16-bit or 24-bit?
- Thread starter westg8
- Start date
Dougie Johnson
New member
Recording in 32 bit is always better quality than 16 bit, but...
Most processors in computers handle 16 bit, some handle 24 bit, and ever fewer are powerful enough for 32 bit.
So depending on your processor, you probably can't run 32 bit, unless you have a Pentium III or an Athlon.
Try running at 32, then if your computer freezes up, demote it to 24, see what that does. If it still locks up, stay at 16.
DJ
Most processors in computers handle 16 bit, some handle 24 bit, and ever fewer are powerful enough for 32 bit.
So depending on your processor, you probably can't run 32 bit, unless you have a Pentium III or an Athlon.
Try running at 32, then if your computer freezes up, demote it to 24, see what that does. If it still locks up, stay at 16.
DJ
AlChuck
Well-known member
????
It's not the computer's CPU that's the problem here, Dougie. The A/D converters are the issue -- well, that, and also the sheer amount of data that has to be moved through the system bus between memory and the hard disk. There are many 16-, 20-, and, lately, 24-bit A/D converters on soundcards, but no 32-bit converters, at least not on the consumer market yet.
Confusion arises because once the audio data is sampled at whatever resolution, calculations that are performed by software processing are usually done using 32-bit floating point numbers. Recording software talks about 32-bit files and 32-bit floating point precision and stuff like that.
It's not the computer's CPU that's the problem here, Dougie. The A/D converters are the issue -- well, that, and also the sheer amount of data that has to be moved through the system bus between memory and the hard disk. There are many 16-, 20-, and, lately, 24-bit A/D converters on soundcards, but no 32-bit converters, at least not on the consumer market yet.
Confusion arises because once the audio data is sampled at whatever resolution, calculations that are performed by software processing are usually done using 32-bit floating point numbers. Recording software talks about 32-bit files and 32-bit floating point precision and stuff like that.
So I should record in 16-bit? I have a Pentium II, 400 MHz and under the performance tab under the 'system' icon in the control panel, it says file system: 32-bit and virtual memory 32-bit? Does this have anything to do with it? Also, i recorded a couple tracks about 3 min long each and, all the tracks still ran, although some would not play at times. Does this mean I should just go with 16-bit then? Thanks for all the comments thus far.
Dougie Johnson
New member
I see what you're sayin, AL Chuck.
And yes, westg8-I think recording at 16 bit is gonna be perfect.
DJ
And yes, westg8-I think recording at 16 bit is gonna be perfect.
DJ
skippy
New member
On the subject of A/D converters and bit depth: I just made something of a study of the current state of the art, and I don't think that 32-bit converters are likely to happen at all soon.
Further, I don't think that you'd buy much of anything by using them, even if they did exist: the current crop of 24-bit converters are already producing more *theoretical* dynamic range than we can use, and the lower 2-3 bits are really down in the thermal noise of the preamps and input electronics. Realistically, the *usable* resolution of the 24-bit converters in current use is about 20-22 bits. They put out a 24-bit sample, to be sure, but the lower couple of bits are really dominated by the noise of the electronics.
As an example: the 24-bit AK5393 converters that are all the rage now (and are used in my Fostex D1624) are spec'd by the manufacturer to support a maximum dynamic range of 117dB. These are regarded as very good converters, and they do sound great. But wait: 24 bits give you a theoretical dynamic range of 144.5 dB (20*log10(2E24)). So where did the rest go? Component tolerances, noise, yadda yadda. 117dB tells me that those converters actually do about 19.5 bits of real usable, linear conversion- and the rest (the 4 lsb's) are basically noise. And that's not bad, not bad at all- given that the early crop of 16-bit converters in the '80s could barely do 14 bits...
However, there is much to be said for doing *signal processing* that uses a full 32-bit signal path, in order to preserve the mathematical headroom to manipulate multiple 24-bit inputs. Take the case of a simple mixer: adding together 2 fullscale 24-bit signals requires 25 bits to have headroom to handle the worst-case peak. Summing 16 fullscale 24-bit signals requires 28 bits. If you add in effects on each channel before the summing operation, such as an EQ with say a +6dB boost, it gets worse. So having a 32-bit (or deeper!) data path for the numbercrunching makes a great deal of sense- even if your converters only convert 16, or 20, or 24. The software should take this higher resolution and dither it down to your desired output resolution only at the last possible minute, to preserve all the original resolution of the samples...
So, if I read this correctly, recording in "32-bit" mode refers to the resolution of the downstream processing, and that should be a win. Pro systems are doing obscene thnigs like 48- and 64-bit internal resolutions, these days. Anyway, that's my reading of it. Anybody know a definitive answer?
Further, I don't think that you'd buy much of anything by using them, even if they did exist: the current crop of 24-bit converters are already producing more *theoretical* dynamic range than we can use, and the lower 2-3 bits are really down in the thermal noise of the preamps and input electronics. Realistically, the *usable* resolution of the 24-bit converters in current use is about 20-22 bits. They put out a 24-bit sample, to be sure, but the lower couple of bits are really dominated by the noise of the electronics.
As an example: the 24-bit AK5393 converters that are all the rage now (and are used in my Fostex D1624) are spec'd by the manufacturer to support a maximum dynamic range of 117dB. These are regarded as very good converters, and they do sound great. But wait: 24 bits give you a theoretical dynamic range of 144.5 dB (20*log10(2E24)). So where did the rest go? Component tolerances, noise, yadda yadda. 117dB tells me that those converters actually do about 19.5 bits of real usable, linear conversion- and the rest (the 4 lsb's) are basically noise. And that's not bad, not bad at all- given that the early crop of 16-bit converters in the '80s could barely do 14 bits...
However, there is much to be said for doing *signal processing* that uses a full 32-bit signal path, in order to preserve the mathematical headroom to manipulate multiple 24-bit inputs. Take the case of a simple mixer: adding together 2 fullscale 24-bit signals requires 25 bits to have headroom to handle the worst-case peak. Summing 16 fullscale 24-bit signals requires 28 bits. If you add in effects on each channel before the summing operation, such as an EQ with say a +6dB boost, it gets worse. So having a 32-bit (or deeper!) data path for the numbercrunching makes a great deal of sense- even if your converters only convert 16, or 20, or 24. The software should take this higher resolution and dither it down to your desired output resolution only at the last possible minute, to preserve all the original resolution of the samples...
So, if I read this correctly, recording in "32-bit" mode refers to the resolution of the downstream processing, and that should be a win. Pro systems are doing obscene thnigs like 48- and 64-bit internal resolutions, these days. Anyway, that's my reading of it. Anybody know a definitive answer?
AlChuck
Well-known member
westg8,
The number of bits used in the A/D converters for each sample is completely different than the virtual memory or disk access bit sizes.
This from somewhere on the 'net(specifically, http://www.oznet.ksu.edu/dp_iet/Annual_Conference_1996/32bit.htm), with some spelling corrections by me:
Probably not the best description but hopefully good enough.
skippy,
Great post about the A/D conversion process.
The number of bits used in the A/D converters for each sample is completely different than the virtual memory or disk access bit sizes.
This from somewhere on the 'net(specifically, http://www.oznet.ksu.edu/dp_iet/Annual_Conference_1996/32bit.htm), with some spelling corrections by me:
Probably not the best description but hopefully good enough.
skippy,
Great post about the A/D conversion process.
since i'm spankin' new to this stuff. So i should record in 16-bit then? Thanks!
AlChuck
Well-known member
There's not really a hard-and-fast answer here. All else being equal, record at the highest sample resolution that your audio card allows. If it's a Sound Blaster, that'll be 16 bits. If it's something else, it might be 16 or it might be 20 or 24 bits. For the latter, you can choose pretty much any of the 32 bit file formats supported, unless you are going to transfer the file somewhere that requires it to be in a specific format. And also remember that when you finally want to make an audio CD from a recording you'll have to save your final mixed stereo file to 16-bit, 44.1 kHz format.
But all else is not equal. If you can record 24-bit sample data, your file sizes will be significantly bigger, which is a factor. It also requires an overall faster system to be able to handle all the extra data streaming around. For a given PC, with 24-bit recordings, you'll have fewer tracks and/or plug-ins that you can use before overtaxing the system as compared to 16-bit samples. So if many many tracks are important to you, this might outweight the benefits of recording at the higher resolution.
But all else is not equal. If you can record 24-bit sample data, your file sizes will be significantly bigger, which is a factor. It also requires an overall faster system to be able to handle all the extra data streaming around. For a given PC, with 24-bit recordings, you'll have fewer tracks and/or plug-ins that you can use before overtaxing the system as compared to 16-bit samples. So if many many tracks are important to you, this might outweight the benefits of recording at the higher resolution.
Jerry Kahn
New member
Simple: If your setup allows it, 20 or 24 bit audio sounds shitloads better than 16-- shitloads.
jk
PS -- shitloads
jk
PS -- shitloads
John Sayers
Solar Power!!
Great post Skippy - but I agree with Jerry - if you can do 20/24 bit.... do it!! 
cheers
john
cheers
john
The Green Hornet
New member
YO Bit-people:
Bit by bit, I think I get the idea. Thanks for the information.
Green Hornet
Bit by bit, I think I get the idea. Thanks for the information.
Green Hornet
skippy
New member
In these days of oversampling converters, bit depth wins over sampling rate (IMNSHO, anyway). These days, the analog artifacting of old due to the antialiasing filters is pretty much a thing of the past.
Recording at 24 bits allows for more flexibility in setting levels, and some folks would describe this as "more headroom". Truth be told, in the digital world, nothing provides more headroom: you can *always* turn the knob up until it crunches! However, 24 bit encoding pushes the noise floor down somewhat, so you can record at more conservative levels and still get good signal-to-noise (in this case, signal to noise+distortion, because you're primarily concerned about the distortion characteristics of the A/D-D/A chain for low-level signals).
16 bits gives reasonable dynamic range, and preserves reasonable noise characteristics for low-level signals, as long as you use it all. 24 bit encoding provides a much more benign environment for low-level signals. Just listen to your reverb tails in the two environments, as the reverb decays: that's my classical test for it with good converters. Even I can hear a certain "graininess" in the 16-bit stuff that is less obvious in the 24 bit version, and I'm dadgum near deaf these days.
To get best linearity and best S/N at 16 bits, conventional wisdom holds that you must record right up at 0dB on the peaks, with no margin for error. With 24 bit, you get equivalent or better performance, even if you back off to -6dB or more for the peaks. It makes for a much more relaxed tracking environment, when you don't feel that you have to max the bejeezus out of everything- much more of an oldtimey analog feel, when the tape could "handle it"...
If that's more headroom, it's a psychological thing, *not* a circuit thing as it was for analog tape. I don't care: I'll take it, and I'd rather be more relaxed any day!
However, as in all things, there's a caveat: some "24-bit" converters out there are really only good for 15-16 bits of actual resolution and linearity. We can thank the markerters for that one... In which case, your "extra headroom" is illusory, because those low bits are just junk.... Check the specs: if the S/N and/or dynamic range of your 24-bit converters is 90dB or less (and there are some that are much worse, and are still marketed as 24-bit) you might as well just record at 16 bits: 24 ain't-a-gonna help, and you're going to have to fish your signal up out of the noise floor anyway, and work as hard as you would with simple honest 16-bit conversion to get equivalent results.
Confusing? You bet. However, consider this: there are some 20-bit converters that actually produce more usable linearity, which controls both S/N+D and dynamic range, than many of the low-cost 24-bit guys.
Here's an attempt to clear up the confusion I just created. My example of the AK converters above stands. They are "24-bit" converters, but produce essentially 20 bits of usable linear range. I'm happy: I get 4 bits more resolution than a "perfect" 16-bit converter, so I can back off 2 bits worth (6dB) and still have 2 bits better resolution and linearity than the ideal 16-bit example. And less ulcers worrying about crunches and overs! Make sense?
Somebody with cheapo pseudo-24-bit converters that really are only good for 16 bits, and who backs off 2 bits worth, will be recording only 14 bits above their noise floor. And that's a long way from optimal: that's a _lose_.
So what are your converters good for, in terms of _linear_ range? That's what makes the "extra headroom" magically appear. Those 8 extra bits don't help much unless what they contain has some relevance to the _signal_, no? Caveat emptor... Hope that helps.
Recording at 24 bits allows for more flexibility in setting levels, and some folks would describe this as "more headroom". Truth be told, in the digital world, nothing provides more headroom: you can *always* turn the knob up until it crunches! However, 24 bit encoding pushes the noise floor down somewhat, so you can record at more conservative levels and still get good signal-to-noise (in this case, signal to noise+distortion, because you're primarily concerned about the distortion characteristics of the A/D-D/A chain for low-level signals).
16 bits gives reasonable dynamic range, and preserves reasonable noise characteristics for low-level signals, as long as you use it all. 24 bit encoding provides a much more benign environment for low-level signals. Just listen to your reverb tails in the two environments, as the reverb decays: that's my classical test for it with good converters. Even I can hear a certain "graininess" in the 16-bit stuff that is less obvious in the 24 bit version, and I'm dadgum near deaf these days.
To get best linearity and best S/N at 16 bits, conventional wisdom holds that you must record right up at 0dB on the peaks, with no margin for error. With 24 bit, you get equivalent or better performance, even if you back off to -6dB or more for the peaks. It makes for a much more relaxed tracking environment, when you don't feel that you have to max the bejeezus out of everything- much more of an oldtimey analog feel, when the tape could "handle it"...
If that's more headroom, it's a psychological thing, *not* a circuit thing as it was for analog tape. I don't care: I'll take it, and I'd rather be more relaxed any day!
However, as in all things, there's a caveat: some "24-bit" converters out there are really only good for 15-16 bits of actual resolution and linearity. We can thank the markerters for that one... In which case, your "extra headroom" is illusory, because those low bits are just junk.... Check the specs: if the S/N and/or dynamic range of your 24-bit converters is 90dB or less (and there are some that are much worse, and are still marketed as 24-bit) you might as well just record at 16 bits: 24 ain't-a-gonna help, and you're going to have to fish your signal up out of the noise floor anyway, and work as hard as you would with simple honest 16-bit conversion to get equivalent results.
Confusing? You bet. However, consider this: there are some 20-bit converters that actually produce more usable linearity, which controls both S/N+D and dynamic range, than many of the low-cost 24-bit guys.
Here's an attempt to clear up the confusion I just created. My example of the AK converters above stands. They are "24-bit" converters, but produce essentially 20 bits of usable linear range. I'm happy: I get 4 bits more resolution than a "perfect" 16-bit converter, so I can back off 2 bits worth (6dB) and still have 2 bits better resolution and linearity than the ideal 16-bit example. And less ulcers worrying about crunches and overs! Make sense?
Somebody with cheapo pseudo-24-bit converters that really are only good for 16 bits, and who backs off 2 bits worth, will be recording only 14 bits above their noise floor. And that's a long way from optimal: that's a _lose_.
So what are your converters good for, in terms of _linear_ range? That's what makes the "extra headroom" magically appear. Those 8 extra bits don't help much unless what they contain has some relevance to the _signal_, no? Caveat emptor... Hope that helps.