We can not hear group-delay under 100Hz?

[DBMandrake]

Originally Posted by Barleywater

Yes, percussion on track is only sound with hard transient. Pano's Butterworth filtering for this is form of shelving filter, and from group delay perspective looks something like this:

Attachment 321838

Very mild compared to temporal aberration with speaker in living room measured at listening position:

Attachment 321837

Above is raw cardioid woofer setup with previously posted DSP correction results in #36

You can't directly compare these two graphs (for audibility) though - you're comparing apples to oranges because the first one is group delay that is caused only by excess phase (the frequency response is still flat with pano's summed filters) but the second one is clearly group delay that is caused largely by amplitude response non flatness, eg group delay that is due to both minimum phase and excess phase components but with minimum phase components dominating.

Therefore you can't tell anything about how "audible" this group delay is in comparison to the first, because most of it is caused by a very non-flat amplitude response which will be audible in and of itself.

As I said in my previous post, lumping together the audibility of group delay and excess group delay (pano's test has only excess group delay) has been going on throughout the thread and only serves to confuse the issue.

The only valid test to determine whether group delay by itself is audible is to maintain identical amplitude responses which means we are really studying excess group delay, not group delay.

Any test where we're altering the frequency response to produce group delay proves nothing except that frequency response alterations are audible...

Out of interest, try showing us the excess group delay instead of group delay on your bottom measurement, you'll find that across most of the bass region the excess group delay will be relatively flat even in a room, but there may be one or two frequencies where it peaks up really high, typically where there is either a notch in the amplitude response or where the reflections are higher in amplitude than the direct signal...this is useful because it shows us where (with the current speaker configuration and listening position) we cannot EQ out the response error with normal minimum phase EQ...

[Barleywater]

Originally Posted by DBMandrake

You can't directly compare these two graphs (for audibility) though - you're comparing apples to oranges because the first one is group delay that is caused only by excess phase (the frequency response is still flat with pano's summed filters) but the second one is clearly group delay that is caused largely by amplitude response non flatness, eg group delay that is due to both minimum phase and excess phase components but with minimum phase components dominating.

Therefore you can't tell anything about how "audible" this group delay is in comparison to the first, because most of it is caused by a very non-flat amplitude response which will be audible in and of itself.

As I said in my previous post, lumping together the audibility of group delay and excess group delay (pano's test has only excess group delay) has been going on throughout the thread and only serves to confuse the issue.

The only valid test to determine whether group delay by itself is audible is to maintain identical amplitude responses which means we are really studying excess group delay, not group delay.

Any test where we're altering the frequency response to produce group delay proves nothing except that frequency response alterations are audible...

Out of interest, try showing us the excess group delay instead of group delay on your bottom measurement, you'll find that across most of the bass region the excess group delay will be relatively flat even in a room, but there may be one or two frequencies where it peaks up really high, typically where there is either a notch in the amplitude response or where the reflections are higher in amplitude than the direct signal...this is useful because it shows us where (with the current speaker configuration and listening position) we cannot EQ out the response error with normal minimum phase EQ...

All most my exact point. Pano's first phase distorting filter is of same type all ready present in most recording/playback chains and speakers.

You can't know what a square wave sounds like until you can produce it acoustically. Real sounds in music all start with causal moment, with minimum phase and very quickly morph to non-minimum phase as rest of instrument starts radiating, followed by immediate environment of instrument. So circle the wagons; compare waveform recorded of instrument, that has been corrected for minimum phase effects of microphone system, and what this sounds like played back on speaker v real instrument.

When I analyze my own recording and playback chain, it readily shows minimum phase impact on content <200Hz, providing basis of masking very effect being discussed, even when using expensive headphones.

Now here is confusion:

As I said in my previous post, lumping together the audibility of group delay and excess group delay (pano's test has only excess group delay) has been going on throughout the thread and only serves to confuse the issue.

Pano's test with "Josie" is convolution with sum of two minimum phase filters, indexed to causal start of each resulting in perfectly minimum phase filter. Zero excess phase.

Points about excess phase are well taken into consideration with corrections I produce. In case of posted sub setup room response is speaker response at listening position. This is excess phase of raw cardioid sub:

sub raw excess gd.png

In above first big excess phase event is above range of sub. For main speaker measurement and correction due consideration is given, and excess phase between two measurements is adjusted for when integrating filters from different measurements.

Fully corrected system is basis for comparison, and I find my setup reproduces waveform of source to much greater fidelity than any uncorrected system I've worked with.

[DBMandrake]

Originally Posted by Barleywater

Now here is confusion:

Pano's test with "Josie" is convolution with sum of two minimum phase filters, indexed to causal start of each resulting in perfectly minimum phase filter. Zero excess phase.

Not true at all, unless you have a different definition of minimum phase to everyone else. Two minimum phase filters can and usually do sum to a non-minimum phase result.

Take a 2nd order L/R low pass and high pass. Individually they are minimum phase. Sum their outputs together and the amplitude response is perfectly flat however the sum is not minimum phase because there is a 180 degree total phase rotation from low frequencies to high frequencies. It forms an all-pass filter.

The excess phase / excess group delay without any change of amplitude response was the whole point of pano's test.

[Pano]

At least I hope that was the point (and that I got it right).

[gedlee]

Originally Posted by DBMandrake

Not true at all, unless you have a different definition of minimum phase to everyone else. Two minimum phase filters can and usually do sum to a non-minimum phase result.

Take a 2nd order L/R low pass and high pass. Individually they are minimum phase. Sum their outputs together and the amplitude response is perfectly flat however the sum is not minimum phase because there is a 180 degree total phase rotation from low frequencies to high frequencies. It forms an all-pass filter.

The excess phase / excess group delay without any change of amplitude response was the whole point of pano's test.

You have to be careful here with the term "sum". It sometimes means "cascade", in which case the two MP filters will always yield a MP result and "adding" different signal paths, which does not guarantee a MP result from two MP filters. I see confusion above in this regard.

[thierry38efd]

Originally Posted by Barleywater

Points about excess phase are well taken into consideration with corrections I produce. In case of posted sub setup room response is speaker response at listening position. This is excess phase of raw cardioid sub:



In above first big excess phase event is above range of sub. For main speaker measurement and correction due consideration is given, and excess phase between two measurements is adjusted for when integrating filters from different measurements.

Fully corrected system is basis for comparison, and I find my setup reproduces waveform of source to much greater fidelity than any uncorrected system I've worked with.

Impressive !
where can we see this cardiod subwoofer ? any pictures around there ?
what kind of DSP are you using ? (standalone or computer ).

[DBMandrake]

Originally Posted by gedlee

You have to be careful here with the term "sum". It sometimes means "cascade", in which case the two MP filters will always yield a MP result and "adding" different signal paths, which does not guarantee a MP result from two MP filters. I see confusion above in this regard.

Wouldn't cascading of two filters be multiplication of the transfer functions instead of algebraic summing ?

In any case, given the context of pano's test signal I can't see how there could be any confusion, the result of cascading a high and low pass filter is never going to be an all-pass with a flat amplitude response, which was a key part of the test.

[Pano]

What I was attempting to duplicate was a 4th order HP and LP recombined, as they would be in the air. Is this different from combining them electrically or in software?

[DBMandrake]

Originally Posted by Pano

What I was attempting to duplicate was a 4th order HP and LP recombined, as they would be in the air. Is this different from combining them electrically or in software?

Nope, at least not on the design axis with the drivers time aligned. Non time aligned drivers or going off the vertical axis would obviously introduce a time delay between the two (which you could also simulate) while both horizontal and vertical off axis would change the individual transfer functions a bit due to off axis beaming of the drivers, thus the summed off axis result.

So it would be a valid test of the perceived on axis response but probably not of the changes to the reverberant field or early reflections. (since summing in air is only the same as the electrical summing on axis)

Early sidewall reflections would only be affected by the changes due to driver beaming mind you, not by differences in delay, so for a speaker with uniform horizontal off axis performance it would be representative of early sidewall reflections too.

[thierry38efd]

I tried to hear a difference between the 3 files,headphone and loudspeaker.
...nothing.

here's the modification on the file
it's about 4-5 ms delay below 125 Hz.
it's like the woofer is 4.7 fts stepping back.

24 db/oct recombined.



and the group delay modification