NOS (non-oversampling) DACs - Dan Lavry comments

[emmodad]

this may be interesting for some of you, especially if you've not had the pleasure of suffering through years of EE study, Digital Signal Processing theory, and DAC design, or spent much time in the entertaining über-tech world of the AES......

 

those who have had such good times, know that in the world of dsp audio professionals, there is a very short list of people who truely understand the fundamentals and application of dsp, system design, analog and digital design, etc etc. -- and make products which prove it.

 

not hobbyists who "design" products; not "producer/composers who started recording" and spout drivel in online fora about dsp; not people who come from on-chip interconnect design and are stepping far outside of their expertise in pontificating about signal processing...

 

Dan's is one of the names often mentioned by the leaders in the field as someone who truely understands the technology of digital signal processing as applied in creation of pro audio products (as context, other highly-regarded names you may encounter in the world related to CA are such as Daniel Weiss, without a doubt one of the gold-standard dsp guys; Bob Adams (Analog Devices, you owe it to yourself to read everything he's published about data conversion and sample-rate conversion); Jim "JJ" Johnston (one of the fathers of perceptual audio coding among many other signal processing achievements); Rich Cabot (tech brilliance behind the creation of Audio Precision and many other signal processing developments); Karlheinz Brandenburg, Harald Popp, Jurgen Herre and others at Fraunhofer; John Strawn, Mike Story, Bob Stuart...)

 

why the buildup about Dan?

 

simple: there is a lot of posting in audio fora about the alleged "superiority" of NOS DAC design. Generally, when you do some forensic investigation and track back through web postings, "reviews," references, etc; you find that the posters generally have little more than subjective info ("I like how it sounds") or anecdotal stories ("I've heard that..."; "my customers say..."; "this guy has posted a lot and says"); and as a rule few have any idea at all about the fundamentals of digital signal processing, sampled-data systems, analog and digital filter design, data conversion fundamentals, overall audio system design....

 

well, Dan is the real deal. He posts based in technology and fact. Any occasion when he makes time to post (especially in fora filled with die-hard subjectivists) and explain technology in a way to shortcut years of academic study, it's worth reading.

 

over in head-fi land right now, there's an interesting thread where Dan is taking pains (with some obvious frustration, he may be blunt but that is his nature) to explain the technology fundamentals underlying concerns about NOS design.

 

worth a read, enjoy:

 

http://www.head-fi.org/forums/f7/nos-dac-marketing-bs-438220/

 

[icebreaker]

Lavry's comments are a very interesting read. I believe he says that if your NOS DAC is actually an NOS DAC(many aren't) then it may not have anti-imaging filters. If that is the case then your NOS DAC is decreasing the performance of your amp, speakers, headphones, etc. due to the image energy beyond the range of human hearing. (see post#15) I wonder if that energy could cause any long term damage to your equipment? Shortened life due to being overworked? I shall read on...

 

[davidR]

I'd like to hear Peter's thoughts on this especially concerning Lavry's comments after his inital post on headfi.

 

[audiozorro]

Dan's article and the 5 pages of posts is a bit much. After I read the below post on the first page it was all downhill from there.

 

I don't think that you quite understood Dan's post. His point was not so much that NOS DAC's are inferior, but rather that there is no such thing as a non oversampling DAC anymore. All the DAC's that I know of subject the signal to some kind of oversampling, whether they claim to be NOS or not.

 

I remember the TV show called "To Tell The Truth" from many years ago. It had three contestants, all claiming to be the same person, and the panelists ( Peggy Cass, Kitty Carlisle and somebody else) would ask the questions, then vote on who they felt was the real person. To Tell The Truth was very entertaining and we might find it equally amusing if readers would post their lists of real and false NOS DACs. So,

 

Will the real NOS DAC please stand up?

 

[PeterSt]

At first, last Friday, I created a post so long that it occupied 5 pages in a Word document. Saved the document and went on.

Then, Monday, I created a post for in the Head-Fi thread itself, found myself politely asking questions to mr Lavry, put that aside too, and went on.

 

Today, where mr Lavry denotes the thread as "gone completely nuts". I'd rather say he himself has lost a few. There is no sense in posting anything, because this is not a discussion, but a God at word. No, shouting.

Not for me.

 

Since the man is talking to himself only, only listenes to himself, and carefully avoids anything that could be cross to his ideas, sadly I learned exactly NOTHING from it, while it easily could have been different.

Read his 27 page paper as well, and the only thing coming from that is that he knows his stuff, but in that one direction only.

 

To summarize my 5 pages :

It seems I can't find any way to let sound OS (read : upsampled 4x) and filtered better. Spent last weekend to create a filter for 2x upsampling (stopped when it measured best), but it didn't even last one hour at Sunday evening.

I listened to the fsx4 + filter for some 4 weeks, and while at first all sounded the most black and "quiet", switching back to NOS made me realize I had been listening to a dead cat and wasted 4 weeks of potential good listening time. And no, I don't like listening to distortion.

 

The only thing I can come up with, against the theory of "reconstructing the wave as how it was", is that this cannot be true at all because of the ringing which goes along with it. No free lunches here.

The only thing I can see at this time is that it *must* be so that the ringing will create even more anomalies than the imperfect(ly reconstructed) wave. The net result is just worse.

 

Stating (but that is in the other D.L. CA-thread I think) that a higher sample rate than 44.1 isn't necessary because "everything" can be resonstructed anyway, to me sounds like nonsense, because you'd always have that ringing which a native (e.g.) 24/96 doesn't carry (but leave the stream alone).

 

For fun I created a filter that didn't protect against clipping. After a first impact of a firm bass wave (say the good plunk on a bass), the echo of the filter was audible for SECONDS (at an interval of 300ms or so). If that doesn't clip, what do you hear ? the same echo but without notice because it is bass. IOW, fuzzyness.

 

For the next few days I may try linear interpolation again. At least there won't be a filter at play *and* it removes the aliasing from the audio band.

So yes, I am in the middle of deliberately applying the wrong solutions, only to obtain a better result opposed to NOS/Filterless for once.

 

Peter

 

PS: I guess I'll end up with some super smart interpolating filter, because it seems the only good way for everyting and all. So far that needs smartness beyond my own. haha

 

[BobH]

Just carry on doing what you're doing and try to enjoy yourself!!

 

There are many valid ways of building of an amplifier - Class A, Class A/B, Class D, Class T, Class of 69, whatever!

 

There are many valid ways of building speaker drivers and speaker cabinets - paper, aluminium, mylar film, transmission line, closed, ported, horn, electrostatic, multi-driver, single driver, learner driver, whatever!

 

Same goes for everything else. Unless it's a Nos Dac, - or it's a Not-Really-A-Nos-Dac - or it's a I-Say-I-Am-But-I'm-Not-A-Nos-Dac-Dac - apparently!

 

I'm sure Mr Lavry had the best of intentions in choosing to air his views in the way that he did, but the bottom line is it really doesn't matter if he can prove that the implementation is technically inferior. It is a pointless proof if people still think the technically flawed implementation sounds better! It then becomes simply technically technically flawed and not practically technically flawed. If you get my drift!

 

This sort of discussion is rife in audio forum's everywhere and it always ends in tears simply because there is no such thing as 'The Right Way'! If there was such thing, it would be dirt cheap and we'd all have one!

 

[Synfreak]

... that Peter has also not completely got the point(s) Mr. Lavry was talking about.

 

Peter, at one point you talk of using NOS DACs would be better, and at the same post you mention that you will use a "interpolating" filter to get the best results.

So ... what do you mean a good designed oversampling filter (outside or build into a DAC chip) should (and would) do? How would you like to interpolate on an TRUE NOS DAC? there are now "new samplepoints" where you can fit in your interpolations.

 

And the mentioned pre- and post ringing is not a failure by "bad design", it is because you look at a band limited signal!

There is no 100% perfect squarewave, it is just impossible.

The higher the samplingrate (or the "oversampling" filter, and the characteristics of the filter), the less ringing (amplitude) you might get to see, but it will never be eliminated completely without forcing some other problems (i.e. in the frequency domain). So you will end up making compromises.

 

Badly my english is definately not good enough to get into detail (the friendly neighbourhood idiot might chime in here), and I´m far off to be a audio-skilled technician, too.

 

Peter, I hope you don`t read this as attack against your person, but I think you mix up things, which should not be mixed up at all.

 

Yours

Harald

 

[barrows]

Bob:

 

While I really agree with you in spirit, I feel it is important to point out that there is a "right way" and that audio reproduction is objective in nature-ultimately we are only trying to accurately reproduce something (sounds in time) accurately. The reason that the sound we like to get out of our systems is subjective in nature (and I agree that this is the case) is because technically we are so far from getting it "right" and knowing how to do that, that we are left with a bunch of choices between things which all sound "wrong". We end up choosing components that sound pleasing to us, because nothing really sounds "right".

In other words, we have a long way to go in audio reproduction and high end sound!

I have a great deal of respect for engineers like Dan Lavry, who take a technical approach to audio reproduction, in order to get us closer to achieving and discovering the "right way". They may take missteps along the path-but ultimately they are on the right path.

 

[BobH]

And I concede fully that both the industry and the hobby need, and benefit enormously, from the talents of engineers such as Mr Lavry. Where I normally come unstuck is those 'grey' areas, if you will, ending up as discussions of merit on audio forums.

 

For example, let's suppose that the industry decides to bin NOS Dacs as a technically inferior solution. We can do better, let's move on. Now let's suppose that one stubborn individual thinks, 'No, I can fix this!', and carries on regardless. Now, what happens if that one stubborn individual makes a break-through? All the non-technical people have been conditioned to accept that NOS Dacs are rubbish and along comes this guy with a brand new NOS Dac! He now needs to fight public opinion to get his product accepted.

 

This process, of continual 'one-upmanship bickering, happens all the time within the scientific communities - proof/dis-proof new idea - proof/dis-proof/proof etc, etc. A continual cycle of improvement, regression and knowledge-gaining that is, and this is the important bit for me, normally kept within the closed community of those concerned. Increasingly in our hobby, and particularly within the 'digital' field, we see this sort of 'healthy discussion' taking place in public.

 

We've had many such threads on this forum - it's my way or the highway - which I don't think are ultimately helpful, either to those of us on the hobby side of the line or indeed to the manufacturers themselves! The threads by I_S are particularly useful here - they are technically informative to those interested in such things and promote the view that to make any kind of Dac well, compromises will have been taken. This let's us know that liking a NOS Dac is not, by definition, a bad thing. Yes, it is compromised but the other way of doing it will be compromised as well! A manufacturer of OS Dacs promoting the view that, actually, all NOS Dacs are either bad, or lying about their NOS-ness, is not really that helpful.

 

I stick that sort of information in the same pile as that belonging to manufacturers who tell me that the only good speaker is an active speaker, that all Dacs are the same and that the only good usb implementation is mine! As has been said many, many times - manufacturers and dealers need to be very careful indeed when they choose to actively participate in general forums. Sometimes even the very best of intentions can back-fire!

 

[Synfreak]

Hi!

 

A short time ago I stumbled (by chance) over this PDF:

http://www.acourate.com/freedownload/SRCFilterInvestigations.pdf

which is based on some "theoretical" thinking about how oversampling "works".

It isn`t a scientific paper, so there might be some errors and shortcomings in there, but if you go thru this, you might get the point on how oversampling "works", and where the tradeoffs may lie.

 

Cheers

Harald

 

[PeterSt]

Hey hey, there's no way I can be offended by a post like yours. But I guess it is the underlaying "real subject" that makes it like that, and ... which may not have been 5 pages in its original form for no reason.

 

To summarize (ok attempt) that subject :

1. Whatever I try, NOS/Filterless sounds best;

2. I want to know why, hence what is actually going wrong with OS.

 

So, from this springs that I had to built an NOS/Filterless DAC in the first place, in order to have the options of OS and filtering. Thus, thinking about this a bit ... it won't go the other way around !

 

Next thing (if I not already have said it), I *understand* the theories, I don't see where they flaw (but for ringing maybe, see before post), and while I understand the theories, I don't understand why they don't work out.

 

Besides all, and for a possible better understanding, try to imagine me as someone who will spend infinit time (and money when needed) to at last understand what's not understandable (by me at least) at first.

May I recall the reason to create a software player ? This was the most simple : I could not understand why different players -all checked by me for being bit perfect- sounded different. Well, one solution : make one yourself.

 

Yeah yeah, stupid, crazy, name it.

 

So you see, the same I'm doing now with a DAC. Not out of ignorance, but I try to understand what I hear, so I will dive into some serious project to get there. Where ?

a. NOS is not good at all;

or

b. It is, because ... and you overlooked this and that.

 

I will get there, may it be a. or b.

 

What confuses, is the a. So, I *have* to explore the oversampling and/or (!) filtering, in order to proove for myself that indeed it isn't working. What may come from it, is that it just does (which so far fails).

Thus, keep in mind that I am working on the very same DAC always, which of course is the best to compare because all hadware stays the same. Apples and apples.

 

Allright. More to the subject, it may be hard for most to see what this is actually about. Thus, this is not about a single subject with an on/off switch, no, it is about many parameters of which currently someone like Dan Lavry tells us that alhough there's many parameters, there's really one switch. Switch that Off (NOS) and you're bad. Switch it On, and you're good all the way.

This is not so in my humble opinion.

 

Barrows, if it were for me, you hit the nail on the head, and what you express is what I have been expressing for a long time now :

 

There are so many, many things wrong with audio, that all of the other things mask the one.

 

So, for example, if you'd look at an FFT from NOS/Filterless this is one big mess. Btw, don't forget to use something like 18KHz, and don't use 1KHz. 1KHz is "allright". Now, I could "show" you output like this (telling you 30% distortion !!), and nicely filtered output telling 0.004% distortion, and you will have the most hard time to find a difference at listening. Even the knowledge (hence no ABX) of what you listen to is the good or the bad one, doesn't help. BUT :

 

This highly depends on the interface in the first place. So, don't underestimate this, and while, say, an USB connection indeed may show a huge difference, SPDIF does not. This is about just one of these zillion things wrong in audio, and what I meant before, such a one thing may mask all of the others.

 

My before 5 pages tried to show these kind of parameters, and tried to explain what happened or did not happen by empirical finding. I explained that the more close we get to 1:1 reproduction, the BETTER NOS sounds. I tried to explain that the better we represent the WORSE measuring situation, the better it becomes for net result.

 

But I have no scientific proof, so I just as well leave to me saying such a thing. Period.

 

In the past I have been connecting many of these in fact strange results, to nature. So, this is a bit in the area of "but our ears and brain filter too", although I would never say exactly that. This is more in the area of "when SPL meets the original SPL during the recording, all suddenly fits" (yeah, think about this).

I have more of these high-flying subjects, which just came to me by implicit empirical findings, but next should be explained.

 

So, again, and this should be the summarized message (leaving out the five pages), I have been working step by step from various angles (starting at the playback software) on better 1:1 reproduction, always knowing that -theoretically- the closer I would get, the more digital it SHOULD become (for sound). It NEVER happened that way, and it ALWAYS went the exact other way around. And this was always with NOS (where NOS is really NOS) and no such thing as a filter onboard (and no means no here). So be aware, I talk about 1:1 reproduction of what is in the file, which actually is the wrong approach (because the file is already wrong) ...

 

Back to the parameters within NOS/OS, filtering/filterless -and again try to imagine FFTs with a complete mess, less mess or without mess- this "mess" can be at various places. For instance, upsampling alone takes away the mess in the audio band (aliasing), which at least is something my brain tells me I should hear it when in there. Now, leaving out the filtering, only allows my amps to get stressed, *if* they do at all.

This is not so difficult to underatand; when my amps can cope, I can do without filtering, *that* phenomenon possibly being the harmful thing to good sound. One thing - doing it that (simple) way, leaves me with linear interpolation, that by itself implying harmonic distortion.

 

Yes ?

 

No. No, because it is not that black and white, and everything is relative. For fun, try to think like this :

When linear interpolation brings me HD, what brings doing nothing ? aha, MORE HD. But wait a minute, doing nothing already sounded good (read : better than OS/Filter). Now tell me why upsampling a bit in the wrong way should not be better than doing nothing ...

 

Possibly you may say "yeah, well, uhmm, could be", and for theories I would agree. But here is nature again, and at interpolating in a linear fasion you'd molest nature. The nature of doing nothing at all is more nature, and this is (IMO) because the stepping (more rough now than with upsampling) within itself is consistent. Thus, breaking in on that, and cross it with creating a step in the middle while nature dictates that to be at 90% or whatever, is harmful and audible.

 

To understand the above better, it is important to know and see that sound doesn't come to you from a single small one cycle wave, but it is composed of the long term interaction of many of those repeating cycles, where "repeating" spreads over several of these individual cycles, before you have a repeating pattern. It is these repeating patterns we hear as "a sound". It may go too far to state that such a repeating pattern may reconstruct the original wave, but I think I'm not far off at suggesting it.

 

For those still here (what page I'm at ?), I often refer to dithering;

Those who understand the working of good dithering, may be more in touch with this thinking. Thus, what is good dithering ? this is the most random adding noise as possible, and when random is not really random, well, you'll perceive a tone. Think the other way around with NOS/Filterless; the long term disrupted wave, may not come out so disrupted as we expect. Look at a scope;

 

If you would be able to capture the disrupted wave in a complete standing still picture, *there* you would have the wrong sound you will be listening to. A disformed sine or whatever it intended to be. But, at sampling the wave at the sample rate (on the scope), you won't see this disformed wave. You will see a vague sine again. Yes, the "vague" will denote harmonics, the same the FFT will show you, but the general outcome is the sine, because that is the main line (it was like that originally, so it will stay like that looking at the long term).

 

Now what ?

Now it may come down to how audible the (false) harmonics are, and I have always been thinking that where the false harmonics are created by the exact same wave, meaning at the exact same time, the main "sound" from it may overvoice the harmonics. This too is something you can see from the FFT although it may be some harder to accept that you are looking to one frequency only, and how to know that the loudest frequency shown (the one test tone you setup) does or does not overvoice its false harmonics.

Of course, you could just listen to that tone, and undoubtedly you will hear the "square" harmonics (uhmm, if you have an NOS DAC haha), but this is a test tone, and in real life a. such a setup sine doesn't exist in the first place, b. it is not about continues tones, c. it will be in the middle of sheer infinit other frequencies.

 

Again back to the parameters, have a tad of upsampling (2 x is enough) and you're done with the roll off (which otherwise starts at 5KHz). This is always good.

Second, interpolate such that nature will not be violated. I have said it elsewhere, it can well be that just RANDOM interpolation is a good solution here. Easy to try (for me). Just avoid creating unnatural patterns.

Lastly, use amplifiers which can cope with the high frequency stuff left.

Done.

Maybe.

 

Of course, dictating special amplifiers is not much realistic, but I would do it (ehh, try to find them).

 

Now, all of these elements may flaw hugely. If they do, apart from me seeing it all wrongly -and which most probably is just the case- I again would be stuck with NOS still sounding better without good reason.

 

AND THIS CAN NOT BE SO

 

And therefore - no matter how good and consistent theories for OS and all are - there must be something wrong with it.

 

Or call me deaf.

Peter