retro-go icon indicating copy to clipboard operation
retro-go copied to clipboard

Published 20 hours ago • verified publisher icon ducalex

rg_input_read_battery: noise to signal ratio

Open mtojek opened this issue 2 years ago • 14 comments

Hi,

Today I spent some time examining the battery status. I connect GPIO36 to GND via 10k resistor, and to VCC via 10k resistor. I measured the VCC voltage and it's ~4.787V. Unfortunately I can't get that value reported by retro-go:

STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:80.20 (4.80V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:80.80 (4.81V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:81.20 (4.81V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:87.60 (4.88V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:87.40 (4.87V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:87.80 (4.88V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:85.60 (4.86V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:97.80 (4.98V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:70.40 (4.70V) <--- !!!!
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:80.80 (4.81V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:88.00 (4.88V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:86.00 (4.86V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:88.20 (4.88V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:88.40 (4.88V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:87.00 (4.87V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:80.60 (4.81V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:100.00 (5.09V) <--- !!!!
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:88.20 (4.88V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:88.80 (4.89V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:80.80 (4.81V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:88.20 (4.88V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:79.60 (4.80V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:86.00 (4.86V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:66.40 (4.66V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:88.00 (4.88V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:87.60 (4.88V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:81.40 (4.81V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:86.40 (4.86V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:86.60 (4.87V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:88.20 (4.88V)
STACK:4892, HEAP:62+3483 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:87.80 (4.88V)

I tried:

  1. Put extra 104 capacitor and grounded it.
  2. Set the proper eFuse value (1177mv). BTW should this value be configurable via targets?
  3. Increased number of samples to 16.

Values are oscillating around 4.88V, but I'd like to achieve a higher precision.

Do you have any clues how can I improve readings?

Thanks for all suggestions!

mtojek avatar Nov 01 '22 20:11 mtojek

I think you're near the top of the range with that voltage. Maybe try a 1:3 ratio in your divider instead? Adding a rolling average could help with stability but probably not accuracy. Unfortunately the esp32's ADC is famously bad :( .

As for the vref in target, well it varies per chip so I don't think it's a device-specific thing.

ducalex avatar Nov 01 '22 20:11 ducalex

Well, with 3 resistors (1:3 ratio), things also don't look promissing:

STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:83.60 (4.84V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:84.80 (4.85V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:89.90 (4.90V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:84.20 (4.84V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:90.80 (4.91V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:86.90 (4.87V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:89.90 (4.90V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:100.00 (5.08V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:83.90 (4.84V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:90.50 (4.91V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:89.60 (4.90V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:90.20 (4.90V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:82.40 (4.82V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:100.00 (5.19V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:91.10 (4.91V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:93.80 (4.94V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:90.20 (4.90V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:94.70 (4.95V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:89.60 (4.90V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:91.70 (4.92V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:82.40 (4.82V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:90.80 (4.91V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:89.90 (4.90V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:89.00 (4.89V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:91.10 (4.91V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:81.80 (4.82V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:91.70 (4.92V)
STACK:4892, HEAP:63+3635 (48+3392), BUSY:0.00, FPS:0.00 (SKIP:0, PART:0, FULL:0), BATT:90.20 (4.90V)

Do you thing that this is an outcome of bad calibration or I should add some filters.

BTW is there any default/official/recommended schematics of the console? It might be a good idea to note some tricks for different platforms.

mtojek avatar Nov 01 '22 21:11 mtojek

Odroid-Go official document: https://wiki.odroid.com/odroid_go/odroid_go

KD-MM2 avatar Nov 01 '22 21:11 KD-MM2

Hi,

Today I spent some time examining the battery status. I connect GPIO36 to GND via 10k resistor, and to VCC via 10k resistor. I measured the VCC voltage and it's ~4.787V. Unfortunately I can't get that value reported by retro-go:

2x10k is too low. It'll drain your battery too. Try 2x100k or maybe 2x470k and a capacitor on the input pin. I use 2x100k like on the original odroid go and it also isn't total stable, but good enough.

mikeinkl1 avatar Nov 02 '22 01:11 mikeinkl1

We could consider adding back a rolling average, or at least re-enable the old averaging (the two commented lines in rg_input_read_battery) but 100mV fluctuation is quite a lot even for the esp32 so there are likely electrical improvements to be made too.

ducalex avatar Nov 02 '22 18:11 ducalex

Yes, main issue is probably the hardware, and this may be related the screen flicker and audio noise as well. Rolling average will help, but finding out why the launcher screen flickers but not the menu screen would be better. Does the launcher read the sd card on every cycle perhaps? If so, can that be changed?

mikeinkl1 avatar Nov 02 '22 19:11 mikeinkl1

Does the launcher read the sd card on every cycle perhaps?

It doesn't. There used to be a background task but it was disabled when I added wifi. There is currento no sd card access at all when idle.

Wifi could be the cause (if enabled). When transmitting it can easily use over 150mA for a few ms, this is a frequent cause of instability in esp32 projects.

ducalex avatar Nov 03 '22 01:11 ducalex

Yes.. I just unplugged the sd card and the flicker was still there. I dont think it is the wifi either as the flicker already occurred in the previous releases. There must be something else that gets called in the launcher loop, but not when the menu is open.. next suspect would be the screen? Is this reset or redrawn on every cycle? Maybe time to buy an oscilloscope and see whats really going on.

mikeinkl1 avatar Nov 03 '22 04:11 mikeinkl1

The screen is redrawn 10x per second when idle. That's the only thing that's happening when the launcher is idle in the current version.

You can tune the redrawing by editing launcher/main.c:

        else if (rg_system_timer() >= next_idle_event)
        {
            gui.idle_counter++;
            gui.joystick = 0;
            prev_joystick = 0;
            gui_event(TAB_IDLE, tab);
            next_idle_event = rg_system_timer() + 100000; <---- or you can increase that number for even less fps
            redraw_pending = true; <------------ comment this line
        }

But frankly the emulators do a lot more redrawing so it would be weird if that were the issue.

ducalex avatar Nov 03 '22 18:11 ducalex

yes thanks, that fixed the issue.

Edit: the little preview images now don't show, wifi and battery icons are not updated, so i would not call it a fix. I did some digging and the cultprit is really rg_gui_flush();. And yes, the noise is also present in the emulators, but normally not audible, only if the amp is all the way open. It is worse with full refresh. The flicker probably isn't visible at 30Hz or so, but at 10Hz it is quite noticeable. I changed the GUI update to 1 second, which makes the flicker barely noticeable. Maybe it can be improved by doing a partial refresh also for the rg_gui, but really it is a hardware/power issue.

mikeinkl1 avatar Nov 04 '22 09:11 mikeinkl1

I added averaging to retro-go a while ago that might help + the electrical changes you have made.

Is this issue still relevant?

ducalex avatar Dec 14 '22 21:12 ducalex

I tried compiling 1.36.3 but it hangs here: /home/mike/retrogo/retro-go-1.36.3/components/retro-go/rg_storage.c:338:37: error: 'struct stat' has no member named 'st_mtim'; did you mean 'st_mtime'? result->mtime = statbuf.st_mtim.tv_sec; ^~~~~~~ st_mtime

mikeinkl1 avatar Dec 16 '22 01:12 mikeinkl1

The stat and dirent structs are so annoying, it's always a pita to make portable use of it.

It's now fixed in dev branch but you can make the change on your local branch: https://github.com/ducalex/retro-go/commit/cfa842fa65efe2a8b17aa5b1c46cf9a2b66deb1a

ducalex avatar Dec 16 '22 01:12 ducalex

Thanks.. it compiles now but the noise is pretty much the same. I reduced the refresh rate to 1Hz which makes it hardly noticeable.

mikeinkl1 avatar Dec 16 '22 02:12 mikeinkl1

I've added thresholds to reach before the UI/console is updated, this eliminates the noise issue (well, it hides it).

In other words:

  • the percent value will only be updated when the difference is at least 1%
  • the voltage value will only be updated when the difference is at least 10mV

This fixes the flicker in the battery icon, it fixes the jumps in the serial console, etc. Note that battery is still being read every second so any large change will reflected immediately, so we're not losing anything.

(First implemented in f6b8df6de1cb6135b8c2d5514114046b07ca6330 which is now merged to master, made configurable in 0e237db07a8708576730c2faf6705297a139eb49 which is not yet merged to master).

In my eyes this solve the issue as best as we possibly can, so I'm closing this issue!

There were a few tangential issues in this thread. If they are still relevant, please open a separate issue :)

ducalex avatar Feb 28 '24 23:02 ducalex