Hi all, As some one who is fairly new to computer based recording in general, but slowly getting my head round it all I wonder if someone could take the time out to explain to me what the actual (audible) difference between these three settings are. As far as I have gathered the higher the bit, the wider the frequency range? Is that correct. Also, what does "float" mean? (as in 32 bit "float"). I hear that recording in "24 bit" and dithering down to "16 bit" for cd is higher quality than just doing the whole thing in 16 bit. What about doing it in "32 bit float" and dithering from there. (these seem to be the three settings available to me in Logic Plat 6). And what does that really mean "higher quality" - can that be explained in a tangible way? How do these different settings then affect your CPU and I/O loads? Thanks for your time, John James www.johnjamesmusic.co.uk

On a fine day, 02-02-2004, John James wrote: >As some one who is fairly new to computer based recording in >general, but slowly getting my head round it all I wonder if someone >could take the time out to explain to me what the actual (audible) >difference between these three settings are. As far as I have >gathered the higher the bit, the wider the frequency range? Is that >correct. Also, what does "float" mean? (as in 32 bit "float"). The bit depth describes the length of the (binary) number that's used to encode signal levels. An 8-bit sample uses only 2^8 (2 to the power 8) = 256 different levels. So if you record e.g. a sine wave at 8-bits and then look at it, you'll see a 'stepped' waveform consisting of only 256 different levels. Small variations in signal level will be lost, because they simply can't be distinguished with only 256 levels. With 16 bits you already have 2^16 = 65536 levels available. Needless to say that your 'stepped sine' will look (and sound) much smoother, and more "real sine-like" than the 8-bit version. 24 bits give 2^24 = 16,777,216 different levels and is thus again a massive improvement: even very small fluctuations in signal level can be faithfully recorded. The maximum attainable frequency is independent of this, and is determined by the sampling frequency (i.e. how many times per second is a sample recorded). As a rule of thumb: the max. reproducible frequency is half of the sampling frequency. So a CD-quality 44.1 kHz recording can contain frequencies up to 22.05 kHz, and a 96 kHz recording can go up to 48 kHz. Bit depth and sampling frequency are independent: in principle you can sample at 8 bits, 96 kHz (high top-frequency but low resolution), or 24 bits, 22 kHz (low top-frequency, but high resolution), or any other mix. So: bit depth determines the vertical resolution (level) and sampling frequency determines the horizontal resolution (time). Increasing bit depth has a greater effect on audible fidelity than increasing sampling frequency has. I.e. the difference between 16 and 24 bit is very well worth it. Increasing the sampling frequency from 44.1 to e.g. 48 kHz is not really worth it (esp. since you then have to "downsample" again to 44.1 at the end, because that's what CDs use, and downsampling is tricky and can produce artifacts that negate the 'gain' you hoped to get from using a higher frequency). >I hear that recording in "24 bit" and dithering down to "16 bit" for >cd is higher quality than just doing the whole thing in 16 bit. Correct. That's because all the effects-processing, summing of audio signals, etc, etc (in short: all the stuff that goes on in a digital audio system) necessarily introduces rounding errors and such. If these occur in a 16-million level system (24 bit) they'll be far less audible than when they occur in a 64-thousand level system (16 bit). The concept is rather straightforward: if you're off by one on a 256-point scale (8 bit) the error is 1/256 of the max signal level. If you make the same off-by-one mistake in a 24 bit system, the error is only 1/16-million-th of the full scale. The tricky part of course is how to get from your nice clean 24-bit recording to the standard 16 bits used by CDs. There's a whole bunch of mathematics involved in doing this "nicely", which is all implemented in Logic's 'dithering' algorithms (press the Bounce button on the Output object, and in the bounce dialog you'll notice a popup with 3 dithering options: that's the stuff that can turn 24 bits into 16). So you needn't worry about that. Just record at 24 bits (if your hardware supports that!), and when it comes to bouncing the final mix for CD, pick one of the dithering options. Warning: never dither more than once, and only do it at the very last stage of production. E.g. if your recording is going to a mastering engineer, just bounce at 24 bit and let him/her do the dithering. >What about doing it in "32 bit float" and dithering from there. >(these seem to be the three settings available to me in Logic Plat >6). The '32 bit float' is Logic's internal processing resolution. You can't record at 32 bit float. Logic simply uses an even higher resolution internally to calculate effects, audio-summing, and such. Completely transparent and not something to worry about. The positive spin-off of Logic's approach is that you have plenty of headroom in the audio mixer and will find it very hard to cause real digital clipping. The only place where this plays a role for the user is when setting up the "Freeze" options. Normally Logic will use 32-bit float for its freeze files. If you want to use the freeze files in other programs, or if you want to manually place (a copy of) a freeze file on an Arrange track, you need to tell Logic to create 24 bit (or 16 bit) freeze files, since 32-bit-float isn't a format any software can use. If you don't do any of this, but use 'freeze' the plain regular way, just leave this set to '32 bit float'. Since the freeze files then already are in the format Logic uses internally, you'll save some CPU cycles (a 16/24 bit file has to be converted to 32-bit float by Logic internally, before Logic can start doing calculations). >How do these different settings then affect your CPU and I/O loads? The higher the bit-depth, the bigger the file. A 24-bit file is 1.5 times a 16-bit file. So that means more disk activity (and more disk space). About cpu use: I think this shouldn't be affected -- both 16 and 24 bit are converted to 32 bit float anyway, so processing is always done in the same resolution, regardless of the input format. Finally, about the "float" in "32 bit float". "Normal" 16 and 24 bit files use whole numbers (integers) -- i.e. numbers like 256 and 345678 without a decimal dot in them. A "float" is a computer number with a so-called "floating point" -- like 56.8765. Generally speaking, there's plenty of reasons for preferring one over the other in various situations, and here in the LUG there have been heated debates over whether using floats is better than using integers for audio or not -- but that's the kind of stuff you probably don't want to get into at all (and neither do I :). To sum it up: record at the highest bit-depth your hardware supports (i.e. preferably 24 bit), use a standard 44.1 kHz sampling frequency, and dither down to 16 bits at the very last step. -- Hendrik Jan Veenstra h @ k n o w a r e . n l Omega Art: http://www.omega-art.com/

Hendrik Just had to say that your précis of bit-depth, sample-rate and dither principles was one of the most concise, helpful and clear (not to mention accurate - amazing how many people will happily spout nonsense on these subjects!) accounts I have ever read. Thank God there are some people around who are still prepared to take the time to explain these things, and do it in a clear and meaningful fashion using everyday terminology. Thank you!! I do have one question for you though, which is always an issue for me with DAW's. If you're bussing out to a mixer instead of mixing in the box (in my case 48 channels of 24bit audio out to external boxes and a Mackie D8B), does Logic automatically dither down from it's internal mixer's bit depth (32 bit float in this case) to 24 bit fixed, or does one manually need to add dither to all those output busses? I'm always concerned that hosts do not dither by default, meaning that these signals are being truncated. As far as I understand, whenever bit depth changes, dither should be used. Indeed I've had this confirmed by a couple of very knowledgeable sources, even when converting from 32bit float to 24bit fixed. When I used Cubase SX and Nuendo, I had to use busses routed to individual outputs with Waves L2 limiters across every bus dithering down to 24 bits, to ensure that all my bus outs were being looked after properly. This is because (as I understand it) the bus outs in Nuendo do not automatically apply dithering when dropping from 32bit float to the output bit depth. Should I be doing the same with Logic, or does Logic use a fully dithered mixer, including all the individual outs? Thanks for any feedback. Jules

On a fine day, 02-02-2004, Jules Bromley wrote: >Just had to say that your précis of bit-depth, sample-rate and dither >principles was one of the most concise, helpful and clear (not to mention >accurate - amazing how many people will happily spout nonsense on these >subjects!) accounts I have ever read. <blush> Thank you for the compliments -- I'm glad you liked it (and agreed with the technical stuff) >I do have one question for you though, which is always an issue for me with >DAW's. If you're bussing out to a mixer instead of mixing in the box (in my >case 48 channels of 24bit audio out to external boxes and a Mackie D8B), >does Logic automatically dither down from it's internal mixer's bit depth >(32 bit float in this case) to 24 bit fixed, or does one manually need to >add dither to all those output busses? Woh... I don't think my 'expertise' stretches that far. My guess would be that it's "obvious" that Logic dithers to 24 bit, but that's just based on... well, nothing much really :). My intuition would say something like: if you use an I/O plug-in, the external gear surely gets the audio delivered in a useable format (I hope), which means 24 or 16 bit. Likewise if you assign a bus or output object to a physical output, Logic surely delivers the audio to the output in a usable format as well, shouldn't it? I mean, is there any gear out there that can actually read/use 32-bit float? I wouldn't know, but have always thought the answer to that is: no. And if that's correct and 32-bit float is typical for Logic but for nothing much else, then it would be bizarre if the output carried a 32-bit float signal... But then you say: >When I used Cubase SX and Nuendo, I had to use busses routed >to individual outputs with Waves L2 limiters across every bus dithering down >to 24 bits, to ensure that all my bus outs were being looked after properly. >This is because (as I understand it) the bus outs in Nuendo do not >automatically apply dithering when dropping from 32bit float to the output >bit depth. Now this has me all confused, meaning the "obviousness" of the foregoing maybe (or: apparently?) isn't that obvious at all. But... since 32 *float* is so fundamentally different from 24 bit *integer*, would external 24-bit gear be able to do anything sensible with 32 float _at all_? I mean, I can imagine how a 16-bit machine simply truncates incoming 24-bit data -- that's easy enough. But I can't really imagine what a 16- or 24-bit machine is supposed to do with 32-bit float data -- except dither it (which would make the entire matter a moot point -- who cares who does the dithering, the external box or Logic?), or reject it, or turn it into horrible noise or something like that. And: if you would have to dither to 24 bit, then how would you go about doing that? Does Logic offer a dither plug-in? I honestly wouldn't know, as I've never needed one... Having to use a 3rd party plug-in (like the Waves L2) would be extremely silly. So the best idea I can offer (which isn't very good, I know) is: just try it. If the result sounds 'regular' then most likely some form of dithering takes place somewhere -- either in Logic or in the external machine. If there's no sound at all, or it all sounds completely wrong, then you're potential problem apparently is an actual problem. I know this isn't much of an answer, but I really wouldn't know the _real_ answer. I'm just a simple mathematician who happens to understand the theoretical concepts, but who's never seen a spdif-cable in his life :-). >As far as I understand, whenever bit depth changes, dither should be used. Correct. The alternative is plain truncation of the data (i.e. throwing away the lower bits) which is undesirable. >Should I be doing the same with Logic, or does Logic use a fully >dithered mixer, including all the individual outs? I hope someone else can offer a definitive answer to this. I would be curious as well (if only for theoretical reasons :). -- Hendrik Jan Veenstra h @ k n o w a r e . n l Omega Art: http://www.omega-art.com/

>Hi all, > >As some one who is fairly new to computer based recording in general, but slowly getting my head round it all I wonder if someone could take the time out to explain to me what the actual (audible) difference between these three settings are. As far as I have gathered the higher the bit, the wider the frequency range? > Most analog audio signals are continuous signals. When an audio signal is digitized, it is converted to a discrete approximation of the original continuous signal. Two approximations are made. The first is "Sampling Rate". This refers to the number times (per second) that a "snapshot" is taken of the audio signal. (kind of like a series of still frames approximating continuous movement in a motion picture. The more frames per second, the better the appearance of continuous motion.) With audio signals, greater sampling frequency leads to a less "dull" sound. Generally the higher the sampling rate the better (but there are limits to the audible benefits of higher rates...so, 44.1 KHz is more than sufficient). The second approximation is "Bit Depth". This is the resolution of each "snapshot." This is analagous to the sharpness of the still frame photo in the example above (or analagous to the number of pixels in a digital camera. The more pixels, the better the image). The greater the bit depth, the more accurately each snapshot (sample) represents the original audio signal. I would recommend 24-bit depth recording if possible. But you are ok if you record at at least 16-bit. I record at 20-bit (my digital mixer limits me at 20-bit) and I have no problem with my recordings (except for the massive human error I introduce into the process). Regarding CPU load, sampling rate and bit depth reflect the amount of information that is obtained from the original audio signal. The greater the sampling rate and bit depth, the more information there is about the original audio signal and the more room it takes to store and process the information. Again, 44.1 KHz and 16-bit to 24-bit are standard for today's computers. I would go for 44.1 KHz and 24-bit if possible. Dave > >

Hi Hendrik > Woh... I don't think my 'expertise' stretches that far. My > guess would be that it's "obvious" that Logic dithers to 24 > bit, but that's just based on... well, nothing much really > :). My intuition would say something like: if you use an I/O > plug-in, the external gear surely gets the audio delivered in > a useable format (I hope), which means 24 or 16 bit. Well Logic certainly 'delivers' audio in a usable format (16 or 24bit fixed point), I have no problem with that. The issue I was trying to get to the bottom of is 'how' it gets to that final format. ie. if all it's internal processing is at 32bit float, how does it convert to a lower, usable, fixed bit depth? My concern is that the mixer simply truncates at the output stage, unless one specifically adds dither as the final process in the mixer. Common sense would seem to me to suggest that all outputs should be dithered by default because, as you rightly point out, no hardware (ok I'm generalizing) can work with 32bit float files, and therefore bit depth will always have to be reduced as audio leaves the host environment. My concern is that this does not necessarily take place. I think I'm right in saying that ProTools offers a choice of either a dithered or undithered mixer, which seems to make sense, and is certainly what the pro market it's aimed at would and should demand. IMHO Logic should also be offering this same functionality. Indeed it may already dither all signals by default, I just can't find out whether it does or not. If it doesn't, then I would choose to add dither manually to every output bus, but I don't want to do this if it's already being done discretely, for obvious reasons. So many of us are insistent upon using the best converters, the highest resolution plugins, the highest bit depth and sample rate etc. and yet this issue rarely seems to surface. I guess many people are now mixing entirely in-the-box, but a lot still use mixers and/or 'real' outboard of one kind or another, and I would have thought this was fairly fundamental. Oh well, maybe a passing developer might like to shine some light and put us out of our misery!! Jules

On Monday, February 2, 2004, at 07:14 PM, Hendrik Jan Veenstra wrote: >> does Logic automatically dither down from it's internal mixer's bit >> depth >> (32 bit float in this case) to 24 bit fixed very short answer ;-) there really is no need to dither to 24 fixed. dithering adds noise at level below lsb - with 24 bits this is so microscopic as to be unnecessary. steve parker.

Thanks for your input Steve > very short answer ;-) > there really is no need to dither to 24 fixed. > dithering adds noise at level below lsb - with 24 bits this > is so microscopic as to be unnecessary. Please don't think me pedantic or argumentative but .... lots of people say they can't hear the difference between 16 bits and 24bits, yet the industry is well and truly committed to 24 bit, not to mention 96K (which I confess I can't hear the benefits of blind). All of these microscopic differences eventually add up to an audible difference, and many of us rightly or wrongly set our standards as high as our resources (and capabilities!) allow. For me, dithering when changing bit depth is a basic and fundamental principal of digital audio and I think most engineers, especially in the mastering field, would agree. Can we at least agree, that technically, dithering is the correct thing to do (otherwise why do the more respected plugin developers insist on it, even at far greater bit-depths?)? If so, I think it's only reasonable to want to know whether Logic is dithering or not. Finally, I hope you don't mind me quoting Joe Bryan of Universal Audio (below), whose views on all things audio I have the greatest respect for. Sorry if this thread is getting slightly anal, but I think there's a basic question which we could do with an answer to. Jules --------- Converting from 32-bit float to 24-bit fixed still requires dither because signals below -6dB have more resolution in float than they do in fixed. It's true that 32-bit float only maintains 24-bits of resolution, but the extra 8-bits used for scaling *maintain* this resolution at all signal levels. Think of it this way: float *always* maintains 24-bit resolution, regardless of level, and the worst case error is 1-bit or less. On the other hand, 24-bit fixed *only* maintains 24-bit resolution when the signal is between -6dB and 0dBFS. When the signal goes below this range, the upper bits are essentially unused (they're duplicates of the sign bit), so the signal resolution degrades as the level goes down. Every 6dB drop looses a bit of resolution. To make matters worse, headroom bits must be maintained (typically 2-3 bits), so the effective signal resolution at nominal levels is only 21-22 bits. Therefore, since fixed point resolution is less at lower signal levels while float resolution is unchanged at lower levels, dither must be applied when converting from float to fixed. Incidentally, dither can increase the effective resolution of a signal beyond the theoretical 6.028dB x N-bits resolution. The information carrying capabilities of perceptual data representations (like audio or visual signals) include not just the instantaneous signal representation (each individual sample), but also the spectral energy profile. Dither rearranges the quantization error energy in a way that maximizes information transfer. The total error energy is actually higher, but it's reorganized in a way that minimizes the impact these errors have on perception. We reccommend always leaving your audio data in 32-bit float format until the final mastering stage. Once all signal processing has been performed, resample (if necessary), and apply dither as the last steps before burning to the target media. The UAD-1 performs its internal calculations at a higher bit-depth than IEEE float, and uses UltraDither to reduce the internal resolution before transferring the data back to the host. For plugs like the CS-1, Nigel, DreamVerb, etc. that consist of multiple processing elements chained internally, the UAD-1 leaves the data in its internal resolution between processing elements.

jules, i take your point 100% about 16 bit etc. however dither added below the lsb of 24 bit is *so* tiny that thousands of operations will not bring it into any kind of audibility. dither is a compromise between raising the noise floor and spreading the quantization distortion. the distortion at (the lsb of) 24 bits is inaudible so a final "truncate" from 32 bits is ok. steve parker.

On a fine day, 02-02-2004, Jules Bromley wrote: >Well Logic certainly 'delivers' audio in a usable format (16 or 24bit fixed >point), I have no problem with that. The issue I was trying to get to the >bottom of is 'how' it gets to that final format. ie. if all it's internal >processing is at 32bit float, how does it convert to a lower, usable, fixed >bit depth? My concern is that the mixer simply truncates at the output >stage, unless one specifically adds dither as the final process in the >mixer. Aha, now I understand what you're getting at. Makes more sense than my original interpretation :-). Sorry, had a long day yesterday when I wrote the reply. Yes, this seems like sensible question -- but see Steve Parker's reply, which in turn seems to be a sensible reply. It could indeed very well be that truncating 32 to 24 bits doesn't really matter, from a sonic perspective. -- Hendrik Jan Veenstra h @ k n o w a r e . n l Omega Art: http://www.omega-art.com/

Hi Steve > i take your point 100% about 16 bit etc. > however dither added below the lsb of 24 bit is *so* tiny > that thousands of operations will not bring it into any kind > of audibility. > dither is a compromise between raising the noise floor and > spreading the quantization distortion. > the distortion at (the lsb of) 24 bits is inaudible so a > final "truncate" from 32 bits is ok. > steve parker. We should probably leave the argument of whether dithering from 32bit float to 24bit fixed is advisable aside for now. Much as I totally respect your opinion, too many other knowledgeable sources are categorical that all bit depth reductions should be accompanied by dither. Do you happen to know Steve if Logic is or isn't applying dither within the mixer at the output stage? I'll certainly agree that this is even more critical for anyone bussing out 16 bit audio streams. Jules

Jules Bromley wrote: > Well Logic certainly 'delivers' audio in a usable format (16 or 24bit fixed > point), I have no problem with that. The issue I was trying to get to the > bottom of is 'how' it gets to that final format. ie. if all it's internal > processing is at 32bit float[..] 32bit float audio usually consists of several components in contrast to a simple fixed point 32bit audio sample. IEEE754 consists of a 23bit mantissa with another hidden bit, which represent the audio data itself. The rest is basically scaling information of this 24bit audio, you can easily alter the absolute dimension by setting the appropriate scaling bit. So, the conversion between 32bit float and 24bit fix point audio shouldn't be too complex at all, since it's basically a truncation of the rated audio material with dithering around the lsb. > [..] unless one specifically adds dither as the final process in the > mixer. Common sense would seem to me to suggest that all outputs should be > dithered by default because, as you rightly point out, no hardware (ok I'm > generalizing) can work with 32bit float files, and therefore bit depth will > always have to be reduced as audio leaves the host environment. My concern > is that this does not necessarily take place. As long as you work with decent audio equipment, you can relax and trust in the manufacturers to care about your music :) Please do not forget we are talking about the last 6dB of a 144dB capable signal. This is more something to concern when measuring, but it's by far not as critical as it was with the "16bit situation". > I think I'm right in saying that ProTools offers a choice of either a > dithered or undithered mixer, which seems to make sense, and is certainly > what the pro market it's aimed at would and should demand. IMHO Logic > should also be offering this same functionality. Another point, or maybe more something like a feature request. > [..]all signals by default, I just can't find out whether it does or not. Well that should be easy to figure out. Simply generate a 1kHz sine wave with a high resolution without any additional signal components. Then, send it to a bus, alter its volume by -40dB or something alike and rerecord it. Then have a look at its lsb structure. If you find a noise around the LSB, it was dithered (likely). If you are interested in the exact dithering algorithm, notch out the 1kHz with the hardest notch you can find (beware of filter resonances), and normalize the rest. If you now do a fft over the noise, you'll get an idea of what was added to the original 1kHz after sending it through the bus and compare it with some curves of known dithering shapes. I assume it will be triangular noise, but actually I'm too lazy to perform the test myself ;-) > If it doesn't, then I would choose to add dither manually to every output > bus, but I don't want to do this if it's already being done discretely, for > obvious reasons. Well, that depends on what you do with the busses. If they are send through analog equipment like a console, it is very likely the noise floor of the outboard chain will add some "natural" dither to the signal anyway. So, no need for additional dither anymore. But if the output of the bus is going to be a final product in the digital domain, it will need some dithering, well but as I already said, I believe Logic does that anyway. > Oh well, maybe a passing developer might like to shine some light and put us > out of our misery!! Yeah - there is definately still too much superstition and hell of a bunch of rumors about digital audio :) I guess the answer is pretty simple, like "the routine converting 32bit float to 24bit fix point handles dithering anyway", but one never knows. Regards, Daniel

> >Well Logic certainly 'delivers' audio in a usable format (16 or 24bit fixed > >point), I have no problem with that. The issue I was trying to get to the > >bottom of is 'how' it gets to that final format. ie. if all it's internal > >processing is at 32bit float, how does it convert to a lower, usable, fixed > >bit depth? My concern is that the mixer simply truncates at the output > >stage, unless one specifically adds dither as the final process in the > >mixer. > > Aha, now I understand what you're getting at. Makes more sense than > my original interpretation :-). Sorry, had a long day yesterday when > I wrote the reply. Yes, this seems like sensible question -- but see > Steve Parker's reply, which in turn seems to be a sensible reply. It > could indeed very well be that truncating 32 to 24 bits doesn't > really matter, from a sonic perspective. Hi Guys, This whole debate is something I was confused about when I first started using logic and I came up with my own explanation which may or may not be correct. Either way I thought I'd share it with you and see what you think. I don't think there is any dithering taking place at all on output objects. I think the 32bit float resolution is used purely to provide additional headroom in mixing tracks. Say you are working in 24bits and mixing two audio signals. If the combined signal has an amplitude which is greater than that representable by 24bits you would usually get a clip. However if you use 32bits internally it can still be represented "without clipping". If you try to send this to an 24 bit output your signal will clip, and I've observed this. By reducing the volume on the outputs you are reducing the amplitude of the signal into a range representable by 24bits and hence no clipping. Does this makes sense to anyone? Tas.

http://www.xroadz.com/MOTU/BTD.html hi. the above link is from the motu-mac list....but is quite interesting as regards this issue. steve parker.