EQ as represented by the graphics, or not

[CinningBao]

My Sunday morning thought

EQ display of four different plugins - (clockwise from bottom left) Apple AU Parametric EQ, Voxengo Voxformer, Logic built-in Channel EQ, and Ableton built-in 8band EQ..

Screen Shot 2017-07-23 at 10.59.20.jpg

I was just wondering, with regard to the 'bending' of the EQ shape near the top end of the Ableton EQ and Apple AU Paramteric EQ, if this is an accurate represention of what the code is doing? Logic EQ nor the Voxformer EQ (a limited comparison, I know) display this characteristic so is Ableton's EQ curve doing asymmetrical gaining near the top end or is it an oversight of the Logic EQ GUI designers not to include this kind of filtering in display (assuming it is just 'the way' EQ's work near the nyquist frequency)


This might not be the most appropriate forum choice for this post, please Mods, move as you see fit.

[lobanov]

Nevermind. Use your ears instead of fancy GUIs. Nobody cares how your curves look, nobody sees it.

Compare the sound of two eqs. Do they differ? Is this difference considerable or vital? That's it!

[CinningBao]

I should have qualified this with "I'm not using GUI's to make my EQ alterations" - this question has nothing to do with my ears!

this is perhaps why I should have posted this in DSP.. I'm wondering if the GUIs are representative of what the code actually doing. Are EQ curves which behave like the AbletonEQ8 and AU Parametric EQ actually doing what the GUI describes.

Maybe I should just do some analysis here..

[whyterabbyt]

The EQ curves drawn are an abstraction for UI purposes, not an analysis of the actual curve. And it kinda looks like the axis scaling might be different on those plugins anyway, affecting that abstraction.

[lobanov]

Why do you need to know it? Curiosity? Any practical value of this knowledge is doubtful. If you are not a developer.
Every GUI is a set of symbols, these symbols must not give us an absolutely exact representation of real processes. There is no need in it. And I think it's impossible.

[MogwaiBoy]

Or it's not just a GUI quirk and it's actually what's called "pre-warping" or "cramping" - a problem of it mathematically always wanting to be 0db gain at the Nyquist. This can be solved with oversampling (eg: HOFA IQ-EQ) or clever code (eg. Pro-Q2).

[CinningBao]

Thanks whyterabbyt, is that what you did when you last coded an EQ GUI?
Yes, there may well be vertical scaling differences between the GUIs, my observation is not based on the scaling at all.

I observed that some of these EQ plugins exhibit ASYMMETRY in the high-frequency range and wondered

a) if any EQ algorithms do actually behave like this
and
b) if they do, why choose one method over the other

[whyterabbyt]

Yes, there may well be vertical scaling differences between the GUIs, my observation is not based on the scaling at all.

I observed that some of these EQ plugins exhibit ASYMMETRY in the high-frequency range and wondered

You see asymmetry of behaviour on the horizontal axis (frequency), but when I say 'difference in axis scaling' discount it because of the vertical?

[random_id]

One of the most common EQ algorithm used by developers will produce this phenomenon. For peak filters, the gain at nyquist will be 0. For shelf filters, the gain at nyguist will match the gain of the filter (which cramps everything as the frequency is closer to nyquist). For a low pass, the gain at nyquist will be 0 (which also cramps everything).

My guess is that the visual curves are pretty accurate. The developers are just using the same algorithm to calculate the filter to calculate the frequency response. If anything, you are seeing that not all developers use the same EQ filters. Oversampling will help, with the added cost of CPU cycles and a steep filter at nyquist. There are also alternative algorithms that don't produce these same effects. I would definitely try a bunch out and find the ones that you like.

[CinningBao]

@lobanov - simple curiosity, after observing this asymmetry in the GUI, I did some quick analyses and it does appear that the AU Parametric EQ exhibits this asymmetry at the output as well as the GUI istself, unless the analyser in Channel EQ is reporting incorrectly as well!

@MogwaiBoy - I think you might be closer to a real answer.. Steve Hoffman (a mastering engineer who clearly knows his stuff) says "..the Sony Oxford EQ plug-in produces a fully de-cramped and symmetrical EQ response .." while implying this symmetry is the preferred design, and also how analog EQs deliver their loveliness.

So it seems, not surprisingly, that the old AU Parametric EQ and the Ableton EQ use an inferior (cheaper on CPU?) EQ algorithm which doesn't fully de-cramp the signal, but the Logic Channel EQ and Voxformer do, do some degree, ensure the gains are symmetrical either side of the specified frequency.

[CinningBao]

Yes, there may well be vertical scaling differences between the GUIs, my observation is not based on the scaling at all.

I observed that some of these EQ plugins exhibit ASYMMETRY in the high-frequency range and wondered

You see asymmetry of behaviour on the horizontal axis (frequency), but when I say 'difference in axis scaling' discount it because of the vertical?

Not at all, amplitude scaling, unless it changes across the frequency range (which I suspect not!) is not part of this observation. Like I said, it is the difference either side of the selected frequency which piqued my interest with this today.

[CinningBao]

One of the most common EQ algorithm used by developers will produce this phenomenon. For peak filters, the gain at nyquist will be 0. For shelf filters, the gain at nyguist will match the gain of the filter (which cramps everything as the frequency is closer to nyquist). For a low pass, the gain at nyquist will be 0 (which also cramps everything).

My guess is that the visual curves are pretty accurate. The developers are just using the same algorithm to calculate the filter to calculate the frequency response. If anything, you are seeing that not all developers use the same EQ filters. Oversampling will help, with the added cost of CPU cycles and a steep filter at nyquist. There are also alternative algorithms that don't produce these same effects. I would definitely try a bunch out and find the ones that you like.

Now this is a GOOOOD answer - using the same algorithm to generate the graphics is good practise, surely!
Ah, and you mention a steep filter at nyquist - this could well explain the asymmetry of the curves nearer nyquist.

My curiosity has been satisfied! (unless it's obvious I missed a crucial aspect)
cheers

[BRBWaffles]

Yeah, the UI is just a guide. Some are better than others. I've found Pro-Q to be the most accurate, and ReaEQ to be the absolute least accurate out of any EQ interface I've used. ReaEQ seems to always be off with the UI. A frequency will be very clearly popping up in the analyzer, but to isolate it accurately you actually need to make your EQ move slightly to the right of the frequency shown. It's bloody annoying.

[camsr]

One problem to watch out for is when you have a warped EQ (represented graphically or not), the peak of the amplitude may be in the wrong location also... a peaking filter may not peak where it shows on the graphics.

[FabienTDR]

This is one reason why TDL plugins are more demanding on the CPU. They hold their promises under any circumstance, much like anybody would expect from a professional tool.

This not only affects the Nyquist warping as shown in this thread, it also affects boost/cut symmetry and aliasing. It immensely affects the precision (and audible results) of dynamics processing.

Ironically, many digital EQs and compressors fail at covering their own parameter ranges. Most of the offered operational ranges are broken. None of them warns: "Be careful above 5kHz, the effective curves will then not even look close to what you'd expect". You also won't read: "At the end of the day, and due to our lack of antialiasing, pseudo analogue saturation won't be harmonic at all."