Also, does anyone know of any digital consoles that are less expensive than the Soundtracs or Euphonix that use this
floating point principle ( Sony DMX ??? ).
Thanks.

floating point doesn't have greater dynamic range than fixed point...

where did you get that idea??

it's the bit depth that matters....

and my experience with floating point is that fixed point sounds better!

so what is the problem here?? can't you sum all you channels without distortion??

maybe you should look into a protools system if you're tired of your yamahas... been a while since i worked with them....

or some analog summing...

A heard that 40 bit floating point processing would give me more headroom than the Yamaha. Is this not the case ?
I get fairly quickly in the red on my stereo bus in the Yamaha when
I use a lot of tracks.

iznogood wrote: floating point doesn't have greater dynamic range than fixed point...

where did you get that idea??

.

Iznogood, indeed.
A 32-bit floating point system has virtually unlimited internal headroom!! As oposed to your(and my) pro tools system which tolerates no internal peaks on buses or channels.

As long as the final output of a floating point system is scaled down so that the DA, or 16/24 bit output does not peak, your headroom is endless.

This is the big advantage of floating point vs. Fixed point.
Do the test yourself, try to create internal overload in cubase or Nuendo! Det kommer du ikke at klare.

Thanks for your replies.
Does this also mean that you have more headroom on the 2-bus bus
itself with floating point ?
How does it actually make a difference on the 2-bus bus as opposed to fixed point ?
Will it actually make a dynamic difference at the end ?
Sorry, still a little fuzzy about that.

Turner wrote: Thanks for your replies.
Does this also mean that you have more headroom on the 2-bus bus
itself with floating point ?
How does it actually make a difference on the 2-bus bus as opposed to fixed point ?
Will it actually make a dynamic difference at the end ?
Sorry, still a little fuzzy about that.

Headroom on the 2-bus, in the sense that you can't overload the 2-bus, yes. But you still have to keep within 0dBfs for 24/16 bit digital output, DA, or bouncing to any other digital format.
32bit floating does not give you a larger dynamic range in the end format(unless it's 32bit floating), because of the above reduction.
It's purely internal headroom.

To me, infinite internal headroom is really pointless. counter intuitive actually.

Most people are used to the fact that you have finite headroom and that you should stay bellow 0dbf. In this respect, fixed point is closer to "analog" than floating point math.

why not just keep the faders down???

mixed alot of records.... never had the masterbus peak alot

shmaba wrote: [quote=iznogood]floating point doesn't have greater dynamic range than fixed point...

where did you get that idea??

.

Iznogood, indeed.
A 32-bit floating point system has virtually unlimited internal headroom!! As oposed to your(and my) pro tools system which tolerates no internal peaks on buses or channels.

As long as the final output of a floating point system is scaled down so that the DA, or 16/24 bit output does not peak, your headroom is endless.

This is the big advantage of floating point vs. Fixed point.
Do the test yourself, try to create internal overload in cubase or Nuendo! Det kommer du ikke at klare.

I assume that you are talking about PT LE, right?

Thanks for your answers everyone.
What is the difference sonicly in end result then going out the
2-bus bus, between a mixer that uses fixed or floating internal
processing then. Will there be a more open sound coming out the
2 mix bus ?
That's what's finally the most important thing, does it make any difference in end result ?

it makes a difference...

but the most important thing to achieve good sound is good monitoring and talent/ears

iznogood wrote: it makes a difference...

but the most important thing to achieve good sound is good monitoring and talent/ears

I'll second that.

To answer the rest:
32-bit floating includes PT LE, Logic, Cubase, Nuendo, and virtually any other native system.

PT TDM is 48bit fixed point and as NO internal headroom above 0dBfs. The in and output of the plug in bus in PT TDM is 24bit.

Analouge mixers usually have quite a lot of internal headroom, and often price goes hand in hand with the amount of it- especially regarding the mix bus- discrete Neve or API's have A LOT of internal headroom.

The end result is influenced by the time you need to dedicate to keeping the gain staging straight.

I usually work with PT TDM, but everytime I work with Nuendo, I'm thrilled with having one less worry.
Altough all things added up, I do prefer PT TDM.

Thanks for the answers, Shmaba.
Just a couple a more questions.
Why do you like PT TDM more than Logic ? Is it because of the plug-ins or interface ? I use midi a lot, combined with audiotracks.
And also, when doing final mixes for music-production, do you mix
in the box or do you go external ( digital or analog ? ).
I think that the O2/R is my weakest link in the chain.
( it's not only because of floating or fixed point of course, it also has to do with the preamps, dynamics, eq and effects in the mixer ).
Should I go for a digital higher level mixer with floating point or digitally controlled analog to get better results ?
Just want to ameliorate my audio chain.
Thanks for any suggestions.

shmaba wrote:
PT TDM is 48bit fixed point and as NO internal headroom above 0dBfs. The in and output of the plug in bus in PT TDM is 24bit.

A small tidbit.

True. Full scale is full scale. But the design in PT TDM is such that even with all channels going full blast at 0dB the mix bus will not overload (the guys who did that design knew what they did). The part that may overload is right after the master fader. But if you turn that one down, you are fine.

Gunnar

That remark is an important one Gunnar.
If I do this on the Yamaha it won't work.
I mean, even if I don't have tracks going 0 dB, the master will overload fairly quickly. This actually struck me very soon after I bought the mixer. It gets worse the more tracks I have.
I have to check out the remark that there would be more internal headroom in Logic than Protools.
I always had the impression that Logic would go in the red faster
than Protools. Anyway, I have to check it out.
Do you guys mix in the box ? If not, which console do you use or
do you have any preference. Thanks.

shmaba wrote: [quote=iznogood]it makes a difference...

but the most important thing to achieve good sound is good monitoring and talent/ears

I'll second that.

To answer the rest:
32-bit floating includes PT LE, Logic, Cubase, Nuendo, and virtually any other native system.

PT TDM is 48bit fixed point and as NO internal headroom above 0dBfs. The in and output of the plug in bus in PT TDM is 24bit.

Analouge mixers usually have quite a lot of internal headroom, and often price goes hand in hand with the amount of it- especially regarding the mix bus- discrete Neve or API's have A LOT of internal headroom.

The end result is influenced by the time you need to dedicate to keeping the gain staging straight.

I usually work with PT TDM, but everytime I work with Nuendo, I'm thrilled with having one less worry.
Altough all things added up, I do prefer PT TDM.

Two things.
1. SAW is fixed point and it is native.
2. There seems to be confusion as to what "headroom" is. There is no headroom technically in digital. Period. There comes a point, where all available bits have been used, and that is 0. This is not an analog thing where the circuit performs best at +4dBV (0), and then you have some headroom above that, say +28 on a 9098 or SSL, etc. In digital, 0 is 0ut, all gone, no more, we are now using a nice flat line. Theoretically, a 48 bit engine floating or fixed should have the same "dynamic range" (1 bit = 6dB). The difference is going to be the accuracy of the math.

The sonic differences between floating and fixed point can be extremely subtle, or extremely dramatic, depending on the compared DAW's or digital consoles, etc. If you are recording and sampling with prosumer or hobbiest gear and facilities, it is really a moot point anyway.

The sonic differences between floating and fixed point can be extremely subtle, or extremely dramatic, depending on the compared DAW's or digital consoles, etc. If you are recording and sampling with prosumer or hobbiest gear and facilities, it is really a moot point anyway.

If one has more than one channel of audio that is sampled and then mixed to maintain near 0 full scale, then you WILL overload the mix bus most likely. This is not an error on the mix buses part, or the mix engines part. It is operator error. If you are having overload issues, drop all of your levels.

The biggest error that I see when mixing local projects is a limiter has been used on ever freakin' channel. There is no RMS left! They are trying to use the illusive 24th bit on every channel. This sucks. It takes the dynamics out of the music, but also uses up too many bits to represent the signal whose dynamic range required half as many bits. Again, 0 in digital is not like 0 in analog. 0 in digital should be around -12 to -16. Apogee recommends -12. Digi's 192s came calibrated to -14. YMMV.

That is a very oversimplified statement.

In 32 bit float you have virtually unlimited additional headroom above "0dBfs", I state "0dBfs" because you have to stay within that number when converting to 16 or 24, and 0dBfs is stated in those programs as the 16/24 0dBfs- The internal 0dBfs is much, much, much higher.

It's as simple as that, but hey, if you don't get it, don't go over 0dBfs anywhere. And since youre on PT you can't do it anyway.

YMMV , Indeed. Read up.

Turner, both Logic and PT "hit red" at the same point, the difference is, that in PT this is the overload point, your audio will digitally distort above that point. While in Logic(Nuendo, Cubase), as long as you're speaking internal busing, it really doesn't matter sonically.

In regards to any output to DA or 16/24 bit, you have to stay within 0dBfs.

You may be the one needing to read up, as you do not understand what happens exactly.

James Moorer, former Sonic Solutions and now Adobe engineer submitted a paper on the subject, how 48 bit fixed beats 32 floating (search AES library).

When comparing double-precision fixed-point (parrallel 24 bit) vs. single-precision floating-point DSP, he states “…there is an advantage to using integer arithmetic in general, in that most integer (24-bit, fixed-point) arithmetic units have very wide accumulators, such as 48 to 56 bits, whereas 32-bit floating-point arithmetic units generally have only 24 or 32 bits of mantissa precision in the accumulator. This can lead to serious signal degradation, especially with low-frequency filters.”

The signal degradation mentioned translates into 3 or 4 bits of precision. When multiple operations are performed on a signal, that can add up. Accumulator = temporary memory location, or register, that stores the result of an addition or multiplication operation. Any bits that don't fit in the accumulator must be thrown out, usually via a rounding operation.

sheet wrote:
When comparing double-precision fixed-point (parrallel 24 bit) vs. single-precision floating-point DSP, he states “…there is an advantage to using integer arithmetic in general, in that most integer (24-bit, fixed-point) arithmetic units have very wide accumulators, such as 48 to 56 bits, whereas 32-bit floating-point arithmetic units generally have only 24 or 32 bits of mantissa precision in the accumulator. This can lead to serious signal degradation, especially with low-frequency filters.”

Well, things are never really easy when you want to prove something. The default size accumulator on the Intel platform running on Windows is the so called 53 bit format. You may read a bit more about the format here
http://maven.smith.edu/~thiebaut/ArtOfAssembly/CH14/CH14-3.html#HEADING3-6

But back to the basic issue. Both fixed and floating point "buses" can be made to sound good. If anything a fixed point bus may have a small advantage there. The disadvantage from a user point of view is that there is an absolute stop, where the bus starts to overload. In PT HD this absolute level is well above where you will end up unless you do some really crazy things. On some other mixers the stop is very easily seen. On a floating point system, the full stop is always (in my experience) well above the level where a user would end up.

So from a user point of view, float is much more convenient.

But, to program a good DAW algorithm has some pitfalls. A straigh-off "naiive" implementation might have weak spots. There are some, albeit older, implementations that sound hearably worse due to this. Modern well-rounded implementations in general sound good enough for most purposes. My favourite application, Samplitude, is considered to sound good by some people with good ears. You may find similar experiences from people using most other platforms.

Gunnar.

shmaba wrote: A 32-bit floating point system has virtually unlimited internal headroom!! As oposed to your(and my) pro tools system which tolerates no internal peaks on buses or channels.

As long as the final output of a floating point system is scaled down so that the DA, or 16/24 bit output does not peak, your headroom is endless.

This is the big advantage of floating point vs. Fixed point.
Do the test yourself, try to create internal overload in cubase or Nuendo!

Umm. Sorta'.

Protools TDM systems have enough headroom that you can sum together 64 channels of simultaneous full scale data without clipping. That is an ENORMOUS amount of headroom that is far in excess of any practical mixing scenario from an engineer that uses conventional mixing methods.

While floating point systems do theoretically provide additional headroom we must keep in mind that it is with constant and fixed dynamic range. As the use of the headroom floats upward so too does the noise floor. This does not happen in a fixed point system such as Protools TDM etc.

It is also probably pertinent to discuss the downsides of 32 bit floating point systems, such as the repeated truncation distortion from having to reduce its wordlength to 25 bits internally after various stages without dither.

Nika

shmaba wrote: PT TDM is 48bit fixed point and as NO internal headroom above 0dBfs.

This is absolutely erroneous as I described above. Protools TDM has 30dB of headroom internally. I doubt you could exceed this in any non-pathological mixing environment.

Analouge mixers usually have quite a lot of internal headroom,

Though not as much as Protools TDM. Further, the noisefloor on the analog mixers is higher, meaning less dynamic range. As someone else said, analog mixers are also far more analgous to fixed point nomenclature than floating point nomenclature.

Nika

Turner wrote: That remark is an important one Gunnar.
If I do this on the Yamaha it won't work.
I mean, even if I don't have tracks going 0 dB, the master will overload fairly quickly. This actually struck me very soon after I bought the mixer. It gets worse the more tracks I have.

I would not be convinced that it's a floating vs. fixed issue in your consoles. It sounds like an implementation issue. It sounds like the o2rs are poorly designed, but this has nothing to do with being floating or fixed. For what it's worth, the Sony Oxford console was entirely fixed point.

Nika

sheet wrote: 2. There seems to be confusion as to what "headroom" is. There is no headroom technically in digital. Period. There comes a point, where all available bits have been used, and that is 0. This is not an analog thing where the circuit performs best at +4dBV (0), and then you have some headroom above that, say +28 on a 9098 or SSL, etc. In digital, 0 is 0ut, all gone, no more, we are now using a nice flat line.

If one has more than one channel of audio that is sampled and then mixed to maintain near 0 full scale, then you WILL overload the mix bus most likely. This is not an error on the mix buses part, or the mix engines part. It is operator error. If you are having overload issues, drop all of your levels.

This is not correct. If you sum two full scale channels on a properly designed digital mixer (such as, for example, Protools TDM) then the internal architecture (we call it headroom even though it is linear) allows the numbers to still pile up properly, continuing to add with a few bits to the left extra to cover the overflow. Then the master fader is used to scale it back within a reasonable region when it is time for the data to leave the system.

Nika

Thanks Nika!

I'm learning something everytime i read something you have written!

Wow, a lot of discussion goin on here.
What I learn from your posts and looking up the
internal processing of a few digital consoles, I would have to conclude the following :
Fixed or floating point processing does not have such a lot of impact
on what comes out of the mixer as I thought.
I mean, the Sony Oxford or the DM 2000 both use 32 bit fixed point
processing, but there's a big difference in sonic performance not ?
There are other things that must be more important to determine
the quality of the mixer. What things are they ?
I should think AD/DA conversion, but what other determining
factors ? Sorry if this seems a little oversimplified for some of you.
Thanks for any comments.

Turner wrote:
Fixed or floating point processing does not have such a lot of impact
on what comes out of the mixer as I thought.
I mean, the Sony Oxford or the DM 2000 both use 32 bit fixed point
processing, but there's a big difference in sonic performance not ?
There are other things that must be more important to determine
the quality of the mixer. What things are they ?
I should think AD/DA conversion, but what other determining
factors ?

IMPLEMENTATION!

It's not whether it's floating or fixed but rather how they program it whatever it is!

If you really want to go off the deep end in your research just keep this word close at hand: dither. A lof of it comes down to how these manufacturers implement dither.

Cheers!
Nika

Thanks for this remark Nika !
It makes sense to me. I also understand that a lot of dithering is going on in the console and that the dither at the final stage is crucial. Have to check out your book.
A paper on the O2/R states :
"You might want to use another dither arythmic than the one in the O2/R, before going to your storage device".
Is it wise to install such a dithering device and which one would you prefer ? How can you know the quality of dithering in a console?
Thank you and to everybody who has participated in this topic.

It is very difficult to ascertain the quality of dither inside a console. Just for example, however, at one point in the lifespan of the Sony Oxford console roughly 1/3 of the 450 dedicated DSP chips in it were used for the generation and application of dither. That's 150 DSP chips - equivalent to, say, 20 Protools systems or so - just for the generation of dither. Dither is a crucial aspect of implementation.

Nika