The ABEC simulations here show that if you want constant directivity, you shouldn't use the modelled U-frame much beyond 100 Hz.
That is how it's used in the Bitches Brew, with/without subs depending on how loud it is to be played and the size of the venue.
"My" single 15OB350 dipole is planned to be used from 80 or 100 Hz up to 400 Hz. It maintains the dipole nulls past 400 Hz as shown in its polar maps that follow
BB
Below the dipole will be a shallow mount 12" sealed sub, which will no doubt give me more than 6 db over the U-frame at 30 Hz. My innovation, insofar as it is that, is only to use a shallow mount sub facing up, flat on the floor, just behind the dipole, taking up just 116 mm of vertical space.
That is how it's used in the Bitches Brew, with/without subs depending on how loud it is to be played and the size of the venue.
"My" single 15OB350 dipole is planned to be used from 80 or 100 Hz up to 400 Hz. It maintains the dipole nulls past 400 Hz as shown in its polar maps that follow
Below the dipole will be a shallow mount 12" sealed sub, which will no doubt give me more than 6 db over the U-frame at 30 Hz. My innovation, insofar as it is that, is only to use a shallow mount sub facing up, flat on the floor, just behind the dipole, taking up just 116 mm of vertical space.
These are super interesting comments. Why is it that you find the 200-300 Hz range more problematic? The pathlength difference being under 4 msec would be the same for all frequencies, so curious why the same thing--say at 500 Hz--would be less of a problem.
With @perrymarshall's Bitch's Brews I believe there's significant overlap between the upper and lower SB 15's... would that make it behave a bit more like the stacked vertical array of WWWW you described (and which I've seen done in OB elsewhere)? Would that suggest that the Birch version with nothing in the middle of the baffle might have less of this potential for smearing (or however you'd describe the impact), or is it partially negated by 200-300 Hz coming from even higher up the baffle?
I'm wondering how things would change if the coax-based Live Edge dipoles had their mid/high drivers suspended 4" or more above the main baffle in this kind of a nude configuration? I know that wasn't what @perrymarshall was targeting for visual impact, of course, but it could also make for some very different aesthetics and having Constant Directivity drivers suspended in free space sounds kind of fun.
P.S. None of these questions are implied criticisms of any particular designer or speaker design!
The goal is to have NO REFLECTIONS reaching the listener until at least 4msec after the direct sound. The floor bounce is such a reflection, and it will arrive "too soon". Regarding reflections:
4msec is around 1.5m roughly, or 60 inches. I have seen this threshold listed as 6msec as well. By "strong reflections" I am here alluding to both diffraction and reflection, since strong diffraction creates a new apparent source much like a reflection does. Diffraction from sharp cabinet edges, or for example from surface mounting a tweeter, will combine with the direct sound and cause problems (cancellation, comb filtering, etc.) when the propagating wave from these apparent sources reaches the listener before about 4msec after the direct sound reaches the listener.
Using multiple large radiators helps to smear out the reflection from boundaries and edge diffraction, so taking one out an leaving a hole there is probably not improving the design. OTOH, for the midrange, minimizing the baffle area immediate adajecent to the driver will likely smooth and improve the responses off axis.
- 0-4msec: there should not be any strong reflections reaching the listener during this time period
- 4-40msec: the brain suppresses reflections during this period, and they are not perceived in the same way as a microphone measures the frequency response (e.g. they are NOT perceived as cancellation or comb filtering in the frequency response). The more the frequency response of the reflected signal matches that of the direct sound, the better the suppression.
- >40msec: strong reflections of sufficient amplitude will be perceived as "echo" or "flutter"
4msec is around 1.5m roughly, or 60 inches. I have seen this threshold listed as 6msec as well. By "strong reflections" I am here alluding to both diffraction and reflection, since strong diffraction creates a new apparent source much like a reflection does. Diffraction from sharp cabinet edges, or for example from surface mounting a tweeter, will combine with the direct sound and cause problems (cancellation, comb filtering, etc.) when the propagating wave from these apparent sources reaches the listener before about 4msec after the direct sound reaches the listener.
Using multiple large radiators helps to smear out the reflection from boundaries and edge diffraction, so taking one out an leaving a hole there is probably not improving the design. OTOH, for the midrange, minimizing the baffle area immediate adajecent to the driver will likely smooth and improve the responses off axis.
Last edited:
It’s not absolutely clear to me what we are looking at. Red is both drivers, blue is driver 1 and Green is driver 2?
When you are only running one at a time, are you assuming that the cone motion of the other woofer is zero?
Or are you only assuming that the voltage is zero and that the air from the active driver is pushing the other cone back and forth?
This is why on the live edge dipoles the coax is 38” off the floor and the sub is against the floor. The floor coupling cannot be avoided so we use it for reinforcement in a band where the reflected wave can’t be confused as a 2nd signal. @dayneger
The first time I saw this done was the Boston Acoustics A200 and many designers do this.
Roy Allison published a couple of papers in the 1970s on the topic of floor and wall reflections and suggested designs in which the woofer was at the floor and the midrange high above it (e.g. 40") with a crossover around 300 Hz IIRC. These driver locations and this crossover frequency are just as valid for an OB or dipole system as for a boxed one. The first dip from the floor reflection at one point was commonly referred to as the "Allison dip":
http://villchurblog.com/roy-allison/
http://villchurblog.com/roy-allison/
Sorry. I should have clarified that at the start; meant to, but forgot.
In the SPL charts, green is the radiation from the rear of the woofer(s). Dark blue is the radiation from the front of the woofer(s) diaphragm. Red is the sum of the two. In the SPL chart for the two woofer sim, green includes the radiation from the rears of both woofers; similarly for red and blue.
In the abec model, each side of each diaphragm is assigned to a DrvGroup. This is done in the elements.txt file. In the LE.txt file, where LE stands for lumped element, a netlist of drivers and filter and potentially cross over components is created with DrvGroups being the links between BEM modelling in the elements file, LE modelling, and fields and spectrum "observations". One can observe the SPL, fields and the spectrum of any Drv_Group.
In VACS, the graphical display utility, one can extract curves from the polar maps and export the polar data. Then, with a name_change utility such as AdvandedRenamer, one can convert the names of the exported files to suit their use as Vituix directivity files.
As I implied before, in the two woofer sim, I should have assigned the diaphragm sides of the two woofers to different drive groups to ensure that they were effectively connected in parallel rather than series.
The one-woofer sims include only a single woofer model; the 2nd woofer is commented out of the model. I don't believe there is any way to simulate the effect of one cone pushing the other cone back and forth in ABEC. I think this is more of an issue for measurement than simulation. If measuring a single driver with other drivers in the vicinity, one would want those other drivers to remain connected to amplifiers so that any voltage induced in an idle driver's VC would be dissipated in the amp's near zero output impedance.
Last edited:
I don't believe I ever replied to the hole in the baffle suggestion. The reason is I'm not sure which is better: a baffle with a hole in it below the coax driver or a naked coax driver suspended by a magnet mount above a baffle that covers just the woofer.
I tried simulating a naked driver and don't trust the results. They are just too pessimistic to believed. OTOH, I have yet to find an outright error in the simulation model. I set that aside and went on to simulating holes in the baffle. This idea does seem to work but the response is widening as it approaches crossover to the tweeter, suggesting that a naked or half naked driver might be better. OTOH, it does seem to work well enough so I may stick with it.
Ironically, the hole in the baffle, which was introduced to fix the vertical response of the coax_mid, which was tilted by the lack of any substantial baffle above the driver but substantial baffle below it, had the effect of tilting the woofer response. The fix for that was to introduce a hole in the baffle below the woofer. That is a interesting story also but I will save the details for another day.
I tried simulating a naked driver and don't trust the results. They are just too pessimistic to believed. OTOH, I have yet to find an outright error in the simulation model. I set that aside and went on to simulating holes in the baffle. This idea does seem to work but the response is widening as it approaches crossover to the tweeter, suggesting that a naked or half naked driver might be better. OTOH, it does seem to work well enough so I may stick with it.
Ironically, the hole in the baffle, which was introduced to fix the vertical response of the coax_mid, which was tilted by the lack of any substantial baffle above the driver but substantial baffle below it, had the effect of tilting the woofer response. The fix for that was to introduce a hole in the baffle below the woofer. That is a interesting story also but I will save the details for another day.
Wave length of 200 Hz is 1,7 m, 400 Hz is 0.85 m, so U frame is not able to block sound from back to be canceled with front radiation. Probably missing nulls are from simulation imperfection. Dipole peak is definitely on 200 Hz, so dipole nulls are higher. U-frame speaker have on dipole peak nearly omnidirectional polar plot, this is exactly what is seen on your polar plot simulations.
Last edited:
I needed to redo the U-frame sims with each diaphragm side assigned to a different drive group. The ABEC documentation seems to have an implicit assumption that there is only one driver in a drive group, which is counter-intuitive given that its called a group. With this recoding that places the two drivers in parallel, they sum to 6 db more than a single driver as expected
In the course of doing this I found that in my prior orgy of spinning off multiple simulation models to get a handle on what was going on, I had the wrong driver parameters in my LE script, those for the 5208C_LF instead of the 15OB350. Correcting that yields more SPL but doesn't affect the directivity. This escaped notice for so long because I have been focused on directivity, confident the SPL could be sorted out later because ABEC doesn't require that the directivity be re-computed after changes to the LE script.
Here are the new SPL curves
The red is once again the sum of all four diaphragm sides. 1001 and 1002 are front and rear,
respectively for the lower woofer; 1003 and 1004 for the upper woofer. The next two charts chose these separately for each woofer.
Interesting that there a pronounced dipole peak and null for the upper woofer, where the wings are shallow, but not for the lower woofer. The output below 100 Hz is about the same for both woofers. Were I not planning sealed sub plus dipole instead of a dual woofer U-frame, I would look into this further.
The slanted wing profile may not yield more low end SPL but it does reduce the (out of passband) dip in the composite response due to the dipole null.
In the course of doing this I found that in my prior orgy of spinning off multiple simulation models to get a handle on what was going on, I had the wrong driver parameters in my LE script, those for the 5208C_LF instead of the 15OB350. Correcting that yields more SPL but doesn't affect the directivity. This escaped notice for so long because I have been focused on directivity, confident the SPL could be sorted out later because ABEC doesn't require that the directivity be re-computed after changes to the LE script.
Here are the new SPL curves
The red is once again the sum of all four diaphragm sides. 1001 and 1002 are front and rear,
respectively for the lower woofer; 1003 and 1004 for the upper woofer. The next two charts chose these separately for each woofer.
Interesting that there a pronounced dipole peak and null for the upper woofer, where the wings are shallow, but not for the lower woofer. The output below 100 Hz is about the same for both woofers. Were I not planning sealed sub plus dipole instead of a dual woofer U-frame, I would look into this further.
The slanted wing profile may not yield more low end SPL but it does reduce the (out of passband) dip in the composite response due to the dipole null.
and the fields views
40 Hz
The last two images are 160 Hz, close to the dipole peak, and 210 Hz, close to the dipole null.
How high can this U-frame be used? It depends how constant you want the directivity to be and how far down the driver you are crossing it over to can reach without strain. The side nulls are still distinct at 80 Hz but already widening by 106 Hz. By 160 Hz, they are almost gone.
40 Hz
The last two images are 160 Hz, close to the dipole peak, and 210 Hz, close to the dipole null.
How high can this U-frame be used? It depends how constant you want the directivity to be and how far down the driver you are crossing it over to can reach without strain. The side nulls are still distinct at 80 Hz but already widening by 106 Hz. By 160 Hz, they are almost gone.
From 2 posts above, this image:
![Curves_H[1].jpg](https://www.diyaudio.com/community/attachments/curves_h-1-jpg.1245324/ "Curves_H[1].jpg")
...shows the classic signs of "pinching" above the dipole peak. This occurs when the front to back pathlength exceeds some threshold (that isn't all that high). If this driver must reproduce low frequencies you are stuck with this behavior because making the pathlength smaller to reduce or eliminate the response pinching causes the LF losses to increase. Thus, you must cross this configuration over between 100Hz and 150Hz if you want to avoid the pinched region where the pattern has broadened. IMO all large-baffle OB systems will have these sort of response problems for one or more drivers in the system, and mitigating those with the crossover is tricky.
...shows the classic signs of "pinching" above the dipole peak. This occurs when the front to back pathlength exceeds some threshold (that isn't all that high). If this driver must reproduce low frequencies you are stuck with this behavior because making the pathlength smaller to reduce or eliminate the response pinching causes the LF losses to increase. Thus, you must cross this configuration over between 100Hz and 150Hz if you want to avoid the pinched region where the pattern has broadened. IMO all large-baffle OB systems will have these sort of response problems for one or more drivers in the system, and mitigating those with the crossover is tricky.
There are two related strategies that will reduce the pinching:
1. keep the source(s) - the drivers - the same and make the front to back pathlength smaller
2. keep the front to back pathlength the same and make the source(s) larger
Both of these try to make the ratio of source size to front to back pathlength closer to but not quite 1, e.g. 1.25 to 1.5 is a good range to shoot for. The smaller the pathlength the higher will be the frequency of the dipole peak.
If the band in question is the lowest one in the system, it also has to have enough output down to 30-40Hz and this requires that the source(s) are large because the pathlength can only be up to 1.25 to 1.5 times that. You can use a single large driver like an 18" or you can use multiple smaller drivers. I have built a couple of systems using four 8" drivers in a 20x20 to 24x24 inch panel, for example, and have just recently started using a system with stereo short u- or h-frames housing a high Xmax 18" driver. The 20x20" panel can operate up to about 300Hz, and the 'frame up to a little over 200Hz, limited by where the dipole peak is located and the response starts to fall into the null above it.
Once you have figured out what you want for the lowest frequency band, you then have to design the next higher band(s) in a way that delivers a good pattern for them as well. The same sort of ratio rule will apply, and this has some implications for the baffle size and shape around these drivers.
1. keep the source(s) - the drivers - the same and make the front to back pathlength smaller
2. keep the front to back pathlength the same and make the source(s) larger
Both of these try to make the ratio of source size to front to back pathlength closer to but not quite 1, e.g. 1.25 to 1.5 is a good range to shoot for. The smaller the pathlength the higher will be the frequency of the dipole peak.
If the band in question is the lowest one in the system, it also has to have enough output down to 30-40Hz and this requires that the source(s) are large because the pathlength can only be up to 1.25 to 1.5 times that. You can use a single large driver like an 18" or you can use multiple smaller drivers. I have built a couple of systems using four 8" drivers in a 20x20 to 24x24 inch panel, for example, and have just recently started using a system with stereo short u- or h-frames housing a high Xmax 18" driver. The 20x20" panel can operate up to about 300Hz, and the 'frame up to a little over 200Hz, limited by where the dipole peak is located and the response starts to fall into the null above it.
Once you have figured out what you want for the lowest frequency band, you then have to design the next higher band(s) in a way that delivers a good pattern for them as well. The same sort of ratio rule will apply, and this has some implications for the baffle size and shape around these drivers.
I tried just dropping a coax driver into the model by itself and what I saw at first wasn't pretty. I ran multiple simulations dropping one element at a time from the model to root cause the peaks and valleys in the rear response. Finally, I shrank the simulation down to just enough of the baffle to fit the driver, and worked on that until the response at least looked reasonable. I'm only showing some of the variations I tried in the drawing, just to give a flavor.
There seemed to be some combing with the floor I had in U-frame model. This is needed for woofer sim to prevent an acoustic short circuit at the bottom. It apparently is only part of the story for the midwoofer simulation. The rear wave valleys are reduced with the floor omitted; they don't disappear completely until the wings also are removed from the model. That version of the sim had what I hope is simulation anomaly showing a spike at about 1500 Hz.
Ultimately, I left this floor panel out of the midwoofer simulations because it clouds the issue. In the real world, the rear wave reflecting off this panel will be matched by the front wave reflecting off the full floor.
And of course, I had some of the usual coding errors or bad judgment calls that I found and corrected along the way.
The response is sensitive to baffle thickness and to details of the cone and magnet heights. I went back over the data sheet drawing to interpolate these dimensions with greater accuracy. Likely, the model could be improved with parameters measured directly on the driver. I settled on a 54mm thick baffle; it was incrementally better than earlier 36mm thickness results.
Next I went back to the full model and shrank down the top part of the wings to more closely resemble the BB sketch.
I partitioned the baffle into a top section just big enough for the driver and a lower section that could be modelled with a coarser grid as simulation time had grown quite a bit.
Here is what I simulated:
Here is its SPL response. Its still not pretty but it looks useable and might even be reasonably accurate below 1100 Hz. In any case, its what the model predicts.
Here are the polar maps and curves derived from them:
I partitioned the baffle into a top section just big enough for the driver and a lower section that could be modelled with a coarser grid as simulation time had grown quite a bit.
Here is what I simulated:
Here is its SPL response. Its still not pretty but it looks useable and might even be reasonably accurate below 1100 Hz. In any case, its what the model predicts.
Here are the polar maps and curves derived from them:
This sequence of field views shows the midwoofer response changing over its operating range.
Deep side nulls are present until about 800 Hz and then start to bend forward. At the same time, the 15" cone is starting to beam, which likely is the cause of the bending forward of the side nulls.
Looking at the 1100 Hz fields, we see nulls being bent forward towards the ceiling and floor, which may be a good thing. This narrowing of response doesn't preclude a transparent crossover to the CD/tweeter because it is has the same directivity.
Deep side nulls are present until about 800 Hz and then start to bend forward. At the same time, the 15" cone is starting to beam, which likely is the cause of the bending forward of the side nulls.
Looking at the 1100 Hz fields, we see nulls being bent forward towards the ceiling and floor, which may be a good thing. This narrowing of response doesn't preclude a transparent crossover to the CD/tweeter because it is has the same directivity.
Attachments
Originally, I thought it would be difficult to get a dipole response all the way up to 1100 Hz, especially so for a 15" driver.
I undertook these simulations to see if that were the case. It seems not so much that dipole side nulls are missing but that they are bent forward due to driver beaming and a very useable response exists (just) past 1100 Hz.
I would continue to export directivity files for the woofer and midwoofer to be included in a Vituix system simulation except for the fact that the system thus modelled has already been built.
If anyone wants these simulation models, just PM me and I will zip up the working directory.
I undertook these simulations to see if that were the case. It seems not so much that dipole side nulls are missing but that they are bent forward due to driver beaming and a very useable response exists (just) past 1100 Hz.
I would continue to export directivity files for the woofer and midwoofer to be included in a Vituix system simulation except for the fact that the system thus modelled has already been built.
If anyone wants these simulation models, just PM me and I will zip up the working directory.
There's a reason the crossover is at 1100Hz, in fact the woofer drive signal has a peak built in at 900. Here is the Flex Eight EQ curve for the bass-mid section:
The notches at 40Hz and 75Hz fix modes in my room. Notice the filtering above 1100 is a high shelf filter, not a traditional crossover slope.
And below are VituixCad sims of the series crossover:
The notches at 40Hz and 75Hz fix modes in my room. Notice the filtering above 1100 is a high shelf filter, not a traditional crossover slope.
And below are VituixCad sims of the series crossover:
Last edited: