How to choose the Timer Input Capture Filter value

[GBert.2]

I'm using an STM32 to control a DC motor with a dual hall encoder feedback.

To monitor motor position I want to use a General Purpose Timer (16bits) in Encoder Mode.

As my input is noisy, I want to filter the signal using the Input Capture filter.

My problem: I don't find any information about the ICTimer.

Here's what I found in the AN4013 - STM32 cross-series timer overview Application note

> "Program the filter by writing the IC1F[3:0] bits in the CCMRx register, and the prescaler by writing the IC1PSC[1:0] if needed"

What kind of filter is this?

How do I choose the value for IC1F?

What is the cutoff frequency?

Your help is appreciated.

[KnarfB]

You are not specifying the chip, so I took G4 series as an example. Ref. manual RM0440 Rev 6 says in 28.3.9 Input capture mode "Let’s imagine that, when toggling, the input signal is not stable during at must 5 internal clock cycles. We must program a filter duration longer than these 5 clock cycles.". So the filter is just debouncing transient transitions. It is named "Filter downcounter" in the figures.

ICPS is a prescaler. No idea what it is used for except for extending the counter range.

hth

KnarfB

[waclawek.jan]

Which STM32?

> As my input is noisy

Specify.

> What kind of filter is this?

It's very simple: if input level changes, output changes only after filter samples the same level for the prescribed number (N) of samples. You can select several levels of N and several prescalers for the sampling frequency.

JW

[Tesla DeLorean]

Wouldn't the prescalers divide the input pulses, so you'd need 4 or 8, or whatever, to get a single count?

The clock divider of the filter drops the sampling frequency, ie clocking of the synchronizer circuit

[waclawek.jan]

Sorry for the confusion with the many prescalers:

• there is a common prescaler for filters in all 4 channels (and for dead-time detection where applicable), set in TIMx_CR1.CKD (resulting in fDTS)
• individual settings of TIMx_CCMRx.ICxF use a further subdivision of either directly the timer's internal clock (***_INT) or its prescaled version (fDTS)
• as Clive said above, IC1PSC determines, after how many filtered edges will an actual capture occur.

JW

[GBert.2]

Hi @Community member​ ,

Thanks for you answer.

> It's very simple: if input level changes, output changes only after filter samples the same level for the prescribed number (N) of samples. You can select several levels of N and several prescalers for the sampling frequency.

What you're describing seems like a counter.

If it's the case, how is the hysteresis working?

> as Clive said above, IC1PSC determines, after how many filtered edges will an actual capture occur.

From my observation, I don't think it's the case.

The filter seems to behave like a low pass filter: filtering out rapid variations, regardless of the number of edges.

[waclawek.jan]

> What you're describing seems like a counter.

You can look at it as a counter. This is exactly how debouncing algorithms work in software, too.

> If it's the case, how is the hysteresis working?

What hysteresis are you referring to?

>> as Clive said above, IC1PSC determines, after how many filtered edges will an actual capture occur.

>  From my observation, I don't think it's the case.

> The filter seems to behave like a low pass filter: filtering out rapid variations, regardless of the number of edges.

ICxPSC does not control the filter itself.

JW

[GBert.2]

Hi everybody,

I think I don't understand what you're saying.

When I implement a low pass filter with a counter, I put two hysteresis thresholds.

Here's a rapid drawing about the relationship between the input, the counter and the output.

 

[waclawek.jan]

I fail to see how this works as a filter.

The following figure from AN4776 illustrates how the input capture filter rejects short pulses (i.e. noise):

It's not a perfect solution, of course. It introduces delay, and it's aimed at removing random switching noise not periodic interference (in which case it may produce confusing results).

JW

[GBert.2]

The AN4776 had the details needed.