Dynamic Range and Bit Depth

Bob Katz Leave a Comment


MBlock: Bob,
Would you please help me to settle a debate I’ve been having with a friend of mine? I maintain that technically it is wrong to digitize audio in the -10 to -20 (peak) range. You’re just throwing away bits.”

Bob Katz: Hi, Mark. In general, from the point of view of the A/D converter, more is better, as long as you are not overloading any intermediate analog stage in line in order to do it. And as long as you have an accurate overload indicator. (See my article on levels: Part I)

OtherGuy: “Your interest in making a track as loud as possible would seem to sacrifice dynamic range.”

BK: Making an initial recording peak near 0 dBFS is not the same as making the final mixed production as loud as possible. Making the track peak near 0 actually maximizes the dynamic range of the recording with respect to the signal to (dither) ratio of the recorded multitrack medium. Afterward, in post production, you can alter (lower) the mixed or post produced level of that previously recorded track to any point you wish, so, once again, the dynamic range of the final production is also not sacrificed. Unless I totally misunderstood the poster’s motives, he is wrong.

MB: “No, just the opposite. The dynamic range is the difference (in dB) between the noise floor and the maximum signal a system can tolerate without audible distortion.
(The term “audible distortion” requires a definition in the analog world, but with a digital signal it’s clear cut.) Full 16 bit dynamic range requires peaks near 0.”

BK: Correct, as I just said. But “Requires” is a dangerous word. For example, if I were recording a live performance direct to 2-track with a world-class properly dithered 16-bit A/D converter, and one peak in the entire hour hit 0 dBFS, but the second movement was all at -20 dBFS (the original range of the players), it would probably just be fine. The sound would be world-class. As long as you did not have to raise the gain of the second movement in post production or do any post production at all. The point being that in direct to 2-track of live material, if the original dynamic range is good-sounding, then you have made a proper recording. But if you have to do any post production on it, potentially needing to raise or lower some tracks, then it pays to record the 2 track (or in this case, more likely, multitrack) 24 bit with good (“24 bit”) A/Ds, so that you have more room above the quantization noise floor when getting into post, where you might have to raise or lower gains. More on this in a moment.

OG: “It seems that your rationale forces you to constantly lower the music level in the mix anyway.”

BK: That’s fine, not a problem. By recording “hot”, you have made a good, clean original recording which you can then later lower in level inthe mix or post stage. That’s called “having your cake and eating it, too.” It’s a good idea: Maximizing the record level of A/Ds, especially the cheap, shoddy ones, is a very good thing. Another way of thinking of it is you are increasing the signal to garbage ratio of the original recording, where the garbage is at the bottom of the A/D converter. More on this, as we get into the “24 bit rules”….

MB: “My aim is simply to bring audio into the Avid at maximum resolution, and that means digitizing as hot as possible without distortion.”

BK: Correct. But as ADCs have gotten better it is no longer necessary to digitize to full scale. Regardless, as I said, you are preserving dynamic range if you are simply recording the unaltered microphone output to the ADC at near full scale.

OG: “You are equating bit depth with volume. Are you sure that is really true? By definition, this would imply that a 24-bit system can achieve louder levels and a wider dynamic range. most commercial music is highly compressed to start with and really doesn’t cover this full dynamic spread. As a result, if you have 20dB of actual (mixed) dynamic range with “loudness” that falls in the bottom, middle
or top of the available range, it really doesn’t end up sounding different – or so it would appear to me.”

BK: You are more and more correct for the more modern ADCs recording at 24 bits. Set your average level to -20 dBFS and let the peaks fall where they may. However, the older and more inferior ADCs may benefit from “maximizing”.
MB:My understanding is that digital audio sounds best as you approach maximum level; it sounds pretty nasty down in the bottom of the range.
BK: Shall we clarify that to say: An Analog to digital conversion sounds best as you approach max. level….. (within reason).  On the D/A side and mixing side, it’s a bit more complex, and it is possible to remain clean without needing to hit full scale. Otherwise you could never mix productions with reasonably wide dynamic range that are designed to be reproduced well (gotta sweat the soft stuff….).

MB: “I try to record with pre-mixing in mind. I have the playback levels all set at zero on the console (10dB down from the top of the fader.) I set my record levels so that I have a pretty good mix coming back with all faders in this ruler-straight line. If a guitar peaking at full scale digital is too loud, I can either pull down the record level or the monitor level.

BK: As long as you are talking about the A/D conversion (tracking) situation in 24-bit then I have to agree, you do not have to “maximize” your record side as long as your average forte levels measured with an averaging meter are working at or higher than about -20 dBFS.

OG: “Now… There is no difference between recording the guitar at -12 or pulling it down 12dB in the final mix.”

BK: There certainly is, from the point of view of the sound quality of the original recorded track, in a 16-bit recording, which I think your friend is referring to. As we moved to 24-bit, I would correct that to say, “there is no perceptual difference between….” meaning that there is enough signal to noise ratio in good 24-bit ADCs to allow more latitude in record level.

OG: “And, you are not really losing any resolution. If you are recording at 24bits, even if you record at -40dB you still have 24bits of resolution in the data you are recording.

BK: Depends on what you do with that data and how you gain stage it in the final. If that -40 dB represents a pianissimo passage that is going to remain that way and not be turned up, then you haven’t lost any resolution. If instead that -40 dB represents the peak level you are putting on the 24 bit recorder, then you have lost almost 8 bits of resolution! So I agree with him if his forte passages are at -20 dBFS approximately, average-reading meter.

OG: In my definition of “resolution”, resolution and noise are directly related. The closer the recording is to the quantization distortion noise) level (the lower the recorded level), the lower the resolution of the recording, and the fewer effective bits it has. If I record to a 24 bit recorder peaking at 48 dB below peak level, I have essentially made a 16 bit recording, from the point of view of resolution, and signal to garbage ratio. Resolution has nothing to do with record level. Resolution and noise floor both change values when talking 16bit vs 24bit, but they don’t really have much to do with each other.

BK: My definition of resolution has EVERYTHING to do with record level. You can have a high resolution tape recorder but if the highest peak of the entire program is -24 dBFS, then you have probably lost 4 bits of resolution which you will never get back. If you have to raise it later in mastering it may not sound as good as if you peaked your max. peak to full scale. The difference will be subtle with modern ADCs, perhaps inaudible, but theoretically I would try to avoid making a recording whose highest peak doesn’t reach at least, say, -10 dBFS peak. This is just a conservative application of the principles…. not too low, not too high  :-).

OG: “When you record at lower levels, you still have the same resolution, the only thing that changes is the noise floor. With 16bits the noise floor is always 96dB down from full digital level. If you record at -10 your noise floor is 86dB below the signal. Still very good. Since audio reference levels are at -18 or whatever it is for digital audio on video for the final mix, then it doesn’t matter if you record at the lower level and mix with the master fader up, or record at the higher level on each track and pull the master fader down. The final noise floor is the same. Quantizing error is a function of the converters used, and does not figure in to the recording levels or resolution. So, it doesn’t matter, and I would just get everyone to agree on one way to do it so that everything is easier and you don’t have to switch back & forth between methods.”

BK: I disagree with his stating that resolution is the same when the signal is closer to the noise. The closer the signal to the noise, the less resolution that low level signal will have, at least in terms of signal to noise ratio. Yes, the noise floor is the same IF in the end case you do not raise the gain. If the original soft signal remains reproduced soft in the final, then you have not worsened the perceived noise of the final product.
Now, if the original A/D and recorder was 24 bit, there is considerable leeway in how far you pushed the original track, absolutely, but it is still true if the original track was recorded rather low, and for esthetic reasons you must raise the level of the mix fader—-then the level of the original quantization noise and distortion of of the original A/D conversion gets raised above the  noise floor in the final mixdown. This is true whether you have a floating point or fixed point mixer.

Hope this helps,

Leave a Reply

Your email address will not be published. Required fields are marked *