• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Proper Definition of High-Resolution Music

CuteStudio

Active Member
Joined
Nov 23, 2017
Messages
119
Likes
65
it is inaccurate and IMHO misleading to refer to upsampling as decompression. Upsampling does not decompress anything or add anything.

A couple of misundstandings here, not of the technology but of what I was trying to say.
My experience is that upsampled audio sounds better, this is through Behringer Ultramatch and Ultracurve boxes, and is I suspect simply a function of the easier analog filtering offered by the higher rate. Without knowing the internals I cannot however be sure, perhaps it's a different effect.

Upsampling I view as decompression. Analog data has infinite points per second, but we only need X per second for any given max frequency as per Nyquist. What actually happens in the upsampling is that the software adds points in - in the case of 2x upsampling it calculates and adds points between the original ones (ideally not moving the original ones). I've programmed this stuff with convolutions - you end up with more (or less if you want!) points. This in my mind is adding something, it's not really extra data, it's simply calculating what would have been there if the sample rate was higher.

Upsampling is only really done for easier filtering in my view, we don't technically _need_ the extra points, but it helps if we add them in mathematically and then pass the result it through a simpler analog filter to recreate the infinite samples per second of Analog. Then because of the intermediate points each step is less in magnitude and a decent (non brick wall) filter can operate.

I hope that answers my view of the digitising and re-conversion back to analog as a form of compression -> store/streaming -> expansion.
 

tmtomh

Major Contributor
Forum Donor
Joined
Aug 14, 2018
Messages
2,636
Likes
7,488
A couple of misundstandings here, not of the technology but of what I was trying to say.
My experience is that upsampled audio sounds better, this is through Behringer Ultramatch and Ultracurve boxes, and is I suspect simply a function of the easier analog filtering offered by the higher rate. Without knowing the internals I cannot however be sure, perhaps it's a different effect.

Upsampling I view as decompression. Analog data has infinite points per second, but we only need X per second for any given max frequency as per Nyquist. What actually happens in the upsampling is that the software adds points in - in the case of 2x upsampling it calculates and adds points between the original ones (ideally not moving the original ones). I've programmed this stuff with convolutions - you end up with more (or less if you want!) points. This in my mind is adding something, it's not really extra data, it's simply calculating what would have been there if the sample rate was higher.

Upsampling is only really done for easier filtering in my view, we don't technically _need_ the extra points, but it helps if we add them in mathematically and then pass the result it through a simpler analog filter to recreate the infinite samples per second of Analog. Then because of the intermediate points each step is less in magnitude and a decent (non brick wall) filter can operate.

I hope that answers my view of the digitising and re-conversion back to analog as a form of compression -> store/streaming -> expansion.

Thanks for your reply. This all makes sense, but I still don't see how it makes sense to call this decompression. Any predictive power of an upsampling algorithm will be mooted by the fact that the original sample-rate source, per Nyquist, already has sufficient data points/samples to accurately "predict" aka reconstruct the analogue waveform in the DAC process. And even putting that aside, I'm still not understanding how such predictive power in the digital realm can be understood as a form of decompression.
 

RayDunzl

Grand Contributor
Central Scrutinizer
Joined
Mar 9, 2016
Messages
13,200
Likes
16,981
Location
Riverview FL
Analog data has infinite points per second


As does a digitally stored signal after the reconstruction filter returns it to the analog domain.
 

Krunok

Major Contributor
Joined
Mar 25, 2018
Messages
4,600
Likes
3,065
Location
Zg, Cro
I found that there was quite a difference between listening to the raw 44.1/16 and having it upscaled to 88.2/24 and then listening to it.
This suggests there are two aspects to the rate/quantisation discussion:
1. Is there any info lost that may be useful
2. Upsampling is a form of 'decompression' (in the data set size realm) and should offer better sound.

For me the upsampled 88.2 was a more relaxed listen with a better long term listenability.

Did you know what samples you were listening or it was a proper blind test?

Btw, upscaling doesn't add anything to the orginal sound so SQ remains absolutely the same.
 

CuteStudio

Active Member
Joined
Nov 23, 2017
Messages
119
Likes
65
As does a digitally stored signal after the reconstruction filter returns it to the analog domain.

Yes, exactly. Doing a bit of the reconstruction with a DSP allows the ultimate analog filter's job a bit easier - so it sounds better.
Perhaps 'Interpolation' is technically more accurate the 'Decompression', for me there's not really any difference.
The data (waveform) is reconstructed to/using a known set of rules, if you do this for FLAC to WAV or ZIP to Spreadsheet people are happy with the word 'decompression', but for transposing a digital time/level matrix to an analog shape - people may not like the word, I don't know why - the concept for me is the same.
 

CuteStudio

Active Member
Joined
Nov 23, 2017
Messages
119
Likes
65
Btw, upscaling doesn't add anything to the orginal sound so SQ remains absolutely the same.

I think this depends upon the analog filter at the end doesn't it? Straight 44.1 needs a brick-wall filter which may not sound as nice as an 88.2 filter which could be much simpler for greater effect.
So you are correct, it doesn't 'add' anything, but it can 'remove' work for the filter - a case where less is more.
 

Krunok

Major Contributor
Joined
Mar 25, 2018
Messages
4,600
Likes
3,065
Location
Zg, Cro
True, but I doubt you would be able to notice the difference in a proper blind test.
 

Eirikur

Senior Member
Joined
Jul 9, 2019
Messages
318
Likes
509
Straight 44.1 needs a brick-wall filter which may not sound as nice as an 88.2 filter which could be much simpler for greater effect.
But what DAC would do straight 44.1?
Way back in the eighties 4x oversampling was already the norm, soon followed by 16x oversampling and 18bits internal resolution, see e.g. the BurrBrown PCM58P datasheet I happened to study a couple of days ago.

To push my point once again: if you're inadvertently using a bad resampler like Windows the difference may well be audible (and pre-upsampling might save you if it exactly matches the set output rate!):
resamplerSweep.png

and to finish it off
resampler1kHz.png

These comparisons are obtained from infinitewave.ca
 
Last edited:

Tool

Active Member
Joined
Jun 29, 2019
Messages
145
Likes
92
From my point of view it's not only about 24/48,88.2,96,176.4 or 192 becouse if recording has Dr below 9 or 10 its crap no matter what resolution you listen to. And it's getting worse and worse unfortunatly.
 

Tool

Active Member
Joined
Jun 29, 2019
Messages
145
Likes
92
Fortunately there are still first releases to get. So I'm not worried. And I've got mine;)
 
Top Bottom