search

esp-rs / esp-hal

no_std Hardware Abstraction Layers for ESP32 microcontrollers
https://docs.esp-rs.org/esp-hal/
Apache License 2.0
789 stars 218 forks source link

Double check results from calibrated ADC readings #1203

Open bjoernQ opened 9 months ago

bjoernQ commented 9 months ago

At least on C6 we see suspicious results.

e.g. for AdcCalCurve

PIN2 ADC reading = 7734 mV
PIN2 ADC reading = 6681 mV
PIN2 ADC reading = 6832 mV
PIN2 ADC reading = 6992 mV

for AdcCalLine

PIN2 ADC reading = 7985 mV
PIN2 ADC reading = 7343 mV
PIN2 ADC reading = 7506 mV
PIN2 ADC reading = 7007 mV
PIN2 ADC reading = 7017 mV
PIN2 ADC reading = 7506 mV

AdcCalBasic

PIN2 ADC reading = 8223 mV
PIN2 ADC reading = 7218 mV
PIN2 ADC reading = 7385 mV
PIN2 ADC reading = 7553 mV
PIN2 ADC reading = 7207 mV
PIN2 ADC reading = 7385 mV
jessebraham commented 8 months ago

Not sure what's going on here, but based on my testing C6 worked without issue. I did see incorrect readings for the C3 and S3, however.

Possibly due to different chip revisions or something? Not sure, needs further investigating. I did double-check the curve coefficients at the very least and they seem to be more or less correct.

jessebraham commented 5 months ago

I've just done a quick round of testing on all devkits I have available to me. I've included the revisions as reported by espflash board-info as this may make a difference. Values listed in each column are when connected to the GND and 3V3 pins respectively, and are printed by the adc_cal and adc examples.

Anyway, results are all over the place, so we need to spend some time fixing all this for sure.

Chip None (GND/3V3) Basic (GND/3V3) Line (GND/3V3) Curve (GND/3V3)
ESP32 (v3.0) 4095/4095 — — —
ESP32-C2 (v1.0) 2138/4095 0/3450 0/3343 —
ESP32-C3 (v0.3) 1690/4095 7/4095 5/2960 6/2860
ESP32-C6 (v0.0) 2160/4081 1/3393 0/3262 0/3250
ESP32-H2 (v0.0) 2160/4081 — — —
ESP32-S2 (v0.0) 8192/3470 — — —
ESP32-S3 (v0.1) 1960/4095 0/4096 0/3456 0/64562

I know it's a bit of a pain, but if 1 or 2 other people would be able to do this same test that would be helpful. More data won't hurt here.

bjoernQ commented 5 months ago
Chip None (GND/3V3) Basic (GND/3V3) Line (GND/3V3) Curve (GND/3V3)
ESP32 (v1.0) 0/4095 — — —
ESP32-C2 (v2.0) 2155/4095 0/3460 0/3308 —
ESP32-C3 (v0.3) 1674/4095 10/4095 10/2885 13/2796
ESP32-C6 (v0.0) 2145/4081 1/3377 5/35367❓ 0/34789 ❓
ESP32-H2 (v0.1) 2144/4081 — — —
ESP32-S2 (v0.0) 2825/8191 — — —
ESP32-S3 (v0.1) 1971/4095 0/4095 0/3540 0/3159
JurajSadel commented 5 months ago
Chip None (GND/3V3) Basic (GND/3V3) Line (GND/3V3) Curve (GND/3V3)
ESP32 (v3.0) 3830/4095 — — —
ESP32-C2 (v2.0) 2180/4095 1/3431 1/3374 —
ESP32-C3 (v0.3) 1693/4095 3/4095 3/2949 3/2852
ESP32-C6 (v0.0) 2160/4081 0/3313 0/3295 0/3283
ESP32-H2 (v0.2) 2160/4081 — — —
ESP32-S2 (v0.0) 2821/8191 — — —

I don't have S3 right now.

lure23 commented 5 months ago

My results with adc_cal example from esp-hal (d2a9389):

Board Chip None Basic Line Curve
ESP32-C3-DevKitC-02 "1.1" ESP32-C3 (v0.4) 1680..1689 / 4095 0 / 4095 0..1 / 3014 1..2 / 2906
ESP32-C3-DevKit-RUST-1 "1.2a 04/2022" ESP32-C3 (v0.4) 1680..1690 / 4095 0..18 / 4095 5..17 / 3040 6..14 / 2927
ESP32-C6-DevKitM-1 "1.0" ESP32-C6 (v0.0) 2176..2177 / 4081 0..1 / 3281..3297 0..1 / 3304..3321 0..1 / 3292..3309

Values "gnd / 3v3" like from the other people. I don't know what to make of this. Hope it helps.

The quoted strings are what's seen on the underside of the dev board PCB. DevKitC was controlled by uart, others by JTAG (don't think that matters).

jessebraham commented 5 months ago

Thanks everybody for the additional info. I will start digging into this next week most likely.

sourcebox commented 5 months ago

I did some more detailled measurements with my ESP32-S3-DevKitC-1. I connected a pot to GPIO3 and set some inbetween voltages. As you can see, all calibrated results below 0.3V are near 0, so there's a large gap, while the raw values are still changing. Above 3.0V, the curve calibration is completely broken.

Voltage None Basic Line Curve
0.0 1973 0 0 0
0.1 2095 0 0 0
0.2 2233 0 0 0
0.3 2348 93 83 83
0.4 2473 222 188 195
0.5 2582 331 275 285
0.6 2711 463 385 393
0.7 2857 614 512 518
0.8 2957 717 585 597
0.9 3079 829 689 694
1.0 3207 955 805 794
1.5 3870 1628 1349 1323
2.0 4095 2180 1830 1785
2.5 4095 2967 2465 2404
3.0 4095 3800 3172 64400
3.2 4095 4071 3395 65534

After doing these measurements, I did a quick check against the C example from the IDF and it shows that calibrated results start at about 25mV of input voltage. They are not very accurate, but they are not zeroed.