the longer formula below refer to 3.3V, and .3V is below minimum operating voltage for this part

enter image description here

3
stm32adcstm32f0
I don't see anything obviously wrong with your code, and I can confirm that .3 instead of 3.3 in the reference manual is indeed a typo (the copy I found online didn't have that error). One wild guess as to the problem - have you perhaps left the Vssa pin floating, instead of connected to ground? (Assuming that you are using a STM32F042 variant that actually has a separate Vssa pin.) Your error in calculating Vdda is suspiciously close to one diode drop, which seems like a plausible outcome if the negative reference voltage is floating. – jasonharper
That's an interesting idea, but the pin (pin 32) is connected to GND: imgur.com/gMo2GsH Interestingly, the thermal pad isn't connected to anything. – iter
That schematic is VERY wrong - it shows the part number for the UFQFNP32 variant of the part, but has the pins labelled according to the LQFP32 variant (which doesn't even have a thermal pad). On the UFQFPN32, pins 16 & 32 are additional Port B I/O pins, and the thermal pad is your only ground connection, absolutely required for proper operation. Basically, your chip is only seeing ground via the ESD protection diodes on some I/O pins, and 2.62V is an accurate measurement of the power supply as received by the chip. – jasonharper
Owwww...... You are exactly right. A previous version of the schematic called for a LQFP32, then it got changed to UFQFNP32, and I guess the hardware guys didn't read the datasheet closely enough. This looks pretty bad... I'm surprised the chip is working at all, and pretty well at that (in the digital domain). Obviously have to fix this in the next spin of the board. As a stop-gap measure, would it help anything to set PB2 and PB8 as inputs instead of high Z, to connect more of the circuitry in the chip to ground, or are the ESD diodes the only path to ground? – iter
Don't think inputs vs. hi-Z would make any difference. Setting the pins to output LOW might actually get a better path to ground, although that's living dangerously - even a momentary output HIGH state is likely to fry something. If there's a spot under the chip with no traces, you might be able to mill through from the back side and actually connect to the thermal pad. (Hey, it could be worse - I've seen a similar "wrong footprint" board layout, where the two footprints were basically rotated by 90° - absolutely nothing was connected to a usable pin.) – jasonharper

3 Answers

1
votes

I'm currently developing an ADC driver for STM32L4. During implementation I encounter almost the same problem. In my opinion the first formula enter image description here

is not calculating the VDDA, but VREF+. It's the voltage against which the ADC is evaluating the ADC-IN channels. Further the VREFINT_DATA is not measured VREF+ voltage, but an internal reference voltage which is controller dependent. In my case it defined in controller datasheet: enter image description here

Here's a pic how I am using the posted formulas: enter image description here

Some comments: ln 102: calculating VREF+ not VDDA

ln 105-110: calculate all ranks/configured sequence

ln 108: calculate voltage measured by ADCpin_x

ln 109: multiply by gain to get real value

In my opinion by calculating the VREF+ for each conversion sequence, I'll get better results, because so some ripples on VREF+ are compensated.

1
votes

as @Artur said Vref + is not Vdda, but usually (that's how I have it in my hardware design) Vref + is connected to Vdda (with the corresponding filters according to the datasheet), so calculating Vdda is the same as calculating Vref +.

I will show you how to calculate vdda, as I have it based on the STM32L431.

. You must first configure the ADC to measure VREFINT:

void MX_ADC1_Init(void)
{

    /* USER CODE BEGIN ADC1_Init 0 */

    /* USER CODE END ADC1_Init 0 */

    ADC_ChannelConfTypeDef sConfig = {0};

    /* USER CODE BEGIN ADC1_Init 1 */

    /* USER CODE END ADC1_Init 1 */
    /** Common config
     */
    hadc1.Instance = ADC1;
    hadc1.Init.ClockPrescaler = ADC_CLOCK_SYNC_PCLK_DIV4;
    hadc1.Init.Resolution = ADC_RESOLUTION_12B;
    hadc1.Init.DataAlign = ADC_DATAALIGN_RIGHT;
    hadc1.Init.ScanConvMode = ADC_SCAN_DISABLE;
    hadc1.Init.EOCSelection = ADC_EOC_SINGLE_CONV;
    hadc1.Init.LowPowerAutoWait = DISABLE;
    hadc1.Init.ContinuousConvMode = DISABLE;
    hadc1.Init.NbrOfConversion = 1;
    hadc1.Init.DiscontinuousConvMode = DISABLE;
    hadc1.Init.ExternalTrigConv = ADC_SOFTWARE_START;
    hadc1.Init.ExternalTrigConvEdge = ADC_EXTERNALTRIGCONVEDGE_NONE;
    hadc1.Init.DMAContinuousRequests = DISABLE;
    hadc1.Init.Overrun = ADC_OVR_DATA_PRESERVED;
    hadc1.Init.OversamplingMode = DISABLE;
    if (HAL_ADC_Init(&hadc1) != HAL_OK)
    {
        Error_Handler();
    }
    /** Configure Regular Channel
     */
    sConfig.Channel = ADC_CHANNEL_VREFINT;
    sConfig.Rank = ADC_REGULAR_RANK_1;
    sConfig.SamplingTime = ADC_SAMPLETIME_247CYCLES_5;
    sConfig.SingleDiff = ADC_SINGLE_ENDED;
    sConfig.OffsetNumber = ADC_OFFSET_NONE;
    sConfig.Offset = 0;
    if (HAL_ADC_ConfigChannel(&hadc1, &sConfig) != HAL_OK)
    {
        Error_Handler();
    }
    /* USER CODE BEGIN ADC1_Init 2 */

    /* USER CODE END ADC1_Init 2 */
}

. Now I will show you the code where the equation is executed:

vdda = 3.0 * (VREFINT_CAL /average);
vch = VREF * (average / ADC_RESOLUTION);

log("vdd = %.5f - ", vdda);
log("vchn = %.5f", vch);

where:

#define ADC_RESOLUTION 4095.0           // adc resolution 12 bits
#define VREFINT_CAL 1655.00             // Raw data acquired at a temperature of 30 °C (± 5 °C), VDDA = VREF+ = 3.0 V (± 10 mV)
#define VREF 3.3                        // voltage reference 3.3V

note:

'average' is an average of 256 samples taken by the adc (it's just a simple filter).

'log' is a function created by me similar to printf for the uart.

'VREFINT_CAL' varies according to the model.

result:

vdd = 3.28035 - vchn = 1.21343

as we see VREFINT matches the datasheet (1.212V typ.):

VREFINT

0
votes

Actually it is calculating Vdda, since the Vref calculation is very simple, you have to read the corresponding channel of the ADC with a sample time longer than the one marked in the data sheet (usually 10 us). If Vdda is 2.0 V, a value of 4095 corresponds to 2.0 (or more) V absolute (related GND). In a linear way, the value of Vref will be much higher than if it is read with Vdda = 3.30 V. Therefore, the compensation of the values ​​read with 2.0 V is necessary to know the absolute values ​​of voltage that the ADC is measuring. If they are not compensated, they will be values ​​relative to the voltage level that Vdda has at that moment. In addition, the power supply value is achieved, which will be useful in order not to go beyond the specifications of the microcontroller.