16/44.1 vs. 24/176.4

[Josef]

Wow we created some controversy here, great fun

Like Peter, I am a bit skeptical about how ADC is treating 16/44 - if I wanted to be really, really nasty (no hard feelings please:) I might surmise that 'dynamic decimation' is a marketing term for 'just chop 3 out of 4 samples'

But seriously: Version I posted (let's call it 'DND' - see below) is simply a more realistic scenario because software is more likely to be better at what it does (because it's easier to experiment & update) and also likely to _do more_ than hardware - Specifically, version I posted has been processed minimally (& maximally at same time ) like this:

1. Downsampling from 176 to 44: There are many complex algorithms here i.e. this is much more complicated than simply keeping every 4th sample (not saying your ADC does that - just illustrating the point of complexity and likelihood that sw will beat hw every time)
2. Noise Shaping - this is theoretically not needed but in my experience I found it to be essential when 'downsizing' to 44/16. Again, there is a multitude of algorithms to choose from...
3. Dithering - i.e. cutting 24 to 16 bits. This one was to me always more impressive (even magical!) than first 2 as, in theory, it affects only the last 16th bit (hey, how important can it be whether it's 0 or 1??)  and yet dithering method can impact a definitive sound signature - And of course, there are a zillion algorithms to choose...

Point being - to get _good_ 16/44 you need to choose from many, many different algorithms which is easy (although time-consuming) to do with software but often impossible to do with hardware alone, and results are bound to be waaay betetr with software approach.

Anyway - who cares about all this technical mumbo jumbo stuff - let's discuss how this sounds to us!

24/176 does not work on my hw but I am fairly certain that 16-bit 'DND' version is superior to 'raw' 16/44.

It is hard to put it in words but while listening I remembered an occasion when I was experimenting with manipulating WAVs I was loading to my iPod. I was using Foobar and turned off both Noise Shaping & Dithering controls (no changes were being made to music apart from ReplayGain).

Initially, I was surprised how much more immediate everything sounded - vocals were 'in my head' and music was very much 'in your face'. After listening for longer time however, I realized that effect of no NS & D was, in a way, similar to comparing 'flat' & 'heavily compressed' track. Compressed track always sounds 'better' because there is 'more there' until one gets a headache and starts seeing square waves in sleep

'Raw' 16/44 sounds eerily similar: It's like a rough diamond - all information, bits (pun intended & pieces, atoms & molecules, are there but it's not cut yet and certainly not polished - I assume 24/176 will sound similar: more flow, more ease, more polish - simply more 'music' as opposed to 'sounds'.

But maybe I'm mistaken: would people with capable hardware care to share their thoughts of both 'native' 24/176, 'raw' & 'DND' 16-bit and, especially interesting for us fans of XX, AP upsampled 16-bits?

PS.
>Which program did you use for the downsampling/decimation?

Mani, I'm not sure it's OK to talk about other software on this forum so I'll send you an email!

[manisandher]

Firstly, let me ask a favour of you - please take what comes below in the spirit in which it’s meant. I’m not here to convince you of anything. I’m just here to share my experiences that you may or may not find useful or interesting.

18 months ago, I bought a couple of ‘HRx’ 24/176.4 WAV albums (burned on DVD-Rs) from Reference Recordings and was really taken aback by their SQ. There was a ‘quality’ that they possessed that I hadn’t heard before from any of my ripped or downloaded 16/44.1 files. This quality is hard to put into words. Extra low-level information? Smoother? More dynamic? I’m not sure that any of these manage to describe the particular quality that I heard. The only really accurate description that I can offer is ‘life’ – the music ebbs and flows as if it’s alive. (But please don’t mistake this description for ‘life-like’ – you would never mistake these files being played back on my system for the real thing. A better analogy would be with a painting – great paintings have so much ‘life’ in them, but they’re not necessarily ‘life-like’ at all. Do you understand what I mean?)

And this got me thinking about 16/44.1 vs. 24/176.4. Are the former inherently inferior to the latter in an audible way? If so, is there anything that can be done about this?

I wanted to be as objective as possible. Of course, it’s impossible to be totally objective, but I wanted to eliminate as many of my own biases as possible. The best way seemed to be to get an analogue source, record it at the two different resolutions and compare the recordings with the original analogue source. An assumption that I’m making here is the absolute quality of the playback system (preamp, amp, speakers, cables, etc) becomes irrelevant because it is the same system used in all cases. Of course, this is only valid if the system is resolving enough to allow any differences to heard – and I believe it is.

Method

Before making the comparison between the vinyl, native-24/176.4 and native-16/44.1 files, I wanted to ensure that my digital chain was totally transparent.
I tested the following two routes:
1. phono stage -> AD -> DA (i.e. internal)
2. phono stage -> AD -> AES cable -> RME PCI card -> RME mixer -> AES cable -> DA
Irrespective of sample rate, I found that there is absolutely NO difference in sound between these two routes. (Incidentally for 2, I knew that I really was passing through the RME mixer because I could hear any changes I made to levels, muting, etc in the RME mixer.) From this, I concluded that the AES cables, the RME PCI card, the RME mixer and the RME driver are all transparent enough for the purposes of this assessment.

Objective Assessment

(Yes once again, I know that this cannot be truly objective if I am making the assessment...) I played the vinyl and was looking for the digital file that sounded closest to it. IN COMPARISON TO THE VINYL, here is my ranking:
1) native-24/176.4
- sounds identical
- cannot distinguish in double blind listening tests
2) (XX) QAI of native-16/44.1 = (PM) 1:4 Interpolation of native-16/44.1 = (Software) 4:1 Decimation of native-24/176.4
- tonal balance maintained
- less depth
- rounded transients
- less low-level detail
3) (XX) QAP of native-16/44.1
- sharper
- leading edges (over?) emphasised
- cymbals sound thicker and much less delicate – the initial strike is more emphasised but the shimmer/decay is attenuated
- less body and weight to instruments – they sound smaller and ‘cheaper’
4) native-16/44.1
- brighter
- more edgy
- more forward

I suspect that most people would come to a similar conclusion here. I’ve played the vinyl to a number of (non-audiophile) people and then these digital files. Asking them which digital file sounds most similar to the vinyl, I get very similar results to above, although they often find it difficult to articulate exactly why they think what they do.

Subjective Assessment

Here, I consider which digital ‘sound’ I most like and can most happily live with. My ranking is:
1) native-24/176.4
- I don’t long for anything else, this has it all
- it is detailed and yet totally smooth and easy on the ear too
- it has the all-important ‘life’ quality
2) native-16/44.1
- this has ‘life’
- it breathes
- although it is brighter and more edgy than the native-24/176.4, I can live with this
- it doesn’t annoy (too much)
- good boogie factor
3) (XX) QAP of native-16/44.1
- the ‘life’ of the native-16/44.1 is maintained
- sounds very impressive on first playing
- tight as hell
- all instruments/voices are absolutely delineated and focused
- phenomenal boogie factor
- but gets fatiguing after a while
4) (XX) QAI of native-16/44.1 = (PM) 1:4 Interpolation of native-16/44.1 = (Software) 4:1 Decimation of native-24/176.4
- very smooth and easy to listen to
- all the ‘life’ has been sucked out of the music
- great for innocuous background music

Conclusion

No, I couldn’t get 16/44.1 to sound as good as the vinyl source or 24/176.4. But I can get it to sound ‘good enough’ to listen to, and more importantly, to ENJOY. Playing the native-16/44.1 file hits this spot. Using QAP with my setup doesn’t. But I’m holding out for OAP on a NOS1... which I'm hoping will do it.

Cheers.
Mani.

[PeterSt]

I listened to the 44 first and thought, well it sounds good. I don't expect the hires to sound much better, and got a bit bored of the one-note hihat, one-note sax and one-note drum.

Hey Gerard,

I suppose this is nitpicking, but I never heard a hihat in there ! What's used there is a ride cymbal which really sounds different. Just for fun : http://en.wikipedia.org/wiki/Ride_cymbal

But let's say this won't bother you anymore in near future. Haha.

Peter

[PeterSt]

This is going to be complicated ...

Let me first paste in what I wrote to Mani earlier via email (I had the files somewhat earlier than they were posted here) :

First listened to the 44.1. It sounded just gorgious to me. I tried to imagine what the 176 could bring more. Listened through the whole track ... no complaints.

Next the 176. Hmm ... Right from the start I seemed (!) to notice more air. Later (listening a second time) the same perception. But, somehow I can't listen this one through. It doesn't fit. The piano at the right (assumed I have my L/R cables right) wouldn't be the leading "rythm", while I think it is/should. Later (maybe the 4th time) I perceived it as a stupid cut "ploink" thing. Besides, the sax (is it a sax ?) clearly annoyes with the 176.

Notice : I was just comparing again (Josef's decimated version incorporated) and after listening several times to pieces of the 44.1 versions, I compared with the 176.4, and the exact same as before : the piano doesn't want to work.

For me the 44.1 version(s) win (8x AP), although it is hard to find more reasons than the piano. I have some allright, but I'm not sure what they tell :

Firstly I got the focus somehow on the sax after 56 seconds (up to 1:20 or so) in the track. It seems that it contains a distortion in the lower regions in Josef's version. I was comparing this to Mani's 44.1 version, and thought that one sounded more natural.

When I lastly compared it to the 176.4 "original", I found the same "distortion" to be in there. So, was Josef's version a better copy ?

In order to get better merits of this all, I listened to 2:50 up to 3:20. This is right after the drums start to be soloish. I think it might be important that I listened to this part first with the 176.4 version. All sounded very clean and as should. And, this is exactly why I wanted to hear that very part in the 44.1 versions. Here's is the strange part :
Both 44.1 versions contained an echoish sound on the drums. A sound of which "you" most probably would say it doesn't belong there. If you'd hear what I mean, you would be convinced that it must be something strange, because the lower tom hits this is about, don't contain any higher frequency "noises" you perceive here (and which were not there in the 176.4). And, if I hadn't been recoding my own drum set the other day I would never have heard it ... this is the sound of the drum kit itself. Partly it is the snare of the snare drum, but if you were next to such a kit and tick against the poles and sides of drums and anything, you'd have a metalish sound just of "loose" things.

From that moment on I thought that the distortion in the lower regions of the sax just are details which the Mani-44.1 coudldn't bring. However, they are there in the 176.4 version as well (as in Josef's 44.1 version).

Conclusions ? none. Or none other than that everything makes a difference.
For me the 176.4 version shows more air, but nothing which makes the happening better. The contrary.

Keep in mind the context in which this is to be interpreted :

1. What exactly happens to the resulting file when the 176.4 is recorded;
2. What will exactly happen LESS to the resulting file when the 44.1 is recorded (which internally still is a 176.4 according to Mani's description);
3. What will exactly happen otherwise when the 176.4 is decimated manually (which can go per a 1000 means as per Josef's description).

Ad 1.
It is quite important to know that very many of the CDs we listen to, are recorded by the same device Mani used to take this recording. And, what at least I am trying to, is how to benefit (with playback means) from the knowledge of what really happened to our CDs when they were produced.

Peter


PS: In the mean time I have another idea to look at this all. Give me a few minutes ...

[GerardA]

I suppose this is nitpicking

Haha, yes thinking of it.. Never to old to learn/think!
So much for the rest of my remarks!
One-note is not true too, and is it just an ordinary sax?

One way it is interesting to find out what instruments you can hear, maybe a new hobby..

[PeterSt]

Ok, I wanted to do this more extensively, but it really takes too much time. Anyway, below (also see next posts) you see a few pictures.

The first one here shows Josef's 44.1 take at the top, and Mani's at the bottom;

To me it is clear that Josef's version is better. It shows more detail, and it better follows the original;
Hard to explain by means of just this one picture, but what I see, I see back everywhere. Look at the bottom channel of the bottom picture (Mani's version) for an example. You there see two situations (one at 1/3 of the screen, and one at 1/2) where two adjacent samples have no difference in amplitude. With this in mind, already in this one picture you can see that the amplitude (volume) steps generally are smaller in Mani's version.

Notice : It is not said at all that larger steps are better, but it is about following the original better (see next posts).

[PeterSt]

What we see here is Mani's version compared with the 176.4 "original". The original has 4 steps sideways where the 44.1 has 1.

Look at the top channel just before the middle; The original shows the tendency of the downgoing slope being higher in amplitude than the upgoing slope preceeding it; You can see that the samples at the other side (the right side) of the slope are higher in amplitude than at the left side. However, Mani's version shows it the other way around, and a downwards going tendency.
If you look more to the right, you see similar happening at the top channel (Mani's version) where the wave disappears under the horizontal line. It shows an upwards tendency, while the original clearly shows a donwards tendency (compare with the left side of that wave).

Try to see more of this yourself ...

[PeterSt]

Here we see Josef's version against the "original". Read the text from the previous post again, and see that here things are as expected.
Well, as how I see it.

Also look at the second peak from the left, and what happens in front of it. Look at the down - plus upgoing small slope and compare it with the original. Yes, you can sense Mani's version (previous post) as better matching, but in the mean time that shows nothing of a smooth going up again, where this one does. But, doubtful in this case.

But, just look, and "feel" that this is al a better representation;
I have done it for several of these pictures, and they all give me this same feeling.

[PeterSt]

Lastly a more zoomed out comparison between Josef's (top) and Mani's (bottom).

Without seeing the original (which I couldn't locate in 30 minutes for this position, so I gave up), again you can sense the better representation of the top one. If you look at the 2s138.6ms position you see a downgoing slope with no sense in the bottom picture. No guarantee when looking at this only, but seeing more of these picture really makes me say this.

Here too, you can clearly see (if not dizzy by now) that many of the slopes go just the other way around. So, one of them must be wrong, and that this is Mani's is clear to me, but also follows from the few examples in the before posts.

But :
... see last post about this ...

[manisandher]

Haha... this is what they meant by 'dynamic decimation'

Looking forward to your next post before I respond...

Mani.

[PeterSt]

... it is not said that the one "decimator" is better than the other. I didn't look for it (takes too much time), but what Josef should have done, and probably has not, is starting his "recording" at the very same sample Mani did with his 44.1 take. Now, already that is impossible, because Mani took two recordings, and the sampler with its discrete fixed sample rate will never have started at the same (analogue) wave position as the previous time. But would that have happened by pure coincidence (with a change of 1:174000), then Josef should have started his recording at a common denominator sample of the 44.1 version of Mani's. When you don't do that, all is shifted a little, and where Mani's 44.1 virtual sampler would have started at the first sample, Josef's will have started om e.g. the 2nd, and all will be totally different because of it.
Look again at one of the 44.1 versions against the 176.4 version, and you'll get what I mean. Just look one 176400 sample more sideways, and up going slopes will be downgoing slopes.

Is this all too stupidly looking at details ? No, I don't think so. I hear a difference between the two versions (Mani, Josef), and the files *are* different. But also I heard similarities between 176400 and Josef's take (see earlier post), which just as well is explained now.


Allright. Don't try to understand this. But if we are spending our time on subjects like this anyway, we must do it well. I don't say I just did, but it really comes down to these kind of details;
If files are different they should sound different. If they do not, better wonder what can be outbettered on your system.

And oh, nothing tells me that Josef's file should sound different from the 176.4 original. This depends on many other things (like the analogue workout and how *that* compares). But all sure tells me that Mani's 44.1 MUST sound different.

Peter

[manisandher]

WOW! Great analysis Peter. Thanks.

But all sure tells me that Mani's 44.1 MUST sound different.

I agree with this 100%. What I call the native-16/44.1 sounds the most different from the vinyl. And the native-24/176.4 sounds the most similar to the vinyl (actually identical, as far as I can tell on my system). Josef's 16/44.1 definitely sounds more similar to the vinyl than my 16/44.1.

Mani.

[manisandher]

Josef's 16/44.1 definitely sounds more similar to the vinyl than my 16/44.1.

BUT... I prefer the sound of my 16/44.1...

EDIT: Pacific Microsonics strongly recommends using the 1:4 Interpolation in the PM2 for 16/44.1, but I don't like the way that sounds either.

Peter, as Josef's 16/44.1 has been derived directly from the original 24/176.4, would it be possible to create another graph comparing QAP applied to his 16/44.1 vs. the original 24/176.4? This would give us an excellent idea of how accurately AP can reconstruct a hires file.

Mani.

[PeterSt]

Yes, that's the difficult part of it all. I too like your version better, though technically (listening) the 176.4 sounds better. So, this even goes further.

Maybe it is more important to recognize that either version sounds very good to begin with, and that somehow we are (suddenly, as found out) very close to letting whatever it was sound very close to the original - by means of digital recording and playback.

Assuming that these details already matter (and why not), I could for fun make an AI version of your 176.4 and show that. But I already know that won't work, because no common points can be found then. It will be totally different (I already tried that once).

So, in the far end it comes down to what happens beyond what we just looked at, and this will be our DACs. I can reason out what it will be with my DAC (namely, nothing as far as it concenrns the digital part), but e.g. yours ?
But careful, because "nothing" is not necessarily good, and all now depends on how good Arc Prediction is (of which I can *not* make a file without programming something for it (opposed to AI this happens in memory, and not as a pre-pre process which goes to disk first)).

But ok, since I made that 16/44.1 recording of a live drum session (people in here may not know it) which can't be distinguished from real at playing it back, I'm quite confident.

Now Mani, when are you in the neighbourhood again ? It sure looks like we can have some interesting times;
I'm not obsessed with this (you are I think, haha), but I really think if we can go down to the bottom of this al, and learn how to avoid mistakes, playback can be better again. So yes, that would come down to playback anticipating on recording devices, but since it can be done in software (and the NOS1), why not !
Better plan more than an ordinary one night.

Peter

[manisandher]

I'm not obsessed with this (you are I think, haha)...

Well, work has been very quiet over the summer, but will kick off in earnest in about 2 weeks time. I've just been trying to get as much done as possible around the house and on the hifi before I start living on a plane and in hotels again. My dream is to come back home after a work trip, spend time with my family and then relax in front of the hifi, just listening to music... without obsessing over how it sounds!

One thing I've been thinking about though. Your recordings - you're using the FF800, right? This uses delta-sigma ADC chips. Now, I have no idea how these things work on the ADC side, but they must be OK. I mean, they can't be mangling up the sound, because you're getting such good results from your live recordings. So, would there be any advantage in using a multi-bit ADC chip, like the one in the PM2?

Mani.