I am not reading out of context, look at this article
dsd
This article
http://www.uemc.polito.it/papers/opampsusc_01.pdf
Tell me what I took out of context if you like.
Perhaps you can apply more than in real life, but it's more limited than real life as well.
See this thread for example with an amplifier simulation at 0.000003% THD
http://www.diyaudio.com/forums/solid-state/255236-opamp-based-power-amp.html
Is that real or fantasy.
Anyway yes you are right, I may try simulation later.
I don't follow this comment, sorry.
The simple point is that oversampling can / will increase bit-depth / dynamic range by reducing the quantization noise and shifting it.
I was replying to DF96's comment here......
Here is a picture of DSD noise
This is increased resolution, in the sense of dynamic range resolution, via the oversampling and noise shifting.
Not only this, oversampling increases the samples per second in the 20 - 20 listening bandwidth, thus increases the sense of perceived detail, whether in theory or in reality.
In other words, it increases dynamic range resolution plus it increases time resolution as well.
It's not only "to make filter design easier", like DF96 was saying.
______
It seems like 16-bit / 192 kHz has an increased 20 - 20 time resolution as well.
Filter tap length in the reconstruction filter results in a theoretically increased time resolution as well.
There, we now have three "increased time resolution" factors, in the 20 - 20 listening bandwidth!!
I'm not saying they're humanly perceivable, I'm just saying that they exist.
The Xiph article does not say this anywhere, the Xiph youtube video quickly and incorrectly whisks over it.
https://www.xiph.org/~xiphmont/demo/neil-young.html
https://www.youtube.com/watch?v=cIQ9IXSUzuM#t=20m54s
Here is a picture which illustrates a predictive sine in white, versus the actual samples in blue
Professor Mark Csele: Color Organ Page
An externally hosted image should be here but it was not working when we last tested it.
-
Last edited:
If I understood correctly, he was talking about oversampled PCM vs non-oversampled PCM. You're comparing PCM vs DSD now.
This picture is propaganda created by the proponents of DSD. It doesn't reflect reality vis-a-vis the relative noise levels inherent in hi-res PCM and DSD.
its certainly not SACD single bit DSD - that peaks above 0 dB by 1.4 MHz
doesn't even manage 96 dB spot noise at 20k - see figs 4,5 Accuphase DP-85 SACD player Measurements | Stereophile.com
and of course actual commercial music recording dynamic range would be an informative curve to include too
doesn't even manage 96 dB spot noise at 20k - see figs 4,5 Accuphase DP-85 SACD player Measurements | Stereophile.com
and of course actual commercial music recording dynamic range would be an informative curve to include too
This experiment can give you insight in DSP.
But you may be very surprised by the outcome.
It's counter intuitive for sure.
Let us know what results you get.
Key point here is: Oversampling in combination with noise shaping.
Oversampling by itself can only increase dynamic range by just a tiny bit in very certain circumstances (if noise is decorrelated).
Here you imply that sampling theory is incorrect. And of cause it isn't.
A real world signal has an certain amount of noise in it and it has a limited frequency range. If the digital system has a lower amount of noise and bigger frequency range, it can reproduce the real world signal perfectly, this includes its timing resolution.
Not to mention the Second Law. But first principles and basic understanding may be too mundane for creative thinkers.
I took the picture from the very interesting Positive Feedback article, they took it from here
Playback Designs DSD Explained by Andreas Koch
I'm interested in amplitude resolution and how that would affect micro dynamic range, quiver, perhaps I can reduce the amplitude resolution in Audacity or some other software.
Even a cassette tape can display a wide macro dynamic range.
I'm interested in time resolution as well, like samples per second, especially since the Chord device has recently opened up this rarely visited parameter.
In fact I don't see how 192 kHz sample rate playback at 4x speed differs very much from the equivalent in extended tap length.
Playback Designs DSD Explained by Andreas Koch
I'm interested in amplitude resolution and how that would affect micro dynamic range, quiver, perhaps I can reduce the amplitude resolution in Audacity or some other software.
Even a cassette tape can display a wide macro dynamic range.
I'm interested in time resolution as well, like samples per second, especially since the Chord device has recently opened up this rarely visited parameter.
In fact I don't see how 192 kHz sample rate playback at 4x speed differs very much from the equivalent in extended tap length.
With a 1 bit digital system you have 2^1=2 steps of amplitude.
With a 2 bit digital system you have 2^2=4 steps of amplitude.
With a 3 bit digital system you have 2^3=8 steps of amplitude.
...
With a 16 bit digital system you have 2^16=65536 steps of amplitude.
With a 24 bit digital system you have 2^24=16777216 steps of amplitude.
So reduction of amplitude resolution is just bit depth reduction.
And now we can come back to this threads original question: All digital systems have quantisation noise. What matters is if this quantisation noise is correlated to the signal and at what level the quantisation noise is compared to the signal.
If you decorrelate the quantisation noise to the signal, with dither, you get just noise. The spectrum of that noise depends on the kind of dither you apply.
If the quantisation noise level is to low compared to the signal you get masking and you can't hear it.
Only in very extreme conditions, very low level signals and very high listening levels, people can hear this correlated quantisation noise with redbook audio. Its called birdsinging.
Its an audiofool myth that you need higher bitrates to increase timing resolution. See my previous post.
What do you mean with "micro dynamic range"?
yes, pretty trivial using audacity or sox. Attenuate the signal by desired amount, truncate, dither and re-amplify.
Time resolution is not depending on sample rate (as long as the nyquist criteria is satisfied).
Kastor L, it's really not clear to me what exactly you are interested in achieving here, or why. Are you trying to determine what would constitute theoretically perfect digital reproduction, or merely audibly perfect digital reproduction, or what? Furthermore, why are you interested in determining such things, what do you hope to achieve via this discussion? I mean, I honestly cannot tell whether you are attempting to learn something from us for genuine enlightenment, or are simply attempting to argue with us for controversy's sake. You seem unwilling to engage in an discipline study of the fundamental engineering principles involved, yet you continue to adamantly disagree with those who have. We'd like to help you to learn, but it's beginning to appear that you don't truly want such help. Perhaps, you can clarify your purpose?
Are the PS enough noise free for the better digital shematic you talk about ? Or does it need necessairly battery PS with bigs caps filtering because the noise floor and to be drifted free ? Just a urban legend ( does the PS are quiet enough ?) ?
I try to understand when our hears are more dependent of the PS design in a DAC than the shematic around the digital stuffs itselves ? Are not the LDO regs or their needed embeded compactness to be near to the active devices their main default but a needed trade off (EMT, etc...)... all things being equal of course (with a same output : being current or analogic stage)
Or do you talk just in theory, putting layout issues because OT in this discussion ?
I try to understand when our hears are more dependent of the PS design in a DAC than the shematic around the digital stuffs itselves ? Are not the LDO regs or their needed embeded compactness to be near to the active devices their main default but a needed trade off (EMT, etc...)... all things being equal of course (with a same output : being current or analogic stage)
Or do you talk just in theory, putting layout issues because OT in this discussion ?
+1
This is hopeless. It's impossible to have a meaningful discussion about, for example, linear phase and minimum phase filters with somebody who doesn't even know what those words mean.
I think you need to be much more specific here so I don't answer you incorrectly.
Are you saying that I'm saying Shannon-Nyquist is incorrect? If yes, how?
You do not think oversampling increases timing resolution / sense of detail, or you think it does increase it?
Do you think filter tap length increases timing resolution / sense of detail / resolution?
Thx
Last edited:
I recommend you direct your theories towards Asahi Kasei Microsystems, Wolfson and A/B listening tests of various filter designs.
You may be correct in theory that without frequencies above the Nyquist rate that minimum, linear and maximum phase will all sound / perform identically, minus latency, I'm really not sure.
Either way in real-world systems it seems like we have to deal with these frequencies.
It's like saying, without MHz interference we don't need a slew rate higher than 5V/uS.
Without ice-cream cones, we do not need ice-cream, yet, we have the ice-cream cones, yet you seem to prefer to think we do not have them.
Last edited:
Yes!
Sampling theorem states: If a function x(t) contains no frequencies higher than W cps, it is completely determined by giving its ordinates at a series of points spaced 1/2W seconds apart. see:https://en.wikipedia.org/wiki/Nyquist–Shannon_sampling_theorem
Iaw: If a signal is band limited (and a real world signal always is), a digital system with a sampling frequency of (a bit more) twice that bandwith can COMPLETELY reproduce it.
So for humans everything we can hear can be COMPLETELY reproduced by a 50kHz samplingrate. (I'm being very careful here, good implemented redbook is enough).
Oversampling makes the engineering of the anti-aliasing and reconstruction filters much easier. The most part of the filtering can be done in the digital domain. All modern converters use oversampling.
In DSP you use oversampling when frequencies above fs/2 are generated. This reduces/eliminates aliasing artefacts.
Here we're talking about the digital part of the anti-aliasing and reconstruction filters. Perfect filters have infinite taps, but human perception is not perfect and so we only need "blameless" filters. "Blameless" means we can't hear a difference between input and output.
The anti-aliasing and reconstruction filters need to be of a certain quality, and therefore have a certain amount of taps, to be "blameless".
The time when anti-aliasing and reconstruction filters were not properly designed (or digital systems in general) is decades ago. Even very cheap modern converters have excellent performance and are "blameless". Of cause there are exceptions where there is no reconstruction filter or oversampling being used, but you'll always have some nutters around.
Although I can't directly answer your questions, I hope this will explain it.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.