Concerns and suggestions regarding current sensing & control (especially in B-G431B-ESC1)

[Antun_Skuric]

Hey Guys,

I am happy to hear some proper control theory questions and remarques, I think that we dont have them enough.

@Larock1234

• A lot of people seem to struggle with current control because of various reasons (see below). This is fine, current control is hard. However, I often read in this forum to use a low-pass filter if the current readings are not good and to adjust it as slow as tau= 0.01s. I don’t think this is good practice, because with this, the current controller will never be able to regulate the current in a satisfactory way as typical motor time constants are around tau_motor = 0.0002, …, 0.001. By using those very slow filters, we create slower systems that appear to be better controllable, but in reality we just hide the fast motor winding dynamics.

This is true. The low pass filter is really not intended for smoothing the torque control dynamics (or making it more stable) but to filter the current values. And it is absolutely true that having the time constant around 0.1 is far to big and even strange to be working at all. (BTW the same is true for the velocity Low pass filter).

However, the low pass filters in the simplefoc are placed on the D and Q current not on phase currents so even though they are very crucial for the good dynamics they will not ruin the vector alignment. That is probably why such values even work.

In my opinion, whenever such solutions are necessary it means either one of the two things (which you also very nicely pointed out):

• PI gains are off and they need to be re-tuned ( lowered in most cases )
• Current sensing is very noisy

@Larock1234

• At the same time, the gains for the current controller seem to be recommended to be chosen rather arbitrarily and in general very low. Instead, we should start to rely on control theory. For example, in case no current low-pass filter is used, there is a very elegant way of cancelling a pole in the open-loop system, by setting K_i = K_p*R/L (because SimpleFOC uses a parallel PI) and choosing K_p with K_p = L *omega, where omega is the desired bandwidth. See for example here: Teaching Your PI Controller to Behave (Part III) - Industrial - Technical articles - TI E2E support forums. A typical value for omega is around 1500 rad/s, which is in the same range as the intrinsic motor winding dynamics and which should usually be feasible with a current control sampling frequency of ca. 20kHz and reasonable noise. Of course, a similar procedure could be done in discrete time or with an additional filter.

Yeah recommending the gains is very tricky and as many of our users are using gimbal motors with higher phase resistance and lower currents many of our examples are using relatively high current control gains. And you are definitely right that we need a better way for tuning the current loops. For now we support the SimpleFOCStudio which can be used to find good parameters but its still hand tuning and requires at least a good idea of initial values.

I would really really like to automate the current control loop and I’ve been looking into some papers to determine the best ways and I hope to be able to do so soon. And I’m definitely open to any type of collaboration on this subject.

However, an important thing to say is that we are searching to find a general solution to these problems, not a solution which works on one board/driver/motor/etc. So this adds a lot of complexity to the problem definition and the way it needs to be implemented.
It is very likely that we will never be able to reach the performance of the STM or Microchip firmware for BLDC motors simply because we are choosing not to make our code so hardware specific.

@Larock1234

This brings me to the next point. Unfortunately, I didn’t have enough time to check everything, but it seems like (at least with the B-G431B-ESC1) we have an issue with the current measurements as they are very noisy. I eventually ended up using the ST Motor Control Firmware and it gave me much better current readings which a noise that was very acceptable for the motor I used. This confirms that it’s not a hardware issue. Maybe, we (I will help as soon as time allows) should double check again the alignment, timing and settings of the ADC for the current readings. It seems to me like the noisy measurements are kind of a deal-breaker for SimpleFOC, which is very sad.

The code for the BG431B board is actually a contributed code from our community
https://community.simplefoc.com/t/b-g431b-esc1-current-control/521

and the current sensing is very specific to this board. I personally do not even own one and did not test it at all.
The PWM/ADC alignment is done using injected channels and the code uses the integrated opamps in the STMG4 chip. There is probably room for improvement in terms of performance and we’d be happy to collaborate over this.

All of our stm32 based low-side sensing is based on the ADC injected channels and TRGO events that are triggering the ADC conversion exactly in the middle of the time the low-side mosfets are active (which is effectively the start of the center aligned duty cycle). This is not the most optimal synchronization technique, I think that the STM motor suite uses the fourth timer channel to trigger the ADC which will usually result in a bit more precise measurements. But at this point we decided not to use this option because we do not want to use additional timer channel for the ADC sync as the user might prefer to use it for something else. But this might change in future.

So at the moment, to avoid potential problematic situations due to the ADC conversion lasting to long and entering the high-side activation of certain mosfets we suggest using the PWM frequencies that do not exceed 20kHz. This in most cases results in behavior.

However there are many things that can be done to improve the low-side and inline current sensing implementation that we have currently. They are really in their early stage of developement.

@Larock1234

At the same time, maybe we need to rethink the architecture and the fact that the current controllers are independent of the current measurement. It is usually beneficial to have the current controller updates aligned with the current readings (see for example the code for the TI or ST motor control dev boards). Current control is really sensitive and we want to make sure to guarantee regular, timely and quick (>= 20kHz) updates. Maybe reconsider using interrupts coupled to the ADC readings?

This would once again be very very hardware specific. It is doable of course but at the moment the choices that we’re taking are oriented towards the cross-platform implementations more than to the optimal performance.

On the separate note, SimpleFOC loop times in the low-side current sensing mode are sometimes faster than the PWM frequency and the ADC conversion which means that effectively you get more than one update of the current target per PWM duty cycle. So we might be able to achieve the behavior you’re describing even without the interrupts.

@Larock1234

Yes, I see that the board uses low side current sensing. So the ADCs have to be carefully aligned with the PWMs (which might or might not be the case, I didn’t have time to check, but I rather suspect it not to be the case). Well from what I understand, low side sensing has benefits in terms of circuit design as the opamps are closer to ground. Anyway, I don’t think we need to discuss this, I didn’t design the hardware

The PWM and the ADC conversion is aligned for the low-side current sensing at the center of the low-side active interval. Here is the diagram.

As I said already starting the conversion at the center is not the best place to start the ADC conversion, it would be better to start it a bit before to be measuring the currents at the center of the all phases low part of the duty cycle. But as long as the timing is consistent and the PWM frequency is not too high the current measurements should be good.

Finally the current control in the simplefoc is at its early stage and I am not really happy with the results yet, I hope we will be able to improve its performance in the future. And we are open to any type of collaboration
Current control is complicated and requires some experience and having more control engineers onboard would be great!

@Larock1234

Edit: Btw, why does SimpleFOC actually support both, inline and low-side for this specific board (according to the doc) if the hardware is built for low-side?

We don’t actually. We only support the low-side current sensing for this board. This is a legacy issue because when the guys implemented the low-side current sensing on this board the library did not have lowsidecurrent sense interface so they used the inlinecurrentsense class. Later we’ve added the low-side current sense interface to the simplefoc and for this board we kept both implementations. So using inline or low-side current sensing interfaces will both use the same low-side code.