What do you think makes NOS sound different?

Bulls eye Marcel! So this is what a NOS DAC really sound like.

Its like one sharp image and one over-sharpened image... it looks.... crap.

The above justifies my thesis that the "fault" in NOS DAC compensates for some other deficiency in the system - and that system will never be better than the NOS DAC.

//

The counter to this argument is that the over-sharpened image is being reproduced accurately by an NOS DAC. Distortions are a function of a nonlinear transfer function whereupon each signal point can be in an expansive relationship to the input signal. Hence signal distortions alone can result in an over-sharpened, over-contrasted image as not therefore a function of NOS. It is often the case that in my listening experience using NOS DAC's that it can sound the equal of an over-sharpened image. In contrast, using an OS DAC the image becomes more opaque and obscure, appearing as a more palatable sharp image.

In the piece God give me strength, using my oversampled S.M.S.L M100 DAC, the 176KHz file sounded considerably worse as the more I listened to it. No redeeming quality if you will. This was under circumstances that the signal was being subsequently filtered by my DAC. The only competition was between 44.1KHz and the 88.2KHz files with the 44.1KHz still having an advantage in terms of more expansive imaging with more palatable vocal inflections sounding more lifelike and real.

Things like voices, acoustic guitars and pianos are all instruments subconsciously comparable as a reference to real instruments. This is unlike variant independent sources of sound in some piece that often requires guessing what the sound engineer had done.

There seems a clear limit to using sequential oversampling, that if in doing so things don't improve, suggests a fault in the oversampling method. This can hold true for any sequential filtering being modest or otherwise. This isn't necessarily a law of diminishing returns, rather of negative returns.
 
The counter to this argument is that the over-sharpened image is being reproduced accurately by an NOS DAC. ......

No, the sharpening is created by the DAC - the NOS thingy... or actually absence of correct filtering - its faulty...

But maybe you mean if you blur/soften it?

What you see here is a non reciprocal process which ends up in information loss if you try it...

//
 

Attachments

  • _MG_6956 copy sharpedx100 then blurred.jpeg
    _MG_6956 copy sharpedx100 then blurred.jpeg
    135.1 KB · Views: 175
Last edited:
But with a really funky and poor monitor and ill behaving graphics card - the sharpened picture might look better than on one with perfect, truthful picture reproduction. (hint: monitor and graphics card being the equivalent of the rest of the sound systems i.e. amp and speakers).

//
 
How many NOS Dacs have been auditioned by JA.


So using the listening experiences he had with NOS Dacs must be treated with great care.
Auditioning like he does is called subjective listening impression. It does not have the necessary objectivity to assess any meaningful comparison result. It does serve well for advertisement though. Readers just have to know the difference between marketing and science.
 
So what is the conclusion? Upsampled 4X without aa filter played by OS sounds the same as NOS 44.1K without filter or not?

I'm uncertain of exactly what you are asking. Final conclusions are not complete, as our experiments have not yet concluded. For the moment, it appears that high performance 88.2 OS (as performed by the PGGB software resampler) essentially sounds the same as 44.1 NOS.
 
The only way to really find out what a NOS DAC with only zero order hold reconstruction sounds like to a cat is to die, reincarnate as a cat and find personnel with such a DAC, but as an experiment to get a first impression, I've scaled everything by two octaves.

The attached zip contains a 10 second music excerpt band limited to 5 kHz, this is roughly what an OS DAC playing 44.1 kHz sample rate audio should sound like to a cat (two octaves too little on the treble side).

The other has the band-limited signal subsampled at 11025 Hz and all samples repeated four times. This is roughly what a NOS DAC playing 44.1 kHz sample rate audio should sound like to a cat (two octaves too little on the treble side and spectral copies).

Listening to these files, I think the reason why NOS DACs without reconstruction filter sound acceptable to humans is precisely that we can't hear the images.
To illustrate Marcel's sound tracks with spectra, this is how they look like.
It's obvious that the left 11.025Khz NOS Dac spectrum sounds terrible because up to our hearing limit of at least 18Khz we have already two images of which one is even reversed.
The spectrum at the right representing the output of an 4 times OS Dac at 44.1Khz of the same file is clean up to 20Khz.


Hans
.
 

Attachments

  • 5Khz.jpg
    5Khz.jpg
    180 KB · Views: 155
No, the sharpening is created by the DAC - the NOS thingy... or actually absence of correct filtering - its faulty...

But maybe you mean if you blur/soften it?

What you see here is a non reciprocal process which ends up in information loss if you try it...

//

TNT,
Sorry, but I have a problem with comparing those two spectra with sharpening.
By analogue filtering afterwards, the process is "reciprocal" in your terms other than sharpening a JPEG that is non reciprocal.

Hans
 
I see @MarcelvdG reply, it is all wrong, bearing all consequences of a band limiting tunnel syndrome in the scientific thinking process.

Really? Then what did NOS DACs sound like the last time you were a cat?

Regarding the files I posted this morning (post #1332), normally audio for CDs would be steeply low-pass filtered at 20 kHz, sampled at 44.1 kHz and then either replayed through a normal DAC with 20 kHz low-pass filter, or a NOS DAC that just keeps reproducing a sample until the next one comes in (with no other filter than this hold function). I did basically the same, but with all frequencies scaled by 2 octaves to get an impression what CD audio must sound like to someone who can hear up to 80 kHz.

Of course a far better experiment is to die and reincarnate as a cat living at a home with a NOS DAC, but there are some practical difficulties with that.

I was thinking a lot about high frequency roll-off and whether I can hear it or not. Surely this should be significant amount around 8-12kHz (it is an area I can still hear despite of my age). I came to the conclusion that I don't notice such roll-off on my NOS R2R-11.

Is it a presence of Nyquist images cancelling roll-off? This is a big question. I think a clue is bringing DCs rep on the headfi forum. He says that roll-off is due to the unfiltered image adding in antiphase to the high frequencies below 20kHz. In the same way a characteristic NOS lush bass is because it is an area where interference comes in-phase. I know that roll-off is calculated mathematically, but I didn't try to challenge this statement. It is clear to me that such interference can only happen on a non-linear audio chain.

The sinc roll-off has nothing to do with the presence of Nyquist images. You just pass the signal though a filter with an impulse response equal to a rectangular of 1/44100 seconds wide, that filter gives you the roll-off.

However, when you play back 22040 Hz and look at what the NOS DAC output looks like on a scope, you will see something of which the amplitude goes up and down periodically. That's because both the signal at 22040 Hz and its image at 22060 Hz contribute to what you see on the scope. Measure with a spectrum analyser with a sufficiently narrow resolution bandwidth and you will just see a steady 22040 Hz with an amplitude that's about 3.91846 dB too low.

Now it comes to my point (which some part I have expressed already). As our sensors are digital, can a presence of Nyquist images cancel high frequency roll-off? We can't hear these frequencies as a sinewaves, but it is an energy that enters our ears, it cannot be disputed. The same way, I do not experience a bass lush on my R2R-11.

I disagree with your claim that our sensors are digital. From the hair cells onward, everything becomes quite non-linear, but that's not the same as digital (discretized in time and momentary value). No-one would call the output signal of the limiter chain in an FM radio digital, even though a limiter chain is quite non-linear and has essentially a two-valued output.

When the images are in the frequency range usually considered audible for humans, they certainly make the signal sound less dull and more distorted, as shown by the files of post #1332.
 
TNT,
Sorry, but I have a problem with comparing those two spectra with sharpening.
By analogue filtering afterwards, the process is "reciprocal" in your terms other than sharpening a JPEG that is non reciprocal.

Hans

Isn't sharpening done by spatial filtering? Basically the spatial equivalent of turning up the treble?

In any case, I understand TNT's analogy, as compared to the file that's only filtered, the version with images sounds less dull and awfully distorted.
 
In the piece God give me strength, using my oversampled S.M.S.L M100 DAC, the 176KHz file sounded considerably worse as the more I listened to it. No redeeming quality if you will. This was under circumstances that the signal was being subsequently filtered by my DAC. The only competition was between 44.1KHz and the 88.2KHz files with the 44.1KHz still having an advantage in terms of more expansive imaging with more palatable vocal inflections sounding more lifelike and real.
176k file was upsampled with additional noise-shaping, it is most probably a reason. Remember, I asked three times a question what options were used to upsample 88.2k file and didn't receive direct reply, it is only found in the description of the last test. Other than that, 176k file may approach a level where jitter is perceived bad.
 
I disagree with your claim that our sensors are digital. From the hair cells onward, everything becomes quite non-linear, but that's not the same as digital (discretized in time and momentary value). No-one would call the output signal of the limiter chain in an FM radio digital, even though a limiter chain is quite non-linear and has essentially a two-valued output.

When the images are in the frequency range usually considered audible for humans, they certainly make the signal sound less dull and more distorted, as shown by the files of post #1332.
I don't accept this argument. It is a set of thousands of sensors each tuned to a different frequency and there is a trigger point when an attached nerve is activated. Fully digital. Comparison to the FM modulation makes no sense.

The last paragraph bring us together. However you should accept a fact that additional energy contained in images stimulate a moment where sensors are triggered. These frequencies do not have to be heard as a continuous sine waves (which actually I pointed in my post), it is an upper level function in our brain of recognising objects.
 
Last edited:
Apparently you learned a different definition of digital than I did. The analogy with FM is that in natural neural networks, signals are often encoded in firing rates, that is, frequencies.

The hypothesis that spectral copies that sound awful when they are in a frequency range that you can hear well, enhance the sound quality when they are in the ultrasonic region seems rather far-fetched to me.
 
To illustrate Marcel's sound tracks with spectra, this is how they look like.
It's obvious that the left 11.025Khz NOS Dac spectrum sounds terrible because up to our hearing limit of at least 18Khz we have already two images of which one is even reversed.
The spectrum at the right representing the output of an 4 times OS Dac at 44.1Khz of the same file is clean up to 20Khz.
Hans
Just to illustrate how subjective the listening impressions are, those two files sound indistinguishable to me. :wave2:
The audiogram curves fall off sharply right there, at 6kHz. 😀

George
 
Apparently you learned a different definition of digital than I did. The analogy with FM is that in natural neural networks, signals are often encoded in firing rates, that is, frequencies.

The hypothesis that spectral copies that sound awful when they are in a frequency range that you can hear well, enhance the sound quality when they are in the ultrasonic region seems rather far-fetched to me.
It is before neutral networks where sensors receive kicks. It is a verified fact. Neural networks inner working are just a hypothesis.

I didn't make a comment about your experiment in my last post other than you are now closer to my way of thinking. A little bit. 🙂 I am sure you will find a weak starting point, take a time.
 
Last edited:
Do you think the 44.1 kHz NOS perceived quality is influenced by the presence of mirror images? This is definitely not the case with 88.1 kHz oversampled but not digitally filtered audio data.

Hi, lcsaszar,

That is a facet of the fundamental question predicating this thread. Which is; why don't OS and NOS essentially sound the same? While we've not fully concluded the analysis, indications are that it's not due to whether the image-bands have been filtered or not, but rather due to exactly how such filtering, when applied, is performed. So, while NOS means utilizing no OS interpolation-filter, high-performance upsampling is a OS interpolation-filter. Yet they both produce sonics which essentially sound the same. This indicates that the great majority of OS filters, which are typically co-resident on the DAC chip, are somehow introducing audible artifacts having little or nothing to do with their co-incident filtering of the image-bands. I don't wish to go in to more than this before thinking about and writing the investigation's final report.
 
Last edited: