[icebear]

In photo processing there is the option to "rez up" a file for printing. By means of various forms of interpolation (e.g. bi-cubic) a file of e.g. 10x15 inch and 120 dpi (18MP) can be printed to a size of 30x45 inch and keeping the resolution the same i.e. 120 dpi. (144MP).
 
Obviously the viewing distance can or even needs to be increased with the bigger picture to see it in it's entirety rather than pixel peeing with your nose against the print. Usually a printer has a max. dpi resolution which is defined by the droplet size and printing speed. Increasing the resolution beyond what the printer is capable of to get onto the paper is no problem with a today's powerful photo processing applications. Although it does obviously not make much sense ... does it?

 
So where am I going with this ...
 
Does it make any real world sense, other than marketing, to increase the frequency way into the MHz range when you simply can't step further away to hear the "bigger picture"? As the filters anyway shave the signal off above 20kHz, it's like you increased the file size enormously but can't transfer anything that is transferable onto the "audio canvas". You aren't actually printing that 30x45 inch print you stick to the original 10x15 size but did the "rez up" thing anyway. Now you have the 144MP and squeeze it into 10x15 which gives a theoretical 960dpi. ... in other words a printer error message.

 
You can measure audio signal differences, jitter in 2ms windows, distortion at -160dB (some claim) yet you can't pixel peep with your ears.
 
What's you take on the number's game?
Are we already way beyond what the printer can transfer to paper or in audio terms the speakers and headphones can transport to our ears?
 
Maybe this is more philosophical than scientific ... anyway, I listen to music not to numbers

 

 

[spruce music]

I really think we are in to apples and oranges here.  Where there are similarities it isn't in terms of increased frequency response.
 
Increasing resolution for prints or screens increases frequency of the lines spatially, but the frequency of the light itself is unaltered.
 
So going beyond 20 khz to mhz isn't at all like screen resolution.  It would be like shifting the viewable spectrum beyond normal into the ultraviolet ranges.
 
When you move closer to a print or screen the percentage of your field of view is increased which lets you see finer details spatially.
 
When you move closer to a real sound source you also get an angular increase along with a loudness increase.  This is like the difference in close miking or distant miking with a stereo pair of microphones.  You might not for instance spatially resolve distance between an oboist and clarinetist from 100 feet away with either your ears or a pair of mics.  Move to within 10 feet and you would. 
 
Now as a matter of perspective would you wish to view the oboist or clarinetist from 6 inches away?  Resolved detail is higher, but the perspective is unnatural and you can't even see the whole view of the instrument.  You do sometimes get close miking equivalent to that, but it too is unnatural in perspective.

 

[watchnerd]

 
 
Does it make any real world sense, other than marketing, to increase the frequency way into the MHz range when you simply can't step further away to hear the "bigger picture"? As the filters anyway shave the signal off above 20kHz, it's like you increased the file size enormously but can't transfer anything that is transferable onto the "audio canvas". You aren't actually printing that 30x45 inch print you stick to the original 10x15 size but did the "rez up" thing anyway. Now you have the 144MP and squeeze it into 10x15 which gives a theoretical 960dpi. ... in other words a printer error message.

 

 
Good analogies make things easier to understand.
 
Comparing audio to digital photos makes things harder to understand.
 
But to the core of your point:
 
Extending home audio playback into the ultrasonic makes just as much sense as extending my home television into the ultraviolet.

 

[icebear]

...OK it looks like the analogy more distracted than helped to get my point across, sorry

.
My point is what are all the high rez formats worth? These have been around now for ...10 years? Still no valid proof that any of these are a significant improvement over RB-CD for the end user. Point well taken that a higher rez recording gives more leeway for the production side of things.
 
But there is no end in sight of the numbers game, moar is battar

16bit, 20 bit, 24bit, 32bit, 64bit
44.1, 48, 96, 192, ...768, DXD
DSD2x, 4x, 8x.
164000, 500000, 1 Million taps...

 
Once the original sound has been captured, is it possible to improve the signal in the audible spectrum JUST by oversampling to a high rez format and NOT by altered mastering? I highly doubt that.

 

[watchnerd]

   
Once the original sound has been captured, is it possible to improve the signal in the audible spectrum JUST by oversampling to a high rez format and NOT by altered mastering? I highly doubt that.

 
Depends, if you're using a whacko non-oversampling DAC without a filter, not only does it help, it's a necessity (basically oversampling on a computer instead of in the DAC).
 
Otherwise, the preponderance of the evidence is that Redbook and high resolution PCM of the same master are transparent to one another in ABX tests.
 
There are even some arguments out there for why 24/192khz is actually bad: 
 
https://people.xiph.org/~xiphmont/demo/neil-young.html

 

[skhan007]

Can one of you please provide a simplistic definition of Tap Length? I'm being told in another thread about a DAC that is going to produce 1 million taps and this is, apparently, very important. Sounds impressive, but what does it mean? Something to do with a piece of data being analyzed/filtered 1 million times? How does this make a given song sound better? Thanks in advance!

 

[spruce music]

  Can one of you please provide a simplistic definition of Tap Length? I'm being told in another thread about a DAC that is going to produce 1 million taps and this is, apparently, very important. Sounds impressive, but what does it mean? Something to do with a piece of data being analyzed/filtered 1 million times? How does this make a given song sound better? Thanks in advance!

Okay, no expert myself and hopefully this isn't too simplified to illuminate.
 
Imagine we are doing a 5 tap filter.  What does that mean?  It means we fill a buffer with 5 samples.  We do math operations to the samples in the buffer.  We output the result of the math done on the 5 samples in the buffer.  We take a new sample into the buffer, and push the earliest sample out of the buffer. Repeat the math again, so on and so forth.
 
So just a super simple filter in this example averages all 5 sample values and puts out the averaged value of 5 samples.  Next sample comes along, we push out one sample, get a new sample in buffer position one and average those five samples and put out a new sample on the new average.  This will end up being essentially a low order low pass filter.  You cannot jump from the lowest sample value to the highest at the output in one sample because it will get averaged out to some lower value for the 5 most recent samples. 
 
Now it can get much more complicated.  I might average 5 samples and add that to the next incoming sample and average the new total.  That will be a different kind of low pass filter. So there can be a recursive quality.  Now once you get to implementing steep or complex filtering the mathematical operations on the various bits gets complicated.  At some point there is a limit to using a 5 tap(5 sample) digital filter and errors can occur or at least irregularities in the filter output.   If I go to a 10 tap (read 10 sample) filter to perform the math then results can be more precise and/or more complex.  A 100 tap filter is even better still.  A million tap filter even better still. 
 
To those who know digital filters yes I am taking some liberties so unless you think I have given the wrong idea please allow me that. 
 
Just for your information many DAC chips with their onboard digital filtering have 128 or 256 tap filters. 
 
Does going to 1 million taps make an audibly better result?   I doubt it.  It may make more a more nearly perfect result according to theory and one that can be measured as a minor improvement.  Audibility I think will be at best very small and possibly nothing.

 

[spruce music]

Some illustrations of filter taps.  This is from the Equalization in Audacity.  You can select between 21 taps and 8192 taps for EQ. 
 
First up is a 21 tap filter which is a shelving EQ which drops 24 db in one octave from 1khz to 2 khz. The blue line is what I am asking for.  The green line is what the actual filter will be with 21 taps.  You will notice it lowers the level and rounds of the EQ curve I am requesting as it can't fully do the EQ with only 21 taps to work with.  You can click on the images to see a larger version.
 

 
Here is the same EQ using 253 taps and it almost perfectly creates the EQ response I am asking to get.
 

 
Now using 21 taps I ask for a -120 db shelve over just one octave.  Getting close to the digital brickwall filters.  You will see the very uneven response higher up in frequencies above the 2 khz range.  21 taps isn't enough to do a good job with this.
 

 
Now I do the same thing with 253 taps and the result is much better though still not perfect.  Even 253 taps is inadequate for this steep a filter.
 

 
Now I use 1817 taps which is capable of giving almost exactly what I am asking to get.
 

 

[skhan007]

Okay, no expert myself and hopefully this isn't too simplified to illuminate.

Imagine we are doing a 5 tap filter.  What does that mean?  It means we fill a buffer with 5 samples.  We do math operations to the samples in the buffer.  We output the result of the math done on the 5 samples in the buffer.  We take a new sample into the buffer, and push the earliest sample out of the buffer. Repeat the math again, so on and so forth.

So just a super simple filter in this example averages all 5 sample values and puts out the averaged value of 5 samples.  Next sample comes along, we push out one sample, get a new sample in buffer position one and average those five samples and put out a new sample on the new average.  This will end up being essentially a low order low pass filter.  You cannot jump from the lowest sample value to the highest at the output in one sample because it will get averaged out to some lower value for the 5 most recent samples. 

Now it can get much more complicated.  I might average 5 samples and add that to the next incoming sample and average the new total.  That will be a different kind of low pass filter. So there can be a recursive quality.  Now once you get to implementing steep or complex filtering the mathematical operations on the various bits gets complicated.  At some point there is a limit to using a 5 tap(5 sample) digital filter and errors can occur or at least irregularities in the filter output.   If I go to a 10 tap (read 10 sample) filter to perform the math then results can be more precise and/or more complex.  A 100 tap filter is even better still.  A million tap filter even better still. 

To those who know digital filters yes I am taking some liberties so unless you think I have given the wrong idea please allow me that. 

Just for your information many DAC chips with their onboard digital filtering have 128 or 256 tap filters. 

Does going to 1 million taps make an audibly better result?   I doubt it.  It may make more a more nearly perfect result according to theory and one that can be measured as a minor improvement.  Audibility I think will be at best very small and possibly nothing.

Thank you!! Ok, it sounds like diminishing returns, as at some point, the benefit of increased tap length may result in statistical accuracy, but it may not be audible. Sounds similar to statistical vs. clinical significance in medical research.

 

[skhan007]

Some illustrations of filter taps.  This is from the Equalization in Audacity.  You can select between 21 taps and 8192 taps for EQ. 

First up is a 21 tap filter which is a shelving EQ which drops 24 db in one octave from 1khz to 2 khz. The blue line is what I am asking for.  The green line is what the actual filter will be with 21 taps.  You will notice it lowers the level and rounds of the EQ curve I am requesting as it can't fully do the EQ with only 21 taps to work with.  You can click on the images to see a larger version.




Here is the same EQ using 253 taps and it almost perfectly creates the EQ response I am asking to get.




Now using 21 taps I ask for a -120 db shelve over just one octave.  Getting close to the digital brickwall filters.  You will see the very uneven response higher up in frequencies above the 2 khz range.  21 taps isn't enough to do a good job with this.




Now I do the same thing with 253 taps and the result is much better though still not perfect.  Even 253 taps is inadequate for this steep a filter.




Now I use 1817 taps which is capable of giving almost exactly what I am asking to get.

Oh wow, this is cool to see, visually. I'm going to study this a bit to absorb it all. Thank you!!

 

[watchnerd]

  Can one of you please provide a simplistic definition of Tap Length? I'm being told in another thread about a DAC that is going to produce 1 million taps and this is, apparently, very important. Sounds impressive, but what does it mean? Something to do with a piece of data being analyzed/filtered 1 million times? How does this make a given song sound better? Thanks in advance!

 
Is this DAC using a DSP processors instead of a dedicated DAC chip?
 
Usually "tap count wars" are the domain of DSP processors like SHARC.

 

[spruce music]

   
Is this DAC using a DSP processors instead of a dedicated DAC chip?
 
Usually "tap count wars" are the domain of DSP processors like SHARC.

I imagine it is the new Chord Blu MkII Mscaler being discussed.  Soon to be available with 1,015,808 Taps. It will involve beyond just filtering upsampling and such. 
 
 
http://www.chordelectronics.co.uk/product/blu-mk-2/
 
All with the inconvenience of actual disc replay.  It does have digital ins and outs so disc replay is not the only method of use.  Yet not 48 khz multiples which is a strange omission.
 
No word on pricing though rumored to be in the $20,000 range.

 

[watchnerd]

 
I imagine it is the new Chord Blu MkII Mscaler being discussed.  Soon to be available with 1,015,808 Taps. It will involve beyond just filtering upsampling and such. 
 
 
http://www.chordelectronics.co.uk/product/blu-mk-2/
 
All with the inconvenience of actual disc replay.  It does have digital ins and outs so disc replay is not the only method of use.  Yet not 48 khz multiples which is a strange omission.
 
No word on pricing though rumored to be in the $20,000 range.

 
I find this fascinating.
 
It appears that more and more DAC makers have decided they need to differentiate themselves off-the-shelf DS chip in a box (possibly because of all the super cheap stuff Asia) by innovating on the digital side (as opposed to on the analog side).
 
More and more seem to be implementing custom filters and other 'beyond Redbook' notions. Chord appears to be one, the rumored upcoming Schiit Manhattan might be another.

 

[RRod]

 
I imagine it is the new Chord Blu MkII Mscaler being discussed.  Soon to be available with 1,015,808 Taps. It will involve beyond just filtering upsampling and such. 

 
1.21 Gigataps!

 

[spruce music]

1.21 Gigataps!

1.21 megataps?