Why NOS actually may make sense.

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
You may be interested in AES paper by Rob Stuart (2014).
The sharper the filter slope the worse the distortion of transient timing. It's known that the attack phase of sounds is the most important phase of any sound.

I included the pdf. It is a must-read.

Thanks for the article. It is certainly interesting, I'll definitely look up some of their references about sampling of non-band-limited signals.

It's a pity that there are some statements in it that really make no sense at all. For example, "It is now widely accepted that one key benefit of higher sample rates isn't conveying spectral information beyond human hearing, but the opportunity to tackle the dispersive properties of brick-wall filtering." If you can't hear anything above 20 kHz or 24 kHz or whatever under any circumstance, there is by definition no way you can hear the effect of an ideal low-pass at 20 kHz or 24 kHz, no matter how dispersive it may be. If you do hear the pre-ringing, that means that the presence or absence of signal above the cut-off frequency affects your perception, in other words, that you can hear the effect of signal above the cut-off frequency, at least when played back in combination with the rest of the signal.

Hans van Maanen was really much clearer about this in his article in Linear Audio volume 5. He is a firm believer in the audibility of high-frequency time smearing, and he pointed out that tests with sine waves do not necessarily fully characterize the human auditory system because it is not linear and time invariant. Hence, it could well be that you hear something at frequencies that are inaudible when you listen to sine waves. I'm not sure whether he is right or wrong about that, but at least his story was clear. Then again, Linear Audio is not peer reviewed, so he didn't have to come up with compromise texts to please the reviewers.

By the way, the Rob Stuart and Peter Craven article has some overlap with R. Lagadec and T. G. Stockham, "Dispersive models for A-to-D and D-to-A conversion systems", AES preprint 2097, 1984. That article also contains an interesting discussion about passband ripple of linear phase filters.
 
I think their point is that with fairly high sample rates (96 kHz) and music of which the spectrum already rolls off naturally, you can get away with a relatively smooth filter and still keep the aliasing products well below the noise floor. Of course that means that the choice of anti-aliasing filter becomes dependent on the characteristics of the music; if you would want to record a church organ, you would need to switch to a steeper filter when the bats wake up.

Yes I get that - however in terms of playing music back I'm stuck with source material which is by and large all 44k1 sample rate. Although the acoustic sources are covered in that paper and do roll off naturally, I didn't see any consideration of the accidental pick-up of out-of-band components. If you go over to Inside Your High-Res Music: Testing 1 2 3!!! | Page 2 | Audio Science Review (ASR) Forum you'll see spectral examination of some higher rate recorded music where it looks like they couldn't keep out some ultrasonic components.


I think the point here is that errors may accumulate over the entire signal chain. Many people would not object against a piece of equipment that rolls off by 1 dB at 20 kHz, but cascade many of them (microphone, microphone preamplifier, ADC, DAC, preamplifier, power amplifier, loudspeaker) and it becomes a different story.
Yes I accept that as a point, however I find when someone exaggerates to make a point that it loses persuasiveness. When I was designing line level stages in my day job I took -0.1dB @ 20kHz as my benchmark.

I noted with interest the paper suggesting a triangular shape as DAC output. How to achieve that in hardware though - the paper gave no indication? I recall a paper by Hawksford where he proposed a raised-cosine output.
 
Last edited:
Disabled Account
Joined 2015
OK, oon_the_kid. You talked about the reconstruction filter guesses the in between values. I wouldn't say it's guessing. But the FIR algoritm is designed to just "reconstruct" every sine wave it sees. That's why it looks so good at the scope.
But in order to make these nice sine waves it introduces a "momentum" in the interpolated "in between" samples. So if the sine wave rapidly changes it's amplitude, it doesn't really reacts fast enough.
So in the real world with complex signals that comes from normal music, something is definitively happening with the signal.

OK, I think it's nice to look at the differences between NOS and OS this way.

The OS DAC introduces this momentum effect in the audio band ( even if it's very tiny ) and hoists up all the amplitude related dist far up in the spectrum ( which we humans cant really hear )
On the other hand, the NOS DAC doesn't do anything with the audio band, but introduces a lot of overtones around the sampling frequency. And those overtones are not really time related, they are appearing instantly with no "smearing " effect.

So the OS DAC creates time related dist in the audio band, while the NOS DAC creates amplitude related dist in an area that we humans cant' really hear. Which seems best?

Personally I modified my DAC to NOS recently and I haven't decided yet which I prefer. But thankfully, the the shifting between NOS and OS is accomplished by software - a small button on the front of the DAC.
But this evaluation will take time. I'm not very much for A/B tests - an impression usually takes weeks or months to settle.
 
OK, I think it's nice to look at the differences between NOS and OS this way.

The OS DAC introduces this momentum effect in the audio band ( even if it's very tiny ) and hoists up all the amplitude related dist far up in the spectrum ( which we humans cant really hear )
On the other hand, the NOS DAC doesn't do anything with the audio band, but introduces a lot of overtones around the sampling frequency. And those overtones are not really time related, they are appearing instantly with no "smearing " effect.

So the OS DAC creates time related dist in the audio band, while the NOS DAC creates amplitude related dist in an area that we humans cant' really hear. Which seems best?

Actually your non-oversampling DAC with staircase waveform output messes up the signal in the audio band by its zeroth-order hold function. An oversampling DAC or a non-oversampling DAC with peaking analogue filter may reproduce the audio band part of the signal correctly.
 
Further to what DF96 and Ken Newton wrote earlier, I managed to find a bit of time to try and understand the data hold, zero order hold, first-order hold, reconstruction of the original signal from a sampled signal and aliasing a bit better. My (selfish/egoistic) reason for doing this was to prove to myself that what I hear from my “true” (no filters of any kind) NOS DAC does sound good because there might be a scientific proof for that. It’s not that I really needed a proof; however, this thread seems to be attracting nice, civilized, sensible comments that motivated me to read and discover more. So, I stumbled upon another outstanding paper. Very concise. It seems to be written by Richard Tymerski (© Portland State University 2017). The web page and the pdf is here: http://web.cecs.pdx.edu/~tymerski/ece452/Chapter3.pdf

One of (not many!) conclusions/summaries: “The ZOH does not approximate an ideal low pass filter very well. Higher order holds do a better job but are more complex and have more time delay, which reduces stability margin. So ZOH’s get used a lot” (page 10).

It seems that all filters remove more or less, aliasing frequencies with some degree of accuracy and reveal the original signal (as Ken pointed out quite nicely – thanks); however, they also affect the critical time-domain relationship between what should arrive at our ears at a particular moment in time, and what actually does arrive at our ears slightly shifted in time domain. With a complex material and bad filters, the time-domain component AND the amplitude of the signal could be wrong. Our ears are extremely sensitive to that time-domain component. We are starting to see a positive attitude from DAC designers lately, towards implementing a very low phase noise crystal oscillators, specified at 1Hz or even 0.1Hz. With this in mind, wouldn’t then a true (no filters – period) NOS DAC be superior?

I would love to hear what other members think about this…
 
Ken Newton said:
I suggest thinking of the desired signal band more as being revealed by the reconstruction filter, rather than as being created by it.
Good way of putting it.

I remember going though a phase a few years ago when I believed that a triangular output from a DAC (instead of ZOH) would give better audio. Then I discovered (IIRC) that this would given an even worse HF droop than ZOH. As is often the case in digital audio, naive intuition leads us astray.

My own belief is that people prefer NOS because the ZOH HF droop adds some 'smoothness' (few NOS DACs compensate for this) and the first order images in the mid-20s of kHz (derived from signals in the high teens of kHz) put back some vague approximation of the percussive noise which was removed by the anti-aliasing filter at the studio.
 
Member
Joined 2017
Paid Member
I can agree with the difference between NOS and OS DAC, though I haven't experienced a NOS DAC. What I'm interested in is the difference between sampling rates; 44.1k,96k, and 192k. Because NOS of high sampling rate is almost same as OS. It means the purest NOS is 44.1k. I guess someone who likes 44.1k NOS will not like 192k NOS if the image spectrum which remains in 44.1k NOS dominates the effect.
 
Running a multibit DAC faster degrades its basic specs - no DS that I've studied has better figures for THD+N at higher rates than at lower. This is easily understandable in that the more updates per unit time a DAC's asked to make, the greater proportion of time it spends settling.

A few years ago I did a comparison between NOS and 2XOS on the same DAC chip and preferred NOS. However now I reckon this preference was largely influenced by the I/V stage I was then using which was fairly sensitive to RF energy. The 2XOS at the time sounded slightly more 'washed out' or 'greyer'. Since I've adopted a steep LPF between DAC and I/V stage I now find I prefer 2XOS for its additional 'air' or sense of spaciousness.

These results taken together leave me to believe people prefer NOS because of its more benign RF spectrum.
 
The only one I'm aware of (but which is long obsolete now) is CS4303 which has RTZ output coding.
 

Attachments

  • CS4303a.png
    CS4303a.png
    58.5 KB · Views: 302
  • cs4303b.png
    cs4303b.png
    73.1 KB · Views: 262
The ideal NOS DAC would output infinitely narrow pulses (Dirac-delta).

If so, the 'ideal NOS DAC' could be approached in practice by running the DAC as fast as it can possibly go (16X OS maybe?) and feeding in mostly zero samples in between the baseband samples. But the output level would need to be made up somehow - so with 16X OS we'd need to parallel 16 DACs to get to where we were with a single DAC on ZOH.

I very much doubt that this 'ideal NOS DAC' would sound as good as the basic NOS for the reasons DF96 has mentioned. With a passive anti-imaging filter though it would be definitely worth a try.
 
Something that's vague to me is this:

The early CD players, which were NOS, were criticized in the audiophile press for poor sound quality. The reason often given was the action of the brickwall filter (i.e., the analog reconstruction filter after the DAC).

Then, roughly beginning with 2nd gen. players, several manuf's began incorporating 2x oversampling (DF). And audiophile opinions improved.
3rd gens. were up to 4x oversampling ... and later gens. were up 8x ... yada, yada.
Audiophile acceptance increased as the CDP evolved.

But was the audio acceptance factor mostly determined by the use of DF's? Certainly, other factors in subsequent generations also improved: better decoder and DAC chips; improved analog output stages; PCB layout; topology; etc.

Also, recall there were a few third-party CD players in first-generation days -- notably Meridian -- that tweaked the basic Philips model. And squeezed out better sonics.

All that said, I do believe that OS and NOS designs each have their own sonic
"signature" ...as do single-bit vs. multi-bit.

All things equal, I'd probably vote for multi-bit with oversampling.

But I've hardly heard every modern design, including: dCS ring DAC; or the latest FPGA DACs (as used by Chord, etc); or the latest AKM- or ESS-based designs; or the esoteric NOS units from TotalDAC or Zanden.
 
Last edited:
First, make it clear that I have no intention of generating any controversy, I have carefully used the conditional.
I have put the term 'analog' since it seems that analog systems lack temporary errors, at least that seems to show measures like these, that I know made by a professional audio company, not hifi:

http://www.6moons.com/audioreviews/feickert3/13.gif

So it seems that the DSD, in this aspect, behaves like the analogue, maybe we should consider establishing the DSD as a preferred standard against the PCM, it would also be more convenient for the consumer, since the DAC could be dispensed with and thus make it cheaper costs.
I just read this. What they do are recommendations, but it does not speak at all of the dynamic range, which I think is a serious problem in digital audio with recordings with a maximum of 4dB, and one of the reasons why analog recordings sound better.
It would not hurt to recommend also that there are at least two categories: for the house and for the car. For the house recommend at least 30dB dynamic range (there are vinyls with 28dB, the digital should be better) and at least 10dB for the car, in my experience (totally subjective) as soon as it goes down 10dB the sound becomes unbearable.

Recording Academy Producers & Engineers Wing Publishes New and Updated Recommendations for Recording Deliverables and Hi-Res Music Production
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.