No, I have been trying to actually be helpful and informative for people trying to understand this article and video, given that referring to what was done here as a digital system is functionally meaningless and misleading. I am pretty sure the pedantry started with you talking about CPUs being electronic.

But at this point I believe your use of language in this discussion to be so absurdly reductive that I do not think this conversation is salvageable. What you appear to take issue with seems to keep changing and refining until the argument appears to be that any system that communicates information is apparently a digital communications system, so long as you can imagine an arbitrary scheme to interpret at least one bit of information from the signal, regardless of whether that was the message intended to be communicated. It would appear we have ended up with a “digital” storage system wherein a human observes the signal and decides one bit of whether or not it looks like a bird. Though I suppose “bird” was the only symbol in the protocol, so it may actually be a zero-bit system.

If the scheme successfully communicates zero bits of information, you might as well stop bothering to listen for the signal, and miss nothing of value. In fact I should probably do exactly that with this thread.

Yes, that would be a digital modulation. That is decidedly not what is being done with the bird. The input data is the “PNG” of the bird, which is then not digitally modulated, but converted to an analogue signal and later redigitized. If the file has been converted to a series of pulses at different frequencies (the equivalent of your black and white squares) that would be a digital modulation. I am not arguing that this is not possible. My original comment explicitly says I would like to see a follow up with actual modulation. But just because it is possible to run dialup over an analogue phone line does not mean that calling your grandma on that same phone is a digital communication system. Some computers back in the day could modulate and record data on commercial audio cassettes. That does not mean that if I record something off the radio and play it back later that’s a digital copy of the song.

I suppose you have caught me out slightly lacking in precision or pedantry. A digital to analogue modulation scheme is able to exactly reconstruct the original digital signal within the design tolerances for noise and distortion. Yes, eventually a signal may degrade or be corrupted, but prior to that point the reproduction is literally and exactly perfect. That exactitude is just about the definition of a digital system. This bird system is incapable of reproducing the input image of the bird exactly. It is not a digital communication system, unless you consider the “PNG” of the bird to have not been the message being carried.

It isn’t a digital channel because it does not reproduce digital data. Unless it’s a one-bit signal of “does this look like a bird? yes/no”, but then the human making that assessment is part of the channel. To claim this is a digital system would require us to be so reductive as to redefine the meaning of the word.

Yes, the near-identical sentences (only drawing a distinction between the processes where one exists) would indicate that. The “heard by the bird” and “reproduced by the bird” steps were also the same. But this is necessary context to make clear the digital data (“bit-stream”) that is being modulated into the signal.

It is far from “exactly the same”. The similarity is only in that both go through the same analogue channel. The entire point is that the modulated signal can be reconstructed exactly, while the spectrogram cannot.

The article title says they converted a PNG and the bird was able to “recall the file”, and yet it produced an indisputably different file. That it looks vaguely the same to the cursory human observer does not make it the same file.

Precisely… And digital modulation’s entire purpose is for a digital signal to survive those distortions bit-for-bit perfect. Even if we call the digitally-generated spectrogram digital information, the bird simply did not reproduce it exactly. Whatever time, frequency, and amplitude resolution you apply to the signal, if it’s low enough that the bird reproduced the signal exactly within that discretized scheme, then it simply did not achieve 2 MB/s. I would bet that the Shannon capacity of this bird is simply nowhere near 2 MB/s.

Exactly. Digital logic, when implemented in analogue, generally have to have forbidden zones where a signal in that range is considerer invalid. Regardless of implementation, digital is about the discretized logic of the system. That is explicitly the whole point of digital: Minor analogue distortion does not change the information content of the signal unless it is so bad as to flip a bit.

The whole sequence is:

• Digitally synthesized spectrogram (lossless)
• Played through a DAC and speaker to produce an analogue signal (lossy)
• Heard by the bird (analogue, lossy)
• Reproduced by the bird (analogue, lossy)
• Captured by an ADC as a digital audio signal (lossy)
• Spectrum-analysed to observe a similar (but corrupted) reproduction of the shape in the original spectrogram

To be transferring digital information, we would instead need to modulate and demodulate the digital signal (exactly like an old modem) so that the analogue corruption does not affect the digital signal:

• Image file (lossless)
• Bit stream (lossless)
• Analogue modulation of bit stream played through DAC (lossy)
• Heard by the bird (lossy)
• Reproduced by the bird (lossy)
• Demodulated to recover exact bit stream despite distortion (lossless again)
• Decode bit stream to recover original image file, bit-for-bit perfect

I extremely doubt that this bird is capable of 2MB/s. For reference that would make it 280+ times fast than dialup, and barely slower than ADSL. This setup is basically just using the bird instead of a telephone line.

Not to be a wet blanket, but every time this comes up I get annoyed by some factual inaccuracies in the articles about this. It is not digital! He drew an image on a computer, but converted it to an analogue spectrogram to store on the bird. That’s neat as hell, but it’s not digital. The image that he got back was slightly corrupted.

Now I would be fascinated to see a follow-up seeing if you can actually modulate a digital signal and have is survive a round trip through the bird bit-for-bit accurate. I suspect in reality it would be much lower data rate, but definitely not nothing!

I am familiar with these alternatives. My experiment was specific in wanting FIDO2 and I ended up figuring out the issue. It was the intersection of a couple of weird behaviours that made debugging very confusing, but it works exactly as I expected it would once those are resolved. I guess we can consider this a proof of concept that you can indeed use FIDO2 tokens as an external SSH host key (though as I said below whether this is practically useful is another matter entirely).

FYI I got it working, please see the update above.

Got the client working (mostly) without issues again, though trying to imitate my process for host keys as closely as possible I did encounter some weirdness that led me to this open bug: https://bugzilla.mindrot.org/show_bug.cgi?id=3355 So that may be the source of my issues… If they keys I was using suddenly were secretly requiring touch, it would explain a lot. I can’t right now but I will do another experiment with host keys when I can. Still would love to see if anyone else is able to reproduce this behaviour or get it working.

I previously have had no issues using this as a client key, but honestly I wouldn’t put it past myself to have picked the wrong key and not noticed. Maybe I will give it another try, but I don’t think that’s it.

Yeah, the rough idea is to use any old FIDO2 key as a USB HSM. Not necessarily looking for a very practical solution (the easy fix would be to just encrypt the drive), but curious. What inspired this, though not necessarily the final application, is Nix secret distribution tools that use the host key as the secret recipient. This means that theoretically if you have the host identity tied to an external HSM or similar you could have the same image deploy as different machines based on what security key you have plugged in.

I possibly should put greater emphasis on the fact that I am talking about the host key, not the client key. This guide is for a client identity key. I am talking about host keys as you find in /etc/ssh/ssh_host_ed25519, etc.