Any EQ Pros in the crowd? How sharp can EQ shaping be in Equalizer APO?

[Fraxo]

Is it possible for the Graphic EQ in Equalizer APO to perform such hard cuts and sharp shapes?
I had a guy insisting, saying it's not possible...
As it looks - Equalizer APO's Graphic EQ with Variable Bands shows the possibility to perform any type of shaping, as it's represented by a straight line and you can freely set points that stretch that line in a straight form, so no curves, bells, smoothened edged etc... As simple as it gets (visually), point to point connect the dots without limitations.

He claimed it defies the rules of physics or requires infinite blah blah and that there's no way it actually sounds like what it shows, that sharply and accurately. So although it sounds silly to me - I won't claim otherwise without researching so here it is...
May I pick your brains for some explanations on this matter?
Any test that could be proving one way or another?

Highly appreciated

 

[Joachim Herbert]

Your friend is right. Measure to find out.

 

[ernestcarl]

Regardless of how many adjustable points there are, there’s always a Q. Take an actual before and after measurement. It may be easier to visualize EQ predictions through REW’s EQ module or with rePhase — in the latter you can apply hundreds of PEQ points.

 

[charleski]

You’re basically asking if there are limits on the slope of the filters used. As far as I understand the docs, EqualizerAPO uses minimum phase IIR filters. The slope is then dependent on the order of the filter, which can be arbitrarily large. You can add as many filter taps as your computational hardware can support. The downside is latency, as the long filters needed for a massively high-order filter mean the audio output gets delayed a corresponding amount.

Amir’s review of the Chord M-Scalar shows a high-order filter that achieves a very steep slope.

 

[Fraxo]

Your friend is right. Measure to find out.

What would be a good way to measure to prove or disprove it?
Something undeniable.... Please suggest, as I'm willing to try.

 

[Joachim Herbert]

Just dial in your eq and measure fr.

 

[dasdoing]

What would be a good way to measure to prove or disprove it?
Something undeniable.... Please suggest, as I'm willing to try.

or you need a internal loopback in an interface,
or you could loopback fisicly conecting output to input,
or you use a "virtual cable" software.

you than just meassure the output in REW

 

[charleski]

Here you go. I installed EQAPO on a VirtualCable input and measured a sweep in REW:

The result is very similar to that predicted by the analysis panel in the editor.
You get a better stopband by giving the filters slightly more leeway:

Again, this matches what's seen in the analysis panel.

 

[fpitas]

Those sharp-cutoff filters are going to ring forever....they'll also have a tremendous peak in group delay right before they head down, which is related.

 

[tomtoo]

What should be the need for such sharp filters in audio?

 

[fpitas]

What should be the need for such sharp filters in audio?

I'll turn it around and say if you need anything like that in your system, it has huge problems, probably insoluble. And if it didn't have big problems before, it will now!

 

[charleski]

What should be the need for such sharp filters in audio?

Well there's antialiasing of course, as alluded to previously. There's not much use for them in the audible band, apart from perhaps a very narrow notch filter to take out a bad ground loop in a recording.

This is what the group delay looks like for the 2nd filter I used:

Since this is minimum phase, phase shift is way off the charts.

 

[Cars-N-Cans]

What would be a good way to measure to prove or disprove it?
Something undeniable.... Please suggest, as I'm willing to try.

Like others have said, do a loop-back with REW using the input on the soundcard, or a line-in, microphone input, etc. I think most cards should be ok with having the cable directly hook back to the input, but I did have a cheaper SoundBlaster card that I had to use my TASCAM recorder as a buffer to stop it from misbehaving. Start with the volume at 0 and then check the levels as you gradually increase it until REW finds the level to be acceptable. I would start with a flat response at first and make sure its actually flat and not ringing or doing something else unpleasant in the actual measurement. If that looks ok, then go thru the calibration procedure before taking any measurements. I think the users manual for REW outlines this fairly well if I recall.

For what its worth, you would be surprised at what you can catch with a loopback if you suspect something is not working properly. With the desired corrections in place I would definitely be looking at the group delay as well if you are using headphones. Having it wander around drunkenly like what is shown charlski's post would not be too swift. Probably ok at lower frequencies, but above 1 kHz where the spatial information is it could potentially corrupt the imaging if its too extreme. Just my 2¢.

 

[Fraxo]

So I actually don't need go that extreme, I usually create rather smoothened shapes, but drawing them this way is easier for me when accuracy is necessary.

 

[dogelition]

As far as I understand the docs, EqualizerAPO uses minimum phase IIR filters.

It does, but not for the Graphic EQ. From my limited understanding of DSP and looking at the code: I believe it takes the (magnitude) frequency response that you draw, derives the imaginary frequency response with the specified magnitude and minimum phase, and then does the filtering via multiplication in the frequency domain (which is equivalent to convolution in the time domain). So, you really get the magnitude response that you draw, but the result might not be pretty in the time domain (ripples etc.).

Please correct me if I'm wrong on any of this

 

[charleski]

It does, but not for the Graphic EQ. From my limited understanding of DSP and looking at the code: I believe it takes the (magnitude) frequency response that you draw, derives the imaginary frequency response with the specified magnitude and minimum phase, and then does the filtering via multiplication in the frequency domain (which is equivalent to convolution in the time domain). So, you really get the magnitude response that you draw, but the result might not be pretty in the time domain (ripples etc.).

Please correct me if I'm wrong on any of this

The code has virtually no comments, so I didn’t try to look at it too hard, it it seems he does a Z-transform, convolves, and then Z-transforms back.

 

[Fraxo]

It does, but not for the Graphic EQ. From my limited understanding of DSP and looking at the code: I believe it takes the (magnitude) frequency response that you draw, derives the imaginary frequency response with the specified magnitude and minimum phase, and then does the filtering via multiplication in the frequency domain (which is equivalent to convolution in the time domain). So, you really get the magnitude response that you draw, but the result might not be pretty in the time domain (ripples etc.).

Please correct me if I'm wrong on any of this

The code has virtually no comments, so I didn’t try to look at it too hard, it it seems he does a Z-transform, convolves, and then Z-transforms back.

Hey guys, thank you for the knowledge sharing.
You've mentioned the time domain not being pretty in such manipulation, but I'm not sure if it's due to the difference in EQ forms or due to the sharpness only.
To better specify my question: if I were to draw an almost identical eq shape with the shown EQ APO Graphic EQ, a random parameteic EQ and also with let's say Fabfilter Pro Q3. Would there be a significant (or any) difference in the phase and time domain? Anyt other difference in result? I mean, it's possible to shape the dB/oct pretty aggressively (up to 90dB/oct) to create sharp shapes, so I wonder if the actual EQ methods (IRR / Butterworth / anything else I'll regurgitate without truly understanding) affect the time domain differently although it appears to be an exact same EQ shaping.

 

[charleski]

Hey guys, thank you for the knowledge sharing.
You've mentioned the time domain not being pretty in such manipulation, but I'm not sure if it's due to the difference in EQ forms or due to the sharpness only.
To better specify my question: if I were to draw an almost identical eq shape with the shown EQ APO Graphic EQ, a random parameteic EQ and also with let's say Fabfilter Pro Q3. Would there be a significant (or any) difference in the phase and time domain? Anyt other difference in result? I mean, it's possible to shape the dB/oct pretty aggressively (up to 90dB/oct) to create sharp shapes, so I wonder if the actual EQ methods (IRR / Butterworth / anything else I'll regurgitate without truly understanding) affect the time domain differently although it appears to be an exact same EQ shaping.

You can see the time-domain differences between the varying filter types here:

11.3. Common IIR filters — Digital Signals Theory

Butterworth and Chebyshev Type II are more 'front-loaded' and have a higher initial peak in the ringing which dies away faster. Since this a single cycle I very much doubt that the difference is audible.