Interface Mystery

[Joakim]

The ADC is beyond dreadful. It looks like it's running at half rate (22.05 kHz) with the worst anti-aliasing filter ever, possibly none, and then upsampled to 44.1 kHz, again with poor or no filter.

I guess it has nothing to do with the ADC at all, just buggy firmware or a malfunctioned/defective unit.

I've learnt since yesterday that it's a 44.1 kHz 24-bit Cirrus Logic ADC. I'm fairly sure the problem isn't with the ADC itself. It almost seems like the device is missing a stage or component that's in place for every other recorder and interface. I haven't been able to find another example of a device doing this.

 

[Blumlein 88]

Thanks for the replies! I really appreciate everyone's take on this.

It is starting to seem like the device is doing something unusual. You will always have (mostly subtle) differences between sample rates and implementations of anti-aliasing filters, but this seems different as it's aliasing like crazy without any frequency content anywhere near Nyquist. To sum it up as succinctly as I can:

View attachment 62307

In that direction folded down opposite angle as your actual sweep that will be an image. You get mirror images around each multiple of the nyquist frequency of your DAC. This is one of those, but an unusually strong one. Could be from 2x or 4x or more the base bandwidth. The way this one looks, and that it is sloped at the same rate as your main sweep I say it is from the 24 khz-48 khz range. Yet the white noise profile didn't indicate this would be this strong. Maybe the filter has some strange behavior so it filters less well at strong signal levels though I can't imagine how that would happen or work.

 

[mansr]

I've learnt since yesterday that it's a 44.1 kHz 24-bit Cirrus Logic ADC. I'm fairly sure the problem isn't with the ADC itself. It almost seems like the device is missing a stage or component that's in place for every other recorder and interface. I haven't been able to find another example of a device doing this.

I'm starting to think broken software is the most likely explanation. For example, swapping samples pairwise (1, 0, 3, 2 order) has more or less the effect we see.

Could you make two recordings, of a tone at 5 kHz and the other at 15 kHz, and post the files? Use the RME to play the tones.

 

[Joakim]

I've seen similar things but not as bad as yours:
https://hydrogenaud.io/index.php?topic=104547.msg858925#msg858925

That's really interesting. This is the first time I've seen anything like what I'm seeing. Does it do the same thing in the audible spectrum?

So, what recorder is it? Tell us please.

No damn way. In Assembly?! This must be a very old device, or very limited in terms of UI or something.. I really wanna see what this is lol. I can't imagine anyone actually programming in assembly in the modern day anything substantial outside of calculators or something like recorders using non-LCD screens from 2 decades ago or something..

It's a device released in 2019, selling for $1,500+ and with a small but loyal following around the world. It also has a really nice touch screen.

I'm wary of saying what it is before I've figured out what's going on. For all I know, this is a problem unique to my device and I don't want to cast aspersions that could end up hurting someone's business. It is a unique thing for sure, with many great qualities aside from the issues in question. The maker also has a flawless history of fixing problems in response to customer feedback so I want to give them the chance to do that if at all possible. Once this gets resolved, or it turns out it can't be, I'll reveal what it is.

 

[Joakim]

You get mirror images around each multiple of the nyquist frequency of your DAC. This is one of those, but an unusually strong one.

I think this has to be it. I'm trying to find out how filters are implemented to deal with this but most information about images is found in technical papers and the maths has so far been beyond me. The fact that the signal and its mirror are filtered out around 22K makes me think there is an anti-alising filter in there, just like the manufacturer says, but that another filter might be missing. Other recording devices seem to be removing the mirror images around multiples of Nyquist just fine.

 

[Joakim]

Could you make two recordings, of a tone at 5 kHz and the other at 15 kHz, and post the files? Use the RME to play the tones.

Absolutely. The behavior is completely predictable:

• A 5 kHz sine tone will be recorded as two tones. One at 5 kHz and another at 22.05 kHz - 5 kHz = 17.05 kHz
• A 15 kHz sine tone will also be recorded as two tones. One at 15 kHz and another at 22.05 kHz - 5 kHz = 7.05 kHz

It's the same as for any other sound. Recorded tones and harmonics have a more or less equally loud copy of themselves at Nyquist minus their own frequency.

 

[mansr]

Please post the recordings, not screenshots of spectrograms.

 

[Joakim]

Right, let's see if this works:

 

[mansr]

Hah! The samples in the files are pairwise equal. That explains everything. Probably a bug in that fancy assembly code (that's what happens when you code in assembly for no good reason), unless they're stupid enough to do this on purpose.

Can the device record in stereo?

 

[bennetng]

It's a device released in 2019, selling for $1,500+ and with a small but loyal following around the world. It also has a really nice touch screen.

So likely not from popular brands like Tascam or Zoom, I am relieved, thanks.

I've learnt since yesterday that it's a 44.1 kHz 24-bit Cirrus Logic ADC.

I believe even in the old Wolfson and Crystal days they have ADCs with higher sample rates. If it is the recorder company's decision to only support 44.1k then it sounds ridiculous, especially for a > $1500 recorder.

 

[Joakim]

Hah! The samples in the files are pairwise equal. That explains everything. Probably a bug in that fancy assembly code (that's what happens when you code in assembly for no good reason), unless they're stupid enough to do this on purpose.

Can the device record in stereo?

Indeed it can; that's one of its selling points. I don't understand what the samples being pairwise equal means but this sounds like good news! Would you mind explaining a bit more?

 

[Joakim]

If it is the recorder company's decision to only support 44.1k then it sounds ridiculous, especially for a > $1500 recorder.

It definitely has its quirks. I promise to say what it is as soon as this gets solved, one way or the other.

 

[mansr]

Indeed it can; that's one of its selling points. I don't understand what the samples being pairwise equal means but this sounds like good news! Would you mind explaining a bit more?

The first 10 samples in one of the files have these values:
11918
11918
-1399
-1399
-10770
-10770
10243
10243
2359
2359

Notice the pattern?

 

[mansr]

Could you do a stereo recording? The ADC chip almost certainly outputs stereo (the I2S interface demands this), and maybe something is broken in the conversion to mono.

 

[Joakim]

Notice the pattern?

I do, and it looks very similar to what you see in the third image from my original post. There are definitely 44,100 sample points per second but they come in identical pairs. Ok, one stereo recording coming up!

 

[Joakim]

Actually, two stereo recordings. One is of a sine sweep from 1-20,000 Hz and the other goes up to 80,000 Hz.

(Made a mistake and had to re-upload the files)

 

[mansr]

Actually, two stereo recordings. One is of a sine sweep from 1-20,000 Hz and the other goes up to 80,000 Hz.

(Made a mistake and had to re-upload the files)

Same deal there, each sample value repeated twice. I was thinking that maybe the pattern was different for stereo.

Have you reported the problem to the manufacturer?

 

[Joakim]

Yup, I just downloaded Audition to make use of its 'Export Data' function and found the same thing. I've sent the information to the manufacturer and am waiting to hear back. I'm hopeful it's a software bug that can be fixed and grateful to you and everyone else who has pitched in. I'll report back as soon as there's an update!

 

[DonH56]

Is it a "NOS, filterless" DAC? That is the crowd that eschews things like output (DAC) image filters or input (ADC) antialiasing filters as "messing up the sound".

 

[Joakim]

That was my thought too at first but apparently not. I'm not sure about image filters but there is an anti-aliasing filter in place. The manufacturer told me he had tweaked the cutoff frequency of it.