Is 24 bit really as safe as you think? Maybe not...

Just because 24bit offers you more available range digitally, doesn't mean you should completely disregard recording levels. It does mean you have less to worry about relative to 16bit though and I think that general advice may be misinterpreted.

Philip


+1 ... Great thread ... Noise is oft problematic when compressing my vocs 30+ dcbs with M/U gain.  Noise so oft lurks behind that sweet vox.

Glad I'm not the only one suffering with noise that oft doesn't gate out well.

My current Pre-amp-mic noise solutions that don't always seem to work:

1) Dampen the mic on full
2) Record toward the middle of a 'lush' room
3) Prevent early reflections
4) Turn off things that hum (d&rn computer fan won't go off ... so it remains behind the cardiode mic)
5) Mic set on Cardiode
6) Mic HPF'd > 70-80 Hz
7) A strong gate in Voxformer
8) Bass traps and recording booths
9) Retakes (often a sweet cure)

more bits meant more samples per second

mattplaysguitar


Little additions:

Theoretical noise floor of 16bit is -96dB
Theoretical noise floor of 24bit is -144dB

From my tests, I found the noisefloor in my room (this was with mic in listening position, when doing proper recordings I use acoustic panels which considerably lower the noisefloor) was above -96dB. I don't have an actual number, but it sounded louder than the 16bit noisefloor. This digital noisefloor was still perceivable as the amp noise wasn't loud enough to mask it completely. In terms of actual noisefloor numbers, they would have been pretty much the same, only 24bit sounded better, if you know what I'm saying. If I had spent the time setting up the room I might have got my room noise down lower than the 16bit noisefloor with the right mic and amp gain.

mattplaysguitar


I should re-phrase, the potential dynamic range is 96dB. The theoretical perfectly clean recording at 16 bits has a -96dB noisefloor But in reality it will be a little higher with dithering noise and any other sound. We don't usually achieve this because of our environment and gear.

And on the room side of things, I'm comparing it to the particular recordings I made and where the room sound sits in SONAR. It has no relation to the SPL levels you are referring to. And do you mean dB(A) or dB(C) or just the unweighted SPL value?

mattplaysguitar


Everyone here has heard the saying "record in 24 bits and then you can put your preamp down and not have to worry about the noisefloor." I just conducted some tests that prove this is NOT necessarily true. So, the test:

- White noise (sample, not generated) played back through my monitors at the same volume
- Recorded with mic preamp set to 100%, 50% and 0%
- Recorded signals matched in volume

So what did I find? When listening to the white noise signals, comparing one another, they all sounded pretty darn similar. The main difference I heard was in the silent section at the end of the recording. I found the 100% one to be very quiet, the hum of my computer and the background noise were pretty much the same level. Compared the 50% one and the background noise/hiss was now probably twice as loud. Quite noticeable. Now comparing to 0% I found the background noise to be EXTREME. This is not an issue with 24 bit recording, this is an issue with my preamps and/or my A/D converters in my interface.

I also performed the same test recording in 16 bits. The changes in noise levels were pretty much the same. On my first listen, I thought them to be pretty much the same. Though I did notice a slight difference in the 0% preamp gain. After further listening and analysing I can now hear on all of them, a grainy, harsh, high pitched scracthing noise in the background - the 16bit noisefloor.

What does this mean for recording? That old saying holds some value, 24 bit is certainly better than 16 bit, but it is not the magic tool for noise-free problems. It's very important, but there are other huge things to consider - namely the quality of your gear (using you are recording in an acoustically noise-free environment). All gear has noise, but the amount of noise in your preamps at different levels may actually be a lot louder than you expect. They were for me.

This will change how I record. At the very least, I will pay more attention to getting that preamp volume as high as I safely can. I need to do more tests, but I may consider looking into compression during recording if the source is very dynamic. Lowering the preamp just a little bit does put noticeable amounts of noise in the recording. Whether it is low enough or not is another questions, but it is certainly there, and certainly significant in terms of the differences in pre-amp levels.


None of this is new, I knew it was there, but I just never realised it was as noticeable as this. Your results may differ with different gear. I know these are not perfect tests, but they do show that more attention needs to be payed to getting optimal signal to noise ratios. Any issues with my test that I did not realise?

I should also mention the dynamic range we are talking about here too. The gain in sonar I needed to match the 50% to the 100% was 24dB. This introduced about a 6dB increase in noise (visual inspection in Voxengo Span as well as using my ears) in the high frequency range. So if it were linear (which I'm sure it's not) you're looking at 1dB of noise being introduced for every 4dB lowering of your preamp. Significant, I think. More tests should be done to check this. For the 50% do the 0% comparison, 15.8dB of gain is needed to match the 50% version, and about 12dB of noise is introduced as a result. So that's nearly 1dB of noise for every 1dB lowering of your preamp (clearly it is not linear). From the limited samples taken, it looks like the lower you get the preamp, the more of a difference small gain changes make in the introduction of noise.


My gear:
Yamaha HS-80ms (played the white noise)
Edirol UA-25EX soundcard (the source of all the noise!)
sm58 (I couldn't be bothered getting my other mics out)

Processing:
Internal sonar gain on the 0% to match the level of the 100% sample was done partially with Sonitus EQ (24dB gain) and 15.8dB gain with the trim in sonar.

ohhey

drewfx1

ohhey



The main reason to use 24bit has nothing to do with the noise floor, it has no control over that.  The idea is that there are more loudness values available to make the wav smoother.   So it doesn't prevent noise it just keeps the waveform higher quality at all levels.  And yes, this is most important and noticeable at the lowest levels because 16bit conversion favors the loudest stuff and lets the quiet stuff suffer.  So you can't say 16bit is adding noise it's just damaging the lowest level audio.

This is just a different way of saying 24bit has less quantization noise.  Quantization noise is defined to be the difference between the actual (analog) value and the (quantized) digital at the point in time that the sample is taken. The waveorm is "smoother" only to the extent that the magnitude of the quantization error is less - i.e. less quantization noise.

The problem I have with the word "noise" is that it's easy to confuse with other sources of noise that are "in addition to" the signal. Something that is introduced by the recording device.  I think a better term to use with be "distortion" rather then noise.  If you say Quantization distortion that be better.