16/44.1 vs. 24/176.4

[manisandher]

... (and besides, Mani's version keeps on sounding the best to me)

To me too... not as good as the vinyl or the 24/176.4... but very engaging and enjoyable, with zero fatigue.

Mani.

[Josef]

>A few minutes ago I already typed "but the quantization noise should be audible ?" ... and then I thought to listen for it. Josef, your both versions are full of it. Mani's has NONE.

Yes, of course
Both 'my' versions were downsampled from 176 to 44 - AFAIK By definition, downsampling will create quantization noise, won't it? The question is not whether it is there (it has to be) but does it negatively affect the sound?

Now, if you say Mani's version has no quantization noise that would also make sense if ADC did _not_ dynamically downsample 176 to 44 but, rather, as suspected, just kept every 4th sample - right?

>You can hear it right in the beginning, when the needle is still in the lead-in.

Lead-in lasts maybe 2 secs on Mani's 16/44 but only maybe 0.5 sec on 'my' version - Sorry, but I could not hear any weird noises in 0.5 secs, it's just too short a time before drum starts hitting - Or did you mean something else?

(see also next post)

[PeterSt]

Hi Josef,

Awaitening your "next post" ...

Still not sure yet, but I think I start to understand better what to make of your earlier post (the one from 6 hours back or so);

Yes, quantization noise will be there, but therefore is the Anti Alias filter. So, in your decimation you didn't apply that, right ? (or a bad one perhaps ?). Anyway, this is not about dither ...

Again, I may understand you wrongly, but taking out 1 out of 4 samples (at going from 176.4 to 44.1) doesn't work, as it doesn't work the other way around (inject samples at positions by "normal" math); That needs an Anti Imaging filter ... (AI). One exception : inject the samples at the proper position and it's ok (this is AP ... hopefully ).

I once started out with decimation myself, and did this for DXD (which is 24/352.8 ) as provided by 2L. I didn't have a 24/352.8 DAC at that time, so I couldn't check the original. And thus, the quantization noise (I know now) was all over. However, instead of interpreting it like that I called 2L and told them their classical samples were quite noise ...
I'm still red on the cheeks from that conversation. So, they created a properly decimated version especially for me, and it was completely out of noise.
An AA filter just does the trick.

Yes, Mani's version has a longer lead in to listen to, but at comparing this afternoon I could notice the relatively loud noise from (indeed) your half of a second from any distance. Careful here, because I have horn speakers, and they unveil quite some more than any "normal" speaker. But moreover, you can listen "in" the horn with your ear, and there you hear everything.
Also one must know how quantization noise sounds. This is a whizzling quite unlike the ticking ans "sliding" (which you can also hear) of the needel over empy space (which is not empty, because it has to be lead). In dB's I would estimate your "general noise" at 12-15 (over two times more noise).

I don't think this is a matter of "will it be audible or not", because it is nothing like dither (needed). It's just plain distortion which digs itself into anything. As a matter of fact, before listening to the lead in with my ear in the horn, I first listened to the piano, and some later in the track when the drums get more heavy you can hear how that distorts (notice the piano is in the right speaker, so listening to the drum which is in the left, but halls over to the right is quite easy to do (without hurting)). This is not the case with Mani's version.

Lastly (but you may know it), dither is about the decimation in bits (not sample rate) and making more of it than actually there. The more really random the dither is, the better it works. If not really random, patterns will emerge, and it just takes away the sound.
Notice that with proper dither even with 1 bit you can still recognize music. RME has (or had) a test file for it.

Peter

[PeterSt]

To be clear :

but, rather, as suspected, just kept every 4th sample - right?

Totally impossible.

[manisandher]

Now, if you say Mani's version has no quantization noise that would also make sense if ADC did _not_ dynamically downsample 176 to 44 but, rather, as suspected, just kept every 4th sample - right?

Josef, why do you 'suspect' that Keith Johnson and Michael Ritter (of former Pacific Microsonics) would be lying about the decimation processes in the Model Two?

I can create a 16/44.1 file from the native-24/176.4 file using the Model Two's digital-to-digital DSP engine. To do this, I will have to 'record' the Model Two's digital output as it downsamples the native-24/176.4 file to 16/44.1, so the two files will not be aligned or of quite the same length. Let me know if this will likely help us understand the Model Two's decimation process.

Mani.

[PeterSt]

Wait ...

I didn't look it up, nor will I ever try, but there exists at least some "dynamical" filtering, invented by some Japanese guru. IIRC it was implemented in some Denon model.

This meant that different filtering was applied to higher frequencies opposed to lower. Or better : that lower weren't subject to it, or something like that.

Also, I don't think that Josef was "accusing" as such, but merely has an idea about what "dynamically" could ever mean, besides commercial talk. So, I only want to point out that such a thing really existed.

Btw, that Japanese guy himself has been talking about it in DIYAudio IIRC (not working at the company concerned anymore).

2c

[manisandher]

OK, here's more on 'dynamic decimation' from Michael Ritter:

"A filter designer who has to make a 'brick wall' filter at 22 kHz is confronted with conflicting requirements. You want to have flat frequency response out to at least 20 kHz, but you can't have any energy above 22 kHz or you will get alias distortion. This requires a very sharp multipole filter with a very steep transition between the passband and the stopband, which has a number of distortive effects on the signal. It smears transients and causes significant ripples in the passband. If you try to simplify the filter, then to avoid totally unacceptable aliasing you have to start rolling off at 13 to 15 kHz, and even then the signal will not be completely cut off by 22 kHz.

We slightly delay the 88.2kHz [or 176.4KHz] signal, not enough to cause any sync problems but enough that we can do a continuous Fast Fourier Transform. The resultant information is digitally analyzed in real time by an algorithm that determines, based upon a model of the mechanics of hearing and psychoacoustics, what is perceptually dominant in the signal from instant to instant. And that information is used to optimize the decimation filter. One moment you might have a sudden sharp transient, so it uses a filter with minimum time dispersion to pass the transient cleanly. The next instant, there might be a cymbal crash, so it uses a filter that minimizes alias distortion. All the filters are the same length, so you are not getting a phase shift as this is going on.

We never simply truncate. We have a palette of four 16-bit dither options. The reduced distortion-sharper transient response and reduced aliasing-becomes part of the digital recording and will be heard on any player, whether it has HDCD decoding or not."

HTH.

Mani

[Josef]

Guys "next post" is hopefully coming quickly but in meantime:

>>>More importantly, I suspected that ADC also did some limited/poor quality dithering or no dithering at all i.e. it just chopped last 8 of 24 bits.

>This is not the case. The Model Two provides a number of 16-bit dither options, which I understand are pretty sophisticated.

Yes - I found the Instruction Manual (that's why you need to wait for "next post" ) and I stand corrected: there are no less than 7 different dithering algorithms, lol

Here however, we have a small problem: you mentioned that you did NOT use any HDCD processing but looking at manual under dither it is called "HDCD 16-bit Dithers"? It also says that DITHER1 algorithm is default. Is this just marketing again (i.e. dithering has nothing with HDCD encoding) or does this imply that _some_ HDCD internal processing is still being done if dithering is used?

Because if it is then we are running in circles - of course we cannot HDCD encode via software (at least I can't)...

[Josef]

And, just to clarify: did you use mode AD_44.1 or AD+DD44 or something else for 16/44 version? And I assume word length was set to HDCD16?

[Josef]

And oh:

>Josef, why do you 'suspect' that Keith Johnson and Michael Ritter (of former Pacific Microsonics) would be lying about the decimation processes in the Model Two?

Please, where did I state they were lying? People use marketing terms 'freely' all the time and these guys would not be the first so it does not hurt being suspicious - just look at that guy's Goodwin's site - PM Model 2 has processing power of '200 million instructions per second' - Wow that's as fast as 3x 486/66MHz computers or almost half of Pentium Pro from 1996
 
Let's keep this discussion open no need to get excited (and especially not over some product )

>Yes, quantization noise will be there, but therefore is the Anti Alias filter. So, in your decimation you didn't apply that, right ? (or a bad one perhaps ?). Anyway, this is not about dither ...

Alias was used - but only the default, I'll try something else.
And you may be right that we should focus on aliasing filter: I was just thinking - if version 2 without dithering but with different aliasing is sounding closer to native then maybe dither is not important.
Still - I think it will be important at least to a point: I think I know how I can test this (if I can find a piece of software I bought long time ago but never used, lol    

[manisandher]

Is this just marketing again (i.e. dithering has nothing with HDCD encoding)...

My understanding is that dithering has nothing to do with HDCD decoding, so shouldn't be an issue here.

I didn't use any of the three HDCD 16-bit amplitude encoding options.

Mani.

[manisandher]

And, just to clarify: did you use mode AD_44.1 or AD+DD44 or something else for 16/44 version? And I assume word length was set to HDCD16?

AD_44.1 with wordlength set to HDCD16. BUT... Peak Extension and Low Level Extension were NOT applied.

(As explained earlier, my native-16/44.1 file was saved as a 20-bit file by the recording software - but the signal is 16-bit.)

Mani.

[PeterSt]

Let's keep this discussion open no need to get excited

The most important, but I'm also the most sure we all are.

Both open and excited.

Well, I am. I really think it is important.
Just wait until you heard what I have been up to all the time. And, I think I'm finished with it, but I am not.
Not by a mile. I'm fairly sure of that.

[manisandher]

... then we are running in circles...

I really don't think there is an issue with comparing the native-16/44.1 file (encoded with 'HDCD dither') with the native-24/176.4 file. However, if anyone believes that this is not totally fair, then fine, just use Josef's original 16/44.1 file for the comparison.

Mani.

[GerardA]

Well, the technical part is over my head, although I get reminiscences of the mp3/atrac/dcc explanations from 20 years ago.

The piano still intrigues me, is the rhytm missing on 44 kHz too?
To me it sounds like the piano is missing some of it's impact because of resonances from the recordplayer-arm.
I don't have this record but on my player I like the pianosound a lot.
If only I could record at 192/176, not even 44 at the moment.

A nice explanation can be found here: http://www.thefunkfirm.co.uk/arms/fxr_II.html

If this is the case, a very good record player could give us a perfect digital source??