Issue with a short pulse during initialization of two PWM pins

[klinwer]

On a ESP32-C3-DevKitM-1 with firmware esp32c3-20230426-v1.20.0 a short pulse (~100ms) is released during the pwm initialization.

1. pwm0 = PWM(Pin(3), freq=200, duty_u16=0) without any pulse as expected.
2. pwm1 = PWM(Pin(4), freq=200, duty_u16=0) -> A short pulse is released on Pin(3), Pin(4) without any pulse as expected.

I did some tests with different pins and got everytime the same result. The initialization of the second Pin releases a pulse on the fist one.

Is this caused by micropython or is it hardware related?

[robert-hh]

I could not replicate that with the nightly build. Please try the recent nightly build.

[ricksorensen]

wink- maybe initialize a PWM with a disconnected GPIO pin first to get over the first initialization without driving a real component?

[klinwer]

That's a nice idea. My solution is to use an aditional pin:

• init pwm enable
  pin_pwm_enable = Pin(6), Pin.OUT)
  pin_pwm_enable.value(0)

• Init pwm gpios
  pwm0 = PWM(Pin(3), freq=200, duty_u16=0) # create PWM object from a pin
  pwm1 = PWM(Pin(4), freq=200, duty_u16=0) # create PWM object from a pin
  pin_pwm_enable.value(1)

This is already working in combination with two AND gates.

[ricksorensen]

I saw the same results with my UM TinyS3. I looked at the esp32 machine_pwm.c code.
I noticed the configure_channel function set the the ledc_channel_config with a non-zero duty cycle that looked like half. This function is called when timers/channels are being initialzed first time during the PWM init (constructor) via mp_machine_pwm_init_helper function. (line 544). After this, the PWM freq and duty cycle are set for the particular instance.
The constructor sequence seems to be:

• Find timer to use
• Find channel to use
• if channel not configured- configure_channel()
• set frequency
• if duty cycle passed in - set duty cycle to that ; else - set default duty to 50%, 10 bit resolution

So .... I changed line 199 in configure_channel

        .duty =  (1 << (timers[TIMER_IDX(self->mode, self->timer)].duty_resolution)) / 2,

to

        .duty = 0,

A short bout of initial testing indicates that initial pulse is absent, and if a non-zero duty cycle is given constructor the PWM starts as expected ... more evaluation is needed to look at behaviors on soft-reset, deinit(), and init()

[robert-hh]

There is a PR #10854 updating PWM for ESP32, which is present since long time. It may be interesting if that behaves identical.

[ricksorensen]

PR #10854 seems to have addressed this problem - I believe it changes the configure_channel function to set the duty to the requested duty cycle - but I didn't see the leading pulse from the tests above. I just used the machine_pwm.c file from that PR and not all the changes.