Horn Honk $$ WANTED $$

[Elias]

Hello thend,

I think the wavelet already has the required temporal resolution to see the reflections without the need for any extra tube.

- Elias

That's why I invent last year my experimental device for measure the horn reflection.
It can measure the bandwith (with equalisation) of reflection in the horn.
I think it's the perfect test device for this topic, because I have invent it for that... (reflection, diffraction, honk)

Experimental highlight of horn reflection's

 

[thend]

Hello Ellias,

I think the wavelet already has the required temporal resolution to see the reflections without the need for any extra tube.

Maybe, it's just a question of visibility, we can focus on echo (reflection) only.

Today, I have suggested to developer of FuzzMeasure to integrate the wavelet analysis, he said he works on it since 6 month.

 

[doug20]

I wonder when HOLM will have CSD or wavelets. Maybe we should request it.

 

[Paul W]

FWIW, here is a CSD plot from SoundEasy. (ND1460A on an XT1464 @1M)

 

[Jmmlc]

Hello Paul,

The CSD you show to us is typical of a CSD obtained with a very small apodization time.

Many people think they will obtain a better time resolution in choosing a small value of the apodization time. At high frequency this is true and we can obtain a kind of vertical wall (in the waterfall presentation of the CSD) often used by compression driver builders in order to show how fast is their driver.

But in the same time a small value of the apodization time leads to artifcats that are clearly visible on the low frequency part of your CSD graph.

You'll see here attached the effect of the apodization time on a CSD graph (the one corresponding to a Celestion driver previously mentionned).

On left the largest apodization time in ARTA (1ms), on right the smallest one (0.02ms).

IMHO the CSD on the left gives a better presentation of the behaviour of the driver.

Best regards from Paris, France

Jean-Michel Le Cléac'h

FWIW, here is a CSD plot from SoundEasy. (ND1460A on an XT1464 @1M)

 

[doug20]

Great! A new term to figure out

"apodization time"

 

[Jmmlc]

Hello Doug20,

Well may be it is better to call it "apodizing time"

Great! A new term to figure out

"apodization time"

That's the kind of parameter that if you don't optimize it, the resulting CSD is meaningless...

Best regards from Paris, France

Jean-Michel Le Cléac'h

 

[thend]

Wavelets are reality. CSD is a mathematical artefact.

I'm pretty agree.

For frequency domain the frequency window is sufficient, the temporal windowing does the rest.
This:

An externally hosted image should be here but it was not working when we last tested it.

is not this:

An externally hosted image should be here but it was not working when we last tested it.

just a form of consequence.

 

[Paul W]

Hello Paul,

The CSD you show to us is typical of a CSD obtained with a very small apodization time.

Thanks for the tip! I'll do some experiments next time I fire up the measurement system.

 

[EarlK]

Back to Honk

Here are the FRs ( no EQ ), GDs, & CSDs of some components that I use in my SR work ( @ 1500 hz xover, 24 db LR ) . There's a single 22uF cap in line as a HP .

The overlays are 0 & @ 30 degrees Horizontal at 50 cm ( This was done in my living room close to lots of stuff ).

• The driver is the JBL 2431H & the horn is a small 60 by 40 waveguide ( I can't remember the model # off-hand ).
• Driver has a high Fs , just less than 1000 hz .
• This is a tough driver to try to use in HiFi / it really needs a tweeter / & doesn't really react that great to a big horn ( I'll post that data later ).
• Motivated self interest could lead me to search for honk within this setup later this spring or maybe this summer .
• To date, I haven't ever heard any honk from this setup.
• FWIW, I too, identify CLASSIC HONK as existing at those frequencies occuring below 1K .

Here are 2 ARTA files ( zipped ) for those who want to slice & dice this data a bit more .

>< cheers

 

[Elias]

Hello panomaniac,

Please take a look of this wavelet plot from your 811-802 combination. On the right is my understanding of what happens after 2ms. There is a microphone stand reflection, small object reflection and 1st room reflection marked in the picture:

An externally hosted image should be here but it was not working when we last tested it.

Then let's compare with your ARTA CSD. I took the liberty of modifying your picture, hope you don't mind I marked the same phenomena in the the plot too.

An externally hosted image should be here but it was not working when we last tested it.

Interesting is the reflection just above 10kHz in the wavelet plot, which shows like a looong decay in the CSD. One might actually be fooled to think there is a strong resonance! Have to be careful with the CSD, or it will try to make a trick on you

The small object reflection can come from anywhere from the room, it can be anything, like a small photoframe on the table! That's why it's 'dangerous' to draw too many conclusions from measurement made inside, unless you know what is going on, like you know from the wavelet plot for example.

- Elias

FWIW, here is the ARTA sonogram from my 811-802D measurement.
Mine shows much more energy in the region of 1300Hz than yours does.

 

[Elias]

Hello Jean-Michel,

The level of the reflections/diffractions is interesting and we should find the level that matters.

If we look at what happens before 2ms and see the previously posted Altec horns:

An externally hosted image should be here but it was not working when we last tested it.

An externally hosted image should be here but it was not working when we last tested it.

An externally hosted image should be here but it was not working when we last tested it.

They all have reflections at -25dB or lower. Only less than 500us there is reflections higher than -20dB.


Also the Le Cléac'h J321 has -25 dB reflections, but of course they are less in numbers compared to others! Maybe it's the total amount that counts? The human perception contains an integrator.


- Elias

Hello Elias,

-25 dB ( = 5%) is not what can be called a strong reflection.


If the pale blue artifact you indicated at 1ms is due to a reflection this means that the reflector is at a distance of 34/2 = 17cm from the diaphragm.
This is exactly the double distance between the diaphragm of the TD2001 and the throat of the horn. This is a double refelection. (there is may be also on you wavelet graph a 0.5ms artifact corresponding to a single reflection between diaphragm and throat).

You'll find in attached file a schematic of the TD2001 showing the 8.5cm distance from diaphragm to throat.

The presence of a reflection at 1ms significates also that my measured Jerzual 321horn built by Marco Henry, is not perfectly fitted to the TD2001. The angle at throat of the horn is probably a bit different of the output angle of the driver. This can surely be improved as Martin Seddon does on its Azura Horns.

 

[Elias]

Great! A new term to figure out

"apodization time"

It's the time when CSD is pulling your leg and trying to make an apodization on you!

 

[Pano]

Elias - thanks for the plots. I have tried to chase down what the reflections might be. It's hard to determine. I did run a plot with a CD slipped over the mic to act as a nice round reflector. It showed up well as a reflection about 18cm away. The distance from the mic element to the CD and back.

I thought I had saved it, but looks like not. =( Easy to do again, tho. I'll keep chasing those reflections.

 

[EarlK]

This Horn Should Honk !

Very Nice Elias ,

There are certainly a few of us who are convinced of the usefulness of Wavelet Analysis when used with other analytical tools .

I think you need to commercialize the front end ( that you've written for Octave ) / so that the rest of us can get on board the "Wavelet Train " .

• Here's a CSD that you may find interesting ( using a very small Apodising Time setting ) .
• I'm sure there's lots of useless info ( for locating HORN HONK ) in this pic .
• The horn is a huge old RCA 60 degree radial / similar to the Altec 311-60 ( minus the mouth vanes ) .
• Rough dimensions of the horn are ;
  • Mouth ; 13H x 24.5W ( 33cm x @ 62.23 cm )
  • Depth ( to driver ) @ 29" ( 73.66 cm )
  • Depth ( to diaphragm ) @ 30" ( 76.2 cm )

Here are the Group Delay and FR plots for the same .

<> cheers

ps ; heres the ARTA file zipped .

 

[Pano]

New Waveguides.

My new assortment of waveguides/horns came this afternoon.
Got the Dayton 12" bolt on, the 8" and 10" rounds and the little 6x6" eliptical jobs. 6" x 6" Elliptical Waveguide
Will post some measurements soon.

It is easy to hear the difference in the horns just by holding them up to your ear and using them as an ear trumpet.

The deeper the horn, the more gain. But the more coloration, too. The 12" round guide is very clear, but is low gain.

I did put an 806 driver on this Dayton horn.

An externally hosted image should be here but it was not working when we last tested it.

With the same 840Hz HP as the 511-806 combo, it measures pretty much the same. However it does not sound the same. The 511 is darker and bigger sounding. The 12" dayton is brighter, even tho it does not measure brighter and needs to be turned down at least 3dB to match subjectively.

Subjectively I would say that the 511 has some honk, and the Dayton has glare. Tho the more open and detailed sound of the Dayton is often welcome, it can quickly get on my nerves. Somewhere in between the 2 would be nice. (811?)

As I said, I'll be doing some measuring and will post impulse files soon.

 

[mige0]

Here I attach a multiresolution wavelet tranform of the ideal 1ms reflections. It shows correctly the comb filtering of the fundamental wave, where as CSD fails to do this (as seen in your plot). Of course naturally the 1ms reflections are also present.

An externally hosted image should be here but it was not working when we last tested it.

Simply awesome !

CSD is a mathematical artefact.

I'd rather call it a "math transformation for visualisation" – as in a wider sense, IMO all math is kinda "arteficial" anyway


############

Let's come back to honkiness. The different CSD, Wavelets graphs and discussions until now have not succeed in defining its character.

IMHO:

1) honkiness is not an high frequency thing, much more important is to look for features appearing at f<1kHz. (even a Le Cléac'h horn can sound honk if the high pass filter used with it is set at a too much low frequency)

2) honkiness is not a low SPL level thing, much more important is to look for features appearing at relative level over -20dB

At high frequency, reflections, HOMs, etc. don't lead to honkness but to harshness, brighthness,....

Honkiness doesn't sound the same for an hyperbolic type horn and for a waveguide/conical horn. (I guess that large reflection from mouth to diaphragm charcterize the honkiness in waveguides/conical horns and that reactance increase nearby Fc dominate in the honkiness for exponential, hypex,...horns)

All of this is only the results of my own experience on horns.

Thanks for your summary on what you take as horn honk, very much appreciated – I consider myself to simply not have enough audition experience to link my impression to what "usually" is considered to be honk – with the exception of "guttural sound" that I strongly link to the group delay issue.

Also - over the course - I got the impression there is no wide agreement about any definition regarding "honk" - definition here not meant in any scientific sense – just no common sense to that semantic term as it seems – no?

Now the discussion, has evolved in a strange direction on the basis of your wavelets analysis. I guess readers could wrongly consider that the temporal effect is more audible than the frequency effect which is surely not.

I do not exactly follow your arguments.
If we look at this plot and agree that it shows us "kinda reality" we immediately see what I mean:

FR at t=0 is hell of a difference to after occurring massive reflection.
This is simply so as comb filtering does not happen *before* first delay time.

The consequence I see is that we will have hard times to equalize such a horn – either we get FR at t=0 (meaning before reflection) flat and FR after first delay time rugged – or vice versa.

Also – this same mis-behaviour / mis – consistency in time is repeated on every room reflection again !


I really do not think this very *time behaviour* of horn honk is of no concern.

Sure - there *is* medicine for this – but certainly not with simple (electronic) measures...


Michael

 

[Elias]

Hello Jean-Michel,

Don't worry, the reason I speak for the wavelets is because I have found them to be very useful and insight giving to understand more about the reasons behind observations.

I don't consider CSD to be close to what we (auditorily) perceive. There is no such thing build into human brain like CSD perception. Instead, I think wavelets are closer to realism. They have time-frequency information at the same portions like music signal has.

Your honk definition is interesting, as you suggest it being a low freq phenomenon. What do you think is the best parameter to define it? Group delay? Frequency response variation? Directivity?

About the discussion in this thread, I think it's evolving in a very good direction because we are getting to know more what is going on in a horn in both frequency and time domain. If we are going to rank which domain is 'more audible' we are deep into psychoacoustics, and no definite answers can be given.

Let's just remember that frequency domain is not reality. Frequency is a mathematical transformation from the time domain. Time domain is where we live in. Temporal effects are the cause, frequency effects are the consequence. I think it is better engineering to remove the cause than tame the consequences. Wavelet, if so defined, will give good temporal visibility and thus indicate which is the cause of an observed phenomenon. I think!

- Elias

I don't want to dismiss your work for sure. I simply doubt that wavelets will be the unique solution to the problem of how honkiness can be seen and analyzed.

For me the reality of both wavelets, CSD, Wigner-Deville,etc... is all the same. They are tools to help us to understand how our acoustical devices behave. If we use them correctly both those methods can give similar results.

I don't take in account the graphical appearance in itself. The important thing is the information content, not the beauty of the graph.(That's why when I use CSD myself I avoid smoothing). CSD in ARTA may be consider by many as little games, they don't look good, but the information it contain is fully reliable and can be associated to what we perceive.

Let's come back to honkiness. The different CSD, Wavelets graphs and discussions until now have not succeed in defining its character.

IMHO:

1) honkiness is not an high frequency thing, much more important is to look for features appearing at f<1kHz. (even a Le Cléac'h horn can sound honk if the high pass filter used with it is set at a too much low frequency)

2) honkiness is not a low SPL level thing, much more important is to look for features appearing at relative level over -20dB

At high frequency, reflections, HOMs, etc. don't lead to honkness but to harshness, brighthness,....

Honkiness doesn't sound the same for an hyperbolic type horn and for a waveguide/conical horn. (I guess that large reflection from mouth to diaphragm charcterize the honkiness in waveguides/conical horns and that reactance increase nearby Fc dominate in the honkiness for exponential, hypex,...horns)

All of this is only the results of my own experience on horns.

Now the discussion, has evolved in a strange direction on the basis of your wavelets analysis. I guess readers could wrongly consider that the temporal effect is more audible than the frequency effect which is surely not. IMHO in your example, the comb filtering effect (or whatever effect - not only "wide frequency range"- due to the interference between a direct wave and a reflected wave) with SPL peaks at 1, 2, 3, 4, 5 , 6, 7kHz, ... is what we perceive (in the precise example you took), not the 1ms time separation between reflections. That's why it is important to use an analysis method that doesn't favor the temporal appearance over the frequency appearance. Also better to use a method that can bring information at frequency below 1kHz.

Best regards from Paris, France.

Jean-Michel Le Cléac'h

 

[mige0]

An externally hosted image should be here but it was not working when we last tested it.

An externally hosted image should be here but it was not working when we last tested it.

An externally hosted image should be here but it was not working when we last tested it.

An externally hosted image should be here but it was not working when we last tested it.

From above – lets not forget that comb filtering effects form horns are as complex as the involved horn contours itself (*including* the compression driver just like JMLC has outlined) – meaning towards low frequencies there *always* will be some limitation as horn dimensions get too small to create "text book" reflection at mouth – and depending on points of strong diffraction placed in space (manta ray example comes into mind as an extreme), there are different delay times involved.

For more complex contours things get mixed more and more but also "fade out" the issue of reflection (the alignment of diffraction I / we are after ).


So we not necessarily have to look at mic stand reflection unless its clearly outside the "clean" measuring window - do not know Panos setup...

Michael

 

[ZilchLab]

It would be interesting if someone could post the impulse response of an OS waveguide....

Geddes posted some in the Horn vs. Waveguide thread, early pages. You'll see everyone doing their favorite analyses of them there....