Hello forum, I'm (still) working on a data acquisition system based on STM32H743XIH6 MCU, I'm pushing the 16bit ADC to the limits and I'm stuck on a corner.
Let me summarize first the actual code: 1) TIM1 trigger every 1us ADC1 acquisition 2) ADC1 acquire 4 ch and move data via DMA to a ADCbuf[4] (4ch scan @1MSa/s) 3) HAL_ADC_ConvCpltCallback is called (every 1us) when 2) is complete and does some calculations on ADCbuf[4]. (After 100k executions=4x100k samples a flag is risen triggering data output on the main loop).
Long story short, It works flawlessly up to 960kHz, at 1MHz randomly the callback is called again before the previous call get completely served. As far I have understood the problem is that 3) is called with a 1MHz occurrance and, context saving\restore, demand too much time to get the system online without losing any samples.
The solution I'm trying to follow is to relax this 1us callback timing this way: 1) TIM1 trigger every 1us ADC1 acquisition 2) ADC1 acquire 4 ch and append data via DMA to a ADCbuf[40] (ADC still acquire 4ch @ 1MSa/s) 3) When 10 acquisition are completed (every 10us), something (a callback?) is called and does some calculations on a (4x)10 samples batch. (After 10k executions, 4x100k samples, a flag is risen triggering data output on the main loop).
Below follow some code, how do i reconfigure ADC\DMA in order to get this acquisition scheme? Thanks in advance for all the precious infos, Luca
Don't use the ADC interrupt/callback, but the DMA. Config circular DMA which is "endless". Config Half transfer and transfer complete interrupts. In each DMA interrupt callback, process the half of the buffer that is finished while the ADC keeps filling the other half.
Don't use the ADC interrupt/callback, but the DMA. Config circular DMA which is "endless". Config Half transfer and transfer complete interrupts. In each DMA interrupt callback, process the half of the buffer that is finished while the ADC keeps filling the other half.
Hi @KnarfB , do you mean using HAL_ADC_ConvHalfCpltCallback() beside the HAL_ADC_ConvCpltCallback() (which I'm already using)? That's interesting since I didn't realized before that I could lose samples if the ADC overwrite the very first data if I'm not serving interrupt quickly (before the new sampling group 1-10).
DMA is already circular, therefore should I just do the following 4->40 and add the Half callback?
Just a quick follow-up on the design above, implementing the two-step (4x)10 sample acquisition scheme the code is running smoothly without any overruns. Thanks for the hint.
while (1){ //todo: Main Loop
if(ADCcmplt){ // Print data on Buffer full (was: TIM14 interrupt (10Hz) )
sprintf(text,"A:%.6f,B:%.6f,C:%.6f,D:%.6f\r", //SumR and SumA are already computed, need just to sqrt(sum/N) and convert to volts
(q*sqrt (SumRO[0]/Xcnt))/k[j[0]], //CH1 RMS
(q*sqrt (SumRO[1]/Xcnt))/k[j[1]], //CH2 RMS
(q*sqrt (SumRO[2]/Xcnt))/k[j[2]], //CH3 RMS
(q*sqrt (SumRO[3]/Xcnt))/k[j[3]] //CH4 RMS
);
usb_print(text);
HAL_GPIO_TogglePin(LED_GPIO_Port, LED_Pin);
ADCcmplt=0; //Retrigger system output
}
HAL_Delay(1);
/* USER CODE END WHILE */
/* USER CODE BEGIN 3 */
}
Even if code is ok, and data has been validated with bench tests comparing data with lab instrumentation I'm still confused by the data read during debugging in STM32CubeIDE. Let me expand: 1- I've ADCdata[40] buffer initialized to {0}, 2- Breaking code execution on the very first execution of HAL_ADC_ConvHalfCpltCallback(), I get all the 40 samples, not only the first 20 ones!
3. Breaking code execution on the very first execution of HAL_ADC_ConvHalfCpltCallback(), I get all the 40 samples overwritten, not only the latter 20s!
And so on...
I was expecting only a partial update (Half: 0-19 and Cmplt 20-39) but this is not happening. Am I misinterpreting how the callbacks are called?
