Yeah, I need to keep reminding myself not to respond to anything that neither is sensible nor honest.
Bear, is that really a square wave or a low level 1KHz sine wave? Looks like a low level sine wave to me. Of course, according to SY, nothing much matters. '-)
It is what passes for a squarewave
Here, by Pedja Rogic:
https://www.audialonline.com/articles/high-resolution-audio-asynchronous-usb-oversampling-upsampling-and-stuff/#
It is a square wave reproduced by today's DACs...
SY, the point is that it is NOT a square wave, although that is what it is supposed to be... while the DAC will do a pretty fine sinewave, the issue is IF the obvious "flaw" in the digital reconstruction WRT a squarewave, and the cascading of said "flawed" filters between the record side and the playback side contributes to an increase in error (one that can not be directly assessed without the access to the record AND playback side).
Clearly my old tube amp will do a better job at making good looking 1kHz squarewaves than the multi-thousand dollar (or cheap) DAC will, no??
The latest SOTA amp will do a 20kHz square wave almost flawlessly.
Afaik, a now ancient Spectral amp will do more than a 20kHz square wave...
So, clearly there is a type of distortion inherent in the DAC replay system. One that is based on the mathematics, that to be overcome would require a different approach than the one now used.
But, no doubt, it's not audible, so why worry about it. 🙄
PS. those are actual measured results from 2016 of two DACs from Stereophile mag, SQUAREWAVEs, 1kHz.
As I mentioned a bandwidth limited tube amp does a better job, so take the bandwidth argument and run it into the round file.
The issue is somewhat different than merely the "bandwidth".
Here, by Pedja Rogic:
https://www.audialonline.com/articles/high-resolution-audio-asynchronous-usb-oversampling-upsampling-and-stuff/#
It is a square wave reproduced by today's DACs...
SY, the point is that it is NOT a square wave, although that is what it is supposed to be... while the DAC will do a pretty fine sinewave, the issue is IF the obvious "flaw" in the digital reconstruction WRT a squarewave, and the cascading of said "flawed" filters between the record side and the playback side contributes to an increase in error (one that can not be directly assessed without the access to the record AND playback side).
Clearly my old tube amp will do a better job at making good looking 1kHz squarewaves than the multi-thousand dollar (or cheap) DAC will, no??
The latest SOTA amp will do a 20kHz square wave almost flawlessly.
Afaik, a now ancient Spectral amp will do more than a 20kHz square wave...
So, clearly there is a type of distortion inherent in the DAC replay system. One that is based on the mathematics, that to be overcome would require a different approach than the one now used.
But, no doubt, it's not audible, so why worry about it. 🙄
PS. those are actual measured results from 2016 of two DACs from Stereophile mag, SQUAREWAVEs, 1kHz.
As I mentioned a bandwidth limited tube amp does a better job, so take the bandwidth argument and run it into the round file.
The issue is somewhat different than merely the "bandwidth".
Last edited:
congratulations (i coincidentally bumped into the leader/leading candidate of an opposition party here yesterday. It's general election time next month. To get some, maybe easier to give some, i've got a lot i want to get)
Bear, I am pretty sure that it is a low level undithered 1KHz sine wave. That is what they measure in CD reviews.
yes, quite so, undithered... 😀
gotta chum the waters here, you know?
But the fully dithered reconstruction still shows significant overshoot on leading and trailing edged and ringing... more than anyone here would likely accept in an opamp or power amp... fwiw.
gotta chum the waters here, you know?
But the fully dithered reconstruction still shows significant overshoot on leading and trailing edged and ringing... more than anyone here would likely accept in an opamp or power amp... fwiw.
Regarding DAC reconstruction filters, if you make a very sharp brickwall filter (analog or digital) and hit it with a sudden step it's going to ring. Some FIR filters will pre-ring. Nothing mysterious about that. Using sharp filters like that is one way to make a DAC. Another way is to up-convert the sample rate to something much higher than 44.1, then covert to analog at the high sample rate, then use a gradual slope reconstruction filter that doesn't ring like a sharp filter does. Take your choice.
Last edited:
Oh my, is someone constructing signals that violate Shannon-Nyquist and wondering why things end up badly?
(Just because you made it in software rather than lowpass-ADC, doesn't mean that aliasing isn't going to be a big problem)
(Just because you made it in software rather than lowpass-ADC, doesn't mean that aliasing isn't going to be a big problem)
Oh, I don't know. People can learn how to play a time-feel or groove just like they can learn how to hit a pitch, or even to sing every individual inflection in a vibrato. It helps a lot to have a teacher that can clearly explain what to do and then demonstrate it. Some people who can play a tight groove don't know exactly what it is they are doing to make it sound the way it does, so they can't explain how to do it to somebody else.
Well the idea, again goes something like Danley's method for elevating certain distortion products up by running the signal through the DUT "N" times - each time through the characteristic difference products (distortion) are increased.
What I was thinking about is IF there is a "best" combination of digital filter on the A/D side and then the D/A side? Conversely, might there be a relatively terrible combination? Filters on both sides considered individually might be quite reasonable?
The evidence that this may be so is that there is ringing post/pre on the impulse and the square wave reconstruction shows not just ringing, but overshoot on the leading and trailing edges, with ripple in the passband.
That's all... no need to get into issues of how the squarewave was produced, or which DAC is doing what. Not at this point. It's a general question relating to the effects of cascading two unknown filters. Which is obviously unavoidable in today's technology.
And fwiw, running up the oversampling or sampling rate doesn't seem to eliminate the artifacts, if a filter is used. Reduced is great, of course. So, I would expect that the effect, IF it is there, and audible as well would be most pronounced on redbook...
remember, stay clam at all times!
What I was thinking about is IF there is a "best" combination of digital filter on the A/D side and then the D/A side? Conversely, might there be a relatively terrible combination? Filters on both sides considered individually might be quite reasonable?
The evidence that this may be so is that there is ringing post/pre on the impulse and the square wave reconstruction shows not just ringing, but overshoot on the leading and trailing edges, with ripple in the passband.
That's all... no need to get into issues of how the squarewave was produced, or which DAC is doing what. Not at this point. It's a general question relating to the effects of cascading two unknown filters. Which is obviously unavoidable in today's technology.
And fwiw, running up the oversampling or sampling rate doesn't seem to eliminate the artifacts, if a filter is used. Reduced is great, of course. So, I would expect that the effect, IF it is there, and audible as well would be most pronounced on redbook...
remember, stay clam at all times!
Last edited:
I don't think there is a best combination. For CDs, if you could assure that your input signal contained no frequencies above 20kHz, then you wouldn't necessarily need an A/D input filter. In that case, it would be there only for insurance. If you could assure that you could limit all input frequencies to something higher than 20kHz, that's fine too. Just sample fast enough so no aliasing can occur and you don't need an input filter. It's only there to prevent aliasing.
If you do sample very fast, you could then down convert the sample rate using software back to 44.1 or whatever you want. Some of the software for that is very good, with processing artifacts down 160 db or so. I sure can't hear any distortion from such software.
The reconstruction filter is a different matter. Sampled data is like a series of dots on a graph, evenly spaced along the time axis (assuming no jitter). The reconstruction filter is there to smoothly and accurately connect the dots, thus providing a continuous time analog output.
Bear if you don't want to do the homework and become facile in the theory but on the other hand throw out random criticism there's no basis for a discussion.
We should remember to emphasize that anti-aliasing filtering must still be done when converting to a lower sample rate. Otherwise, we'll spend three days on that detour.
All good fortune,
Chris
- Status
- Not open for further replies.