That's not inevitable, you can have a nice flat impedance, but it will beam badly.You might want to increase "loading", which inevitably makes wavy impedance graph [...]
Yeah it's always a multifaced problem, I can imagine it's not that easy to define what to optimize for. Beaming can be done without loading as well, with circular arc profile, as an example.
After several years of "optimization" and taking all kinds of things into consideration, I have the Gen2 horns - that's in a way my answer.
For sure, can't expect anything less, as much of a trade secret as any. Perhaps some other hobbyists have tips to share.
Well, if one line isn't good enough to rate a waveguide, I'll probably rate the -1 -3 -6dB lines. It'll be just a little bit more work to rate three lines instead of one. Ideally I'd like a 'true' constant directivity waveguide anyway.
My scripts were also done mostly with ChatGPT even though I know just enough Python to do it myself. It's just quicker for simple things like that. 😀
I've added the script for rating the radiation impedance to the zip. If you want to save a few seconds, you can copy most of my run_abec_fast.py script. Might not make much sense for circ symmetric simulations, but for longer ones it's going to be necessary. It's really easy, too.
I'll try to adjust my scripts in the next days so I can run Bayesian optimization automatically.
My scripts were also done mostly with ChatGPT even though I know just enough Python to do it myself. It's just quicker for simple things like that. 😀
I've added the script for rating the radiation impedance to the zip. If you want to save a few seconds, you can copy most of my run_abec_fast.py script. Might not make much sense for circ symmetric simulations, but for longer ones it's going to be necessary. It's really easy, too.
-extract in the ath.exe folder
-edit the config.ini with your file paths
-run _pixelpos.py to set up the coordinates of the pixels to check for ABEC progess
-edit the base_template.txt to your liking. Use double {{}}, as single {} are for the variable parameters
-edit params.ini
-run start_the_whole_thing.py
To rate the Re Impedance first run 'rename_radimp.py' so the filename contains the parameters used. Then run 'rate_radimp.py'. If you like to change the rating weighting or algorithm, do so in the rate_radimp.py. To remove the numbers from the rating script run 'undo_radimp.py' before.
-edit the config.ini with your file paths
-run _pixelpos.py to set up the coordinates of the pixels to check for ABEC progess
-edit the base_template.txt to your liking. Use double {{}}, as single {} are for the variable parameters
-edit params.ini
-run start_the_whole_thing.py
To rate the Re Impedance first run 'rename_radimp.py' so the filename contains the parameters used. Then run 'rate_radimp.py'. If you like to change the rating weighting or algorithm, do so in the rate_radimp.py. To remove the numbers from the rating script run 'undo_radimp.py' before.
Attachments
I just want to share the measurements I took with my passive crossover prototype for the g520extended with La Voce DF10.171K.
I intentionally left the rise on the high end of the spectrum. This crossover has a 3rd order filter, followed by parallel notch filter and a shelf filter. I aimed for flat and 800hz crossover.
The measurements are taken at around 1m and gated 5ms and smoothed 1/12. The off axis angles were done without measuring roughly 10degrees each
up to 60.
I intentionally left the rise on the high end of the spectrum. This crossover has a 3rd order filter, followed by parallel notch filter and a shelf filter. I aimed for flat and 800hz crossover.
The measurements are taken at around 1m and gated 5ms and smoothed 1/12. The off axis angles were done without measuring roughly 10degrees each
up to 60.
Hi, It's thrilling but still kinda lot of work even though AI helpers can do heavylifting. This could somewhat undermine mabat possibilities to monetize his work to some extent, so not sure if such thread is entirely good idea... I'm doing this for personal project, purhased one of his and tried to persuade mabat do commercial for my target but no wind, which is understandable as we have different motivations, so no other option but to try to develop my own optimization algorithm 🙂 Sharing on a concept level is fine I think, but distributing complete code perhaps not. Although, even with finished shared codebase not everyone are able to run it so, well, what ever mabat feels about it.This is thrilling... why not open up a new thread "ATH auto optimisations" ... ? 🙂
Would be interesting to follow....
//
Please go ahead, maybe you manage to do something I couldn't 🙂
What was the target? I may have missed that.tried to persuade mabat do commercial for my target but no wind
I would try to dampen down the peak just above 1 kHz. This is typically possible by changing the values of the parts a bit, without a topological change in the crossover, and in a simulator this should be an easy task.I just want to share the measurements I took with my passive crossover prototype for the g520extended with La Voce DF10.171K.
BTW, for a 800 Hz crossover I think you could use a short adapter instead as well: https://at-horns.eu/gen2m.html#df10171k
It may be much easier to do passively than the long one. I would definitely try the short first in that case.
Last edited:
It was the "practical constant directivity", where listening window, say +/-20deg or so are all flat. Idea is to have same sound for the whole listening window, a practical constant directivity. Examples of "real constant directivity", where DI is flat, doesn't seem to provide it, as highs rise in level in listening window to make DI flat, which makes the listening window not constant directivity. This is my quest for uniform sound on the listening window, so if it doesn't make commercial sense to you I'll do it for my personal project myself.Please go ahead, maybe you manage to do something I couldn't 🙂
What was the target? I may have missed that.
Post in thread 'Acoustic Horn Design – The Easy Way (Ath4)' https://www.diyaudio.com/community/...-design-the-easy-way-ath4.338806/post-7829699
This is not "flat enough" for you (the blue trace is at +-20deg)?
It's roughly -2 dB from 700 Hz to 10 kHz for 20 deg. This can be reduced of course, but there's a cost (which I wouldn't take).
It's roughly -2 dB from 700 Hz to 10 kHz for 20 deg. This can be reduced of course, but there's a cost (which I wouldn't take).
Actually what's my first wish is bit wider pattern than A460G2 and more constant. You published intent for constant directivity waveguide here https://www.diyaudio.com/community/...-design-the-easy-way-ath4.338806/post-7814678 but only for 1" so I started to roll my own. And if you look at listening window response for the constant DI device it's not constant but rising, and what I'd like is have it constant, middle of the two examples you have. Screenshot and illustration from behind the link.

While these all likely sound great, it's just that if I need to roll my own I'll do it exactly how I want it since it's about the same trouble and cost to build. Since I do not know any trade-offs not sure what those are. I think it's just a device between the two and with ~same compromises as these have, which are about none.

While these all likely sound great, it's just that if I need to roll my own I'll do it exactly how I want it since it's about the same trouble and cost to build. Since I do not know any trade-offs not sure what those are. I think it's just a device between the two and with ~same compromises as these have, which are about none.
That's more or less the approach I'm using in the genetic algorithm I developed for this - although only for the -3dB and -6dB lines.Well, if one line isn't good enough to rate a waveguide, I'll probably rate the -1 -3 -6dB lines. It'll be just a little bit more work to rate three lines instead of one. Ideally I'd like a 'true' constant directivity waveguide anyway.
Finding a good "scoring" system is the bulk of the work. I'd recommend weighting the frequencies in the 800-16khz range more as that should be the priority for optimization - at least to begin with. I also experimented with applying a slightly harsher "penalty" for a falling DI as that allowed me to get better results with less iterations.
A flat listening window may be a better target than the flat DI, which was obviously my target for that particular design (A460D). In fact, it's somewhere in between in the end and I tend to think about it as "in tolerance" (+/- 1dB for 800 - 16k everywhere in the listening window).
Last edited:
Yeah, it enables tonality of direct sound stay same when toe-in is changed for example, or if people moves in the room, compared to system where the listening window response tilts either way per angle. For example I have quite short listening distance in a domestic room, if I lean back at main listening position I have different angle to speakers than if I lean forward, and tonality changes. I want this be more constant. There is lot's of reasons why tonality changes moving in the room closer / further from speakers, like at which azimuth angle the speakers are to head and head shadowing changes things, early reflections change a bit, and so on, so perhaps this is not the ideal either. It's still something that I think would be great at the moment, at least until I've tried it 🙂
I suspect you won't be satisfied with this kind of LW flatness but this is all I can do without a considerable more effort (a 1.4" throat).
But I also think you already have someting close (also based on the Gen2).
(per 5 deg)
But I also think you already have someting close (also based on the Gen2).
(per 5 deg)
Hi, yeah thats about where I'm at currently 🙂 Will try evolution algorithm to it. It's really close though, but G2 is bit smoother still so wanted to try 🙂
- Home
- Loudspeakers
- Multi-Way
- Acoustic Horn Design – The Easy Way (Ath4)