Theoretical benefits to hi-res if only max 20kHz signal?

[adamdea]

Just create yourself such an impulse and apply a lowpass or resample it. The post and preringing only happens around fs/2 so above 20kHz. I did some and with a gentle filter kicking in around 20khz there is virtualy no ringing left.
Of cause the picture of simply the impulse, not its spectrum is welcome for marketing

I agree that the impulse response is very misleading since it represents the response of the filter to an impossible input.
BUt still the effect of pre-ringing is confusing. Although it represents to effect of the filter cutting off frequencies above the stop band AFAIK it supposedly spreads out energy throughout the passband. This is where i get confused because it supposedly does so even though the filter has a flat response in the passband- how can it do that (PAGING TEDDY RAY)

[adamdea]

You also need to consider the sampling rate of the human ear and the "inter-sample" interpolation (and all of the other heavyweight processing) in the brain...

Absolutely, and i think that saying you can detect the time domain effects is just sneaking through the back door the argument that you can hear over 20KhZ

[Phil Leigh]

Absolutely, and i think that saying you can detect the time domain effects is just sneaking through the back door the argument that you can hear over 20KhZ

Agreed. Most of this is hogwash dreamt up by people who heard early digital and felt it was bad (which it mostly was)... And then invented a whole fake mythology about why it was bad and why it would never be good.
Fools.

[Wombat]

I agree that the impulse response is very misleading since it represents the response of the filter to an impossible input.
BUt still the effect of pre-ringing is confusing. Although it represents to effect of the filter cutting off frequencies above the stop band AFAIK it supposedly spreads out energy throughout the passband. This is where i get confused because it supposedly does so even though the filter has a flat response in the passband- how can it do that (PAGING TEDDY RAY)

I lately did some in a different forum, first some kind of Apodizing against steep linear.
http://imageshack.us/photo/my-images/718/impulsapo.png/
Second is a gentle linear:
http://imageshack.us/photo/my-images/12/sox90a.png/

You see it shouldn´t touch the audible band at all.

[adamdea]

This is very intersting- what does the unfiltered signal look like?

[Wombat]

Original is 24/96 full scale impulse that looks exactly like the signal below 20khz on these linears up to 48kHz.

[evdplancke]

Nyquist theorema tells there is no benefits but is based on the following assumptions:
1) sampled signal is made of perfect dirac pulses... practically they will look like square pulses resulting in a quite bad low pass sin x/x filtering that zeroes at 1/T where T is the width of the pulse (staircase sampled signal being the worst with 1/T equals to sample freq)
2) sampled pulse has a constant time period... but in reality time period is varying due to jitter
3) signal reconstruction is using perfect brickwall filters... but in practice it is very difficult to have steep filter slope without artifacts

In conclusion, there might be a (big?) gap between theory and practice.

According to the 3 assumptions above, here are the (audible?) benefits of 96khz:
1) higher sample freq is reducing the low pass effect of square pulses, rejecting the resulting zeroes to higher freq
2) doubling the number of pulses adds redundancy to the signal that increases immunity to jitter
3) less steep brickwall filters are needed for signal reconstruction, reducing artifacts and improving linearity

So hi res might technically improve the accuracy of reconstructed signal even if baseband signal does not contain any freq above 20 khz.

The question remains however: are those benefits audible or not? I'd say it makes it at least more robust against less than optimal receiver implementation.

[Phil Leigh]

Nyquist theorema tells there is no benefits but is based on the following assumptions:
1) sampled signal is made of perfect dirac pulses... practically they will look like square pulses resulting in a quite bad low pass sin x/x filtering that zeroes at 1/T where T is the width of the pulse (staircase sampled signal being the worst with 1/T equals to sample freq)
2) sampled pulse has a constant time period... but in reality time period is varying due to jitter
3) signal reconstruction is using perfect brickwall filters... but in practice it is very difficult to have steep filter slope without artifacts

In conclusion, there might be a (big?) gap between theory and practice.

According to the 3 assumptions above, here are the (audible?) benefits of 96khz:
1) higher sample freq is reducing the low pass effect of square pulses, rejecting the resulting zeroes to higher freq
2) doubling the number of pulses adds redundancy to the signal that increases immunity to jitter
3) less steep brickwall filters are needed for signal reconstruction, reducing artifacts and improving linearity

So hi res might technically improve the accuracy of reconstructed signal even if baseband signal does not contain any freq above 20 khz.

The question remains however: are those benefits audible or not? I'd say it makes it at least more robust against less than optimal receiver implementation.

There's still no irrefutable evidence that anyone can hear the difference - you'd think by now the Net would have at least a few conclusive studies....?

[adamdea]

Nyquist theorema tells there is no benefits but is based on the following assumptions:
1) sampled signal is made of perfect dirac pulses... practically they will look like square pulses resulting in a quite bad low pass sin x/x filtering that zeroes at 1/T where T is the width of the pulse (staircase sampled signal being the worst with 1/T equals to sample freq)
2) sampled pulse has a constant time period... but in reality time period is varying due to jitter
3) signal reconstruction is using perfect brickwall filters... but in practice it is very difficult to have steep filter slope without artifacts

In conclusion, there might be a (big?) gap between theory and practice.

According to the 3 assumptions above, here are the (audible?) benefits of 96khz:
1) higher sample freq is reducing the low pass effect of square pulses, rejecting the resulting zeroes to higher freq
2) doubling the number of pulses adds redundancy to the signal that increases immunity to jitter
3) less steep brickwall filters are needed for signal reconstruction, reducing artifacts and improving linearity

So hi res might technically improve the accuracy of reconstructed signal even if baseband signal does not contain any freq above 20 khz.

The question remains however: are those benefits audible or not? I'd say it makes it at least more robust against less than optimal receiver implementation.

I am not really sure what you mean by 1)except as a corollary of 3).
and I'm not sure about 2) either -surely a higher sampling rate would require more accurate timing in the ADC? not that I think this is an issue.

The real point is that doubling fs halves the quantisation noise in the audible spectrum.This the equivalent of adding 1 bit albeit very inefficiently since you have doubled the data rate whilst increasing the amount of information by the same amount you could have achieved by adding one bit which would have increased the data rate by 1/16.It does give you some room to play with for noise shaping though.

I think 3) is the real time domain issue. But it's worth pointing out that the fundamental issue is that increasing fs enables you to have less steep filter in the anti alias stage of ADC. But it all begs the question - given the ability to create very steep digital filters with negligible passband ripple or phase issues, what are the artefacts?

I note however that daniel weiss indicated in an interview that he thought 16/88 would be more beneficial as a consumer format than 24/48. I don't think he spelt this out but I assume it was because he thought there was chance that there was something in the time domain issues.

Even then though it's worth stressing that there should not be any time domain issue if the pre-filtering pre ADC signal has no energy above 20kHz - there will not be any time smear then![I Hold my breath and wait to be shown up as an idiot]

Equally If there is no signal above 20KHz in the 24/96 file then either there was none in the pre recording sound [no time smear with sensible filtering] or there was but it has been filtered out anyway ie the time smear will have occured. . IN that case I can't see how there could be any worthwhile improvement in a 96kHz file over a 44.1 kHz.

[evdplancke]

I am not really sure what you mean by 1)except as a corollary of 3).

1) is different from 3) because the low pass effect will affect the baseband signal: a convolution of a square pulse with perfect dirac pulse train in the time domain corresponds to multiplying the signal in the frequency domain by a very poor sinx/x lowpass filter. This has nothing to do with the reconstruction filter used for 3): this is a lowpass distortion BEFORE reconstruction, that is definitely altering the baseband signal in an extent inversely proportional to the square pulse width.