DSD64, DSD128, DSD256 or DSD512?
DSD64 is roughly equivalent of 48/32 PCM, but without the brickwall digital filter artifacts.
-
WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
Octave Music Don Grusin High Resolution Music Analysis (Video)
- Thread starter amirm
- Start date
DSD64, DSD128, DSD256 or DSD512?
DSD64 is roughly equivalent of 48/32 PCM, but without the brickwall digital filter artifacts.
We've been improving the amplifier. It is certainly not easy implementation, but it works great. For sure you need fancy PSU, just like for any DAC reference voltage. Biggest challenges are the board design and the other high power switching design.
Yes, in my opinion it can compete. But it is certainly more complex and expensive though. So not necessarily for someone looking for low BOM solution. Already the hardware needed to run the related software costs more than BOM of those NCore/Purify amps. But OTOH, if you anyway want to do a "streaming amp" or similar you can see from NAD and others, you will need some computer for the software anyway. And then it is not so bad in the end. Plus you can run all the digital room correction at the same time as well. Moore's law is solving cost of the computing hardware all the time, it is already much easier than back in 2018.
In the recording this thread is (supposed to be) discussing.
This is what I got from my 18 years old Sanza Clip Zip using 48k sampling rate, 16 bits.
2400 Hz square wave, no sign of any phase problem.
Spectrum of a 10 kHz sine wave. Little spurs at ~970 kHz and then twice of that. Looks clean enough to me.
2400 Hz square wave, no sign of any phase problem.
Spectrum of a 10 kHz sine wave. Little spurs at ~970 kHz and then twice of that. Looks clean enough to me.
Quite a bit of Gibbs ringing, but that is to be expected with low rate PCM.
Did you check phase vs frequency response?
Best way to check is to make 0 - 22.05 kHz (or 0 - 24 kHz in case you use 48 kHz sampling rate) 10 second long linear sweep with analyzer in peak hold mode, like the ones I've presented. Then let it run for a minute or so to gather proper peak spectrum. That will give you good overview how things look like. Single tones like 10 kHz are hard to spot and don't really tell much.
But overall looks a bit like Cirrus Logic DAC chips, similar noise bump as you would get also with DSD. So it's SDM DAC chip.
It shows the textbook case of a bandlimited square wave. Certainly not the response of the Butterworth filter you showed in post #944.
Why don't you just tell us your exact procedure and equipment used to generate the graphs in your post #956, and see if someone (likely not me as I don't currently have access to better instruments) can replicate them.
How would you say since it is swallowed by the Gibbs ringing? So you didn't measure it's phase response? How much phase shift does it have (in degrees) at 20 kHz?
Point of the DSD square was to show that it looks pretty much like non-bandlimited square. Which will also expose any analog post-filter effects.
Here's Marantz HD-DAC1 playing 7 kHz square at 192k PCM rate:
Not so much effect visible yet from the Butterworth analog filter. Because it disappears largely in the Gibbs overshoot.
But if we do the same on the same DAC with DSD128 we begin so see more clearly:
Because now it actually hits the post-filter big time and otherwise begins to look like a proper square.
Some DACs may use Bessel type filter and then it looks better in this respect. But then you need even higher order filter with higher corner frequency to avoid rolling off the highs. That's why I used special transient optimized analog filter in my DAC design to avoid such overshoot and also keep phase response nice and avoid early roll-off.
I have told many times. 10 second long 0 dBFS linear sweep 0 - 22.05 kHz, 44.1/32 TPDF dithered WAV. Analyzer set to 1M point FFT running at 10 MHz sampling rate, peak hold spectrum. Use some sensible window function like Blackman or flat-top. 0 dBr calibrated to 0 dBFS using 1 kHz test tone. Let the tone run in repeat at least for 1 minute or longer until the spectrum stops changing.
Three essential things, 0 dBFS linear sweep from DC to Nyquist, enough bandwidth and analyzer in peak hold mode.
Last edited:
Pink noise is not good. White noise works better, but it takes forever on peak hold to fill up all bins to 0 dBr.
Here is the theoretical bandlimited square wave constructed from summing h1, h3, h5, h7, and h9. If phase is an issue with the output signal from my Clip Zip, it will not look like this.
Put a -40 deg phase shift in h9 as shown in your filter calculations, it changes to this.
dc655321
Major Contributor
- Joined
- Mar 4, 2018
- Messages
- 1,597
- Likes
- 2,236
BW and peak hold are not issue for AD2, but fft size is severely constrained by limited resources. It’s not really designed for such analyses. Could just offload data for external analysis.
For what order and kind of filter? That looks like high order minimum-phase filter.
If we go back to the original and look at step response of fifth order Butterworth filter:
If we then change to 4th order Chebyshev that also meets the spec:
You can see that the step response got much worse.
And then for 5th order Bessel:
And still the phase response looks like this:
But problem is that it would have early frequency response roll-off:
I generally want phase shift at 20 kHz to be less than 10 degrees. And frequency response no more than 0.1 dB down at 20 kHz.
What is shown is that phase shift in the anti-imaging filter in a dirt cheap Clip Zip is already a total non-issue.
Last edited:
So I'm quite happy with the DAC running at DSD512:
(you can ignore the extra interferences that are thanks to PSU and DAC boards without chassis on table with wiring hassle)
This has 4th order analog filter with fc=100 kHz. But of course the D/A conversion section itself is 32 element, so essentially 32-tap linear phase analog FIR.
Holo Spring 2 DAC at DSD512:
DSD 1024:
1.4112 MHz PCM:
6 dB level difference in noise floor because of 6 dB output level difference between it's R2R and DSD DACs.
705.6 kHz PCM:
Now you can see the image around 705.6 kHz sampling rate.
So you can see Spring has pretty steep analog filter in it's output stage. I don't have 7 kHz square wave plot of it at the moment, but I can check later when I get back home.
(you can ignore the extra interferences that are thanks to PSU and DAC boards without chassis on table with wiring hassle)
This has 4th order analog filter with fc=100 kHz. But of course the D/A conversion section itself is 32 element, so essentially 32-tap linear phase analog FIR.
Holo Spring 2 DAC at DSD512:
DSD 1024:
1.4112 MHz PCM:
6 dB level difference in noise floor because of 6 dB output level difference between it's R2R and DSD DACs.
705.6 kHz PCM:
Now you can see the image around 705.6 kHz sampling rate.
So you can see Spring has pretty steep analog filter in it's output stage. I don't have 7 kHz square wave plot of it at the moment, but I can check later when I get back home.
Which DAC is that ?So I'm quite happy with the DAC running at DSD512:
(you can ignore the extra interferences that
Why no DSD noise humps ! Good analogue section design.
Last edited:
My DSC1 design.
Because it is proper one. Of course there are others, like Holo.
EVGA NUAudio at DSD256:
Now, we are back to the original scheduled programming of most of the story can be told by SINAD.
So .... FUD on phase shift problem: Debunked
This you showed in post #956 is just a poor implemented DAC. Not that it matters much in real life as very few audio amplifiers has frequency bandwidth to a 300 kHz, no transducer reproduces them, and we can't hear them. It is shown that a bottom tier device sold >15 years ago did not have this problem. Another FUD debunked.
Similar threads
- Locked
