What are the usual suspects with AD converter noise? How to improve converters?

[living sounds]

I´m also curious about A/D noise. I like to capture in 192khz and at that samplefreq it makes that increase of noise from 50khz above. It does not distrub me and it is not in lower samplereq, but I would like to know where it comes from. Maybe there is the lowcut filter not active anymore because it is not relevant at that high samplefreq?

It´s the same if the converter is Master or WC slave. It´s directly connected from the converter to RME AES-soundcard.

Noiseshaping. All modern DS-converters do it to a degree. It's a way to trade better SNR in the audible range (or even the more sensitive areas of hearing) against that above and also a function of the algorithmic processing to get useable performance out of 1-bit high-oversampling conversion. You can apply a lowcut filter, but that again will be a tradeoff.

 

[canidoit]

This is the noise I am referring to, but I get different results from different EQ programs:
In Logic EQ, the noise shows up on the graph but visually only the low sections, but I can hear the mid to high frequency noise. So I boosted it by 24+ gain in the EQ as shown in the example below to see the curve of the noise. This is why I think it is pink noise.


On Pro-Q3 eq the noise shape is opposite. Here it is, not gain boosted.


Here it is gain boosted on the Logic EQ with both graph results.


Would this sort of behaviour be typical for AD converters?

 

[abbey road d enfer]

I´m also curious about A/D noise. I like to capture in 192khz and at that samplefreq it makes that increase of noise from 50khz above. It does not distrub me and it is not in lower samplereq, but I would like to know where it comes from. Maybe there is the lowcut filter not active anymore because it is not relevant at that high samplefreq?

It´s the same if the converter is Master or WC slave. It´s directly connected from the converter to RME AES-soundcard.

Not sure if I fully understand your question, but many converters use noise-shaping, which consists in displacing part of the noise from the audible range to above. It results in extremely quiet conversion, but if you want to use the converter for low-level ultrasonics, it becomes almost unusable. Of course it happens only at dual (88.2/96k) or quad (192kHz) speed.

 

[JohnRoberts]

I am reluctant to dive in and reveal how much I don't know about high performance conversion. My experience consulting on a DSP platform a couple decades ago (after I left Peavey) revealed a number of ways that a codec convertor's noise floor could be compromised. From memory they were very sensitive to PCB layout (like how the sundry ground pins were interconnected, if that was easy there would only be one ground pin). I also recall some noise related to clocking signals.

I wouldn't expect any of these to change after a SKU is manufactured and in the field but don't know... I am not the convertor expert here.

JR

 

[ccaudle]

but I can hear the mid to high frequency noise.

That Pro-Q3 graph seems to indicate that the noise is around 110dB below full scale digital. That goes back to my earlier question about gain staging, what kind of gain path do you have that you can hear -110dB FS noise? If you have the monitor gain such that full scale is around 110 dB SPL (which I would consider very loud for domestic use, that is more like concert sound levels), the noise would be at 0dB SPL, essentially inaudible in any facility which was not designed with extreme sound isolation.
So if you can hear that noise floor, that would imply you have quite a lot of gain downstream from the converters. That is usually a backwards way to work, you want your gain in the early stages for best noise performance.

 

[Newmarket]

Not sure if I fully understand your question, but many converters use noise-shaping, which consists in displacing part of the noise from the audible range to above. It results in extremely quiet conversion, but if you want to use the converter for low-level ultrasonics, it becomes almost unusable. Of course it happens only at dual (88.2/96k) or quad (192kHz) speed.

Not sure that the sampling rate consideration holds there. If we are talking about the same thing - "Noise Shaping" is/was a thing with '48kHz' converter devices (that oversample internally)
I used AKM devices that did this but from a random search (see references to noise shaping)
https://www.ti.com/lit/ds/symlink/pcm1801.pdf?ts=1654373317097&ref_url=https%3A%2F%2Fwww.google.hu%2F

 

[Newmarket]

I wouldn't expect any of these to change after a SKU is manufactured and in the field but don't know... I am not the convertor expert here.

Yes - I'm unclear if the question is related to ADC "noise" in general or to an increase with a particular device over time.

 

[Newmarket]

They may do that but the improvements often are the results of wishful thinking. The only significant point of improvement, for a given D/A chip is in the opamp used in the I/V converter.

Yes - in most ("Never Say Never" !) cases changing passive components isn't going anywhere unless there are obvious deficiencies (eg on very inexpensive consumer DACs). There could be an "Audiophile" case for using Bulk Metal Foil resistors in place of Metal Film (anything serious now uses at least 1% Tolerance Metal Film) but I wouldn't really go down that route. And in reality you won't do much better than on-chip laser trimmed resistors. These give both tight absolute tolerance and - arguably more important - even tighter ratiometric tracking.
Ballpark figures - 10ppm per deg C absolute with 2ppm per deg C ratiometric.
Of course, in terms of "resistor noise" you are limited to the values that the manufacturer chooses but I don't think that's really an issue with devices targeted at audio applications.

 

[Newmarket]

Another possible culprit for adc noise is nyquist zone noise folding. I’m not familiar with audio specific adc design, but noise at higher frequencies (hundreds of kHz) can be downconverted in the analog front end of an ADC and show up in the audio band.

Yes. Although it's debatable if it is can really be classed as "noise" in that it is essentially a "spurious tone".
Not to trivialise it, but if you present an ADC with a frequency that is >= (half the sampling frequency) then the result will "fold back" into the bandwidth. Blame Mr Nyquist .
This applies to all sampling / data acquisition systems.
Hence the need to lpf signals. This gets more complex when you take into account Oversampling / Noise Shaping / Decimation etc.
But to bring it back to basics - it's aliasing and the same mechanism that makes wagon wheels sometimes appear to be going in reverse on "Westerns"