See I, II and III.
But I'll give you some more basics.
When you approach the limits of quantization, regardless of bit depth, you will see an increase in distortion. That's why it's called quantization distortion.
Think about it. -90.3 dBFS * 2^15 (that's 16 bit quantization) = +/- 1.
Below that there's only 0, which is digital silence.
So at -96.33 dBFS all samples will round down to 0 = silence and the error would now be 100% since the whole signal is missing, right?
Well no, not if you do it properly and dither or do noise shaping. That will decorrelate the quantization error from the signal, and so you just get a well-defined noise floor and not distortion products as seen above.
.. but again, don't judge that by its looks either. Have you ever seen a DSD stream? Sounds pretty good considering how that looks, right?
Last edited:
Hi,
The assumption in the OP is completely wrong,
its pure nonsense, nothing to at all discuss.
rgds, sreten.
The assumption in the OP is completely wrong,
its pure nonsense, nothing to at all discuss.
rgds, sreten.
Agree, -90dB is very low, but usually sine waves of dac-s, especially 16/44 are hidden.
As author of this topic said, you must be more opened for new ideas, maybe good, maybe bad, but it's silly to stuck in some digital theorem for telegraphy, when even Sony go ahead.
Ever heard about Rob Watts?
Here, you'll laugh that man, because he don't stuck in telegraph theorem.
Thanks, I go in the reading mode.
As author of this topic said, you must be more opened for new ideas, maybe good, maybe bad, but it's silly to stuck in some digital theorem for telegraphy, when even Sony go ahead.
Ever heard about Rob Watts?
Here, you'll laugh that man, because he don't stuck in telegraph theorem.
Thanks, I go in the reading mode.
Last edited:
What do you mean by hidden? I'm sorry but that first sentence doesn't make any sense.
I'm also not the author of this topic and I am open for new ideas.
It's silly to say we're stuck in "telegraph theorem" and I'm not even talking about the grammar here. 😛
As an analogy, this discussion is like talking about how multiplication works because apparently there is some lack of knowledge or at least confusion about it. The basics of the basics. Now you come along and say it's stupid that we're stuck in multiplication?
I'll tell you what is stupid. To expect a clean sine when I've explicitly pointed out the I, II and III. That's like expecting 2*3 to equal 5.
And you also cannot change the sampling theorem just like you cannot change the rules of multiplication.
Biggest apologies that I can't offer anything on the topic.
But I have to chime in. First, Xnor, thank you so much for giving information that can help lead to an understanding. Second, while there may not be aggressive words, I find it well out of the spirit of the forum when the usual cast literally just log in to tell people they're wrong and to go away and come back when they don't have anymore questions. What the **** for?
Also there's no on going record of claims that DAC's, bits, resolutions, filters, etc don't sound different. Quiet the opposite. That is among the people whom have HEARD them...................................
But I have to chime in. First, Xnor, thank you so much for giving information that can help lead to an understanding. Second, while there may not be aggressive words, I find it well out of the spirit of the forum when the usual cast literally just log in to tell people they're wrong and to go away and come back when they don't have anymore questions. What the **** for?
Also there's no on going record of claims that DAC's, bits, resolutions, filters, etc don't sound different. Quiet the opposite. That is among the people whom have HEARD them...................................
Last edited:
I presume that this "telegraph theorem" is reference to Oliver Heaviside's theorem, the basic of todays signal integrity stuff..... stuff that I use nearly every day....
He also gave us twisted pair cabling I believe....
Who is Rob Watts any links anybody.....
He also gave us twisted pair cabling I believe....
Who is Rob Watts any links anybody.....
He's a digital audio designer who used to have his own hifi company (Deltec Precision Audio) but nowadays is better known for his designs for Chord (Hugo and successors) which use a largeish FPGA to implement long FIR filters.
Cheers.
had a look at both sites, some nice looking kit, don't know what it sounds like. if they are using Virtex FPGA's at Chord then they will be following the basics from the telegraphers theorem (equations) I should imagine... or has Nironiro got something mixed up?
had a look at both sites, some nice looking kit, don't know what it sounds like. if they are using Virtex FPGA's at Chord then they will be following the basics from the telegraphers theorem (equations) I should imagine... or has Nironiro got something mixed up?
Here is a different approach to improve NOS:
Use full-scale peak-to-peak of 300V instead of 3V. This can be achieved by some tube circuit. Then reduce the pulse width to 1/100, i.e. about 2.27 us, e.g. by using a differentiator (C-R circuit). So we get a very narrow pulse of high amplitude, that has the same energy content as the original NOS analog signal. The rationale behind it is that samplig occurs at discrete times at the ADC, and there is no information about the analog waveform between these discrete times. The reconstructed waveform after the DAC will also have samples at discrete times, and no information (zero signal) between the samples.
Would it work?
Use full-scale peak-to-peak of 300V instead of 3V. This can be achieved by some tube circuit. Then reduce the pulse width to 1/100, i.e. about 2.27 us, e.g. by using a differentiator (C-R circuit). So we get a very narrow pulse of high amplitude, that has the same energy content as the original NOS analog signal. The rationale behind it is that samplig occurs at discrete times at the ADC, and there is no information about the analog waveform between these discrete times. The reconstructed waveform after the DAC will also have samples at discrete times, and no information (zero signal) between the samples.
Would it work?
20khz
From a NOS Dac to a oversampling Dac if correctly filtered there is about 4khz on
band pass because the level dac is a comb filter .....🙂
From a NOS Dac to a oversampling Dac if correctly filtered there is about 4khz on
band pass because the level dac is a comb filter .....🙂
Could someone translate this from gibberish into English?
I wash born here, an I wash raished here, and dad gum it, I am gonna die here, an no sidewindin' bushwackin', hornswagglin' cracker croaker is gonna rouin me bishen cutter.
All it would do is almost eliminate the sinc frequency response of a normal sample-and-hold NOS output. It might be more likely to fry whatever comes next in the audio chain, unless followed by a passive low pass filter. Oops - we just replaced one rolloff by another rolloff!lcsaszar said:
In what sense is this an improvement? In both cases you have enough information to fully recover the original bandlimited analogue waveform (if you choose to do so by using a reconstruction filter) as the samples fully encapsulate the waveform between the samples - that is what bandlimiting does for you.
No, still can't understand what you are saying. Can you try again?gumo73 said:
In dsp terms this is upsampling or zero stuffing which comes before filtering (=> interpolation).
For example, if you upsample by a factor 2 you have to insert 1 zero after each sample. In the frequency domain you'll get everything mirrored up perfectly from half the original sampling rate.
So you still have to filter with a low pass that suppresses these images and has 2x DC gain to make up for the "loss" of energy.
The advantage of this in dsp is that you don't get any droop like you do with zero- or first-order hold, so all you'll need for interpolation is a simple low pass.
--
On your question, I do not think this will work. The input signal is not a square wave.
Imagine a 10 Hz stairstep wave .. each tiny step will result in a tiny pulse at the differentiator output. Now imagine a high frequency stairstep wave with the same amplitude, where the samples fall on the extremes ... huge steps will result in big pulses.
Besides, there is a phase shift and other practical issues.
Last edited:
No, not even a little bit. The small part I understood (Shannon-Nyquist magically doesn't work for 44.1k sampling and 16k reproduction) was thoroughly wrong, and the rest was non-comprehensible.
I don't think he needs to read any books on that topic. 😉
But I'll still give this a try, because I want to understand where the misunderstandings lie:
Where does the number 16 kHz come from?
What does "you can't going up" (I guess you mean "you can't go above") mean?
What happens above that frequency that doesn't happen below it?
But I'll still give this a try, because I want to understand where the misunderstandings lie:
Where does the number 16 kHz come from?
What does "you can't going up" (I guess you mean "you can't go above") mean?
What happens above that frequency that doesn't happen below it?
Ahahah no comment the lies ....😎
do you kow how the dac work ?!? what is his frequency response..... is a mystery for some pepole....
Last edited:
OK, I will attempt to translate this:gumo73 said:
"If you have a normal sample-and-hold DAC working at 44.1kHz you get a comb filter (sinc frequency response) which means that by 16kHz you already have a small amount of HF rolloff which needs to be compensated for if you require a completely flat frequency response."
I am losing interest in why this is thought to be a contribution to the discussion.
- Status
- Not open for further replies.