Today we got our dishwasher fixed under warranty. To diagnose it, the tech stuck a photosensor+magnet onto the front panel over the green status LED and started reading diagnostics. It turns out the status LED is actually a serial port*, and it's continually transmitting status.
I can think of so many gadgets that could and should use this trick...
*It's probably something other than regular serial -- a UART's TX would flicker. I don't know any details.
Effectively, this is an opto-isolator. Except it's not all in one device. Wikipedia has a good article on them, and notes that they can manage 10s of KHz. (See MIDI.)
The nice thing is that there's zero extra hardware -- every gadget already has an LED or seven, and if your MCU has a spare DMA channel, you can stream a bitstream to it (however it's encoded) with low overhead. If it's a short status message, you can just let the DMA loop over it and there's no MCU overhead.
@kbob sounds close to the smart gas meters that are apparently the sole reason you can buy a USB IrDA interface in 2025:
https://en.m.wikipedia.org/wiki/IrDA#Power_meters
Lacking specialized electronics, many power meter implementations utilize a bit-banged SIR phy, running at 9600 BAUD using a minimum-width pulse (i.e. 3/16 of a 115.2KBAUD pulse) to save energy. To drive the LED, a computer-controlled pin is turned on and off at the right time. Cross-talk from the LED to the receiving PIN diode is extreme, so the protocol is half-duplex.
i've got two of them on my desk. they attach to the meters with magnets and run on a Silicon Labs CP210x chip.
i may not be able to get them talking to my PalmOS devices, because the full IrDA stack is a Whole Thing, but maybe…
@kbob DMA channel?
If it's broken you can bitbang a serial stream without issue. Add a IR receiver and you can have bidirectional serial with not much issue. Most LEDs are easily able to do serial at 9k6 and probably at 115k2, which is trivially fast enough (even 9k6) with a decent protocol to transfer diagnostics data (other than full log files).
Note that a LED flickering with at least 1 start bit per 10 bits, at 9k6, would be flickering on and off at least 1000 times per second, so you wouldn't necessarily even notice that it flickers.
@dascandy The reason I suggested a DMA channel is that bit banging keeps the MCU busy. The MCU presumably has other things to do, otherwise why is it there?
Very low end MCUs don't have DMA, but it's a common feature these days.
And yes, if you add hardware you can add capabilities. I was talking about a zero hardware feature, which is what this appears to be.
@kbob Cool! Similar to IEC 62056, I guess...
Out of interest, what make is the dishwasher?
@LovesTha What?
The dishwasher has a green LED on the steel front panel. The tech put a magnetic device onto it that covered the LED and had a cable running to his computer. He said it was a photo sensor and was reading diagnostic data from the LED. He didn't do anything to start data transmission. I extrapolated that the dishwasher must always be transmitting diagnostic info by modulating the LED output. At least that's what I think was happening.
@kbob oh, I communicated badly.
How and technical things were immediately obvious.
That it was 2025 and I'm 42 before I learned that such a thing was real is my in credulousness.
This has amazed me with child like wonder
@kbob that’s a clever trick, and way less disruptive than most self diagnostics systems.
At a guess it’s encoding data in the difference between, eg, 80% on / 20% off duty cycle and 90% on / 10% off duty cycle. Providing the PWM is fast enough and the “high” and “low” duty cycles are reasonably balanced I’d expect it just looks like a LED around 85% on by (fast) PWM. And just needs good oversampling on the reader to decode.
@jaseg if the bits were sent slowly then I’d agree there could be visible flicker. But if they’re sent rapidly enough with a good (fast) PWM controller so it wasn’t just one on/off per cycle, then the flickering would only be visible to a (fast) camera. (Eg send the duty cycle 10 times per bit, rather than once per bit.)
Whitening (for ~equal 1s and 0s) is a good idea anyway, and could probably be combined with some forward error correction.
CC @kbob
Which makes me wonder if it only transmits when the magnetic device is attached.
The LED would be a steady green (or whatever) under normal use, but when a magnet is sensed, data is transmitted. Since the LED is covered, you wouldn't see any flickering.
@danderson @kbob our smoke detectors can output their serial number, number of alarms, battery state etc. by modulating their piezzo. A smartphone can be used to do the "mandatory" self test once a year and see the history per smoke detector. Its a neat system. They somehow measure if its day or night and sound the low battery alarm only at day - well worth the premium price. I already lost a few nights of sleep hunting the stupid beeping thing that woke me up at 2 AM.
Edit: brand is Hekatron ; so far they had no false alarms (vs. the cheap ones) but detected accidental kitchen oopsies reliably.
@kbob @phooky the best PDF i found about this from miele was: https://www.tie.ro/wp-content/uploads/2013/05/Smart_electronics_in_home_appliances_-_MIELE_-_TIE2013.pdf#page=24
The topology seems that behind the "optical interface" we could also do TX
(Also cool to see they use Super-H for Touch, Page 13)
@phooky still better than a Electrolux/IKEA oven which has knobs -but they tend to not work or cancel a timer when there is a bit of steam buildup and the resistance goes haywire :-)
(instead of using dedicated pins/traces)
@kbob
Very clever and very hostile to user repairs and probably third party repairs - at least a visible error code can be looked up
I absolutely hate that implementation
@kbob while we don't ship it in the production firmware, the Oxide server's service processor firmware has a build option that transmits kernel profiling data encoded NRZI over the status LED. It's a handy trick.
A lot of MCUs even have hardware left over from IrDA that you can use to frequency modulate a UART output!
I've been trying to figure out how to read this sort of a channel from a phone, without additional hardware... that part's hard so far, at reasonable bitrates. LG appliances can emit encoded audio chirps that a phone can pick up, maybe that's the way. (I was unable to convince my colleagues to include an audio output device on the server.)
@cliffle Is that the Hubris kernel?
I'd think it would be hard for a phone to isolate an audio chirp in a datacenter. So much fan noise...
@kbob the more of our racks you use, the less of a problem that fan noise will become. 😉
(But seriously we're dramatically quieter)
(...omg I should try to modulate the data _onto the fans_)
I came very close to implementing that at my last job: https://honk.bewilderbeest.net/u/zev/h/HW2GcrgzthGts7DXR8
@shironeko @kbob it depends on the phone, but, most don't have a physical shutter and use CCD-level "shutters," usually rolling rather than global.
A lot of the CCD controllers can do absurdly high frame rates if you window the sensor narrowly to e.g. the center four pixels. But I don't think mobile OSes tend to expose that.
@shironeko @kbob So, delightfully, I did an optical illusion with this many years ago, which would draw ghostly matrix images in photos taken with rolling shutter digital cameras.
I actually hadn't considered using this for data reception though. That's great. You'd need time sync patterns of some sort.
Kernel Bob 
Hanneke
post-Gundam cryptography
Peter Bindels
Nicolás Alvarez
mal3aby
LovesTha🥧
rawe
Samuel ✅ moved to raphus.socia
Markus Redeker 🐘
Andrew
Daniel Carosone
Ewen McNeill
jaseg
John M. Gamble
Dave Anderson
Klaus Vink Slott
notsoloud
TunafishTiger
grapefrukt
Daniel O'Connor
Jim Spath
StarkRG
sabik
The Penguin of Evil
Chuck
Chip
Workshopshed
Stefan Baur 8 * 💉
phooky
TAL
Oli
JohnB
cliffle
zev
shironeko