Digital EQ Fact & Myth.

[Lagerfeldt]

Very interesting thread that sparked a two pages thread on my own forum.

How about filtering ringing, which can differ even with minimum phase equalizers?

How can they sound the same if the filter ringing is so different?

EQ Artifacts

Just a couple of examples:


Very short but some amplitude.


Lower ampltiude but much longer ringing.




Linear phase of course can and will have pre-echo but that's not the discussion here.

[zmix]

Nice one, Holger!

[Lagerfeldt]

Just came to think of this, and I'm genuinely interested in getting a real scientific explanation of how this affects the other findings in this thread.

My own tests concluded that I can get most minimum phase EQs sounding equal down to around -80 dBFS something if I tweak to 0.01 decimals.

I think I can safely conclude that linear phase/minimum phase apart, the biggest difference in well designed digital equalizers are the functions available and the interface. I like the 2x, Inverse, A/B, morphing slider, multichannel and M/S options in the Flux Epure - but it's apparently quite similar in sound to most other well designed digital equalizers.

So are the functions and interface worth the extra price? Yes, I think so.

But I'll be more hesitant in recommending beginners to buy a new plug-in equalizer. Instead they should spend it on dynamic plug-ins or just save up for a hardware equalizer instead.

Talking of dynamics, I was quite succesful in emulating the Waves SSL master bus compressor using the new circuit type emulation in Logic 8's compressor, but that's another story ;-)

[David Rick]

Originally Posted by Lagerfeldt

My own tests concluded that I can get most minimum phase EQs sounding equal down to around -80 dBFS something if I tweak to 0.01 decimals.

The residual difference is undoubtedly due to variations in internal architecture, leading to different residual errors at the output. From the standpoint of user-accessible controls, both EQ's are probably attempting to do the same thing, but one is doing it more perfectly than the other. Still, the differences are very small, and probably only a mastering engineer would care. For day-to-day mixing, the EQ's are basically equivalent, and you should pick whichever one is more comfortable to use.

Now if you begin to use extreme settings, you may find that the two EQ's in question behave rather more differently. That's when internal architecture starts to matter, because some internal nodes may operate at considerably more gain than is seen at the output, and limitations on internal precision may become audible when Q and/or Gain are pushed to extremes.

Earlier you wrote:

How about filtering ringing, which can differ even with minimum phase equalizers?

How can they sound the same if the filter ringing is so different?

In this case, you have two EQ's that aren't set the same, whatever the markings on the user controls say. An ideal EQ is a linear device, and if it is minimum phase then there is a one-to-one correspondence between impulse response and frequency response. So if you see that the impulse response is different, then so is the frequency response.

Why wasn't the difference audible in your case? Well, maybe the extra "hang-over" in the impulse response was masked by something else in the music. Or (if you were actually listening to the test signal), maybe the difference was simply below your psychoacoustic "just noticeable difference limen".

David L. Rick

[David Rick]

Originally Posted by Luny Tune

Absolutely, David. Don't hold back!!

Regarding your "harmonics EQ"... A sort of "parametric exciter"...? Maybe it could work fine on a recording in poor shape and even add emphasis to a recorded signal already in good shape. For certain applications, in short. It sounds to me like an "eq" for saving signals rather than for making what's already good better.

Yes, I imagine it would find use primarily as a mastering tool. Got a track with a muddy bass line? Add some harmonics just to the bass to make it stand out better.

At the mix stage, you'd obviously have lots of options to process the bass track alone, and a conventional saturation plug in (or a tried and true slow attack compressor) could end up accomplishing the same thing.

But for folks like me who do a lot of "live to two-track" work, tools that only work on an isolated instrument or vocal aren't very useful. I don't bill myself as a mastering engineer, but much of my post-event audio production looks a lot like mastering.

Speaking of radical "EQ" plug ins that are useful for rescuing bad recordings in mastering, I've got to put in a plug for Duane Wise's "Dynamic Parametric EQ" tool (Quartet DynPEQ, distributed by Sonic Studio). To understand what this thing really is, look at this explanation on Duane's web site. In the demo I heard, he took a really bad jazz trio recording and turned it into something you might actually want to broadcast. Too bad we have to wait for the VST version!

David L. Rick
Seventh String Recording

[rhythminmind]

Originally Posted by space2012

do a simple test,
import a .wav song from a cd, that is super recorded/mastered

duplicate that song/track
use native eq in A, another "SSL" in B,
do your magic,
then change the phase in one track,
if what you say its true, they should cancel 100% and nothing will come out in the master channel.
absolute silence.

This is just what i did check page 1 of this thread.

[cjogo]

I have went through my racks of analog/ tube EQ's, in the past 30 years ---ended up with Behringer DEQ 2496 and the Channel strip in our DAW > simple ~affordable & clean Keeps the cost down for the clients.

[chrisj]

Originally Posted by David Rick

And to take the discussion in another direction, it seems like people value certain old hardware EQ's for sonic characteristics (distortion) some of which happen even with no boost or cut. Well, suppose you could "boost" a particular frequency region, but the plug in didn't really boost it -- it just created more 2nd or 3rd harmonics (but only for stuff in the "boosted" frequency region). Would that be useful. Does somebody already make such a plug in?

Yes, I do. Airwindows Audio Unit Plugins

But it really isn't QUITE what you're talking about. What it's doing is taking advantage of the fact that nonlinear transforms like distortions produce big changes in apparent loudness and also apparent depth/distance without altering the hottest peaks.

So you get to pick a cutoff, and then alter the bands above and below this- but rather than applying gain changes you're applying positive or negative saturation. If you're boosting it's obviously creating harmonics, though not in a eventide-harmonizer way- and the end result is not that additional harmonics are apparent, but that the band seems to come forward or drop back in the sound image.

Actually this one's due for an upgrade (free to existing customers btw) to turn it from two bands into three bands, because that would be more useful. The two band one is all too much like proof of concept and while it works, it hasn't been exciting people. A three band one would be capable of crazy great stuff like taking a track and dropping the mids back a bit for some flashy hi-fi-ism, or focussing them- and the basic algorithms are unusually good sounding though not very flexible.

Anyway, yes, I did one of those it does what you say, it just doesn't isolate it so that the only change is adding harmonics. They become a major artifact of how apparent gain is changed for a band, and are intentional, but you hear them as closeness or farness and not as harmonics.

[David Rick]

Chris, can you explain what you mean by "negative saturation"? I tried to understand how to cut with this kind of algorithm, and didn't get anywhere. Canceling existing harmonic content isn't easy, because we don't know its phase relationship to the fundamental.

David

[chrisj]

It's not directly harmonic content, it's just an effect of the transfer function. Like you can produce harmonics (when you input a sine wave) using Chebyshev functions... inverse saturation is just like taking a wave and saturating it, but backwards. The peaks remain hot but the quieter places on the transfer function become even quieter. The reason people don't do this is it sounds like crap, but it also causes the apparent sound source to appear farther away- has some uses as a building block. I'm going to revise Shelves and hopefully I can produce demos that will make this more apparent. I'm picturing doing a mid-suck effect using this- ought to do nice things.

[Kenton]

I like the idea of a plugin that would allow me to control how much my mids suck.
Sounds like you're onto something Chris...............

[Dan]

Originally Posted by chrisj

It's not directly harmonic content, it's just an effect of the transfer function. Like you can produce harmonics (when you input a sine wave) using Chebyshev functions... inverse saturation is just like taking a wave and saturating it, but backwards. The peaks remain hot but the quieter places on the transfer function become even quieter. The reason people don't do this is it sounds like crap, but it also causes the apparent sound source to appear farther away- has some uses as a building block. I'm going to revise Shelves and hopefully I can produce demos that will make this more apparent. I'm picturing doing a mid-suck effect using this- ought to do nice things.

That would be useful for midi orchestral tracks, and other stuff.

[jdtrbn]

Originally Posted by chrisj

It's not directly harmonic content, it's just an effect of the transfer function. Like you can produce harmonics (when you input a sine wave) using Chebyshev functions... inverse saturation is just like taking a wave and saturating it, but backwards. The peaks remain hot but the quieter places on the transfer function become even quieter.

How do you remove e.g. second harmonics? There isn't a transform function that would map a value to the value before saturating uniquely. I.e. such inverse saturation doesn't exist. For odd harmonics it's an interesting thought.

[Paul Frindle]

Originally Posted by chrisj

It's not directly harmonic content, it's just an effect of the transfer function. Like you can produce harmonics (when you input a sine wave) using Chebyshev functions... inverse saturation is just like taking a wave and saturating it, but backwards. The peaks remain hot but the quieter places on the transfer function become even quieter. The reason people don't do this is it sounds like crap, but it also causes the apparent sound source to appear farther away- has some uses as a building block. I'm going to revise Shelves and hopefully I can produce demos that will make this more apparent. I'm picturing doing a mid-suck effect using this- ought to do nice things.

Reading this prompts me to say that such techniques are in use in the Inflator plug-in when the curve is et to -ve.

Also much of the basic sound of many class B tube designs can be attributed to this effect as well. One of the tricks in designing these amps was to get the bias value to produce this effect to exactly the best degree. It seems to increase the punch by actually expanding the dynamic range on an immediate (non time related) basis. But of course it produces distortion as well - over most of the signal range.

Anyway you can make a freq response version of this by using an EQ with the right bump response feeding the inflator then mixing it back with the original signal that has an EQ in exact symmetrical cut. You can make some very interesting and subtle sounds like this :-)

[lukasbrooklyn]

Originally Posted by David Rick

Gee, I guess I put everyone to sleep with nerdy DSP stuff before I got to this part:



And to take the discussion in another direction, it seems like people value certain old hardware EQ's for sonic characteristics (distortion) some of which happen even with no boost or cut. Well, suppose you could "boost" a particular frequency region, but the plug in didn't really boost it -- it just created more 2nd or 3rd harmonics (but only for stuff in the "boosted" frequency region). Would that be useful. Does somebody already make such a plug in?

David L. Rick
Seventh String Recording

voxengo harmoniEQ i believe. Voxengo

[fader8]

Originally Posted by David Rick

In this case, you have two EQ's that aren't set the same, whatever the markings on the user controls say. An ideal EQ is a linear device, and if it is minimum phase then there is a one-to-one correspondence between impulse response and frequency response. So if you see that the impulse response is different, then so is the frequency response.

Hi David,
That was my exact assumption before I made the measurements that Holger is referencing above. I can assure you that this assumption is incorrect. In fact, it was because of differences in audible artifacts I was "hearing" that prompted me to make these measurements in the first place. In particular, Logic's Channel EQ. Many have heard that one "sing". There must be "something" these designers are doing that creates this.

Anyway, if you check my site you'll see that I set up the same filter response for all measurements. In fact, I made sure the difference in the transfer functions were within 0.1dB. Yes, it took a while to set all these up, but I was a man on a mission!

Here's my EQ page.

Originally Posted by David Rick

Why wasn't the difference audible in your case? Well, maybe the extra "hang-over" in the impulse response was masked by something else in the music. Or (if you were actually listening to the test signal), maybe the difference was simply below your psychoacoustic "just noticeable difference limen".

Agreed! The audibility of these artifacts is extremely dependent on signal content and are easily masked in the presence of sustained waveforms. However, with the right transient or fast gated signal, it becomes entirely audible.

Interestingly, even though the Flux EQ exhibits worse measured characteristics than many, it is still no where near the Logic Channel EQ in audibility. Generally speaking, it's a great sounding EQ. As I mention on the page, these measurements shouldn't be used as a sole criteria for selecting an EQ, but it is important to keep these characteristics in mind for mix troubleshooting.

fader8

[David Rick]

fader8,

Getting to the bottom of this might reveal something interesting about one or both of these equalizers. Are you game?

Normal scientific method would be for someone to try to replicate your test and show the same result. Unfortunately I can't help, since I own neither Logic nor Waves licenses. So I suggest that you try some different test methodologies to see if the results match.

The tone bursts you used are rectangularly windowed, which means they are causing a lot of spectral splatter. To control this, you could use a more gradual window. Common choices are Cosine and Blackman windows.

Since you are good at making two-channel measurements, how about making a direct comparison between the Waves and Logic linear phase eq's (set as in your previous test) to verify that the difference curve is indeed flat?

Next step would be to measure the impulse responses of both products and compare the results. I don't know if spectrafoo can do this, but ARTA can.

David L. Rick

[Paul Frindle]

Originally Posted by David Rick

fader8,


The tone bursts you used are rectangularly windowed, which means they are causing a lot of spectral splatter. To control this, you could use a more gradual window. Common choices are Cosine and Blackman windows.

Next step would be to measure the impulse responses of both products and compare the results. I don't know if spectrafoo can do this, but ARTA can.

David L. Rick

This is correct :-) And the other thing to remember is that the linear phase EQs themselves must also be windowed in the same way - as they give rise to a potentially infinite preamble series in time.

The fact that the whole time series can never be represented in total will affect the freq response too - so it's a good idea to look at those responses at very tight accuracy using a slowly swept sine wave, to see if they are indeed the same.. You might have a surprise!

You might also consider that there are various ways of making a steady state response with varying degrees of 'ringing'. In fact you can actually make a steady state flat response system that still exhibits ringing at a single freq band!

[fader8]

Originally Posted by David Rick

Getting to the bottom of this might reveal something interesting about one or both of these equalizers. Are you game?

Hey, I'm game for anything that sheds light on the subject!

Originally Posted by David Rick

The tone bursts you used are rectangularly windowed, which means they are causing a lot of spectral splatter. To control this, you could use a more gradual window. Common choices are Cosine and Blackman windows.

OK, I think that you guys might be misunderstanding my test procedure.

In either Logic or Pro Tools, I send pink noise to a bus and pick it up on two mono aux channels. So each aux is receiving an identical signal.

Each aux is sent into its own mono output. I tap into those channels (via RME Totalmix, very useful) and they feed Spectrafoo's 2 analyzer channels.

In the transfer function window, magnitude and phase null completely. The analyzer is set to use a Blackman window, btw.

OK, back in the host, I insert an EQ on one of the auxes and after making sure any plug-in delay compensation is perfect, I adjust the EQ to match a common reference curve I have saved in Spectrafoo. I can zoom the window full screen to see that I'm easily matching the new response curve to my reference curve within 0.1dB, and in most cases better. I'm using a high frequency resolution, slow integration time. By the way, I've done this with a slow sweep and I get the same results but I use pink noise as it is closer to a musical signal in dynamic characteristics than a sweep ever could be.

OK, I turn off the pink noise and paste my recording of a 700Hz 2-cycle burst onto a track. Then I bounce that through the EQ I've just adjusted to my reference curve.

The resulting waveforms you see on my page are simply waveform displays from a sample editor of the bounced file. So you see, I'm not simply pumping a burst into a filter and windowing the output with an FFT. This is actual real-world, in-situ application of the EQ plug-in.

In other words, what you see is what you get when you employ that plug-in with your host.

Originally Posted by David Rick

Since you are good at making two-channel measurements, how about making a direct comparison between the Waves and Logic linear phase eq's (set as in your previous test) to verify that the difference curve is indeed flat?

I think I can do that as the files bounce out with the same timing and sample count. I'll give it a shot.

With all that said, understand that I'm not defending these measurements as the be-all end-all definitive test for EQ's. I do have a quite a few years under my belt doing audio and acoustic testing, but I'm not a DSP engineer by any stretch. But I do enjoy exploring the finer elements of this stuff!

[fader8]

Originally Posted by Paul Frindle

You might also consider that there are various ways of making a steady state response with varying degrees of 'ringing'. In fact you can actually make a steady state flat response system that still exhibits ringing at a single freq band!

Would that be a linear phase all-pass filter then?

FYI Paul, I'm saving my sheckles for your DSM plug. Hurry up with the AU version, will ya? Very excited about that plug.

[Paul Frindle]

Originally Posted by fader8

Would that be a linear phase all-pass filter then?

In a way - yes. But you can make an all-pass 'almost anything you like' - it does not have to be so called linear phase.

I made just such an object in my early days at SSL (almost by accident), when I was struggling to find a way of getting the excess noise out of assignable analogue EQ circuits for a proposed new project. The idea was that a feedback loop of all-pass sections would exhibit less noise than the usual state variable circuits - because the loop gain would remain the same at all freqs regardless of setting etc..

Anyway, fiddling with the all-pass resonating 'flat response EQ' was quite fun as it had some pretty 'interesting' sonic properties - not at all for the faint hearted though - LOL!

The fact is that you can make most anything you can imagine - but whether this produces a useful sound is another matter entirely of course.

BTW - I do fully understand the test set-up described - and the impulse response test proposed by David is still a very good idea indeed - if you want to find out what's really going on :-)

FYI Paul, I'm saving my sheckles for your DSM plug. Hurry up with the AU version, will ya? Very excited about that plug.

The AU version was released couple of weeks ago :-)

[sage691]

Originally Posted by rhythminmind

I would like to get some technical examples of EQ algorithm differences. I understand compression algorithms. These can be drastically different. How one digital EQ is actually better then another gets fuzzy. I'm not talking about features or interface but audio integrity.

I have recently switched to sonar 7PE. This has made me reevaluate my plugin arsenal.
I have been using Christian's plugin analyzer to compare my EQ's
Christian’s Blog » Programs

I have been comparing the
Duende channelstrip
urs CSP
Sonars Native mixer EQ
VC-64 (kjaerhus audio GAC-1 port)
& LP-64

The EQ variables that i am aware of are
frequency response
filter slope / Q
& phase

I have been able to recreate the the slope, phase, & sonic characteristic of the Duende & CSP with built in native EQ . Or with any fully variable parametric EQ for that matter.
As i said compressors are a different story. Duendes program dependent comp is it's own beast

Besides linear phase, internal bit resolution, or to free cpu with external DSP i'm now questioning why anyone needs anything but there native EQ.
I can't find technical reasons.
Please inform.

BECAUSE THE EXPENSIVE PLUGIN EQs HAVE MORE ELABORATE, COLORFUL GRAPHICS!!! THAT'S WHY THEY COST SO MUCH MORE!!!

[fader8]

Originally Posted by Paul Frindle

In a way - yes. But you can make an all-pass 'almost anything you like' - it does not have to be so called linear phase.

Ha, that was meant to be a joke. (A geeky joke, nonetheless.) I guess a linear phase all-pass would simply be a delay!

Originally Posted by Paul Frindle

Anyway, fiddling with the all-pass resonating 'flat response EQ' was quite fun as it had some pretty 'interesting' sonic properties - not at all for the faint hearted though - LOL!

The fact is that you can make most anything you can imagine - but whether this produces a useful sound is another matter entirely of course.

Well, I've used them a good bit in Kyma for various oddball stuff, but I recently picked up DDMF's IIEQ Pro plug-in and as a result have been using it in mixing. It allows any band to be config'd as an all-pass and you can serial or parallell wire the bands. Chain a few together and shove it in a duplicate reverb return channel . . . .

Or, just enough to smear a snare transient. It can knock off a couple dB as the peaks aren't so additive anymore.

Originally Posted by Paul Frindle

BTW - I do fully understand the test set-up described - and the impulse response test proposed by David is still a very good idea indeed - if you want to find out what's really going on :-)

I'll get back to this stuff in a few days. Can't stop what I'm doing right now, (except of course to procrastinate on web forums!)

Originally Posted by Paul Frindle

The AU version was released couple of weeks ago :-)

Doh! You can't do that. I don't have enough sheckles yet!

[ericwall]

Bump!

This thread doesn't deserve to sink into oblivion yet IMHO.

I haven't read every single word I must admit, and I got lost somewhere around the talk about "negative distorsion" but is there a general consensus still that all digital EQ's sound the same ???
Wow. In that case it's Scoop of the year if you ask me, and in that case my ears have once again proved that I should trust no one but them. I admit I'm an ITB/UAD/YADA YADA sceptic, but still:

Is just noise really enough to make us believe we've got analog magic going on in our ITB-mixes?

Anyone got the UAD Neve EQ's and care to make a null test against some stock EQ's?

Something tells me we'll be discussing some manufacturer's advertising-methods if this is the case.

/EW

[rhythminmind]

Originally Posted by ericwall

Bump!

but is there a general consensus still that all digital EQ's sound the same ???

/EW

Thats a yes from me. You can add Oxford EQ to the list. Oxford was tricky with it's scaling Q's. I would love AES to standardize Q settings. Check the .jpg. FYI people should run there tests at 96k so not to be mislead by oversampling EQ's.

[jdtrbn]

Originally Posted by LPK

The Analog Button on the Waves plugins does exactly one thing:

It adds noise, and nothing else (at least for API if I remember correctly, and definetly for the SSL EQ)
I've done the nulltest, and yes, it's just that.
No saturation, no this, no that, no nothing .. just noise.

Actually an analyzer software shows that there will also be more harmonics. If I remember right the normal mode gives you the first 4 harmonics and the analog mode gives you some higher than that. Usually higher order harmonics aren't that pleasant, so it's understandable that there's a separate mode for noise and that (though noise and higher order harmonics are very low in level).

[EstateMatt]

hrm.... this gives me an idea.

[EstateMatt]

Originally Posted by Lagerfeldt

Very interesting thread that sparked a two pages thread on my own forum.

How about filtering ringing, which can differ even with minimum phase equalizers?

Lager, Is there anyway to see these artifacts using a pulse as an input?

[rhythminmind]

Even my beloved Duende channel. The numeric value's are nowhere near the same. I had to go blind to get the plots to match. This is most likely why the duende stands out to my ears. Well that and the compressor =). But once again a Digital parametric EQ is a Digital parametric EQ.

[rhythminmind]

Some more for ya