ANOTHER Example of Why I HATE DSD and Why Customers Who Bought Sony's Boloney Are So Annoying

[Miska]

3 hours ago, Ryan Berry said:

And impedes manufacturers from doing what we know makes a large positive change in the units.  I fail to see the advantages of attempting to drive the DSD standard on listeners.

 

It doesn't, instead it leaves you more money to focus on the actual conversion and analog sections and doing things you know. And better allows you to omit cheap COTS DAC chips and go for discrete designs of your own. I much more see that use of COTS DAC chips impedes you from doing many things.

 

3 hours ago, Ryan Berry said:

All it's really doing in the end is bypassing filters

 

Of course, that is half of the point. Other half of the point is that with a suitable DAC it bypasses both digital filters and modulator. Data feed going straight to the D/A conversion section. Which is the point.

 

As I said, with PCM, even at 16x rate inputs that many DACs already support, it is not yet enough to perform all the necessary oversampling needed for the modulator. And DAC chips resort to cheap-ass methods going up from there such as sample-and-hold (copying same sample multiple times) or linear interpolation which are both a big no-no for audio. This is before going to things like modulator implementation details.

 

I have one discrete R2R DAC that can accept 32x rate inputs (there are DACs from at least two vendors on the market) and at those rates on such DACs, when suitable noise-shaped dithers at correct word lengths are performed, you can get pretty decent results with PCM too. Then when combined with digital filters that have enough stop-band attenuation the analog filters actually manage to remove most of the digital images and the reconstruction becomes fairly complete.

 

As you know, for complete and accurate reconstruction of the analog waveform, there must be no image frequencies left anywhere above Nyquist frequency of the original content. If the first image level is let's say at -72 dB, it means that your reconstruction is accurate only to about 12-bit resolution. Now with my best filters, image levels are below -240 dB. And apodizing filters remove aliasing band problems of A/D converters.

 

Sure, if you can provide a DAC that has let's say 12 MHz 32-bit PCM input, I can happily send that kind of data there too. But then the question becomes what kind of modulator implementation follows?

 

3 hours ago, Ryan Berry said:

that even tricks people into thinking the DSD is somehow better

 

I have provided quite a bit of measurement results as well.

 

[Miska]

2 hours ago, asdf1000 said:

1bit SDM (DSD) is quite different to 5 or 6 bit SDM...

 

It is not, implementing those higher bit count modulators in a cheap way is just easier. That is really only "benefit".

 

Beauty of DSD is that you can easily make for example "5-bit" SDM conversion section convert it to analog. You can see example of such in my DSC1 design. There you have "5-bit DSD D/A converter" in action. It's a bit dummy though and I know how to massively improve it.

 

By the way ESS Sabre for example works in a pretty similar way with 6-bit (it has 64 elements) and is "multiple bits in time".

 

Nice thing with 1-bit SDM is also that you can make "power DACs" straight from it. It is also used by modern MEMS microphones natively. So for example in your mobile phone, both microphone and speaker/headphone amplifier use 1-bit SDM natively. So people listening Spotify on their mobile phones are practically listening to "DSD" without knowing about it.

 

[Miska]

1 hour ago, asdf1000 said:

As you know, Rob Watts (designer of all current Chord Electronics UK DACs) has different thoughts. But we don't need to rehash all that again.

 

Well, not so much, since he is using specifically those very much simpler modulator types. OTOH, he has so far made so many misstatements about DSD, that I have proven wrong, that I don't put much weight on that.

 

We know for example how RME ADI-2 running in DSD Direct mode performs, compared to how similarly priced Chord Mojo for example performs...

 

He won't be able to run anything like my filters and EC modulators on the FPGA's he's currently using.

 

Nice thing with computers is that for less than 5000€ you can get about 1 TFLOPS (double precision) of computing power that allows doing all kinds of things!

[Miska]

2 hours ago, asdf1000 said:

A better comparison (cost and non-portable) is probably i9-9900K build + RME ADI-2 DAC against Chord Hugo TT2 ....

 

Computer you anyway have for other purposes, so you can count that out. I don't have Hugo TT2 to measure, but I doubt it is much better in terms of modulator performance.

 

2 hours ago, asdf1000 said:

And for headphones setup you'd need to add a HPA4 to probably match TT2's headphone output performance (especially power).

 

No, you don't need anything like HPA4 to match TT2's headphone output performance. Schiit Magni 2 is probably enough. No need to have much power for headphones either.

 

If you could actually use headphone output of the ADI-2 it would be also technically good.

 

[Miska]

1 hour ago, asdf1000 said:

The 'Output drive' slide is most interesting to me and then the APx555 measurements.

 

Yet another set of measurements that get cut to 20 kHz so it doesn't tell anything about the reconstruction accuracy.

 

[Miska]

8 hours ago, asdf1000 said:

What should it go to? I can try to ask him 😀

 

I'm using 90 kHz and 5 MHz bands usually. If necessary also 200 MHz, but it is rarely necessary.

 

1 kHz tone at 5 MHz 1M point FFT with peak-hold

0 - 22.05 kHz linear sweep at 5 MHz 1M point FFT with peak-hold (sweep should be couple of minutes long and let repeat couple of times so that the peak hold fills up properly)

 

[Miska]

8 hours ago, CG said:

Wouldn't there also be some value in shorter sweep times but lots and lots of repeats to capture possible random events?

 

Yes, and another aspect is that in order for the sweep frequencies (and related harmonics) to properly hit the FFT window and get recorded by the peak hold, you need either slow sweep, or many repeats (preferably both). So that there are no gaps...