Analyze FOC timing of AMCBLDC and AMC2C

[Nicogene]

          Let me try to recap what I got from a F2F alignment with @sgiraz.

1. There's pretty much NO difference between using STD or CMSIS ops on AMCBLDC and AMC2C.
2. On the AMCBLDC, the FOC is taking up $8$ $\mu s$ and NOT $17$ $\mu s$. The difference is represented by the ADC operations.
3. FOC takes up $14$ $\mu s$ on AMC2C, longer than on AMCBLDC.

• Point 1 completes this issue ✅ We go with STD. See FOC: use STD instead of CMSIS #75.
• Follow-up question on Point 2: "Why does ADC last that long compared to FOC? Is there any margin for improvement?"
• Follow-up question on Point 3: "Why is AMC2C slower than AMCBLDC?" We know this is an ongoing study and we're probably at the right point to contact ST.

cc @Nicogene @sgiraz @marcoaccame @maggia80 @mfussi66

Originally posted by @pattacini in #67 (comment)

[sgiraz]

➡️ On AMCBLDC

In point 2 I want to remark that the $8.3$ $\mu s$ are taken to execute the AMC_BLDC_step_FOC() function when the control mode is IDLE.
When the control mode is set to CURRENT instead, it takes $10.7$ $\mu s$, see the scope plot below:

I Apologize for the mistake/confusion!

By the way, the EventViewer shows that the whole DMA interrupt takes about $15$ $\mu s$ :

Measuring with the scope we got $13.1$ $\mu s$ measuring the whole onCurrents_FOC_innerloop function:

...And $14.34$ $\mu s$ considering the whole DMA:

static void adc2_HalfTransferComplete_cb(ADC_HandleTypeDef *hadc)
{
    // TODO: start measuring from here
    
    GPIOB->BSRR = (uint32_t)0x0020; // 0x0004

    
#if defined(MOTORHAL_changes) 
    rawCph1 = adc2_Buffer[0].cph1;
    rawCph2 = adc2_Buffer[0].cph2;
    rawCph3 = adc2_Buffer[0].cph3;
    pwmSetCurrents_cb(rawCph1, rawCph2, rawCph3);
#if !defined(MOTORHALCONFIG_DONTUSE_RUNTIMECURR_FILTERING)
    analogMovingAverageFromISR(&cph1Filter, rawCph1); 
    analogMovingAverageFromISR(&cph2Filter, rawCph2);
    analogMovingAverageFromISR(&cph3Filter, rawCph3);   
    

#endif    
#else
    rawCph1 = adc2_Buffer[0].cph1;
    rawCph2 = adc2_Buffer[0].cph2;
    rawCph3 = adc2_Buffer[0].cph3;
    pwmSetCurrents_cb(rawCph1, rawCph2, rawCph3);
    analogMovingAverageFromISR(&cph1Filter, rawCph1);
    analogMovingAverageFromISR(&cph2Filter, rawCph2);
    analogMovingAverageFromISR(&cph3Filter, rawCph3);
#endif

    GPIOB->BRR = (uint32_t)0x0020; // 0x0004
}

In summary (when in CURRENT control mode):

AMC_BLDC_step_FOC() function	$10.7$ $\mu s$
onCurrents_FOC_innerloop function	$13.1$ $\mu s$
the whole DMA	$14.34$ $\mu s$

cc @Nicogene @pattacini @marcoaccame @mfussi66

[sgiraz]

➡️ AMC2C (66.6 KHz)

When the control mode is set to CURRENT instead, the AMC_BLDC_step_FOC() function takes $20$ $\mu s$, see the scope plot below:

The the whole onCurrents_FOC_innerloop function takes $25.45$ $\mu s$:

The whole DMA takes $26.86$ $\mu s$:

With the EventViewer we got about $29$ $\mu s$ :

In summary (when in CURRENT control mode):

AMC_BLDC_step_FOC() function	$20$ $\mu s$
onCurrents_FOC_innerloop function	$25.45$ $\mu s$
the whole DMA	$26.86$ $\mu s$

cc @pattacini @marcoaccame @maggia80 @Nicogene

[Nicogene]

Great analysis @sgiraz !

To summarize we have

	AMC2C (66.6 KHz)	AMCBLDC
AMC_BLDC_step_FOC() function	$20$ $\mu s$	$10.7$ $\mu s$
onCurrents_FOC_innerloop function	$25.45$ $\mu s$	$13.1$ $\mu s$
the whole DMA	$26.86$ $\mu s$	$14.34$ $\mu s$

"Why is AMC2C slower than AMCBLDC?" We know this is an ongoing study and we're probably at the right point to contact ST.

A ~ 2x factor seems a lot, maybe it is the case of contacting ST

[Nicogene]

The analysis of @sgiraz pointed out an important timing issue on amc2c, and this will be further investigate contacting ST, after that we hopefully know how to align the performance of the two boards.

Closing.

[pattacini]

Interesting timing for a comparison:

Source: https://www.mathworks.com/campaigns/offers/next/field-oriented-control-techniques-white-paper.html

Note

We're not that bad, after all!

cc @marcoaccame @sgiraz @mfussi66 @Nicogene @maggia80