How should the FMAC of STM32G474 be configured to obtain correct calculation data?

Question

Hi,I encountered problems during the development process of using FMAC. To illustrate the problem simply, I used FMAC to implement an RC filter, by which I mean to illustrate the calculation problems of FMAC.RC filter:10K+10nFIts discrete z-domain expression：（Ts=1e-6,Tustin）so I get fixed-point parameters :I referred to the example program of AN5305 and used a 1-us timer to trigger ADC sampling. After the ADC sampling is completed, DMA writes to the WriteData of FMAC, and then the ReadData is processed in the interrupt as follows:As shown above, my current ADC sample is 0, and my Vref is set to 800. I think the output of FMAC as an RC filter should be around 800 in the DC case, but actually it is not. It is a relatively small value. Why is that?May I ask if there is a problem with my configuration of FMAC?

Saket_Om · Accepted Answer

Hello ​@TakingLeePlease found some issues or incoherence in your code that could be the cause of your dysfonctionnement: 1- HAL_FMAC_FilterPreload is called with the wrong parameters.  aInputValues is sized with ARRAY_SIZE (1) but INPUT_BUFFER_SIZE is 3.  This may cause the function to read two values beyond the array. 2- HAL_FMAC_FilterStart is called with the wrong parameters  With your FMAC configuration, the expected output size should configure the number of data for DMA transfer.  If you want a one-shot operation, use 1U. 3- error in CNTRL_1p1zFloat:  The local variable "acc" is not initialized and you did a first "+=" operation. The result could be unpredictable.  I recommend making an initial assignment instead of a compoud assignment. 4- FMAC works with Q1.15 format data, while your "CNTRL_1p1zFloat" function works in float data.  This may naturally create small differences in results. 5- The A1 coefficient is inconsistent:    - For FMAC, the coefficient A1 is set to 0x8147 (Q1.15 format) which is the approxmatively the float value -0.989.   - For CNTRL_1p1zFloat, the coefficient A1 is set to the float value +0.99 which is the 0x7EB8 in Q1.15 format.    The FMAC implements the IIR formula: IIR filter Y = B*X + A*Y’. By using a negative A coefficient the formula is resume to Y = B*X - A*Y’  Then if this equation is fine for you, then CNTRL_1p1zFloat should used the same negative value.  Otherwise, if you want the initial IIR filter formula Y = B*X + A*Y’, then you must use a positive coefficient    Which coefficicent do you want to keep?Best regardsOmar

Saket_Om · Answer

Hello ​@TakingLeeAbout your question "May I ask if there is a problem with my configuration of FMAC?" and based on the information provided, the FMAC configuration seems to be correct.A quick reminder regarding the use of variables in callbacks: they must be declared as "volatile".Then we invite you to verify that the "Fmac_output" array is correctly declared as volatile in your application.Otherwise, about your problem, we have no enough data to make a correct analysis.Then could you give us more details about your ADC configuration ?What sampling time is it using ?Is the raw ADC measurement correct relative to your input signal ?

Sign up

Login with SSO

Login to the community

Login with SSO