Issue with DMA+PWM based signal generator

Question

I want to generate some signal on GPIO pin on Raspberry PI Zero W using DMA control blocks and PWM to pace data. However I faced with some problems while trying to generate signal at ~10-50kHz.

When I configure timings to generate 10us high state followed by 10us low state in loop (basically looping DMA control blocks) I see following picture on GPIO pin output:

In general it generates 10us pulses but sometimes something goes wrong and DMA changes GPIO state with wrong timings. It looks like this "interrupts" happens periodically each ~115us (about 8.5kHz).

Here is minimal reproducible code that generates signal on GPIO pin 23 and compiles fine on Raspberry PI Zero W: https://www.codepile.net/pile/v2JJ8L2q (I'm just modified example from https://iosoft.blog/2020/05/25/raspberry-pi-dma-programming/)

What I'm doing wrong?

Answer 1

From my experience PWM (both soft and hard) on RPI is seriously affected - at least in Python. For simple LED dimmer, where I really want no flickering when LEDs are very low light, I had to go for pigpiod. Then the problem is much less visible - at least on LEDs. The only disadvantage of pigpiod is: it consumes the CPU (deamon). However, while doing the same dimmer on Pico (in micropython) I did not notice such behaviour.

Answer 2

The problem with using DMA to write to I/O pins is that they don't have a FIFO.

In theory DMA should produce a steady steam of transfers with precise timing, but in reality that isn't true; there are gaps, then a few back-to-back transfers to catch up. That is why the peripherals such as UART, SPI and SMI have FIFOs; they bridge the gaps so that the peripheral sees a steady stream of data.

If you want to generate precise timing, SMI is the best option, see this post