• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

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

THW

Member
Patreon Donor
Joined
Nov 21, 2018
Messages
70
Likes
62
#1
not an expert in engineering, but from what I’m gathering from others outside here, “multibit” is supposed to be better because it “preserves the original samples”... but if said “multibit” design distorts the signal, wouldn’t that mean that it also isn’t really good at preserving the original signal in the first place?

and delta-sigma is supposedly inferior because it’s a simulation... but then what’s so bad about a very well executed simulation?

based on this, wouldn’t the implementation of both design types and the design as a whole matter more than whether or not one is “delta-sigma” or “multibit”?

just a horse sense take on this, lol
 

SIY

Major Contributor
Technical Expert
Joined
Apr 6, 2018
Messages
2,216
Likes
3,581
Location
Phoenix, AZ
#2
not an expert in engineering, but from what I’m gathering from others outside here, “multibit” is supposed to be better because it “preserves the original samples”...
Congratulations, you have just discovered that the vast majority of things said on audio forums is total ignorant nonsense. :cool:
 

THW

Member
Patreon Donor
Joined
Nov 21, 2018
Messages
70
Likes
62
#3
Congratulations, you have just discovered that the vast majority of things said on audio forums is total ignorant nonsense. :cool:
Honestly I don’t know much lol, I’m just a normal person trying to figure out what is actually good for me

But even as someone who doesn’t know much, more and more things simply don’t add up about this subject the more I think about it
 

SIY

Major Contributor
Technical Expert
Joined
Apr 6, 2018
Messages
2,216
Likes
3,581
Location
Phoenix, AZ
#4
Don't worry, most people who make confident assertions about technical issues in audio don't know much either. Dunning-Kruger is the rule, not the exception. Hand-waving is the usual style of audio "expertise," so ignore anything that is not supported by cold, hard data.

Look at the DAC reviews on this site (where things are actually measured!) and notice which units have the highest performance. The ONLY thing that counts is what comes out of the analog outputs- the nature of the black box to get that result is far less important.
 

THW

Member
Patreon Donor
Joined
Nov 21, 2018
Messages
70
Likes
62
#6
Don't worry, most people who make confident assertions about technical issues in audio don't know much either. Dunning-Kruger is the rule, not the exception. Hand-waving is the usual style of audio "expertise," so ignore anything that is not supported by cold, hard data.

Look at the DAC reviews on this site (where things are actually measured!) and notice which units have the highest performance. The ONLY thing that counts is what comes out of the analog outputs- the nature of the black box to get that result is far less important.
Speaking of measurements, it just seems strange to me that “subjectivists” would reject them outright, claiming that you can’t say how good something is if you only look at the measurements

But most of the reviewers that have been reliable to me so far have been able to link their listening experience with what is shown in the measurements, which makes that line of argument even stranger to me
 

maxxevv

Addicted to Fun and Learning
Joined
Apr 12, 2018
Messages
658
Likes
476
#9
It was part of a bigger thread. Can't recall which one either Let me search through my much shorter message list to see if I can find it.

Not completely sure but was it part of the Yiggy or Jotu thread ?
 
Last edited:
Joined
Nov 4, 2017
Messages
40
Likes
28
Location
East Coast, USA
#12
Congratulations, you have just discovered that the vast majority of things said on audio forums is total ignorant nonsense. :cool:
Thank you for the morning's first laugh. I'd amend your sentiment to say that "the vast majority of things said is total ignorant nonsense." It seems to be getting worse every day, or maybe as I grow older my willingness to tolerate it is diminishing.
 

solderdude

Major Contributor
Joined
Jul 21, 2018
Messages
1,440
Likes
2,250
Location
The Neverlands
#13
not an expert in engineering, but from what I’m gathering from others outside here, “multibit” is supposed to be better because it “preserves the original samples”... but if said “multibit” design distorts the signal, wouldn’t that mean that it also isn’t really good at preserving the original signal in the first place?

and delta-sigma is supposedly inferior because it’s a simulation... but then what’s so bad about a very well executed simulation?
The 'preserving sample value' is BS with a bit of truth in it. I'll try to explain. Say we want to create a certain DC voltage with an R2R DAC. Then we apply the 'digital word' for that value to the DAC chip and out comes a steady DC voltage with no HF noise.
When we want to do the same with a DS chip then we apply the 'digital word' and out comes an average value of the same voltage that comes out of the R2R DAC but it has some HF 'noise' on it as well. That noise is above the audible range and low in amplitude so not audible. Result is similar as that of the R2R.

Both DAC chips will thus give the same DC output voltage. (assuming both are say 3V FSD).

Things get a little different when the output voltage is NOT static any more as in the description above. Music is an AC signal that has a varying voltage and is not a pure sine wave.

Now let's assume the 'digital words' describe a perfect sinewave of say 5kHz.
The R2R chip would show the familiar 'stepped' signal at its output. A sine-wave with gross deviations in amplitude as an R2R chip 'holds' the output value for the sample period at the value that is represented by the input 'word'. That 'bit perfect value' thus would only be a bit perfect value at a point in time (say a picosecond) and NOT during the total time the sample value is present.
The 'error' that exists in the time period the sample is held can be enormous with a 5kHz full scale 5kHz (arbitrary) sinewave.

yes, there are people idiots claiming a NOS filterless DAC is the BOMB. It is not 'accurate' nor bit perfect (99.something%) of the time. The rest of the time the signal is wrong in amplitude and timing.

It can be improved by adding a very steep (reconstruction) filter at the output of the R2R DAC chip. This 'smoothes' transitions between the sample values and removes the errors quite well. This works quite decently in CDplayers. The early non-oversampling ones.
It does not work for USB DAC's that can do many bitrates. In that case there would have to be a steep filter which is set differently for each sample rate. It does not exist.
What manufacturers do is either set the reconstruction filter at a higher than the lowest sample rate needs frequency or use no filter at all.
So the bit perfect argument is pure nonsense for R2R DAC's it does not exist.

The DS is different. It is much more complicated then I will describe but the reality is not far off.
A DS DAC chip looks at a previous value, the current value and next value (in reality it looks at way more values before and after the current value) and calculates which values would have been in between these 3 'sample' values and then creates a bunch of values in between the known values.
While doing so it makes small 'mistakes' as it can only make a limited number of 'values' (just a few bits available). These errors are in a very high frequency, way above the audible limit, and are filtered out with a simple analog filter.
The 'reconstruction part' is done in the chip by the DS part and not in the analog domain after the chip (as is essential in R2R).

In the end, when looking at the output of the DS circuit, it is just as 'accurate' as that of a R2R chip WITH a steep reconstruction filter behind it.

Both are thus equally 'not bit perfect' and the argument is bogus.

Most R2R DAC's do oversampling though and these thus generate (calculate) some 'expected' sample values in between 2 sample values.
This is done by digital filtering.
In most oversampling R2R DACs there are thus more 'invented' (carefully calculated) values than bit perfect values.
There are different types of 'upsamplers/filters' that do it somewhat different.
All depends on the digital filter IF there are actual 'bit perfect' samples present at all (sample values with the bexact same word value as the one presented to the DAC chip).
OS DAC's have filters looking waaayyy before and or after the actual bit perfect value and thus even the sample values that could have been bit perfect may not even be that. From what I understand from SCHIIT is that their filter always uses the actual bit perfect values and calculates the values in between where other manuf. (supposedly) recalculate every sample in their filters.
Take it for what it's worth. The SCHIIT method is NOT more accurate though.

So there you have it. The bit perfect story for R2R is utter nonsense. When it comes to linearity DS is better than R2R and when it comes to noise levels the 'signature' is different as in frequency range and the amplitude of it per frequency.
 
Last edited:

watchnerd

Major Contributor
Beer Hero
Joined
Dec 8, 2016
Messages
4,419
Likes
1,488
Location
Seattle Area
#15
not an expert in engineering, but from what I’m gathering from others outside here, “multibit” is supposed to be better because it “preserves the original samples”... but if said “multibit” design distorts the signal, wouldn’t that mean that it also isn’t really good at preserving the original signal in the first place?

and delta-sigma is supposedly inferior because it’s a simulation... but then what’s so bad about a very well executed simulation?

based on this, wouldn’t the implementation of both design types and the design as a whole matter more than whether or not one is “delta-sigma” or “multibit”?

just a horse sense take on this, lol
Well, here is the other fact:

Even if the "preserving the original samples" idea mattered, the vast majority of ADCs used in recording are Sigma Delta...so how can they be "preserved"?
 

bennetng

Addicted to Fun and Learning
Joined
Nov 15, 2017
Messages
532
Likes
389
#16
Actually IIRC there was a more in-depth discussion elsewhere. Could be imagining it ofc...
https://www.audiosciencereview.com/...ts-of-schiit-yggdrasil-v2-dac.3607/post-87474
The linearity and anharmonic distortion issues of the yggy are partially caused by truncation as the Yggy is only capable of 20(21?) bits. Adding dither can somehow improve the measurement but also violates bit-perfectness... their design theory.

A lot of people interpret bit-perfectness and digital accuracy in a strange way. Obsession with analog volume control is another example.
https://www.audiosciencereview.com/...lity-in-windows-using-wasapi.5272/post-117704
 

THW

Member
Patreon Donor
Joined
Nov 21, 2018
Messages
70
Likes
62
#17
The 'preserving sample value' is BS with a bit of truth in it. I'll try to explain. Say we want to create a certain DC voltage with an R2R DAC. Then we apply the 'digital word' for that value to the DAC chip and out comes a steady DC voltage with no HF noise.
When we want to do the same with a DS chip then we apply the 'digital word' and out comes an average value of the same voltage that comes out of the R2R DAC but it has some HF'noise' on it as well.

Both DAC chips will give the same DC voltage. (assuming for fun both are say 3V FSD).

Things get a little different when the output voltage is NOT static any more as in the description above. Music is an AC signal that has a varying voltage and is not a pure sine wave.

Now let's assume the 'digital words' describe a perfect sinewave of say 5kHz.
The R2R chip would show the familiar 'stepped' signal at its output. A sine-wave with gross deviations in amplitude as an R2R chip 'holds' the output value for the sample period at the value that is represented by the input 'word'. That 'bit perfect value' thus would only be a bit perfect value at a point in time (say a picosecond) and NOT during the total time the sample value is present.
The 'error' that exists in the time period the sample is held can be enormous with a 5kHz full scale 5kHz (arbitrary) sinewave.

yes, there are people idiots claiming a NOS filterless DAC is the BOMB. It is not 'accurate' nor bit perfect 99.something of the time. The rest of the time the signal is wrong.

It could be improved by adding a very steep (reconstruction) filter at the output of the R2R DAC chip. This 'smoothes' transitions between the sample values and removes the errors quite well. This works quite decently in CDplayers. The early non-oversampling ones.
It does not work for USB DAC's that can do many bitrates. In that case there would have to be a steep filter which is set differently for each sample rate. It does not exist.
What manufacturers do is either set the reconstruction filter at a higher than the lowest sample rate needs frequency or use no filter at all.
So the bit perfect argument is pure nonsense for R2R DAC's it does not exist.

The DS is different. It is much more complicated then I will describe but the theory is not far off. A DS DAC chip looks at a previous value, the current value and next value (in reality it looks at way more values before and after the current value) and calculates which values would have been in between these 3 'sample' values and then creates it. While doing so it makes small 'mistakes' as it can only make a limited number of 'values'. These errors are very high frequency, way above the audible limit and filtered out with a simple analog filter. The 'reconstruction part' is done in the chip by the DS part.

In the end when looking at the output of the DS chip it is just as 'accurate' as that of a R2R chip WITH a steep reconstruction filter behind it.

Both are thus equally 'not bit perfect' and the argument is bogus.

Most R2R DAC's do oversampling though and these thus generate (calculate) some 'expected' sample values in between 2 sample values.
This is done by digital filtering.
In most oversampling R2R DACs there are thus more 'invented' values than bit perfect values. All depends on the digital filter also IF there are actual bit perfect samples at all. OS DAC's have filters looking way before and or after the actual bit perfect value and thus even the sample values that could have been bit perfect may not even be that. From what I understand from SCHIIT is that their filter always uses the actual bit perfect values and calculates the values in between where other manuf. (supposedly) recalculate every sample in their filters.
Take it for what it's worth. The SCHIIT method is NOT more accurate though.

So there you have it. The bit perfect story for R2R is utter nonsense. When it comes to linearity DS is better than R2R and when it comes to noise levels the 'signature' is different as in frequency range and the amplitude of it per frequency.
that’s actually quite informative and easy to understand for someone who doesn’t really study these things, thanks!
 

JJB70

Major Contributor
Joined
Aug 17, 2018
Messages
1,075
Likes
1,767
Location
Milton Keynes, England
#20
People like simple marketing hooks, something that can be presented as being better therefore if you get a device with "X" it must be better than the alternatives. Technically it may be complete nonsense but these things can be remarkably effective in terms of marketing, especially if helped along by some clever guerrilla marketing to get a bit of bottom up momentum.
 

Similar threads

Top Bottom