Enjoy The Music article on Hires worth a look

[Jud]

For me, this is really hitting the nail on the head again, as far as what I wanted to get to with my "bit perfect" thread on CA.  With the challenges of decimation on the one hand and interpolation on the other, and *knowing* all the converters will sound different (and therefore none can be ultimately "right," though it is possible to get very, very close), why do this down and up conversion? 

There were plainly reasons once upon a time, and to a certain extent these procedures have become set in stone now in the recording industry even though the technology has progressed to the point where we should no longer need them.

A couple of other comments:

- Though Peter has referred to decimation and interpolation as the same sort of process, I wonder if decimation might even be potentially more destructive.  With interpolation there may be the possibility of retaining all the original samples, while with decimation you know you will be throwing some away, a higher proportion the higher the sample rate of the original.  So is it better to sample in the A/D step at 44.1, or to sample at 176.4 and decimate?

- Very interesting comment from praphan that he finds himself preferring the interpolated 44.1 version of a recording to the "native" resolution, though the two are quite close.  With something as good as Peter's software, and so customizable, the dividing line between more accurate and more pleasing has got to be *very* difficult to ascertain.  My usual test (how different various recordings sound from each other, or to flip it around, how much of its own sound the hardware or software you're listening to imparts to the music) might not be enough in such circumstances. 

Really, for such a task one would ultimately like to be able to compare "native" and decimated-then-interpolated versions from a session where one was present for the recording.

[PeterSt]

(sort of)

LOL Jud. I know you mentioned it before but I forgot to look (or was busy maybe ). I just did at light speed but it has almost become impossible to say something with sense.
Maybe later, when I can really spend some time on it.

Btw, I also saw some nice matches between what I said here and what was suggested by some (which is never SP...etc. )
But I can promise you, I never read it until just now. Nor it can have been read by others what I wrote today. Funny. And it wasn't even you motivating it. Still you did overthere ...

Sorry to be cryptic. I think it's funny ...
One day we will be there I guess.

Peter

[esimms86]

While I acknowledge that Barry Diament's recordings are not everyone's musical taste, the same can also be said about any recording ever made. Most of the music that any of us listen to is recorded with close miking and that goes for the overwhelming majority of your and my favorites. At the same time, I agree with the notion that musicians and engineers have a right to use nontraditional means of recordings to fit their own vision of how the music should sound. I have AIX, 2L and Naxos discs that bravely go that route producing interesting musical experiments that I also enjoy. Though part of a niche market, I don't find those recordings to be unlistenable and I also don't think it impossile to compare sound quality when listening to differently formatted recordings that are not exactly one's cup of tea.

Still, I have to admit that the bulk of music recordings that most of us(myself included) want to hear are available as popular CD's in redbook format and, short of a flood of legacy recordings by companies like WMG becoming downloadable for Neil Young's Ponos device(and I don't even want to begin to get into a debate right now about that enterprise which has yet to surface in the marketplace) and usable by folks using non-Ponos playback systems, our favorite recordings will continue to only be available in redbook format.  I find it interesting that some folks on the CA website have dissed the HDTracks 24/96 download of Crosby, Stills and Nash while Steve Hoffman's clearly superior remaster of the original recording is available as a redbook CD. I've listened to the Hoffman remaster as an XXHE upsampled file playing through my NOS1 and have not been left wanting.

The point that Peter makes about downsampling a 24/192 wav file to 16/44.1, effectively decimating, if you will, the original recording, is an excellent one. While not the same thing as, say, downsampling a 24/44.1 or 24/192 Beatles recording to an mp3, it is still a move in the same direction. I agree with Jud's proposition that the ultimate test is to listen to and compare recordings of performances where you were present at the actual performance.

Esau

[Scroobius]

Isn't downsampling more of a problem than upsampling?. Upsampling moves aliasing images outside the audio band the more upsampling the further outside the audio band (a key reason to upsample). Whereas downsampling moves the aliasing images inside the audio band (yuk) so downsampling has to involve a brick wall filter which is the only way the effect of those aliasing images can be mitigated.

So why not take an analogue recording and create one digital master at say 24/192 and another at 16/44 and then compare them? Peter - can you do that with the PMII? I guess maybe it does not which would be a great shame.

P

[CoenP]

Synchronous downsampling is easy (ie from 192 to 96 or to 48). Compare this to Peter's AP algorithm: allways exponents of 2 (2-4-8-16). Downsampling is the same, just drop half the bits. I am shure it will make a difference to the reconstruction however which bits you drop.

Async interpolation needs interpolation (ie from 96 to 88.4). This is done by finite bandwidth filters. Here you go. Options and implementations are wide open. Many ways to mess it up.

The only way to truely compare is indeed to record 44.1 vs 192. But beware, adcs that can record 44.1 ' native' have not been made for many years (if ever). With 44.1 coming out of an adc there is allways a high risc that some dsp going on since professional recording standard is 48 kHz (or multiples). The best way to create an 44.1 kHz master is to start out at ie 176.4kHz and decemate from there. No dsp involved to do this!

Regards, Coen

[PeterSt]

Coen ... No way downsampling works like that !
Sorry ...

(with Upsampled Arc Prediction, yes because that is a lossless means; but that's quite another beast)

[CoenP]

Coen ... No way downsampling works like that !
Sorry ...

(with Upsampled Arc Prediction, yes because that is a lossless means; but that's quite another beast)

Ok I've havent been stating which adc process I assumed, but sigma delta adc chips are doing just that: dropping samples for synchronous conversion (but this also has the prupose to establish the sample value in this process).
Halving the number of samples in sync conversion is mathematically correct because in the higher sampling rate all the bandwidth of the halved rate is allready present and no interpolation is necessary to obtain the samples within that reduced bandwidth (the aliasing is a reconstruction problem in this case). That is the same reason why you need an AA filter for asynchronous conversion of sample rates (otherwise you would interpolate aliasses into the lower rate).

Then there is also the unmentioned issue of reducing the bits from 24 to 16....

Am I missing a point somewhere?

Regards, Coen

Added: from a theoretical vieuwpoint it does not matter which samples you drop in sync conversion. Both are perfect descriptions of the same signal PROVIDED that the conversion back to analog is perfectly limited to the nyquist frequency. You can find some flash on the net illustrating this counter intuitive phenomenon.

[PeterSt]

Coen, not to debate it, but to give some counterweight to people who might like to pick up something : still not true.
I'd say : try Linear Interpolation and hook up an analyser (if you don't hear it in the first place). That's just injecting samples like you describe ...
(I should remove that stupid option )

Peter

[CoenP]

Coen, not to debate it, but to give some counterweight to people who might like to pick up something : still not true.

I chewed a little on my post over dishwashing and you are right (as allways )! What I posted is not true and contains a flaw and you will end up with aliasing into the target bandwidth.

My post only makes sense if you filtered out all frequencies above the target nyquist freqency in advance. Only then you will (at least theoretically) end up with samplesets that all represent the same waveform (with synchrounous downsampling).

My synchronous 'experiment':
So take a 176.4 kHz original (88.4 kHz audio bandwidth). Filter it to 22.05kHz audio bandwidth digitally with a brick wall filter and then you will end up with four sets of 44.1 kHz samples all representing the same waveform. So no new samples on the time axis, but recalculated sample amplitudes.

So no way to downsample without filtering it first. This is the basis of Shannon!

The adc chips contain many of these filters. You will have to down sample from 24 MHz or something to 192k...

I'd say : try Linear Interpolation and hook up an analyser (if you don't hear it in the first place). That's just injecting samples like you describe ...
(I should remove that stupid option )

Well this is not what I had in mind of course. With asynchronous conversion you also change the timebase of the samples (only once every x samples you end up with a sample timed on the original timebase). The filter will determine the amplitude of all samples, so no one in his right mind will use Lin Interp for that.

I have allways wondered what the above synchronous experiment will result in. Do the four 44.1kHz sub files all sound indentical after proper filtering of the 176.4kHz 'motherfile', like they theoretically should?
This would be quite illuminating. Actually my idea is that they will NOT...

Back to the sq of hirez material.
Theoretically the hires in native samplerate contains much more original information than the downsampled 44.1 red book variant. You will have to filter and reduce bits by some inherently lossy process to get from the master to red book format whatever you record with these days. I don't buy 'native' 44.1k.
In practice you seem to have ameliorated this loss to something insignificant, like we experience with xx and the NOS1.
Thanks for that!

Regards, Coen