Page 3 of 6

Posted: Wed Jan 28, 2004 11:40 pm
by krabapple
thomh wrote:
krabapple wrote:
thomh wrote:Let us talk a bit about bits.


AFAIK, there is a trade-off here. With noise reduction we introduce non-linearity into a linear system. When we talk about the resolution of a system we are actually talking about resolution in amplitude domain (dynamic range) and resolution in the time domain (bandwidth). Noise reduction only affects the amplitude domain. You get a bigger dynamic range, but the overall system resolution does not get any better. In fact, it makes it worse. Witness how during crescendo passages you can hear the noise floor rising. So with a combination of smaller and louder signals, those smaller ones will get buried in the noise floor. So the payoff is more simple dynamic range at the expense of linear resolution.


But that rising noise floor problem wouldn't apply, would it, if NR is only applied to 'quiet' parts -- as is the case for e.g. the Rolling Stones SACDs?

I'm not saying overall system resolution gets better, though. I'm just wondering if it necessarily has to be worse, if NR is applied conscientiously.

Posted: Thu Jan 29, 2004 12:23 am
by krabapple
Apparently the eyes of god are upon us:
//


Forum folk,

Thanks and I appreciate the gesture, but you don't have to send me warning emails when an SH poster posts unflattering stuff about me, or this place at another site. Thanks for thinking of me though.

That being said, I was emailed this from a well-meaning soul. It's a post by Thomh from another site. I don't care if you talk about me someplace else, just don't put words in my mouth!

The post:


From Thomh:
--------------------------------------------------------------------------------

Let us talk a bit about bits.

Bitdepth (word size) is directly related to dynamic range, so the more bits there is available the lower the broadband noise floor and hence the potential for more information.

It is very important to keep in mind that when we talk about dynamic range we are also talking about resolution. The two are unbreakable. You increase one and you also increase the other. Think about it, it makes perfect sense.

What you also need to factor in here is that digital processing bits is different from AD and DA bits. When processing a lot of data one looses bit accuracy because of rounding errors. A lot of DSP causes some error and that is where more bits come in. This is why more bits during the recording stage is a *very* good thing. And why when you go from 16 to let's say 20 bits, you will probably hear an improvement.

Equally important is that once we have a signal with a certain dynamic range, as is the case with an analog master tape, we CANNOT IMPROVE IT. Sampling can only capture what is there.

So given a master tape where there is nothing above 15 kHz then 16 AD bits is plenty and so is 44.1 kHz sampling rate. Increasing the bitdepth or the sampling rate will do nothing.

Ryan is correct in saying that the lower the noise floor the more you are able to reveal things like echo decay and other low level details. But remember, *once a dynamic range is set then it cannot be improved upon*.

What SH wants you to believe is that a 30+ year old analog master tape which, according to him, has nothing above 15 kHz somehow exceeds the bandwidth set for Redbook. A *good* analog tape deck probably has dynamic range of around 70 - 80 dB. That is equal to about 14 bits (2^14 gives a DR of 84 dB). A 16/44.1 kHz sampling should be able to capture all the information contained on this tape including the echo.

So there is something else at work here. As noted by Krab, his CD and SACD layers used *different mastering chains*. Can we assume a less than perfect PCM conversion is culprit here? Or has he proven the sampling theorem wrong?
---------------------------------------------------------------------------

Steve responds:

I don't want you to believe anything. All I'm saying is TRY IT FOR YOURSELF.

Thomh, Krab said I use different mastering chains. I do not. I use the same chain with a split feed, as I've mentioned here many times. PLEASE, if you wish to address something with me or about me, DO IT HERE!



A mastering chain that bifurcates at the A/D stage is no longer the same chain. Any confirmed audible difference that results -- and so far, none have been confirmed properly for your recordings -- *could* be due to CD vs DSD per se, OR to equipment differences in conversion and playback.

This is an inherent problem in comparing formats. It remains even if the comparison is between an SACD layer, and a Redbook copy of the SACD layer...which is how some other hybrids are made.

So, the order of proof should be: first, prove that you actually hear the difference (this means using a bias-controlled level-matched protocol)

Second, prove that the difference is due to the formats, and not the equipment.

Lastly, I'm willing to DO IT THERE, if the draconian rules against expression of skepticism are changed, as the Gorts know.

Over to you , thomh.

Posted: Thu Jan 29, 2004 2:02 am
by thomh
lukpac wrote:
thomh wrote:A waveform has 2 dimensions, amplitude and time. If you sample a 20 kHz signal at 44.1 kHz, 96 kHz or higher, it is within Nyquist of all and is captured *100%*. No amount of handwaving from the audiophools will change that. This is a *fundamental* principle of the sampling theorem. What this means is that you get a wave of the *exact* amplitude you put in and positioned in time *exactly* as it went in.


Not exactly. The *exact* amplitude falls somewhere in an infinite range, while digital is inherently finite. On a CD, for example, you can have ~65k values at any point in time, while in reality what you're sampling will almost always fall somewhere in between two of those values. That isn't to say that 65k values aren't enough (I'll leave that open to debate), but to say that you get the *exact* amplitude is untrue.


I am *not* talking about a system with infite dynamic range and infinite bandwidth. That is an impossibility as it requires infinite energy and time. I thought that was given, but I should perhaps have spelled it out.

All discussions here relate to a system with *limited bandwidth*. So as long as the Nyquist criterion is met, i.e. the sampled waveform has all components below 1/2 the sampling rate, any finite energy waveform will be encoded completely.

It's only the analog/vinyl crowd that talks about "infinite" resolution and other techno-babble.

As for ~65k being enough? Well, considering it spells out a dynamic range of 96 dB, I'd say it pretty much can encompass all recorded music to date.

Posted: Thu Jan 29, 2004 2:06 am
by thomh
krabapple wrote:But that rising noise floor problem wouldn't apply, would it, if NR is only applied to 'quiet' parts -- as is the case for e.g. the Rolling Stones SACDs?

I'm not saying overall system resolution gets better, though. I'm just wondering if it necessarily has to be worse, if NR is applied conscientiously.


I am tempted to agree with you there, Krab.

Posted: Thu Jan 29, 2004 8:14 am
by lukpac
thomh wrote:All discussions here relate to a system with *limited bandwidth*. So as long as the Nyquist criterion is met, i.e. the sampled waveform has all components below 1/2 the sampling rate, any finite energy waveform will be encoded completely.


Again, you bring up nothing about the sampling depth here. I can sample at 44kHz using *1* bit, but what I get certainly won't be a "complete" encoding of the original signal.

Even if a signal has little dynamic range, there will still be an infinite range of values between its loudest and softest points.

And, no, I'm not with the "analog/vinyl crowd".

Posted: Mon Feb 02, 2004 1:05 pm
by thomh
lukpac wrote:Again, you bring up nothing about the sampling depth here. I can sample at 44kHz using *1* bit, but what I get certainly won't be a "complete" encoding of the original signal.


What has been said along is that all signals <Fs/2 will be sampled equally accurately for amplitude, phase and frequency. All signals >=Fs/2 cannot be sampled *at all* and as such cannot be considered part of the equation.

If the "original signal" is <Fs/2, you will get a "complete encoding". The short story is that due to the non-linearity of your 1-bit system, the accuracy of the *representation* in the amplitude domain will be totally useless due to distortion and noise modulation.

Are you suggesting that more bits = a more accurate capture of the original signal, i.e. the more bits you add the more you approach a 100% capture?

Bitdepth (word size) is directly related to dynamic range which is a direct measure of the *accuracy of representation* in the amplitude domain. So the more bits there is available the deeper the broadband noise floor and hence the greater potential there is for resolving the sampled signal.

Posted: Tue Feb 03, 2004 5:01 pm
by thomh
krabapple wrote:Apparently the eyes of god are upon us:


Apparently not god himself, but one of his loyal servants.


----------------------------------------

Steve responds:

I don't want you to believe anything. All I'm saying is TRY IT FOR YOURSELF.

Thomh, Krab said I use different mastering chains. I do not. I use the same chain with a split feed, as I've mentioned here many times. PLEASE, if you wish to address something with me or about me, DO IT HERE!

-----------------------------------------

Krab responds:

A mastering chain that bifurcates at the A/D stage is no longer the same chain. Any confirmed audible difference that results -- and so far, none have been confirmed properly for your recordings -- *could* be due to CD vs DSD per se, OR to equipment differences in conversion and playback.

This is an inherent problem in comparing formats. It remains even if the comparison is between an SACD layer, and a Redbook copy of the SACD layer...which is how some other hybrids are made.

So, the order of proof should be: first, prove that you actually hear the difference (this means using a bias-controlled level-matched protocol)

Second, prove that the difference is due to the formats, and not the equipment.

Lastly, I'm willing to DO IT THERE, if the draconian rules against expression of skepticism are changed, as the Gorts know.

Over to you , thomh.


Well, let's take a look at this flimsy evidence which was put forth.

The SACD arrived up here by UPS today so I had a chance to listen to it. I used both my Sony DVP-NS905 players so that I avoided having to switch between layers and instead I could just flip back and forth between the players instantly.

Yes, I do notice a difference in the echo between the two layers of the SACD of Willie & The Poor Boys. The SACD's got more!

But, I also notice a difference between the CD layer of the SACD and the DCC Willie & The Poor Boys CD. The DCC got more! In fact, it rivals the SACD layer.

So what kind of crap converter is SH using to master his CDs nowadays?

Posted: Tue Feb 03, 2004 8:49 pm
by krabapple
This echo difference you speak of -- which I assume is apparent enough to obviate the need for a blind comparison -- if it's *this* audible, it shoudl show up as a visible diff in a .wav view of the various versions....for the SACD, you'll have to capture the analog feed , but the two CDs can be captured with no intervening analog step (not that a good A/D conversion is going to make a difference, but still) It should also of course yield different statistics for each wav.

Can you give it a whirl and post the result, time permitting? It won't prove that the diff was due to DSD vs SACD or anything like that, but I'm interested to see HOW different these are.

Posted: Tue Feb 03, 2004 9:02 pm
by krabapple
That's the disheartening thing, Krab. I *found* it on SH.tv...

http://www.stevehoffman.tv/forums/showt ... adid=27313

Ryan


Revisited this SHTv today to see if there has been any movement. The discussion seems hung up on jitter.

Would someone please note that the measuements done in the prismsound article did NOT support the claim that jitter (either mfr or playback) could be the culprit in the supposed audible differences? Which is amazing, since Julian Dunn virtually made a career , in the last part of his life, of warning about jitter.

How about noting, too, that the *listening* results reported in the article indicated that the perception of audible difference was dubious at best?

Posted: Tue Feb 03, 2004 11:40 pm
by Rspaight
I'm having too much fun with LeeS at the moment...

Besides, I've got Homero finding "differences in differences" between playback chains. This could get interesting...

Ryan

Posted: Sat Feb 07, 2004 5:49 pm
by thomh
krabapple wrote:This echo difference you speak of -- which I assume is apparent enough to obviate the need for a blind comparison -- if it's this* audible, it shoudl show up as a visible diff in a .wav view of the various versions....for the SACD, you'll have to capture the analog feed , but the two CDs can be captured with no intervening analog step (not that a good A/D conversion is going to make a difference, but still) It should also of course yield different statistics for each wav.

Can you give it a whirl and post the result, time permitting? It won't prove that the diff was due to DSD vs SACD or anything like that, but I'm interested to see HOW different these are.


OK, I tried what you suggested.

I plugged the analog outs of my SACD player directly to my LynxTwo SoundCard and did a quick and dirty sampling of Fortunate Son from the SACD layer using 16/44.1kHz. The same song from the CD layer was just extracted from the disc. The wav file, which I have linked here, contains the 4 snare drum hits from the CD layer followed by the same 4 snare hits from my 16/44.1kHz resampling of the SACD layer.

Notice how my version has a much fuller and deeper echo which disappears into the noisefloor, while the echo on SH's CD layer does not do this. You have to have a good soundcard inorder to hear it. Better yet, record the wav to a CD-R and play it on your CD player.

So this again begs the question: Why this difference in SH's CD layer when, obviously, it is not a problem for 16/44.1kHz to capture this echo?

Also notice from the JPG links that the CD layer is mastered quite hot. In fact, it clips over times.

Here are the files:

http://home.online.no/~thomh/Fortunate_Son_CD.JPG
http://home.online.no/~thomh/Fortunate_Son_SACD.JPG
http://home.online.no/~thomh/Fortunate_Son_CD_SACD.wav

Posted: Sat Feb 07, 2004 11:09 pm
by Rspaight
What's that sound I just heard? Sounded like another panicked email from a FLO lurker to Steve's personal email...

Ryan

Posted: Sun Feb 08, 2004 2:33 am
by krabapple
Listening over a pair of typical H/K computer speakers via WinAmp,
the only real difference I can claim to hear is that the CD capture is louder than the SACD...which is born out by inspection of the .wav images. I'll give it a more critical listen later. One fun test, though, would be to present these to an 'audiophile' on SH.tv who doesn't know which is which, and ask which sounds better. Betcha he picks the CD (loudness effect).

Here's the cool edit stats for the CD part of the drumbeats .wav (not being terribnly fussy about selection start and endpoints, just placing them by eye :

Left Right
Min Sample Value: -28137 -24513
Max Sample Value: 31057 27518
Peak Amplitude: -.47 dB -1.52 dB
Possibly Clipped: 0 0
DC Offset: -.014 .057
Minimum RMS Power: -27.6 dB -28.77 dB
Maximum RMS Power: -9.7 dB -11 dB
Average RMS Power: -19.58 dB -20.73 dB
Total RMS Power: -18.62 dB -19.73 dB
Actual Bit Depth: 16 Bits 16 Bits

Using RMS Window of 50 ms

and for the SACD part:

Left Right
Min Sample Value: -23985 -20832
Max Sample Value: 26223 23030
Peak Amplitude: -1.94 dB -3.06 dB
Possibly Clipped: 0 0
DC Offset: 0 .003
Minimum RMS Power: -29.02 dB -30.27 dB
Maximum RMS Power: -11.13 dB -12.5 dB
Average RMS Power: -21.05 dB -22.27 dB
Total RMS Power: -20.1 dB -21.28 dB
Actual Bit Depth: 16 Bits 16 Bits

Using RMS Window of 50 ms

Posted: Sun Feb 08, 2004 3:27 am
by thomh
Listen through a pair of headphones. It does not scream out you but the last 4 snare hits exhibit a fuller and deeper echo.

Here are the stats for Fortunate Son:

SACD
----

Min Sample Value: -32768 -30383
Max Sample Value: 32759 30397
Peak Amplitude: 0 dB -.65 dB
Possibly Clipped: 2 0
DC Offset: -.002 -.002
Minimum RMS Power: -96.34 dB -96.34 dB
Maximum RMS Power: -9.99 dB -10.24 dB
Average RMS Power: -17.44 dB -17.18 dB
Total RMS Power: -16.68 dB -16.52 dB
Actual Bit Depth: 16 Bits 16 Bits

Using RMS Window of 50 ms

CD
--

Min Sample Value: -32768 -32768
Max Sample Value: 32767 32767
Peak Amplitude: 0 dB -.01 dB
Possibly Clipped: 226 50
DC Offset: -.001 .062
Minimum RMS Power: -69.89 dB -70.51 dB
Maximum RMS Power:-8.57 dB -8.75 dB
Average RMS Power: -16.13 dB -15.8 dB
Total RMS Power: -15.31 dB -15.09 dB
Actual Bit Depth: 16 Bits 16 Bits

Using RMS Window of 50 ms

Posted: Sun Feb 08, 2004 10:13 pm
by krabapple
But given the level difference, all bets are off. Clearly these aren't the 'same' mastering, and there ain't no way in hell it can be definitely attributed to DSD vs PCM, as Hoffman does.