Which bit depth conversion methodology going from mix to master to CD ?

Guys, is there a particualr advisable methodology WRT bit depth conversion when going from mix to master to CD?
 
I'm running projects at 44.1/24 bit.  Should I be mixing out to 24 bit or instead to 32 bit (which seems to be the default selection every time I go to export)?  And then master to what...32 or 24 bit before final step of 16-bit conversion for CD ?   Or should I just master to 16 bit directly for CD?
 
Lots of possible options, just wondering if there's some advisable bit depth conversion methodology I should adhere to for best sound quality....
 
Sonic

44.1/24 bits for recording and mixing.  The 32 bit export ain't for 32 bit file but is an upsampling process internal fx - if you are using the 64 bit engine just leave that it as is.  You want to "work" on the highest resolution file (and keep that file if you need to go back to make more changes).  I put my 24 bit Mix Master back into the SONAR project. 
 
Then I pull the 44.1/24 bit MIX Master file into Sound Forge and do whatever work needs be done (top and tail, limiting, etc.), then export it as final MASTERED 16 bit file.  Then back to the 24 bit to export it again as an MP3.

I work 32 bit floating internally, I then export 24bit 44.1khz then use CD Architect to Dither to 16bit 44.1kHz when cutting.

Thanks for the replies guys.
 
So...sounds like the 32-bit default setting when exporting is attributed to the fact I'm running with the 64-bit mix engine enabled?   I usually select 24 bit when exporting.  Or should I keep it 32 bit and wait until I master to drop it to 24 bit?
 

Wookiee
I work 32 bit floating internally, I then export 24bit 44.1khz then use CD Architect to Dither to 16bit 44.1kHz when cutting. 

 
This sounds like what I'm doing....most likely a result after I researched this extensively many years ago, but I can't recall.   Here's my process:
I have the 64-bit mix engine enabled, mix to 24-bit (no dither).  Then master with dither to 24 bit.  Then if I need to go to CD I use the highest quality conversion algorithm in R8Brain to convert to 16 bit.   And if I need to go MP3 I use the 24 bit mastered file.   

Does this methodology seem correct?  Please let me know if I have any flaws in there....

Sonic

for me, only 64bit floating point could defeat the 48bit (56bit actually) of Avid hardware mixing, with more than double bits of 24bit, extra dynamic headroom avoiding internal clippings, and can still keep full 24bit detail even through completely muted at the first stage but boosted back at the next stage, no any sounds lost. And also less rounding errors (like when 2 is going to be divided by 3, pi, or square root of 2). And finally, just kept all details inside 64bit, dithering from 64bit to 16bit so you can hear the incredibly unbelievable dynamic range and sensible SNR........

* Recording - whatever the audio interface produce, normally 24bit.
    - there is no interfaces which have meaningful 24bit resolution, but most of them are better then 16bit. The hint is in the specification: SnR or alike which is 96dB for 16bit (you can easily find that most interfaces have  over 100dB, but no of them have 144dB required for 24bit)
    - and so recording into 32 means saving at least 12 useless bits (into 64 format, ~44 bits pro samples will be useless)
* Processing - 64bit. Even simple mixing 2 channels "damage" the last significant bit, complex algorithms  use 1000s of mathematical computations. While that does not mean 10 operations always destroy last 10 significant bits, in practice that is much less, the result depends from the algorithm (usually moderate to good quality) and the programmer (experience of which in the music world is unpredictable). So the difference between 32bit and 64bit processing can shift into audible range (~16 bit). Note that up-sampling (f.e. to 96kHz) in some situations sounds "better" (you can find some examples from Craig) for the same reason.
* Intermediate saving (rendering/mix export/etc) - 32bit. That is "floating point" format with 24bit precision bits for any number. Even in worse case scenario, it will take ~8 intermediate savings till the difference to 64bit become audible (so you process -> export -> import -> process -> export -> import ... etc). I have not seen examples in the Internet that such degradation ever happens in practice. Note that 24bit format is fixed point, that means low amplitude samples have low precision (down to 1bit!) and only full amplitude samples have 24bit precision. The degradation is 1bit loose pro ~6dB. F.e. if the upper level of samples in your rendered file is -18dB, you use at most 21bit when exporting into 24bit format.
* Master output - 24bit or 32bit. After mastering the level should be normalized. Future processing is not going to increase the level of low amplitude samples, it will only reduce the precision when moving toward 16bit CD format. So 24bit fixed point is more then sufficient.
 
Finally. Dithering should be applied only once, especially "good" algos (noise shaping). By definition of that process, dithering adds noise to the signal (that is a trade, not a magic, you get more noise for less quantization distortion, a good trade for listening but questionable for future processing).  

Use an 8 track, one-inch tape machine instead of 16 heads.  And mix down to a half-track, 1/4-inch tape, not a 1/4- inch, quarter track.  More sound per square inch of tape.  Almost like higher bit depths and sample rate.  Always keep your resolution at the max for each step.

SonicExplorer
Interesting discussion.  So, wouldn't it be easiet to simply master inside a project then?  Therefore you can simply export to 24 (or to 16 with dither) just ONE time, and all the internal processing remains at whatever the global default render depth is set to.  This eliminates the need to worry about the various settings along the way. And if you want to toggle between mix and mastering work simply disable the FX bin on the mastering bus.  Assuming you have the CPU power it sure seems like this approach is a lot easier than trying to break things into different steps (mix, master, dither/CD) using separate templates and/or different software hosts.
 
Sonic
 
 

I do exactly this

azslow3
* Recording - whatever the audio interface produce, normally 24bit.
    - there is no interfaces which have meaningful 24bit resolution, but most of them are better then 16bit. The hint is in the specification: SnR or alike which is 96dB for 16bit (you can easily find that most interfaces have  over 100dB, but no of them have 144dB required for 24bit)
    - and so recording into 32 means saving at least 12 useless bits (into 64 format, ~44 bits pro samples will be useless)
* Processing - 64bit. Even simple mixing 2 channels "damage" the last significant bit, complex algorithms  use 1000s of mathematical computations. While that does not mean 10 operations always destroy last 10 significant bits, in practice that is much less, the result depends from the algorithm (usually moderate to good quality) and the programmer (experience of which in the music world is unpredictable). So the difference between 32bit and 64bit processing can shift into audible range (~16 bit). Note that up-sampling (f.e. to 96kHz) in some situations sounds "better" (you can find some examples from Craig) for the same reason.
* Intermediate saving (rendering/mix export/etc) - 32bit. That is "floating point" format with 24bit precision bits for any number. Even in worse case scenario, it will take ~8 intermediate savings till the difference to 64bit become audible (so you process -> export -> import -> process -> export -> import ... etc). I have not seen examples in the Internet that such degradation ever happens in practice. Note that 24bit format is fixed point, that means low amplitude samples have low precision (down to 1bit!) and only full amplitude samples have 24bit precision. The degradation is 1bit loose pro ~6dB. F.e. if the upper level of samples in your rendered file is -18dB, you use at most 21bit when exporting into 24bit format.
* Master output - 24bit or 32bit. After mastering the level should be normalized. Future processing is not going to increase the level of low amplitude samples, it will only reduce the precision when moving toward 16bit CD format. So 24bit fixed point is more then sufficient.
 
Finally. Dithering should be applied only once, especially "good" algos (noise shaping). By definition of that process, dithering adds noise to the signal (that is a trade, not a magic, you get more noise for less quantization distortion, a good trade for listening but questionable for future processing).  

You obviously didn't read my post, because it's about rounding errors and internal dynamic range. Those bits aren't wasted in the mixing and mastering process. And who listens to 16 bit CD anymore anyway? I don't!. And you don't understand 64 bit fp masters for digital distrubutiin. I upload 64bitfp audio files to sound cloud at whatever sample rate I recorded too. So rather than a crappy 16 bit file that gets squashed....a 64bitfp HD audio gets squashed. No one gets digital conversion. If I could be bothered I'd write a paper on this...but as the old saying goes 'you can lead a horse to water, but you can't teach it to drink 64bit fp master files solves the HD audio problem. Even if you still only deliver a final 24bit master for actual listening. Which is the lowest format I use and listen too. Crapify has let the team down...instead of really sonically degraded audio files people listen to...with today's tech there should have been a 24 bit boom. Where all 24 bit commercial masters are delivered from 64bitfp masters. Again 64 bit fp solves the HD digital audio problem. It's more complex than than that...the secret sauce so to speak...but I can safely assure you that I could prove without a doubt that you the listener would never need a crap l.p. player or any other so called hi-fi equipment or even any over-priced analouge paper weight again - so long as people stuck with 64bit fp. I have the academic qualifications to say this!

Ben

Here's something I'm curious to do a quick poll on: how long do people generally take to master a track? Doesn't matter for now how you define "master", am just curious. Am expecting wildly varying answers, and there's no "right" answer to it.

I don't think there's anything wrong with mixing and mastering inside the same project as long as you use a separate mastering bus and are disciplined enough to disable the FX bin until you are ready to master. It is very helpful to be able to quickly & easily hear how a mastered final result might sound before finalizing a mix.  In fact, I think this ability to quickly toggle back and forth is incredibly helpful in learning how the mastering process impacts the final result.  You can actually learn how tweaks in the mix translate during mastering.
 
Anyway.... as to the subject of CD's:  Yes, CD's are mandatory if you are a serious pro artist.  Most labels won't even consider any arrangements without physical product to distribute, and in many parts of the world there is still a good demand for physical product.   Personally I wish everything would go electronic, makes it easier on the artist and engineer/producer, but we are likely a long way away from that.
 
Sonic
 
 

BenMMusTech
You obviously didn't read my post, because it's about rounding errors and internal dynamic range. Those bits aren't wasted in the mixing and mastering process. And who listens to 16 bit CD anymore anyway? I don't!.

I did... and that was the reason I have written my post, with some reference to what going on in the real world
I mean in the real world there is no 32bit DACs and no floating point DACs. So the upper format physically used by audio devices is 24bit fixed point (otherwise something should convert on the fly) and even that is more then top converters can really reproduce. But I have mentioned that for mixing and mastering 64bit is a good idea.
 
Not to mention there is no known prove there is a single person in the world who can distinguish between original 24/32/64 and (properly prepared) 16bit signal, on any equipment. But well, at least technically the signal can be different (between 16 and 24, not between 24 and 32/64).
 
I had a friend with SACDs and all related hardware ("special" cables, etc.). I know that is a religion and as such no arguments can help to claim anything else
 
Here, in Germany, if I want to buy music, I order CDs. For some reason digital delivery audio (and books) cost more or equal to the physical delivery. And I listen them on crappy equipment, mostly in my car... But I have to deal with DACs/ADCs and the precision of calculations in other domain.

Some of this discussion is over my head but FWIW I will share that AFAIK Sonar 6 began supporting a render depth of 64 bit fp/export (previously in S5 was 32 bit).  Even though both S5 and S6 had 64-bit mix engines.   And although the final 24 bit mixes were identical when bit-compared, I felt I could definitely hear a difference on playback inside of Sonar when comparing S5 and S6.  Nobody ever did have an explanation for that though.  S6 sounded ..... more...analog or smooth than S5.  My ears are really sensitive to upper frequencies and harmonics so I'm guessing it was something in that realm.  But then again, maybe it was all in my head....never made sense to me how two different versions of Sonar could seem to sound different inside the project yet the mixes summed identically.   ??