Denon DCD-900NE Review (CD Player)

[NTTY]

@antcollinet meant that these have “perfect” analog output, so they don’t need “further treatment”. And all of them use oversampling filters (and most probably analog too).

 

[NTTY]

@NTTY your graphs and comments are wonderful, edifying. They make it clear how dither plays out. So it's not a trick but a smart way to reach the limits of the medium, and most good player has it in its guts, that is good news. Thank you a lot, you made my day!

Your reviews are exciting.

"Vive le CD !"

Thanks to you for your comments

Yes, the good news is that dither is now part of our recordings, so that improves all converters out there. Now, there are multiple ways to add dither and multiple types of “noise shaping” techniques available on top. Noise shaping is the same as dither but the added noise is “shaped” so that 1) noise floor is decreased in the frequencies of interest, 2) noise added is pushed away from audio band.

The AES standard for measurements asks for TPDF dither (Triangular Probability Densitity Function), which is standard dither, and the one that I use in my test files.

————
Flo

 

[antcollinet]

Are you saying that for instance the S.M.S.L SU-1 (SINAD 116, [→ASR review]) is achieving its good performance without using any filters?

Of course I'm not.

Im saying they are D/A converters with an excellent analogue output which doesn't need further treatment to that analogue output. That is what you were asking for:

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.

 

[Timmy]

Well, from my understanding the work of a D/A converter is done, as soon as an analogue signal has been generated. Any further treatment of that analogue signal is no part of a D/A conversion (because a D/A conversion demands digital data as input), but rather that of an A/A conversion.

 

[Scytales]

Well, from my understanding the work of a D/A converter is done, as soon as an analogue signal has been generated. Any further treatment of that analogue signal is no part of a D/A conversion (because a D/A conversion demands digital data as input), but rather that of an A/A conversion.

You are mistaken.

Filtering, including in the analogue domain, is part of the conversion process from analogue to digital or digital to analogue.

That's because digital systems are based on sampled signals and that the Nyquist-Shannon theorem tells us that you can unambiguously reconstruct any sampled signal as long as said signal have been sampled from a band-limited signal whose highest frequency of interest is strictly less than half the sampling frequency.

Problem is : the Nyquist-Shannon theorem is only true mathematically as long as the axiom that the sampled signal is band-limited through a perfect low-pass filter, ie a filter with infinite attenuation rate from a frequency just below half the sampling frequency, is satisfied. And, from a Hi-fi perspective, this filter should have perfectly flat frequency response without ripple or attenuation near the corner frequency and unadulterated phase response in its pass-band.

As this axiom and ideal requirements cannot be satisfied in the actual physical world where we all live because no mechanical or electrical device of any kind can achieve something such as infinity (which is only a thought-experiment), every designers have to deal with compromises in the design of the filtering process, both at the digitisation stage (A/D) and the reconstruction stage (D/A). Hence different strategies to filter the out of band signal at the A/D or the D/A stages. Here lies part of the art of conversion to or from digital.

 

[antcollinet]

Well, from my understanding the work of a D/A converter is done, as soon as an analogue signal has been generated. Any further treatment of that analogue signal is no part of a D/A conversion (because a D/A conversion demands digital data as input), but rather that of an A/A conversion.

You seem to be confusing a DAC chip with a DAC. The chip is one component of a DAC - it is not a complete DAC. A DAC is the device you buy from someone like Topping which is the complete system from digtal input to analogue output. You don't have a converted analogue signal till it appears on the analogue output of a complete DAC.

The DAC chip does not output an analogue signal (of course it doesn't, it is not a complete DAC) - it outputs something that looks like a DSD 1 bit digital stream. It doesn't become analogue until after the filtering.

Just as dough output from the mixer doesn't become bread until after the baking.

 

[AnalogSteph]

The DAC chip does not output an analogue signal (of course it doesn't, it is not a complete DAC) - it outputs something that looks like a DSD 1 bit digital stream. It doesn't become analogue until after the filtering.

To be fair, this is the status quo of the early '90s (see e.g. CS4303 datasheet). The introduction of switched-capacitor filters for synchronous post-filtering was a big deal at the time - not only did it make the life of following electronics easier, it dramatically reduced jitter sensitivity, which was always a major pain point for 1-bit DACs. There was a good article in The Audio Critic at the time (sporting David Rich).

You can still find DAC chips with rather nasty levels of ultrasonic modulator noise in this day and age, notably the CS4392 or rather its integrated cousin in the CS4272's DAC section. It has some internal filtering but not nearly enough. Even the post-filter circuitry suggested by the datasheet is not entirely adequate, and a lot of cheaper audio interfaces do not even follow that. (If you noticed the wonkiness with the headphone output in Amir's Arturia Minifuse review, that IMHO is entirely the AP's analog frontend being upset by ultrasonic noise in its higher gain ranges.) It is, admittedly, a very old economy-focused design that was attempting to squeeze a lot of dynamic range out of what I think is a 1-bit or dual bit modulator by using a high order of noise shaping... and N-th-order noise shaping requires N-th-order filters to flatten out the noise again.

You can also buy the polar opposite like ESS DACs that are claimed to have noise levels flat to 200 kHz.

 

[antcollinet]

To be fair, this is the status quo of the early '90s (see e.g. CS4303 datasheet). The introduction of switched-capacitor filters for synchronous post-filtering was a big deal at the time - not only did it make the life of following electronics easier, it dramatically reduced jitter sensitivity, which was always a major pain point for 1-bit DACs. There was a good article in The Audio Critic at the time (sporting David Rich).

You can still find DAC chips with rather nasty levels of ultrasonic modulator noise in this day and age, notably the CS4392 or rather its integrated cousin in the CS4272's DAC section. It has some internal filtering but not nearly enough. Even the post-filter circuitry suggested by the datasheet is not entirely adequate, and a lot of cheaper audio interfaces do not even follow that. (If you noticed the wonkiness with the headphone output in Amir's Arturia Minifuse review, that IMHO is entirely the AP's analog frontend being upset by ultrasonic noise in its higher gain ranges.) It is, admittedly, a very old economy-focused design that was attempting to squeeze a lot of dynamic range out of what I think is a 1-bit or dual bit modulator by using a high order of noise shaping... and N-th-order noise shaping requires N-th-order filters to flatten out the noise again.

You can also buy the polar opposite like ESS DACs that are claimed to have noise levels flat to 200 kHz.

I really do need to get a better handle at the detail level on what DACs are doing these days.

 

[Billy Budapest]

Interesting about the trick to select a different filter to get better results when a test situation is detected. They were doing this a long time before VW it seems !

Right—I was about to say that this reminds me of VW Dieselgate—although VW had to pay hefty fines to the U.S. government after they were caught cheating.

 

[Timmy]

Right—I was about to say that this reminds me of VW Dieselgate—although VW had to pay hefty fines to the U.S. government after they were caught cheating.

Actually, I can’t find anything wrong with that Denon behaviour here: It makes the »measuring guys« happy – in case one is around and testing. And if else, if someone just wants to enjoy that unit and its sound, then the Denon serves those folks too

 

[Billy Budapest]

You seem to be confusing a DAC chip with a DAC. The chip is one component of a DAC - it is not a complete DAC. A DAC is the device you buy from someone like Topping which is the complete system from digtal input to analogue output. You don't have a converted analogue signal till it appears on the analogue output of a complete DAC.

The DAC chip does not output an analogue signal (of course it doesn't, it is not a complete DAC) - it outputs something that looks like a DSD 1 bit digital stream. It doesn't become analogue until after the filtering.

Just as dough output from the mixer doesn't become bread until after the baking.

That’s true for delta-sigma DACs as well as any manner of PDM conversion (PWM, DSD, etc.). However, it doesn’t describe the operation of old style PCM DAC chips like resistor ladder designs.

 

[NTTY]

Actually, I can’t find anything wrong with that Denon behaviour here: It makes the »measuring guys« happy – in case one is around and testing. And if else, if someone just wants to enjoy that unit and its sound, then the Denon serves those folks too

That is why I kept it

 

[antcollinet]

That’s true for delta-sigma DACs as well as any manner of PDM conversion (PWM, DSD, etc.). However, it doesn’t describe the operation of old style PCM DAC chips like resistor ladder designs.

True - with them you typically get a sample/hold “stair step” type waveform out - which is also only a partial conversion. You still need the analogue filter to complete the process.

 

[Timmy]

You still need the analogue filter to complete the process.

And that analogue filter is fed with analogue data – correct?

 

[antcollinet]

And that analogue filter is fed with analogue data – correct?

An imperfect analoge signal. In that it is a poor analogue of the signal it is supposed to represent. The subsequent analogue filter removes those imperfections (as far as is possible with such a system).

Don’t think for a minute you hve found some sort of a “gotcha”. The r2r dac chip (or discrete circuit if that is what it is) is still only doing part of the job and needs a reconstruction filter to complete it. And then not (in most cases) as effectively as a modern DS design.

 

[Timmy]

An imperfect analoge signal.

I understand. And has that imperfect analogue signal been converted from digital before?

 

[antcollinet]

I understand. And has that imperfect analogue signal been converted from digital before?

You seem to be playing stupid games. Youll be winning stupid prizes.

 

[Timmy]

You seem to be playing stupid games. Youll be winning stupid prizes.

I take this as a somewhat circumlocutory way of answering »Yes« to my question, OK?

 

[antcollinet]

I take this as a somewhat circumlocutory way of answering »Yes« to my question, OK?

If I were you I’d take it as advice against stupid point scoring modes of discussion you seem to be engaging in.

Either make your point clearly in one post or move on. Either way, I’ll not be engaging with you further.

 

[Timmy]

your point

Well – the point is, that as soon as an analogue signal has been processed from digital, the DA conversion is complete. How can there be any reasonable doubt about it?!

Any further treatment of the resulting analogue signal, may it be necessary or not, is not part of that conversion. Instead, it may rather be evidence to the fact that the precedent conversion process has some basic flaws.

The goal for the coming future now has to be to develop a DA conversion that is free of any noteworthy flaws. AI may be of some help to reach that goal.