drivers/ws281x: improve timing for ESP32x

[gschorcht]

Contribution description

This PR provides a small change which improves the timing for ESP32x SoCs.

If overhead like the loop control or the calculation of the waiting times for the next bit are performed while waiting for the end of the LOW phase, the time required for such operations is included in the LOW phase. This makes both the LOW phase and the period more precise.

According to the definitions

#define WS281X_T_DATA_NS                (1250U)
#define WS281X_T_DATA_ONE_NS            (650U)
#define WS281X_T_DATA_ZERO_NS           (325U)

the following timing is used:

Parameter	Value	Master	this PR
1-Bit Period	1250	1820	1400
1-Bit HIGH	650	690	710
1-Bit LOW	600	1130	690
0-Bit Period	1250	1880	1400
0-Bit HIGH	350	380	400
0-Bit LOW	900	1500	1000

Timing with current master:

Timing with this PR:

Testing procedure

tests/driver_ws281x should still work.

Issues/PRs references

[riot-ci]

Murdock results

✔️ PASSED

c40e015 drivers/ws281x: improve timing for ESP32x

Success	Failures	Total	Runtime
6882	0	6882	10m:32s

Artifacts

• Documentation preview

[maribu]

bors merge

Nice one, thx :)

[bors]

Build succeeded:

• Murdock
• static-tests
• tools-build-success

[gschorcht]

Thanks for reviewing and merging.

BTW, I already have a prototypic driver version that uses the RMT controller of the ESP32x SoC. With the RMT it is possible to generate arbitrary digital waveforms like NEC remote control signals or even WS2812 signals.

The advantages are that the CPU is not busy when generating the WS218x signal and the phase times are accurate to nanoseconds. The disadvantage is that the configuration of the RMT is so complex due to its flexibility that I simply used the ESP-IDF API for RMT. However, this requires about 6 kBytes more ROM, about 3 kBytes more IRAM, and 88 bytes more RAM.

I also decided to use the ESP-IDF API to keep the flexibility of the RMT for other applications. The WS218x driver just needs only one channel of 4 or 8 available channels. We could try to hard-code the configuration of the RMT for WS218x driver, but this would prevent other applications using RMT if the WS281x driver is used.

[gschorcht]

However, this requires about 6 kBytes more ROM, about 3 kBytes more IRAM, and 88 bytes more RAM.

We could make the use of the hardware driver optional.

[maribu]

Sounds like a plan. Since the ws281x backend can be flexible be chosen, one could even opt for the bit-banging driver if ROM and RAM is scarce. But given that all ESP32 MCUs are in the high end hardware leage, most use cases likely can affort to spend RAM and ROM for it. So I'd even say the RMT should be the default backend on ESP32, and people with special use cases that are heavy on memory can opt for the bit-banging driver instead.