STM32H745 Can't get temperature

[jgauthier]

I've made several attempts to get the temperature from the ADC and all end up resulting in negative numbers.

I'm using an ADC reference IC that provides a 2.048 reference.

I've tried a few different approaches.

The simplest approach I've done is trying to use the built in macros:

__HAL_ADC_CALC_TEMPERATURE, and __LL_ADC_CALC_TEMPERATURE.

  serprintf("%i\n", __HAL_ADC_CALC_TEMPERATURE(2048, aADC3ConvertedData[0], ADC_RESOLUTION_12B));
  serprintf("%i\n", __LL_ADC_CALC_TEMPERATURE(2048, aADC3ConvertedData[0], LL_ADC_RESOLUTION_12B));

Where aADC3ConvertedData is an array that consists of temperature, vref, and vbat.

These macros give me ~-135C. (one calls the other anyway, I think)

I've attempted to get the vref using this macro:

vref =  __LL_ADC_CALC_VREFANALOG_VOLTAGE(aADC3ConvertedData[1], ADC_RESOLUTION_12B);

That gave me the same result.

Lastly, I've used the datasheet function:

There is at least one caveat with this function

1) TS_CAL1 and TS_CAL2 were measured with a vref of 3300mV

(They are stored as 12251, and 16254 respectively)

I've converted the values to my vref of 2048mV using the following process:

3300/65535
Out[128]: 0.05035477225909819
 
2048/65535
Out[129]: 0.03125047684443427
 
12251*0.05035477225909819
Out[132]: 616.896314946212
 
 
616/0.03125047684443427
Out[135]: 19711.69921875
 
16254*0.05035477225909819
Out[136]: 818.4664682993821
 
818.4664682993821/0.03125047684443427
Out[141]: 26190.527343750004
 
TS_CAL1 = 19740.380859375
 
TS_CAL2 = 26190.527343750004

Raw ADC value for temperature (TS_DATA) is 6050.

Plugging into the above equation gives me -139C.

I've left everything at 16b since this is what is recommended in stm32h7_hal_adc.h

Is there something else I have not considered?

[S.Ma]

Check and adapt a Nucleo ADC example which is closet to your chip, which is supposed to measure temperature.

[Piranha]

Not the main issues, but still...

• You have to divide by 65536.
• The rounding of 616.8... is incorrect.
• The TS_CAL1 result in lines 12 and 20 doesn't match.

Check my post in this topic:

https://community.st.com/s/question/0D53W00002167nhSAA/wrong-voltage-reading

[jgauthier]

I see I made a few minor issues, but still something isn't right.

I took a slightly different approach to scale the TS_CAL1 and TS_CAL2 values from 3300mV to 2000mV:

TS_CAL1 = 12251

(12251/65536)*3300
Out[28]: 616.8869018554688

So, this is a "voltage" representation based on this discussion.

I should be able to convert that voltage reading into a scaled version of the ADC signal at my VREF:

616.8869018554688*65536/2000
Out[29]: 20214.15

Validating that:

(20214.15/65536)*2000
Out[31]: 616.8869018554688

I do the same with TS_CAL2 and I get the results of

26819.1

My reading, which is around ~6500 is already scaled to 2000mV. Gives me ~-135C.

However, upon reflection why is my ADC giving me 6050 for temperature?

If TS_CAL1 is taken at 30C, and it's scaled value is 20214.15 then this should be significantly higher since I'm at room temperature. I understand this is not a linear value, but it should be much higher.

[jgauthier]

Additionally, I also have an STM32G491. However, it's using a 3.3V ref. I noticed TS_CAL1, and TS_CAL2 are 1034 and 1384 respectively.

The raw ADC temperature reading is 245.

I applied a heat gun to both MCPUs it increase the temperature and neither one of them changed significantly at all.

My Cube settings for G491

and for H745

[waclawek.jan]

1.5/2.5 cycles sampling time is certainly incorrect. Read datasheet.

JW

[Pavel A.]

> I applied a heat gun to both MCPUs it increase the temperature and neither one of them changed significantly at all.

Then you have a problem with the A/D conversion?

[jgauthier]

This definitely seems to be the cause. Definite disconnect on my end about how it works. Still something isn't clear though. The datasheet excerpt is this:

 

My ADC is clocked at 120Mhz, however I am sing a prescaler of 2, so it is effectively 60Mhz.

This seems to imply I can select 12.5 cycles?

I noticed with 12.5 cycles this is still not accurate. But as I increase cycles the accuracy increases. For instance, at 12.5 cycles my ADC reading is still ~500 which ends up being -110C.

If I increase the cycles to 92.5 then my temperature is 44C. Closer, and possible.

If I increase the cycles to 247.5 then my temperature calculates to 29C. This is probably correct.

So, I guess I am trying to get a little further understanding of this.

Based on what I've read, Tconv would be (247.5+12.5)/60M

About 4us. So, with this in mind, I am converting 230769 times per second.

Is that the right thinking? The sampling is adequate.

Thanks for pointing me in the right direction.

[Chris1]

Refer to Temperature sensor characteristics in datasheet, should have something like:

  ADC sampling time when reading the temperature: 9 μs