Topic: Analog Transfers, why 24 bit?

[Sloan Simpson]

Some folks can't hear the difference between 16 0 24 bits......to bad for them they are missing out on some great things.

What are we missing? Can you describe the difference?

[runonce]

When i do a live concert recording now in 24/96, i can hear a big difference between that one and a down sampled 16/44.1 version.

And that difference is...? Better...worse?

[runonce]

Is this analogy appropriate?

If the net catches the butterfly...is it any more caught by a bigger, finer net?

[raymonda]

Live mic feed processed through an ad.....then da.......monitors....or headphones or analog transfers processed and listen back the same way.

Not repeatable, therefore not a proper test.

So........8 bits is enough........well maybe 8 bits of headroom but what happens when you have 4 bits of headroom and you are recording at 8 bits.....much of your resolution will be at 4 bits.

We are talking about analog transfers, there is no need to leave headroom, certainly not 24dB.  That would be incompetent engineering.  Properly recorded, you could safely truncate such a 50dB dynamic range source to 12 bit.

Some folks can't hear the difference between 16 0 24 bits......to bad for them they are missing out on some great things.

Rather than insulting my hearing, try understanding the science behind digital audio instead.

Or, give me a 24 bit file of your choosing to convert and then take an ABX test.

You have got to be kidding me. I'm not trying to give you a test to repeat. I can repeat my test with myself over and over again a thousand times. If you want....Please come over to my home studio....we can have a beer and we'll see how well you can discern between 8 bits.....16 bits......or 24. For those that can not hear the difference between 8 bit and 24........I feel sorry for you. And again......for pcm the premise that analog resolution is equal to 8 bit is totally wrong head. People are confusing dynamic range with resolution. 

Again.....no insult meant......but if anyone can't hear the difference between 8 bit pcm and 16 or 24 bit.....they have defective hearing.......they are certainly not ears that I would trust.

[DSatz]

DATBRAD, the idea that "more bit depth is always better" is not well based in what happens in the real world of signals and noise.

Can we go back to the world of analog tape for a moment? One of the stock examples in discussions of noise is that if you copy from one analog tape to another, with everything kept strictly equal (same type of recorder and tape, same levels, same use of Dolby, etc.), the noise level on the copy will be 3 dB higher than on the original. That's because the tape for the copy has the same physical noise floor as the original, and the noise from the two combined, identical sources is twice what you started with. OK, but the unspoken assumption is that all noise in the original recording came from the inherent noise of the tape. Since that's never true in the real world, we'll come back to that.

Also, what if you found that the peaks on the original recording only went to -10 dB? Then as we all understand, the quality of the transfer could be improved by raising the peak levels to 0 dB on the copy (or even higher, because this is analog). Now the noise buildup from making the copy is no longer 3 dB; with a 10 dB spectrum-wide boost it will be under 1 dB. You could say that the 10 dB higher noise floor of the original tape is drowning out all but a tiny amount of the second tape's inherent noise.

Finally let's focus on that <1 dB noise increase. To me the most important thing about it is that it's essentially inaudible, even at extremely high listening levels. You can tell when continuous tones go up 1 dB in volume if they're loud enough--but you can't reliably hear a <1 dB change in constantly shifting, random noise at background levels. In the presence of program material at much higher levels, you can forget about it.

These simple situations don't change much if you move into the digital realm. Once you have enough bits so that the inherent noise of your digital recording channel is ~10 dB below the noise floor of the analog signal, the whole recording process adds <1 dB of noise. And yes, if you added more bits, you could reduce the fraction of that <1 dB of noise that's being added; your theoretical concept is not entirely wrong in that respect. But each extra bit reduces the noise floor by a smaller and smaller fraction of that fraction of a dB. Nothing audible is gained as a result.

--best regards

P.S. added later: Your concern about the low-level signals being represented by a smaller number of bits is unfounded. Low-level signals are still represented by all the bits; digital sample values are fractions, not integers, and "zeroes after the decimal point" (as in 0.000442 or its binary equivalent) are meaningful and indispensable parts of the value being expressed.

[edtyre]

Is this analogy appropriate?

If the net catches the butterfly...is it any more caught by a bigger, finer net?

Is this analogy appropriate? You=Asshole

[Marshall7]

Is this analogy appropriate?

If the net catches the butterfly...is it any more caught by a bigger, finer net?

Is this analogy appropriate? You=Asshole

Actually Ed, while I doubt it was his intention...doesn't his analogy actually support your initial point?

[DSatz]

raymonda, "resolution" has no generally accepted definition, and is rarely used by audio engineers unless they're engaged in marketing.

My own definition, if I used this term (which I generally avoid doing, because it seems irresponsible) might be based on the audible difference between the input to a process (such as digital recording) and the output of that same process. If that difference--the inaccuracy or error of the process--is difficult or impossible to hear, the resolution is high. If the difference is easily audible, the resolution is low. If the difference is zero, the process is perfect, and please have your engineer give me a call!

The notion of the error isn't just theoretical, since you can derive an actual error signal by subtracting the input from the output; you can then listen to this signal, measure it in various ways, and try to relate the kinds and amounts of the error to its audibility. Its amplitude isn't the only thing that matters; its relationship to the original signal is important as well, and this is difficult to include in any notion of resolution that is merely one-dimensional (high vs. low).

For example, in an undithered (truncated) digital recording that doesn't have enough bits, the error signal's amplitude will vary in response to changes in level in the program material, and you can hear distorted but recognizable parts of the program material, that are either no longer in the recorded signal, and/or their inverse is present the recording as a form of distortion. That's what sounds so wrong in undithered digital recordings that lack sufficient bit depth. But in a properly dithered digital recording, regardless of bit depth, the error signal consists entirely of random noise. It varies in level somewhat as random noise must do--but it does so "on its own"; there's no correlation between the momentary noise level and anything that's going on in the program material at that time.

Proper dithering lets you hear "down into the noise" and perceive actual details of the sound that the recording has preserved. It gives digital recordings the exact same resolution as a good analog recording with the same signal-to-noise ratio would have. There's no more "stairstep distortion" and the only loss of audible signal components that occurs is due to masking by noise--exactly as with analog recording. The noise that exists in all signals limits the "resolution" of any recording that you can possibly make of them. If your recording system allows the noise of the incoming signal to predominate, to the point where you can't humanly distinguish whether the recording system's noise is there or not, then that's as high-resolution a recording as you can possibly make, analog OR digital, of that signal.

Beyond that point, where digital recordings are concerned, adding more bits doesn't improve the sound any. But a lot of people don't realize this, so they're willing to pay extra for 24/96 transfers of original recordings that have considerably less than 16-bit content to begin with. Insert your favorite pun here regarding the invisible "hand" of the market and what kind of "job" it is performing for those people ...

--best regards

[runonce]

Is this analogy appropriate?

If the net catches the butterfly...is it any more caught by a bigger, finer net?

Is this analogy appropriate? You=Asshole

Well fuck you too!

[datbrad]

DATBRAD, the idea that "more bit depth is always better" is not well based in what happens in the real world of signals and noise.

--best regards

I agree with you that when an analog tape is under-saturated, with levels down -10db as in your example, using 24bit word length will not produce better results than 16bit. However I still contend that analog recordings properly made with levels reaching close to tape saturation will be more faithfully preserved using 24bit PCM.

While these views are my own, they are also shared by the International Association of Sound and Audiovisual Archives, the Co-ordinating Council of Audiovisual Archive Associations, the Southeast Asia-Pacific Audiovisual Archive Association, the Association for Recorded Sound Collections, and the National Recording Preservation Board of the Library of Congress.

24bit for archiving analog recordings is globally accepted as a "best practice" for a variety of reasons. I found a paper produced by NRPB following a symposium on this very topic that may help with understanding why this is. It was published in 2004 when these standards were first being discussed and I think does a far better job than my feable attempt to make the case for 24bit preservation of analog sound recordings.

http://www.clir.org/pubs/reports/pub137/pub137.pdf

Great topic, and one that certainly has folks divided into one of two camps, those that think the highest resolution format available should be used for archiving, and those that only seek a method deemed "good enough".

[datbrad]

Jon, I think there are a couple of reasons that 24bit is preferred as a standard for the preservation of recordings originally made on analog tape that were not mentioned in that paper. One situation would be when there are artifacts from degraded magnetic media. Any attempt to overcome these artifacts with a DAW will be faced with significantly more summing errors if the digital recording of the analog tape was 16bit, compared to 24bit. Properly applied dither works well to address quantitization errors with 16bit to produce better than 16bit resolution, provided there is no requirement to "clean up" the recording. The thing that I get hung up on is the difference in possible values to represent the reconstructed waveform between 16bit and 24bit. I think it's something like 65,000 for 16bit and 16 million for 24 bit. When I consider this fact, I can't help but think this means the reconstructed waveform using 24bit will more faithfully resemble the analog waveform that was recorded compared to what would be seen with one at 16bit. Am I missing something? Would it not be better to have a digital wave form that more closely resembled a smooth sine wave and less like stepped square wave? This is definitely a deep subject.......

[DSatz]

DATBRAD, no. Simply, no. Nearly any time anyone is concerned about increasing the "possible numbers to reconstruct the waveform" or whether there are enough "possible numbers" available for low-level signals, it shows a fundamental misunderstanding. With real-world digital recordings, things actually work in a way that's just the same as with analog: Once the noise of the recording channel is far enough below the program material to be inaudible, that noise doesn't become "more inaudible" if you increase the channel's dynamic range further (e.g. by adding more bits if the recording is digital).

Please take a step back and restudy the issue, and/or I and others here will gladly continue to discuss it with you. But please realize that what you're saying isn't fact- or experience-based, and isn't a matter of individual opinion, either; it's a misunderstanding both in principle and in fact.

That said: Many bright, honest people are misinformed about bit depth and "resolution", and have been for decades in some cases. That is the real explanation for why certain commercial products use "24-bit" quantization when to do so brings no possible audible benefit (and this is apart from the fact that no "24-bit" converter has 24-bit quantization). If you were in marketing, you'd be hard-pressed to risk your company's profitability by saying to audiophiles, "We released this product in a less expensive form because you only delude yourself about what you can hear." It's a much better marketing message to say, "We spared no expense to bring you the highest possible quality, because we care and you deserve it" even when that isn't really true.

--best regards

[sacchini]

I'm impressed of your tech knowledge about this topic.
Since I'm not an audio engineer but I know that the storage cost is not so high, I will be happy to have transfers at 24 bit.
Not all are sure about 24 bit being better (audible) but no one think that 24 bit may be worse.
So for me it's an easy choice...