Varying duty_cycle on PWMOut very occasionally results in a high blip

[kevinjwalters]

I've been using PWMOut for audio use and I heard a few anomalies on 4.0.0-beta.3. I've looked into it briefly and it appears that very occasionally there's a high blip on my PWM. @deshipu asked if it's one cycle on Discord and I think it is although it's difficult to tell on my traditional (non storage) oscilloscope. It can be recreated with this code on a Circuit Playground Express (CPX):
[EDIT by dhalbert to update example to use pwmio]

Press any key to enter the REPL. Use CTRL-D to reload.
Adafruit CircuitPython 4.0.0-beta.3 on 2019-03-07; Adafruit CircuitPlayground Express with samd21g18
>>> import board, pulseio, time
>>> p6 = pwmio.PWMOut(board.A6, duty_cycle=32768, frequency=100, variable_frequency=True)
>>> while True:
...     for i in range(4096, 28672, 1000):
...         p6.duty_cycle = i
...         time.sleep(0.1)
...     for i in range(28672, 4096, -1000):
...         p6.duty_cycle = i
...         time.sleep(0.1)

If you watch the square wave for about 30 seconds you're likely to see one or two.

I tried same thing on board.A1 and watched it for a few minutes and I've not seen any problem there.

[kevinjwalters]

I can connect up a logic analyser and try and capture it to check duration of the high if this is needed?

[kevinjwalters]

From discord:

@sommersoft well on the CPX, A6 (PB09) is a TC only, where A1 (PA05) is a TCC only. so, that could be a vector to dive into

[dhalbert]

@kevinjwalters I'll check the TC code, but could you also check using pin A1 and see if you also see a blip? I'm not sure the TC's have the same kind of double-buffering synchronization you need to avoid glitches like this.

[kevinjwalters]

I've checked A1 and didn't noticed any problems over a 4 minute period.

[kevinjwalters]

I put an Ikalogic SQ25 logic analyser on A6 to verify what I was hearing and seeing on non-storage oscilloscope, it does appear to be a single cycle blip which has the effect of joining with two either side into one long high "triple" pulse. To me these seem random but the code is modifying duty_cycle with values that differ very slightly over time so it could be there's a particular transition between two values or two value ranges that creates this.

[dhalbert]

The TC's are not double-buffered like the TCC's, so changing the duty cycle can cause these glitches. There's some discussion of a workaround here: https://medium.com/@nebsp/smoothly-changing-a-timers-frequency-on-the-arduino-zero-1f6cd285369a, though I have to study this a bit.

[kevinjwalters]

Thanks for looking at this. I'm still surprised these are so noticeable when the pwm is used for sound, I only found this because I heard the audio being a bit clicky.

[tannewt]

@dhalbert Do you plan on doing this for 4.0? Seems like we could do this in 4.x.

[dhalbert]

I'll move it to 4.x. I know what I want to do, but I need to test it too.

[dhalbert]

@kevinjwalters I am experimenting with adjusting the TC count when the duty cycle is modified, to make smoother transitions. Would you like to try this UF2? I haven't tested it myself.
cpx-smooth-pwmout.uf2.zip
It's a build of https://github.com/dhalbert/circuitpython/tree/smooth-tc-pwmout

[kevinjwalters]

Sure, might have time this weekend to have a look, thanks.