Research and implementation of oversampling technology based on stm32F1

1. Problem statement: 
In the 2016 TI Cup Electronic Design Competition, Question G, a simple electronic scale, one of the requirements is as follows: 

It can be calculated that the accuracy required by this question is 
(500-5)/0.01=49500

However, the built-in ADC of stm32F1 has only 12-bit accuracy, which means that it can collect a maximum of 2 to the 12th power of data (2^12=4096), which obviously does not meet the requirements of the question. Through oversampling technology, the required accuracy can be achieved.

2 Extraction of oversampled data

 
Note: The p here is the additional p bits of precision required

Assuming p=1, 4^1=4 data need to be collected. Assuming that the 4 data are 
1101 
, 1100 
, 1011, 
and 1010,  respectively
, R represents the number of sampling bits. Here, R=4. 
The sum of the four data is 101110 (R+2p=6, so a 6-bit data is obtained after accumulation). 
Then shift the 6-bit data right by 1 bit (p=1), and you can get 10111 (R+p=5). This value is the sampling result with 1-bit accuracy improved.

3 Prerequisites for oversampling 
(1) There must be some noise in the input signal. The noise must be white noise with power evenly distributed throughout the useful frequency band. 
(2) The amplitude of the noise must be large enough to affect the input signal so that the ADC conversion result can randomly flip at least 1 bit. Otherwise, all input signals will convert to the same value, and the decimation operation on these values ​​will not improve the accuracy.

Note: Here you can refer to the link for some articles "Research and Implementation of Oversampling Technology Based on STM32"

4. Program Implementation 

In my own program, I did not implement it completely according to the reference material. My program:

    while (1)
    {

        ADC_ConvertedValueLocal =(float) ADC_ConvertedValue/4096*3.3; //Read the AD conversion value, the variable ADC_ConvertedValue here is the AD conversion value
        printf("\r\n The current AD value = %d \r\n", ADC_ConvertedValue); 
        printf("\r\n The current AD value = %f V \r\n",ADC_ConvertedValueLocal); 

    for(h=0;h<10;h++)
        { //The outermost loop here is to print the converted value 10 times in a row for easy observation
            for(j=0;j<500;j++)
            { for(i=0;i>8; // Shift right by 8 bits to get 12 bits of precision
                        Adctemp4=Adctemp4>>4; // right shift 4 bits to get 16 bits of precision
                    //printf("16λ¾«¶ÎΪ£º%d\n",Adctemp4);
                    //printf("12λ¾«¶ÈΪ£º%d\n",Adctemp8);
                    Adctemp=0; sum=sum+Adctemp4;
            }
            adcaverage=sum/500; //Take the 16-bit precision value 500 times, that is, take the average value of multiple samples
            sum=0; printf("adcaverage=%d\n",adcaverage); //Print out the average value of 500 consecutive values ​​with 16-bit precision
    } 
    Delay(0xffffee);  
    }123456789101112131415161718192021222324252627282930313233

5 Notes 
(1) In the program, the sampling frequency of ADC is 9MHz. Obviously, the sampling frequency is too high in the sampling of electronic scales. You can choose to reduce the ADC sampling frequency to further improve the accuracy. 
(2) For more details, please refer to an article in Baidu Library, "Research and Implementation of Oversampling Technology Based on STM32". This article has a good analysis. I also basically wrote the program according to the ideas of this article. This blog is just my own experience. Please take care of the poor writing.

The website of "Research and Implementation of Oversampling Technology Based on STM32" 
is: 
http://wenku.baidu.com/link?url=5r_i7r38zSqDF0-cQlWqSBp64UHeXmPKAMGgnKk4u-FYkYdzgltG5btIFz6x0Wyi_8Q8MzoRdo4AHVhGs9MN4fxigSqJUC3O1XPjzefCdeC

• top

• 0

• Step on