Marc's Public Blog - Arduino Hacking

This is a display driver for Framebuffer::GFX that displays on RGB Panels using https://github.com/mrcodetastic/ESP32-HUB75-MatrixPanel-DMA . It runs on top of the same code that https://github.com/marcmerlin/Framebuffer_GFX and therefore provides access to the usual trio of APIs: FastLED, Adafruit::GFX (via Adafruit::NeoMatrix compatibility using FastLED::NeoMatrix) and LEDMatrix compatibility too. To get example demo code, go to: https://github.com/marcmerlin/FastLED_NeoMatrix_SmartMatrix_LEDMatrix_GFX_Demos

Here is a video demo of a 24bpp framebuffer based set of demos that simply can't run on ESP32-HUB75-MatrixPanel-DMA without this framebuffer driver:

Installation

How to use?

Why not use mrcodetastic's library gives Adafruit::GFX compatibility?

What you get is this:

Low Level Drv|Glue Driver for FrameBuffer::GFX
FastLED     - FastLED_NeoMatrix  -------------\     FastLED CRGB Array 
ESP32-HUB75-MatrixPanel -,                    |
FastLED_ESP32-HUB75-MatrixPanel_FrameBuffer_GFX
SmartMatrix - SmartMatrix_GFX -----------------\    24bit FB storage        API Support
                                                \   CRGB methods like
AF:ILI9341-\                                     \  scale8/fadeToBlackBy
AF:SSD1331  |--- FastLED_ArduinoGFX_TFT ----------|        |               ,FastLED API
AF:ST7735  /     FastLED_SPITFT_GFX (for AF)      |        |              / (XY 2D to 1D mapping)
ArduinoGFX/  AF:Adafruit (ArduinoGFX is all in 1) |        |             

                                                  |        |            

ArduinoOnPc-FastLED-GFX-LEDMatrix arduino         - FrameBuffer::GFX------ Adafruit::NeoMatrix +
emulation for linux / Raspberry Pi:               |        |             \ Adafruit::GFX APIs
----------------------------------               /    Adafruit::GFX       \ 
rpi-rgb-led-matrix - FastLED_RPIRGBPanel_GFX ---/   LEDMatrix (optional)   `LEDMatrix API
ArduinoOnPC X11/linux - FastLED_TFTWrapper_GFX 

FastLED_SDL (linux)   -  FastLED_NeoMatrix   -

So you write against any of the 3 APIs in Framebuffer::GFX, and you get to run on all hardware backends without having to change your code, including switching from teensy to ESP32, or rPi to display on RGBPanels.

you can run your code natively on linux

I should stress that for RGBPanels, if you're writing code for ESP32 and ESP32-HUB75-MatrixPanel, you absolutely want to use this glue driver, because when you decide to scale up, you can switch to an rPi, get 3 parallel outputs for a display that is 3 times as large or refreshes 3 times faster, and all your code will just work with no changes required outside of the matrix init. So the same code can run on these bigger displays (those run with rpi-rgb-panel):

or on bigger arrays with triple channel using rpi-rgb-panel, no code changes needed

And one more reminder of how awesome it is to be able to compile your arduino code on linux, run it on linux with full display, and debug it on linux with gdb and ASAN to find your memory leaks and dangling pointers. You get this for free by using FrameBuffer::GFX for all your code.

Github: https://github.com/marcmerlin/FastLED_ESP32-HUB75-MatrixPanel_FrameBuffer_GFX

new board this year, radio communicator

People working hard:

and fun stuff

Met someone who made LED shoes, which predictably broke because they can't take the flexing:

I came decked out:

Random slides and talks:

SAOs are still fun things to play:

Evening parties:

A fun workshop was a 64x64 matrix connected to an ESP32 S3 with adafruit shield, I actually ended up spending more time on that because blinky :):

a new badge :)

Good times, looking forward to next year

eyes are drawn with math, they aren't sprites or animated gifs

So, I already built a 64x64 Matrix the hard way in 2018, including early uses of the ESP32 FastLED parallel output code that was still being written in 2018 when I built it. Building the matrix from scratch with 64 strips laid out one by one, was a pain, it took close to a week just to build. Code-wise, it took a little while, but I had a sweet running 110fps 16 parallel channel output setup, it was lovely.

professional wiring work, haha

yeah, that's why I wanted to use a nice expander board this time around

not counting that I had to add level shifters to get full 110fps speed from 3.3V output to 5V pixels

but eh, it did work and it survived 2 burning mans until the playa ate the pixels from the inside

I was honestly quite sad about my 4096 pixel array that took so much effort having been eaten by the corrosive playa, so when I saw pieces of pre-made matrices at a more reasonable price, I I kind of impulse bought 6 bunches 10x60 pre-made strips of much better quality just before the Trump tariffs came in. It was still $500 just in LEDs tough, but that's actually a good price for that many high quality pixels. I however figured I'd try using pixxelblaze with it because progress and not writing my own code for everything (although it was already written, haha). I also hoped to use the PB expander board to help with wiring.
I also was curious to try out the library of 2D patterns available with pixelblaze. In the end I found around 40 2D patterns that looked decent enough. Is 40 a lot? It's not bad, but when using my own Framebuffer::GFX in C++, I've easily gathered over 200 demos that are overall better due to more speed and obviously a lot faster (almost unlimited speed limited by the LEDs themselves).

I figured I'd live with the limitations of Pixelblaze and the limited amount of demos compared to C++ framebuffers, But things didn't really work out as planned. Namely:

Here are pictures of the build

all 6 sub matrices connected, turns out single power was good enough even if the matrix power wire was a bit thin and ran a bit hot

my 300W 12V power supply was definitely overkill, note the small step down converter to power the 5V PB from 12V

power was good

I tried to split the output in two by using a spare PB pico I had laying around

coordinate mapping was a huge pain due to lack of docs

with 2 devices, without magic in the code, a single PB would not know to display the left or right half

In the end, I gave up and went with a single 3600 pixel output, and make peace with patterns that ran as slow as 3 to 5fps:

I used a 110V power cord to re-inject 12V power in the middle, not fully required but nicer on wires

sadly my setup didn't come with the right plug to connect to the output and backfeed power from the other side, so I made my own from spare connectors

it worked without the power backfeed, but it was better with it

now came the job of connecting 60*5=300 knots between the sub-sections with twisty ties

didn't take too long, time for install

wee!

for a display that doesn't have a framebuffer and things are drawn with math, not bad

and it looks cool from inside the house too :)

Do you want the same demos without spending all the time it took me to download them one by one? Marc's Favorite Pixelblaze 2D demos pbb config you can directly install

The magic file above will install everything you need all at once, you'll just have to re-set Wifi, change the name and resolution.

If anyone is interested, here are the demos I settled on, the ones prefixed with '_' were downloaded from https://electromage.com/patterns :

Prototype of new version:

The previous v6 was an attempt at simplyfying v5 by integrating the batteries directly to the back of the panels. In the case of Lipo batteries that I cannot let run flat or they'll be damaged forever, and need for extra wiring for the balance connector, it was simpler to attach everything to the back of the panels, but it made the panels quite heavy and more bulky looking.
That outfit served me well, but outside of bulk and weight, one issue was getting the lipo batteries through airports. They are totally legal and allowed, but they had to be removed from the outfit every time I flew, which was a pain in unwiring/re-wiring, but even then, I would get stopped more than half the time I went through Xray while they got a supervisor to confirm that yes indeed the batteries were fine. A few times, I got less educated people who reacted in lesser ways, and called more people, a few times causing delays of 30mn or more. To be more specific:

The batteries are in green below, they showed up pretty reliably on Xrays and the whole thing would get flagged and then they would have to look at them and the wires, and often were not super happy:

So, I ended up making a removable battery pack that was wired correctly, and putting it the right way was tricky, so set it up to be correct and in one piece that could be removed and re-added with less work for each flight:

But it didn't take long to find out that even carrying the batteries separately that way still made TSA unhappy at times, and unless I unplugged absolutely everything and rebuild the whole thing every time, they would not be very happy and that was a pain for me to take apart and rebuild every single time. They still occasionally said it looked like C-4.

I wasn't being a contratrian for fun, and see how many world TSAs I could win an argument with, I just didn't really have many other options, I did need those big batteries because my outfit does need lots of power, and 5V USB battery packs simply do not output enough amps.

tried with USB packs

it was not fun, I had to segment power use in different busses plugged into different ports of battery packs, very cumbersome and the first tone to die could shut down everything

But after multiple years (and to be fair, it was many flights and most were fine, but the 5-10% that were not, were _not_ fun), technology finally evolved into whas that gave me better options: using less scary looking USB powerbanks that were finally able to output enough watts to power my panels without dipping power enough to cause reboots and crashes from the CPUs.

so I built a new system with dual 20V input (each one is sufficient on its own)

I have 2 USB-PD selector boards that output into a diode to avoid backfeed, and this goes into a DC-DC converter that takes the 20V down to 5V at up to 20A although probably only 5A max is really needed

It of course runs fine from a single battery:

At the same time, I made a big upgrade on the panel side by using flexible PCB panels that are differnetly thinner. At the same time I switched to RPI0 2W which is slightly more powerful than an RPI3a while being much smaller:

Now the whole thing looks nicer and lighter:

I did switch the big ESP32 to a smaller smaller ESP32 C3 (2 shown in this picture), which can still output to an LED strip if I really want/need. I also tried to switch to a smaller electrodragon passive-3 board that was custom designed to fit the RPI0 2W. It doesn't have the room for the level shifters, but turns out for an ABCDE panel, the passive board worked just as well (top on the picture vs the bigger red one in the middle):

The old setup had 300Wh of battery capacity, and that was real capacity, no BS. I measured it to give me a bit over 18H of runtime, which was very nice indeed, and helpful for places like tomorrowland where I had to camp and go 3 nights without ability to recharge.

This new setup differs a good amount because:

Same on youtube:

But of course, nothing can be simple, this new design brought some new issues:

After using it the first time at ASOT NL 2025 the wooden frame got detached in no time, failure of the 2.5mm screws

very early version without clips on both sides

just a bit of tape to hold things together, clips on sides and top

it was built until the flight and I didn't have time to add a protective frame

the panels were so cheaply made that the screws receptacles came unglued

Another issue I didn't like with these ABCDE panels is that the screws were 2.5mm instead of 3mm for unknown reasons and it was hard to attach to something that small.

Switching to a different 128x64 P2 Flex Panel vendor, but ending up with ABC panels that barely worked with rpi-rgb-panel

Sadly those ABC panels had different chips that required huge gpio slowdowns or suffered from noise and display issues:

new chips

It took a few weeks, and very kind help from board707 and multiple issues to make a new line addressing shift register pusher that was faster and didn't have the noise/corruption issue:

once I figured all that out, finally got the ABC panels to work almost as fast as ABCDE and now have a nice screet/pixel protector

USB powerbank failures

It took some tracking down to catch real time failures:

Flexible PCB solder point failures

flexing the PCB can cause any of those points to fail, causing the above

and they fail in different ways :(

Oh no, it looks terrible when it fails in the wrong place

Easier Replacing of Panels

This is not a great long term solution, but I have no better idea for now outside of going back to bigger and bulkier non flexible panels.

New Pants

looks more tidy

end result

New LED Hats

this was my old Dreamstate Hat, it looks cool but it's heavy

so I made new ones that are much lighter

funny thing is the very first time I flew with the new hats, I was detained because of them :)

Upgraded LED Shoes with bright Shoelaces

I got new brighter shoelaces but the batteries didnt even last a day

So I made a connector to plug into the shoe's bigger batteries I upgraded

End result:

Extending battery life

I have a "few" battery packs, different sizes and weights, but the weight adds up:

The good news is that I'm now able to use smaller packs that can still output 60Wh, but only up to 12V. The bigger packs can do 100W at 20V and output up to 75Wh on a good day, although most of mine often seem to do less (they are of course rated for 99Wh)

in the process, I found out the Wh totalizer in these meters, is garbage and can't be trusted

nicer/bigger 100W capable pack (more than I need) with better totalizer

new test at 15V instead of 20V (the red boards can be changed to ask any voltage) with diodes to prevent backfeed

measuring the loss between 15V input and 5V output

trying again with 12V, we can also note the voltage drop at 1.4A

some surprising results, some 75Wh packs emptied quicker than expected, only got 65Wh from thsi one

a fun part of this exercise is figuring out if the Wh totalizers are even good

Here are the results I got so far:

So, what comes out of this is:

The smaller packs seem to work almost as well for a much smaller size, their main downside is they don't charge nearly as fast (the 100W packs can charge in 1.5h at max speed given the proper power supply).

got a few extra colorful yard lights ;)

To debug some early issues I had with the pixelblaze, I soldered a few extra wires to add serial monitoring:

the serial port is temporarily going to a rPi3a which in turn makes it available over an ssh connection

up left if the pixelblaze pro expansion board that gives 8 channels, handles power distribution and converts the 12V to 5V for the pixelblaze itself

For the pixels, I picked rolls of 1000 pixels at 5cm and 10cm pitch from Ray Wu, trusted seller of blinky stuff, and one bonus of not doing this installation a few years earlier is that new pixels have been designed: WS2818.

The last bit "more efficient" actually worked to my advantage as I was able to make a string with 600 pixels and not have to re-inject power anywhere. I measured line voltage at the end and it was only 5V but the pixels still worked great at the reduced voltage.

The advice I got on the leds are awesome group was to try hot glue, and it was a good idea. Thankfully I had a battery powered hot glue gun, which was a must have. The other important bit was a solder reflow hot air gun, which did have to be plugged in the few times I had to unglue strips to move them after I found a better routing or visual pattern:

And in all, I actually need 3 tools, the 3rd one was a battery operated air can replacement which allowed me to blow cold air on the hot glue points and have them dry in 10 to 15 seconds instead of 1mn. That was a lifesaver:

After laying strips, house looked like this:

After the first 2 days of work, I had 4 strips, around 1500 LEDs:

or on youtube:

But it really got better once I added 2 more strips and upp'ed the count to 2000, which was a lot more visually pleasing:

or on youtube:

This is my current demo list:

These look interested but are too fast:

1) Oh, I need to finish writing code to get the SAO badge holder to do something fun

2) Mmmh, why does this python global variable thing doessn't work in the function

3) Goes to re-learn python, with help from gemini and how python forks global variables by default in functions so what you write to them isn't saved at local scope (oh my, why did they do that?)

4) after more hacking, get a proper demo working:

5) publish code https://github.com/marcmerlin/2024-Supercon-8-Add-On-Badge/blob/main/README.md (actual code is here: https://github.com/marcmerlin/2024-Supercon-8-Add-On-Badge/blob/main/software/software/main.py )

6) Oh, I need to make a video demo of the SAO, but...

7) oh yeah, there was that hat on my desk I needed to add LEDs on for an upcoming party, start hacking on that

8) mmmh, the hat won't work long enough with a USB battery, start hacking on a lipo solution

9) ah, lipos work, but the front LED panels rated for 5V start failing around 3.5V, so I need to make a dual lipo solution, one for the neopixels and one for the RGB panels

10) test and measure the dual lipo solution and figure out it will last close to 12H, that's good. Ok, hat works mostly, post that https://www.instagram.com/p/DCYEl1jJd_-
11) Ah yes, the SAO video I was supposed to work on: https://www.youtube.com/watch?v=z7MSywf7js4

12) During that time, I get interesting distracting packages on my doorstep, like lovely flexible LED panels that will hopefully power my future outfit: https://rpi-rgb-led-matrix.discourse.group/t/rpi0-2w-speed-compared-to-rpi3a/913/2

13) While I'm testing them, also try some new rPi0 2w chips and microcontrollers to see if they can power my LED outfit with a smaller form factor (and yes they can): https://rpi-rgb-led-matrix.discourse.group/t/rpi0-2w-speed-compared-to-rpi3a/913 . In the interest of briefty, I'm going to skip the part that included migrating from raspbian to a new dietpi linux distro and port back all my custom changes on it :)

14) That said, I had a long-ish discussion and planning with the electrodragon folks in China on how to make a new 3 channel daughter board that is much smaller and would decrease the over footprint of all new picture frames like the one I built, and of course my LED outfit.

15) but wait, now I also find out there is a new beaglebone solution that does 6 channels output instead of 3? Ok, make a post about it: https://rpi-rgb-led-matrix.discourse.group/t/interesting-6-port-board-beaglebone-pocketcroller-v2/918/2

16) and file a couple of RFE bugs to look into it: https://github.com/hzeller/rpi-rgb-led-matrix/issues/1733 + https://github.com/FalconChristmas/fpp/issues/2054

17) Oh, I just found this old video on my laptop, why did I never post it? Ok, fix that: https://marc.merlins.org/perso/linux/post_2021-01-23_Linux_conf_au-2021-LED-Talk.html

18) Mmmh, so when I quickly posted those pictures from Neil deGrasse Tyson who just gave a talk, I got a reminder it was annoying for my blog system not to be able to link to videos directly without having to upload them to youtube.

19) Ok, learn about html5 video tag, hack my blog system to support videos, and push a new blog with my first video tag: https://marc.merlins.org/perso/outings/post_2024-11-14_Neil-deGrasse-Tyson-back-at-San-Jose-Center-for-Performing-Arts_-An-Astrophysicist-Goes-To-The-Movies_-the-Sequel.html

20) ok, but why did I have to rescue my video from my phone because the H265 recompressed version did not work in the blog? Ah, because of royalties on H265 being too high, most browsers refuse to add support for it: http://marc.merlins.org/perso/linux/post_2024-11-17_Vidrename-Script_-Clean-up-Normalize-Video-Filenames-and-Add-Codec-And-Container-Format-With-Mplayer.html

21) Ok, sure, but now I need to upgrade my vidcomp script to support VP9 and AV1: http://marc.merlins.org/perso/linux/post_2024-11-17_Vidcomp-Script_-recompress-videos-on-linux-with-mencoder-or-ffmpeg-to-div3_-div4_-div5_-xvid_-mpeg4_-h264_-h265_-VP9_-and-AV1.html

22) then I find an Email from a chinese engineer who is willing to talk to me about making better LED shoes. That's exciting, let me reply to him first

23) ok, now I can recompress my videos to VP9, which I benchmarked to somehow be faster to compress and still smaller file size than AV1 (not supposed to be the case, but that's what I'm getting). On the plus side, VP9 works with more browsers than AV1.

24) what as I doing again? (now days have passed). Oh yes, I was supposed to write a blog on hackaday supercon: http://marc.merlins.org/perso/electronics/post_2024-11-01_Pasadena-Hackaday-Supercon-2024.html

25) of course the ridiculous amount of time spent to allow these 2 videos in that post, is a bit hard to justify, but it's really solving a problem I've had for over 10 years, and never got to the top of the pile, because, well, you see what happens when I try to put something on top of the pile :)

26) and I was good, I didn't even play with the micropython rpi micro I received in the mail this week, or the electromage pixelblaze and I/O board, or more LED panels that somehow showed up at my doorstep :)

27) mmmh, and there are all those webcams I ordered and need to configure, but now it's clear I'm going to have to re-install my camera controller, shinobi, from scratch, and that's yet another rabithole...

28) in the meantime I have to finish my LED hat, and design my new shirt and pants for Dreamstate in 4 days... Wish me luck :)