[virtualaudio]

A few notes based on:

1) Personal evaluation. (Though Not Bob Katz or G. Massenburg, I have been composing and engineering for many years).

2) The most reliable sources (Bob Katz, and many others - though I would like to point out that even with top flight engineers, opinions and tastes vary).

3) My (limited) knowledge of dsp coding (I don't code, but know just enough to make questionable assumptions).

Okay, fundamentals of dsp eq coding:

The two most useful types of digital eq are minimum phase and linear phase; there are others, but their functionality is not as common or useful. Maximum phase comes to mind (a filter that purposefully rings and distorts phase).
There a several ways to write an eq dsp code. According to apparently knowledgeable people in this forum, the most common fundamental method to produce the 'nuts and bolts' heart of a MP eq is a bi-quadratic equation (I hope I got that right; I'm in a hurry).

For minimum phase digital eq, there are a number of ways to arrange the filters (series, parallel, etc). These variations affect the gain/q dependency. They have nothing to do with harmonic distortion. (See the Algorithmix Blue description). I do not know if they affect the phase; if anyone can chime in here...

There are several 'corrective' or 'adjustable' methods for phase response (if a linear phase eq can be designed, then a minimum phase eq with a variant phase response should be possible? There are eq's that have different phase possibilities for each filter, so logically..) and hi frequency curve accuracy. (See the Eiosis Air Eq description). This is the part I am admittedly least knowledgeable on.

Many dsp coders 'emulate' 'analog models' by using the same basic equation (bi-quad for mp), restricting the filter points, controlling the gain/q dependency and putting a gui on it that looks like the hardware. A few add noise (of various kinds - see the TRacks noise rise at 300Hz, and the discussion on the Waves SSL series, where the 'Analog' button adds low level white noise (is it white?) and/or some harmonic content (again Waves, particularly API - thanks Fader8 for your informative webblog, and everyone else here that has taken the time to run tests on various eq's).

Some eq's upsample in an attempt to calculate more accurately. This procedure has it's own drawbacks (SRC is non trivial, and to do it right requires a big CPU hit - I saw a website with aliasing test results for sample rate converters, both hardware and software; if I can find the link, I will include it here). Of course, upsampling-downsampling with the same multiplier (48 - 192 - 48) is easier than with a different multiplier (48 - 44.1), but still, it can color the sound. On the other hand, it spreads the errors out over a larger range. If done very well, it can have advantages. Katz says that he is not sure it has any meaningful significance for eq, but can definitely improve dynamic behavior, like compressors, limiters, etc. On major factor in src is the filtering used to restrict the end product to the nyquist frequency to avoid aliasing. See Bob Katz' book for detailed information.

The bit-depth (wordlength) calculation also affects accuracy; a 24 bit calculation rounds enough errors off enough to potentially be perceptible, especially after repeated calculation. Floating point calculations have the added benefit of being able to handle digital overs. Of course a poorly implemented algorithm won't be helped by a 64 bit wordlength or upsampling. But, in eq's especially, 24 bit calculations are too limited to be accurate for high end work (I am not sure I can buy LPK's assertion that info lower than 144dB is meaningless; especially with cumulative calculations).

Okay, those seem to be the fundamentals for dsp eq coding, particularly minimum phase.

Now, my observations:

First,
The Sonnox and Sonalksis eq's sound better to me that my internal DAW eq (Cubase SX3) or the Waves Ren eq. I do not believe it is due to a different qain/q dependency. No way. What could it be then?

The Algorithmix Blue sounds better to me than any other minimum phase eq I have ever heard. (Perhaps this is due to high quality up-sampling and 80 bit floating point calculations); according to them, any well designed, properly implemented analytical digital peq can match any eq setup (I am guessing they mean any, including analog, minus harmonic artifacts that are generated by analog circuitry - thus 'analytical'). I had the opportunity to demo the Blue and I loved the way it sounded, but the cost and dsp hit are very high, too high for my budget. If they are using the cookbook bi-quad dsp, they are adding extra steps to minimize artifacts and maximize accuracy. They say they are using proprietary algorithms..

Which brings me to:
As a previous poster stated, there is a lot of competition in the dsp coding world, and the ROI is slim. From personal experience, DSP coders can be very secretive about their methods. A few people that have posted here have serious coding experience; how much are they willing to divulge? I am not calling the accuracy of their statements into question, but if all the points I have made are valid, then there are ways to make an eq that sounds much better than another eq with the same q/gain dependency, and one may wonder what extra steps are taken to acheive this result.. the manufacturers are generally less than willing to divulge hard earned coding methods. Yes, some marketing is snake oil; but how can we really know the 'extras' that are taking place in a dsp like, say, the UAD Neve 1073? I have also demoed that, and I have to say that it sounds very good; not quite like the hardware, but I don't think I could get my DAW eq to sound like it, and even with an eq like the Algo Blue, it would take a long time to dial in the same settings, and then one would have to add comparable harmonic generation to match their 'analog circuit' modeling (see Fader8's webBlog for the harmonic info on this and other eq's). Which leads to -
Workflow:
If a high quality eq emulation helps you because you are used to that particular piece of hardware, then I say use it. As a previous poster stated, they did the work to dial in the gain/q dependencies (and add some harmonics to help it sound analog); do you really want to go through the trouble, or do you want to be an efficient engineer? And yes, modeling still does not (and perhaps never will) sound exactly like the original, but if it is very close and costs 10 times less, then some of us can benefit from the ease of use due to familiarity..

The above points were focused on minimum phase eq's; linear phase is another matter:

I have also had the opportunity to demo the Algorithmix Red and Orange eq's, and compare them against the Waves LP and the Ozone 3 & 4 LP eq's; I now own the Red and use it frequently, and can say that it definitely sounds better than the Waves and Ozone. Furthermore, refer to Bob Katz' book regarding the exceptionally low pre-post echo of the Red as compared with the Orange, Weiss and other LP eq implementations. There is definitely something quantifiable going on here. (The Red is a 'frequency domain' eq - I think that means an FIR design that has somehow avoided the pitfalls of most FIR designs - see above about proprietary coding and the secretiveness of developers. The Orange is a 'time domain' eq; IIR, I believe - though someone said earlier in this thread that LP eq's were typically IIR followed by FIR, or at least that is one way to do it; I hope I am not misrepresenting). Bob and others call it 'the sweetest LP eq I have ever heard' and 'more analog than analog'.. but hey, maybe they are just hawking a product.. I love it.

Moving on:

If you can run the same source material through two different eq's, match the q/gain (match the eq curve exactly, not by knob readings, but by fft and/or another method like twiddling the settings until you get it right), flip the polarity of one and then sum them and get absolute silence, they are absolutely equivalent, at least at that setting. That is a null test. Period. If there is any residue left, they are not equivalent, which may or may not have major significance regarding the 'sound' of the eq. That is the rub. What low level (ambient) information is psychoacoustically significant? That is a whole debate in itself.. Waves seems to think a little white noise helps their SSL plugs sound 'analog'.. Hell, Katz makes several points about noise and masking, the perception of digital as being sterile, the addition of noise as not necessarily bad, etc. What works, what doesn't, where do we draw the line? (This is related to the analog summing controversy, and in a sense, to the whole debate about analog vs. digital, modeling, etc., yes?)
~~~
So, maybe 90% of minimum phase eq's use the same basic method, but there seem to be a few ways to modify this calculation and/or increase the accuracy (bit depth/sample rate) of it to produce results that are superior (or more aurally desirable) to other eq's. But yes, there are lines, fuzzy to be sure, but lines where annoootthhhher damned modeled eq plugin is a snake oil ripoff (depending on your POV and workflow needs, preferences and budget). But if it helps you get the sound you want..
I personally don't want to spend the time to dial in a 1073 curve, and would rather use the UAD, but that is my preference. Good curves, a little harmonics, and.. I don't know what else, and the developers aren't telling (and no, it is not nano-tubes on the card!). I don't think I could get the nice Voxengo Overtone Geq to sound like the UAD 1073, but they are different animals; maybe the HarmoniEQ.. I'll have to demo and try it.. I have not seen a defintive answer on the dynamic and program response of certain modeled eq's; is there a dynamic 'transformer reactivity' in, say, the UAD stuff that would increase saturation at higher levels? I can say that their 33609 compressor does this, but it is also a cpu hog.

If there are only a few ways to skin this cat, and all the dsp and modeling knowledge obtained by all the developers were combined, I guess an eq could be designed to do anything possible in the digital domain, including coming as close as mathematically possible to all the most desirable hardware, with a bunch of great presets (1073, Manley, Weiss, Pultec, etc), with great GUI functionality. I don't expect to see this any time soon, but would love to have it. Then I would be satisfied with one all purpose eq plugin!
Until then, with the subtle variations in coding, efforts at modeling and economic and mathematical limits of developers, I say Caveat Emptor, but get what sounds good to you and enhances your workflow; you don't need 100 plugin eq's, but I don't think the one in your DAW is enough unless it is really great (and I don't think the Cubase SX3 one is, at least; can't speak for other DAW's).

Sorry for the rambling and incomplete post; I don't have the time to write an organized dissertation. And I apologize if I misrepresented anyone's words, meanings or opinions; some of my post is subjective, and I may be wrong about a few technical point, and would be happy to be corrected by anyone more knowledgeable.

I forgot to mention one last gem from Bob Katz' book: Mastering Audio.
To paraphrase, each of the various methods we have to measure the performance of any audio signal processor only offers a specific 'window' to look into the inner workings; no one (or even two or three) measurements can tell us enough to have a comprehensive evaluation of the 'real-world' behavior of the signal processor; which is to say that for any processor with a sophisticated design (I say this to exclude simple mathematically predictable dsp coded algorithms), only one or two measurements is not enough to say what the psychoacoustic effect the processor may exhibit in any given real world application, or what anomaly might be 'important' and what is trivial.. (I say 'simple mathematically predictable dsp coded algorithms' to distinguish, say, a common 2nd order butterworth filter from the PSP Vintage Warmer; pick your analogy.. and to extend the idea, I am not even sure Sonnox knows exactly what their Inflator does in every application... or if they do, they don't specify in the manual; really, to use the words 'statistical' and 'probability' in the same sentence!)


George Piazza
.:O O:. G.C. N.O.
Tabula Rasa
~~~~~~~~~~~~
V.A.S.T. - Virtual Audio Studio Tech
Mastering, Mixing, Recording & Restoration
Soundtrack/Sound Design for Film / Video
MySpace.com - Tabula Rasa / V.A.S.T. / George Piazza - 45 - Male - New Orleans, Louisiana - www.myspace.com/georgepiazza