Multi Bit DACs vs. Delta Sigma DACs

[laurent]

Thanks Coops,

Do those Dacs have something in common due to their "common" technology?

By the way, on the Ashtray thread on the Dac7 you mentioned that you were getting an unit for test. Did you finally try this Dac?

 

[Purite Audio]

Laurent Hi, well they all lacked the 'metallic' edge that a lot of dacs have, re the DAC 7 no I still haven't heard one, have you compared yours to the CD8? Keith.

 

[laurent]

Hi Coops,

Thanks for the info, so multi bit dacs are more analog like.

No I haven't compared the Dac7 with the Cd8. I did rapidly listen to the cd5 which I liked but did not compare t with the dac7. Certainly like many here I went for computer audio because I was fed up of having so many cds, loosing them and not having enough room for all of them... So no more cd player even though they may outperfrom computer and dac.

 

To Mani,

Your pacific microsonoc model two is a multi bit dac. Since the Berkeley Audio design a kind of legacy of Pacific Microsonic, is the Berkeley Alpha Dac also a multi bit dac?

If it is the case, I think Chris has done a great review of a multi bit dac.

Since I like a large soundstage, something I didn't find in the Weiss Dac2, I think I really should have a look at the Alpha Dac.

The best set up I have heard so far was a Weiss Dac2 connected in Xlr to a Dac7.

I really would like to hear a Weiss Af1i and a alpha dac together.

 

[manisandher]

Hi Laurent,

 

You know, I have no idea whether the Alpha DAC is multi-bit or delta-sigma. (Does anyone else know?) If I had to guess, I'd say delta-sigma... those companies that use multi-bit chips nowadays tend to state so, and I haven't heard this of the Alpha. But this is only a guess. If the Alpha does use multi-bit chips, then I think it becomes a far more interesting product...

 

As far as the AFI1 is concerned, I love the fact that it is a minimalist product with no software-controlled mixer, etc. For 2-channel computer audio, this simply isn't required, in my experience. One day, I will do a detailed comparison of the AFI1 with the RMEFF800 & MOTU896HD (both acting simply as computer interfaces and not DACs). In the "bits is bits" world, there should be no difference in SQ between the three...

 

Mani.

 

[manisandher]

I_S,

 

You said, "DSD ( as used on SACD ), which is basically using delta-sigma as a medium. This runs at 2.8224MHz, 1 bit ( which is much slower than the 11MHz rate I mentioned earlier ) - this is problematic, as there is less spectrum to spread the noise over, so the noise rises very sharply above 20kHz."

 

Actually, SACD beocmes more noisey than 16-bit PCM from about 8KHz onwards! Not only that, but the massive averaging taking place just kills transients, sucking all the blood out of the music in a much more severe way than delta-sigma. (Having been an early adopter of SACD, I now think it was one of the biggest swindles in audio and am so glad that hi-rez is PCM-based.)

 

You said, "And lastly, thanks to the work of Fourier, Nyquist, and Shannon, we know that for a band-limited signal, we can reproduce it EXACTLY by combining sine waves ( of difference frequency, amplitude and phase )."

 

I know I'm on very thin ice here (it's been a VERY long time since I last looked at my physics undergraduate texts) but isn't this only true for periodic waves? [EDIT: But I suppose that's what you mean by 'band limited'.] Don't VERY sharp and irregular transients get lost? This certainly seems to be the audible effect of delta-sigma, and to a much greater extent, SACD.

 

Mani.

 

[Synfreak]

Hi Mani!

 

It might not fit into your scenario of how (multi-) delta-sigma DACs are working, and how this _could_ screw up the sound, but DSD (name it sigma-delta) is way better in (re-)covering sharp transients as linear PCM, just because of its much wider bandwith. The downside is - of course - that a lot of overtone-information / frequency content will be "masked" by the relatively high (noise-shaped) noise.

But this isn´t as bad, as long the downstream components (Pre- and/or Poweramps) aren´t going berserk because of this high emnergy content in the higher registers.

 

To me (and I´m listening to a PCM 1704 based player), it is much more important how the digital (and analogue) filterstages are designed.

We might actually hear more "filtersound" as we hear the "DAC design" (and Bob Katz seems to think the same direction...).

 

Sidenote:

Most of the modern ADCs are sigma-delta designs.

 

Cheers

Harald

 

[manisandher]

Hi Harald,

 

I totally agree that the filters are a big (one of the biggest???) factor. Perhaps another thread should be started on this...

 

You said, "It might not fit into your scenario of how (multi-) delta-sigma DACs are working, and how this _could_ screw up the sound..."

 

I'm an experimental physicist by training (I spent years locked away in a lab creating 'superconducting tunnel junctions' for work in phonon spectroscopy, looking at the electron-phonon interaction in 2-DEGs). The reason why I was an experimentalist was because I wasn't very good at the theory! But there's one thing I learned... repeatable data is always right in the end. If the hypothesis doesn't fit the repeatable data... change the hypothesis, not the data.

 

In this case, the 'data' is my experience listening to these things. I spent a fortune on SACD and really wanted it to be good. But it simply wasn't - to my ears, the transients were killed. I've heard the same 'effects' in delta-sigma DACs, but to a lesser degree.

 

Now, can my 'scenario of how these things work' explain what I'm hearing? I don't know. But if it can't, then I'll change my 'scenario', not my experience of what I'm hearing.

 

This is all very subjective of course. I have a friend who has the full-fledged dCS rig. He likes the sound of 16-bit PCM converted to DSD. I don't. His second choice is upsampling. I don't like this either...

 

Mani.

 

[PeterSt]

Hi Harald - please allow me :

 

but DSD (name it sigma-delta) is way better in (re-)covering sharp transients as linear PCM, just because of its much wider bandwith.

 

I am sorry, but this doesn't suffice for me. This time (not yet anyway) for a change I won't explain the how's and why's as I see it, but I sure like you to explain how you see things are going for this aspect. And oh, I'm from the league of understanding and next being able to explain, and will not just copy a one liner to a next one who wants to know.

 

Might you need some adrenaline, I have a one liner too : the lower the resolution, the higher the transients.

If you want me to explain that, I will, but I think it would be really nice when you are first on yours (and please don't feel attacked or anything, because we are just nicely working out some things; if you turn out to be right, then I am obviously wrong. No problem with that).

 

Thanks,

Peter

 

 

 

[Synfreak]

@ Mani:

Wouldn`t it be a challange - and of interest in a common sense - to get a picture "why" you like the (untouched) pcm "sound" more as a upsampled (be it pcm or dsd) example of the same audio?

_Maybe_ there is a kind of "distortion" (in what way ever) which "adds" up for the sound you like?

Don`t know, but this might be a point to get some evaluation done.

 

@ Peter:

The transient of which you are talking about seem not to be the same I talked about to in my post ;-)

Are you talking about resolution in the frequency or in the amplitude domain?

 

To catch the (high) overtones of short attacks/transients, the higher the sampling rate, the better.

 

Cheers

Harald

 

 

 

 

[manisandher]

Yep, maybe there is a type of distortion that I like...

 

I think trying things and evaluating for yourself is really important. But of course, it's more easily said than done - there are just so many variables, aren't there?

 

But here's what I'm planning for my 'ultimate evaluation':

 

1. get a very good analogue source (preferably open-reel master tape)

2. digitize with WaveLab and Model Two (at 24/88.2 and 24/176.4)

3. playback with WaveLab

4. compare with analogue source (preferably double-blind)

 

If I cannot distinguish between the analogue and the digital, then my 'computer audio search' is over.

 

Now, all I need is a good analogue source! If anyone has one, and would like to be involved in this 'ultimate evaluation', let me know (I'm based in the UK).

 

Mani.

 

[PeterSt]

Hey Harald,

 

It seems that you may be looking at the wrong side of things ...

 

To catch the (high) overtones of short attacks/transients, the higher the sampling rate, the better.

 

... which would count for the ADC.

 

Are you talking about resolution in the frequency or in the amplitude domain?

 

Amplitude.

 

Still your turn I think. :-)

Peter

 

[Purite Audio]

Slightly off topic, but does anyone still record onto magnetic tape anymore?

 

[Audio_ELF]

Well a quick Google got me this "Clients come to this Manhattan recording studio, for the great vibe, gear and Mike Caffrey's production and engineering. Known for great drum sounds, this is one of the few recording studios that still regularly uses analog tape as well as Pro Tools HD (8.0) and Radar for recording. If you've heard Avril Lavigne's song Keep Holding On from the movie Eragon, or her next single Girlfriend, you've heard these drum sounds." from Monster Island Recording Studios

 

Eloise

 

[PeterSt]

(edit : please read my next post first !!)

 

Wouldn`t it be a challange - and of interest in a common sense - to get a picture "why" you like the (untouched) pcm "sound" more as a upsampled (be it pcm or dsd) example of the same audio?

_Maybe_ there is a kind of "distortion" (in what way ever) which "adds" up for the sound you like?

 

I am sorry again, but the time it was allowed to think like this, is behind us.

It is behind "us" since my own multibit doesn't show one bit of of distortion above the noise (all up to 96KHz). But in case you want to know, it is around 90dB down (http://www.computeraudiophile.com/content/Playback-Engines#comment-21808 and don't look at the noise level here, because it shows wrong there).

 

I know exactly how the harmonic distortion of NOS sounds (if I'm allowed to translate this to NOS), and it will be snappier, more fresh, and less dead than without it. However, without it as in my multi bit situation, horns have just the nature of the horn involved (instead of all leaning towards trumpets), playback is black as can be instead of the opposite, no requirement are felt from the remainder of the chain (which most certainly feels the opposite when the HD is still in it), the resolution is as high, but in the end higher though observed in the LOWER regions of the frequency band), all albums play, instead of only those allowing for NOS (for instance, I never got hiphop doing well ... laugh), generally you can say that there is no stress at all, while with the "normal" HD things can get rather the opposite (this is a bit similar to the feeling the rest of the chain is able to cope), there is no haze or flare or anything on top of anything (which almost always is the case with HD NOS), and in the end ...

 

hahaha, the sound is very similar to Delta-Sigma. One big difference though :

 

Delta-Sigma sounds completely dead (I call it a dead bird), there are no transients at all (meaning : yeah, that bad it is), and there is no way instruments can be recognized out of a full orchestra playing.

Don't blame my loudspeakers or anything.

 

I have told it more often : when I have a couple of good ears over, and the quest(ion) is the same, I let the guys guess those instruments, and while most often 10 different ones are proposed with a look in the eyes "who cares", I switch to NOS, and there's always one option left for the instrument, the audience smashing Delta-Sigma like Jimmy Hendrix would his guitar.

And this is also with the heavy distorting NOS, might someone think distortion makes the instruments real. Haha.

 

So, multi bit NOS is in another leage, whether people believe it or not, but it is not in the good leage and Delta-Sigma might win when you for instance don't care about not recognizing instruments (which for most can't be done anyway because of the remainder of the chain) and like dead birds in the first place, but multi-bit NOS with very low noise level (which is key) and proper filtering is yet in another leage, and the best of all worlds.

 

No HD, no transients killed.

 

Ah, the latter was to sort out. Sorry. hehe

Peter

 

2c added :

PS: I have an ESS Sabre here, and considered it the best analogue representation when I first heard it. After two days listening to it, I threw it out of the listening room because there was no fun in listening (dead bird) and my brain got tortured because of wanting to know those instruments (which is an amiss if you're only used to it). What I have today, is as analogue (just urging for Mancini and all) but now everyone jumping and dancing in front of you.

 

[PeterSt]

I overlooked this part, which is more explicit than, well, I saw it (thus overlooked it) :

 

why you like the (untouched) pcm "sound" more as a upsampled (be it pcm or dsd) example of the same audio?

 

To me this indicates that you (Harald) are more on the track I am, than I initially read/thought. I mean, this won't unjustify the contents of my last post, but it may unjustify the post being there in the first place.

 

I let it stay because it is as informative (well, I hope), and it may be a good stepup when i_s turns up with his most awaitened filter part.

 

There, I assume, the pieces of the puzzle will start to fit (which doesn't say (yet) that we have all the pieces).

 

Thanks,

Peter

 

[manisandher]

Thanks Peter. I've never heard a 24-bit NOS DAC, but looking forward to doing so sometime in the future...

 

Harald, you said, "why you like the (untouched) pcm "sound" more as a upsampled (be it pcm or dsd) example of the same audio?"

 

I remember doing a small 'experiment' when I downloaded my first hi-rez files from Linn quite a while ago now. I downloaded two versions of the same song - a 24/88.2 file and a 16/44.1. I then compared three 'versions':

1) the 24/88.2 file

2) the 16/44.1 file

3) the 16/44.1 file upsampled to 24/88.2

 

For sure, 1) sounded the best.

 

BUT... there was no question in my mind that 2) sounded closer to 1) than 3) did.

 

Certainly not conclusive, I know, but just one of the reasons why I believe what I do...

 

Mani.

 

[cfmsp]

 

just curious,

 

What product(s) did you use for dithering and upsampling in example 3?

 

and which did you do first?

 

clay

 

 

[PeterSt]

Hi Mani,

 

To be honest (you know I always like to be), I only hope you didn't come to those conclusions by using XX. I mean, the version you have contains no filtering, or anyway not that I recall when you were helping me out with those graphs. Remember the roll off we saw there ? that's a typical "no filter after upsampling" situation, no matter you looked at it through the (Delta-Sigma) Fireface because you will have been looking digital only).

With the AA (Anti Alias) checkbox ticked the story is different, but soundwise I never liked that too. And besides, 2 times upsampling is not enough, and it will stress your amps all the way (I say this, because how to do it 4 times the proper way, knowing you won't be having a DAC allowing for that).

 

Anyway, I think I never said it, but I too never liked upsampling. Today it is just a necessity to get it right, and I won't go back because of the enormous (positive) difference.

To make the story complete, it won't go without filter if you want the HD out of the way OUTSIDE the audio band. Inside is no problem without filter.

My amps can bear everything up to 200KHz, but the filter still makes the difference. Could be the loudspeakers which can't cope.

 

If you - never mind the virtues opposed to Delta-Sigma - can only sense downsides of multibit NOS (one is enough), think those out of the way, and that is what I have.

And let's try to keep in mind : with a 96KHz DAC it can't be done.

 

Peter

 

 

 

[manisandher]

"I say this, because how to do it 4 times the proper way, knowing you won't be having a DAC allowing for that"

 

Peter, not sure what you're refering to here, but both the Model Two and D70 are capable of accepting 4fs upsampling.

 

Mani.

 

[manisandher]

Peter, this was before you implemented upsampling in XXHE... which you know I never liked.

 

No, I burned a CD from the 16/44.1 file using a Yamaha burner and its 'audio quality' setting.

 

I then compared the downloaded 24/88.2 file (XXHE->FF800->Esoteric D70) to the burned file (Esoteric P70->D70) with and without 2fs upsampling.

 

[EDIT: i.e. the 2fs upsampling was carried out by the P70 transport, which is capable of 4fs too]

 

(IIRC, I got pretty similar results when I upsampled with Foobar).

 

FWIW, I don't use the upsampling capabilities of the Pacific Microsonics either when playing 16/44.1 files. Why would I want to? They sound fine as they are!

 

Mani.

 

[idiot_savant]

Righty-ho, I'm back, with a bit about filtering, why it's needed, how some people do it, and ( hopefully ) a super-duper google spreadsheet to illustrate some points!

 

So, we know that in digital audio, we sample the audio at a frequent, consistent rate ( say 44100 times a second ). We note the value of the sample, store it, and when we want to play it back, we feed the same data into our hypothetical dac ( imagine our simple 8 bit DAC from earlier ) at the same speed. Voila! We get back our original waveform! Or... do we?

 

Most people think that if you sample a 20kHz sinewave at 44100, you get a triangle shaped waveform out, because there are only 2 samples per cycle, right? I've linked to a google spreadsheet that ( theoretically ) anyone can open - go to the bottom and press "Edit" to begin the fun and games - it will plot data that represents a sine wave sampled at 44.1k ( red book )

 

http://spreadsheets.google.com/pub?key=tV3yza5CKyn7wn_5V7DpDLA&single=true&gid=0&output=html

 

IF something has happened, you will see a spreadsheet with a graph on it. Go to the bottom and press "Edit". There should be a frequency box that you can change. If you type in say 1000 into the box, you will see a nice sinusoid type of thing. Brilliant!

Now, try say 6000, 20000, 24100, and finally 43100.

 

Let me know if this works, and what you think,

 

your friendly neighbourhood idiot

 

 

[cfmsp]

"Let me know if this works, and what you think,"

 

Well, if your spreadsheet worked, which I believe it did, the model on which we're basing digital music reproduction doesn't.

 

 

thanks

clay

 

 

 

 

[idiot_savant]

Thanks clay, now this is where I can start to explain what is actually happening with sampling - I'm desperately trying to avoid the maths, so here we go:

The spreadsheet isn't quite what happens with a NOS filterless DAC, but it is close! Every DAC will have the ability to hold the output between samples ( so the graph will have flat bits between samples, and only change on sample transitions ).

 

What you are seeing here is imaging. The act of sampling something means that you are linking it forever and ever with the clock that is doing the sampling. The fascinating thing is that the set of samples you create can represent the thing you want to sample, but also it's mirror in frequency around the clock you sampled it at, and that mirror.

Whenever you sample, each point can represent a point on an infinite spectrum, mirrored around the sample clock.

So, in our example, we put in 1000Hz - what happens is that you also create tones at 43100Hz, 45100Hz, 87200Hz, etc. etc. which are basically all mixed together. Now because of the scale of the graph, these higher frequency components are invisible compared to the tone we have sampled.

When we put in 20000Hz, we also get a tone ( of the same size ) at 24100Hz. Because they are relatively close in frequency, we start to get horrible beating between them.

 

So, does this mean that all digital audio is broken?

 

Well, this is where filters come in. If you experiment in the spreadsheet, you will see that as the frequency you type goes over 22050, the frequency of the sinusoid appears to come down - this is aliasing. Any frequency over the sample rate over 2 will "fold" back over Fs/2, so if you just use a filter to ensure that nothing over Fs/2 gets into the ADC, you can avoid this...

So, how can the DAC avoid recreating all of these infinite high frequency tones? Well, if it knows that the original ADC had a filter around Fs/2, it therefore knows that there cannot be anything to reproduce above Fs/2. And this is where the anti-imaging or reconstruction filter comes in ( which can be done digitally, or in analogue ).

 

I'll let people think about what this means,

 

your friendly neighbourhood idiot

 

EDIT: it occurs to be people may get hung up on the spreadsheet - it's very simple, and if you were to put the resulting data through a properly filtered DAC, you would get out perfect sinewaves ( for frequencies up to 20k or so ). /EDIT

 

[PeterSt]

If you type in say 1000 into the box, you will see a nice sinusoid type of thing. Brilliant!

 

i_s, at first I thought I didn't know what a sinus was, next I thought you don't, and then I saw half of the time it doesn't work. It looks as if my PC is too slow or something.

 

But when it's working ... very nice job.

 

Btw, it looks like above 44100 it is not working and very often the most nice sines come from it. I'm not sure that happens all the time, and while I thought I just saw 87000 at some intended shape, now it's a nice sine.

 

Although it may be correct by some theories I don't know, if I look on the scope for e.g. 11025 or 22050 I see perfect squares (which may not be perfect at all). http://www.phasure.com/index.php?topic=642.msg4806#msg4806.

(please don't read the withgoing text, or we may loose some objectivity here).

 

Peter

 

 

[PeterSt]

"I say this, because how to do it 4 times the proper way, knowing you won't be having a DAC allowing for that"

 

Peter, not sure what you're refering to here, but both the Model Two and D70 are capable of accepting 4fs upsampling.

 

Sorry Mani, my mistake ! I had in my mind that you told you could use 96KHz only. The D70 I knew anyway and in the end I know you must still have it somewhere. I guess I was too fast on reading your upsample experiments to 88.2 only ...

 

Peter