Sign in with
Sign up | Sign in
Your question

Ear Training: group delay?

Last response: in Home Audio
Share
Anonymous
January 31, 2005 1:40:50 AM

Archived from groups: rec.audio.pro (More info?)

Hi All:

Was reading an article in the latest issue of mix that talks
about digital EQ and distortions associated with IIR filters
and such. Now, I've heard a lot of talk in this news_group
about group delay artifacts associated with analog (and
IIR) EQ but don't know that I can identify said artifacts.
How can one train their ears to recognize or perhaps
even quantify these group delay effects?

Later...

Ron Capik
--
January 31, 2005 1:40:51 AM

Archived from groups: rec.audio.pro (More info?)

Take a two way speaker and move the tweeter a few inches closer or
farther from you compared to the woofer. Whatever effect you think you
may hear is similar to what an IIR filter might cause.

Group delay variation (in the audio range) would be quantified as the
change in time delay in ms for one frequency compared to another.
Mark
Anonymous
January 31, 2005 6:10:05 AM

Archived from groups: rec.audio.pro (More info?)

Mark, are you sure that your speaker maneuver gives a fair
approximation of group delay? There's a crossover region between the
frequency range of any two drivers, in which both are active. When you
move one of them relative to the other, then in that range of overlap
you're changing the phase relationship of two sound sources, and thus
varying the pattern of reinforcement vs. cancellation by rather large
amounts. It seems to me that the frequency response effects (several
dB) would be far more pronounced than the time dispersion effects.
--best regards
Related resources
Can't find your answer ? Ask !
Anonymous
January 31, 2005 12:59:53 PM

Archived from groups: rec.audio.pro (More info?)

Ron Capik <r.capik@worldnet.att.net> wrote:
>
>Was reading an article in the latest issue of mix that talks
>about digital EQ and distortions associated with IIR filters
>and such. Now, I've heard a lot of talk in this news_group
>about group delay artifacts associated with analog (and
>IIR) EQ but don't know that I can identify said artifacts.
>How can one train their ears to recognize or perhaps
>even quantify these group delay effects?

Little Labs makes a box with an all-pass network in it, that you can
use to add measured amounts of group delay in any band you want. They
sell it for reducing comb filtering problems, something it does very well,
but you could probably use it as an ear-training tool as well.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."
January 31, 2005 1:00:36 PM

Archived from groups: rec.audio.pro (More info?)

OK, I take your point... the speaker experiment will also create freq
response errors in the crossover region and these freq response errors
will mask any real or imagined effect due to group delay distortion. I
agree.

I do have a small disagreement however with a point in the second
article. It seems to be describing a digital FIR filter and the
article says all EQ filters have group delay distortion. FIR filters
can be designed to create any desired (within reason) frequency
response and at the same time have no group delay distortion at all.
IIR filters however are more like standard analog filters (minimum
phase). However, I agree with the conclusion, that group delay
distortion on the order of one wavelength is not audible in any case.
(unless the sound is later combined with an un-delayed version in which
case constructive and destructive interference that creates freq
response errors.)

Now if you take it to an extreme however, GD will be audible as
described in the article. If you imagine a system with 1 second of
additional delay to high frequencies compared to low, then you pass a
click through it, you'll heat a boom then a tick. If the delay is
large enough, the delayed sound is heard as a separate sound, similar
to an echo. But if the delay is small, less than 1 wavelength as would
be the case for any EQ filter, the delay is too small for the sound to
be heard separately.

Mark
January 31, 2005 1:07:20 PM

Archived from groups: rec.audio.pro (More info?)

Ethan Winer wrote:
> David,
>
> > It seems to me that the frequency response effects (several dB)
would be
> far more pronounced than the time dispersion effects. <
>
> Bingo. Here's an article that addresses exactly this issue:
>
> http://www.ethanwiner.com/phase.html
>
> And this one talks about the role of phase shift in equalizers:
>
> http://www.ethanwiner.com/EQPhase.html
>
> --Ethan


OK, I take your point... the speaker experiment will also create freq
response errors in the crossover region and these freq response errors
will mask any real or imagined effect due to group delay distortion. I
agree.

I do have a small disagreement however with a point in the second
article. It seems to be describing a digital FIR filter and the
article saya all EQ filters have group delay distortion. FIR filters
can be designed to create any desired (within reason) frequency
response and at the same time have no group delay distortion at all.
IIR filters however are more like standard analog filters (minimum
phase). However, I agree with the conclusion, that group delay
distortion on the order of one wavelength is not audible in any case.
(unless the sound is later combined with an undelayed version in which
case constructive and destructive interference that creates freq
response errors.)

Now if you take it to an extreame however, GD will be audible as
described in the article. If you imagine a system with 1 second of
additinal delay to high frequencies compared to low, then you pass a
click through it, you'll heat a boom then a tick. If the delay is
large enough, the delayed sound is heard as a seperate sound, similar
to an echo. But if the delay is small, less than 1 wavelenth as would
be the case for any EQ filter, the delay is too small for the sound to
be heard seperatly.

Mark
January 31, 2005 1:09:05 PM

Archived from groups: rec.audio.pro (More info?)

Ethan Winer wrote:
> David,
>
> > It seems to me that the frequency response effects (several dB)
would be
> far more pronounced than the time dispersion effects. <
>
> Bingo. Here's an article that addresses exactly this issue:
>
> http://www.ethanwiner.com/phase.html
>
> And this one talks about the role of phase shift in equalizers:
>
> http://www.ethanwiner.com/EQPhase.html
>
> --Ethan


OK, I take your point... the speaker experiment will also create freq
response errors in the crossover region and these freq response errors
will mask any real or imagined effect due to group delay distortion. I
agree.

I do have a small disagreement however with a point in the second
article. It seems to be describing a digital FIR filter and the
article says all EQ filters have group delay distortion. FIR filters
can be designed to create any desired (within reason) frequency
response and at the same time have no group delay distortion at all.
IIR filters however are more like standard analog filters (minimum
phase). However, I agree with the conclusion, that group delay
distortion on the order of one wavelength is not audible in any case.
(unless the sound is later combined with an un-delayed version in which
case constructive and destructive interference that creates freq
response errors.)

Now if you take it to an extreme however, GD will be audible as
described in the article. If you imagine a system with 1 second of
additional delay to high frequencies compared to low, then you pass a
click through it, you'll heat a boom then a tick. If the delay is
large enough, the delayed sound is heard as a separate sound, similar
to an echo. But if the delay is small, less than 1 wavelength as would
be the case for any EQ filter, the delay is too small for the sound to
be heard separately.

Mark
Anonymous
January 31, 2005 2:00:36 PM

Archived from groups: rec.audio.pro (More info?)

Mark wrote:

> Now if you take it to an extreme however, GD will be audible as
> described in the article. If you imagine a system with 1 second of
> additional delay to high frequencies compared to low, then you pass a
> click through it, you'll heat a boom then a tick. If the delay is
> large enough, the delayed sound is heard as a separate sound, similar
> to an echo. But if the delay is small, less than 1 wavelength as would
> be the case for any EQ filter, the delay is too small for the sound to
> be heard separately.

What an interesting idea. As you say it is possible to
create any response with a long enough FIR. I'd be very
curious to hear what an allpass that continuously changed
the group delay from a small to a very large value like a
second or from large to small. Could be a very interesting
effect.


Bob
--

"Things should be described as simply as possible, but no
simpler."

A. Einstein
Anonymous
January 31, 2005 2:12:22 PM

Archived from groups: rec.audio.pro (More info?)

David,

> It seems to me that the frequency response effects (several dB) would be
far more pronounced than the time dispersion effects. <

Bingo. Here's an article that addresses exactly this issue:

http://www.ethanwiner.com/phase.html

And this one talks about the role of phase shift in equalizers:

http://www.ethanwiner.com/EQPhase.html

--Ethan
Anonymous
January 31, 2005 6:00:44 PM

Archived from groups: rec.audio.pro (More info?)

"Ron Capik" <r.capik@worldnet.att.net> wrote in message
news:41FD61CC.119A4C2@worldnet.att.net
> Hi All:
>
> Was reading an article in the latest issue of mix that talks
> about digital EQ and distortions associated with IIR filters
> and such. Now, I've heard a lot of talk in this news_group
> about group delay artifacts associated with analog (and
> IIR) EQ but don't know that I can identify said artifacts.
> How can one train their ears to recognize or perhaps
> even quantify these group delay effects?

Take some musical samples, and add some group delay of varying kinds and in
various amounts. Then use a DBT comparator as a tool for evaluatiing your
performance as a listener as you compare the modified files to the
originals. You can find the DBT comparators at www.pcabx.com .
February 1, 2005 1:28:43 PM

Archived from groups: rec.audio.pro (More info?)

Excellent!

Mark
Anonymous
February 1, 2005 3:20:10 PM

Archived from groups: rec.audio.pro (More info?)

Mark,

> the article says all EQ filters have group delay distortion <

That's why I was careful to say, "unless it uses special trickery."

> Now if you take it to an extreme however <

Agreed, as I learned at Arny Kruger's ABX site. But in *typical" amounts
phase shift is inaudible.

--Ethan
Anonymous
February 1, 2005 3:25:44 PM

Archived from groups: rec.audio.pro (More info?)

Bob,

> I'd be very curious to hear what an allpass that continuously changed the
group delay from a small to a very large value like a second or from large
to small. Could be a very interesting effect. <

It sounds sort of like a Leslie simulator. When I was designing my Stereo
Synthesizer DIY project for R-e/p magazine years ago
(www.ethanwiner.com/St-Synth.html) I temporarily disconnected the wire that
sums the plain and shifted versions, and listened to only the shifted
version. While the knob was being turned the music became "animated" and you
could hear spatial changes and also slight pitch shifts - similar to
changing the delay time on a digital delay unit as music passes through it.
As soon as I stopped turning the knob the music became static again, and
sounded no different than the plain unshifted version. This was a real
ear-opener for me, and proved that phase shift alone is benign.

--Ethan
Anonymous
February 2, 2005 1:38:11 AM

Archived from groups: rec.audio.pro (More info?)

Ron Capik wrote:
> Was reading an article in the latest issue of mix that talks
> about digital EQ and distortions associated with IIR filters
> and such. Now, I've heard a lot of talk in this news_group
> about group delay artifacts associated with analog (and
> IIR) EQ but don't know that I can identify said artifacts.
> How can one train their ears to recognize or perhaps
> even quantify these group delay effects?

So, this reminds me of something I've been wondering about.

Is there a web site I could visit or maybe a CD I could buy
that is sort of a "gallery of distortion"?

It would be nice to have some signal (a little musical passage) and
then a giant set of the same signal but with various distortions
applied. You could have a million different things, like:

* L and R channel out of phase
* transistor amp clipping
* tube amp clipping
* a million other kinds of clipping
* low-pass filter cutting out things above 10 kHz
* high-pass filter cutting out things below 50 Hz
* harmonic distortion
* stereo crosstalk
* compression
* pumping
* pink noise added to signal
* white noise added to signal
* low-frequency rumble added to signal
* other LP distortions
* hum at mains frequency
* intermodulation distortion
* digital overflow
* comb filtering
* bit reduction without dithering
* some bad dithering algorithm
* mp3
* doppler distortion :-)
* cone going past maximum excursion
* things resonating that shouldn't
* jitter
* digital sampling that doesn't properly filter out
stuff above nyquist frequency (aliasing)
* sound of an oxidized connector
* wow
* flutter
* bad azimuth
* tape saturation
* tape pre-echo
* digital audio with dropouts due to bad CD ripping
* digital audio with uncorrected read errors from damaged CD
* wireless transmitter / receiver not tuned to quite the same
frequency

It could be kind of fun and educational to listen and compare all
these types of distortion. You could train your ear, and maybe
even have (if it were a web site) a computer-generated quiz where
you identify types of distortion by ear.

- Logan
February 2, 2005 10:04:45 AM

Archived from groups: rec.audio.pro (More info?)

I think ALL this stuff was on the last commercial pop music CD I
bought. :-)

Mark
February 2, 2005 12:31:53 PM

Archived from groups: rec.audio.pro (More info?)

Here's how I think of it...

Imagine a sound source the creates two tone busts at the same time, one
is 100 Hz and the other is 10 kHz. They both start and stop at the
same time.

When you hear them together, they make a certain sound. You can
imagine it.

Now, imagine that compared to the 100 Hz tone, the 10 kHz tone gets
delayed by 100 ms. (This is a huge amount of group delay at 10 kHz)
Now you can easily imagine that the sound is different. The 100 Hz
sound is heard first and the 10 kHz tone lags. Clearly this delay is
audible. But this is an example of extreme group delay.

Now consider an example of more typical group delay. Consider that
compared to the 100 Hz tone, the 10 kHz tone is delayed by only 1 or 2
cycles at 10 kHz or about 0.2 ms. Do you think you can hear a
difference now? I don't think so. The group delay created by
typical analog (or digital IIR) EQ filters is not audible.

Mark
February 2, 2005 5:51:05 PM

Archived from groups: rec.audio.pro (More info?)

FFT combined with coefficient changes and IFFT "can" be used for
filtering. It isn't often done because it is not a very efficient
approach.

Mark
Anonymous
February 2, 2005 7:11:02 PM

Archived from groups: rec.audio.pro (More info?)

Ethan Winer wrote:

> Mark,
>
> > the article says all EQ filters have group delay distortion <
>
> That's why I was careful to say, "unless it uses special trickery."
>
> > Now if you take it to an extreme however <
>
> Agreed, as I learned at Arny Kruger's ABX site. But in *typical" amounts
> phase shift is inaudible.
>
> --Ethan

OK, here's the paragraph from mix magazine that prompted my
question on recognition of group-delay effects, where, in his
"The Fast Lane" column, Stephen St. Croix makes the conjecture:
" ... While IIR digital EQ and it's unfortunate phase shift and group
delay error is OK for certain effects like mimicking old-time mush,
it is desperately lame for real EQ--EQ that you don't hear, but
gets the job done. IIR, being a model of analog, is no better than analog.
.... "

From the above statement it would seem that any application of [ IIR ] EQ
will turn one's audio to mush. Maybe my question should have been more
along the lines of how bad is IIR as compared to FIR... but then I'd guess
that it also depends a lot on the IIR algorithm and the FIR window and lots
of other stuff... and who knows what algorithms the various digital EQ
plug-ins use.
Then too, Mr. St. Croix's ears are likely much better trained than mine.

Later...

Ron Capik
--
Anonymous
February 2, 2005 7:11:03 PM

Archived from groups: rec.audio.pro (More info?)

Ron,

> While IIR digital EQ and it's unfortunate phase shift and group delay
error is OK for certain effects like mimicking old-time mush, it is
desperately lame for real EQ <

Now think about that statement. Stephen is claiming that *every single
equalizer* ever made - but discounting the ones he designs and/or sells, of
course - turns everything passing through them to mush. Does that sound at
all reasonable?

> EQ that you don't hear, but gets the job done <

How is this even possible? Via magic energy fields and karmic auras? How
could it "get the job done" if you don't hear anything change?

> Mr. St. Croix's ears are likely much better trained than mine. <

This is also known as "The Emperor's New Clothes" effect, and I give you far
more credit than that!

--Ethan
Anonymous
February 2, 2005 7:11:03 PM

Archived from groups: rec.audio.pro (More info?)

Ron Capik wrote:

> OK, here's the paragraph from mix magazine that prompted my
> question on recognition of group-delay effects, where, in his
> "The Fast Lane" column, Stephen St. Croix makes the conjecture:
> " ... While IIR digital EQ and it's unfortunate phase shift and group
> delay error is OK for certain effects like mimicking old-time mush,
> it is desperately lame for real EQ--EQ that you don't hear, but
> gets the job done. IIR, being a model of analog, is no better than analog.
> ... "

Utter nonsense. One of Voxengo's equalizers, I am pretty
sure, allows you to switch between minimum and linear phase
so I recommend people use their own ears to find the truth
here. I don't like linear phase EQ at all and _much_ prefer
minimum phase. To me it is a question of which sounds more
natural and nature does not admit of linear phase filtering.
It can only be achieved with DSP methods.


Bob
--

"Things should be described as simply as possible, but no
simpler."

A. Einstein
Anonymous
February 2, 2005 7:11:04 PM

Archived from groups: rec.audio.pro (More info?)

Ethan Winer <ethanw at ethanwiner dot com> wrote:
>
>> EQ that you don't hear, but gets the job done <
>
>How is this even possible? Via magic energy fields and karmic auras? How
>could it "get the job done" if you don't hear anything change?

That was my big complaint when I was playing around with an FIR filter
system. It seemed like it took a huge amount of additional EQ in order
to get the same sound I could get with an old-style IIR filter. I'd find
myself doing comparatively dramatic cuts and boosts to get subtle effects,
and that's not a good thing.

>> Mr. St. Croix's ears are likely much better trained than mine. <
>
>This is also known as "The Emperor's New Clothes" effect, and I give you far
>more credit than that!

I think everyone should try FIR filters, and these days with everything gone
to DAWs, I am surprised more people have not. I don't think everyone should
use them, though. I'm not sure anyone should use them.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Anonymous
February 2, 2005 7:11:04 PM

Archived from groups: rec.audio.pro (More info?)

"Ethan Winer" <ethanw at ethanwiner dot com> wrote in message
news:LK-dnZuAucz4iJzfRVn-qg@giganews.com
> Ron,

<quoting Stephen St. Croix in Mix magazine >

>> While IIR digital EQ and it's unfortunate phase shift and group delay
> error is OK for certain effects like mimicking old-time mush, it is
> desperately lame for real EQ <

> Now think about that statement. Stephen is claiming that *every single
> equalizer* ever made - but discounting the ones he designs and/or
> sells, of course - turns everything passing through them to mush.

Well, at least the ones based on IIR.

There seem to be three popular alternatives - IIR, FIR, and FFT.

> Does that sound at all reasonable?

No, but wait, doesn't some of this depend on what we're trying to use the EQ
for?

For example, I might be trying to set up an ABX test for EQs. What do I do
with the EQ under test? Remember, as soon as I set in some non-trivial
amount of eq, my listener(s) are going to be able to pick out the eq for
pretty obvious reasons.

Since Stephen St. Croix seems to be targetting digital eq, perhaps I should
use some *golden* reference analog eq as my standard. Say an Klark-Technik,
Ashley, or Speck. So, my test would then be to tune in some real-world like
eq curve with the reference eq and then adjust the digital eq as well as I
can, to flatten the frequency response curve back out. Now, I ABX compare
this with a straight wire.

Should I constrain my test to compare parametrics to parametrics and
graphics to graphics, or are mixing control formats reasonable?

I dunno it seems like the use of an Eq is highly subjective, and probably
highly dependent on the front panel nut, you know the guy twisting the
knobs.

>> EQ that you don't hear, but gets the job done <

> How is this even possible? Via magic energy fields and karmic auras?
> How could it "get the job done" if you don't hear anything change?

Agreed.

>> Mr. St. Croix's ears are likely much better trained than mine. <

> This is also known as "The Emperor's New Clothes" effect, and I give
> you far more credit than that!

I don't think we need to belabor what happens to a lot of "trained ears"
when they have to get by based on just their ears. IOW, a blind lsitening
test!
Anonymous
February 2, 2005 7:11:04 PM

Archived from groups: rec.audio.pro (More info?)

Bob Cain <arcane@arcanemethods.com> wrote:
>
>Utter nonsense. One of Voxengo's equalizers, I am pretty
>sure, allows you to switch between minimum and linear phase
>so I recommend people use their own ears to find the truth
>here. I don't like linear phase EQ at all and _much_ prefer
>minimum phase. To me it is a question of which sounds more
>natural and nature does not admit of linear phase filtering.
> It can only be achieved with DSP methods.

I have around here a Krohn-Hite filter box which is minimum phase. The
Q is not adjustable, just the turnover frequency and amplitude. It's got
a conventional RLC filter arrangement and another stage of all-pass network
whose controls are coupled mechanically through a bunch of gears to the
controls on the RLC filter so that they follow in such a way that the phase
shift of the all-pass network is always cancelling out the phase shift of
the filter. I believe this was probably intended for communications system
simulation in the early sixties some time. It sounds terrible.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Anonymous
February 2, 2005 7:11:05 PM

Archived from groups: rec.audio.pro (More info?)

Arny Krueger <arnyk@hotpop.com> wrote:
>"Ethan Winer" <ethanw at ethanwiner dot com> wrote in message
>> Ron,
>
><quoting Stephen St. Croix in Mix magazine >
>
>>> While IIR digital EQ and it's unfortunate phase shift and group delay
>> error is OK for certain effects like mimicking old-time mush, it is
>> desperately lame for real EQ <
>
>> Now think about that statement. Stephen is claiming that *every single
>> equalizer* ever made - but discounting the ones he designs and/or
>> sells, of course - turns everything passing through them to mush.
>
>Well, at least the ones based on IIR.
>
>There seem to be three popular alternatives - IIR, FIR, and FFT.

FFT is not a filtering method.

IIR filters make up only a very tiny fraction of the EQ systems available
today. There are a few out there... Weiss makes a box that will do FIR
filters, and I think the Oxford has them available. And most of the forensic
audio systems have them available.

I was very disappointed when the Finalizer came out and it didn't include
FIR filters, but after playing around a bit with them I realize they are
not as useful as they seemed at first.

>> Does that sound at all reasonable?
>
>No, but wait, doesn't some of this depend on what we're trying to use the EQ
>for?
>
>For example, I might be trying to set up an ABX test for EQs. What do I do
>with the EQ under test? Remember, as soon as I set in some non-trivial
>amount of eq, my listener(s) are going to be able to pick out the eq for
>pretty obvious reasons.

That's what you use EQ for, though. If you can't pick out the EQ, why use
it?

>Since Stephen St. Croix seems to be targetting digital eq, perhaps I should
>use some *golden* reference analog eq as my standard. Say an Klark-Technik,
>Ashley, or Speck. So, my test would then be to tune in some real-world like
>eq curve with the reference eq and then adjust the digital eq as well as I
>can, to flatten the frequency response curve back out. Now, I ABX compare
>this with a straight wire.

No, if you do this, what you want is a minimum phase IIR filter, not an
FIR filter, because you want the phase response characteristics to match
also.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Anonymous
February 2, 2005 7:11:06 PM

Archived from groups: rec.audio.pro (More info?)

"Scott Dorsey" <kludge@panix.com> wrote in message
news:ctr893$b0l$1@panix2.panix.com
> Arny Krueger <arnyk@hotpop.com> wrote:
>> "Ethan Winer" <ethanw at ethanwiner dot com> wrote in message
>>> Ron,
>>
>> <quoting Stephen St. Croix in Mix magazine >
>>
>>>> While IIR digital EQ and it's unfortunate phase shift and group
>>>> delay
>>> error is OK for certain effects like mimicking old-time mush, it is
>>> desperately lame for real EQ <
>>
>>> Now think about that statement. Stephen is claiming that *every
>>> single equalizer* ever made - but discounting the ones he designs
>>> and/or sells, of course - turns everything passing through them to
>>> mush.
>>
>> Well, at least the ones based on IIR.
>>
>> There seem to be three popular alternatives - IIR, FIR, and FFT.
>
> FFT is not a filtering method.

Hmm, how is Adobe going wrong here, and saying one of their filters uses
this methodology? Other references as well.

> IIR filters make up only a very tiny fraction of the EQ systems
> available today. There are a few out there... Weiss makes a box that
> will do FIR filters, and I think the Oxford has them available. And
> most of the forensic audio systems have them available.

> I was very disappointed when the Finalizer came out and it didn't
> include FIR filters, but after playing around a bit with them I
> realize they are not as useful as they seemed at first.

>>> Does that sound at all reasonable?

>> No, but wait, doesn't some of this depend on what we're trying to
>> use the EQ for?

>> For example, I might be trying to set up an ABX test for EQs. What
>> do I do with the EQ under test? Remember, as soon as I set in some
>> non-trivial amount of eq, my listener(s) are going to be able to
>> pick out the eq for pretty obvious reasons.

> That's what you use EQ for, though. If you can't pick out the EQ,
> why use it?

Good point. But, that is not the whole story about how we use Eq. We often
use Eq to undo things, no?

>> Since Stephen St. Croix seems to be targetting digital eq, perhaps
>> I should use some *golden* reference analog eq as my standard. Say
>> an Klark-Technik, Ashley, or Speck. So, my test would then be to
>> tune in some real-world like eq curve with the reference eq and then
>> adjust the digital eq as well as I can, to flatten the frequency
>> response curve back out. Now, I ABX compare this with a straight
>> wire.

> No, if you do this, what you want is a minimum phase IIR filter, not
> an FIR filter, because you want the phase response characteristics to
> match also.

That's part of the controversy, no?
Anonymous
February 2, 2005 7:11:07 PM

Archived from groups: rec.audio.pro (More info?)

Arny Krueger <arnyk@hotpop.com> wrote:
>"Scott Dorsey" <kludge@panix.com> wrote in message
>> Arny Krueger <arnyk@hotpop.com> wrote:
>>> There seem to be three popular alternatives - IIR, FIR, and FFT.
>>
>> FFT is not a filtering method.
>
>Hmm, how is Adobe going wrong here, and saying one of their filters uses
>this methodology? Other references as well.

Adobe is not exactly a fine resource for information on dsp technology.
I can actually recommend a Shaum's Outline on the subject (as well as
a pretty good Shaum's on the FFT as well). It does seem that anything
that involves frequency domain stuff gets claimed as having something to
do with the FFT even when it's actually a convolution or some other
unrelated transform.

>>> For example, I might be trying to set up an ABX test for EQs. What
>>> do I do with the EQ under test? Remember, as soon as I set in some
>>> non-trivial amount of eq, my listener(s) are going to be able to
>>> pick out the eq for pretty obvious reasons.
>
>> That's what you use EQ for, though. If you can't pick out the EQ,
>> why use it?
>
>Good point. But, that is not the whole story about how we use Eq. We often
>use Eq to undo things, no?

Right. If you want to use equalization to undo something that happened in
the analogue world, you absolutely want to use an IIR filter because you
want to match the minimum phase characteristic of the original problem.

You'll never find linear phase aberrations in the real world, so having a
linear phase filter doesn't help you undo them. The ones that aren't minimum
phase are going to be things like comb filtering from room effects which are
neither minimum phase NOR linear phase and are still harder to deal with
properly.

>>> Since Stephen St. Croix seems to be targetting digital eq, perhaps
>>> I should use some *golden* reference analog eq as my standard. Say
>>> an Klark-Technik, Ashley, or Speck. So, my test would then be to
>>> tune in some real-world like eq curve with the reference eq and then
>>> adjust the digital eq as well as I can, to flatten the frequency
>>> response curve back out. Now, I ABX compare this with a straight
>>> wire.
>
>> No, if you do this, what you want is a minimum phase IIR filter, not
>> an FIR filter, because you want the phase response characteristics to
>> match also.
>
>That's part of the controversy, no?

I don't see any controversy. If you want to compare effects, you need to
compare devices that do the same thing. FIR filters do something totally
different. It might be something powerful and useful, but it's different.
If it weren't different, it wouldn't be worth bothering with.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Anonymous
February 2, 2005 9:12:48 PM

Archived from groups: rec.audio.pro (More info?)

Scott Dorsey wrote:

> No, if you do this, what you want is a minimum phase IIR filter, not an
> FIR filter, because you want the phase response characteristics to match
> also.

While this seems to be canon, I see no basis for the
assumption in either theory or in double blind testing.


Bob
--

"Things should be described as simply as possible, but no
simpler."

A. Einstein
Anonymous
February 2, 2005 9:29:39 PM

Archived from groups: rec.audio.pro (More info?)

Scott Dorsey wrote:

> I have around here a Krohn-Hite filter box which is minimum phase. The
> Q is not adjustable, just the turnover frequency and amplitude. It's got
> a conventional RLC filter arrangement and another stage of all-pass network
> whose controls are coupled mechanically through a bunch of gears to the
> controls on the RLC filter so that they follow in such a way that the phase
> shift of the all-pass network is always cancelling out the phase shift of
> the filter. I believe this was probably intended for communications system
> simulation in the early sixties some time. It sounds terrible.

Scott, how is that all-pass stage implemented?


Bob
--

"Things should be described as simply as possible, but no
simpler."

A. Einstein
Anonymous
February 2, 2005 10:57:22 PM

Archived from groups: rec.audio.pro (More info?)

Mark <makolber@yahoo.com> wrote:
>
>FFT combined with coefficient changes and IFFT "can" be used for
>filtering. It isn't often done because it is not a very efficient
>approach.

Yes, but isn't that basically giving you an FIR filter with more
computational overhead?
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Anonymous
February 3, 2005 1:11:00 AM

Archived from groups: rec.audio.pro (More info?)

Scott Dorsey wrote:

> < ...snip.. >
> >
> >That's part of the controversy, no?
>
> I don't see any controversy. If you want to compare effects, you need to
> compare devices that do the same thing. FIR filters do something totally
> different. It might be something powerful and useful, but it's different.
> If it weren't different, it wouldn't be worth bothering with.
> --scott
>

Hmmm, as I see it, the controversy is the audibility of group-delay "distortion"
in an EQ situation...

Can anyone point out audible examples of non-EQ group-delay distortion?
[ ie: same EQ but different group delays ]
{...no heroic offsets; delays typical of EQ compensation}

Later...

Ron Capik
--
Anonymous
February 3, 2005 1:11:01 AM

Archived from groups: rec.audio.pro (More info?)

Ron Capik <r.capik@worldnet.att.net> wrote:
>Scott Dorsey wrote:
>> < ...snip.. >
>> >
>> >That's part of the controversy, no?
>>
>> I don't see any controversy. If you want to compare effects, you need to
>> compare devices that do the same thing. FIR filters do something totally
>> different. It might be something powerful and useful, but it's different.
>> If it weren't different, it wouldn't be worth bothering with.
>>
>
>Hmmm, as I see it, the controversy is the audibility of group-delay "distortion"
>in an EQ situation...

Oh, it's audible all right. Get a software filter that will switch between
FIR and IIR constructs and you'll hear differences that are not subtle.

>Can anyone point out audible examples of non-EQ group-delay distortion?
>[ ie: same EQ but different group delays ]
>{...no heroic offsets; delays typical of EQ compensation}

If you want to create group delay without EQ, that effect is pretty close
to being inaudible to me. You can get the all-pass gadget from Little Labs
to try it, but it's something that doesn't happen normally so it's not
something anyone is going to hear without going out of their way to do so.
--scott


--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Anonymous
February 3, 2005 1:11:01 AM

Archived from groups: rec.audio.pro (More info?)

"Ron Capik" <r.capik@worldnet.att.net> wrote in message
news:42014F4A.24A75D64@worldnet.att.net


> Can anyone point out audible examples of non-EQ group-delay
> distortion? [ ie: same EQ but different group delays ]
> {...no heroic offsets; delays typical of EQ compensation}

Audition/CEP has two rather interesting filters - both FFT-based.

One equalizer allows you to draw a frequency response curve, and it is
applied, but without the usual minimum phase shift characteristic. In
essence there is no phase shift at all.

The other equalizer allows you to draw a phase response curve, and it is
applied, but without the usual minimum phase amplitude characteristic. In
essence there is no gain at all at any frequency.

Audition also has Scientific filters that are precisely minimum phase and
are implemented with IIR filters.

The parametric equalizer also uses IIR filters, but the frequency splitter
and graphic equalizer use FIR filters.

BTW, the way one implements a FFT-based filter is to take the FFT of the
input wave, and then convolve the amplitude or phase components with the
desired response curves. Then inverse FFT to return the modified wave.
Anonymous
February 3, 2005 1:37:40 AM

Archived from groups: rec.audio.pro (More info?)

Bob Cain <arcane@arcanemethods.com> wrote:
>Scott Dorsey wrote:
>
>> I have around here a Krohn-Hite filter box which is minimum phase. The
>> Q is not adjustable, just the turnover frequency and amplitude. It's got
>> a conventional RLC filter arrangement and another stage of all-pass network
>> whose controls are coupled mechanically through a bunch of gears to the
>> controls on the RLC filter so that they follow in such a way that the phase
>> shift of the all-pass network is always cancelling out the phase shift of
>> the filter. I believe this was probably intended for communications system
>> simulation in the early sixties some time. It sounds terrible.
>
>Scott, how is that all-pass stage implemented?

I gotta dig up the documentation, but I think it's an op-amp built
with 12BY7 video pentodes. Imagine that in an inverting configuration...
input goes through a series resistor to inverting input and through
another series resistor (X) to the non-inverting input. There's a
feedback resistor between output and inverting input, and a huge honking
variable capacitor between the positive input and ground. This gives
you constant amplitude but phase lag. The capacitor and resistor X
get swapped for phase lead. Neither bandwidth nor slew rate are very
good.

I think Jung describes this arrangement. It _might_ be in the Radiotron
Handbook but I am on the road and away from my copy.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Anonymous
February 3, 2005 4:39:50 AM

Archived from groups: rec.audio.pro (More info?)

Scott Dorsey wrote:

>>Scott, how is that all-pass stage implemented?
>
>
> I gotta dig up the documentation, but I think it's an op-amp built
> with 12BY7 video pentodes. Imagine that in an inverting configuration...
> input goes through a series resistor to inverting input and through
> another series resistor (X) to the non-inverting input. There's a
> feedback resistor between output and inverting input, and a huge honking
> variable capacitor between the positive input and ground. This gives
> you constant amplitude but phase lag. The capacitor and resistor X
> get swapped for phase lead. Neither bandwidth nor slew rate are very
> good.

Thanks. I didn't know you could do that. I've been
thinking that you can't get anything but minimum phase with
lumped elements but that doesn't seem to be true. As long
as there are active elements involved, which is also
required for Chris's odd order Butterworth crossover
example, then mixed phase is achievable. Thanks for
straightening me out on that.


Bob
--

"Things should be described as simply as possible, but no
simpler."

A. Einstein
Anonymous
February 3, 2005 7:05:14 AM

Archived from groups: rec.audio.pro (More info?)

On 2 Feb 2005 22:37:40 -0500, kludge@panix.com (Scott Dorsey) wrote:

>I think Jung describes this arrangement. It _might_ be in the Radiotron
>Handbook but I am on the road and away from my copy.

Of course it's in Jung, but I was so shocked at the idea that it
could be in RDH that I tried to find it. Then it occurred to me that
the name "all-pass filter" would likely be post T-S thinking.
Could still be there in the kilopages; still looking.

Some of this bru-ha-ha is a simple mis-statement about three
layers back, FWIW.

And to Ron, there is a purely (I think!) acoustic example of
variable group delay. A loudspeaker with D'Appolito geometry
and odd-order Butterworth crossover filters should have a
summed-to-unity amplitude response over a moderately large
range of axial locations of the high-pass'ed driver, and
(obviously) varying relative time (group delay). If you hold
your head vertical. Don't cough.

Good fortune,

Chris Hornbeck
Anonymous
February 3, 2005 7:05:15 AM

Archived from groups: rec.audio.pro (More info?)

Chris Hornbeck <chrishornbeckremovethis@att.net> wrote:
>On 2 Feb 2005 22:37:40 -0500, kludge@panix.com (Scott Dorsey) wrote:
>
>>I think Jung describes this arrangement. It _might_ be in the Radiotron
>>Handbook but I am on the road and away from my copy.
>
>Of course it's in Jung, but I was so shocked at the idea that it
>could be in RDH that I tried to find it. Then it occurred to me that
>the name "all-pass filter" would likely be post T-S thinking.
>Could still be there in the kilopages; still looking.

I think this is one of the circuits that you can't think about easily except
as an op-amp design. If you ever get a chance to get the 1960s Philbrick
introduction to op-amp circuits book, it's absolutely fascinating reading
and a lot of what is in Jung and the Active Filter Cookbook is already in
there. No parametric filters, though, and some of the more common gyrator
circuits today aren't there either.

I would bet a nickel that this basic design came out of some sort of analogue
computer work.

>Some of this bru-ha-ha is a simple mis-statement about three
>layers back, FWIW.

I've just been keeping with this thread as an opportunity to be pedantic,
actually. It's not like any of this really matters very much.

>And to Ron, there is a purely (I think!) acoustic example of
>variable group delay. A loudspeaker with D'Appolito geometry
>and odd-order Butterworth crossover filters should have a
>summed-to-unity amplitude response over a moderately large
>range of axial locations of the high-pass'ed driver, and
>(obviously) varying relative time (group delay). If you hold
>your head vertical. Don't cough.

That's not purely acoustic! That's a combination of electrical filters
and acoustical combining! See, I can still be pedantic.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Anonymous
February 3, 2005 7:28:59 AM

Archived from groups: rec.audio.pro (More info?)

On 2 Feb 2005 23:18:22 -0500, kludge@panix.com (Scott Dorsey) wrote:

>I've just been keeping with this thread as an opportunity to be pedantic,
>actually. It's not like any of this really matters very much.

Well, for you and Bob it's just fighting a rear-guard action, but
for lotsa folks, it's still important stuff to understand. If
only to get it ahind ye. True for me anyway.

>>And to Ron, there is a purely (I think!) acoustic example of
>>variable group delay. A loudspeaker with D'Appolito geometry
>>and odd-order Butterworth crossover filters should have a
>>summed-to-unity amplitude response over a moderately large
>>range of axial locations of the high-pass'ed driver, and
>>(obviously) varying relative time (group delay). If you hold
>>your head vertical. Don't cough.
>
>That's not purely acoustic! That's a combination of electrical filters
>and acoustical combining! See, I can still be pedantic.

But, but, but, but...

Thanks as always,

Chris Hornbeck
Anonymous
February 3, 2005 7:29:00 AM

Archived from groups: rec.audio.pro (More info?)

Chris Hornbeck <chrishornbeckremovethis@att.net> wrote:
>On 2 Feb 2005 23:18:22 -0500, kludge@panix.com (Scott Dorsey) wrote:
>
>>I've just been keeping with this thread as an opportunity to be pedantic,
>>actually. It's not like any of this really matters very much.
>
>Well, for you and Bob it's just fighting a rear-guard action, but
>for lotsa folks, it's still important stuff to understand. If
>only to get it ahind ye. True for me anyway.

I don't know if it really is important to know the details. I think it
is important for everyone to play with an FIR filter system themselves and
get a sense of what it sounds like and how they might find it useful. I
never much found it useful although it seems like it ought to be a great
thing. I think first people need to find out what it sounds like and if
they want to use it before worrying about anything else.

And I don't think the thread will really end until everybody has done that.
Of course, then it might turn into a different thread....
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Anonymous
February 3, 2005 7:57:02 AM

Archived from groups: rec.audio.pro (More info?)

On 2 Feb 2005 23:18:22 -0500, kludge@panix.com (Scott Dorsey) wrote:

> If you ever get a chance to get the 1960s Philbrick
>introduction to op-amp circuits book, it's absolutely fascinating reading
>and a lot of what is in Jung and the Active Filter Cookbook is already in
>there. No parametric filters, though, and some of the more common gyrator
>circuits today aren't there either.

Thanks so much for the reference. Just some early hits on the web are
great. Thanks!

Chris Hornbeck
Anonymous
February 3, 2005 12:39:28 PM

Archived from groups: rec.audio.pro (More info?)

Bob Cain <arcane@arcanemethods.com> wrote:
>
>Thanks. I didn't know you could do that. I've been
>thinking that you can't get anything but minimum phase with
>lumped elements but that doesn't seem to be true. As long
>as there are active elements involved, which is also
>required for Chris's odd order Butterworth crossover
>example, then mixed phase is achievable. Thanks for
>straightening me out on that.

There is a really nice discussion of this in the Jung book. You can't do
anything but minimum phase with lumped elements unless you use feedback,
which is the key. I think the Active Filter Cookbook also has some discussion
of this configuration.

But, you really have to go out of your way to make anything non-minimum-phase,
which means that the chance anyone will encounter any electronics that are
not minimum phase in the course of an average day is very small.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."
February 3, 2005 1:08:53 PM

Archived from groups: rec.audio.pro (More info?)

Scott Dorsey wrote:
> Mark <makolber@yahoo.com> wrote:
> >
> >FFT combined with coefficient changes and IFFT "can" be used for
> >filtering. It isn't often done because it is not a very efficient
> >approach.
>
> Yes, but isn't that basically giving you an FIR filter with more
> computational overhead?
> --scott
> --
> "C'est un Nagra. C'est suisse, et tres, tres precis."



Yep, that why it's not often done for filters that have a fixed
response.

I think it is used for case as you described above, where you want to
build a filter where the amplitude and phase response each can be
dialed in independently. This can be done directly with the FFT / IFFT
(using complex values) It would be hard to accomplish this with an FIR
because a new filter (set of coefficients) would have to be determined
each time a phase or amplitude knob was tweaked. Do the so called FIR
based EQs actually use FIR filters or do they use FFT / IFFT? If they
really use actual FIR filters, something in the box has to recalculate
and load an entire new set of coefficients each time you change one of
the "sliders". In effect, a new filter has to be designed each time
you move a slider.

I've also been thinking about the audibility of GD distortion and EQ in
general. Maybe its not the GD that is the problem, maybe its the
ringing. If you create a Low pass filter (for example) with a sharp
cutoff using FIR or IIR or whatever, it will "ring" at (or near) the
cutoff frequency. This ring will occur even if the filter is
implemented as an FIR and has a flat group delay. An FIR filter with
a flat GD will pre-ring and post ring. A min phase IIR will only
post ring. Again, the ringing is inherent and due to the steep EQ
change and will happen no matter how the filter is implemented. See
the Gibbs phenomenon. So I think what I'm getting at is that rapid
freq response changes produce ringing and sound bad no matter how they
are created. So if you have a 256 band EQ and you create a steep
change in the freq response, it will probably sound bad regardless of
how it is implemented. If you need to apply EQ and want it to sound
good, the response changes should be done gradually.

So maybe it's not the GD variation from EQ that sounds bad but simply
the ringing cause by rapid frequency response changes.


This may also relate to why 44.1 kHz sampling may sound bad to some
people. There is always a steep low pass filter at around 20 kHz. I
know __I__ can't hear any ringing around 20 kHz, but I suppose with the
right source material and the right speakers someone might be able to
hear ringing at 20 kHz.

And a final comment for anyone wondering about how FIR filters can have
pre-ringing. How does the filter predict the future?. It doesn't.
(All real filters are causal.) The ringing starts when the input
starts. The filter delays the main part of the signal so that the
ringing comes out before the main part of the signal, but the ringing
does not (and obviously cannot) come out before the input went in.

Thanks

Mark
Anonymous
February 3, 2005 1:18:02 PM

Archived from groups: rec.audio.pro (More info?)

Scott Dorsey wrote:

>
> There is a really nice discussion of this in the Jung book. You can't do
> anything but minimum phase with lumped elements unless you use feedback,
> which is the key.

But Chris's example of the high, odd order Butterworth
crossover can only be done with separately amped drivers yet
there is no feedback involved. It's not about it being an
acoustic system because if you replace the drivers with a
summing amp the transfer function remains all-pass.

I'm sure this is all addressed in Jung but I don't have that
to reference. At any rate, I'm always glad to have a
misconception cleared up.


Bob
--

"Things should be described as simply as possible, but no
simpler."

A. Einstein
February 3, 2005 1:19:26 PM

Archived from groups: rec.audio.pro (More info?)

I think you can create non min phase passivly. I have seen video LC
Low pass filters that include passive LC all pass networks to equalize
the GD distortion.

see passive all pass and delay equlilzer sections of:

http://www.filter-solutions.com

(I have no connection to this site)

But I agree with you, these are the exception. Most common circuits
that change the frequency repsonse are min phase.

Mark
Anonymous
February 3, 2005 4:45:17 PM

Archived from groups: rec.audio.pro (More info?)

Bob Cain <arcane@arcanemethods.com> wrote:
>Scott Dorsey wrote:
>
>>
>> There is a really nice discussion of this in the Jung book. You can't do
>> anything but minimum phase with lumped elements unless you use feedback,
>> which is the key.
>
>But Chris's example of the high, odd order Butterworth
>crossover can only be done with separately amped drivers yet
>there is no feedback involved. It's not about it being an
>acoustic system because if you replace the drivers with a
>summing amp the transfer function remains all-pass.

I think it's about it being an acoustic system because it has to do with
delay. Delaying is the key to all of this.

>I'm sure this is all addressed in Jung but I don't have that
>to reference. At any rate, I'm always glad to have a
>misconception cleared up.

Run out right now and order Audio IC Op-Amp Applications immediately!
Or whatever the latest edition is called. It's up there with the
Radiotron Handbook and the Audio Cyclopedia as being a necessary
reference. It doesn't have a lot of the math and is much more of a
cookbook, but it's still something everybody needs.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Anonymous
February 3, 2005 8:24:50 PM

Archived from groups: rec.audio.pro (More info?)

"Scott Dorsey" <kludge@panix.com> wrote in message
news:cttrft$l6v$1@panix2.panix.com...

> Run out right now and order Audio IC Op-Amp Applications immediately!

The Walter Jung book? I thought that had been out of print for years.

Hal Laurent
Baltimore
Anonymous
February 3, 2005 8:49:59 PM

Archived from groups: rec.audio.pro (More info?)

Hal Laurent <laurent@charm.net> wrote:
>"Scott Dorsey" <kludge@panix.com> wrote in message
>
>> Run out right now and order Audio IC Op-Amp Applications immediately!
>
>The Walter Jung book? I thought that had been out of print for years.

It's always going out of print and then being reissued and then going out
of print again. Sam's publishes it, or published it. You gotta have it.
--scott


--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Anonymous
February 3, 2005 9:13:40 PM

Archived from groups: rec.audio.pro (More info?)

"Scott Dorsey" <kludge@panix.com> wrote in message
news:ctu9qn$kjp$1@panix2.panix.com...
> Hal Laurent <laurent@charm.net> wrote:
> >"Scott Dorsey" <kludge@panix.com> wrote in message
> >
> >> Run out right now and order Audio IC Op-Amp Applications immediately!
> >
> >The Walter Jung book? I thought that had been out of print for years.
>
> It's always going out of print and then being reissued and then going out
> of print again. Sam's publishes it, or published it. You gotta have it.

I *do* have it, but I'd stopped recommending it to people 'cause I thought
it was unavailable.

Hal Laurent
Baltimore
Anonymous
February 3, 2005 10:54:28 PM

Archived from groups: rec.audio.pro (More info?)

Hal Laurent <laurent@charm.net> wrote:
>"Scott Dorsey" <kludge@panix.com> wrote in message
>news:ctu9qn$kjp$1@panix2.panix.com...
>> Hal Laurent <laurent@charm.net> wrote:
>> >"Scott Dorsey" <kludge@panix.com> wrote in message
>> >
>> >> Run out right now and order Audio IC Op-Amp Applications immediately!
>> >
>> >The Walter Jung book? I thought that had been out of print for years.
>>
>> It's always going out of print and then being reissued and then going out
>> of print again. Sam's publishes it, or published it. You gotta have it.
>
>I *do* have it, but I'd stopped recommending it to people 'cause I thought
>it was unavailable.

Sadly, the Audio Cyclopedia is also unavailable now, and it's also a book
everyone needs to have. At least the Radiotron Handbook has been reprinted
(and is even available on CD).
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."
Anonymous
February 3, 2005 10:54:29 PM

Archived from groups: rec.audio.pro (More info?)

Scott Dorsey wrote:
>
> Sadly, the Audio Cyclopedia is also unavailable now, and it's also a book
> everyone needs to have.

I bought Glen Ballou's follow-on, is the original that much better?
Anonymous
February 4, 2005 2:42:50 AM

Archived from groups: rec.audio.pro (More info?)

"Scott Dorsey" <kludge@panix.com> wrote in message
news:ctuh44$4q5$1@panix2.panix.com...

> At least the Radiotron Handbook has been reprinted (and is even available
on CD).

That sounds like something I should have. Can you point me to a source?

Hal Laurent
Baltimore
!