This is what I got after some simulations.
Some "weird" wiring and some EQ to flat the response. The resulting impedance is rather low (around 4 Ohms).
The vertical directivity seems to have improved a lot even if after 5k is still raising too much.
The in-room response has been evaluated considering floor, ceiling and front wall attenuated by 10dB.
I think that we can do much better than this
Manuele
Some "weird" wiring and some EQ to flat the response. The resulting impedance is rather low (around 4 Ohms).
The vertical directivity seems to have improved a lot even if after 5k is still raising too much.
The in-room response has been evaluated considering floor, ceiling and front wall attenuated by 10dB.
I think that we can do much better than this
Manuele
What you did there is power-shading, not my favorite kind as one driver is lifting way harder than all the rest.This is what I got after some simulations.
View attachment 1060747
Some "weird" wiring and some EQ to flat the response. The resulting impedance is rather low (around 4 Ohms).
View attachment 1060748
The vertical directivity seems to have improved a lot even if after 5k is still raising too much.
The in-room response has been evaluated considering floor, ceiling and front wall attenuated by 10dB.
I think that we can do much better than this
Manuele
It is the easiest way to widen the pattern on top though, but the driver that makes the top end beautiful is also the one
that lift the hardest on the bottom end. That's not ideal in my view to keep dynamic potential high.
Whatever solution I'd use would make sure all drivers have about an equal share to lift at the bottom end.
IR of the 25 driver array I'm building at the moment. 25x Scan Speak 10F 8414 G10.
Belonging to this graph:
This simulation is a combination of power and frequency shading because I'm gradually low-passing the high frequency as the distance from the Center driver increase. I agree that is not ideal this is why I said “we can do much better than this” 😜😜.
Manuele
It's easier to try just about anything that comes to mind within the simulations. I ran more than 60 simulations before calling it ready for my own project.
The hard part is getting that wider top end without making a single or a pair of drivers work much harder than the rest.
But even within a group of series driver you can filter them, that's what I used in my own filtering to be able to stay with 5 groups of 5 drivers.
In my use case I just needed to make a power dip on a few drivers at a certain frequency, so I used a notch, but you can mimic the use of the coil in
the series of drivers by placing a capacitor parallel to a single driver in the series (shading off it's to end). That way you can do more with the 3x 3 schematics.
The hard part is getting that wider top end without making a single or a pair of drivers work much harder than the rest.
But even within a group of series driver you can filter them, that's what I used in my own filtering to be able to stay with 5 groups of 5 drivers.
In my use case I just needed to make a power dip on a few drivers at a certain frequency, so I used a notch, but you can mimic the use of the coil in
the series of drivers by placing a capacitor parallel to a single driver in the series (shading off it's to end). That way you can do more with the 3x 3 schematics.
The room?
Bit of a tasteless answer, isn't it? But seriously, what matters to me is getting the direct sound as right as possible, cleaning up the first ~20 ms
so I can get a good shot of listening to the queues within the music. Bringing down those first reflections (with absorbing panels) automatically
cleans up everything else, creating a good direct vs indirect mix. I don't want a dead room, I even introduce virtually created late queues
(of a space larger than mine) to fill in what I stole from the room. Main purpose: hide the real room but don't cloud the queues of the recordings.
My mission: getting a believable sound, transporting me to the recording, not bringing them in to my (too cramped) space.
What that looks like can be found in my thread as it isn't that different from the unshaded array:
Here you can see the direct sound and the room's response. Still not up to 100 ms, but it does illustrate my point.
The damping (absorbing panels on first reflection points) is chosen to balance the room response, making it mimic the direct response in balance.
(but slowly declining the higher in frequency one goes, as can also be seen in the DI curve)
It was my attempt to create a RFZ (reflection free zone) although in all honesty one could better name it a reflection reduced zone in a real living room.
But it does work due to the behavior of arrays within the room (omitting floor and ceiling as sources of reflections).
All that and way more can be found here...
Bit of a tasteless answer, isn't it? But seriously, what matters to me is getting the direct sound as right as possible, cleaning up the first ~20 ms
so I can get a good shot of listening to the queues within the music. Bringing down those first reflections (with absorbing panels) automatically
cleans up everything else, creating a good direct vs indirect mix. I don't want a dead room, I even introduce virtually created late queues
(of a space larger than mine) to fill in what I stole from the room. Main purpose: hide the real room but don't cloud the queues of the recordings.
My mission: getting a believable sound, transporting me to the recording, not bringing them in to my (too cramped) space.
What that looks like can be found in my thread as it isn't that different from the unshaded array:
Here you can see the direct sound and the room's response. Still not up to 100 ms, but it does illustrate my point.
The damping (absorbing panels on first reflection points) is chosen to balance the room response, making it mimic the direct response in balance.
(but slowly declining the higher in frequency one goes, as can also be seen in the DI curve)
It was my attempt to create a RFZ (reflection free zone) although in all honesty one could better name it a reflection reduced zone in a real living room.
But it does work due to the behavior of arrays within the room (omitting floor and ceiling as sources of reflections).
All that and way more can be found here...
Last edited:
Those results are promising, with a bit of EQ I can have a reasonable sensitive speaker with a good vertical response. If possible, I would also like to think of a way to make the vertical directivity even more controlled between 3 and 10 kHz, by reducing the "wings" visible in the graph.
A couple of days ago, messing around with the simulations, I realized that Vituixcad has an optimization feature.
I quickly tried to run some simulations and it is possible to optimize several parameters, such as the position of the drivers (in x, y, z) as well as all the values of the passive crossover elements.
I tried to find the best compromise for both the listening windows and power response (70% and 30% weight respectively) to see if vertical directivity could benefit from the optimization, but I was unsuccessful.
I started with a "wide" optimization leaving the software free to act both on the drivers position and the crossover values but the resulting response was not at all attractive.
I then refined the optimization to the crossover parameters only, but I have to repeat the simulation because I didn't really have the opportunity to do it correctly.
@kimmosto @wesayso Do you know if there is a way to optimize only the vertical directivity by exploiting this function of Vituixcad?
Thanks,
Manuele
I wouldn't have that answer for you. I've always done it the hard way, by changing parameters myself.
The so called wings are a byproduct of center to center spacing though. It might be possible to change/
optimize that behavior by changing driver positions.
The picture above is the central drivers only, so a 3 driver array. As can be seen this setup is responsible for that triangle
shape. So playing with position won't do much, playing with the exact values of the inductors might show benefit.
Or using a more elaborate combination of filters to shape the output of the outer 2 of the 3 drivers.
The so called wings are a byproduct of center to center spacing though. It might be possible to change/
optimize that behavior by changing driver positions.
The picture above is the central drivers only, so a 3 driver array. As can be seen this setup is responsible for that triangle
shape. So playing with position won't do much, playing with the exact values of the inductors might show benefit.
Or using a more elaborate combination of filters to shape the output of the outer 2 of the 3 drivers.
I was wondering how much these "wings" can bother listening. My understanding, as a layman and DIYAUDIO novice, is that by reducing the "wings" it is possible to reduce the harmful interactions with the ceiling and floor in those frequencies that are really critical to the human ear, improving the response in the room.
I was taking a look at the theoretical fractal array https://www.diyaudio.com/community/...raight-cbt-with-passive-xos-and-no-eq.330031/ and with my surprise, using Vituixcad's built-in optimizer, the resulting drivers layout is very similar to the Fractal Array's proposal
.
A beginner question: doesn't the use of inductors and capacitors compromise part of the beauty of Full Range drivers to operate without passive components while maintaining coherence and excellent impulse response? Is it something that can be recovered via DRC FIR?
To conclude, my next steps will look like these:
Manuele
I was taking a look at the theoretical fractal array https://www.diyaudio.com/community/...raight-cbt-with-passive-xos-and-no-eq.330031/ and with my surprise, using Vituixcad's built-in optimizer, the resulting drivers layout is very similar to the Fractal Array's proposal
.A beginner question: doesn't the use of inductors and capacitors compromise part of the beauty of Full Range drivers to operate without passive components while maintaining coherence and excellent impulse response? Is it something that can be recovered via DRC FIR?
To conclude, my next steps will look like these:
- Perform some more advanced BEM simulations, using my actual speaker boxes, to understand a little more of the behavior in my real room (not trivial!). I could use AcouSTO (http://acousto.sourceforge.net/), an open source software developed by one of my university professors.
- Play a little more with the position and horizontal angle of the individual drivers as well as simulate some "minimally invasive" filters
- Complete my actual boxes and measure the real performances! (absolutely the best thing to do but the most complicated from time point of view)
Manuele
Don't need to worry about coherence of the impulse and STEP if they still look good after EQ-ing the frequency response flat. You can view it within Vituixcad.
View -> Impulse response -> Total SPL.
Indeed DRC-FIR could reconstruct the IR if needed, but in my 25 driver case, the impulse of the filtered array I use now is closer to ideal than the unfiltered array.
Indeed those lobes that stick out vertically will cause reflections from floor and ceiling in most usual cases. A slanted ceiling may act different, but in these quick samples
I have included floor and ceiling reflections and it is visible in the gray "in-room" prediction in the middle left picture. (the zig/zag top end from 3 to 12 KHz)
The ceiling is placed at 2.5 meter height, distance to speaker 5 meter. One could determine the off axis angle using those dimensions
in a simple sketch. Floor is 1 meter below the microphone, that makes the ceiling 1.5 meter above it. Usually an 1 meter listening
height (ear height when seated) is about right.
Of the steps you mention, getting the measurements done would really help. That fractal array is a thing of beauty, and can be a nice way
to go if you like the form factor it brings. It is a clever and clean solution of an expanding array with exemplary vertical behavior.
This discussion is like deja vu to me and Wesayso. Frequency shading the arrow cures the combing and reduces the barbs on the arrow head of the polar response but at the cost of narrowing the vertical directivity. Shortening the array does the same thing. Wesayso's filters for his own arrays were a good compromise: he reduced the combing is much as possible while keeping the vertical window as tall as he needed it. I did some simulation experiments that eliminated the combing almost completely but resulted in a very narrow window suitable for seated only listening. I think that is what you will have with only 8 drivers, regardless of frequency tapering. If you accept that - seated only - you really just need to optimize response over perhaps +/- 150 mm from nominal ear height.
The 3dB/octave drop still is (and should be) there in my modified/filtered array. As the array still is large enough at each frequency to maintain array behavior.
Look at https://www.diyaudio.com/community/threads/infinite-line-source-analysis.314917/ for the proper math to prove it or the quote below to realize that the 3dB drop per octave isn't related to combing, but simply array behavior:
quite close, thanks to using clean drivers without a ragged top end and the powers of DSP to correct their deviation.
Filtered array with the 3dB/octave drop showing when we remove the EQ. Don't mind any value below 200 Hz, or above 10 KHz
as this information is lacking from the ABEC simulations used for this particular model. Vituixcad will extrapolate the missing data.
The short arrays fall somewhat short on array behavior though. Which is why the tall arrays still are my favorite.
About combing? Not really a problem with the filtered array up to nearly 10 KHz. One could even argue that more combing is heard from
a single driver speaker in a normal regular room with a floor and a ceiling than with an array like the one I have in that same room.
Just look at the orange frequency curve for the difference between a single driver at the listening spot and an array
at the same spot. The restlessness that is left within the filtered array FR curve (5 to 10 KHz) is the floor and ceiling reflections of
the (remaining) lobes due to driver spacing.
While the array tends to hold the frequency response one sees here when moving up, down and left or right, the single driver will
still show these quite large dips in frequency at ever changing (frequency) points, depending on which direction one moves to.
Anyone that has made measurements of a speaker or driver within a room will be familiar with that, yet an array is way less prone
to show large deviations like that which is the reason for me to stick with this concept and spend this much time pursuing even
better results (however small it may be). You'd need the room to cooperate though. That fact seems to be missed at times but
I deem it essential for good and predictable results. FIR EQ tools too are essential to get that last bit out of it. If you do, it really is
a wonderful concept producing mighty clear sound.
Look at https://www.diyaudio.com/community/threads/infinite-line-source-analysis.314917/ for the proper math to prove it or the quote below to realize that the 3dB drop per octave isn't related to combing, but simply array behavior:
In other words: even a perfect seamless source array would still show this 3 dB/octave drop. Our less that perfect array comesPOST #13
C. Infinite Line Source: frequency-domain pressure response (continued)
Our 'normalized transfer function' for the Infinite Line Source (for kr>>1) is:
Hn(kr) ~= SQRT[2/(pi*kr)]
It's worth comparing this expression to a 'normalized transfer function' for the Point Source (developed around page 2, ignoring the time-delay and 'rho' scaling):
Hn(kr) = 1/(4pi*r)
Two interesting aspects of the 'transfer function' for the Point Source:
1. It's independent of frequency ... the frequency response magnitude, measured at any fixed point in space, is FLAT (no dependence on 'k' or 'w')
2. As we increase distance from the Point Source, the response falls off at 6dB for every doubling of distance
How does the Infinite Line Source compare (for kr>>1)? That "square root of 1/kr" relationship in the 'transfer function' tells us two things as well:
1. The frequency response magnitude measured at any fixed point in space is NOT flat ... in fact, it falls off at 3dB for every doubling of frequency
2. As we increase distance from the Infinite Line Source, the response falls off at only 3dB for every doubling of distance
While the first point is rather unfortunate, and would ultimately call for some sophisticated equalization (FIR much preferred, since 3dB/octave is challenging with IIR) and healthy dynamic range in the driving electronics (assuming we can actually build something like an "infinite" line source
It's worth noting that ALL of this has been done, well before me. If i can claim anything at all innovative, so far, it may only be my 'transfer function' view of this domain of acoustics. Also, i certainly expect more pros & cons to be debated as the thread develops
quite close, thanks to using clean drivers without a ragged top end and the powers of DSP to correct their deviation.
Filtered array with the 3dB/octave drop showing when we remove the EQ. Don't mind any value below 200 Hz, or above 10 KHz
as this information is lacking from the ABEC simulations used for this particular model. Vituixcad will extrapolate the missing data.
The short arrays fall somewhat short on array behavior though. Which is why the tall arrays still are my favorite.
About combing? Not really a problem with the filtered array up to nearly 10 KHz. One could even argue that more combing is heard from
a single driver speaker in a normal regular room with a floor and a ceiling than with an array like the one I have in that same room.
Just look at the orange frequency curve for the difference between a single driver at the listening spot and an array
at the same spot. The restlessness that is left within the filtered array FR curve (5 to 10 KHz) is the floor and ceiling reflections of
the (remaining) lobes due to driver spacing.
While the array tends to hold the frequency response one sees here when moving up, down and left or right, the single driver will
still show these quite large dips in frequency at ever changing (frequency) points, depending on which direction one moves to.
Anyone that has made measurements of a speaker or driver within a room will be familiar with that, yet an array is way less prone
to show large deviations like that which is the reason for me to stick with this concept and spend this much time pursuing even
better results (however small it may be
). You'd need the room to cooperate though. That fact seems to be missed at times butI deem it essential for good and predictable results. FIR EQ tools too are essential to get that last bit out of it. If you do, it really is
a wonderful concept producing mighty clear sound
.
Last edited:
My original premise was to low pass each driver at the frequency where its contribution at the LP arrived 90 degrees out of phase compared to that from the driver at array center. If you do this you get zero combing but it requires an amp and filter per driver and results in that very short vertical window. Therefore, you have to compromise so the curing is no longer absolute. You divide the drivers into groups and apply that 90 phase path length difference criteria to each group as a starting point for fine tuning. The center group is unfiltered and doesn't satisfy the path length difference criteria for its outer drivers. You see in Wesayso's last post, he has combing visible in the frequency response only in the top octave. If you enforced that path length distance criteria up to the very highest frequencies, you would end up with only a single driver playing in that highest range. Thus the ever present desire for an array with a tweeter at center. But it turns out to be quite difficult, perhaps impossible ,to match directivities between that tweeter and the rest of the array.