Mäh W.May 26

LG has made a thing called "Smart Diagnosis". On some devices you press a special button for a specific time and the device starts to emit a sequence of tones that sound a bit like a fax machine.

I am looking for
1) people with LG devices who are willing to record the sound and share it with me for further analysis
2) folks with a better understanding of digital signal processing who are interested to help out doing demodulation+decoding.

🧵 A little thread...

#dsp #hacking #audio #appliance

Show threadMäh W.May 26

I quickly stumbled across various YouTube videos when trying to help @koje71 diagnosing their broken washing machine. A very brief search also led me to the following git repository: https://github.com/kabelincho/LG-Smart-Diagnostics-modem -- a nice description, unfortunately no code and no sound snippet. M-FSK seems to be applied. Only one sound sample has been attached in the issue tracked. Unfortunately, that one does not match the description in README.md...

https://chaos.social/@koje71@darmstadt.social/114473587809022276

Show threadMäh W.
Fortunately, koje71@darmstadt.social gave me a recording which I've been trying to analyse. The top 4 peak frequencies here were indeed quite similar to the findings documented on GitHub: 1397 Hz, 1575 Hz, 1778 Hz and 2007 Hz.
May 26 at 8:58pmWeb
Show threadMäh W.May 26
Squaring the magnitude of the signals of the four subbands and low-pass filtering with a cutoff frequency of approx. 2*T_sym_estimated gave me this. Which IMHO doesn't look to bad for three subbands with the higher frequencies. The lowest subband (plotted in blue) doesn't look as good as the other ones. What could the reason be here? Distortion of the channel? Wrong parameters on my side? I don't know if normalizing the "energy envelopes" was a good idea ...
Show threadMäh W.May 26
... but it gave me another plot so that I could compare the subbands and tell if that frequency was active or inactive during the symbol. Zooming in gives an idea about the symbol time and what frequency/symbol is active when. Seems like somehting I would call a "one hot" encoding, maybe that's typical for M-FSK.. but this also means that only four different symbols are available an not 8 or 16 (I may be wrong). I guess from here on it's "only" ...
Show threadMäh W.May 26
... symbol timing recovery (I've played with it but basically I don't have a clue about it). Fortunately, I've got a recording and don't have to do the calculations "online". The rest should be about bit decisions and decoding... Any recommendations? Got you interested and you want to hop on?
Show threadMäh W.May 28
I've now added a first step before starting processing, which is sampling down the input signal to a sample rate of 8 kHz because that contains everything that we need and should speed up calculation dramatically.
Show threadMäh W.May 27
@funkylab ich hoffe es ist nicht unhöflich dich direkt zu fragen. Aber scheint das Signal für dich so plausibel insbesondere das "blaue Subband"? 🙃
Show threadkoje71 🙂May 26
@maehw 👍
Show threadkoje71 🙂May 26
@maehw du machst dir ja mehr Arbeit damit, als wenn ein Alien Signal aufgenommen wurde 😀