Hello Soongsc,
About linearity of phase there have been many improvments in the last 20 years. As for few examples: "
"quasioptimal crossovers", linear FIR filtering, phase equalization...
This is coupled to a better understanding on how loudspeakers ( + horns) distort phase (min phase behaviour or not...) and how to use them for better results according to phase linearity.
This is also well illustrated by the way most new audio analysis software even for non professional purpose have phase analysis or group delay analysis (and CSD, Waterfall....) modules.
Therefore, it is false, IMHO, to say that we have still to deal with very non linear phase response for our audio system and accepting this as a minor unavoidable "side effect".
Best regards from Paris, France
Jean-Michel Le Cléac'h
About linearity of phase there have been many improvments in the last 20 years. As for few examples: "
"quasioptimal crossovers", linear FIR filtering, phase equalization...
This is coupled to a better understanding on how loudspeakers ( + horns) distort phase (min phase behaviour or not...) and how to use them for better results according to phase linearity.
This is also well illustrated by the way most new audio analysis software even for non professional purpose have phase analysis or group delay analysis (and CSD, Waterfall....) modules.
Therefore, it is false, IMHO, to say that we have still to deal with very non linear phase response for our audio system and accepting this as a minor unavoidable "side effect".
Best regards from Paris, France
Jean-Michel Le Cléac'h
Hello Jean-Michel,
Much techniques you have mentioned require digital implementation, is that correct? I have seen latency filters that handle phase in a limited frequency region which seem oaky for crossover adjustments, but not freely tunable over wide frequency bands. Do you know of any commercial application of these techniques "quasioptimal crossovers", linear FIR filtering, phase equalization? I know that there are many ways to handle the crossover points, but do they actually handle the low frequency and high frequency rolloff regions?
Much techniques you have mentioned require digital implementation, is that correct? I have seen latency filters that handle phase in a limited frequency region which seem oaky for crossover adjustments, but not freely tunable over wide frequency bands. Do you know of any commercial application of these techniques "quasioptimal crossovers", linear FIR filtering, phase equalization? I know that there are many ways to handle the crossover points, but do they actually handle the low frequency and high frequency rolloff regions?
Hello Soongsc,
I use my own developped quasioptimal crossover which can be implemented with IIR filtres, active filters or passive filters..
see:
Document and Speadsheet (doc, xls): http://freerider.dyndns.org/anlage/LeCleach1.zip
Theory of operation (ppt): http://freerider.dyndns.org/anlage/LeCleach2.zip
You can read also:
http://www.diyaudio.com/forums/multi-way/121175-lecleach-paper-crossovers.html
Best regards from Paris, France
Jean-Michel Le Cléac'h
I use my own developped quasioptimal crossover which can be implemented with IIR filtres, active filters or passive filters..
see:
Document and Speadsheet (doc, xls): http://freerider.dyndns.org/anlage/LeCleach1.zip
Theory of operation (ppt): http://freerider.dyndns.org/anlage/LeCleach2.zip
You can read also:
http://www.diyaudio.com/forums/multi-way/121175-lecleach-paper-crossovers.html
Best regards from Paris, France
Jean-Michel Le Cléac'h
Below I show simus originally intended for jzagaia and his issue of phase plugs some postings back (and my recommendation for ring radiators) – but they are probably as interesting in the context of the high pass function of a horn and its groupe delay as well
What we see below are conical horns with a round over at mouth and throat (– driver is kept out of the picture as far as possible).
Mouth is kept constant and throat diameter is doubled each time.
To keep things simple, throat rounding is calculated as to the conical tangent meets centre of plain diaphragm – very close to OS...
below for a diaphragm / horn throat of 6.25mm / 1/4"
below for a diaphragm / horn throat of 12.5mm / 1/2"
below for a diaphragm / horn throat of 25mm / 1"
below for a diaphragm / horn throat of 50mm / 2"
below for a diaphragm / horn throat of 100mm / 4"
All in all - it seems there *is* kinda high pass filter originating from the horn (rather than from the driver)
##########
Now here are simus of 3rd order HP filter at 1000 Hz and different characteristic to compare with
Below 3rd order Gaussian HP filter at 1000 Hz
Below 3rd order Bessel HP filter at 1000 Hz
Below 3rd order Butterworth HP filter at 1000 Hz
Below 3rd order Chebychev HP filter at 1000 Hz
All in all – group delay is pretty different but anyway we stay roughly below 1ms for a HP of 1kHz.
For comparison here is what Jean-Michel calculated for my horn contour
http://www.diyaudio.com/forums/multi-way/100392-beyond-ariel-130.html#post2008893
and here a pretty well corresponding 500 Hz 3rd order Butterworth HP
Question for me is now :
– *if* this group delay can be seen as equivalent to any other HP behaviour – why does a (steep slope) XO with direct radiators not have the same pronounced sonic effect in this department as horns ?
Michael
What we see below are conical horns with a round over at mouth and throat (– driver is kept out of the picture as far as possible).
Mouth is kept constant and throat diameter is doubled each time.
To keep things simple, throat rounding is calculated as to the conical tangent meets centre of plain diaphragm – very close to OS...
below for a diaphragm / horn throat of 6.25mm / 1/4"
below for a diaphragm / horn throat of 12.5mm / 1/2"
below for a diaphragm / horn throat of 25mm / 1"
below for a diaphragm / horn throat of 50mm / 2"
below for a diaphragm / horn throat of 100mm / 4"
All in all - it seems there *is* kinda high pass filter originating from the horn (rather than from the driver)
##########
Now here are simus of 3rd order HP filter at 1000 Hz and different characteristic to compare with
Below 3rd order Gaussian HP filter at 1000 Hz
Below 3rd order Bessel HP filter at 1000 Hz
Below 3rd order Butterworth HP filter at 1000 Hz
Below 3rd order Chebychev HP filter at 1000 Hz
All in all – group delay is pretty different but anyway we stay roughly below 1ms for a HP of 1kHz.
For comparison here is what Jean-Michel calculated for my horn contour
http://www.diyaudio.com/forums/multi-way/100392-beyond-ariel-130.html#post2008893
and here a pretty well corresponding 500 Hz 3rd order Butterworth HP
Question for me is now :
– *if* this group delay can be seen as equivalent to any other HP behaviour – why does a (steep slope) XO with direct radiators not have the same pronounced sonic effect in this department as horns ?
Michael
Last edited:
It should be the same. Horn roll-off is very nearly minimum phase, at least the ones I've seen, so there is no difference in group delay between the horn and an electrical filter with the same magnitude response. The only way you get harmonics arriving way before the fundamental is if the fundamental is many dB down in response so the effect 'should' be relatively inaudible.Question for me is now :
– *if* this group delay can be seen as equivalent to any other HP behaviour – why does a (steep slope) XO with direct radiators not have the same pronounced sonic effect in this department as horns ?
Michael
Now delayed reflections caused by mouth diffraction are another matter.
Hello Catapult,
I disagree.
1) there is many musical instruments for which the fundamental is not the most powerful frequency components.
2) The group delay rise when the frequency decreases toward the acoustical cut-off of the horn can be efefctively perceived as, (e.g.) a male voice seems to emanate separately of the instruments from the throat of the horn itself...
Best regards from Paris, France
Jean-Michel Le Cléac'h
I disagree.
1) there is many musical instruments for which the fundamental is not the most powerful frequency components.
2) The group delay rise when the frequency decreases toward the acoustical cut-off of the horn can be efefctively perceived as, (e.g.) a male voice seems to emanate separately of the instruments from the throat of the horn itself...
Best regards from Paris, France
Jean-Michel Le Cléac'h
Hello Michael,
About electrical crossovers:
In fact what you have to consider is not the group delay curve of each filter (LPF and HPF) but the resulting group delay curve of the whole (LPF + HPF).
And for people for which phase distortion is audible, yes they can hear such large group delay variations.
Best regards from Paris, France
Jean-Michel Le Cléac'h
About electrical crossovers:
In fact what you have to consider is not the group delay curve of each filter (LPF and HPF) but the resulting group delay curve of the whole (LPF + HPF).
And for people for which phase distortion is audible, yes they can hear such large group delay variations.
Best regards from Paris, France
Jean-Michel Le Cléac'h
Hello,
It is sometimes said that as phase variation with a horn is comparable to the phase variation of an electrical filter an equalisation based on a phase min beahaviour should be possible.
I never saw any published measure of sur elctrical compensation of the group delay variation of a horn (which is due to its throat reactance).
I used several years ago to work a bit on that subject.
In the attached graph you can see the partial compensation (linearisation) of the group delay curve of a Le Cléac'h horn (Fc = 320Hz, T = 0.8). Nearly one octave of useful bandwith was gained in that example.
Best regards from Paris, France
Jean-Michel Le Cléac'h
It is sometimes said that as phase variation with a horn is comparable to the phase variation of an electrical filter an equalisation based on a phase min beahaviour should be possible.
I never saw any published measure of sur elctrical compensation of the group delay variation of a horn (which is due to its throat reactance).
I used several years ago to work a bit on that subject.
In the attached graph you can see the partial compensation (linearisation) of the group delay curve of a Le Cléac'h horn (Fc = 320Hz, T = 0.8). Nearly one octave of useful bandwith was gained in that example.
Best regards from Paris, France
Jean-Michel Le Cléac'h
Attachments
The cutoff area is also where the CSD decays slowest. It might be difficult to measure under normal conditions, but have you also looked at the CSD before and after implementing the phase equalization? The perception you mention also seems like long CSD time in ordinary drivers.
Hello Soongsc,
From memory, the length of the spectrogram decay (at the time I did that study I didn't possess ARTA nor CSD)near the cutoff is not modified by the Group Delay equalization, the result of which I shown.
You can even see on the pulse response drawn on the attached graph some amelioration of the decay of the pulse response... (but this may due to the partial centering of the low frequency components pulse response around the main HF holding pulse...)
Best regards from Paris, France
Jean-Michel Le Cléac'h
From memory, the length of the spectrogram decay (at the time I did that study I didn't possess ARTA nor CSD)near the cutoff is not modified by the Group Delay equalization, the result of which I shown.
You can even see on the pulse response drawn on the attached graph some amelioration of the decay of the pulse response... (but this may due to the partial centering of the low frequency components pulse response around the main HF holding pulse...)
Best regards from Paris, France
Jean-Michel Le Cléac'h
Jean-Michel,
Now that I look at the impulse and response a bit closer, it does seem to damp out a certain frequency range. But is also seems that it's only possible to improve group delay in a limited frequency range just like a latency filter. Perhaps this might be one reason to tune the horn/guide lower than the driver F0? I've been struggling among these for quite a while. Maybe I should try some filtering to see how this works.
Now that I look at the impulse and response a bit closer, it does seem to damp out a certain frequency range. But is also seems that it's only possible to improve group delay in a limited frequency range just like a latency filter. Perhaps this might be one reason to tune the horn/guide lower than the driver F0? I've been struggling among these for quite a while. Maybe I should try some filtering to see how this works.
I agree - but pronounce "should"
For the slightly larger horn I did for the B&G NEO3 the sonics were pretty strange.
On the other hand I think I never have carefully listened to a direct radiator tweeter on high slope XO *without* mid / bass.
I guess some time I will have to check this out more seriously
It's just that its hard to believe that what I subsumize under horn honk hasn't been explored in each and every aspect yet – given this malady lasts for a roughly 100 years now !Yes its a strange sonic behaviour.
On the other hand I have a feeling that group delay alone is possible not enough for explanation.
If I look back on what I have shown in response shaping here:
http://www.diyaudio.com/forums/multi-way/100392-beyond-ariel-115.html#post1837992
I'd say there is no reason to assume group delay might not be affected accordingly by response shaping.
*This* lead me to the question :
– *if* this group delay can be seen as equivalent to any other HP behaviour – why does a (steep slope) XO with direct radiators not have the same pronounced sonic effect in this department as horns ?
#################
Sure I fully agree – nevertheless...
Regarding its IR, phase, GD, CSD etc. I tend to agree...
This I do not understand ?
You mean excessive phase shifting without touching FR ?
Looks interesting but I don't get the point in "extending the useful bandwidth by one octave" – I mean – the –3dB point of the HP filter is not that much affected – so bandwidth in this configuration is how it is – adding a XO on top (at shifted frequency – not the order increase we probably see here), would completely mess up the current plots.
On the other hand – doing a textbook filter, the group delay at XO point is completely determined.
For the plots I've shown we would have to go Bessel characteristic to severely change group delay at XO point – and even then we can cut it only ~ half
So – summing up – my guess about that honk issue would be that its neither (pure) group delay nor mouth reflection.
Looking at IM, especially Dopple IM – well - *thats* the usual suspicious candidate for me.
Again, hard to believe that this hasn't been nailed...
Michael
Last edited:
Hello,
Sheldon asked about instruments having a frequency spectrum with the level of the fudamentals below the level of some harmonics:
A one minute search on the net
http://physics.info/music/spectrum-human-voice.gif
female voice,
There's life above 20 kilohertz! A survey of musical instrument spectra to 102.4 kHz
trumpet
http://www.zainea.com/v195.gif
human voice on a wovel
http://www.acoustics.org/press/139th/brown.gif
oboe, saxophone, flute
http://www.owlnet.rice.edu/~psyc351/Images/Timbre1.jpg
guitar, bassoon, alto saxophone
Best regards from Paris, France
Jean-Michel Le Cléac'h
Sheldon asked about instruments having a frequency spectrum with the level of the fudamentals below the level of some harmonics:
A one minute search on the net
http://physics.info/music/spectrum-human-voice.gif
female voice,
There's life above 20 kilohertz! A survey of musical instrument spectra to 102.4 kHz
trumpet
http://www.zainea.com/v195.gif
human voice on a wovel
http://www.acoustics.org/press/139th/brown.gif
oboe, saxophone, flute
http://www.owlnet.rice.edu/~psyc351/Images/Timbre1.jpg
guitar, bassoon, alto saxophone
Best regards from Paris, France
Jean-Michel Le Cléac'h
Hello Michael,
Group delay is affected by response shaping if we use equalizers having min phase behaviour. (linear phase or zero phase equalizing iare different animals).
I said that the useful bandwith was increased by nearly 1 octave in the example I gave because I only use for myself the horns inside the interval of frequency for which the group delay (expressed in its equivalent distance travelled at the speed of sound) doesn't varies more than 10 centimeters (4 inches). I always set the crossover Fc at the frequency above which the group delay variation tend to be lesser than 10 centimeters.
In the illustrated case this means I can set the HP crossover at a lower frequency for the GD compensated horn than for the uncompensated.
If you use a Le Cléac'h horn following that method no honkiness is present (many testimonies insist on the point a Le Cléac'h dosen't sound like a horn...)
Best regards, from Paris France
Jean-Michel Le Cléac'h
Group delay is affected by response shaping if we use equalizers having min phase behaviour. (linear phase or zero phase equalizing iare different animals).
I said that the useful bandwith was increased by nearly 1 octave in the example I gave because I only use for myself the horns inside the interval of frequency for which the group delay (expressed in its equivalent distance travelled at the speed of sound) doesn't varies more than 10 centimeters (4 inches). I always set the crossover Fc at the frequency above which the group delay variation tend to be lesser than 10 centimeters.
In the illustrated case this means I can set the HP crossover at a lower frequency for the GD compensated horn than for the uncompensated.
If you use a Le Cléac'h horn following that method no honkiness is present (many testimonies insist on the point a Le Cléac'h dosen't sound like a horn...)
Best regards, from Paris France
Jean-Michel Le Cléac'h
Anytime I second that your contour (operated in the right range) is fantastic - seriously!
What I want to clarify is something different - *if* on the example given you put a XO on top where the GD still looks good for now, *then* GD is back again at XO point - for any solution that can be realized passively at least...
Hence I said, I do not see how you ever will be able to "lower" GD (or increase useful BW) and also its no explanation that direct radiators seem to have quite different sonics in this department under comparable conditions regarding FR phase GD CSD etc....
Michael
What I want to clarify is something different - *if* on the example given you put a XO on top where the GD still looks good for now, *then* GD is back again at XO point - for any solution that can be realized passively at least...
Hence I said, I do not see how you ever will be able to "lower" GD (or increase useful BW) and also its no explanation that direct radiators seem to have quite different sonics in this department under comparable conditions regarding FR phase GD CSD etc....
Michael
Last edited:
