Each different way of viewing measured data has it's own strengths and weaknesses. How one interprets data in a way to quckly improve a design takes some understanding of the usefulness of each.
Elias said:
- I believe I can readily tell the difference / but who knows if we all employ the same vernacular ?
- I believe it is level dependant .
- Overlaid , normalized ( Multiple ) IR plots ( taken in 5 db increments over a 40 db range ) ought to show the onset of these non-linearities within the Group Delay trace .
- Specific "Honk" frequencies ought to show a rapid increase in GD as the critical threshold is reached ( anyhow, so goes my conjecture ).
<> EarlK
Well it's an interesting conjecture.
I can run 110-120dB sweeps if need, tho I'd really rather not.
Another "FWIW". It's been awhile since I played music thru the 511B+806A, so I thought I should. If crossed 2nd order Butterworth at 700Hz and with a shelving HP to tip up the top end, it's pretty flat. Acoustically it is equal to a 3rd order HP at 700Hz and a 4th order LP at 15Khz. Pretty easy to match up to a woofer. (at least on axis). So I did.
It doesn't suck. Certainly not the "In Your Face" sound that the stock Altec crossover has. Non aggressive. But it sure does have that "woody" sound that ZilchLab noticed. It does not have the light, open sound of the multi-cell, even if it measures better. Nice, but not world class. Maybe I should compare it to a full-range driver covering the same band.
Very likely. I'll do some test to try to figure it out.
I'm not sure I see that. But it could be the DAC and/or the ADC.
Hello Elias,
You are wrong, Hornresp is perfecly able to simulate the reflection inside the horn.
I once gave as an example of the ability of Hornresp to predict the reflections the CSD of the pulse response of a tube (you'll find it in attached file).
Results of BEM has also demonstrated that the Le Cléac'h horn has the lowest reflection coefficients at frequency above 1.5 Fc.
Most large spectrum reflections you see resides inside the conical part of the output of the TAD TD2001 itself.
The problem for which I am reluctant to use my own wavelet software is that this method leads to a bad equilibrium between temporal and frequency effects of irregularities. That's evident when we study the breaking of a cone (e.g). Large Q features appear as having a widen frequency range with wavelets...
I largely prefer CSD which allows a better understanding of the origin of the problems.
Best regards from Paris, France
Jean-Michel Le Cléac'h
You are wrong, Hornresp is perfecly able to simulate the reflection inside the horn.
I once gave as an example of the ability of Hornresp to predict the reflections the CSD of the pulse response of a tube (you'll find it in attached file).
Results of BEM has also demonstrated that the Le Cléac'h horn has the lowest reflection coefficients at frequency above 1.5 Fc.
Most large spectrum reflections you see resides inside the conical part of the output of the TAD TD2001 itself.
The problem for which I am reluctant to use my own wavelet software is that this method leads to a bad equilibrium between temporal and frequency effects of irregularities. That's evident when we study the breaking of a cone (e.g). Large Q features appear as having a widen frequency range with wavelets...
I largely prefer CSD which allows a better understanding of the origin of the problems.
Best regards from Paris, France
Jean-Michel Le Cléac'h
Attachments
Hello Michael,
A CSD should not show steps.
Here is, in attached file, the CSD of the .pir you gave the link of which - as I consider it has to be-.
I don't recommand to try to increase the temporal resolution or the frequency resolution of the CSD. It is as if all CSD (fro a given probleme and this means ainside a temporal window) possess a unic quantity of information and there is an optimal time/frequency resolution adapted to you studied problem. If you try to escape this rule, artifcats as I can see in most of CSD shown in this thread appear (only Catapult's CSD is OK for me ...)
Best regards from Paris,
Jean-Michel Le Cléac'h
A CSD should not show steps.
Here is, in attached file, the CSD of the .pir you gave the link of which - as I consider it has to be-.
I don't recommand to try to increase the temporal resolution or the frequency resolution of the CSD. It is as if all CSD (fro a given probleme and this means ainside a temporal window) possess a unic quantity of information and there is an optimal time/frequency resolution adapted to you studied problem. If you try to escape this rule, artifcats as I can see in most of CSD shown in this thread appear (only Catapult's CSD is OK for me ...)
Best regards from Paris,
Jean-Michel Le Cléac'h
Attachments
Last edited:
Re: Horn honk linearity
I think it was conjectured by Dr. Geddes that while horn honk / HOM or whatever is a linear phenomenon, it is the perception of such distortions that is level dependent. They did a study on group delay perception and level dependence was shown. Not very surprising really knowing that many other aspects of hearing are level dependent. Even pitch perception is level dependent!
I think it was conjectured by Dr. Geddes that while horn honk / HOM or whatever is a linear phenomenon, it is the perception of such distortions that is level dependent. They did a study on group delay perception and level dependence was shown. Not very surprising really knowing that many other aspects of hearing are level dependent. Even pitch perception is level dependent!
Well so far I have not concluded. I've gone back and forth, back and forth between the plots.
The driver makes more difference, it seems.
But I do like the sound of the 811 better. It seems a little cleaner and clearer. Maybe it's a polar response thing.
These comparisons are not "usefull"?
http://www.diyaudio.com/forums/multi-way/161627-horn-honk-wanted-21.html#post2118966
That is a perfect post showing what combo worked...
Heck, we can see which one is better from all these....
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.
and then Le Cleach....Damn that is nice!!!
An externally hosted image should be here but it was not working when we last tested it.
Graphing it
When generating graphs like these it is easy to get different looking results from the same file. Using ARTA, for example, I can even get upside down results depending on where the cursors are set. See the 2 CSDs below of the same impulse. (700Hz, 2nd, L-R, high pass).
And with same info the graphs look different using different scales. The same problem exists for all graphs. Advertisers use this to great advantage.
A CSD of sonogram on a 15dB scale looks much different then on a 40dB scale.
So. We should standardize, at least in this thread. I suggest we use.
I'm open for suggestions. If we are all doing graphs in the same software(s) we should at least all be doing them the same.
When generating graphs like these it is easy to get different looking results from the same file. Using ARTA, for example, I can even get upside down results depending on where the cursors are set. See the 2 CSDs below of the same impulse. (700Hz, 2nd, L-R, high pass).
And with same info the graphs look different using different scales. The same problem exists for all graphs. Advertisers use this to great advantage.
A CSD of sonogram on a 15dB scale looks much different then on a 40dB scale.
So. We should standardize, at least in this thread. I suggest we use.
- 40dB scale. This shows deep enough without missing anything important.
- We decide how far (in milliseconds) before and after the impulse peak to set the gate or cursors.
- I suggest 1mS before and 5mS after. But am open to suggestions.
- Other parameters should be set, such as window type, FFT, ect.
I'm open for suggestions. If we are all doing graphs in the same software(s) we should at least all be doing them the same.
Attachments
doug20 said:
- I like very much the Wavelet analysis(es) that Elias has displayed / they are quite useful. Though I don't have a clue as to how to produce them .
- Wavelet analysis is not the same as CSD analysis as you've implied by quoting me .
- The question of the usefulness of analysing CSDs remains & needs to be further explored .
- JM has just stated his position that within certain confines they (CSDs ) do show useful information . He knows better than I , so I'll defer to him .
<> EarlK
Very true....the Wavelet does has CSD properties though (like time decay) and I look at them the same way I do a CSD.
I think people can choose to use what they like. Alone none of this stuff would work for anyone but as part of a whole they can be great.
Besides who doesnt like pretty graphs...its better then text based line charts
As for producing them, I have Octave installed, I can generate plots from wave files. I could not find out how to do a Wavelet though. I googled about it and there some links to WaveLab, so I installed that as a package in Octave but no luck after that. Just zero examples out there in google world
I wonder if Elias wrote a little program to do it all. Octave is a pretty cool little language/environment. Im a python developer, well I management more then I code these days but I can muddle through it. I only spent about 2 hours though....I have to spend more time measuring my new speakers instead.
Last edited:
Well 5 ms seems to me to be the longest period after the peak. I'd go shorter, if it does not skew the results. 3ms takes us out to 1 meter.
Maybe to use the inpedance curve for (quasi)free room reflection.
And maybe a tube between!
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.