The Hells Bells
We (mostly @vineethkumar01) have worked serious magic on our horns, the Hells Bells. After tinkering with several active crossover points, EQ tweaks, and speaker positioning/placements, we've finally found the sweet spot!
The Hells Bells are a little picky about where they live. We had to wheel them around, push, pull, and twist them until they decided to make friends with the room. Thank goodness for those wheels, seriously! Now they're sitting pretty at a 20-degree toe-in and with 8 feet between them.
Ever heard a horn that sounds like a hi-fi speaker? The Hells Bells combines the clean directivity and dynamics of a horn with the smoothness of hi-fi. This bass horn undoubtedly stands out as one of the most refined and effortless bass experiences I've ever encountered. It fills the air with a warm glow that invites Leonard Cohen to dig deeper, and plenty of slam power on reserve for low-digging electronic music.
The marriage of the mk3b2 and the 15PR400 bass horn is indeed a match made in heaven. The cohesion of the tone is seamless.
The common inhibition that horns sound, well, honky/shouty is probably due to bad design, first, and selection of CD, next. It's a complex science (let's be honest, whatever I do not understand is complex science) to get the horn geometry right, and we are grateful that there are still a few wizards willing to share their work with the DIY community. @docali @DonVK @fluid
This project, having roots in early June, was a plunge into unknown waters. Vineeth backed by his strong technical prowess, and I, backed by my strong imbecility, decided it was worth considering and poured all our heart and soul into it.
Even before Vineeth started reading me my rights on the perils of an untested design, I said YES!
It was only the next morning did I realise that it was the hooch that spoketh on my behalf. One foot already off the boat, we thought, might as well dive in and swim the depths.
To obtain a first-hand understanding of the horn's performance, we printed a 3D prototype. The process wasn't as smooth as we hoped, thanks to some software compatibility issues, but Vineeth helped iron out these kinks to make the 3D files usable.
The horn was an ordeal to fabricate and glue the various sections to align seamlessly, but accomplished; partly thanks to my ingenious fabricator, who turned out to be a master of jugaad!
I lugged the horn back home from the workshop and plonked it on an existing 15" subwoofer (that was also a collaborative effort with Vineeth) to get a "feel" of the horn. Not the absolute indicator of performance, but that's all we had to back our horse.
Subjectively, I quite liked what I heard, and Vineeth liked the look of the polars, the MK3B2 showed promise!
We still weren't sure how the final system would sound, and if the final result would be to our liking. We discussed a few options, including a system with a waveguide. But, any untested change we make would still be an unknown. As we had already tested and familiarised with the mk3b2, we thought it held enough merit and was worth pursuing.
Three months of sweaty foreheads, smelly armpits, dusty shirts, bruised and sticky fingers, and we are finally home. Priceless, rewarding experience!
There is indeed magic in a properly designed 2-way 15" horn-based system, something to be experienced first-hand, not read.
@vineethkumar01 has been a rock, always reasoning and articulating every step of the way to ensure maximum success. I cannot thank him enough! It's gut-wrenching that he is 360 km away and isn't here to experience his creation! I'm also grateful to everyone who pitched in with ideas and suggestions on this thread. You have all contributed to the making of a dream speaker!
P.S. When my playlist automatically resumed after connecting, AC/DC's "Hells Bells" thundered down these speakers. This was the very first track I blared through, and man, it sounded enormous! I mean, it was fate to name them that, right? 🎸🔔😄
We (mostly @vineethkumar01) have worked serious magic on our horns, the Hells Bells. After tinkering with several active crossover points, EQ tweaks, and speaker positioning/placements, we've finally found the sweet spot!
The Hells Bells are a little picky about where they live. We had to wheel them around, push, pull, and twist them until they decided to make friends with the room. Thank goodness for those wheels, seriously! Now they're sitting pretty at a 20-degree toe-in and with 8 feet between them.
Ever heard a horn that sounds like a hi-fi speaker? The Hells Bells combines the clean directivity and dynamics of a horn with the smoothness of hi-fi. This bass horn undoubtedly stands out as one of the most refined and effortless bass experiences I've ever encountered. It fills the air with a warm glow that invites Leonard Cohen to dig deeper, and plenty of slam power on reserve for low-digging electronic music.
The marriage of the mk3b2 and the 15PR400 bass horn is indeed a match made in heaven. The cohesion of the tone is seamless.
The common inhibition that horns sound, well, honky/shouty is probably due to bad design, first, and selection of CD, next. It's a complex science (let's be honest, whatever I do not understand is complex science) to get the horn geometry right, and we are grateful that there are still a few wizards willing to share their work with the DIY community. @docali @DonVK @fluid
This project, having roots in early June, was a plunge into unknown waters. Vineeth backed by his strong technical prowess, and I, backed by my strong imbecility, decided it was worth considering and poured all our heart and soul into it.
Even before Vineeth started reading me my rights on the perils of an untested design, I said YES!
It was only the next morning did I realise that it was the hooch that spoketh on my behalf. One foot already off the boat, we thought, might as well dive in and swim the depths.
To obtain a first-hand understanding of the horn's performance, we printed a 3D prototype. The process wasn't as smooth as we hoped, thanks to some software compatibility issues, but Vineeth helped iron out these kinks to make the 3D files usable.
The horn was an ordeal to fabricate and glue the various sections to align seamlessly, but accomplished; partly thanks to my ingenious fabricator, who turned out to be a master of jugaad!
I lugged the horn back home from the workshop and plonked it on an existing 15" subwoofer (that was also a collaborative effort with Vineeth) to get a "feel" of the horn. Not the absolute indicator of performance, but that's all we had to back our horse.
Subjectively, I quite liked what I heard, and Vineeth liked the look of the polars, the MK3B2 showed promise!
We still weren't sure how the final system would sound, and if the final result would be to our liking. We discussed a few options, including a system with a waveguide. But, any untested change we make would still be an unknown. As we had already tested and familiarised with the mk3b2, we thought it held enough merit and was worth pursuing.
Three months of sweaty foreheads, smelly armpits, dusty shirts, bruised and sticky fingers, and we are finally home. Priceless, rewarding experience!
There is indeed magic in a properly designed 2-way 15" horn-based system, something to be experienced first-hand, not read.
@vineethkumar01 has been a rock, always reasoning and articulating every step of the way to ensure maximum success. I cannot thank him enough! It's gut-wrenching that he is 360 km away and isn't here to experience his creation! I'm also grateful to everyone who pitched in with ideas and suggestions on this thread. You have all contributed to the making of a dream speaker!
P.S. When my playlist automatically resumed after connecting, AC/DC's "Hells Bells" thundered down these speakers. This was the very first track I blared through, and man, it sounded enormous! I mean, it was fate to name them that, right? 🎸🔔😄
Last edited:
One more version of the crossover
Reverse null test
With the zobel on woofer removed
Reverse null test
I think both cases above look almost the same. With/without the zobel on woofer.
Now I have a question. Does the EPDR curve look worrying?
Is there any reason why one should look at it when using any other amplifier topology than class B?
Reverse null test
With the zobel on woofer removed
Reverse null test
I think both cases above look almost the same. With/without the zobel on woofer.
Now I have a question. Does the EPDR curve look worrying?
Is there any reason why one should look at it when using any other amplifier topology than class B?
I think I might like your playlist 🙂 Glad you like the result and aren't cursing us for leading you down the garden path.
Will try and proceed with this crossover based on local availability of components
Reverse null test
Reverse null test
Are you sure that the acoustic centres of both drivers are in the same plane? Your sim have no offset defined.
@docali,
Thanks for pointing out the z-offset aspect.
We have done semi-dual channel measurements as per Vituixcad instructions for REW here:
https://kimmosaunisto.net/Software/VituixCAD/VituixCAD_Measurement_REW.pdf
The (vertical) rotation plane for measurements was assumed to contain the tip of the CD horn mouth. The tip of the bass cab mouth was (and is still in the final placement) about 3.5 inches behind it. I think this then turns out similar to the stepped baffle scenario mentioned in the above VituixCAD manual, where we also have to manually enter the offset in the z-direction. I did some calculations in XSIM as well (using the CD horn measured alone, the bass cab measured alone and then both of them connected in parallel and measured once) and tried to derive the z offset. This also turns out close to 3.5 inches.
We had missed this earlier. We will try including the z offset for the woofer and make some adjustments to the crossovers.
Thanks for pointing out the z-offset aspect.
We have done semi-dual channel measurements as per Vituixcad instructions for REW here:
https://kimmosaunisto.net/Software/VituixCAD/VituixCAD_Measurement_REW.pdf
The (vertical) rotation plane for measurements was assumed to contain the tip of the CD horn mouth. The tip of the bass cab mouth was (and is still in the final placement) about 3.5 inches behind it. I think this then turns out similar to the stepped baffle scenario mentioned in the above VituixCAD manual, where we also have to manually enter the offset in the z-direction. I did some calculations in XSIM as well (using the CD horn measured alone, the bass cab measured alone and then both of them connected in parallel and measured once) and tried to derive the z offset. This also turns out close to 3.5 inches.
We had missed this earlier. We will try including the z offset for the woofer and make some adjustments to the crossovers.
Why not using "Estimate IR delay" with loopback and calculate the difference when the mic is at a certain distance at height of each driver?
@docali: Thanks for the suggestion. I took a look at the measurement REW files again.
On the on-axis response of the woofer, I see that the estimated IR delay function gives an estimate of 1.584m
On the on-axis response of the CD horn, for 0-degree angle, for some unknown reason (REW error or loopback system error), it shows a negative delay of -1.115m.
But I am pretty sure, it was measured at precisely 1.5m away from CD horn mouth. For 40 degrees and beyond (up to 90 degrees), the timing delays shown in REW are along expected lines (like 1.6+m)
So it seems like we measured the CD horn on its axis from 1.5m away and the bass horn on its axis from 1.584m away (that 84mm also almost exactly corresponds to the 3.5-inch overhang of the CD horn in front of the bass horn). Now, how do we set the z offset in the crossover simulation?
(VituixCAD says we ideally had to measure the two devices (bass horn & CD horn) on their own planes of rotation & then we have to enter the z offset for the bass horn.)
On the on-axis response of the woofer, I see that the estimated IR delay function gives an estimate of 1.584m
On the on-axis response of the CD horn, for 0-degree angle, for some unknown reason (REW error or loopback system error), it shows a negative delay of -1.115m.
But I am pretty sure, it was measured at precisely 1.5m away from CD horn mouth. For 40 degrees and beyond (up to 90 degrees), the timing delays shown in REW are along expected lines (like 1.6+m)
So it seems like we measured the CD horn on its axis from 1.5m away and the bass horn on its axis from 1.584m away (that 84mm also almost exactly corresponds to the 3.5-inch overhang of the CD horn in front of the bass horn). Now, how do we set the z offset in the crossover simulation?
(VituixCAD says we ideally had to measure the two devices (bass horn & CD horn) on their own planes of rotation & then we have to enter the z offset for the bass horn.)
Hi, VituixCAD coordinates move origin of the measurements, center of rotation. So if there is error of 84mm, tweeter was too far away when measuring, then you must bring the measurement set closer to account the overhang. This is done with negative Z in VituixCAD, z=-84 to the tweeter.
edit.
sorry, read the problem wrong. You could bring the bass forward, so z=-84mm for the bass. Leave tweeter z=0.
Because rotation axis was not on either source acoustic center there is some error on both. But, you could try to estimate the error with simulated data if you wish. Since crossover is rather low, wavelength around 50cm or so, the error isn't likely too big. On-axis data would be alright, but off-axis and thus power and all would have some.
edit.
sorry, read the problem wrong. You could bring the bass forward, so z=-84mm for the bass. Leave tweeter z=0.
Because rotation axis was not on either source acoustic center there is some error on both. But, you could try to estimate the error with simulated data if you wish. Since crossover is rather low, wavelength around 50cm or so, the error isn't likely too big. On-axis data would be alright, but off-axis and thus power and all would have some.
Last edited:
If both dut and mic were not moved, both woofer and tweeter was measured with exact same setup, then you would not add any z or other coordinates if it was dual channel measurement.
The data would contain error still, if rotation axis wasn't acoustic center of each transducer.
I think idea is to measure each transducer exactly same distance away (1.5m in this case) by moving the mic, to prevent error while rotating. Then one would use the coordinate system to move the measured sets within the coordinate system to reflect reality.
But, go with VituixCAD manual what ever that says is likely right 🙂
The data would contain error still, if rotation axis wasn't acoustic center of each transducer.
I think idea is to measure each transducer exactly same distance away (1.5m in this case) by moving the mic, to prevent error while rotating. Then one would use the coordinate system to move the measured sets within the coordinate system to reflect reality.
But, go with VituixCAD manual what ever that says is likely right 🙂
Thanks a lot @tmuikku for the suggestions.
I have few questions. In this case, while measuring the dut was not moved except for rotation. The mic was not moved except to adjust for height so that it faced the driver centre. In such a scenario,
1) How can we simulate the error using VituixCad as you had mentioned above?
2) What should be the plane of rotation for a relatively deep horn such as this one. Is it the vertical plane containing the CD horn mouth tip for the CD horn and for Bass horn, the vertical plane containing the bass horn mouth tip?
I have few questions. In this case, while measuring the dut was not moved except for rotation. The mic was not moved except to adjust for height so that it faced the driver centre. In such a scenario,
1) How can we simulate the error using VituixCad as you had mentioned above?
2) What should be the plane of rotation for a relatively deep horn such as this one. Is it the vertical plane containing the CD horn mouth tip for the CD horn and for Bass horn, the vertical plane containing the bass horn mouth tip?
Hi,
1)
I'm not sure how to do it in purely in VituixCAD so here is fun graph for you to play with
https://www.desmos.com/geometry/vsypcznxi2
It is visual aid to think it through and kind of estimate measurement error if rotation center is not acoustic center of a DUT. Distance to mic varies with measurement angle which makes some SPL error, also path length difference from each driver would change, which means they would have phase difference I think. Please comment if there is some glaring error 😀
The demo does not represent reality either as path length to mic would depend on size and shape of the objects. Anyway, it's something to play with and get some kind of a questimate from error(s) involved.
With deep devices the error is greater and greater, but it's mostly in the off-axis.
2)
I'm measuring my shallow devices using mouth as rotational axis and the small error is fine.
In your case, I'd use the woofer/baffle plane as rotation axis to reduce error on the woofer to zero. I think also the tweeter error would reduce, unless I'm so tired currently I cannot think it through clearly 😀
----
According to the above desmos example, the error in magnitude due to distance varying with measurement angle seems to be within about 1-2db, and phase difference stays below 90deg, which doesn't seem too bad.
If you simulate the crossover with such data, and then prepare to adjust the xo by ear like usual and confirm by measurement, I would like to think the error is not too bad and you'd endup quite similar xo regardless.
edit. updated the link
1)
I'm not sure how to do it in purely in VituixCAD so here is fun graph for you to play with
https://www.desmos.com/geometry/vsypcznxi2
It is visual aid to think it through and kind of estimate measurement error if rotation center is not acoustic center of a DUT. Distance to mic varies with measurement angle which makes some SPL error, also path length difference from each driver would change, which means they would have phase difference I think. Please comment if there is some glaring error 😀
The demo does not represent reality either as path length to mic would depend on size and shape of the objects. Anyway, it's something to play with and get some kind of a questimate from error(s) involved.
With deep devices the error is greater and greater, but it's mostly in the off-axis.
2)
I'm measuring my shallow devices using mouth as rotational axis and the small error is fine.
In your case, I'd use the woofer/baffle plane as rotation axis to reduce error on the woofer to zero. I think also the tweeter error would reduce, unless I'm so tired currently I cannot think it through clearly 😀
----
According to the above desmos example, the error in magnitude due to distance varying with measurement angle seems to be within about 1-2db, and phase difference stays below 90deg, which doesn't seem too bad.
If you simulate the crossover with such data, and then prepare to adjust the xo by ear like usual and confirm by measurement, I would like to think the error is not too bad and you'd endup quite similar xo regardless.
edit. updated the link
Last edited:
eww, checked the desmos sim and there is plenty of errors in the texts, and can't edit new link. Well, skip the texts, hit play button on the k - parameter and scroll down watching how much distance from sources to mic vary with angle (k). Same info visible on the graph, without numbers.
Point was to illustrate, that if the rotation axis is not at the transducer acoustic center, distance to mic varies with measurement angle causing some error in the measured data set. If both transducers have different error, different offset from rotation axis, the error is somewhat greater. You can manipulate the distances in desmos to get some intuition about how big an error could be, and how to get it smaller. For example, ~8cm wiggle on measurement jig has about similar (phase or distance) error to crossover at 700Hz as roughly 2cm error has for 2800Hz crossover.
Point was to illustrate, that if the rotation axis is not at the transducer acoustic center, distance to mic varies with measurement angle causing some error in the measured data set. If both transducers have different error, different offset from rotation axis, the error is somewhat greater. You can manipulate the distances in desmos to get some intuition about how big an error could be, and how to get it smaller. For example, ~8cm wiggle on measurement jig has about similar (phase or distance) error to crossover at 700Hz as roughly 2cm error has for 2800Hz crossover.
Last edited:
For Vituix the baffle plane is where the rotation is meant to happen. This is the same as the instructions for CTA 2034. The program is set up to work with that system. There will be a difference between those style of measurements and measuring from the actual acoustic centre. That is necessary when the true 3D polar information is needed for use in multi element PA rigs. Those measurements can be even more difficult to make properly and tend to be done with other software and supplied in a GLL file.
The difficulty here is that both drivers are in a horn of some kind so a reference plane will have to be decided and the acoustic offset worked out from there.
The difficulty here is that both drivers are in a horn of some kind so a reference plane will have to be decided and the acoustic offset worked out from there.
Oh yeah the bass was a horn as well! pardon being shallow with all the comments without knowing the project too well.
Since it has been quite a project I would personally spend some time with the measurements: do another round with different rotating axis and compare to the current data and see the difference if it's meaningful or not. Then, just implement the xo and do another measurement with the complete speaker. Now with three sets of data error should be visible if significant.
Since it has been quite a project I would personally spend some time with the measurements: do another round with different rotating axis and compare to the current data and see the difference if it's meaningful or not. Then, just implement the xo and do another measurement with the complete speaker. Now with three sets of data error should be visible if significant.
Thanks @tmuikku & @fluid for all the suggestions 🙂
I have to carefully study the data once more then.
When we had taken the polar measurements as per the positioning discussed in above posts, I had designed a prototype crossover for the system that looked like this
I had not set the half-space option in VituixCAD for directivity calculation (hence it probably had a 3dB less directivity across most of the bandwidth as seen in above plot).
Once this crossover was put in place, we also measured the polars of the full speaker with crossover. It looked like below with the directivity caculation done after setting half space option in VituixCAD (since we had taken only 0 to 90 degree horizontal polar measurements). This was a gated measurement with about 5ms gate window and hence probably doesn't have much value below 500 Hz. At the time we also did not realise that the volume control on the amp connected to CD horn was turned down a little bit.
Looking at the nice looking directivity, we adjusted a little bit in DSP and got a response like this:
This sort of looked closer to the original design and hence we left it at that. But I do see a bit of error between the crossover-based responses predicted by VituixCAD and what we got eventually with full speaker polars. This as I understand now is due to the issue with measurement setup By the way the reference axis for this design was 20 degrees off axis from the CD horn centre in the horizontal plane, at the height of the CD horn centre.
But then again since the directivity was looking very nice, we thought @WetFartz could just apply a house curve to suit his subjective tastes and leave it at that.. 🙂
I have to carefully study the data once more then.
When we had taken the polar measurements as per the positioning discussed in above posts, I had designed a prototype crossover for the system that looked like this
I had not set the half-space option in VituixCAD for directivity calculation (hence it probably had a 3dB less directivity across most of the bandwidth as seen in above plot).
Once this crossover was put in place, we also measured the polars of the full speaker with crossover. It looked like below with the directivity caculation done after setting half space option in VituixCAD (since we had taken only 0 to 90 degree horizontal polar measurements). This was a gated measurement with about 5ms gate window and hence probably doesn't have much value below 500 Hz. At the time we also did not realise that the volume control on the amp connected to CD horn was turned down a little bit.
Looking at the nice looking directivity, we adjusted a little bit in DSP and got a response like this:
This sort of looked closer to the original design and hence we left it at that. But I do see a bit of error between the crossover-based responses predicted by VituixCAD and what we got eventually with full speaker polars. This as I understand now is due to the issue with measurement setup By the way the reference axis for this design was 20 degrees off axis from the CD horn centre in the horizontal plane, at the height of the CD horn centre.
But then again since the directivity was looking very nice, we thought @WetFartz could just apply a house curve to suit his subjective tastes and leave it at that.. 🙂
@WetFartz , it's great to see you've got the speakers (HB's 😎) finished, positioned and like the sound. Those polars look incredible and I'm sure they sound as good as they look 😀 @vineethkumar01 thanks for helping and sharing the results. Everyone benefits.
Half space in this context means an infinite baffle or ground plane, not that only 0 to 90 degree measurements were taken in full space. The result will be quite different.
I would measure the speaker complete with the top horn as it is intended to be used. Pick a reference plane, the front of the woofer horn or something else then measure the woofer on it's axis and the horn on it's axis at the same distance from the reference plane. Then you can remove the same time of flight from each measurement and the z difference between horn and woofer is already baked into the measurements and the z position can be 0 for both. Set the Y position to reflect the vertical distances between drivers.
Thanks @fluid.
In fact this is what we had done earlier too.
The reference plane was picked as the imaginary vertical plane containing the tip of the CD horn mouth.
Both the drivers were measured at a distance of 1.5m away from this reference plane on their respective axes.
During the measurement the CD horn mouth overhang in front of the bass horn was about 3.5ish inches (as it is in its current final placement)
Here is a snapshot from the woofer's on axis response estimate IR delay option in REW
The additional 80mm distance over the 1.5m, as shown above is almost the same as distance by which the the bass horn cabinet mouth is behind the CD horn mouth (where our reference plane is)
However, in the CD horn measurement, there was an issue with estimate IR delay. We are pretty sure that we did it 1.5m away from the above mentioned reference plane but the estimate IR delay option gives a very wierd value
I was expecting it to show 1.5m but it shows -1.114m above. This has partly been the cause of all sorts of confusion so far..
We will try to do the measurement once more..
In fact this is what we had done earlier too.
The reference plane was picked as the imaginary vertical plane containing the tip of the CD horn mouth.
Both the drivers were measured at a distance of 1.5m away from this reference plane on their respective axes.
During the measurement the CD horn mouth overhang in front of the bass horn was about 3.5ish inches (as it is in its current final placement)
Here is a snapshot from the woofer's on axis response estimate IR delay option in REW
The additional 80mm distance over the 1.5m, as shown above is almost the same as distance by which the the bass horn cabinet mouth is behind the CD horn mouth (where our reference plane is)
However, in the CD horn measurement, there was an issue with estimate IR delay. We are pretty sure that we did it 1.5m away from the above mentioned reference plane but the estimate IR delay option gives a very wierd value
I was expecting it to show 1.5m but it shows -1.114m above. This has partly been the cause of all sorts of confusion so far..
We will try to do the measurement once more..
The impulse response should show why REW is estimating that value. Estimating picks the highest peak which is not necessarily the right place. A loopback timing reference is always better if it is available.