I keep seeing arguments like this 'against' CD (mainly on forums where vinyl vs CD is still a topic of concern): "16bits offers 96dB of dynamic range but it's not linear: low level signal (e.g. at -48dBFS) only 'get' 8-bits of resolution and that's not enough because we all know how bad 8-bit sound is."

Help me out here with a technically accurate response. My copy of Pohlmann is currently in storage.

I see we're still reading the same SH.tv threads.

I can think of two retorts. First, there is a plethora of evidence that 8 bits at -48dbfs is perfectly reasonable and inaudible for commercial music productions. All evidence to the contrary generally involves listening environments that don't exist in reality (ie, situations where you would crank the volume up 50db), or involve completely unverifiable listening experiences.

Second, people still say that quantization noise masks low-level detail on CD, but somehow, vinyl noise doesn't mask the detail. This completely ignores the use of dither and noise shaping in digital playback, and also ignores the large amount of broadband noise present in vinyl, due to both surface noise and thermal noise - which is, spectrally, almost the same as dither, and also happens to be 20db louder in level. In other words, the low-level detail argument is completely unsupported by numeric evidence, unless one takes another's subjective experience as gospel.

Generally, if you're arguing against people who believe their subjective experience of something is infallible, technical accuracy doesn't get very far.

One word: dither.

One demo: 8-bit sound with dither - which, as expected, demonstrates a 48dB SNR - but nothing worse.

(Noise shaping can even make 10-12 bits sound "good enough" - i.e. 70-80dB perceived SNR)

EDIT: Axon was faster!

Cheers,
David.

I have actually in the past tried to explain the use of dither to such people, but either I didn't do it well enough, or they just couldn't hear it over the LALALALA_LPS_ARE_GREAT noise in their heads However, isn't dither mainly useful when starting from higher bit depth then going to a lower one?

Axon have you tried your reply on SH.tv? I've been banned for ages now, but I still read and fume along. Subjective preference is one thing, but some peopel try to get technical...and that's where they
should be called on it. There's just a sort of silly jujitsu going on in those arguments, and I'd like to see the fallacies pinned down . It bugs me when people say a 16-bit format isn't 'really' 16-bit in this fashion.

I think one fallacy is that people confuse recommendation against use of 16 bits for recording and production, with recommendation against it as a consumer delivery format.

Primarily, yes. Its uses deal with quantization and re-quantization.


Agreed. For production, 24 bits (or even 32 bits) can be very beneficial, but for the end user, those benefits are much less obvious and far less critical. Yes, precision and detail is being sacrificed in the jump from 24 bits per sample to 16 bits per sample, but sacrifices are made with practically any form of consumer format, vinyl included.

Let me put it another way. Let's suppose a purist 16-bit recording of music with some very quiet passages (e.g. fades to silence), with no intervening digital production, released as a CD. I know these hardly exist commercially these days, if ever, but let's suppose. Would dither be used or even be necessary?

Just say

"It's a whole lot better than an LP recorded 48dB down, though".

I already can predict the answer to that: 'But LP resolution is LINEAR!" What follows then is a torturous
debate about the meaning of 'resolution', ending with something along the lines of 'well, I'm not sure what it means technically, but I know it when I hear it' from the other side. Been there, done that.

I've read estimates that the bit-depth equivalent of LP resolution is about 14 bits, compared to CD's 16. So I guess the question to those folks would be, 'is it still 14 bits when the signal is 48 dB down?'

Quantization, in terms of digital audio sampling, is a destructive process by nature. You're taking a theoretically infinite resolution and attempting to represent it with a finite number of bits, the bits themselves having only a binary resolution, and quantization error is a factor in the first sampling. No matter what the bit depth, there is always a degree of error introduced due to the quantization process and the reality of finite bit resolution. Ergo, dither is typically introduced by ADCs, but natural/analog sources of noise, like tape hiss, might be sufficient at fulfilling the same purpose.

In practice, you can be fairly reckless with dither without audibly impacting the signal, so there's no real advantage in trying to minimize its application.

From my reading, dither was typically added, in the analogue domain, in the days when recordings were primarily done at 16 bit. Analogue dither randomized “regular” signals, such as music, and eliminated quantization distortion. Soundcard ADCs do not add dither and it is not used when recording at 24 bit, as far as anything I’ve ever run across.

There is enough broadband noise and tape hiss from vinyl and cassette to adequately dither 16 bit transfers. Some poorer soundcards are so noisy that their noise might serve as a kind of dither but that noise often has regularly repeating patterns ,such as hum and harmonics of same, or noise from switching power supplies, so it is not good dither.

Dither is not used in post recording processing when using 24 bit or floating point, especially the later. It is only used when reducing the bit depth, which is normally not done until everything else is finished. It was used at each step, or at least at many steps, when mixing and mastering was done at 16 bit, but that is a little dicey. Dither is “benign” noise but very many successive applications and it becomes quite audible if not well noise shaped.

I have a related question. Assuming that in a PC the volume adjustment (and other modifications such as equalization) are performed digitally, are these calculations done at greater than 16 bit depth, and is dither then applied before outputting to the DAC?

I think window "Master volume" is analogue.

for the other, I'm not sure.

Does that mean that as long as all of the other sliders are at 100% and no equalization is applied, the 16-bit DAC is getting exactly the bits from the CD (assuming all-lossless) and volume reduction is analog, requiring no dither?

It is possible to digitally control analogue signal levels, such as with some voltage controlled amplifier chips, but except in regard to microphone preamp gain, few soundcards have any analogue variables. Output level changes are normally done in DSP chips prior to the DAC. The Windows mixer, only used with gaming and low-end soundcards, may be entirely software (thus obviously digital), but I’ve never looked into whether there is a standard hardware interface specified for its use.

The DSP chips are usually 24 or more bits (36 bit on the M-Audio Delta series) so quantization errors are too small to matter. I’ve never seen any mention of adding dither. The Echo Mia, and probably others in the Echo line-up, have an option for adding dither to the S/PDIF input. The manual says to use this only when receiving 24 bit data that is to be recorded at 16 bit (who would?). This indicates the ability to provide hardware dither can’t be too big a deal, but there really isn’t much reason to want to. If it were useful, it would almost certainly be a feature touted in the features list.

No LP I've seen was good to 14 bits.

12, maybe.



I wouldn't put it quite that way. There is no such thing as "theoretically infinite resolution" in the real world, signals are carried by electrons that have finite charge, and that charge is of a size that is easily big enough to measure the effects of in standard electronics.




In a word "no".

Sound cards vary so much in their methods that a generalization is unwise, to say the least.

Aside from such considerations, it is well established that a properly bandwidth limited signal is completely captured and reproduced digitally (In this universe. That one inhabited by certain audiophiles runs by different rules, of course). Quantization errors are irrelevant if the bit depth is high enough. 16 bit is high enough except for very low level signals and dither pretty much saves the day there.

Please link!

An old demo of 6 bits with and without dither is available here:
http://mp3decoders.mp3-tech.org/24bit2.html#dither

It clearly shows how dither does exactly what it's supposed to.

Cheers,
David.

http://www.stevehoffman.tv/forums/showthread.php?t=130348

Oh dear.

The problem is these poor deluded people are confusing two things.

Yes, an LP that's carefully mastered without excessive compression will sound much better than a CD that's clippressed to death.

However, as many people heavily "in to" vinyl will attest, that LP can be copied perfectly onto CD and lose nothing (except a few clicks if you carefully declick it!).

Plus, of course, the master tape that was the original source can also be copied onto CD - giving something superior to both the vinyl, and the clipressed CD.

If you think...
master tape > vinyl > CD
...sounds better than...
master tape > CD
...then clearly you like noise and distortion. (Most people do, in moderation.)


And as for this...


Well SACD only has two discrete levels to represent the entire loudness range - how bad should that sound?

I love it when people who don't understand science and engineering try to "prove" things using misunderstood "facts".


I think I once heard the "shortcomings" of 44.1/24, but it wasn't in a statistically significant test...

http://www.hydrogenaudio.org/forums/index....=9311&st=50

Certainly what I heard bares no resemblance to the problems most audiophile report with CDs: IMO they can do cymbals, dynamics, and perfectly smooth sound just fine when mastered properly. These commonly report faults of CD sounds are down to bad mastering. AFAIK any other faults have yet to be identified in a statistically significant way in a blind test using sensible content and listening levels.

Cheers,
David.

That's a great example. To wit:

http://www.stevehoffman.tv/forums/showpost...mp;postcount=66




and just a few posts later




My god....several pros in the music industry might be wrong about digital audio?

Alert the media.

It's easier for studio pros to be fooled. I've been fooled myself in the distant past. Take some rich instrument with a fast attack and decay (Moog synthesizer works nicely . Run it through a high quality reverb. Listen at 16 bits (truncated) and then again at 24 bits. HUGE difference in the reverb trail, especially at high volumes. Now try it again with 16 bit output but properly dithered and the difference goes away.
Most studio software isn't dithering the "monitor" output (dither only gets applied on final mixdown). If you're playing with a mix and decide to switch your soundcard settings between 16bit and 24bit, you'd reach the same conclusion that 16bit "sucks".
This is what dither is for, and why it's so awesome

The 16-bit versus 24-bit debate to me is like this. This is how I understand it.
A dithered 24-bit file to 16-bit file retrieves low level signals which would be lost below the noise floor.
Dithering does not increase low level signal with higher resolution. Low level signals and resolution are two different things. 24-bit files sound better but not by much, but I would still prefer listening to a good 24-bit recording. (19-bit in reality if you have a good DAC.)
But if you cannot find a difference between 16-bit and 24-bit, by all means stick to 16-bit. (especially on a good
cd player or soundcard).

For years, on various forums, I’ve been asking for a sample of any genuine music (i.e. not test signals) for which it is possible to ABX the 24 bit version from a properly resampled and dithered 16 bit version. If you can provide one, you will be the first.