Fair enough. Just wanted to make it clear that the (non-)issue of timing resolution, as stated by the sampling theorem, is still valid, irrespective of if the reconstruction filter is a well-designed one or an accidental, less effective one.
I have a vague memory that Bruno showed in one of his 'masterclass PPTs' that without filtering, the timing resolution of any NOS DAC gets shot to pieces, no matter the sample rate.
Perhaps that's just another way of saying though that inadequate bandlimiting for a DAC will have consequences in terms of timing resolution. In which case perhaps Kastor has a point if he's claiming that beyond the walled garden of Shannon-Nyquist, in practice higher sample rates get subject to less timing degradation (since bandlimiting is never perfect). Bit of a long shot for what he might be saying though?
Perhaps that's just another way of saying though that inadequate bandlimiting for a DAC will have consequences in terms of timing resolution. In which case perhaps Kastor has a point if he's claiming that beyond the walled garden of Shannon-Nyquist, in practice higher sample rates get subject to less timing degradation (since bandlimiting is never perfect). Bit of a long shot for what he might be saying though?
But I guess the key part there is the "no matter the sample rate".
Perhaps - but again, sticking to Shannon-Nyquist is the way to go.
It is of course more important on the input side than on the output - our ears are pretty good low pass filters.
Yes, Nyquist-Shannon defines the point at which the file has all time resolution information in it.
However, we then need to display that time rez information in the final result, like the transducer.
Let's say we want 21 kHz, then 48,000 and 96,000 have both surpassed the limit of "all information", which was at 42,001 per second.
However, the excess of information in the 192,000 file should result in higher time accuracy, in the transducer, if the DAC is really weak.
For instance 2x oversampling and a reconstruction filter with tap length = 16.
There is a terminal velocity at which point the transducer can not alter air faster.
Let's just say it's 21 nanoseconds, then what is the equivalent tap length of 21 nanoseconds?
That seems to define a more realistic tap length than "infinity", at which point the speaker is in theory moving faster than heat inside the sun.
However, we then need to display that time rez information in the final result, like the transducer.
Let's say we want 21 kHz, then 48,000 and 96,000 have both surpassed the limit of "all information", which was at 42,001 per second.
However, the excess of information in the 192,000 file should result in higher time accuracy, in the transducer, if the DAC is really weak.
For instance 2x oversampling and a reconstruction filter with tap length = 16.
There is a terminal velocity at which point the transducer can not alter air faster.
Let's just say it's 21 nanoseconds, then what is the equivalent tap length of 21 nanoseconds?
That seems to define a more realistic tap length than "infinity", at which point the speaker is in theory moving faster than heat inside the sun.
Last edited:
And that "terminal velocity" defines the power bandwidth of your system. If your transducer can't move fast enough to reproduce your 21 kHz sine wave at the amplitude you want, adding harmonics to the signal won't help.
Perhaps in a testing environment that accuracy can be proven, but since the wave-form is reconstructed in the reconstruction-filter, I doubt the improvement will be audible.
If we take two extremes of world view
"Entertain that anything is possible, until proven otherwise"
versus
"Entertain that nothing is true, until proven"
Then neither of them will alter the air movement, in any way, but perhaps that's what you were saying with your comment there =)
Yes, this is one aspect of transducer velocity, but doesn't a transducer need a certain velocity to achieve really low THD as well?
For instance a 1 kHz sine, I can see that "very fast" transducer designs tend to achieve lower THD.
If you know the answer to this it will help a lot, will 64x oversampling / interpolating a 1 kHz sine achieve lower THD at all? Versus non-oversampling or let's say 4x.
Escape velocity?
"Oh, your speaker only does 150 km/h? That can't sound very good!"
Any pointers to references?
No (assuming you measure the THD from 20 Hz to 20 kHz, as you should).
I think you answered Tattoo's post earlier where he discussed Nyquist-Shannon --> oversampling --> reconstruction filter.
If you compare the signal after the filter to reality it seems to ideally need to be minimum-phase to look / perform identical.
There's links to the Cirrus Logic, Wolfson and Julian Dunn papers concerning that earlier in this thread.
The second part, at least in Tattoo's post, is the value "infinity" versus "acceptable" in the filter.
When I noted "A weak DAC" I'm implying a filter which has not fulfilled "acceptable" yet, let alone "perfection" at "infinity".
I don't personally think "infinity" is the correct value, here's why, imagine a speaker deep in the ocean, at a certain point in that specific environment the speed / force / velocity in the speaker movement will cause air pockets which collapse, then convert into different energy, like light!
That would define the "terminal velocity" =)
A filter tap length of let's say 64 is inaccurate and "not acceptable yet", at least according to the theory in Tattoo's post.
If we assume the above, then either 48,000 versus 192,000 samples per second will perform equally inaccurate, or the 192,000 file will display higher accuracy or less accuracy.
Last edited:
Yes, sure, that is what designers / users with electrostatic or balanced-armature drivers may say at times.
I've asked in the past why electrostatic sounds different and can't find any real answers.
If speed is only "moves fast enough for 21 kHz" like you say, then it's not speed, is it?
Then which parameter makes electrostatic and dynamic always sound different?
Julf said:
I'll post a few links to dynamic drivers in the speaker cone thread alater then link to it from here.
Julf said:
Alright.
With X, Y, Z, X = frequency, Y = amplitude, Z = time
Just like any CSD chart
Then I'll assume oversampling / interpolation increases the bandwidth in X and increases the samples per second in Z.
THD is only measuring X, oversampling doesn't affect the THD, it only lowers the noise in X at times, THD+N.
The samples per second in Z is increased, interpolated, like 6x times more samples per second with 6x oversampling.
However they are not "real" samples, just interpolation.
The dispute is if they result in higher perceived detail or if they are invisible.
Alright, that all makes sense.
Hi Fotis,
I think there may be a few answers in this thread
http://www.diyaudio.com/forums/digi...-nos-192-24-dac-pcm1794-waveio-usb-input.html
I can't check it since I'm limited to my phone right now. I'm not sure what they've done precisely nor why, if you know please tell us.
What are the available OSR settings in the WM8741 and the PCM1792 by the way?
Thx
Last edited:
Not sure if you're familiar with the concept of group delay, but here's a fun fact:
Linear phase filters delay all frequency components of the signal by the same amount. Minimum phase filters do not. Minimum phase filters delay different frequencies by different amounts.
It should be obvious that if different frequencies are delayed by different amounts, then the shape of the waveform will be changed. i.e. If you want the signal after the filter to look the same, you need a linear phase filter, not a minimum phase filter.
From the wiki article:
Sure. Lots of people say strange things to promote/defend their products.
Radiation/directivity patterns and lack of pronounced resonances?
And the lowering of noise is not due to oversampling, but noise shaping.
No dispute among those who have actually studied the stuff.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.