Why 24/192 is a bad idea?

That would be very unwise IMO.

For the record, no mod has threatened to do so, and the calls to keep things civil and on-topic are appreciated. I am nonetheless flabbergasted by the inconsistency of the warnings.
 
while i agree generally with Lee that higher bit depth and sampling rates generally/reliably sound better in various ways; 44.1/16 or even 48/16 can sound very good and it's hard to tell what depth/rate you are listening to cold. top level dacs these days do a wonderful job with Redbook. to me the sonic result always has more to do with things other than the actual PCM bit depth/sampling rate.

yes; "when all other things are equal" then bit depth and sample rate matter. and i seem to listen to PCM for longer periods when i am listening to hirez on my server than redbook. it's simply more involving. but obviously that is a very subjective thing. of course, musical enjoyment is all i care about.

OTOH dsd where the native source was analog, is relatively easy to identify cold...although not 100%.

LOL. I have been saying this fromm day one. But hardly anyone quotes me.:) I dare anyone in the Bay Area to prove me wrong. I will bring my Great Northern Sound Company Statement Wadia S7i against any and all for a showdown. Winner buys the beer/wine or should I say, "whine".
 
The use of 58 Khz is not necessary to give you more bandwidth for tones you hear but rather to provide more bandwidth to park the quantization noise and to make it easier to build DACs that have less in-band ripple.

Bob states unequivocally in the article that response extension to 26kHz is required, so he is in fact arguing for 'more bandwidth for tones you hear'. Additional room to park shaping noise is alluring, but still unneccessary.

Noise-shaping unfortunately has not taken off in creating music. People either truncate the bits or subject it to dither.

To be clear, the 'noise shaping' I've been discussing is noise shaped dither. It is still dither, just not white TPDF.
 
No, it is easily done as Bob did. If I said tape has no noise, you would quote the 70 db signal to noise ratio and say, "here it is."

That is proof of a positive hypothesis, not proof of a negative (null) hypothesis like arnyk was talking about. You cannot prove a null hypothesis, you can only confirm it.
 
One man's opinion -- I think everything -- format, resolution, to a point, even gear -- comes in way behind the quality of the recording and mastering. I recommended Jorma Kaukonen's Blue Country Heart the other day. I believe that's a digital recording, but it is so good I'll bet it would even sound good transferred to analog ;).

Tim

Tim,
I was one of the members who bought this recording on your recommendation. Great music, so good we sometimes forget the sound quality. :) However, although the instruments seem nice sounding the CD lacks real dynamics and spaciousness. Also instruments sound loud having equal loudness.

As you seem to like my comments starting with the word but I add: but the sound quality of this particular CD is not an intrinsic limitation of the CD format - my Mark Levinson CDs of Music Maker are much better in these aspects.

The liner notes say that it was an original DSD recording and that there is also a SACD - it would be interesting to know if it has better sound quality.
 
Tim,
I was one of the members who bought this recording on your recommendation. Great music, so good we sometimes forget the sound quality. :) However, although the instruments seem nice sounding the CD lacks real dynamics and spaciousness. Also instruments sound loud having equal loudness.

As you seem to like my comments starting with the word but I add: but the sound quality of this particular CD is not an intrinsic limitation of the CD format - my Mark Levinson CDs of Music Maker are much better in these aspects.

The liner notes say that it was an original DSD recording and that there is also a SACD - it would be interesting to know if it has better sound quality.

my comments about this recording relate to the multi-channel dsd layer of the hybrid SACD of it. i don't recall listening to the redbook layer.
 
Tim,
I was one of the members who bought this recording on your recommendation. Great music, so good we sometimes forget the sound quality. :) However, although the instruments seem nice sounding the CD lacks real dynamics and spaciousness. Also instruments sound loud having equal loudness.

As you seem to like my comments starting with the word but I add: but the sound quality of this particular CD is not an intrinsic limitation of the CD format - my Mark Levinson CDs of Music Maker are much better in these aspects.

The liner notes say that it was an original DSD recording and that there is also a SACD - it would be interesting to know if it has better sound quality.

Won't comment on the spaciousness of studio recordings but will say that there's nothing wrong with the dynamics in my view. Is the guitar as loud as the mandolin? Yes. Is that how it is in nature? No. But I don't think it's excessively compressed, I think it is mixed that way. Listen to the attack of those stringed instruments. It doesn't sound compressed to me. I could be wrong, though. Maybe someone who really knows compression can comment...Bruce?

Tim
 
Won't comment on the spaciousness of studio recordings but will say that there's nothing wrong with the dynamics in my view. Is the guitar as loud as the mandolin? Yes. Is that how it is in nature? No. But I don't think it's excessively compressed, I think it is mixed that way. Listen to the attack of those stringed instruments. It doesn't sound compressed to me. I could be wrong, though. Maybe someone who really knows compression can comment...Bruce?

Tim

Tim,
It seems we got the wrong version. :) Five minutes with google and I immediately found a critic saying nice things about the music and also complaining that the CD was compressed.

The acoustic instruments are well-recorded and the sound is clean and unfettered by needless electronic effect, though in keeping with unfortunate industry practice, the CD was compressed to make it louder than it needed to be for this style of music

However, many people say the SACD is mervelous and well recorded. It has a consequence - the only SACD seller I could find for sale asks usd 150 for it! Happily I do not have a SACD player.
 
Last edited:
Hi Tim,

Won't comment on the spaciousness of studio recordings but will say that there's nothing wrong with the dynamics in my view. Is the guitar as loud as the mandolin? Yes. Is that how it is in nature? No. But I don't think it's excessively compressed, I think it is mixed that way. Listen to the attack of those stringed instruments. It doesn't sound compressed to me. I could be wrong, though. Maybe someone who really knows compression can comment...Bruce?

Tim

I was going to comment earlier that I love the music but sonically, find it more than a little squashed.

Best regards,
Barry
www.soundkeeperrecordings.com
www.barrydiamentaudio.com
 
while i agree generally with Lee that higher bit depth and sampling rates generally/reliably sound better in various ways; 44.1/16 or even 48/16 can sound very good and it's hard to tell what depth/rate you are listening to cold. top level dacs these days do a wonderful job with Redbook. to me the sonic result always has more to do with things other than the actual PCM bit depth/sampling rate.

yes; "when all other things are equal" then bit depth and sample rate matter. and i seem to listen to PCM for longer periods when i am listening to hirez on my server than redbook. it's simply more involving. but obviously that is a very subjective thing. of course, musical enjoyment is all i care about.

OTOH dsd where the native source was analog, is relatively easy to identify cold...although not 100%.

Oh no question that great CD can sound pretty damn good. But if we can get hirez, I agree it is more natural and provides a non-fatiguing listening session.

As for DSD, I find our DSD ADCs get closer to the mic feed than our PCM converters. But that is a whole other discussion.
 
One man's opinion -- I think everything -- format, resolution, to a point, even gear -- comes in way behind the quality of the recording and mastering. I recommended Jorma Kaukonen's Blue Country Heart the other day. I believe that's a digital recording, but it is so good I'll bet it would even sound good transferred to analog ;).

Tim

Tim, in my experience mastering is really important as is the original recording.

Mastering: It's pretty clear if you buy the DCC gold CDs and compare them to regular CD releases. The differences can be startling. I have collected almost all of Steve Hoffman's several hundred masterings for DCC on both gold and silver. It is an expensive hobby but the sonic rewards are great.

Original Recording: Here if we are talking audiophile releases or small ensembles anyway, we are really talking microphone placement and probably to a lesser extent recording chain and mixing choices. I've been moving mics around since 1990 and I still learn something every new session. If the mics are poorly placed then no magical technology or mastering is going to save you. This is why the otherwise great Rudy Van Gelder cannot do better with his piano sound which is often imho poor. Fortunately the Blue Notes sound so damn good in other dimensions we can easily overlook it for most of the time.

On the Hoffman board I often say you need three things: a quality original recording, a quality mastering, and a high resolution format (ime LP, SACD, 24/88.2 or higher PCM).

Caveat, if you really do a reference job on the 16/44.1 like the FIM and LIM K2HDs then regular redbook can be pretty damn amazing too.
 
The liner notes say that it was an original DSD recording and that there is also a SACD - it would be interesting to know if it has better sound quality.

I have both. Interestingly enough the SACD has no hybrid layer. At least mine doesn't have one, its one of the few I have like that. Plus I never found the cd to sound anything but topshelf. My I2S connection dropping that lsb must be working wonders with this disc. ;)
Course, once I got the SACD I didn't play the redbook much.
 
(1) It doesn't mask it, it actually decorrelates it. The distortion peaks eliminated by dither sit well above the dither level.

(2) BTW, is there anything special about your simulated DACs that I should know about (they look like a tone + white noise) or your quantization process? I am able to replictate the simulated DACs easily. So far I'm seeing the behavior I expected: truncation acts like rounding with a DC offset. Also, both your 24->16 conversions show background white noise levels much higher than expected at ~ -130dBFS. Was more white noise injected?

Of course, I'm still looking for any bugs that might be mine. I can up some graphs/test data if you like.

(1) Semantics; the verbiage I was weaned on when first introduced to dither (in a radar system, in the 80's) was that the dither "masked" the quantization noise by decorrelating it. We used specially shaped dither (not white or anything like it) to help decorrelate the noise floor and extract low-level returns ("spurs") through averaging (which evolved into much more complicated DSP filters, but I am a hairy-knuckled analog engineer, not a mathematical analyst or physicist). However, dither had hardly any effect on nonlinearities such as amplifier/mixer THD/IMD or nonlinearities in the data converters themselves. Essentially the dither raised the noise floor (decreased SNR) slightly while masking (decorrelating) quantization spurs, but other nonlinearities such as front-end distortion were correlated to the signal and thus not affected by the added noise (white, pink, brown, or otherwise band-shaped). I suspect we are defining "distortion" differently here... I tend to distinguish between sampling artifacts and system nonlinearities (including those added by such things as threshold bows that cause distortion in the converters) -- the latter is distortion to me. Arguably any sampling induces distortion, again this is the way I was weaned so we could distinguish sampling "distortion" from more conventional distortion terms.

(2) The "DACs" in those simulations are extremely simple, just ceiling functions to create integer steps from a sine wave. I generated the sine wave based upon the IEEE Standard for ADCs (1241) so I did not have to window the FFT. I don't remember off-hand how I truncated; I usually scale to the resolution desired then simply truncate the figures past the decimal (binary) point. The truncation problem has been around for ages, at least in my world, so I did not think anything special of the spurs. Truncation loses information and so causes spurs as if the lsb was corrupted. This is somewhat an artifact of the signals I am using; more random signals with a normal window function may not exhibit it to the same degree. All my programs use the IEEE method so I can't say for sure. I have not thought of it as a d.c. offset (not saying it is wrong, I have not thought of it that way and do not know). I do suspect dither would reduce those particular spurs by randomizing the errors since they are an artifact of truncation and not a "normal" nonlinearity.

I am not sure your concern with the noise floor other than it is higher than ideal? Ideally 16 bits would yield around 144 dB (9N) SFDR, but I ran the simulations on my old, old notebook. It is slow and so I only did 64K-point (I think) FFTs. More points would drop the noise floor; that wasn't needed to show what I wanted to show in that thread and I was too impatient to wait. Few real-world converters hit that sort of noise floor, though I admit I have not looked for a while. My experience is with GS/s stuff and fewer bits; I have not spent a lot of time looking at audio DAC data sheets.
 
I am not sure your concern with the noise floor other than it is higher than ideal?

That it is much higher than only rounding/truncating the 24-bit simulated DAC can account for. My own operation, after producing an approximately identical simulated 24 bit DAC, yields a much lower noise floor + harmonic distortion, which is what the math says should happen. So, we're doing something different, I am interested in what it is.

Ideally 16 bits would yield around 144 dB (9N) SFDR, but I ran the simulations on my old, old notebook. It is slow and so I only did 64K-point (I think) FFTs.

That's 144dB assuming additive gaussian noise, isn't it? I'm attempting to replicate your operation identically with 64K-point FFTs, and without additive noise, I get a lower floor + ~ -100 to -110dB spurs from both truncation and rounding. The only difference between the two is a DC offset in the truncation case.

More points would drop the noise floor; that wasn't needed to show what I wanted to show in that thread and I was too impatient to wait. Few real-world converters hit that sort of noise floor, though I admit I have not looked for a while. My experience is with GS/s stuff and fewer bits; I have not spent a lot of time looking at audio DAC data sheets.

Right. I was trying to replicate exactly what you'd done, not critique the point you were making when you generated the graphs.

Oh, and full agreement-- dither only decorrelates the distortion inherent to input correlation. It does nothing for nonlinearity flaws.
 
The 144 dB comes from the (ugly) math to determine the quantization noise floor for a given number of bits. It is roughly 9N for N bits (vs. the SNR which goes as roughly 6N). There is no additional noise source, just quantization. I don't recall if Guassian is assumed in the derivation but it seems likely; the derivation uses Bessel functions and high-level math that makes my head hurt. :) What are you seeing for a noise floor?

It is also possible I messed up and there is a noise (or jitter) source I forgot to turn off. I used a general-purpose file with a ton of different things in it and may have left something enabled that I shouldn't have... I don't have time to repeat tonight but will try to look into it later this week. It may be as simple as not using enough significant digits in the functions; most of my work has been 6 - 12 bits with a few designs hitting 16+ (no mean feat at several GHz). 1 ppm is not good enough for these 16+-bit simulations.

I can also find out exactly what I did in the program. Since it gave me the result I expected I did not look further. Perhaps I am all wet.
 
I don't have time to repeat tonight but will try to look into it later this week. It may be as simple as not using enough significant digits in the functions; most of my work has been 6 - 12 bits with a few designs hitting 16+ (no mean feat at several GHz). 1 ppm is not good enough for these 16+-bit simulations.

No hurry. I found an unrelated bug in one of my own analysis packages in the course of playing with this, so I'm going to take the time to hunt that down anyway. I didn't get an exact noise floor before ripping it all apart, but it was below -160dB with spurs up to -105dB. Nor have I triple-checked my work yet.
 
To be clear, the 'noise shaping' I've been discussing is noise shaped dither. It is still dither, just not white TPDF.
That is not how noise shaping is talked about or implemented. The two are distinct operations with one shaping the quantization noise and then dither neutralizing distortion. http://en.wikipedia.org/wiki/Noise_shaping

"Noise shaping is a technique typically used in digital audio, image, and video processing, usually in combination with dithering, as part of the process of quantization or bit-depth reduction of a digital signal. Its purpose is to increase the apparent signal to noise ratio of the resultant signal.
[...]

Noise shaping works by putting the quantization error in a feedback loop. Any feedback loop functions as a filter, so by creating a feedback loop for the error itself, the error can be filtered as desired. The simplest example would be:
ef1dce537218273f5801be5312970a99.png

[...]
Noise shaping must also always involve an appropriate amount of dither within the process itself so as to prevent determinable and correlated errors to the signal itself. If dither is not used then noise shaping effectively functions merely as distortion shaping — pushing the distortion energy around to different frequency bands, but it is still distortion. If dither is added to the process as:
36d84c6207fa91515f9d0f1664aadcf6.png
"


So it is clear that the two concepts are distinct and spoken about as such.

Now if you mean dither with different probability distributions, then that is not the use of it by Bob or me in the context you quoted me.
 
To be clear, the 'noise shaping' I've been discussing is noise shaped dither. It is still dither, just not white TPDF.
That is not how noise shaping is talked about or implemented.

My usage is an accepted variant, eg, "Dynamic Range Enhancement Using Noise-Shaped Dither Applied to Signals with and without Pre-emphasis", Authors: Bob Stuart and Rhonda Wilson. "Noise shaped dither" makes it clear that the noise shaping and dither are both happening together in the quantization loop.

However, I concur that it is a separate concept.

Now if you mean dither with different probability distributions, then that is not the use of it by Bob or me in the context you quoted me.

That's not what I meant.
 
Last edited:

About us

  • What’s Best Forum is THE forum for high end audio, product reviews, advice and sharing experiences on the best of everything else. This is THE place where audiophiles and audio companies discuss vintage, contemporary and new audio products, music servers, music streamers, computer audio, digital-to-analog converters, turntables, phono stages, cartridges, reel-to-reel tape machines, speakers, headphones and tube and solid-state amplification. Founded in 2010 What’s Best Forum invites intelligent and courteous people of all interests and backgrounds to describe and discuss the best of everything. From beginners to life-long hobbyists to industry professionals, we enjoy learning about new things and meeting new people, and participating in spirited debates.

Quick Navigation

User Menu