NOS DAC - Marketing BS? - Page 3

I would think that Vinnie of Red Wine Audio would be the best person to answer this. However it is my understanding that MOT's are not supposed to answer questions about their products nor defend them outside of the paid for ads/threads they sponsor in another area of this forum.

no question about it.

I wouldn't call myself an expert when it comes to this topic, but I quite like my Paradisea 3 DAC compared to others I've auditioned... then again, I'm one that will take smooth, warm, and natural to detailed/revealing any day of the week.

I never could quite understand the notion that extreme detail is always the best way to go... call me crazy, but I actually prefer my old '89 Sony TV compared to the majority of high-definition LCD TVs widely available today; the new back-lit LEDs could change me soon...

Isn't the whole idea behind over/upsampling is to fool the mind by adding information that isn't inherently there? Whether it be additional pixels or tones, my belief is that heavy upsampling will always decrease quality (or at least, the mind will catch-on and reveal the magician’s trick). Anyone who has worked with poor means of upsampling photos or videos will know what I'm referring too... now, starting with a good source, SACDs, is a different story because the extra info has been recorded and doesn't need to be created.

Just my two cents... please tell me why and how I'm wrong... yes, I want to hear your opinions!

This thread has my head spinning a little. I have an Ack dAck 2.0 which is supposed to be filterless and non oversampling. I have always liked the sound of this DAC and have had more than a couple of people remark on how good it sounded at meets, surprisingly so, considering that I have mid-fi gear at best.

Now I'm considering a new DAC because I want to be able to tap into the 1000s of songs on my computer instead of always having to play CDs and the Ack dAck has no USB connection. I was considering either getting a comparable NOS DAC with USB input or some kind of USB to SP/DIF converter. After reading this thread I may have to reconsider a NOS DAC.

I am curious as to what the designers of the Ack dAck and Red Wine Audio DACs might have to say about this, although I understand, probably not in this thread.

Um, okay.

Basically every manufacturer has reasons why their product is the best engineered with the best engineering approach.

As people are hinting at, you get the sense that those manufacturing NOS dacs would have their arguments to put forth and would be able to carry on this conversation with others who took a different approach like Lavry.

Lavry's argument seems a bit too dogmatic. I wouldn't say argument over or case closed.

Personally I have heard his dac and compared to the red wine, I strongly prefer the latter. Regardless of whether supposed NOS is truly NOS or whether NOS is technically possible or represents merely an approach with a technical meaning commonly misused. The red wine unit simply sounded much, much better to me. Part of that is preference no doubt.

Maybe my ears are poor, doesn't matter, the red wine created an experience i enjoyed.

But assuming my ears aren't poor, and assuming that is true for some of the other folks, including reviewers, who like the red wine, and assuming that we're not being easily deceived or simply being silly and buying an SUV, or as Mr. Lavry notes, simply not unable to hear high frequencies and duped into thinking that something sounds good, then what? Could it be that there is a technical explanation that accounts for our preferences.

I don't concede that the argument is closed. But even if one could prove that an engineering approach should sound better than an alternative, it doesn't mean that it will.

Here are some OBJECTIVE FACTS, not opinions and wishful thinking:

Click on: http://www.lavryengineering.com/white_papers/sample.pdf

Then look at middle of page 2. There is a plot titled “Filtering of sampled signals”. The 4 solid tones under 22KHz are not filtered. But the tones presented by the “dotted lines” show what would happen when you do not filter the DA output. I only showed the range up to 176.4KHz (4fs), but the tone energy keeps going way up there, and the amplitude is nearly as high as the audio!

So you want to feed all that high frequency right along with your audio to the next device? That is asking for problems! I already wrote a long post here, and apparently some folks did not bother to read it.

Then, go to page 3 and look at the plot called SIN(X)/X plots for X2, X4, X8 and X16 (the bottom plot). These plots represent your frequency response (amplitude vs. frequency). Note that I DID NOT EVEN BOTHER to show a NOS which is fs. I started at X2fs, and even at X2fs you lose around .75dB by the time you get to at 20KHz. At fs (NOS) the loss is around 1.5dB. In short, you lose amplitude as you go to higher audible frequencies.

You can call my argument "dogmatic" or "catmatic" :-). You can declare "case not closed". There is no way that a credible professional engineer will dispute what I said. All they will ever say is the "standard" subjective argument (it sound great). Such statements do not rise up to the task of contrudicting the physics and the electronics. You (or anyone else) can not do that because I state iron clad facts. Facts are not a subject for agreements, disagreements or opinions. There is a sun, Ohms law works and 1+1=2.

Perhaps I am wasting my time here. But I can tell you guys that there is NO SELF RESPECTING ENGINEER that can dispute those plots. This is not at all about opinions, or about what one would like to believe. This is about cut and dry FACTS.

I am amazed at the fact that anyone can believe that TI, Analog Devices, AKM as well as virtually all the gear used by the professional community are all standing by flawed fundamental thinking for the last 20 years. I am amazed at the fact that folks are willing to believe that credible manufacturers of ICs and gear would prefer to ADD circuits (add cost, space, heat…) and all that for a lower performance. This is crazy!

Up-sampling and oversampling are major steps forward, and filtering is there for a reason. They solve the problems I pointed out. Filtering of NOS is a real problem, an unfiltered NOS is even worse. Ups-sampling a DA solves the response problem and enables good filtering of the unwanted high frequency (image energy).

Regards
Dan Lavry
Lavry Engineering

Very well written Dan...PLease don't feel dissuaded by the skeptical outlook of folks here. It takes time to make people see the technical facts for waht they are.

NOS just needs to be killed..and fast. And taht quote of "Music is happening" and it "sounds like music" is utter nonsense.

I have had a NOS DAC for 2 years before i saw the light. It was a filterless and unbuffered output as well and a passive I/V to boot. After listening to a few DACs that were OS side by side with the NOS dac it became quite clear how far ahead the OS DACs are in terms of quality. wasn't all that surprising considering that I was listening to a DAC chip and implementation that was used in mass market electronics back in the 80s.

I would really like to see Dan Lavry, Filburt, cetoole and others take us through a techinical lesson on the science of D to A conversion. At least for those of us interested in the technical side of this would totally appreciate learning more aout this. It would have to be objective of course with plain science and enggineering to back up the arguments..no subjective BS.

I appreciate your frank and honest discussion of the facts. Hopefully all of us here can seek to learn rather than argue.

I do have a question though, for sake of learning and discussion. Let's suppose that you have a NOS DAC with a filter that loses, for example, 6 dB at 20 Khz and 4 dB at 16 Khz, and doesn't lose a significant amount of detail. Then let's suppose that you placed a linear phase EQ after the filter to boost those frequencies back up? Would that have a positive effect on the filtering problem? (I'm not suggesting this is a better idea than oversampling. I am just curious what the implications of this would be.)

As another note, I noticed that on the Valab discussion thread, people have tried the Valab without filters, and the consensus seems to be that it "opens up the sound" but "causes high frequency noise on a 'scope". So I think it's possible that the Valab really is an NOS DAC with significant filtering.

Which DAC? There is a big difference between DA10 and DA924

Fwiw, I prefer bypassed mode on my Stello (= disable oversampling?)

I really do not have the time for a well organized answer. I want to do some real work. But here we go:

First let me note that there are two causes for loss of high frequency in a NOS.

1. The sin(X)/X curve (we call in sinc function) which is what I showed in my paper (see the previous post).

2. The anti imaging analog filter, and even a real poor (low order filter) will cause much additional loss of higher frequencies.

Virtually all DA's in the old days of NOS did include some filtering. Having no filtering is really against good engineering practices.

One may find some SPECIFIC gear; say a power amp that can handle a lot of high frequency with less impact. But then there is a lot of SPECIFIC GEAR out there that will cause a lot of distortions when presented with the high frequency energy. It is always worse without the filter and gear should be designed to remove the high frequency image energy.

Now, with up-sampling, one moves the filter cutoff to much higher frequencies, so the -3dB point (the bandwidth) of the analog filter is way up, and at 20KHz there is no attenuation (or say +/-.1dB flatness response instead of worse then -3dB). ADD TO THAT the sinc response, which with up-sampling to say only X16 is virtually non issue at 20KHz, and with a NOS is ADDITIONAL 1.5dB loss.

I have no desire to enter into a sonic argument about removing of the filter from a NOS in a specific implementation while driving specific gear. With a filter, you suffer less from the high frequency energy but have worse response. Without the filter, you have less worse response (still very poor) but more high frequency image energy issue. Each case sucks in a different way. Do you want to have a headache or a back pain? One will be better then the other, but you can avoid both.

You suggested to EQ the impact of NOS with a high frequency boost. Indeed, we used to do so, to the best of our ability, and we did so by means of a lot of DSP. Analog EQ did not cut it. Why? Because, the shape of all analog filtering is "bound" by poles and zeros. In simpler terms, one can not come up with ANY shape of analog EQ curve, there are "restrictions" in the shape of the curve. It is unfortunate that the curvature of a sinc function (sin(X)/X) is very different then anything that you can approximate with analog filters (poles and zeros). The analogy is trying to cover an ellipse with a circle. It just does not match well. Trying to do so even without consideration to phase is a losing battle. You can try your best, but you will never come anywhere near the flat response of an up sampled DA.

Some of the gear makers did put a lot of DSP (digital processing) into trying to compensate for the sinc issue. And with enough DSP you can get better results. You can find a lot of literature about it in DSP books. I do not think that most NOS DA designers do it, this was rarely done (I did it).

But when you correct for say 6dB at 20KHz with a high frequency EQ, you are also in fact BOOSTING the noise floor at 20KHz by 6dB. In other words, you are sacrificing a bit at high frequencies, and your 16 bit CD is no longer 16 bits at high frequencies (it is 15 bits). Also, when you are boosting the high frequency energy around 20KHz, you can not just make a high pass boost. You need to roll it off by the time you get to 22KHz (Nyquist), or else you are in fact going AGAINST the required filtering (analog anti imaging filter). Such "shelve filter" - boost by 6dB at 20KHz and back to 0dB is impractical. Remember, a single pole is 6dB PER OCTAVE, so 6dB from 20-22KHz calls for around 10 poles just to return it back to “neutral”. And all that for a very rough and real poor approximation. It will take at least 20 precision resistors and caps, plus 5 OPamps per channel, just for the filter correction back to neutralize the filter between 20-22KHz, and the results are going to be terrible.

And all that can go away with just a little bit of oversampling.

You asked a technical question, so I needed to answer it in technical terms. I am sure some readers will be lost...

Regards
Dan Lavry
Lavry Engineering

Thank you. I really do appreciate you taking the time to give a technical answer. I aspire to actually understand this stuff.

I'm not as good at explaining as Dan, and I think he's answered the other questions, but not yours. Over/upsampling isn't done to try to reconstruct or add information. That would be interpolation.

Over/upsampling is more like using more decimal places when you are doing your math to get more accurate results. You haven't changed any values, or added any new values. (Even though it sort of sounds like you are.)

This is useful for interpolation, like when you enlarge a photo and ask Photoshop to smooth out the new pixels. But it is also useful if you are filtering a signal, because the "more decimal places" lets your filter function be much more accurate, precise, and targeted.

OK, my 25 cents, skip to the next post if you are close-minded:

This is a complicated subject and everyone has their own perspective, so no single explanation is going to make sense to everyone.

If you have studied communications theory or differential equations, then you have a head-start, otherwise it's going to be ... different.

I DO have a background in music, mathematics and electronics, and I HAVE studied Dan's white papers and his posts and I agree with everything I've read. I know that some of you will find these things incomprehensible, even suspicious; but I believe that dedicated audio engineers have a greater "motivation" than marketing tactics.

No no and no! The idea is NOT to fool the mind, nor is it to add information that is not there! Where do you get such miss-information?

When you sample a waveform, you "take snapshots" at a certain rate. In the case of CD (for example), you take 44100 samples per second. Each of the samples ("snap shots") represents a value at one point in time. So the question is what happens between the samples. The original waveform (voltage changing over time) exists at ALL times, but the sample points are only values at the exact sample time.

In order to reconstruct the original wave from the sample "points", one needs to "fill in" the "gaps" between the sample points. One does not connect "the dots" with a straight line. We need to do better then that. In fact we can do much better then that. In theory, if we sample just fast enough (Nyquist rule) we can in theory reconstruct the original from the sampled data PERFECTLY. There is no need to sample any faster then a little over twice the audible bandwidth, and one can PERFECTLY RECOVER the original waveform. In theory, for 20KHz audio, 44.1KHz sampling is fast enough.

So how does it work in practice? Look at the specifications of a good DA. When you see .001% distortions, it means that the DA output is within .001% off from the ideal. When you see 100dB, it means the accuracy is one part in 100000 (ten part per million). I did write "Sampling Theory" paper, and it is too technical for some, but the facts are still facts.

So the issue is about how good or poor the implementation is. The fact is, a good up sampler computes the intermediary point with great precision. Say you up sample by X2, then you are in fact computing the additional values right at the middle between each pair of samples, and you do so with great precision. It is very possible to compute the values to better then a part per million accuracy.

In theory, the outcome is the SAME as having taken those additional “in between samples”. The process of "filling the gaps" does involve filtering which is in fact the main part of interpolation. Analog filtering is one way to "connect the dots" into the original shape. Up-sampling is not all that different in theory. Up-sampling does some of the filtering in the digital world, and then you still end up with doing the rest of the "connecting the dots" with an analog filter. But with the digital up-sampling, the analog needs to do only a part of the task, the easier part.

With a NOS, there are huge practical problems - one can not come up with a good enough analog filter. With up-sampling (filtering in digital part of the way) a very good analog filter is very practical. I understand that the concept of connecting the dots to get back the original is difficult to grasp. I also understand that the notion of being able to compute points in between precisely is difficult to grasps. Sorry, but there is no substitute to learning, and sadly, "common sense" and "street smarts" lead the novice to wrong conclusions. The analogy to video and pixels is completely out of place.

I saw many folks that look at a sine cycle of say 20KHz sine wave, and they just do not want to believe that given such a cycle with 3 points, one can connect the "missing part of the wave". Well the electronics does not look at one cycle with 3 points at a time. An quality FIR up sampling interpulator looks many hundreds of points at a time (thus hundreds of 20KHz cycles at a time) to compute each and every new sample value. At 44100Hz (CD sample rate), one gets to "look at" over 4410 samples in 0.1 second, or 441 samples in 0.01 second time, and there is a lot more information there then one cycle and 3 samples... The end result is that each interpulated (up sampled value) falls on near the exact value of where it needs be (to say 10 parts per million or lower accuracy in a good design)

Let me just say that the perfect interpolator and the perfect brick wall filter are the SAME THING. Filtering is what gets one to connect the dots and end up with the ORIGINAL WAVEFORM. The rest is about how to implement that filtering, and NOS is at a great disadvantage. A good up-sampler does NOT add anything to fool the ear. It does a good job of RECONSTRUCTING the ORIGINAL wave (to end up with duplicate of the analog wave BEFORE it was ever sampled).

The only "fooling of the ear" I am aware of is by those spouting all sorts of made up garbage and miss information.

Dan Lavry
Lavry Engineering

now THAT post makes it much clearer!