I assume that you're wondering whether the resulting two non-linear phase shifts would be complementary, thereby producing a net linear phase response without the introduction of pre-echo to the impulse response. Since both the AAF and the AIF filters are low-pass sinc function, I would expect that any non-linear phase shifting of the original input signal to become more severe after passing through both filters, but I haven't done such an analysis. However, I suppose that an complementary phase correcting all-pass function could be added to the playback DSP chain.
Perhaps that was what I was hoping for yes but the thinking hadn't really crystallized in my mind. But probably one of the filters would need time-reversing to achieve that?
I wasn't explicitly thinkng of the phase response though, rather something that Rob Watts has raised in the context of his Hugo DAC, where he claimed that the reconstruction filter had to be both linear phase and extremely long (10^6 taps I believe he mentioned) to get time domain errors below 16bit levels. But perhaps on reflection what you said is just the dual of what I was thinking of - phase distortion could just be another way of looking at 'time domain errors'?
What's been tickling me is - if the original sampling filter wasn't linear phase (IOW min phase as the earliest ADCs didn't use oversampling) then would a min phase AIF make the best job of reconstruction or is a linear-phase one still better?
I like the idea of including phase correction for the in/out filters, perhaps I'll consider this in a future DAC design, thanks for the suggestion Ken
Incidentally it seems that the terms 'pre-echo' and 'pre-ringing' have become rather conflated on this thread. Let's keep to the standard nomenclature that refers to the 'leading' part of the sinc function as 'pre-ringing'. Pre-echo is something that only occurs in certain FIR filters with regular ripples in the FR.
I wasn't explicitly thinkng of the phase response though, rather something that Rob Watts has raised in the context of his Hugo DAC, where he claimed that the reconstruction filter had to be both linear phase and extremely long (10^6 taps I believe he mentioned) to get time domain errors below 16bit levels. But perhaps on reflection what you said is just the dual of what I was thinking of - phase distortion could just be another way of looking at 'time domain errors'?
What's been tickling me is - if the original sampling filter wasn't linear phase (IOW min phase as the earliest ADCs didn't use oversampling) then would a min phase AIF make the best job of reconstruction or is a linear-phase one still better?
I like the idea of including phase correction for the in/out filters, perhaps I'll consider this in a future DAC design, thanks for the suggestion Ken
Incidentally it seems that the terms 'pre-echo' and 'pre-ringing' have become rather conflated on this thread. Let's keep to the standard nomenclature that refers to the 'leading' part of the sinc function as 'pre-ringing'. Pre-echo is something that only occurs in certain FIR filters with regular ripples in the FR.
In addition to that, the signal being sampled must have infinite duration, not only forward in time, but backward as well. I've always wondered how an high frequency single-cycle burst would appear to a sampled system. In other words, if the frequency of the sampled single cycle meets Nyquist, yet it is the most opposite possible to being continuous in duration, would it's eventual analog reconstruction reveal any AAF or AIF impulse response related anomalies upon viewing on a scope?
For example, in a 44.1K sample rate system, would an single cycle of an 20KHz sine wave show any anomalies, particularly at the start or stop instants of the cycle? Would the shape of that 20KHz single cycle's band-limited analog reconstruction be obviously altered? I want to say no, that even though it's a single cycle burst, it's maximum rise and fall times still place it within Nyquist, so no issues. However, I don't feel as certain about that as I would like to.
P.S. - I suppose, that what I was really wondering is whether certain band-limited signal patterns, though fully meeting Nyquist, could still show time domain anomalies upon reconstruction? However, after further consideration, I now suspect that there would only be those time domain anomalies at reconstruction which were introduced by the band-limiting anti-alias filter at the system input.
Last edited:
The last part of your post above, #362, are you saying pre-echo occurs strictly within FR = frequency response?, or did you leave the rest of concept out?
If we're going to use the terms in a crystallized sense it's better to denote them in full detail.
As for the 10 Ghz sampling rate, that's easy, the Princeton paper says we are sampling an "infinite" amount of samples right?
The use of the word infinity annoys me, it's not realistic and it's not a hard number, it can be used for convinience, to mystify discussions like DF96 speaking of infinite this and that earlier in this thread and in marketing concepts as well.
If air movement never exceeds 10 Ghz - unless we fill our entire listening room with some kind of unusual substance - then why should sampling rate?
I took a long shot with a tenth of a nanosecond, it should be more like a sixth of a nanosecond, in which case, hasn't octa-rate DSD far exceeded the terminal velocity of sound already!?
See, I'm not all Egyptian Nicotine airy fairy, I'm debunking marketing and technology as well =)
If we're going to use the terms in a crystallized sense it's better to denote them in full detail.
As for the 10 Ghz sampling rate, that's easy, the Princeton paper says we are sampling an "infinite" amount of samples right?
The use of the word infinity annoys me, it's not realistic and it's not a hard number, it can be used for convinience, to mystify discussions like DF96 speaking of infinite this and that earlier in this thread and in marketing concepts as well.
If air movement never exceeds 10 Ghz - unless we fill our entire listening room with some kind of unusual substance - then why should sampling rate?
I took a long shot with a tenth of a nanosecond, it should be more like a sixth of a nanosecond, in which case, hasn't octa-rate DSD far exceeded the terminal velocity of sound already!?
See, I'm not all Egyptian Nicotine airy fairy, I'm debunking marketing and technology as well =)
Its all dealt with in Julian Dunn's paper - have you ferreted it out yet?
I didn't look at the paper - even at 10GHz its still an initinite amount of time to take infinity samples. So you avoided the question about how its relevant to audio.
I suggest a strictly finite dose of the 'infinite improbability drive' to inoculate you - try reading some Douglas Adams.
Plus, personally I find it a little exciting to imagine a perfect / idealized audio system, whether in writing or later in my life in real practice.
If I can't actually hear the perfection, well, that's secondary =)
______
Abraxalito, how does oversampling in the ADC concern you by the way?, or it doesn't?
Just thinking of your preference for Nos in the D/AC.
I'd like to have a switch to "turn off" oversampling just to see what happens, but that isn't really possible in a Sigma-Delta is it?
I see people speak of Nos, filterless and R2R or non-Sigma-Delta all rather interchangably at times, such as that "Nos" PCM1794 thread? It's not actually non-oversampling is it?
Latency is caused in the filter which "doesn't exist in Nos", yet "a filter is a tenet of the Nos design", can we sub-divide that "filter" reference as well? I'd just like to crystallize the terms which are commonly used.
If I can't actually hear the perfection, well, that's secondary =)
______
Abraxalito, how does oversampling in the ADC concern you by the way?, or it doesn't?
Just thinking of your preference for Nos in the D/AC.
I'd like to have a switch to "turn off" oversampling just to see what happens, but that isn't really possible in a Sigma-Delta is it?
I see people speak of Nos, filterless and R2R or non-Sigma-Delta all rather interchangably at times, such as that "Nos" PCM1794 thread? It's not actually non-oversampling is it?
Latency is caused in the filter which "doesn't exist in Nos", yet "a filter is a tenet of the Nos design", can we sub-divide that "filter" reference as well? I'd just like to crystallize the terms which are commonly used.
I've yet to get into designing ADCs, when I do I shall almost certainly use oversampling to ease the design of the AAF. As far as I'm aware, ADCs don't suffer glitching so there's no pressing need to run them as slow as possible.
Correct about S-D DACs never being in practice non-oversampling. Very confusing to noobs.
Correct about S-D DACs never being in practice non-oversampling. Very confusing to noobs.
No?, double-, quad- and octa-rate DSD?
That's very relevant to audio, DSD is vastly more popular now than it was just three years back and it's still rising.
Edit - Not implying DSD is actually better, just taking note of the current trend.
Last edited:
Sorry, I made a clear error, just saw DSD512 is around 22 Mhz.
My initial thought thinking of the numbers was that it had already exceeded the speed of atomic-level interaction in our atmosphere, which, according to the link I provided is around 5 collisions per nanosecond.
All I wish to highlight is it's is the limit, the terminal velocity, the reality number.
Take 6? gigahertz divided by 22 Mhz, there we are, it's sampling pretty close to the ultimate limit nonethelesser =)
Now the Chord person, he says filter tap length of 1 million right? No further explanation nor detail as to how he arrived at that number.
In fact, in the head-fi post he writes "where is the audible limit? 10,000 taps? 1 million? 10 million? No one knows", elsewhere he writes he has a prototype DAC with some really high number, which "sounds even better".
Isn't he just using "infinity" for marketing? Which no one is questioning! Especially not over there.
Like I said, the Cirrus paper notes "10 microseconds" as the resolution level we can perceive in specific settings.
That's a university study which ended up in a Cirrus paper, looks pretty legit to me.
I will think about how 6 gigahertz relates to audio but the primary reason is just to know the limit numbers and cross reference to them.
______
Julf, I didn't follow your comment that 10 uS is around 5mm, is there a specific non-reality in perveiving a constant long string of 5mm or hair width sounds?
I'm not asking to be provocative, I just don't see any reason why it is unbelievable.
I.e. is there is a specific reason or is it just the intuitive likelihood?
______
Edit - Yes, this relates to the thread topic / thread title - time resolution.
It started with dynamic range resolution, which is fairly discussed in full apart from experiments, like the removal of micro dynamic range which would be interesting.
My initial thought thinking of the numbers was that it had already exceeded the speed of atomic-level interaction in our atmosphere, which, according to the link I provided is around 5 collisions per nanosecond.
All I wish to highlight is it's is the limit, the terminal velocity, the reality number.
Take 6? gigahertz divided by 22 Mhz, there we are, it's sampling pretty close to the ultimate limit nonethelesser =)
Now the Chord person, he says filter tap length of 1 million right? No further explanation nor detail as to how he arrived at that number.
In fact, in the head-fi post he writes "where is the audible limit? 10,000 taps? 1 million? 10 million? No one knows", elsewhere he writes he has a prototype DAC with some really high number, which "sounds even better".
Isn't he just using "infinity" for marketing? Which no one is questioning! Especially not over there.
Like I said, the Cirrus paper notes "10 microseconds" as the resolution level we can perceive in specific settings.
That's a university study which ended up in a Cirrus paper, looks pretty legit to me.
I will think about how 6 gigahertz relates to audio but the primary reason is just to know the limit numbers and cross reference to them.
______
Julf, I didn't follow your comment that 10 uS is around 5mm, is there a specific non-reality in perveiving a constant long string of 5mm or hair width sounds?
I'm not asking to be provocative, I just don't see any reason why it is unbelievable.
I.e. is there is a specific reason or is it just the intuitive likelihood?
______
Edit - Yes, this relates to the thread topic / thread title - time resolution.
It started with dynamic range resolution, which is fairly discussed in full apart from experiments, like the removal of micro dynamic range which would be interesting.
Last edited:
My point is that you cause a larger time delay by moving your ear 5mm. At the hair width level, any sort of air movement (such as caused by you breathing, or air conditioning, or draft, will cause time blurring much larger than your time interval.
So it was intended as a sanity check...
Calculus is one of the most fundamental mathematical tools for science - so it is what you are either though at a decent high school / upper school, or if not, then at first year in college. It has historically been called "the calculus of infinitesimals", or "infinitesimal calculus", because it is totally based on infinites and infinitesimals.
For an engineer, infinity is something very practical (and used every day), not anything abstract and theoretical.
Just to be clear - the last part in post #372 - taking a file with complex and intricate 64 dB dynamic range and flattening it all to let's say 4 dB in the second file is almost the same experiment.
Taking a 64 dB file and then changing it to a second file with 64 dB total and 4 dB strict steps is an example of what I have in mind.
It's still 64 dB, but the intra-step information is removed, that's the difference, just to clarify.
When I can and know how, I'll try it, should be fun.
In a transducer a string of pulses would be fun as well, just to hear at which point they transfer into a single pulse, due to the specific transducer limitations.
Just imho.
Taking a 64 dB file and then changing it to a second file with 64 dB total and 4 dB strict steps is an example of what I have in mind.
It's still 64 dB, but the intra-step information is removed, that's the difference, just to clarify.
When I can and know how, I'll try it, should be fun.
In a transducer a string of pulses would be fun as well, just to hear at which point they transfer into a single pulse, due to the specific transducer limitations.
Just imho.
Last edited:
Alright.
Sorry, but I don't think this illustration is accurate, a dog can still hear a dog whistle on a windy day =)
I am sure you can whistle with a dog whistle with a timing precision better than 95 ns...
The issue wasn't if the sound itself would be audible, but if the timing differences would.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.