esp32: Fix PWM not allowing frequencies < 611 Hz (16 bit resolution). #8210

IhorNehrutsa · 2022-01-25T15:48:26Z

This is another version of
esp32: Fix PWM not allowing frequencies < 611 Hz. #8200

It uses a maximum of 16bit PWM resolution.

Tested on ESP32

IhorNehrutsa · 2022-01-25T16:00:10Z

ESP-IDF doesn't like changing the PWM resolution.

# pwm_test_freq.py

from utime import sleep
from machine import Pin, PWM

F_MIN = 1
F_MAX = 10 #20_000_000

f = F_MIN
delta_f = 1

pin = Pin(26)

p = PWM(pin, f, duty_u16=2**16//2)
while True:
    p.freq(f)
    print(p)
    
    sleep(10)
    
    f += delta_f
    if (f >= F_MAX) or (f <= F_MIN):
        delta_f = -delta_f
        
        break

These tests are executed with uncommented lines

micropython/ports/esp32/machine_pwm.c

Lines 325 to 337 in 30b6ce8

    
               /* 
        
               // Bug: Sometimes duty is not set right now. 
        
               // Not a bug. It's a feature. The duty is applied at the beginning of the next signal period. 
        
               // Bug: It has been experimentally established that the duty is setted during 2 signal periods, but 1 period is expected. 
        
               // See https://github.com/espressif/esp-idf/issues/7288 
        
               if (duty != get_duty_u16(self)) { 
        
                   PWM_DBG("set_duty_u16(%u), get_duty_u16():%u, channel_duty:%d, duty_resolution:%d, freq_hz:%d", duty, get_duty_u16(self), channel_duty, timer.duty_resolution, timer.freq_hz); 
        
                   ets_delay_us(2 * 1000000 / timer.freq_hz); 
        
                   if (duty != get_duty_u16(self)) { 
        
                       PWM_DBG("set_duty_u16(%u), get_duty_u16():%u, channel_duty:%d, duty_resolution:%d, freq_hz:%d", duty, get_duty_u16(self), channel_duty, timer.duty_resolution, timer.freq_hz); 
        
                   } 
        
               } 
        
               */

In this PR

PWM(Pin(26), freq=1, duty_u16=32768, resolution=16, (duty=50.00%, resolution=0.002%), mode=0, channel=0, timer=0)
PWM(Pin(26), freq=2, duty_u16=32768, resolution=16, (duty=50.00%, resolution=0.002%), mode=0, channel=0, timer=0)
PWM(Pin(26), freq=3, duty_u16=32768, resolution=16, (duty=50.00%, resolution=0.002%), mode=0, channel=0, timer=0)
PWM(Pin(26), freq=4, duty_u16=32768, resolution=16, (duty=50.00%, resolution=0.002%), mode=0, channel=0, timer=0)
PWM(Pin(26), freq=5, duty_u16=32768, resolution=16, (duty=50.00%, resolution=0.002%), mode=0, channel=0, timer=0)
PWM(Pin(26), freq=6, duty_u16=32768, resolution=16, (duty=50.00%, resolution=0.002%), mode=0, channel=0, timer=0)
PWM(Pin(26), freq=7, duty_u16=32768, resolution=16, (duty=50.00%, resolution=0.002%), mode=0, channel=0, timer=0)
PWM(Pin(26), freq=8, duty_u16=32768, resolution=16, (duty=50.00%, resolution=0.002%), mode=0, channel=0, timer=0)
PWM(Pin(26), freq=9, duty_u16=32768, resolution=16, (duty=50.00%, resolution=0.002%), mode=0, channel=0, timer=0)

In #8200

set_duty_u16(32768), get_duty_u16():65536, channel_duty:262144, duty_resolution:19, freq_hz:1
PWM(Pin(26), freq=1, duty_u16=32768, resolution=19, (duty=50.00%, resolution=0.000%), mode=0, channel=0, timer=0)
set_duty_u16(32768), get_duty_u16():65536, channel_duty:131072, duty_resolution:18, freq_hz:2
PWM(Pin(26), freq=2, duty_u16=32768, resolution=18, (duty=50.00%, resolution=0.000%), mode=0, channel=0, timer=0)
PWM(Pin(26), freq=3, duty_u16=32768, resolution=18, (duty=50.00%, resolution=0.000%), mode=0, channel=0, timer=0)
set_duty_u16(32768), get_duty_u16():65536, channel_duty:65536, duty_resolution:17, freq_hz:4
PWM(Pin(26), freq=4, duty_u16=32768, resolution=17, (duty=50.00%, resolution=0.001%), mode=0, channel=0, timer=0)
PWM(Pin(26), freq=5, duty_u16=32768, resolution=17, (duty=50.00%, resolution=0.001%), mode=0, channel=0, timer=0)
PWM(Pin(26), freq=6, duty_u16=32768, resolution=17, (duty=50.00%, resolution=0.001%), mode=0, channel=0, timer=0)
PWM(Pin(26), freq=7, duty_u16=32768, resolution=17, (duty=50.00%, resolution=0.001%), mode=0, channel=0, timer=0)
set_duty_u16(32768), get_duty_u16():65536, channel_duty:32768, duty_resolution:16, freq_hz:8
PWM(Pin(26), freq=8, duty_u16=32768, resolution=16, (duty=50.00%, resolution=0.002%), mode=0, channel=0, timer=0)
PWM(Pin(26), freq=9, duty_u16=32768, resolution=16, (duty=50.00%, resolution=0.002%), mode=0, channel=0, timer=0)

robert-hh · 2022-01-25T16:36:29Z

That's what I get if I run #8200 with your script. Board ESP32_GENERIC_SPIRAM. Everything fine, and on the oscilloscope it looks right as well.

PWM(Pin(21), freq=1, duty_u16=32768, resolution=19, (duty=50.00%, resolution=0.000%), mode=0, channel=0, timer=0)
PWM(Pin(21), freq=2, duty_u16=32768, resolution=18, (duty=50.00%, resolution=0.000%), mode=0, channel=0, timer=0)
PWM(Pin(21), freq=3, duty_u16=32768, resolution=18, (duty=50.00%, resolution=0.000%), mode=0, channel=0, timer=0)
PWM(Pin(21), freq=4, duty_u16=32768, resolution=17, (duty=50.00%, resolution=0.001%), mode=0, channel=0, timer=0)
PWM(Pin(21), freq=5, duty_u16=32768, resolution=17, (duty=50.00%, resolution=0.001%), mode=0, channel=0, timer=0)
PWM(Pin(21), freq=6, duty_u16=32768, resolution=17, (duty=50.00%, resolution=0.001%), mode=0, channel=0, timer=0)
PWM(Pin(21), freq=7, duty_u16=32768, resolution=17, (duty=50.00%, resolution=0.001%), mode=0, channel=0, timer=0)
PWM(Pin(21), freq=8, duty_u16=32768, resolution=16, (duty=50.00%, resolution=0.002%), mode=0, channel=0, timer=0)
PWM(Pin(21), freq=9, duty_u16=32768, resolution=16, (duty=50.00%, resolution=0.002%), mode=0, channel=0, timer=0)

robert-hh · 2022-01-25T16:52:07Z

Same result with UM_FEATHERS2 and GENERIC_C3. Only for the C3 I had to use the 10 <= f < 20 range.

robert-hh · 2022-01-25T16:54:46Z

Note: I have a glitch in my build system, in that changes to machine_pwm.c do not cause a re-compile of it. I did not look for the reason yet, but remove machine_pwm.c.obj manually instead to force the re-compile.

robert-hh · 2022-01-25T17:54:42Z

Second note: Sometimes, after a a power cycle. I see the phenomenon that the PWM does start extremely slow at the first frequency set with your test code, whatever it is, 1 Hz or 10 Hz. The next setting then is fine. So there seems to be an uninitialized state somewhere. When starting with 1Hz, the duty_u16 shown is 65538, when starting with 9Hz the duty shown is 262144(2**18). But the PWM output is fine. It runs at the proper frequency and duty cycle.

robert-hh · 2022-01-25T19:17:23Z

I catched an extended cycle at the transition from 3 to 5 Hz, resolution 18 to 17. It shows an extended cycle. On the left side, it's 3 Hz, on the right it's 4 Hz. And sometimes the values are set completely odd. Most likely a problem in the esp-idf.

IhorNehrutsa · 2022-01-25T21:21:01Z

That's what I get if I run #8200 with your script. Board ESP32_GENERIC_SPIRAM. Everything fine, and on the oscilloscope it looks right as well.

Have you uncommented PWM_DBG?

// #define PWM_DBG(...)
#define PWM_DBG(...) mp_printf(&mp_plat_print, __VA_ARGS__); mp_printf(&mp_plat_print, "\n");

P.S. Sorry for the delay and excuse me for spending your time.

robert-hh · 2022-01-26T07:27:55Z

I uncommented the block in set_duty_u16(). Whatever, I have seen this glitch in the PWM trace itself: there are disturbed clock cycles when the resolution changes. Here is another example when switching between 1220 and 1221 Hz (resolution 16->15). The same happens when switching between resolution 15<->14, and so on.

IhorNehrutsa · 2022-01-26T07:39:37Z

Most likely a problem in the esp-idf.

Yes, most likely in

micropython/ports/esp32/machine_pwm.c

Lines 322 to 323 in 30b6ce8

    
           check_esp_err(ledc_set_duty(self->mode, self->channel, channel_duty)); 
        
           check_esp_err(ledc_update_duty(self->mode, self->channel));

robert-hh · 2022-01-26T07:48:36Z

I've seen also states where the PWM does not start at all without a resolution change happening. That's even more strange.

IhorNehrutsa · 2022-01-26T08:59:17Z

It seems that issue is resolved 3 hours ago (but I don't see a commit with fixes):
espressif/esp-idf#7288

@espressif-bot espressif-bot added Resolution: NA Status: Done Resolution: Done and removed Status: In Progress Resolution: NA labels 3 hours ago

IhorNehrutsa · 2022-01-26T11:10:49Z

May I rename
PWFREQ to PWM_FREQ
PWRES to PWM_RES_10_BIT
?

robert-hh · 2022-01-26T12:24:35Z

Sure. But we should make only one PR for that issue. Do you want to keep yours?

IhorNehrutsa · 2022-01-26T12:29:37Z

Yes if you not against.
This PR has a smaller changing of resolution.

robert-hh · 2022-01-26T13:02:58Z

This PR has a smaller changing of resolution.

OK. So I'll close mine then. Tell me, if you are finished, Then I can make a few tests,

IhorNehrutsa · 2022-01-26T13:11:05Z

Ready

robert-hh · 2022-01-26T14:31:51Z

Did a test series with GENERIC_SPIRAM, GENERIC_C3 and UM_FEATHERS2
GENERIC_SPIRAM works down to 1 Hz.
GENERIC_C3 works down to 10Hz
UM_FEATHERS2 works down to 1 Hz, with a glitch: When switching from 10..50Hz to 1Hz it takes a few seconds until the PWM output starts again at 1 Hz. Maybe the known esp-idf glitch, maybe another. Switching from 2..9Hz to 1Hz or >59Hz to 1 Hz is no problem (except for the one below).

The glitch when changing between resolutions still exists.It must be fixed in the esp-idf. The UM_FEATHERS2 has a glitch on EVERY frequency change, sometimes for several signal periods, when the resolution changes as well.

According to that test, the machine_pwm.c code works, but bugs in the esp-idf still cause glitches. But that should have existed already before and is identified by espressif.

dpgeorge · 2022-02-01T05:50:01Z

Thanks for this, rebased and merged in a5e64c2 and 15e65b7 (the last commit contains the 4 clean-up commits).

robert-hh · 2022-02-13T16:59:01Z

@IhorNehrutsa There is another tiny hiccup. For frequencies < 611, the duty rate cannot be set with the constructor after hard reset. It will always be 512. See this post in the forum: https://forum.micropython.org/viewtopic.php?f=18&t=11975&p=65175#p65083

esp32: Fix PWM not allowing frequencies < 611 Hz.

bc4d030

esp32\machine_pwm.c: Remove UI_RES_SHIFT macro.

d32ffd8

IhorNehrutsa added 3 commits January 26, 2022 14:37

esp32/machine_pwm.c: Rename PWFREQ to PWM_FREQ.

a98a019

esp32/machine_pwm.c: Rename PWRES to PWM_RES_10_BIT.

a1c702d

esp32/machine_pwm.c: Use UI_RES_16_BIT flag instead of HIGHEST_PWM_RES.

56b2a59

dpgeorge added the port-esp32 label Jan 27, 2022

dpgeorge mentioned this pull request Jan 29, 2022

PWM on esp32 (v1.18) #8225

Closed

dpgeorge closed this Feb 1, 2022

IhorNehrutsa deleted the pwm611_res16 branch July 10, 2023 07:43

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

esp32: Fix PWM not allowing frequencies < 611 Hz (16 bit resolution). #8210

esp32: Fix PWM not allowing frequencies < 611 Hz (16 bit resolution). #8210

IhorNehrutsa commented Jan 25, 2022 •

edited

IhorNehrutsa commented Jan 25, 2022

robert-hh commented Jan 25, 2022

robert-hh commented Jan 25, 2022

robert-hh commented Jan 25, 2022

robert-hh commented Jan 25, 2022

robert-hh commented Jan 25, 2022

IhorNehrutsa commented Jan 25, 2022 •

edited

robert-hh commented Jan 26, 2022 •

edited

IhorNehrutsa commented Jan 26, 2022

robert-hh commented Jan 26, 2022

IhorNehrutsa commented Jan 26, 2022

IhorNehrutsa commented Jan 26, 2022

robert-hh commented Jan 26, 2022

IhorNehrutsa commented Jan 26, 2022

robert-hh commented Jan 26, 2022

IhorNehrutsa commented Jan 26, 2022

robert-hh commented Jan 26, 2022

dpgeorge commented Feb 1, 2022

robert-hh commented Feb 13, 2022 •

edited

esp32: Fix PWM not allowing frequencies < 611 Hz (16 bit resolution). #8210

esp32: Fix PWM not allowing frequencies < 611 Hz (16 bit resolution). #8210

Conversation

IhorNehrutsa commented Jan 25, 2022 • edited

IhorNehrutsa commented Jan 25, 2022

robert-hh commented Jan 25, 2022

robert-hh commented Jan 25, 2022

robert-hh commented Jan 25, 2022

robert-hh commented Jan 25, 2022

robert-hh commented Jan 25, 2022

IhorNehrutsa commented Jan 25, 2022 • edited

robert-hh commented Jan 26, 2022 • edited

IhorNehrutsa commented Jan 26, 2022

robert-hh commented Jan 26, 2022

IhorNehrutsa commented Jan 26, 2022

IhorNehrutsa commented Jan 26, 2022

robert-hh commented Jan 26, 2022

IhorNehrutsa commented Jan 26, 2022

robert-hh commented Jan 26, 2022

IhorNehrutsa commented Jan 26, 2022

robert-hh commented Jan 26, 2022

dpgeorge commented Feb 1, 2022

robert-hh commented Feb 13, 2022 • edited

IhorNehrutsa commented Jan 25, 2022 •

edited

IhorNehrutsa commented Jan 25, 2022 •

edited

robert-hh commented Jan 26, 2022 •

edited

robert-hh commented Feb 13, 2022 •

edited