How do you judge an EQ?

[Christian Budde]

The test would be more complicated with analog EQs because it must take into account the whole audio signal path
(D/A conversion - out amp - audio cable - EQ - audio cable - in preamp - A/D conversion).

We should also think about the fact that when we use an external analog equalizer for mixing purposes on a digital system, we have to deal with the imperfections of the whole audio signal path, which sometimes can be worse than just blurred transients.

Every AD-converter on the market unfortunatly uses linear-phase downsampling filters (typical sigma-delta converters work with only a handful of bits, but at a very high frequency). You should additionally deconvolve the gear as well. So record IR of the chain with every cable, but (hard-)bypassed EQ and record the impulse response of that chain again, but with EQ in the signal path. Then you have to deconvolve the "overall IR" with the "signal chain IR". I'Ve written a tool for that, which does everything automatically (except the fact that you have to plug), but I havn't got hardware here to try it. I once did some measurement with the soundcraft desk in our studio and later with the small yamaha desk, but havn't found the time to check the results.

Kind regards,

Christian

P.S.: Now I have the tools, but no hardware, is someone willing to help?

[martian]

great information! you won't tell us which eqs were used for your examples? regarding transients, distortion should also play some role or am i wrong here?

....

i think it can be measured better with 1k sine in your sequencer and span, i noticed those things (+ Q) show odd distortion reading in christians analyser but have none when performed in cubase for example.

[martian]

eliosound, thanks very much for the shots.

not that im too concerned how a plug performs 10X in practice but im certianly interested so have tried this with waves, sx Q, posihfobit, gliss, and get the same as your first upper spike with all of them. hydratones a little different though, no transient smearing but the spike gets moved 10 samples forward.

so which eqs are you using, please

[Barbarossa]

Christian, danke zunächst für die ausführlichen Erläuterungen.

You mentioned dynamic convolution. It surely comes closer in modeling non-linearities, but i see a big problem, that can't be covered with the present computer power: correct me if i'm wrong, but today's dynamic convolution is applied always on the complete signal, depending on the signal's amplitude. If the amplitude is lower, then another IR is used, as if the amplitude were higher. So far so good. Different Arbeitspunkte with their different non-linear behaviour are covered by snapshots.
But the problem i see is, the signal is consisting of many different sine-waves, with all different amplitudes and they ony all together sum up to the signals amplitude.
So, to make this linearization exact, it would be necessary, to use for every in the signal contained sine-wave with it's unique amplitude the appropriate IR for this amplitude, and afterwards put it together again. But that would mean with a resolution of only 50Hz 400(!) parallel convolutions, instead of one, wouldn't it? Impossible for today's computer power.

[leban]

dynamic convolution is patented... there are many hidden implementations (and you don't need so many parallel convolutions, it weights about one 1/2, so it easy implementable), but it is patented since 1999
Sintefex rocks!!!!!

[Christian Budde]

i think it can be measured better with 1k sine in your sequencer and span, i noticed those things (+ Q) show odd distortion reading in christians analyser but have none when performed in cubase for example.

It's hard to hear odd harmonic distortion, that's why I create the analyser.

You mentioned dynamic convolution. It surely comes closer in modeling non-linearities, but i see a big problem, that can't be covered with the present computer power: correct me if i'm wrong, but today's dynamic convolution is applied always on the complete signal, depending on the signal's amplitude. If the amplitude is lower, then another IR is used, as if the amplitude were higher. So far so good. Different Arbeitspunkte with their different non-linear behaviour are covered by snapshots.
But the problem i see is, the signal is consisting of many different sine-waves, with all different amplitudes and they ony all together sum up to the signals amplitude.
So, to make this linearization exact, it would be necessary, to use for every in the signal contained sine-wave with it's unique amplitude the appropriate IR for this amplitude, and afterwards put it together again. But that would mean with a resolution of only 50Hz 400(!) parallel convolutions, instead of one, wouldn't it? Impossible for today's computer power.

It's not totally correct. I think you are mixing two different things here. Unfortunately I can't write every detail here, because I have some own thoughts about that and I want to see the thoughts written in a paper first to avoid other people of making a patent out of it.
Basically the sintefex idea is to do a time-domain convolution depending on the level of every input sample. Imagine an IR consisting of 256samples. That would make up 256 MACCs (multiplication & addition) only for the time domain convolution per each sample. That'll be 11289600 MACCs per second per channel. Not to forget the signal level comparision and impulse response blending. All together very heavy. And with that you can only emulate timeconstants of 256/44100 = 5.8ms (!). Typical tube based systems may have time constants of up to a second...

Kind regards,

Christian

[BertKoor]

Personally I think this discussion is better off in the DSP Developers forum instead of here in the Effects section. I think Poonna (the original starter of this thread) has lost interest after page 2.

[Christian Budde]

Personally I think this discussion is better off in the DSP Developers forum instead of here in the Effects section. I think Poonna (the original starter of this thread) has lost interest after page 2.

Oh, you're right, too much already. I'll shut up anyway, having some exams ahead.
Happy judging,

Christian

[defjamm]

P.S.: Now I have the tools, but no hardware, is someone willing to help?

i think that's the biggest problem some very talented dsp-engineers have...the guys who have the hardware often don't have the dsp-knowledge and the guys wo have the knowledge often don't have access to some great hardware.

maybe ask some studios where you live and explain your plans or ask a bigger german software-developer like steinberg or native instruments to team up with and make some money with a product that's better than the rest.

gearslutz.com should be a better place to find guys with great hardware in germany.

i've noticed that the german magazine 'sound&recording' uses your plugin-analyser. maybe you could get some contacts through them...should be really easy if they write an article on that topic.

@Zaphod (giancarlo): it seems you're working on a new convolution product? i hope you will make something really great and won't compromise too much quality for cpu.

[Oliver.Lucas]

Oh, you're right, too much already. I'll shut up anyway, having some exams ahead.
Happy judging,

Christian

Phew, these teachers....never understand how they slow down really important developments with their exams....

Only kidding....and...Christian, you are one of my heroes!

The Electri-q plug is just too good to be true!! Every now and then there's a little plug that you just love to pieces and Electri-q is one of 'em. TRUELY AMAZING.

There is a bug by the way in the EQ if you use a control surface and tweak the EQ type parameter with a knob. It is very likely to cause a crash. (Probably because you step through the different algos way too fast for the CPU to keep up with?)
Also, the values of the fequencies aren't nice to read out on control surfaces since the values are too long)

Keep up the fantastic work and all the best for the exams.

Ollie

[mammath]

There is a bug by the way in the EQ if you use a control surface and tweak the EQ type parameter with a knob. It is very likely to cause a crash. (Probably because you step through the different algos way too fast for the CPU to keep up with?)
Also, the values of the fequencies aren't nice to read out on control surfaces since the values are too long)

Keep up the fantastic work and all the best for the exams.

Ollie

Oliver.Lucas, you should report this in the Electri-Q v1.5 released thread as well:

http://www.kvraudio.com/forum/viewtopic.php?t=140373

[Eiosis]

Thank you eliosound. With a good example and a good english you explained very well what I meant.

You're welcome. We are here to discuss on things which are sometimes very hard to understand and to explain,
and I'm afraid that my english is not far better than yours !!
I believe that there's a lot of non native english speaking persons here.

great information! you won't tell us which eqs were used for your examples?

It would be kind, if you can tell us, what gear you've used.

eliosound, thanks very much for the shots. [...]
so which eqs are you using, please

I don't know if saying which EQ produced the upper impulse response and which one produced the second one is very interesting.
What I could say without upsetting anybody is that they are two software EQs, which run as plugins.
I'm not here to judge anybody's plugin or to give a point of view that may cause other people to judge a product on a single impulse response.
But if you really want to know...

P.S.: Now I have the tools, but no hardware, is someone willing to help?

i think that's the biggest problem some very talented dsp-engineers have...the guys who have the hardware often don't have the dsp-knowledge and the guys wo have the knowledge often don't have access to some great hardware.

I agree, that is one of the key points. I discovered a new world since the day I putted my hands on a high end analog EQ. So I began to compare analog versus digital EQs regarding purely technical points of view, then on the workflow point of view, and so on.. Now I mix using both digital and analog EQs, with different goals and applications for each one.

Even if great hardware is not essential to know how to make great dsp processing algorithms, it is a big plus.

Christian, I would be pleased to make some tests with my hardware analog EQs if you need me to do so.

Fabrice,
Eliosound

[bmanic]

so zaphod and bmanic, can you objectively demonstrate your transient issue? to clarify its digital biquad min phase eq comparison that is the subject matter here for me.

I don't know about digital biquad versus biquad. I personally think if correctly implemented there should be zero difference here. I'm mainly interested in any EQ versus any other EQ. I'm always just a hands-on/"I want results" kind of guy and I'd rather get quick results than try to fight with some plugin that COULD get the same results after understanding completely how it works and what one should do exactly to get those results. Sometimes it might be even impossible. I'm sure some of the magic of old analogue EQs is that they have been designed for specific tasks and frequency ranges that the human ear/brain likes. This is why I'm less interested in orange vs orange comparison but rather "EQ1 that offers bla bla bla bla" versus "EQ2 that offers bla bla bla". I like comparing equalisers that bring something to the mastering table as it's a specific task (some are for instance excellent at boosting the high frequencies without being harsh and some are excellent at cutting away offending frequencies at a narrow band).

I mean, try to get your hands on a pultec clone, I suggest the Tube-tech one but any of the other high end clones will do too. Then tweak it a bit on vocals, accoustic guitar, piano, or similar sound sources, to get the desired results. Now time this and see how long it takes. Last time I checked I got the results in about 15 seconds and was 99% satisfied with it. I've had huge trouble with plugin EQs of old days to get satisfying results (old logic 4.0 comes to mind and the built in Cubase VST eq). However, now that plugins have come much further I feel the difference is rapidly catching up.

furthermore i am of course open to the fact you may have much better hearing and monitoring, but i still wont care if i cant hear it. more power to you if you can which i wont doubt if the tests are objective.
thanks,

I totally agree with this. I'm not 100% certain if my subjective opinion is truly hearing but rather a "fantasy". It could be. However, I keep finding differences in my ABX testing and I can usually also pinpoint what I like about one of the examples over the other so there must be something more to it than simple placebo. Or maybe I'm just very very lucky?

I'll post some more examples as soon as I can. Unfortunately we are moving house with my wife so it's total chaos here! Sorry about that guys. I did however render a small demo of Wavelab/Nuendo Q eq sweep versus "mystery eq" sweep in the same range. This is of course again an orange versus apple comparison but it does show the difference in "accuracy" and how transients are affected. I'll post this right away, once I've cleaned up my web space a bit!

Cheers!
bManic

[bmanic]

dynamic convolution is patented... there are many hidden implementations (and you don't need so many parallel convolutions, it weights about one 1/2, so it easy implementable), but it is patented since 1999
Sintefex rocks!!!!!

Sintefex are sneaky bastards, that's more like it.

I'm still suprised that they were able to pantent it world wide. Then again, people have been able to patent human genes so it seems everything is possible!

Cheers!
bManic

[bmanic]

Basically the sintefex idea is to do a time-domain convolution depending on the level of every input sample. Imagine an IR consisting of 256samples. That would make up 256 MACCs (multiplication & addition) only for the time domain convolution per each sample. That'll be 11289600 MACCs per second per channel. Not to forget the signal level comparision and impulse response blending. All together very heavy. And with that you can only emulate timeconstants of 256/44100 = 5.8ms (!). Typical tube based systems may have time constants of up to a second...

.. and the FX8000, flagship 8 channel model calculates 8 times 2048 sample long impulses at 48kHz. It DOES have a shit load of sharc DSPs in it (over 40 of them).

Cheers!
bManic