Topic: Bits v. kHz

[DSatz]

Hi, uncleyug. Let's try to reason through this carefully with a practical example.

Let's say I set up a pair of microphones to record a concert. I've been to a rehearsal beforehand, so I know what to expect as far as the sound levels are concerned--and as usual, I've allowed 3 or 4 dB of headroom for last-minute enthusiasm on the part of the musicians. From then on, the dynamic range of the recording is up to whatever the musicians decide to do. Whether they use that 3 or 4 dB at the top or not, the recording might be sonically first-rate or not, depending on all the usual variables.

However, 3 - 4 dB is easily audible as a difference in loudness--and even a 1 dB difference or less in playback levels can have a real impact on people's opinions about the quality of a recording. Hi-fi salespeople have known this for years: If you can even sneak the level up just 1/2 or 3/4 of a dB on the pair of loudspeakers you want the customer to buy, people start hearing "more detail" and "more emotion" and "better imaging" and "more soundstage depth"--all sorts of qualitative differences, rather than noticing, "Gee, it sounds a tiny bit louder than it did a moment ago."

So if there's 3 or 4 dB difference between my recording and (say) one that the client got from another engineer a month before, then--well, my recording might be worlds better sounding if the client would only make the comparison with the volume levels set scrupulously equal in playback. But the average musician who's not an engineer will simply listen to one recording, listen to the other, and make a snap judgment: My recording just doesn't have the "life" and "energy" of the other (louder) recording.

By the way, this kind of situation has led many good people to firm but mistaken conclusions about all kinds of things in audio. And those people really have heard what they say they've heard--but the uncontrolled conditions make it quite impossible for them to know why things sounded as they did.

I mean--on one level I really can't argue with the client; he pops in the other guy's CD, and he pops in mine, and the other guy's CD sounds better. If there's something I can do to make my work sound better in such a comparison, the client naturally assumes that I should already have done it--what am I waiting for?

So if I'm in one of those situations where I absolutely must make things sound as good as possible in a comparison, I will recopy my recording so that it peaks at -1 dB or even -1/2 dB below digital full scale. And the moment that kind of signal processing comes into the picture, it's better if I'm starting from a 20- or 24-bit live recording.

Think it through: If I start from a 16-bit live original, then copy that in order to raise the levels by 3.5 dB, then in the hypothetical best case, the noise floor of the result would be 3.5 dB higher than before. But if I'm starting from an 18- or 20- or 24-bit live recording, then the noise floor of the recording doesn't increase at all when I boost the gain by 3.5 dB. The resulting CD will in fact have 3.5 dB wider dynamic range than if I boost a 16-bit recording by the same amount.

If you want to make the comparison more realistic by considering dither, then you can simply consider the dither itself to be the digital noise floor of the recording. The one which starts from a dithered 16-bit original will still be 3.5 dB noisier than the one that starts from dithered 18, 20 or 24 bits.

What mitigates this considerably in many cases is the acoustical noise in the recording venue--but that isn't constant at all times or all frequencies. Our ears and brains are most sensitive in the 2 - 5 kHz region and there, every dB and every bit really counts sometimes, at least in the classical recording that I mainly do.

--best regards

[yug du nord]

^^^Once again man, you've schooled me in a fashion that this simpleton can understand!  Thank you!!!  Makes complete sense.  It's always a pleasure to catch onto a thread that you're involved with.  +T every time I see ya!

[Lil Kim Jong-Il]

Even if today's target is CD, someday you will likely upgrade your playback system to 24-bit playback.  I record at 24/96 and use DVD.  Although I have not heard a difference between 96k and 48k/41k, the difference between 24 and 16 bit is clear on my system and storage is cheap.  Processing to 16-bit is not that big a deal and even with a low powered machine you need to do it only once for a show you intend to archive and circulate.  

If you are mastering down to CD for the car, you aren't likely to perceive any real difference between a native 16-bit version and a dithered 24-bit version.  The ability to capture a high SNR recording with conservative levels at 24-bit is priceless when you are in a situation where monitoring the signal level is not possible and you need to be concerned about large transient signals.  

I would even tend to question if you can get such a thing as a true native-16 bit copy from a 24-bit recorder.  The ADC or firmware may be downsampling to 16-bit a 24-bit native sample stream anyway between the sampler and the wav file.

[sml42]

But if I'm starting from an 18- or 20- or 24-bit live recording, then the noise floor of the recording doesn't increase at all when I boost the gain by 3.5 dB.

To be pedantic, the noise floor does increase, it's just still way way below the 16-bits offered by CD

Slightly more seriously, there was another discussed on these forums some time ago. Think it was originally started by TRB, and damned if I can find it, but the gist of it is as follows: even if you were running at 24-bit, and could set your levels perfectly (ie. coming close to 0dB but no clipping) you shouldn't. Instead you ought to aim to peak at -6dB, possibly even -12dB, and add gain in post-processing. The rationale given was, traditional class A amplifiers have more distortion when you crank them hard, so back it off a little - you have extra bits for the dynamic range. The assumption being, the preamp exhibits such behaviour. Now, that's how the thread went, and for the record I'm not sure I buy into it totally; in practice the real-world means you won't come close to 24-bit, I'd be surprised if my MT2496 gave anything better than an effective 18-bits, so no way would I want to be peaking at -12. Maybe if I had some higher-end gear... maybe not.

[sml42]

However, 3 - 4 dB is easily audible as a difference in loudness--and even a 1 dB difference or less in playback levels can have a real impact on people's opinions about the quality of a recording

On the basis that 'louder is better' I will also point out that a small amount of compression can help achieve this... that is of course a whole differnt discussion, so I will shut up now

[Lil Kim Jong-Il]

that is of course a whole differnt discussion, so I will shut up now

light fuse, step away? 

[digifish_music]

However, 3 - 4 dB is easily audible as a difference in loudness--and even a 1 dB difference or less in playback levels can have a real impact on people's opinions about the quality of a recording

On the basis that 'louder is better' I will also point out that a small amount of compression can help achieve this... that is of course a whole differnt discussion, so I will shut up now

This YouTube video on the 'Loudness War' presents one side of the story...

http://www.youtube.com/watch?v=3Gmex_4hreQ

http://www.youtube.com/watch?v=xkkqsN69Jac

Being in the music production business I am interested to look at how many commercial tracks are mastered, and it's not pretty. That said, I regularly see tracks that look like they have been destroyed by over-compression (judging from the audio-editor peak-view), and yet they sound good at all listening volumes. It does however depend on the style of music.

digifish

[boyacrobat]

loadness war is by grand design.
portals that are dimensional found in sound
must be closed,squashed .

this task has been given to the mastering engineers.
blame them more than the producer of the album, but blame them as well.
most still sleep the big sleep.


g

[DSatz]

just sml, thank you for your postings. Compression is certainly an option to get higher long-term average levels onto the transfer of a recording. When to use it, if ever, is a whole other discussion. But from a technical standpoint, if you're going to use it, it's unquestionably preferable to start out with more than 16 bits in the live recording--otherwise the noise will "breathe" or "pump" from moment to moment as the compressor changes the overall gain.

Noise that rises and falls rapidly in level--even if it remains at fairly low levels overall--is far more noticeable than constant noise of an otherwise similar character and amount. This is why we record with linear PCM rather than logarithmic quantization, which would undoubtedly save bits. The problem with logarithmic quantization is that the noise floor rises and fall with the instantaneous levels of the recorded material, causing a kind of modulation noise. Modulation noise was one of the main shortcomings of analog tape, particularly at slower recording speeds (and to some old-timers I've known, anything slower than 30 ips was considered a slow recording speed).

--When I said that "if I'm starting from an 18- or 20- or 24-bit live recording, then the noise floor of the recording doesn't increase at all when I boost the gain by 3.5 dB" I think I was closer to the truth than you seem to think. I was referring to the noise floor of the resulting 16-bit CD, and I think you were referring to the noise floor of the 24-bit recording that's been boosted 3.5 dB--that difference in assumptions would explain your reply, and make us "both right."

But to be totally honest I must admit that there would be a very, very slight increase even in the noise floor of the 16-bit recording--a few thousandths of a dB. It wouldn't even be measurable, let alone audible, but it could be calculated if the hypothetical recording situation were to be specified in enough detail.

--best regards

P.S.: The only point of leaving some headroom in any recording is to avoid overload. It isn't desirable for its own sake. Headroom of (say) 3 - 4 dB or more that is never, ever used is simply wasted dynamic range. It only leads to increased noise, because listeners have to crank up the volume on their playback systems. Personally I'm happiest if my recordings peak at around -1 or -2 dB.

[ScotK]

Of course the 1-bit folk might have a slightly different answer to this question ...
Thanks for the nice discussion!
scot

[DSatz]

ScotK, what do you mean?

A 1-bit A/D converter can be considered as a roundabout way to get to a multi-bit result. Maybe that's not the whole picture but for most practical purposes other than DSD recording (now a thoroughly dead technology), that's how it is used.

It's an interesting technology, very nice in several respects, while in others, not so much. But once you convert its results into the multi-bit realm (16, 24, whatever), the same limits and concepts apply as with conventional multi-bit A/D converters.

--best regards

[ScotK]

Agreed, DSatz.  Just thinking that in the DSD realm, it's low # of bits and high sampling freq - sort of the opposite
priority we're stating here for "traditional" recording methods.

And yes, once you convert the 1-bit stream to something more genreally listenable, you're right that the same arguments apply.

And out of curiosity, why do you call DSD a thoroughly dead technology? Just that it hasn't garnered much
industry acceptance?  Aren't SACD (not that they are particularly popular) 1-bit?

s

[Gutbucket]

Noise that rises and falls rapidly in level--even if it remains at fairly low levels overall--is far more noticeable than constant noise of an otherwise similar character and amount. This is why we record with linear PCM rather than logarithmic quantization, which would undoubtedly save bits. The problem with logarithmic quantization is that the noise floor rises and fall with the instantaneous levels of the recorded material, causing a kind of modulation noise. Modulation noise was one of the main shortcomings of analog tape, particularly at slower recording speeds (and to some old-timers I've known, anything slower than 30 ips was considered a slow recording speed).

Interesting nugget of knowledge there.
Thanks & +T.

[DSatz]

ScotK, now I see what you were referring to, and of course you're completely right. 1-bit recording is a little like real estate in Tokyo--there's no more room for lateral expansion, so the only direction left to build in is upward. (Or maybe downward, but the mole people already have that market cornered.)

If you constrain an audio data stream by definition to just a single bit of precision, then oversampling becomes necessary for anything more complex than Morse Code messages--thus the sampling frequencies in the (low) MHz range. How the sampling rate of a 1-bit system relates to the audio bandwidth and dynamic range of the system isn't quite as straightforward as it is with conventional multi-bit linear PCM.

--best regards

[ScotK]

DSatz,

I was re-reading some of the 1-bit stuff today and it struck that what you really want is not binary logic, but
tri-state logic, which people have worked on for some time: up, down, or same. I guess with such fast sampling
rates, it doesn't matter that same is represented as up down up down ..., but it's got to be better if you really
did have a single tri-bit instead, though!
s