Graham Sutherland / Polynomial
planning a project to replace the hexlight I take to EMF.
each line (edge) is an LED bar, with its own PCB. each bar is a single dimmable unit.
each point where lines meet (vertex) is a connector block.
I want to plug in a single data input and transport it across the assembly. the challenge: no fixed positions, no configuring jumpers. any piece can go anywhere during assembly. I'll pixel map it after.
the main challenge is doing that given the multiple paths. I wonder if RS485 would work.
Ryan Castellucci (they/them) :nonbinary_flag:
@azonenberg @gsuberland How much stuff do you have that will actually take 48V? I've found 12V to be much more common.
Andrew Zonenberg
@gsuberland fair, i have 48V supplies all around my lab for various bigger projects but I'm planning to use it for rack-level DC distribution to all my switches etc long term
Graham Sutherland / Polynomial
@azonenberg also with 3 or 4 strings I can phase shift the PWM per string to keep the ripple current down.
Graham Sutherland / Polynomial
@azonenberg I already have the 24V supply and getting 3.3V for the MCU starts to be more annoying at 48V.
Andrew Zonenberg
@gsuberland Why 24 not 48? For this kind of thing I'd want to go as high as I can and still be SELV
Graham Sutherland / Polynomial
further thoughts:
- 24V supply rather than 12V
- either 3 strings of 7 LEDs or 4 strings of 6 LEDs depending on Vf specs
- each string PWM'd with a small MOSFET, keeping input capacitance and gate threshold low
- probably STM32F103C8 at 24MHz
- average current sensing feedback through shunt resistor with opamp and RC low pass
that particular STM32 model is cheap due to huge volume production and it's used in Blue Pill so plenty of libraries around.
Graham Sutherland / Polynomial
the fun part is going to be making the bar design cost effective. ideally I'll be able to go with a 4L 1oz/1oz stackup and double sided assembly - LEDs on the front, control circuitry on the back. we shall see.
Graham Sutherland / Polynomial
the current draw through the driver went up but it's still very manageable. it's about 300mApk, average much lower. wouldn't necessarily want to source it from an LDO hanging off a 12V line, but that's ok.
Graham Sutherland / Polynomial
I also updated the hexlight model to include C_L=30pF and a 5kΩ to ground at each bar to simulate worst-case loading on the RS422 bus. the graph came out looking essentially identical.
Graham Sutherland / Polynomial
one thing I will almost certainly do on the real circuit is put a common mode filter on the RS422 lines, probably on each bar. it's a roughly 2x2m installation with a bunch of loops so it's basically a funky antenna.
the installation location is typically on or nearby a bunch of metal shipping containers so that offers some natural shielding against induced currents, but EMF Camp also has Tesla coils around Nullsector so, y'know, pays to be careful.
(and TVS diodes are a given anyway)
Graham Sutherland / Polynomial
a bit more tuning. added dead-time generation to the push-pull drivers to minimise shoot-through current.
Graham Sutherland / Polynomial
that extra noise drops significantly if I turn the output into an RLC low-pass (10ohm + 1uH + 33nF). so I think that's a solid option.
Graham Sutherland / Polynomial
ran some simulations with a slew-rate limited push-pull driver and the hexagon light's vertices modelled as LTRAs. looks perfectly cromulent.
Graham Sutherland / Polynomial
115200 baud over RS485 would get me 120fps and 10-bit dimming control, which is plenty. with slew rate control I can tame the edges down so the frequency content of the data line doesn't exceed about 500kHz, which should be plenty low enough to keep things damped despite the multi-path signal and fairly high parasitic inductance.
Graham Sutherland / Polynomial
each bar would have a 6-bit address configured but that's fine since it's a one-off task, not a per-setup task. pixel mapping is just an interactive thing in the controller - show one segment, locate it, move on. easy.
Graham Sutherland / Polynomial
my thought is that the RS485 data would just get passed through everywhere, multiple paths be damned.
I could stomach having one single "special" connector block where I have termination enabled, since that's easy enough to place on the opposing side, so I do at least have the thing terminated. otherwise I think it'd just ring like crazy.
Ryan Castellucci (they/them) :nonbinary_flag:
@uint8_t @azonenberg @gsuberland Ah. I just use thick wire, lol.
Z̈oé
@ryanc @azonenberg @gsuberland my idea is to run a 48-V distribution and convert it to 12/20/5 V locally, do I don’t need to use too think wiring
GNU social JP is a social network, courtesy of GNU social JP管理人. It runs on GNU social, version 2.0.2-dev, available under the GNU Affero General Public License.
All GNU social JP content and data are available under the Creative Commons Attribution 3.0 license.