[HRTIM ?] PWM glitches when using extreme duty cycle values 0.0 and 1.0

[pierre-haessig]

[HRTIM ?] PWM glitches when using extreme duty cycle values 0.0 and 1.0

Nov-Dec 2025

Using the Ownverter 1.1 board, with 1.0 Power API code, there are glitches when using duty cycle values equal or close to 0.0 or 1.0.

Here are two cases, with slightly different symptoms. Both were created using three-phase islanded inverter application https://github.com/pierre-haessig/ownverter-islanded/tree/completed (using the completed branch, since main is the student version with a few blanks to be filled.)

• sinusoidal duty cycles PWM
• square wave modulation

In all cases, experimental conditions are: three-phase inverter connected through an inductive filter (630 µF) to either a balanced R=10Ω load or R+L load (L=76 mH).

In all cases, the control task period is 0.1 ms = 100 µs (10 kHz)

Sinusoidal duty cycles PWM

In this application, with pure R load, the duty cycles are computed as:

duty_a = duty_offset + duty_amplitude * ot_sin(v_angle);
duty_b = duty_offset + duty_amplitude * ot_sin(v_angle - 2.0/3.0*PI);
duty_c = duty_offset + duty_amplitude * ot_sin(v_angle - 4.0/3.0*PI);

with duty_offset set to 0.5 and duty_amplitude set to 0.5, so that the duties reach 0 or 1 once per electric period (50 Hz/20 ms).

Here are the resulting oscilloscope captures, with incremental zoom around the top of the sinusoid:

• yellow-(C1): inverter voltage (phase to PGND or phase to Neutral, I don't remember)
• pink (C2): current through load
• green and cyan (C3, C4): swiching signals (CHA1 and CHA2 pins)

Observations:

• the sinusoidal current is heavily perturbed 6 times per period, in fact each time one of the three phases has its duty cycle close to 0 or 1.
• zooming to the top of the sinusoid, we see that the glitches corresponds to undue switchings while the PWM output should stay constant.
• the switchings happen at integer multiple of the control period (100 µs), so that they are linked to calls to setDutyCycle, with changing values of duty

Square wave modulation

The square wave modulation is done by setting the duty cycle to 0.0 or 1.0 on appropriate half periods:

if (square_wave) {
    if (v_angle <= PI) duty_a = 1.0;
    else duty_a = 0.0;
    if (ot_modulo_2pi(v_angle - 2.0/3.0*PI) <= PI) duty_b = 1.0;
    else duty_b = 0.0;
    if (ot_modulo_2pi(v_angle - 4.0/3.0*PI) <= PI) duty_c = 1.0;
    else duty_c = 0.0;
}

The resulting oscilloscope captures are obtained here with a pure R load (still with L filter, so that L/R = 63 µs).

Results are even more suprising than with sine PWM.

Zoom around a half period switch point	Zoom around a constant duty region

Observations:

• despite the fact that the duty is kept constant, there almost are periodic glitches (every 300 to 400 µs)
  • it doesn't seem periodic with the control task
• even more surprising, no glitch is visible on the PWM output (CHA1 and CHA2 pins) which are constant!

[luizvilla]

Hello @pierre-haessig,

   Thank you for the issue. I'm going to work on it today. 
   It seems to me there are two separate issues here: 

• A recurrent switching even when the PWM signal from the microcontroller is stable
• A toggle between 0 and 1 for the PWM signals from the microcontroller when values are close to 0 and 1.

I suspect the first one is a driver issue. Could it be that the driver of the mosfets forces switching ? @jalinei

The second one looks like a software glitch due to the way in which the 1 and 0 are done. I'll look into it.

[luizvilla]

I found the second error.

When the duty cycle approaches 1, there is indeed a software glitch.

Here's the duty cycle at 0.9975

Here's the duty cycle at 0.9976

And here's the duty cycle at 0.9997

You can reproduce the error by simply using your main branch code but raising the duty cycle by 0.0001 at a time.

The explanation lies on the fact that the HRTIM PWM's on the STM32G4 do NOT support 1/0 mode.
We implemented it by swapping the two outputs. However, we only swap when we are within 3 counts of the final clock, as per the documentation and the glitch resolution here #110 .

Apparently the criteria for going into swap is too small.
I will enlarge it and test a solution to the problem.

[luizvilla]

I just did another test, looking up the raw values after which the HRTIM gets the glitch.
It is not 3, it is 30...

POW: duty a= 50% o= 50% @50 Hz | Vh  0.00 V, Ih -10.00 A, D1 0.997699 , D1R 13568 , P1 13600, 

Where :

• duty a is the duty cycle in %
• o is the offset
• Vh and Ih are measurements (off because I'm using a spin board for testing)
• D1 is the duty cycle as a float between 1 and 0
• D1R is the duty cycle in uint
• P1 is the period in uint

The answer, obviously, was right in front of my eyes:

This table shows the maximum value depending on the pre-scaler that is set on the clock.

The solution is thus to create a table with these values and use it when testing the moment to swap.

[luizvilla]

I implemented a solution on #145.

@pierre-haessig Can you please :

• add my fork of the Core to your VSCode git graph,
• checkout my branch,
• put your main.cpp code in it and
• put the results here

Best
Luiz

[pierre-haessig]

Hello Luiz, thanks a lot for working on the PWM glitches. I've finally made the test of your branch.

Results:

• unfortunately the bug persists 😢
• but the good news is that your changes have positive effect: half of the glitches are reduced!

See results below (left: old, right: your hrtrim_pathc branch). Specifically, I see:

• highly reduced glitches 3 times per period which corresponds to regions where the duty cycle is close to 1.0 on one of the three phases
• on the other hand, I see no changes for the other 3 times per period where the duty cycle is close to 0.0

---

Also, there is a minor issue with your branch which may actually mean that there are missing bits: in platformio.ini, I cannot select the proper shield

# Selected Shield Name and Version
board_shield = ownverter
board_shield_version = 1_1_0

ownverter version 1_1_0 is on available, only 1_0_0.

[luizvilla]

@pierre-haessig Thank you for the feedback.
Can you please try on the square wave modulation please?