Delta-sigma vs “Multibit”: what’s the big deal?

[Thomas savage]

I'm happy to keep the thread open but let's concentrate as we seem to be doing on new facts and explorations rather than going over old ground.

 

[Bluespower]

You can get more than 16bit precision from dithered 16bit input (I think, somebody else chime in please).

With multibit dac the noise floor is constant wathever level. With delta sigma the noise floor is not constant and performs almost equally good with low level and high level signals. Usually noise floor is lower on delta sigma also. Do you think we can be sensitive to this variarion of the noise floor? (That is not present on the recording) . Maybe we cannot hear it?

 

[Blumlein 88]

You can get more than 16bit precision from dithered 16bit input (I think, somebody else chime in please).

Not sure if I've posted it here (I think so), I created a 24 bit file with -110 db tone. Saved it to undithered 16 bit and dithered. The tone was missing in the undithered file and present in the dithered file.

 

[andreasmaaan]

With multibit dac the noise floor is constant wathever level. With delta sigma the noise floor is not constant and performs almost equally good with low level and high level signals. Usually noise floor is lower on delta sigma also. Do you think we can be sensitive to this variarion of the noise floor? (That is not present on the recording) . Maybe we cannot hear it?

There's no chance we can hear it. If you normally listen at 83dB (moderate), and the noise floor in your listening room is 20dB at its lowest point in the audible spectrum, you won't hear noise lower than around -63dB (maybe a bit lower). Add to that X decibels for whatever your listening level is, and you'll soon see that if a DAC has a noise floor of -96dB or below, you'd have to listen at painfully loud levels to hope to hear it.

Then factor in the equal loudness curves and it gets even harder to hear noise at the very low and very high ends of the audio band.

 

[Blumlein 88]

With multibit dac the noise floor is constant wathever level. With delta sigma the noise floor is not constant and performs almost equally good with low level and high level signals. Usually noise floor is lower on delta sigma also. Do you think we can be sensitive to this variarion of the noise floor? (That is not present on the recording) . Maybe we cannot hear it?

Your first two sentences are incorrect. What you are describing is noise floor modulation. It may or may not be present in either type of DAC. If the modulated noise is high enough it could be heard. Usually it's not.

 

[Bluespower]

Not sure if I've posted it here (I think so), I created a 24 bit file with -110 db tone. Saved it to undithered 16 bit and dithered. The tone was missing in the undithered file and present in the dithered file.

It was present a -110db? That means that to decode properly dithered signal we need a dac that can deal with it. A multibit 16 bit dac would give bad level for this tone because cannot do better than 96 db. So what matters is to decode perfectly reverse way as it's encoded.

 

[RayDunzl]

With multibit dac the noise floor is constant wathever level. With delta sigma the noise floor is not constant and performs almost equally good with low level and high level signals. Usually noise floor is lower on delta sigma also. Do you think we can be sensitive to this variarion of the noise floor? (That is not present on the recording) . Maybe we cannot hear it?

The noise floor of the recorded material (assuming it is not purely electronically generated) will swamp (far exceed) the level of noise any modern DAC creates (IMHO).

 

[graz_lag]

Hoping not to be out of the OP ...

I copy/paste links from the TOTALDAC web page, the French' manufacturers of highly regarded R2R DACs, if not the highest one.
The page "Technical Principles" in particular contains some nice graphs ... hoping these might add some good point of discussion ...
Prices are from 5K€ thru 20K€ ...

http://www.totaldac.com/principles.htm

P.S. I do not have any personal interest in copying/pasting the above-links, but simply proud of Frenchie companies capable of exporting some good high technologies ...

 

[Bluespower]

Do adc used to record on studio or to master analog records are usually multibit or delta sigma?

 

[Blumlein 88]

It was present a -110db? That means that to decode properly dithered signal we need a dac that can deal with it. A multibit 16 bit dac would give bad level for this tone because cannot do better than 96 db. So what matters is to decode perfectly reverse way as it's encoded.

No with with dither a 16 bit DAC could do the same. That was the point of the comment earlier about dither. Dither reduces quantization noise and distortion. Plus allowing signals to be encoded in the noise floor below the smallest bit.

 

[Blumlein 88]

Do adc used to record on studio or to master analog records are usually multibit or delta sigma?

Usually delta sigma.

 

[andreasmaaan]

Do adc used to record on studio or to master analog records are usually multibit or delta sigma?

The chips used in the top of the line studio interfaces are almost always DS to my knowledge.

This list includes both budget and top of the line interfaces, but if you start googling all the ADC chips used you'll get an idea.

 

[RayDunzl]

Not sure if I've posted it here (I think so), I created a 24 bit file with -110 db tone. Saved it to undithered 16 bit and dithered. The tone was missing in the undithered file and present in the dithered file.

Quick example:

 

[March Audio]

I suspect Alan's DAC1 but I may be wrong about that as well.

Have a look at the linearity plots made by Amir of R2R and DS DAC's.

Was indeed my DAC1.

 

[garbulky]

Interestingly Mike Moffat was part of a team that developed a multibit ADC. At the time it was huge and very costly.

 

[Blumlein 88]

I thought I would post some more graphical information. First is a spectrogram of low level linearity for the March DAC1.
What you are seeing would go to a clear background if the level reaches -150 dbFS. Of course the visually noisy background is because the noise level is above -150 dbFs.

Each dash at 12 khz is a 10 second 12 khz tone. I've labeled the bit levels they represent in the test signal. You can see at least the 23rd bit of a possible 24 bits. Ignore the vertical bars as they are just markers for me when reviewing the results.

You aren't going to see that kind of result from a multi-bit DAC.
The 23rd and 24th bits varied around some. The values shown are averages of a few measurements. The 22nd bit varied though not very much. The others are enough above the base level noise not to be bothered.

 

[March Audio]

Ooooohhhh I like that

 

[Bluespower]

Hello happy new year all

My question of the year about the picture attached.
About the soundtype
- Do you know any technical reasons of these differences of naturallness of sound
- To akm the first filter wich is like non oversampling is the most natural
- This natural tone filter is at the same time the one with the worst measurments ans trebble roll-off
@andreasmaaan @Thomas savage @Blumlein 88 @March Audio @Veri

Then shall we conclude that this'is only Marketting?
It seams that for a specialised compagny like akm there exists advantages of non oversampling and that choice of filtering is a trade off between frequency response and naturallness of sound.
Does anybody has technical explications ?
Maybe we could ask them their reasons.

As don't like so much subjective tests with hears but i will tell you my experience as i find interesting to see if our hears agree that we measure. And i'm sure we can hear differences between dacs that measures at level where we shouldn't hear differences.
On my phone that has ak4962 i can test 4 of these akm filters via tinymix command and i actually can hear differences betwen them. Suprizly i don't like very much the 2 minum'phase filters (i was surprised as some compagnies that make expensive dacs like them) , and on linear phase both slow roll off and sharp'roll off are fine to my hears they are just different flavors with less trebles in slow roll off but maybe a little'more natural perception actually. Also i was surpised to find the sharp roll off brick wall filter was pleasant,
It seems lots of dac will have a filter like optimal transcient or nos mode in the future and compagny cares about that altghout it measures less good.

 

[Veri]

About the soundtype
- Do you know any technical reasons of these differences of naturallness of sound
- To akm the first filter wich is like non oversampling is the most natural

Short and simple ever since the "era" of digital audio, people have found treble roll-off to sound more "natural" and analogue. One reason is that a lot of audio has been mastered such to compensate for old tape/pre-CD audio treble roll-off. When the treble is NOT rolled-off in digital, super-clean audio (a new thing at the time) people complained about the "digital"/"harsh" sounding audio. Ever since then, people have still been obsessed to get "analogue" sounding audio like in the "good old time", either via tubes, roll-off/NOS (which also tends to have aliasing in the signal due to sub-optimal suppression of ultrasonics, which again has been shown to be a pleasing sound to some...)

It's probably more complicated than that.. but that's how I see the whole preference for a more "warm" signal. With properly mastered audio it's a non-issue. But we don't live in a perfect world, do we

Then shall we conclude that this'is only Marketting?
It seams that for a specialised compagny like akm there exists advantages of non oversampling and that choice of filtering is a trade off between frequency response and naturallness of sound.

As far as I know, all of their Slow filters are still using proper sampling techniques and are not really "non oversampling".
The whole "velvet sound" sound filter thing is entirely marketing imo, but it's easy to see why. They need to differentiate with the likes of Sabre, TI.

 

[Bluespower]

@Veri
The first filter claimed natural is'the super slow filter and is actually non oversampling.
Here is'a link that tests this fliter and show that it's nos,
https://www.diyaudio.com/forums/digital-source/298653-es9038pro-vs-ak4497-fir-filter.html