Hi, I have a problem with audio acquisition using PDM microphones. After using PDM_Filter, single peaks of high amplitude appear in the data in addition to the correct readings.
I am using 4 microphones MP34DT01-M, STM32H743 and SAI interface.
I take 4096 samples from each microphone, convert PDM to PCM and save to SD card using FATFS.
My code (excerpt) is very simple:
void HAL_SAI_RxCpltCallback (SAI_HandleTypeDef *hsai) //PDM buffer full
{
rxstate = 2; //buffer full flag
real_round++; //next real round of acquisition
}
int main(void)
{
HAL_SAI_Receive_DMA(&hsai_BlockA1, (uint8_t*)PDM_buffer, PDM_buffer_size); //start acquisition
while (1)
{
if (rxstate == 2)//if PDM buffer is full
{
rxstate = 0; //clear the flag
current_round++; //next processed round of acquisition
if (current_round <= last_round) //processes and saves a predetermined number of cycles (rounds)
{
PDM_Filter(&(((uint8_t*)PDM_buffer)[0]), &(((uint16_t*)PCM_buffer)[0]), &PDM1_filter_handler); //MIC_A
PDM_Filter(&(((uint8_t*)PDM_buffer)[1]), &(((uint16_t*)PCM_buffer)[1]), &PDM2_filter_handler); //MIC_B
PDM_Filter(&(((uint8_t*)PDM_buffer)[2]), &(((uint16_t*)PCM_buffer)[2]), &PDM3_filter_handler); //MIC_C
PDM_Filter(&(((uint8_t*)PDM_buffer)[3]), &(((uint16_t*)PCM_buffer)[3]), &PDM4_filter_handler); //MIC_D
if (current_round >= first_round)
{
if(f_write(&fil, PCM_buffer, sizeof(PCM_buffer), &numread) != HAL_OK) //write data to SD card
{
printf("f_write ERROR\n");
}
}
}else //if (current_round > last_round)
{
HAL_SAI_DMAStop(&hsai_BlockA1); //stop acquisition
close_file();
unmount_sd();
break; //exit while loop, end of program
}
}
}
}
In each round (cycle) of 4096 measurements, there are single abnormal values with an amplitude larger than the rest of the signal. In addition, the first cycle has invalid values, but this is OK for me. In contrast, the rest of the data should be correct.
I know that the first 2 TDMA frames are invalid, but there are more such invalid values. Do you have any idea where they can come from?
I attach 2 files with fragments of graphs. On the first one, I consider the values with indexes 4097-4103 (according to RM0433) as normal, but I have no idea where the values with indexes 4567-4576 come from. The situation repeats in every cycle. As an example, I attach the appearance of the beginning of the third cycle.
Maybe this is an important piece of information - it takes longer to write to the SD card than it takes to fill the PDM buffer, so there is no continuity of the written signal, but this is OK for me.
