DAC types and their sonic signature

[Calexico]

That should be very easy to confirm using captures of a piano recording and Paul's software.
I assume with 1-bit chips you mean Delta Sigma DAC chips. These aren't 1-bit also not when you had a couple of beers

Why not making this easy test with every dac reviewed here? It would be much more revealing in a direct way. And would definitely stop the debate about multibit.

 

[solderdude]

You mean comparing the original file with a recorded one and then analysing this.
The problem with it is that one cannot nearly look as 'deep' as one can with regular measurements.
Also the ADC would be in the test loop as well as clock drift.

 

[Calexico]

So actually it's not easy to confirm? So this has never been validated if a multibit render better complex signals?

 

[solderdude]

You seem to have missed the memo where multitone measurements tell you all you need to know about how a DAC handles complex signals.
check out a few of Amirs test results to get a feel of this aspect.

Paul's software comes in handy when all you have is a soundcard and want to compare a bit more objectively.

 

[Calexico]

Complex tones has lot of harmonics at different levels.
Their levels are reducing with time.
No test show if the reducing of level with time is respected or approximated.
No test show this for multiple tones at different levels going quieter with time.

 

[Blumlein 88]

Complex tones has lot of harmonics at different levels.
Their levels are reducing with time.
No test show if the reducing of level with time is respected or approximated.
No test show this for multiple tones at different levels going quieter with time.

That is because such a test isn't needed. You keep searching for something you think is a hard signal for DACs to manage. Why? Many of the standard tests are standard because they stress gear in a way we can learn how well they perform at the edges of their envelope of performance.

What would you expect this test to show you? We can create artificial versions of such signals or record real ones. I think I've posted recordings of cymbals showing what is captured on here at ASR in the past.

 

[zalive]

That wasn't the object of the test nor how pleasing the music is.

Well, you basically want a really good original recording for this purpose to allow as much room for discerning (degradation from DA-AD cycle) as possible.

I remember flac-mp3 test I took several times, on already compressed and degraded originals I missed more than once using the headphones (cheapo, though) . On few other samples I never missed. To me sample used made a difference.

You need to understand technical humor to get it.

However, when you switch one resistor in an R2R configuration fast enough and are using noise shaping it could perform better than a full R2R DAC.

Ok, checked it. No R2R, I made it up in my mind it seems. Instead conversion of everything to DSD256, then DA conversion of this 1-bit bitstream using a transverse filter.

Is this somehow similar to idea you described?

 

[Willem]

I just received my new RME ADI-2 DAC and all I can say is that fortunately it does not seem to have a sonic signature of its own. The next step is to investigate its multiple possiblilities for dsp equalization, tone and balance control, variable loudness etc etc. to get the in-room sound just right. It seems to be just about the only DAC that has all the features we have always deemed necessary for a pre amplifier.

 

[solderdude]

Instead conversion of everything to DSD256, then DA conversion of this 1-bit bitstream using a transverse filter.

Is this somehow similar to idea you described?

Yes, it could be an example of how one could use a single R2R section and 'short' the resistor to ground using something like a DSD stream for instance. That way one could design a DAC and claim it has an R2R 'structure' yet would convert a DSD stream.
More of a joke ... not to put ideas in some 'I want to be different just for the sake of it' designers head for instance.

Well, you basically want a really good original recording for this purpose to allow as much room for discerning (degradation from DA-AD cycle) as possible.

You can complain with @Blumlein 88 and suggest another music piece is to be used.

 

[Calexico]

About converting to dsd if find it's stupid.
If you want dsd then buy an old 1bit delta sigma dac.
They all already convert pcm to dsd.
New dac are multibit delta sigma for a reason.
Buy converting to dsd you loose all the advantage of multibit delta sigma and keep all the inconvenient of 1 bit delta sigma.

 

[Blumlein 88]

Well, you basically want a really good original recording for this purpose to allow as much room for discerning (degradation from DA-AD cycle) as possible.
snip

I used a few different recordings. Some were very minimalist with no processing straight off the microphones. Some were heavily processed. Can't please everyone.

 

[solderdude]

New dac are multibit delta sigma for a reason.

Indeed it is.
Enlighten us why you think that is so.

 

[Calexico]

Because it's better than one bit while keeping cost low. As majority of format is still pcm it would be better to improve pure multibit dacs but it would be too expensive.

I believe dsd is a kind of nos.
If you believe converting to dsd is better then you believe it's better to avoid the filter in the dac then you believe nos has benefits.
By the way the filter is done outside the dac with the computer and this must be proven that they re better than multibit delta-sigma filters.

Ideally there should be a format that would need least convertion and filtering and process as possible.
High res multibit pcm with a simple analog low pass filter in this way is the purest path.
all dsd records need pcm to dsd convertion. (Except for analog recordings that can be sampled in dsd directly)
That mean by converting to dsd you add one more step of converting.

So instead of keeping processing as simple as possible it become very complex.

 

[Wombat]

Apologies in advance but you did say believe:

 

[solderdude]

Because it's better than one bit while keeping cost low.

What is 'better' about it ?

I believe dsd is a kind of nos.

Some also believe in UFO's

If you believe converting to dsd is better then you believe it's better to avoid the filter in the dac then you believe nos has benefits.

This website is about the science behind audio not what one believes.
Your thesis is totally incorrect b.t.w.

Ideally there should be a format that would need least convertion and filtering and process as possible.

it's called DSD ... you don't like it.

High res multibit pcm with a simple analog low pass filter in this way is the purest path.

One can consider 48/24 high res PCM... I assume you mean well over 192/24 ?

all dsd records need pcm to dsd convertion.

Nope, DSD sound processing exists but is more basic only.
There are pure DSD recordings made that are distributed that way.

That mean by converting to dsd you add one more step of converting.

Tell that to Miska...

Your lack of understanding ADC/DAC conversion and how digital actually works is astounding.
I have no idea what you read about this topic and where but either you misunderstood or were misinformed.

 

[Calexico]

Pff sometimes you say really incoherent things. you didn't understant at all what i meant.

R2R NOS: you don't have oversampling filter. -> No oversampling -> NOS
DSD: you bypass oversampling filter of the dac. -> No oversampling -> NOS

Both of these want to bypass the oversampling in the dac.

That's why people prefer to convert to DSD, it,s to bypass the filter of the dac and to make it outside.
So it's the same argument that the ones that prefer multibit NOS dacs to don't have oversampling filter in the dac.

Then
DSD: they have choice to keep kind of NOS with DOP or to do oversampling with computer
R2R NOS: they have the choice to keep NOS or to do oversampling with computer

As there are very fiew native DSD records and lot of PCM records, most songs will have to be converted to DSD, wich is not lossless.

 

[solderdude]

Pff sometimes you say really incoherent things. you didn't understant at all what i meant.

R2R NOS: you don't have oversampling filter. -> No oversampling -> NOS
DSD: you don't have oversampling filter. -> No oversampling -> NOS

Your lack of understanding ADC/DAC conversion and file formats is astounding.

Both of these want to bypass the oversampling in the dac.

Digital files don't want to bypass anything, they have no will nor desire by themselves.
They just need to be converted (correctly) to sound waves in the end.
And this whichever way one desires/chooses to do so, with or without (purposely) changing the digitally described waveform.

Note that those that prefer filterless NOS R2R DACs purposely change the intended waveform thinking they are doing the right thing.
Because it sounds different they assume it is 'better' while the reality is that the reproduced waveform looks nothing like the intended waveform.

That's why people prefer to convert to DSD, it,s to bypass the filter of the dac and to make it outside.
So it's the same argument that the ones that prefer multibit NOS dacs to don't have oversampling filter in the dac.

Some people prefer to convert PCM to DSD because they..
A: claim it sounds better to them
B: Have a DAC that can only convert DSD
C: think it is technically 'better'
D: prefer this based on various theories they have

 

[Veri]

DSD: they have choice to keep kind of NOS with DOP or to do oversampling with computer

???

DoP = DSD over PCM, just a hacky way to transmit DSD over traditional PCM data flow (this requires double the bits over native DSD).
Native DSD transfer, or DoP does the same in the end, just one is more efficient. Neither has the slightest to do with NOS or oversampling....

 

[zalive]

while the reality is that the reproduced waveform looks nothing like the intended waveform.

You mean that stepped look of a waveform from a HF range when produced by non-OS DACs?

Funny, Fourier analysis says the perfect base tone frequency sinus wave is present there in its pure sinus form. Harmonics of the base tone which are obviously as well present there make the resulting waveform look stepped.

 

[hetzer]

1bit delta sigma modulator has briefly 5 problems. I'll explain all without math and as simple as possible.
But this will oversimplify the content and there might be some questions and errors. So I'll leave some link that will help.
Detailed Explanation Simplified Explanation Practical Example
1. Since there are only 2 levels of output available, there is trouble with printing out low level signal.

This results in dense high frequency distortion and this is signal-dependant. This distortion looks and sounds like noise(and is treated like noise) since it is so dense in frequency range, but this 'noise' fluctuates depending on input signal. Engineers call this phenomenon 'Noise floor modulation'. This is not 'high resolution' nor 'analog and natural sound'.

2. This problem also happens in '0' input. Even if you just want a silence, a little bit of DC offset made by deviation of electronic parts makes error.
Delta sigma Modulator is similar to a negative feedback circuit. The difference is that in 1bit delta sigma modulator, the feedback signal can be only 1 or 0. So the DC offset error can't be compensated forever.
The link below is a model of simple 1bit delta sigma modulator and you can check out how this works.
link

This results in the oscillation like below

3. On the other hand. multi-bit delta sigma modulators don't suffer from these problems because dither can be applied to them.
Dither noise is a random noise that eliminates the error and adds noise in cost. But this noise is 'natural' as it is constant. The gaussian dither noise is exactly same as LP noise.
However, dither requires at least additional 2 levels of output. 1bit modulators(with only 2 output level) will be constantly overloaded if dither noise is added into signal.

4. 1bit modulator has more high frequency noise than multi-bit modulator because it needs to shape more noise than 5-6bit modulator. This makes 1bit modulator more sensitive to jitter. In frequency domain, Jitter folds the high frequency signal to low frequency. So the massive amount of high frequency noise in 1bit modulator reduces dynamic range.

Furthermore, this high frequency noise is hard to filter on analogue stage and we need to pay many things(money, resistor noise, complicated circuit, etc) to get a nice attenuation. This is why DAC manufacturers advertise low out-of-band noise.

5. This is the limit of every delta sigma modulator in that they have stability issue when given certain amount of input level.
Every modulator has its modulation depth. 50% of modulation depth means that it can modulate 50% of original signal.
If I convert a PCM file to DSD with a standard modulator that Sony/Philips developed, the volume will be 6dB less on DSD. That's because the maximum modulation depth of 1bit delta-sigma modulator is set to 50%.

From: 'White Paper on Signal Processing for SACD' by two Philips engineer
On the other hand, DAC manufacturers solved this problem by using many tricks. And using multi-bit modulator is one of their trick. Since its SNR is better than 1bit modulator, engineers don't need to use high order modulator to get enough SNR.