[GoldenSound]

Question was "what that matter". What you say is inconvincing at all. It even works in the opposite direction. Old CD's sound much better with NOS DAC. I had to re-discover all my old CD library. New mastering techniques like "loudness war" actually worked against further oversampling, as compressed music is more likely to create intersample overloads. With NOS decoding there is no overloads. But now a loudness was is over.

The latest mastering techniques bring recognition to the intersample overloads, but simultaneously high resolution formats became popular, it brings one more argument for listening in the original format.

This is just preference though.
If you massively prefer the sound of NOS that's fine. But it doesn't mean it's 'as produced in the studio'.
In regards to intersample overs. These are an issue, but not in all tracks, and can be mitigated entirely by ensuring your oversampling has sufficient headroom.

DACs from companies like RME, Chord, Benchmark etc all handle intersample overs scenarios with no issue. As do products like HQPlayer, the MScaler and PGGB.

Intersample overs ARE a problem, and some DACs are susceptible to them, but this is a question of poor DSP design and/or mastering, not oversampling itself.

NOS cause treble roll off, but it is only seen on the FFT plot. Human ear seems benefiting from high frequency images compensating drop off, as it is not noticed.

Do you have any evidence for this? Again, if you personally feel that is the case that's fine, but personal preference is not the same as an inherent truth.

 

[sajunky]

Intersample overs ARE a problem, and some DACs are susceptible to them, but this is a question of poor DSP design and/or mastering, not oversampling itself.

Intersample overs are not in result of a poor DSP design, claiming it is an indicative of a lack of knowledge. It is an inherent result of oversampling. To avoid, a volume must be lowered, reducing dynamics, so quality. It is done based on a probability factor, it doesn't guarantee oversampling free of 'overs'. Only the off-line resampler is able to reduce a volume to a such level that never produce overs. It still reduce dynamics, but only as much as needed. A higher rate of oversampling require a bigger reduction of volume.

 

[GoldenSound]

Intersample overs are not in result of a poor DSP design, claiming it is an indicative of a lack of knowledge. It is an inherent result of oversampling.

It is not an inherent result of oversampling.
Intersample overs are the result of both the mastering not including sufficient headroom, meaning even if all samples are below 0dBfs, the original waveform was not.
You must ensure that your recording and playback chain accomodate the maximum level of the original signal, else you'll get clipping either in the digital or analog domain.

It is because of the fact that when storing a waveform as sampled data, you will not always have a sample at the peak of the waveform. Therefore if you allow yourself no headroom, you will get clipping/intersample overs. Because regardless of the fact that your samples are not maxed, your reference for maximum signal level is too low.

If for example we record a 11025hz sine at a 44.1khz sample rate, and do not include sufficient headroom, then this is what we'd end up with:

All the digital samples are below 0dBfs, so the digital samples themselves are 'fine'. But the signal has still been clipped, and so unless you apply some headroom, the output will be incorrect.

Play it back unaltered, through something like a topping DAC which has no headroom, and it will look like this:

But play it back through a DAC that has the proper headroom (or apply the headroom yourself), and it's fine:

Play it back through a NOS DAC and it becomes an 11025khz square wave, not a sine.

To avoid, a volume must be lowered, reducing dynamics, so quality.

It's not about reducing dynamics. In order to properly reproduce the signal you need to ensure you're not clipping.
Otherwise you could use the same logic to imply that you should do the opposite, crank everything up even if it clips cause you'll get more dynamic range right?

A higher rate of oversampling require a bigger reduction of volume.

The rate of oversampling doesn't dictate the needed volume reduction.
Even just oversampling by 2x on the above waveform, you'd end up with a sample at the peak of the sine.
If you have 3dB headroom applied, it'll look like this:

If you have no headroom applied, you'll get this due to not being able to properly reconstruct:

But if you're upsampling by 32x instead of 2x, you still need the same headroom as 2x:

For a linear phase filter, you should only ever need 3dB headroom.
For a minimum phase filter though you may need more. (Though again, minimum phase filters are technically incorrect but that's a different issue).

I'm guessing by the lack of response you don't have any evidence for the statement below?

 

[sajunky]

It is not an inherent result of oversampling

Wrong. There is no intersample overs without oversampling, so this phenomenon is inherent to oversampling.

 

[GoldenSound]

Wrong. There is no intersample overs without oversampling, so this phenomenon is inherent to oversampling.

The effect is the same in the situations where it occurs.
Intersample overs clipping will just clip a straight line from the previous sample to the next when this occurs, because the interpolated samples cannot be added any higher.
And NOS will just clip a straight line from the previous sample to the next, because it isn't interpolating at all.

Still waiting on the evidence for the previous comment btw

 

[dcguy73]

Measurements aside, the T+ DAC sounds fantastic to me. But please, you guys continue if you like.

 

[sajunky]

The effect is the same in the situations where it occurs.
Intersample overs clipping will just clip a straight line from the previous sample to the next when this occurs, because the interpolated samples cannot be added any higher.
And NOS will just clip a straight line from the previous sample to the next, because it isn't interpolating at all.

Still waiting on the evidence for the previous comment btw

Not at all. Re NOS, you are right for this part, there is no interpolation. However there is no clipping in NOS, never. How did you imagine it would? I am really surprised.

Clipping means that inter-sample receives a maximum digital value, while it should be higher (according to the mathematical calculation).

 

[S Crowther]

What about the difference of opinion between the reviewer and Jussi Lasko about whether bit depth is best set at 16 or 24 bits for high res files?
Subjectively do whichever sounds best to you but technically is there a right or wrong?

 

[LostHisMarbles]

technically is there a right or wrong

'Technically' is also a relative term unfortunately;
- Regarding the actual implementation (and can you call upsampling technical when it is primarily a code? aka software?), not really, not wrong, no. If it is done, it is done.
- In context however? It being about sound? If we don't factor in how after a certain period in time "new" magic was needed to advertise and sell newer equipment, we're down to what any converter is meant or is not meant to be doing.. and the term converter entails it all by itself; convert. You have the actual information that is merely converted -as accurately as each component can, no two alike- into something your analog domain can work with. And then you have the actual information that through alchemies and propietary guessing software* changes the actual information, aka the actual musical information; before converting it to analog. So is that still technically accurate? No. The information on your track and the information 'arriving' on your amplifier are no longer comparable.
* It really is guessing software; sounds demeaning, but if you give it a good read, how it's done and then do some further reading, how processors handle arithmetic functions, yeap..

But as to opinions, to each their own of course
It's by now safe to say that we're well past any accuracy, definition or explanation. Strictly empirical and that's at best. You have size mattering, looks mattering, we're in a new era; sadly. As i keep saying, had people a good ear and the right criteria, we'd have a mere tenth of the brands we currently do. The rest would have closed down, superfluous or surpassed.

 

[rkt31]

How nos with aliasing will not affect and change the original intended sound ? How a playback device adding something can accurately represent the intended recording? Is 1khz fft test plot on nos is totally free of grass? If not then how nos is desirable?

 

[sajunky]

(1) How nos with aliasing will not affect and change the original intended sound ? (2) How a playback device adding something can accurately represent the intended recording? (3) Is 1khz fft test plot on nos is totally free of grass? If not then how nos is desirable?

1. NOS produce the most natural sound you can get. Processing change the original sound, makes it less natural. There is no aliasing during conversion (there are images), but aliasing can happen later, see #2.

2. A presence of unfiltered images can be a problem if images intermodulate on the downstream equipment, it is a place where aliasing happen. It is why NOS device require a wide bandwith amplifier and a low distortions up to at least 100kHz. Though a little distortions from a class A non-feedback operation is better than a sterile clean opamp type sound. A nested feedback amps like Topping A90 should be avoided at all cost, despite of excellent FFT plot.

3. I don't exactly follow a question. A 'grass' in the recording is present and I want to preserve it, another reason for not using opamps. Spikes on the FFT plot must be low, but we can't hear it below a certain limit. We don't really need a figure around -120dB, but it seem there is a chase to bring it lower and lower. Some add a wide-spectrum noise, it gives a similar effect of noise shaping, it equalise spikes increasing slightly a noise floor, but it doesn't improve sound quality. Trust your ears.

NOS gives the most listening pleasure, failing only in very complex material like choir or busy moments of a symphonic orchestra. However for classical music and jazz there is a high resolution music freely available, so these limitations are not valid anymore. 88/96kHz is very satisfaing, 176/192 is the most optimum for my R2R ladder, there is a slight bluring effect above this point. However you might like these smoothed details, especially on the beginning if your experience is in Delta-Sigma territory.

 

[shwnwllms]

What about the difference of opinion between the reviewer and Jussi Lasko about whether bit depth is best set at 16 or 24 bits for high res files?
Subjectively do whichever sounds best to you but technically is there a right or wrong?

16 bits is technically correct because Denafrips are linear to 16 bits. (Based on Jussi’s own measurements when performing a linearity sweep). Holo Audio DACs can be set to 20 bits since they are linear to 20 bits. No R2R DAC on the market is linear to 24 bits.

This is only a factor when upsampling to PCM though. Upsample to DSD and it’s a moot point. This setting doesn’t come into play.

Subjectively DSD sounds better to my ears than PCM on my Pontus II. I leave bit depth set to 16 in case I switch to PCM but mostly I upsample to DSD256 - ASDM7ECV2 - Gauss long (1x) Gauss-hires-lp (Nx)

 

[sajunky]

Holo Audio DACs can be set to 20 bits since they are linear to 20 bits. No R2R DAC on the market is linear to 24 bits.

Holo Audio can be set to 20-bits, as it use internal ultrasonic scrambling, therefore aplying dithering multiple times only deteriorate quality. It had been covered in this post located not far away, just on the previous page.

In other word, such recommendation should not lead to the conclusion that Holo Audio is linear to 20-bits. Not at all. It is measured linear to 20-bit using hundred thousands samples averaged results, filtered and noise shaped in the analyzer, not a response to a single output sample value.

 

[GoldenSound]

Holo Audio can be set to 20-bits, as it use internal ultrasonic scrambling, therefore aplying dithering multiple times only deteriorate quality. It had been covered in this post located not far away, just on the previous page.

In other word, such recommendation should not lead to the conclusion that Holo Audio is linear to 20-bits. Not at all. It is measured linear to 20-bit using hundred thousands samples averaged results, filtered and noise shaped in the analyzer, not a response to a single output sample value.

No they do not. And again, please provide evidence for this before repeating it endlessly.

(And please do not just post saying 'its been proven before' or 'its shown in X thread without providing any specifics)

 

[sajunky]

(And please do not just post saying 'its been proven before' or 'its shown in X thread without providing any specifics)

OK