Denon DCD-900NE (CD Player)

[NTTY]

So the source of the problem lays in the way DACs work?

Correct.

Or is all that unwanted data already be present on the Audio-CD itself, basically?

No, it is not. The analog audio signal is filtered before being converted to digital and recorded on the audio-CD. So only 20Hz to 20kHz find their way on the Audio-CD, as @antcollinet mentionned.

————
Flo

 

[Timmy]

So only 20Hz to 20kHz find their way on the Audio-CD

Hmm ... So whatever beyond-20kHz signals a DAC may have to cope with are not coming from the Audio-CD?

 

[NTTY]

Hmm ... So whatever beyond-20kHz signals a DAC may have to cope with are not coming from the Audio-CD?

Exact. It is the DAC itself, when converting, that generates the unwanted “images” beyond 20kHz.

There are therefore two ways to get rid of them: 1) use an analog filter after the DAC so that everything beyond 20kHz is filtered, 2) use a digital filter that processes and modifies the digital signal to include a coding that mimics an analog filter, before feeding the DAC.

A combinaison of the two is generally required because digital filtering does not remove all images beyond a defined frequency.

————
Flo

 

[antcollinet]

Thanks, but that still seems to be far too complicated to me ...

Do all professionally made Audio-CDs really contain data that is imperfect in a way that a DAC has to enhance it first, before actually converting it?

No, the music is perfectly represented in the digital data. But conversion requires filtering. After filtering the music is perfectly represented in the analogue signal.

It is simply how the conversion to and from digital is done. It is the maths of how it works (Sampling theorem)

Nyquist–Shannon sampling theorem - Wikipedia

en.wikipedia.org

 

[Timmy]

Nyquist–Shannon sampling theorem - Wikipedia

en.wikipedia.org

The only thing I do understand here is the Moiré pattern exemplification.

If that really points at the basic nature of the problem, then it seems to me that the actual source of it is a dumb conversion – where we need an intelligent conversion instead.

Maybe we can look forward to an AI-controlled DAC. Should I call Jen-Hsun Huang once again?

 

[Timmy]

Was just asking this question at ChatGPT:

»Can AI help to improve the quality of digital to analogue conversion performed by DAC hardware devices used in the field of Hifi/Home Audio?«


Back came a quite long answer, ending with this conclusion:

In summary, AI can significantly enhance the performance and quality of DACs used in Hi-Fi and home audio by providing smarter, more adaptive, and personalized digital-to-analog conversion. This leads to better sound quality, more user customization, and ongoing improvements through software updates.

 

[Scytales]

Hi Scytales,

Couple of updates. First of, the AL24 is deactivated on the DCD-SA1, when playing SACD disc (DSD) which means it's only active with PCM input.

Here below, find an overlay of Pink Noise (Test track 8) and 1kHz @-16dBFS (test track 12):

View attachment 389355

I kept the view in dBFS on purpose so that the two traces overlay. You can see noise shaping starting just after 20kHz. Even with the single sine 1kHz, there is a massive amount of noise generated here. Is it part of the SACD recording, part of the DCD-SA1 conversion, or both, I can't know.

Pink Noise (orange trace) on top shows that the noise generated by the noise shaping technique takes over the pink noise itself at roughly 55kHz.

Note that the tracks are only 20sec long. The Pink Noise is random, no periodic as a I use, but it does not change anything.

And as I've put a lot of the same view in linear scale before, and for the sake of comparison, I'll add exactly the same view as above, only change is linear frequency scale:

View attachment 389356

It gives a different perspective, especially as to the noise of concern generated by the Denon DCD-900NE and especially the Marantz CD6007.

I will now test the various music tracks and report later.

EDIT: This is below an analysis of musical content (1700+ FFT averages over more than 5min). I overlaid SACD layer (green), CD Layer (red) and a last option of the DCD-SA1 which allows conversion of DSD to PCM (blue) before going to the DAC (Linear Frequency Scale for a better view):


View attachment 389392

The track (#27) in the Denon audio check SACD is Mahler: Symphony No. 2 in C minor "Resurrection" - 5th movement closing part (F.!) (Conducted by Wenceslav Neumann, Czech Philharmonic Orchestra, the choir, Gabriela Benachikova-Chapova (soprano), Eva Landova (alto))

Note that the view is in dBFS to let the measurements overlap. This is important because the pick of noise shown by green trace here, out of band, reaches -48dbr at 73kHz. Converted to % this is 0.4%...

This view shows that it is the player which is adding this shaped noise (green trace increasing starting from 25kHz). So when processing DSD data, there is a huge amount of noise added. This is not a surprise and I'd encourage you to read the Bartok review from Goldensound who showed massive amount of similar noise added when using DSD filters.

Back to our view above, the CD layer shows exactly the same overlay in audio-band. We see the low pass filter starting at 20kHz, which is the effect of ADC when creating/recording the CD at 44.1kHz.

Last and not least, the blue trace is an optional conversion from DSD to PCM before analog conversion (that the Denon DCD-SA1 offers). This view shows that there is probably noise created by the noise shaping technique on the DSD recording itself, but it is attenuated by the subsequent oversampling filtering of the PCM data stream.

And side note when it comes to the advantages of SACD over CD: at 22kHz we can still see "music", compared to CDA. That said, relative to the max signal, the difference is -80dBr vs -100dBr. In other words, even if the SACD can record and reproduce music beyond 20kHz, its energy is null, at least in this instance.

As per the above view, I'll continue sticking to CDA And so the Denon DCD-900NE is a killer.

--------
Flo

Thank you very much for your effort to produce this measurements!

I was pleased to see on your graph that I was able to interpret correctly the reading I had made a few years ago with an old HP 3581A - that the pink noise track on the Denon SA-CD test disc has a frequency content which goes up to about 50 kHz (the frequency limit of the HP).

The thing I find potentially enlightening on your measurement is that the out of band DSD HF noise spectrum is practically almost constant whatever the signal in the pass band. I don't know if you have been able to see that with other tracks of this test disc. That is a different behaviour than with PCM, with which the out of band high frequency spectrum is correlated to the signal in the pass band.

For the people who are concerned with the DSD HF noise spectrum (although no effect of this noise on downstream equipment is documented yet*), your measurements also show that the Denon does a pretty good job to decimate DSD into PCM, removing much of this HF noise in the process.

* Because you get both the equipment and the knowledge, one thing never done before anywhere that you could bring on the table for ASR is to take measurements of pieces of equipment downstream the Denon SA-CD player, both with pure DSD playback and with DSD decimated to PCM, in order to see if the response of the downstream equipment, for instance a preamp, does change or not in the audio band when subjected to the full DSD HF noise.

 

[antcollinet]

Can AI help to improve the quality of digital to analogue conversion performed by DAC

Improvement is not needed. Current DACs can already perform the conversion to a level of perfection way better than the human ear is able to detect.

 

[Doodski]

Was just asking this question at ChatGPT:

»Can AI help to improve the quality of digital to analogue conversion performed by DAC hardware devices used in the field of Hifi/Home Audio?«


Back came a quite long answer, ending with this conclusion:

In summary, AI can significantly enhance the performance and quality of DACs used in Hi-Fi and home audio by providing smarter, more adaptive, and personalized digital-to-analog conversion. This leads to better sound quality, more user customization, and ongoing improvements through software updates.

The response sounds like a gobbledygook feel good answer that is incorrect. Perhaps ChatGPT needs to be more up front and not give mushy answers instead of hard facts.

 

[AnalogSteph]

So the source of the problem lays in the way DACs work? Or is all that unwanted data already be present on the Audio-CD itself, basically?

It is an inherent property of any discrete-time (including digital) audio. Sampling in the time domain at intervals of 1/fs is equivalent to periodicity in the frequency domain with a period of fs.

We are only interested in everything up to fs/2, all the rest in the ultrasonic range is unwanted and could potentially upset following electronics quite badly.

There are some unconventional applications for this property, including the venerable FM stereo decoder.

 

[Timmy]

Improvement is not needed.

Since the result of the actual d/a conversion is so bad – as was stated here before – that it has to be enhanced (actually: corrected) by a filter, there seems to be a lot room for improvement. We need a core conversion that is intelligent enough to produce a result so close to the original sound that no further enhancement would be necessary.

 

[antcollinet]

Since the result of the actual d/a conversion is so bad

DA conversion necessarily includes the filtering. The filtering is part of the process of conversion. We measure the performance of a DAC after the filter, and the result of that conversion is not "bad" - in fact the performance of all decently designed Dacs is audibly perfect. No improvement is necessary.

 

[NTTY]

Since the result of the actual d/a conversion is so bad – as was stated here before – that it has to be enhanced (actually: corrected) by a filter, there seems to be a lot room for improvement. We need a core conversion that is intelligent enough to produce a result so close to the original sound that no further enhancement would be necessary.

It's what oversampling filters do.
The conversion from digital to analog creates data that need to be filtered. And they are properly filtered by the Oversampling filters + analog filters. Some are better than others. But that's a solved problem.

 

[Mart68]

when we have perfect speakers set up in a perfect acoustic space, with no noise intrusion, and we've been fitted with our new bionic ears, then we can start worrying about problems with filters.

 

[Timmy]

The conversion from digital to analog creates data that need to be filtered.

In other words: The result of D/A conversion, which is analogue data of course, inevitably needs to be filtered. Correct?

That means, at least to my understanding, that the core D/A conversion definitely leaves some room for improvement.

 

[NTTY]

Thank you very much for your effort to produce this measurements!

* Because you get both the equipment and the knowledge, one thing never done before anywhere that you could bring on the table for ASR is to take measurements of pieces of equipment downstream the Denon SA-CD player, both with pure DSD playback and with DSD decimated to PCM, in order to see if the response of the downstream equipment, for instance a preamp, does change or not in the audio band when subjected to the full DSD HF noise.

Thanks for your comments.

That is a very interesting question and not so easy to reply to: challenge accepted! That requires a preamp of high definition too. Let’s see what I can do.

Maybe it’s better that I create a new thread to review the DCD-SA1 first, and there we can deep dive into this specific topic.

————
Flo

 

[NTTY]

In other words: The result of D/A conversion, which is analogue data of course, inevitably needs to be filtered. Correct?

Yes.

That means, at least to my understanding, that the core D/A conversion definitely leaves some room for improvement.

It’s a mathematical consequence of the conversion, which is resolved by oversampling.

Some people are ok with all images created, and they use what we call “Non Oversampling” (NOS) converters. I don’t recommend.

The very first CD Players (eg: Yamaha CD-1 which I could review here) did not have oversampling filter and was relying on analog filter only.

 

[restorer-john]

overlaid with the standard AES IMD test (18kHz + 20kHz) which a lot of reviewers like to use

You are using prime (to SR) frequencies of 17987 and 19997, yes?

 

[Timmy]

That means, at least to my understanding, that the core D/A conversion definitely leaves some room for improvement.

It’s a mathematical consequence of the conversion, which is resolved by oversampling.

A mathematic consequence of that sort of conversion, I am saying.

Again: I’m sure we will all (hopefully) live to see a D/A converter that is capable to produce analogue output that needs no further treatment.

 

[NTTY]

You are using prime (to SR) frequencies of 17987 and 19997, yes?

View attachment 389791

I musts admit I don't, and I need to correct that. Thanks for the information, I did not read carefully enough the instructions.

Here below are the differences, from the Denon DCD-900NE.

IMD AES "Analog":

IMD AES "Digital":

And of course, the Denon goes sharp filter with these tests. And when adding a tone (at 80Hz), it switches back to its standard AL32 filtering.

In which case, this is the "True" analog IMD AES:

And then "true" digital IMD AES:

Of course, this is no longer an IMD AES test, but it shows the behavior of the AL32 filter when recognizing standardized tests.

As you can see, the measurement software (REW) has a low pass filter to stop distortion calculation at 20kHz (greyed out area) . Do you believe I need to extend it for this test?

--------
Flo