On 7/17/2011 6:25 AM, Dale Powell wrote: In Integer math. In floating point math, A + B = C +/- (very small rounding error). At one time the USA's NASA (*National Aeronautics and Space Administration*) used Integer math to track their satellites. I don't know if NASA is still using Integer math or has gone to Floating Point math for the calculations now. Best, Stephen.

Thanks to all of you for your inputs! I'm happy that my ears are probably OK, because, as I said, I wasn't able to hear any difference between mixes done on different DAWs. Just to verify I get it right, 48bit = double precision, which means a greater dynamic range? On Sun, Jul 17, 2011 at 2:25 PM, Dale Powell <dj_kaza(a)hotmail.com> wrote:

Moshe Werner wrote: That 48 bit thing could hark back to Protools when it used the Motorola 56k DSP chips: https://secure.wikimedia.org/wikipedia/en/wiki/Motorola_56000 which are fixed point DPS chips. According to this: https://secure.wikimedia.org/wikipedia/en/wiki/Protools Protools uses 48 bit fixed point arithmetic (possibly doubled 24 bit registers of the Moto 56k or possibly custom hardware). If it is 48 bit fixed point I would think that it is inferior to Ardour which almost certain uses double floating point which has a 53 bit significand. Erik -- ---------------------------------------------------------------------- Erik de Castro Lopo http://www.mega-nerd.com/

On Mon, 18 Jul 2011 11:53:47 +0200 David Adler <david.jo.adler(a)gmail.com> wrote: isn't it useless going way above 24 bits, since that's what in the end the DAC (sound card) will use? as far as I can remember, delta-sigma DACs don't go over 24 bit, do they? cheers renato

David Adler wrote: Do not confuse DAC/ADC bit width with bit widths used during DSP calculations. The two are basically orthogonal. For audio data paths this is true. For the intermediate results during complicated calculations (eg filters and reverbs) double precision makes things easier and is basically available for free. Erik -- ---------------------------------------------------------------------- Erik de Castro Lopo http://www.mega-nerd.com/

David Adler wrote: The actual data values are 32 bit but they are converted to 64 bit before they the arithmetic is done. For instance in Secret Rabbit Code (my code), all data entering and leaving the converter plus the actual filter coefficients are stored as 32 bit floats. However, the inner loop which does the multiply accumulate (similar to what is done when mixing) does: double result = 0.0 ; for ( ..... ) sum += coeff [k] * data [k] ; Specifically all the inputs are 32 bit floats, but all intermediate results are 64 bit. Floating point calculations have problems. Specifically, if you take a long list of numbers with both very large values and very small values, you will get different results depending on whether you add them the smallest to largest vs largest to smallest. For the most accurate results, add from the smallest to largest. This is probably the best known paper on the issues surrounding floating point: http://download.oracle.com/docs/cd/E19422-01/819-3693/ncg_goldberg.html However, the problems of floating point are almost non-existant in comparison to the problems of fixed point. I would be almost certain that Jack works on single presicion float data, but does all the intermediate calculations in double precision. If we assume that the 48 bit arithmetic only represents values in the range (-1.0, 1.0) (this is usually the case when doing audio processing on DSP processors). Consider two values that are to be stored in a 48 bit fixed point register: va = 1.0 / pi vb = 1.0 / (pi * 0x10000000000) In the case of the value va, nearly all of the 48 register bits will be used and we will get close to 48 bits of precision. For the case of vb, a number very much smaller than 1.0, about 40 of the most significant bits will be zeros, leaving only about 8 bits of precision. Now compare the above fixed point prepresentation with the floating point representation where the mantissa would have the same number of bits for both numbers and only the exponents would differ. No, this is much more likely the double precision mode of the Motorola 56000 family of 24 bit fixed point DSP chips. Exactly. Floating point, especially double floating point makes it easier to code, because there's much less of this faffing about required. Erik -- ---------------------------------------------------------------------- Erik de Castro Lopo http://www.mega-nerd.com/

David Adler wrote: I haven't looked, but since the people coding Jack know what they are doing. I assume that Jack just does the right thing. The client has no control over whether Jack does 32 or 64 bit float calculations. Erik -- ---------------------------------------------------------------------- Erik de Castro Lopo http://www.mega-nerd.com/

On Mon, Jul 18, 2011 at 10:10:57PM +1000, Erik de Castro Lopo wrote: Apart from summing at input ports and optionally dithering the output signals Jack doesn't do any audio calculations. The summing is of course what we are discussing. It would probably depend on how Jack's summing code works - if the inner loop is over buffers then the intermediate sums are probably full precision. If it is over samples then the intermediate results are probably stored as floats. -- FA

On Mon, Jul 18, 2011 at 11:02:50AM -0400, Paul Davis wrote: Don't mention it ! :-) -- FA

On Sun, Jul 17, 2011 at 7:25 AM, Dale Powell <dj_kaza(a)hotmail.com> wrote: Isn't all this "A+B=C" stuff (or A*x+B*y=C as one person stated) actually begging the question a bit? Are we certain that every DAW implements their mixer that way? Isn't it possible that some might try to model analog mixers to some degree? I'm not arguing either way; I have no clue.

On Mon, Jul 18, 2011 at 6:03 PM, Folderol <folderol(a)ukfsn.org> wrote: precisely. the "magic" in analog circuits does not come from the way it adds two in-range voltages, which is precisely as stated above. take 0.1V, add 0.18V, get 0.28V. it really is that simple (we're ignoring temporal effects that are similar for actual air pressure waves and thus not relevant to a discussion of artifacts caused by recording and playback). the magic comes, to the extent that its actually magic (let alone actually desirable) from the way it does more complex processing, such as compression, limiting and EQ, and also the way that some media will store the result when C is "out of range". even these can be replicated to any arbitrary accuracy if you're willing to spend the CPU cycles on it.

On Tue, Jul 19, 2011 at 5:47 PM, Paul Coccoli <pcoccoli(a)gmail.com> wrote: there are plugins (typically called "channel strips") that try to do this. and there is mixbus, which definitely tries to do this (though not in the "analog modelling the entire channel strip" way that some rather silly commentators on some online forums think it does). there is also "record" (now bundled into reason) which is similar (but modelled on an SSL console rather than a Harrison one). anybody who is saying they are not even possible in theory is an idiot. there is, however, the practical issue of how much money you want to put into understanding the behaviour of an analog circuit and then the equally practical issue of how many CPU cycles you are willing to devote to emulating that behaviour. both act as a bit of damper on doing this in real life, but not to the point of making it "impossible" (or even close to that).

On Mon, Jul 18, 2011 at 5:03 PM, Folderol <folderol(a)ukfsn.org> wrote: To the extent that modeling actual analog mixers would be useless. The amplifiers in analog mixers do have non-flat spectrum (as all amplifiers have a limited gain-bandwidth product and phase-shift at high frequencies). However, the equipment is so designed as to have a nearly constant phase and gain over the range of audio frequencies. Plus, add noise. Both of which you could do intentionally, instead of having some software black box that does it without telling you :)

On Sun, Jul 17, 2011 at 08:45:38AM +0000, Fons Adriaensen wrote: There's another reason, and, as someone who used to be in the business of selling snake oil professionally, I can speak to that a bit. Look at who is buying. "Pro" audio now is so affordable that people who are decidedly NOT pro's are buying it. They don't have formal audio engineering training or even in some cases a remote familiarity with the basic science or even the scientific method. They don't understand what they are buying, or what they are talking about when they brag or argue about what they bought or why. Advertisers can (and do) get away with making up all kinds of wild lies, and if they don't, in some cases the buyers will do it for them.

On Mon, Jul 18, 2011 at 12:40:06PM +0200, lorenzo cecchi wrote: No they don't. There would be a difference if you take the fader gains and panning from one DAW and apply them in another which has a different panning gain at the center. Apart from this, while doing a mix will just compensate for the different panner center gains by using different fader gains. -- FA