How to generate three level pwm for power converters?

[AAlHa.1]

Hello everyone, I have medium level in stm32 mcu's like using adc's, timers and also HRTIM peripheral in 32f334 and G4 series but I can't figure out the optimum way to generate three level spwm waveforms for using with NPC inverter that I'm designing now. I searched a lot across the web and it's shame the such a great mcu like stm32 doesn't have a single example on how generate such a pwm scheme. Even the applications notes from st doesn't mention that and they don't provide source code for it like

"STDES-PFCBIDIR - 15 kW, three-phase, three-level converter". I prefer using hrtim as I'm familiar with. Thank you in advance

[waclawek.jan]

What is three level PWM? Can you post sketch of the expected waveform?

JW

[AAlHa.1]

this is the concept block diagram and the required pwm waveforms

[waclawek.jan]

How does this differ from any other cycle-by-cycle-controlled PWM?

You probably want to review the timer-DMA examples in Cube, where from a prepared table, related reghisters (CCRx) in timer are changed by DMA triggered from Update of the same timer, through the TIMx_DCR/TIMx_DMAR mechanism.

JW

[AAlHa.1]

Now I know I have to use two output compare (and complementary) units, The problem when I use hrtim, comparing very low and very high values results an unpredictable behavior (the minimum compare value is not 0 and the max compare value is not TIM_PERIOD).

[waclawek.jan]

> The difference is I have to compare the sinewave to two sawtooth signals, not one like the traditional pwm​

Well, that just means using two channels, as you've remarked above. The difference in the table-DMA-driven approach is then, that DMA does not feed one channel but two, that's why using the DCR/DMAR mechanism.

I am not familiar with the detailed working of HRTIM.

JW

[waclawek.jan]

> I am not familiar with the detailed working of HRTIM.

The A-F timing units in HRTIM each have its individual DMA trigger/channel so they can simply work out of two tables determining their individual output waveforms.

For the "idle" state, I'd try to use the fact that

A compare value greater than the period register value will not generate a compare match event.

in which case only the set/reset assigned to Period would be active.

JW