We are very quick to say that human hearing is in the range of 20-20khz but I wonder how many people can actually hear to those extremes.

When I was 40 my upper limit was about 17khz now at 61 it's down to 13.6khz. I reckon there's a lot of psychoacoustic/placebo stuff going on which accounts for lots of things. I'm not doubting the math just people's ears.

rabeach
I never stated that a perfect reconstruction was necessary. Nor did I indicate the interpolation error could be heard. I simply stated a perfect reconstruction does not exist.

 
How do you define "perfect" in the real world? My position is in the real world "perfect" means any imperfections are completely lost in other bigger imperfections that exist in the real world. IOW, far enough below the noise floor due to other problems to be meaningless.
 

And invoking a theorem as justification for not sampling at higher frequencies is flawed in my opinion because the math required for the theorem to be held true is not implemented in real-world systems. So whether sampling at a higher frequency is beneficial on the many varied ADC to DAC systems should not be based on a perceived belief that a perfect reconstruction is occurring.

 
If the imperfections are far enough below other imperfections in the real world that they are completely buried, then they can't be improved upon. Period.
 
Perfectly reconstruction a waveform to the point where any errors are buried beneath the bit depth is only difficult at higher frequencies, at high levels.
 

 
That aside sampling at higher frequencies may sound better on some systems and not on others. My aardvark 24/96 was designed by an extremely competent engineer had very stable clocking and filters with extremely low noise and low distortion. It sounded great at 96kHz although I never worked with it at that frequency. My VS-100 sounds great at 44.1kHz (doesn't touch the aardvark though) but I don't care for the sound it has at 96kHz. Stable clocking and filter design are extremely challenging and are not implemented very well in most commercial systems.

 
Clocking is just not an issue in the real world today. Find a real world converter that you think has clocking problems and try and measure the distortion/noise due to jitter. Good luck with that. 
 
Filter design trade offs are very well understood by many, many people. And it's not hard to evaluate a filter. They pretty much show everything you need to know for SRC filters here:
 
http://src.infinitewave.ca/
 
You will find examples there where, even when clearly visible, none of the issues on any chart for a given algorithm will make it into 24bit audio - IOW it is perfect at 24bit resolution up to a certain frequency where it starts rolling off. Some others only really have issues above 18kHz or so. Most of the really, really poor ones look like they didn't even try (perhaps choosing to use as little CPU as possible instead).
 
Apparently filter design is not so challenging that many, many vendors haven't managed to be able to get it right (at least for SRC).
 
If the same converters do indeed sound different at different sample rates under objective conditions (i.e. under careful double blind testing or quantitative measurement), the interesting question is how the results differ and what changed under the hood and why the vendor made the choices they did. Since we don't generally know, we can only speculate. Here's some speculation: perhaps a vendor might have traded off quality for lower latency, and maybe that was a bad decision.
 

In my opinion empirical data collected on the varied systems in use trumps the belief that a perfect reconstruction is occurring and all systems will sound the same at 44.1kHz. But if the empirical data is ever collected and says otherwise I will promptly reverse my opinion.
 

 
If it's possible for gear to achieve perfect (meaning here any imperfections are buried in the existing noise floor) reconstruction in the real world over the desired frequency range, then it shows that sampling frequency is not the problem.
 
Different gear will have many, many design trade offs so not everything will necessarily sound the same at any SR. And indeed it makes sense to evaluate the trade offs together rather than individually, particularly when we have no way of doing getting at the individual pieces.
 
And if a given piece of gear sounds better (or worse) at one SR vs. another, then it is what it is. 
 
But some vendors making that design trade off for whatever reason, does not mean it's difficult or impossible based on the SR itself. If you can show that is impossible, well by all means go ahead.
 
IMO the bigger problem is when people jump to the wrong conclusions based on limited information and without a full understanding of the technical trade offs and limitations.
 

Math is just a construct many people experience a sensory variance from using higher sampling frequencies. 
 

 
Digital audio is math. Once the signal is digitized, aside from bugs, HW problems and incompetence there is nothing other than math until it gets converted back to analog.
 
If people are hearing a difference due to a digital decimation or interpolation filter, then it is purely a math problem.
 

There is a Burr Brown white paper that shows the implementation of a linear-phase filter somewhere in between a Butterworth and a Bessel response; It may be outdated by now I came across it in 1994.
 
http://www.ti.com/lit/an/sbaa001/sbaa001.pdf

The filter design trade offs are the same today, but the trade off involving available processing power has of course changed dramatically. There are also better tools readily available today for filter design and optimization than in 1994. 

YouDontHasToCallMeJohnson
 
I would go to 88 if I knew all plugs and samples would play together. As 88 is the closest to Lavry's "preferred/recommended" frequency of 60.  But, I have no time to trouble shoot doing so now. 

 
I believe that Lavry was talking more about straight PCM for his preferred rate rather than the massively oversampled sigma-delta converters commonly used today. If you're really worried about inaudible ultrasonics causing IM distortion when using 96kHz, and if you actually had any significant content up there, you could always just filter it out.
 

What about:
Up-sampling to 48?
Up-sampling 44 to 96?
 
Most vid, and tv, users will not be able to hear any difference.
 
Does this matter?
 

Sonar's sample rate conversion (SRC) is excellent and will convert from 44.1kHz to any of the higher rates with no issues.

YouDontHasToCallMeJohnson
After reading the thread, and all the referenced articles I agree with the 'sediment' of some who have written (paraphrased), "this is all moot because most of the sample libraries I use were recorded at 44. So, I will stay with 44 until I am beaten about the head, neck, and chest with a blunt object, or buying new samples. Whichever comes first."
 
I have dozens of unfinished projects. All at 44.
 
Not remembering to change windows and interface and Sonar settings to/from 44/48/88/96 is NOT something I want to experience. Already I exceed my weekly quota of loud exclamational words during a much shorter period.
 
I would go to 88 if I knew all plugs and samples would play together. As 88 is the closest to Lavry's "preferred/recommended" frequency of 60.  But, I have no time to trouble shoot doing so now. 
 
So, near future, when needed, I will choose a new interface that is reviewed to have best sound at 44, that I can afford.
 
What about:
Up-sampling to 48?
Up-sampling 44 to 96?
 
Most vid, and tv, users will not be able to hear any difference.
 
Does this matter?
 
Anybody? Anybody? Buehler? Buehler?
 
 

My conclusion was reached based on Lavry's range of 50-60K given as optimal.  Combined with my desire for less expensive hardware, 48K is close enough to 50K for me.  If my hardware wasn't performing extremely well at 48K, I would prolly go back to 44.1K and be 99.999% as happy  :)
 
Something about "sounding better doesn't mean more transparent" sticks with me too.  I feel like if I wanted to go to 96K and beyond, I would want to do a lot of research on interfaces - and expect to pay a lot of money for an interface that was truly transparent at those rates.
 
I would also expect to do a lot of upgrades to ancillary gear as well.
 
Not a trip I want to take, BUT - I'm an amateur home studio guitar player.  It's easy for me to short-change myself  :)
 
 

drewfx1

rabeach
I never stated that a perfect reconstruction was necessary. Nor did I indicate the interpolation error could be heard. I simply stated a perfect reconstruction does not exist.

 
How do you define "perfect" in the real world? My position is in the real world "perfect" means any imperfections are completely lost in other bigger imperfections that exist in the real world. IOW, far enough below the noise floor due to other problems to be meaningless.
 

And invoking a theorem as justification for not sampling at higher frequencies is flawed in my opinion because the math required for the theorem to be held true is not implemented in real-world systems. So whether sampling at a higher frequency is beneficial on the many varied ADC to DAC systems should not be based on a perceived belief that a perfect reconstruction is occurring.

 
If the imperfections are far enough below other imperfections in the real world that they are completely buried, then they can't be improved upon. Period.
 
Perfectly reconstruction a waveform to the point where any errors are buried beneath the bit depth is only difficult at higher frequencies, at high levels.
 

 
That aside sampling at higher frequencies may sound better on some systems and not on others. My aardvark 24/96 was designed by an extremely competent engineer had very stable clocking and filters with extremely low noise and low distortion. It sounded great at 96kHz although I never worked with it at that frequency. My VS-100 sounds great at 44.1kHz (doesn't touch the aardvark though) but I don't care for the sound it has at 96kHz. Stable clocking and filter design are extremely challenging and are not implemented very well in most commercial systems.

 
Clocking is just not an issue in the real world today. Find a real world converter that you think has clocking problems and try and measure the distortion/noise due to jitter. Good luck with that. 
 
Filter design trade offs are very well understood by many, many people. And it's not hard to evaluate a filter. They pretty much show everything you need to know for SRC filters here:
 
http://src.infinitewave.ca/
 
You will find examples there where, even when clearly visible, none of the issues on any chart for a given algorithm will make it into 24bit audio - IOW it is perfect at 24bit resolution up to a certain frequency where it starts rolling off. Some others only really have issues above 18kHz or so. Most of the really, really poor ones look like they didn't even try (perhaps choosing to use as little CPU as possible instead).
 
Apparently filter design is not so challenging that many, many vendors haven't managed to be able to get it right (at least for SRC).
 
If the same converters do indeed sound different at different sample rates under objective conditions (i.e. under careful double blind testing or quantitative measurement), the interesting question is how the results differ and what changed under the hood and why the vendor made the choices they did. Since we don't generally know, we can only speculate. Here's some speculation: perhaps a vendor might have traded off quality for lower latency, and maybe that was a bad decision.
 

In my opinion empirical data collected on the varied systems in use trumps the belief that a perfect reconstruction is occurring and all systems will sound the same at 44.1kHz. But if the empirical data is ever collected and says otherwise I will promptly reverse my opinion.
 

 
If it's possible for gear to achieve perfect (meaning here any imperfections are buried in the existing noise floor) reconstruction in the real world over the desired frequency range, then it shows that sampling frequency is not the problem.
 
Different gear will have many, many design trade offs so not everything will necessarily sound the same at any SR. And indeed it makes sense to evaluate the trade offs together rather than individually, particularly when we have no way of doing getting at the individual pieces.
 
And if a given piece of gear sounds better (or worse) at one SR vs. another, then it is what it is. 
 
But some vendors making that design trade off for whatever reason, does not mean it's difficult or impossible based on the SR itself. If you can show that is impossible, well by all means go ahead.
 
IMO the bigger problem is when people jump to the wrong conclusions based on limited information and without a full understanding of the technical trade offs and limitations.
 

Math is just a construct many people experience a sensory variance from using higher sampling frequencies. 
 

 
Digital audio is math. Once the signal is digitized, aside from bugs, HW problems and incompetence there is nothing other than math until it gets converted back to analog.
 
If people are hearing a difference due to a digital decimation or interpolation filter, then it is purely a math problem.
 

There is a Burr Brown white paper that shows the implementation of a linear-phase filter somewhere in between a Butterworth and a Bessel response; It may be outdated by now I came across it in 1994.
 
http://www.ti.com/lit/an/sbaa001/sbaa001.pdf

The filter design trade offs are the same today, but the trade off involving available processing power has of course changed dramatically. There are also better tools readily available today for filter design and optimization than in 1994. 

I understand your position on real-world perfection.
 
I would think a real time oscilloscope with a 6GHz bandwidth should be able to analyze jitter with regards to audio ADC to DAC systems. Even if employing Sigma Delta A/D converters at 64 x sampling frequency, I would think the real time oscilloscope would do the job. What am I missing? 
 
I would hope there have been advances in filter design over the past 24 years. If there have been significant advances to the analog ones shown in the Burr Brown document, I would be interested in reviewing the white paper design and included bode plots.