Theoretical benefits to hi-res if only max 20kHz signal?

[Wombat]

Thanks for your very interesting post. So we can conclude from there:
1) there is a benefit of using 48khz instead of 44,1khz because of the reconstruction filter
2) fanatics of NOS DAC will benefit from hires even up to 384khz because the higher the fs, the flatter the sin x/x lowpass in audible frequency band...

Then what about a NOS DAC without output filtering (using ear natural lowpass filtering): this would get rid of pre-ringing filtering artefacts. Combined with ultra hires, wouldn't it be close to the perfect sound?

To 1) there is still this theoretical filterproblem that even in real-world isn´t. It is only a problem for many because it is there theoretical. Can you hear the different filters acting around 20kHz?

To 2) i never listened to a NOS DAC and many seem to love it while others find even the best NOS DAC sounds horrible. The music it spits out without sinc filter is indeed horrible to measure and for sure will benefit of higher sampling frequencies most. Now this is a case to wonder, even when the output is technicaly speaking pretty less (just a few steps) precise it still may be sounding fantastic.

[Phil Leigh]

Thanks for your very interesting post. So we can conclude from there:
1) there is a benefit of using 48khz instead of 44,1khz because of the reconstruction filter
2) fanatics of NOS DAC will benefit from hires even up to 384khz because the higher the fs, the flatter the sin x/x lowpass in audible frequency band...

Then what about a NOS DAC without output filtering (using ear natural lowpass filtering): this would get rid of pre-ringing filtering artefacts. Combined with ultra hires, wouldn't it be close to the perfect sound?

1) IME Only very few people can hear the small difference between 44.1 and 48. I used to be able to 15 years ago when DAT machines were popular but I can't anymore. Even then it was a tiny tiny difference on my top end Sony and I couldn't in all honesty say it was really down to the filtering. It might just have been a better DAC.
2) All of the NOS DAC's I've heard sound uniformly dreadful in the upper mid and top end to me
YMMVof course.

[TheOctavist]

"NOS DACS" are a ruse.

http://www.head-fi.org/t/437340/any-...5#post_5905139

[Mnyb]

"NOS DACS" are a ruse.

http://www.head-fi.org/t/437340/any-...5#post_5905139

But they do sound "different"

Amazing if the punter finds that all dac's sound's similar and then finds the nos dac very different .

how on earth do you get to logic defying conclusion that the NOS dac is an improvement ???

[adamdea]

2) fanatics of NOS DAC will benefit from hires even up to 384khz because the higher the fs, the flatter the sin x/x lowpass in audible frequency band...

Then what about a NOS DAC without output filtering (using ear natural lowpass filtering): this would get rid of pre-ringing filtering artefacts. Combined with ultra hires, wouldn't it be close to the perfect sound?

There is an element of truth in it if you are determined to see a point in NOS dacs, but it is still utterly illogical.
All you are saying is that NOS dacs will work better if oversampling (in the sense of increasing the sample rate from the native rate of the data being decoded) is unnecessary.
Oversampling is unnecessary if the native rate of the data is already at the rate you would have oversampled to.

A nos dac with data fs = 192k will presumably behave in the audible band just like a 4x oversampling dac weith 44.1 data but no filter.

So now with no reconstrction filter and Fs = 192k or whatever we only have to aorry about images at 96Khz plus. This does not remove the problem that some downstream electronics will still not like the spuria at 96Khz. I am also not sure whether the combination of intermodulation distortion and imaging which causes NOS dacs to have such a poor noise floor at 44.1 will also affects them at fs =192 or whatver.

Anyway there is now no reason *not* to have a reconstruction filter. (when i say reason I mean the spurious "reason" which justified nos dacs in the first place). Even if there are audible pre-ringing artefacts at fs=44.1 surely even an audiophile wouldn't think they existed at fs=192k (which you will note is Bruno Putzeys point about apodising filters) .

So at higher sampling rates the nos dac makes no sense whatsoever

[JohnSwenson]

I have a lot to say on this subject, but its too much for one post so I'll start with the topic of NOS DACs since its being discussed right now.

First off a bit of background, I have been building DACs for quite a few years, I have built at least 30 different DACs with many different DAC chips, filters, output stages etc. I have tried hard to come up with at least some correlations between differences in what I have been building and what I have heard, this hasn't been easy, there are so many variables it's hard to narrow things down.

And yes I found there ARE differences in sound, they do not all sound the same. I have done quite a few blind tests with other people listening, they were at least single blind, sometimes double blind. The blind tests do prove that there are differences, but preferences as to what people preferred went all over the place. So the following is going to be my own personal preferences, which are certain to be different than other people.

Before going into details of my NOS experience let me say that I am an electronics engineer, I am well versed in sampling theory and have a moderate acquaintance with DSP, but by no means the worlds greatest living expert.

When I first started Building DACs I was firmly in the oversampling is a great thing camp, it radically improves upon the old style brickwall analog filter etc. Then I heard about NOS DACs and decided to build a couple and see what they sound like. I was sure it was going to be terrible. But low and behold that is not what happened. I found the sound was significantly improved in some areas and significantly degraded in others. This was a big surprise, I was not expecting the improvement at all. At this point I had no clue WHY it sounded better.

I tried several NOS DAC designs of my own and several from other people. What I found was that a lot of the NOS DAC designs "out there" were seriously flawed in many ways, it was obvious the people designing them really didn't know what they were doing. Frequently I could make them sound much better by some simple changes.

Even with some of these bad design choices they all definitely have a "NOS sound". To me the improvement part is an increase in subtle detail, being able to hear subtle details of the acoustic environment, being able to hear subtle nuances in performance that purvey emotional content better. At the same time, the sound is "dirty", its rough around the edges. When you go back to oversampling it sounds much cleaner, but also flat and boring in comparison.

My supposition is that the people that like NOS DACs are willing to ignore the bad parts in order to gain the good parts. It might also have to do with age, young people with good hearing may be bothered much more by the "dirty sound" (presumably the infamous aliasing).

I then spent a couple years trying to find out why the NOS DACs sound better. I won't go into the full journey here, but what I eventually concluded was that it was the digital filters themselves getting in the way. I could build my own digital filters that kept the good qualities of the NOS sound but without the "dirt". I wasn't doing anything special with my filters, just a good proper implementation in an FPGA. The only thing I can think of as to why the commercial ones do not sound so good is that they are NOT properly implemented.

In order to properly do a brickwall filter for 44.1 takes a fair amount of hardware resources, my guess is that the manufacturers are cutting corners to save money. They are doing just barely enough to get decent numbers in the data sheets. In particular I'm leaning towards the practice of cascading several small filters rather than using one properly implemented large filter. Looking at data sheet plots of filter performance you can frequently see this cascading of filters.

An interesting occurrence happened early in my DAC quest (before my first NOS DAC) I had an inexpensive DVD player and decided to see if I could get it to sound better. I did a number of analog and PS improvements which improved things significantly, but I also noticed that the DAC chip had a "slow rolloff" filter setting as well as the default brickwall filter. So I built a little board that let me reprogram the registers in the DAC chip so I could set the filter type. I found that I liked the slow rolloff much better. In blind tests most people liked the slow better, but several hated it. Looking at the data sheet plots I found that the slow filter was implemented as a single filter but the brickwall was three cascaded filters.

I have done similar tests with many other DAC chips and with my own digital filters and its turning out to be a fairly decent correlation that the biggest impact is not the filter function itself but if its implemented as cascaded filters. Filter functions DO make a difference, but if they are all implemented as cascaded filters they all don't sound so great. With filters implemented as single proper filters (enough internal bits for the number of taps and enough bits for the coefficients) differences in filter functions CAN be heard but they are not very large. Just getting the filters implemented properly is the biggie.

This brings up the issue of software upsampling and NOS DACs. First off NOS does not mean no filter, just not a digital filter, you can still put an analog filter on a NOS DAC. If the builtin digital filters are the problem, it seems that a good NOS DAC playing upsampled files that were generated with a properly implemented software filter should provide good sound. And my experience is that indeed it does, especially if you put a 2 or 3 pole analog filter after the NOS DAC to get rid of residual high frequency noise. Note this has to be a GOOD NOS DAC, not one of those cheap 16 bit ones from people that have no clue what they are doing.

There are a lot of people that are doing software upsampling and feeding the results into soundcards and external DACs that I think are trying to do the same sort of thing, but the data is still going through the compromised digital filter in the DAC chip. It would be much better if they fed it through a good NOS DAC.

An interesting side bar on this is an early experiment I did. I had been reading about people that stacked 8 1543 DAC chips, I tried this in two different ways. One group of people spread the chips out on a board (see the picture somewhere up in this thread), but others actually stacked the DAC chips on top of each other. I tried both and found the stacked on top of each other approach sounded much better. Note this was EXACTLY the same circuit, just a different physical layout of the chips. The difference was that the stacked chips got HOT. Doing so cut off the airflow so they got to much higher temperatures.

I hypothesized that it was this higher temperature rather better linearity or lower noise that made the improvement. I decided to test this by gluing a power resistor to a single chip and pumping DC through it to raise the temperature. I added a thermocouple so I could check the temperature of the chip (well the temperature of the case, not the actual chip). I then very slowly raised the temperature of the chip and low and behold it sounded way better as the temperature went up high. (still not all that great, the single chip by itself is a pretty bad sounding DAC chip) So all that theory that it sounded better because of the increased bit depth because of stacked chips was hooey, it sounded better because it got HOT. BTW the one with the 8 chips spread out on the board sounded worse than a single chip, that was just a bad idea.

Things HAVE been getting better. The latest crop of chips seem to not have as bad digital implementations as previous ones (with the decrease in cost of compute hardware, its probably cheaper to just do it right than spend the money trying to develop creative corner cutting). That doesn't mean they are perfect. Every one I have tried I have been able to make sound better by disabling the internal digital filter and using a properly implemented external filter. The only ones I have found that seem to do a pretty good job of their internal filters are the Sabre chips.

Well there you have it, some of my exploration of NOS DACs.

John S.

[Phil Leigh]

So John, are you saying that on 44.1 material you prefer a NOS with a good analogue filter set for 22Khz vs an ASRC DAC running internally at 384 or 768 with an appropriate digital filter?

Just curious.

That "dirt" you refer to is exactly what I don't like about all of the NOS DAC's I've heard. Of course this doesn't mean ALL NOS DAC's are empirically bad... simply that I personally haven't heard what I consider to be a good one.

Interesting about the heat phenomena. The heat must be causing the Johnson noise to rise inside the DAC, no:? - maybe the noise is having a "dithering" effect?
regards
Phil

[Archimago]

Thanks for sharing your experience with DAC's John!

Over the years I have played with a cheap DIY 4-chip TDA1543 design similar to the stuff on eBay as well as a more expensive Mhdt Constantine (TDA1545 I think). Can't say I was enamoured with the sound... I found the highs a bit too rolled off for my taste so happily went back to the Transporter sound.

Wondering if there was a commercially available NOS DAC you think represents a good design. I don't have time anymore to fool around with DIY's.

As for the slow roll-off filter, any opinions on the Transporter's AK4396 slow-roll vs. standard filter?

[adamdea]

This brings up the issue of software upsampling and NOS DACs. First off NOS does not mean no filter, just not a digital filter, you can still put an analog filter on a NOS DAC. If the builtin digital filters are the problem, it seems that a good NOS DAC playing upsampled files that were generated with a properly implemented software filter should provide good sound. And my experience is that indeed it does, especially if you put a 2 or 3 pole analog filter after the NOS DAC to get rid of residual high frequency noise. Note this has to be a GOOD NOS DAC, not one of those cheap 16 bit ones from people that have no clue what they are doing.

There are a lot of people that are doing software upsampling and feeding the results into soundcards and external DACs that I think are trying to do the same sort of thing, but the data is still going through the compromised digital filter in the DAC chip. It would be much better if they fed it through a good NOS DAC.

John this is fascinating stuff, but the terminology is getting confusing to me. NOS literally stands for non oversampling (?) - but tends to be associated with either no filter or analogue output filter.

Forgive my ignorance but how do "upsampled files that were generated with a properly implemented software filter" differ from the up/oversampling which would go on in an ordinary OS dac? It seems to me that you clearly have both oversampling and a digital filter here- your just doing it in software.


For the sake of simplifying Inguz all my files are (courtesy of Phil's tweak) upsampled by sox to 96 kHz before streaming to my dac. Is this what you have in mind? I can't see how you can go beyond 96kHz via a squeezebox

If the processing power in the dac is the limiting factor I see why it might be better to finesse this in software. I hope I am not making a fool of myself by I can't see how this differs in principle from the classic OS dac with a final analog filter to remove images above half the oversampling frequency. Would this be NOS in the sense that Kusunoki meant it http://www.sakurasystems.com/articles/Kusunoki.html

Presumably your proposal will only work with a multi bit dac which
natively works at the upsampled frequency not a delta sigma one as those will presumably inevitably have their own filters.

[I fear I am betraying my ignorance but I have never been able to grasp the difference between upsampling and oversampling]

[JohnSwenson]

Thanks for sharing your experience with DAC's John!

Over the years I have played with a cheap DIY 4-chip TDA1543 design similar to the stuff on eBay as well as a more expensive Mhdt Constantine (TDA1545 I think). Can't say I was enamoured with the sound... I found the highs a bit too rolled off for my taste so happily went back to the Transporter sound.

Wondering if there was a commercially available NOS DAC you think represents a good design. I don't have time anymore to fool around with DIY's.

As for the slow roll-off filter, any opinions on the Transporter's AK4396 slow-roll vs. standard filter?

If I were going to buy a NOS DAC today I would get the Audio-gd DAC-19, it uses a pair of 1704K chips (my personal favorites). It also does what I do, it implements its own digital filters in an FPGA (strangely enough exactly the same FPGA chip I use in my latest DAC). You can also set it to NOS mode if you desire.

Its not the cheapest DAC nor the most expensive. I think it does a good job all around (except for maybe the USB input).

I have not had a change to listen to the Transporter's slow rolloff mode so I can't comment on that.

John S.