digitalPulse() jitter/offset

[espruino-discuss3]

Posted at 2018-11-06 by barbiani

Hello,

I was evaluating if the digitalPulse() (on ESP32 v2.00) timing would meet the requirements of a project and found that:

pinMode(D32, 'output');
setInterval(()=> {
      digitalPulse(D5, 1, 2);
}, 50);

Produces pulses from about 2096us to 2160us.

Being generated by hardware I wonder if it can do any better. Maybe the original espruino does not have the jitter?

Thank you.

[espruino-discuss3]

Posted at 2018-11-06 by Robin

Tue 2018.11.06

Although I'll agree the accuracy isn't specified,

https://www.espruino.com/Reference#l__global_digitalPulse

the implication is millisecond(s) wide pulses.

Have you tried using an array of pulse values (see description in link above) and just a single digitalPulse() to rule out whether setInterval() is adding unnecessary (likely) intermittent delay?

Considering:

'It uses a hardware timer to produce accurate pulses'

For a 2msec pulse 4%-8% error is rather odd. . . .

---

Posted at 2018-11-06 by barbiani

I understand that a single call to digitalPulse may have better accuracy between pulses.. but as the application will fire a pulse after another sensor decodes a pattern. The setInterval came from the test code I added to simulate the hardware.

The question now is if the original espruino does better than the ESP32 port.

---

Posted at 2018-11-12 by @gfwilliams

if the original espruino does better than the ESP32 port.

Yes - the official boards use an IRQ-based timer. Just tried this on a Pico and it's exactly 2ms to as near as I can measure on my digital scope (so well within 1%)... And if that's not good enough there's some documentation on how to use the timer peripherals in the device itself.

There was a recent change in ESP8266 courtesy of @MaBecker that fixed the issues that has, and it does look like the ESP32 is using IRQs so I'm not sure why there would be the variance.

I just had a quick look and @jumjum it could be that gettimeofday may be using the normal low-speed RTC, which might explain some timekeeping issues: https://github.com/espressif/esp-idf/blob/master/examples/protocols/sntp/README.md#timekeeping

While the timer is accurate, it's set using the current time as a base - so if the timekeeping is 'grainy' it can screw things up.

---

Posted at 2018-11-12 by Wilberforce

We are using both timers as suggested:

https://github.com/espruino/EspruinoBuildTools/blob/master/esp32/build/app/sdkconfig#L200
CONFIG_ESP32_TIME_SYSCALL_USE_RTC_FRC1=y

---

Posted at 2018-11-12 by barbiani

Well... then their timer implementation is not good! I was hoping to see at most 4us jitter.

The idf has a timer jobs linked list. Does espruino run another one on top of it?

---

Posted at 2018-11-13 by @gfwilliams

Does espruino run another one on top of it?

Yes, but for something like digitalPulse there's one job queued up, so there is no real overhead involved.

Espruino's timer implementation could be improved (eg. for digitalPulse there's no need to use the current time at all), but it's reasonably low priority at the moment as it's quite a big change and it works fine on official boards. There's an issue for it here: espruino/Espruino#1444

If you really want accurate timing without actually buying an official board it's possible there's some spare timer hardware on ESP32 that you could set up yourself with peek and poke.

---

Posted at 2018-11-13 by Wilberforce

If you set up a pwn output is the timing for this accurate?

---

Posted at 2018-11-14 by barbiani

analogWrite(D5,0.5, { freq : 3000 } ); is perfectly stable. Can not see jumps down to 5ns per division. Frequency reads 3003hz.

digitalPulse(D5, 1, 1); // 1.106ms
digitalPulse(D5, 1, .2); // 0.5ms

---

Posted at 2018-11-14 by Wilberforce

I think this might be the issue here:

https://github.com/espruino/Espruino/blob/9bb8c66015d8b9fa1aaa80a4eaab217e2f308be0/targets/esp32/jshardware.c#L623

If the timer is less than 30 - it's set the 30. @jumjum - can you recall why?

---

Posted at 2018-11-15 by Wilberforce

As a possible work around - you might be able to use the pwm output and do a setwatch on that - and in the setwatch- use that to drive another pin low then high and low again. You can use a variable flag to say if you want the out put in the output pin.

Just an idea.

---

Posted at 2018-11-15 by barbiani

So, is the pwm software driven? Or espruino will use hw interrupts for the pwm edges?

---

Posted at 2018-11-15 by Wilberforce

No - the pwm on the Esp32 is hardware driven.

Or another thought - you could send an array of o,1,1,1,0 and then adjust the number of 1s to be close to 2us

---

Posted at 2018-11-15 by @gfwilliams

Or maybe:

analogWrite(D5,0.002, { freq : 1 } );
setTimeout(function() { digitalWrite(D5,0); }, 500);

So use the hardware PWM to do the output, then cancel it in software after the first one - leave the duty cycle low enough that any jitter in execution by Espruino won't be an issue.

---

Posted at 2018-11-15 by JumJum

Sorry, no bell is ringing on that.
Could be in there from beginning by nkolban ?

---

Posted at 2018-11-15 by @gfwilliams

Git blame to the rescue! espruino/Espruino@1b3f15a

I imagine it will break if you don't at least limit it to 1us, but it feels like make 30 was there for a reason :)

---

Posted at 2018-11-18 by barbiani

Looks like the original esp32 port used RMT then changed to the Espruino style. I wonder why.

@gfwilliams Can the timer functions be used with or changed to use the floating point delay to achieve sub ms timing?

Sub millisecond would only make sense in the digitalPulse function, because I see that:

setInterval(()=> {
    D5.toggle();
  }, 5);

Gives a high pulse from 3.6 to 7ms.

---

Posted at 2018-11-18 by Wilberforce

Looks like the original esp32 port used RMT then changed to the Espruino style. I wonder why.

What do you mean by espruino style?

---

Posted at 2018-11-18 by barbiani

That came from the source code comments!

The RMT code is commented out... with it and added code from the nrf52, it says.

---

Posted at 2018-11-19 by @gfwilliams

Can the timer functions be used with or changed to use the floating point delay to achieve sub ms timing?

That won't be your issue. The timer functions are currently using us timing anyway.

espruino style?

He's probably referring to it using jstTimer rather than what amounted to a while(getTime()<endTime); loop.