What am I looking at here? (UR22mkII distortion at 44.1 kHz)

[FF_CCSa1F]

Hello, I'm a power amp guy so you'll have to excuse my ignorance about these digital gremlins!

Just for fun, I decided to do a bit of contrast and compare between my various audio devices by hooking them up to my HP 339A distortion meter, and I stumbled upon something that baffles my analogue brain:

The only difference between those pictures is the sample rate sent to the UR22mkII (using WASAPI or ASIO through Foobar2000), and the difference is huge! -74.8 dB THD+N at 44.1 kHz vs. 86.4 dB at 192 kHz. To make matters worse, the 30 kHz low-pass filter was enabled on the 339A, so it's not even particularly high-frequency noise. I did the same test on a Yamaha AG03 which uses the same Cirrus Logic CS4270 CODEC, with very similar results. If I do the same test on the Xonar DX sound card installed in the same PC, it performs much better, landing at somewhere around -95 dB THD+N, so I don't think I have a problem with my testing set-up.

Can anyone shed some light on what I'm seeing here? The frequency of the distortion seems to be roughly 43 kHz by just counting the peaks between the 1 kHz cycles. That's not an exact science though, so it could be off by a few kHz. The 30 kHz LPF is down about 8 dB at 40 kHz, so there really needs to be quite a bit of noise up there to show up on the measurement.

...and here's a what the frequency is according to a DSO (16 sample average, ignore the bottom frequency counter, it tells only lies):

Thanks,
FF

 

[mansr]

Can anyone shed some light on what I'm seeing here? The frequency of the distortion seems to be roughly 43 kHz by just counting the peaks between the 1 kHz cycles.

You're seeing images of the 1 kHz tone at 44.1±1 kHz. Below is the spectrum of a 1 kHz tone played back on a UR242 (same CS4270 DAC chip) at 48 kHz sample rate. Note the prominent spikes at 1 kHz offsets around the first few multiples of the sample rate.

Increasing the sample rate to 192 kHz, all the images below that frequency go away:

Playing back white noise we see the full frequency response of the interpolation filter. Oh, dear.

Again, increasing the sample rate to 192 kHz moves the crap up along with it:

It is not a very good DAC.

 

[FF_CCSa1F]

You're seeing images of the 1 kHz tone at 44.1±1 kHz. Below is the spectrum of a 1 kHz tone played back on a UR242 (same CS4270 DAC chip) at 48 kHz sample rate.
It is not a very good DAC.

Thank you very much for this educational post!

I thought I was going crazy, but clearly not! I'm just testing a terrible interface!

 

[bennetng]

Could you explain the spikes at 192kHz and 384kHz? I've seen similar stuff in other reviews as well. It does not happen if it is simulated with software without actual analog conversion.

https://www.stereophile.com/content/okto-dac8-stereo-da-processor-measurements

See? The filter that goes through fs/2 does not have a high spike at near 70kHz. Again, not about audibility, but why?

 

[mansr]

Could you explain the spikes at 192kHz and 384kHz? I've seen similar stuff in other reviews as well. It does not happen if it is simulated with software without actual analog conversion.

I'm not sure what causes it, but it's quite common. Maybe it's the sample clock leaking through to the analogue output.

 

[bennetng]

I'm not sure what causes it, but it's quite common. Maybe it's the sample clock leaking through to the analogue output.

I recall you measured the UR242 and it showed a lot of jitter, so perhaps they are somehow related...

 

[mansr]

I recall you measured the UR242 and it showed a lot of jitter

That's putting it mildly:

 

[AnalogSteph]

Playing back white noise we see the full frequency response of the interpolation filter. Oh, dear.

[...]

It is not a very good DAC.

Interesting. I never realized that the CS4270 doesn't have the same filter characteristics as the CS4272, and features a stopband attenuation merely in the 50s of dB. And even then there's a sneaky "Note 10" accompanying that spec that says

Single and Double Speed Mode Measurement Bandwidth is from stopband to 3 Fs. Quad Speed Mode Measurement Bandwidth is from stopband to 1.34 Fs.

So these "ears" around 4fs in single speed mode and 2fs in quad speed mode are cleverly excluded.
Those seem to be quite typical for decimation filters, here's the filter response for a '90s CD player DAC:

Also, I never realized that the UR22 MkII uses the lowly CS4270 in this price class. Cheapskates. (Maybe solid build was deemed more important than ultimate dynamic range.) The Focusrites (2i2 and Solo) are using the substantially higher-spec CS4272, even Behringer does.

BTW, I hope you guys realize how ancient these codecs are by now. The CS4272 came out in early 2003, CS4270 some time in 2005. They still are performing quite well on the recording side though - 4271 and 4272 are pretty much the same as a CS5361 (save for a lower-performance single-ended input for the 4271) - honestly I'd take that over an AK5552, especially at 48k/96k. The DAC side looks like a CS4392. The CS4270 by contrast seems to have inherited its filters from the CS5341, while the DAC side looks more like a CS4391(A), both rather more pedestrian.

CS4271 and 72 seem to cost almost the same while CS4270 is little more than half the price, no wonder manufacturers are trying to save on BOM cost with this one.