Heard these recently, really loved them, and now plan to build them. Others embarking on the same journey, please share! Ideally this thread can concentrate the latest design notes, guides and user builds.
The speakers: https://evo2.org/wp-content/uploads/2023/07/Bitches_Brew_Open_Baffle_Live_Edge_Speakers.pdf
A note on the passive crossover: https://www.diyaudio.com/community/...over-vs-parallel-for-subs-open-baffle.404680/
Another note on the crossover: https://www.diyaudio.com/community/threads/open-baffle-bass-reflex-hybrid.393837/page-3#post-7250690
And the latest updates from @perrymarshall: https://www.diyaudio.com/community/...on-updated-design.405104/page-19#post-7938344
My plan is to biamp, keeping the passive crossover between the 15OB350 and 15CXN88 mid-woofer. It seems a more elegant and reasonable solution than triamping, although the triamp route is tempting. I'll do a 5º tilt and shrink the sidewings just a little (for an overall depth of 23" instead of 26"). Design files attached. Perry, off the top of your head do you know how much low end is lost per ~3" or 6" of shrinkage on the sidewings?
Re: FIR filtering and miniDSP. From what I can tell, the miniDSP Flex Eight has FIR filters on the inputs, the Flex Balanced has them on the four outputs, and both share the same total number of taps. It looks like the Flex HTx and SHD do NOT allow user-created FIR filters. I assume Perry's files will only work on the Flex Eight, but it shouldn't be difficult to copy the settings over to other units.
I'm tempted to use the miniDSP SHD instead of Flex, with Dirac enabled. Streaming, preamp, crossover all in one box would be nice and tidy. @perrymarshall, am I correct in thinking that the main compromise with that approach is forfeiting the phase correction and phase benefits of FIR filtering in the upper range EQ (top 2 octaves?)? But Dirac should take care of that anyway? Any other issues or complications to know about if using the SHD instead of Flex?
Some lovely user builds, for reference:
@Balthazarp
@Subsonic1050:
The speakers: https://evo2.org/wp-content/uploads/2023/07/Bitches_Brew_Open_Baffle_Live_Edge_Speakers.pdf
A note on the passive crossover: https://www.diyaudio.com/community/...over-vs-parallel-for-subs-open-baffle.404680/
Another note on the crossover: https://www.diyaudio.com/community/threads/open-baffle-bass-reflex-hybrid.393837/page-3#post-7250690
And the latest updates from @perrymarshall: https://www.diyaudio.com/community/...on-updated-design.405104/page-19#post-7938344
My plan is to biamp, keeping the passive crossover between the 15OB350 and 15CXN88 mid-woofer. It seems a more elegant and reasonable solution than triamping, although the triamp route is tempting. I'll do a 5º tilt and shrink the sidewings just a little (for an overall depth of 23" instead of 26"). Design files attached. Perry, off the top of your head do you know how much low end is lost per ~3" or 6" of shrinkage on the sidewings?
Re: FIR filtering and miniDSP. From what I can tell, the miniDSP Flex Eight has FIR filters on the inputs, the Flex Balanced has them on the four outputs, and both share the same total number of taps. It looks like the Flex HTx and SHD do NOT allow user-created FIR filters. I assume Perry's files will only work on the Flex Eight, but it shouldn't be difficult to copy the settings over to other units.
I'm tempted to use the miniDSP SHD instead of Flex, with Dirac enabled. Streaming, preamp, crossover all in one box would be nice and tidy. @perrymarshall, am I correct in thinking that the main compromise with that approach is forfeiting the phase correction and phase benefits of FIR filtering in the upper range EQ (top 2 octaves?)? But Dirac should take care of that anyway? Any other issues or complications to know about if using the SHD instead of Flex?
Some lovely user builds, for reference:
@Balthazarp
Hi!
This is my version of Perry Marshalls "The Bitches Brew" open baffle. First real speaker project and I could not be more happy with the outcome. But as this is my first speaker I have built, if you have any technical questions please read the original designers notes here on the forum, he has uploaded a PDF with all info you could possibly want!
The crossovers and wiring is still not final, just so you know.
Kind Regards
Jakob P
This is my version of Perry Marshalls "The Bitches Brew" open baffle. First real speaker project and I could not be more happy with the outcome. But as this is my first speaker I have built, if you have any technical questions please read the original designers notes here on the forum, he has uploaded a PDF with all info you could possibly want!
The crossovers and wiring is still not final, just so you know.
Kind Regards
Jakob P
@Subsonic1050:
Attachments
Yes, Dirac should take care of that anyway. I'm not aware of any other issues or complications with using the SHD. You can use the MiniDSP Device Console and load in the settings to the Flex Eight module, then hand copy the filter settings into the SHD. Not difficult.
In that case you are skipping the FIR files and then using DIRAC to 1) EQ your room and 2) straighten out the phase.
For those who use the Flex Eight which has 2 FIR inputs, I've attached an FIR "bin" file zipped up. You can use this file for both the Bitches Brews and the Live Edge Dipoles.
This file does not alter amplitude (frequency response) at all. The only thing it does is a 360 degree maxphase rotation. It corrects the phase response back to flat so you get excellent impulse and step response. If you load in this file to each of the 2 Flex Eight input channels, it will fix the phase problems and does NOT require you to buy FIR software, understand FIR or anything else.
I'd like to show you, measurement-wise, what it does. These measurements are from the Live Edge Dipoles (18" Eminence + 8" Radian coax) but the Bitches Brews will exhibit very similar characteristics:
STEP RESPONSE WITHOUT FIR PHASE CORRECTION
Above - You can see the tweeter polarity go negative at 9.4msec at the very beginning. Less obvious is the downward step after 10msec, then it turns back up. That's not an oscillation, that's the 18" woofer having opposite polarity of the 8" mid. Generally speaking this is considered a VERY good step response for a speaker. Even without phase correction it's much better than most speakers, especially 3 ways. That's because the acoustic slopes are effectively 12dB per octave across the board, and the channel delay settings in the DSP ensure that the whole speaker is "minimum phase". It's very easy for the FIR to "unwrap" these phase distortions at the next step.
STEP RESPONSE WITH FIR PHASE CORRECTION
Above - the tweeter is now firing in same phase as the woofer. Now the decay tapers down more slowly with no phase reversal between midrange and woofer. It's almost unheard of to attain a step response like this from a speaker and this is one reason why they are so transparent and image so well.
(The extremely consistent Constant Directivity radiation pattern is another reason. They deliver excellent phase response off axis as well - so the timing cues are good everywhere in the room, not just the listener position.)
One of my most-admired speaker designers, John Dunlavy, always insisted this (time response) is the starting point of a speaker design because it's the hardest thing to get right. Most designers don't even want to think about it. He was way ahead of his time.
IMPULSE RESPONSE WITHOUT FIR PHASE CORRECTION
A perfect impulse would be an upside down capital T. Here you see the tweeter go negative, then the midrange goes positive and settles down.
IMPULSE RESPONSE WITH FIR PHASE CORRECTION
Above you can see that 80% of the impulse is contained in that tiny window at 25msec with a tiny overshoot coming back down and some settling. Impulses this clean from a speaker are almost unheard of. Most speakers butcher phase response. This looks more like the specs of a CD player than a 3-way speaker.
Again, the FIR files are in the ZIP which is attached and should work fairly well for either design.
I love your much-improved drawings.
You asked:
Think of it percentage-wise. Every 50% reduction of the sidewing length will cost you 6dB in amplitude at low frequencies, and doubles the cutoff frequency insofar as it is caused by the smaller baffle. (In other words ignoring the woofer's natural response.)
A good rule of thumb for dipole cutoff is: F = 2000/(Length of minimum path from front to back)
So the original design has an AVERAGE sidewing depth of 18" (measured at the point between the two subwoofers) the tape measure path length from front to back is about 50 inches. That gives you a dipole roll off of around 40Hz.
If you cut the sidewing depth down to 9 inches your dipole roll off will start at roughly 80Hz and you'll lose 6dB across the board below that.
Practically speaking, that means you can compensate for the smaller wings with DSP bass boost until the woofers start complaining. You won't get any complaints at 35-60Hz, but you will run out of steam rapidly below 35Hz if your wings are short.
So if you make a system like Jakob P with minimal wings above, and you like to play your music loud, just roll off the DSP below 35Hz so your woofers are not helplessly flapping at 20Hz and you'll still get plenty of slam and SPL from 40Hz up.
Great summation, thanks for starting this thread.
Thanks Perry, the measurements and explanation are helpful.
And very good to know re: low end and sidewing depth. Assuming the original is ~50" and 40Hz, then a 44" length would move the rolloff up to just ~45Hz or so (also assuming that the geometry is comparable). So it seems like there's some flexibility there, although it is surprising to me that others chose to eliminate them completely (i.e. make them open). I do not plan to use subwoofers so will keep them close to original.
Related: if adding more bass boost via DSP on the bi-amp model, is there a concern that that will affect the mid-woofer performance?
And very good to know re: low end and sidewing depth. Assuming the original is ~50" and 40Hz, then a 44" length would move the rolloff up to just ~45Hz or so (also assuming that the geometry is comparable). So it seems like there's some flexibility there, although it is surprising to me that others chose to eliminate them completely (i.e. make them open). I do not plan to use subwoofers so will keep them close to original.
Related: if adding more bass boost via DSP on the bi-amp model, is there a concern that that will affect the mid-woofer performance?
Made a parts list to get an order for crossover components placed and figured I'd share it here. Cabinets/cabinet material not included. Note that the Flex 8 instead of SHD and a more reasonable amplifier such as the Buckeye NC252MP 4-channel would save almost $2K off the total here.
For those of you who are FIR filter geeks - I've been experimenting with improvements to the design. I'm using Eclipse Audio FIR designer.
The following measurements were taken after I had tweaked the standard IIR filters in the Flex Eight as much as I could to get a flat response. This includes careful time alignment of the drivers and my deliberate pursuit of low Q 12dB/octave acoustic slopes between the midrange and tweeter.
I used FuzzMeasure to generate an impulse response. Most measurement tools will do this. (Many measurement tools calculate phase response from the frequency response curve instead of actually measuring the phase. Exporting the impulse and converting it to FFT is a good way to get accurate phase information.)
Then I uploaded the impulse file into ChatGPT and asked it to give me a FFT in a CSV file with frequency in first column, magnitude in DB in 2nd column and phase in degrees in the 3rd column. ChatGPT does this easily (how cool is that?). I told it to window 100msec starting just before the impulse and ignore everything else.
Then I loaded it into FIR designer. The blue curve is the amplitude (which is only accurate above a few hundred Hz) and the red curve is the phase.
As you can see the frequency response is quite flat from 300Hz to 12KHz and then you start seeing breakup of the compression horn.
You can see that the phase response slowly and evenly "wraps" downward just as a 3 way with true 12dB/octave acoustic slopes should Woofers normal polarity, midrange reverse polarity, tweeter normal polarity. This is exactly as expected. It's the most realistically conservative way I know to design a speaker using minimum phase filters.
The goal is that the speaker has no "not minimum phase" behaviors so that the FIR filters can perfectly unwrap the phase response. In a 3 way with all 12dB/octave slopes, with the midrange wired in reverse polarity, all the surfaces are basically moving the same direction at all frequencies.
Since the MiniDSP Flex Eight only has 2 FIR filters on the inputs, rather than separate FIRs for each driver output, I have to take care to keep the entire speaker minimum phase, which is rarely done.
In the "FIR Phase Adjust" screen I've built 3 filters which delicately unwrap the phase rotations. Maxphase 1st order at 150Hz; Maxphase 1st order at 3500 Hz; and Maxphase 2nd order filter at 20KHz with a bandwidth of 0.25 octave.
These push the phase the exact opposite direction as you can see in the upper right.
Image above, Lower right: Blue is the frequency response magnitude (still unchanged).
Image above, Faint red is the uncorrected phase response.
Image above, Bright red is the corrected phase response. Nearly flat +/- 45 degrees from 150Hz to about 15KHz. For a speaker this is exemplary.
~
Below - the grey dot at 480Hz indicates where magnitude correction kicks in. Above that frequency, the system is being EQ'd to 1/48th octave precision. The upper right GREEN curve is the correction curve.
The faint blue indicates the uncorrected magnitude response of the system. The Dark Blue indicates the flattened response.
The light red indicates the uncorrected phase response of the system. The Dark Red indicates the phase correction curve.
~
Below:
The upper right is the 2000 samples of the FIR filter, with windowing.
Center right is the red phase correction curve of the filter; and the blue magnitude correction of the filter.
Lower right is the error of the FIR filter compared to the ideal that it's trying to achieve. Fractions of a dB and a few degrees phase.
Note: It is possible to skip the "FIR Phase Adjust" step in the FIR Designer software and just have the "Auto Phase" do all the correction in one step.
However because the "Auto Mag" and "Auto Phase" curves are specific to a very specific individual set of drivers; and since your mileage may vary with whatever you get from the factory - I can export just this generic FIR curve as you see here, and it will work with anyone's system. You get to decide if you want to use DIRAC or your own FIR to smooth out the 1/48th octave bumps.
This is the export screen with no 1/48th octave amplitude and phase correction. Any Bitches Brew owner should be able to use this FIR filter to correct phase and you're on your own for amplitude correction using the other tools that any DSP system offers you. This post contains a ZIP file with the .bin FIR filters for this.
So what's the punch line of all this? Measurements that exceed 99.9% of speakers at any price. Most speakers butcher phase, impulse and step response. Have all kinds of timing errors.
I insist you can hear the superiority of a phase-accurate speaker in the form of better imaging and very fine high definition resolution and deep texture.
To put it another way: The Bitches Brews have the qualities that people love in high end headphones, planars, electrostatic speakers like the Quads and Martin Logans, and full range drivers - the transparency and "open window" resolution - but the Bitches Brews also achieve SLAM, outstanding bass and 20-30dB more dynamic range.
It's no accident many electrostatics and planars have an inviting transparent quality. It's because they are phase coherent.
It's no accident some audiophiles prefer shallow slope crossovers. But this design achieves the promise of those shallow slopes. Most speakers attempt but fail.
~
Below, 1/24th octave frequency response in a real room with no gating. Includes room reflections which is why the curve is bumpy below 1KHz (and not particularlly accurate below 50Hz). For a real room this is quite good. Notice it has completely smoothed the breakup modes above 12KHz.
~
Below, the phase response is flat from 70Hz-20KHz:
Below, the impulse response looks more like an amplifier or CD player or DAC than a speaker:
An ideal measurement will look like a perfect upside-down T. The above graph literally means the speaker can start and stop on a dime. Most of the energy of the pulse is contained in a 1/10,000 of a second window. Speakers with impulse response this slender and accurate are almost unheard of.
And the step response is nearly perfect as well:
(This month's Stereophile mag features a digital speaker, the Grimm LS1C, which gives a comparably good step response. That model with subs costs $45,000.)
Practically speaking what accurate phase and time information means is incredible microdyamics and resolution.
Imaging is incredible. Very large and deep sound stage. Separate of instruments is extreme. The dynamics are vibrant and colorful.
The accurate phase and impulse response extends far off-axis because of the Constant Directivity design. Owing to the coax, there are no lobing problems such as you get with separate drivers. Reflected energy coming from the rest of the room is very coherent.
The Bitches Brews hold their own with the most respected and expensive designs in the world, as others have attested.
The following measurements were taken after I had tweaked the standard IIR filters in the Flex Eight as much as I could to get a flat response. This includes careful time alignment of the drivers and my deliberate pursuit of low Q 12dB/octave acoustic slopes between the midrange and tweeter.
I used FuzzMeasure to generate an impulse response. Most measurement tools will do this. (Many measurement tools calculate phase response from the frequency response curve instead of actually measuring the phase. Exporting the impulse and converting it to FFT is a good way to get accurate phase information.)
Then I uploaded the impulse file into ChatGPT and asked it to give me a FFT in a CSV file with frequency in first column, magnitude in DB in 2nd column and phase in degrees in the 3rd column. ChatGPT does this easily (how cool is that?). I told it to window 100msec starting just before the impulse and ignore everything else.
Then I loaded it into FIR designer. The blue curve is the amplitude (which is only accurate above a few hundred Hz) and the red curve is the phase.
As you can see the frequency response is quite flat from 300Hz to 12KHz and then you start seeing breakup of the compression horn.
You can see that the phase response slowly and evenly "wraps" downward just as a 3 way with true 12dB/octave acoustic slopes should Woofers normal polarity, midrange reverse polarity, tweeter normal polarity. This is exactly as expected. It's the most realistically conservative way I know to design a speaker using minimum phase filters.
The goal is that the speaker has no "not minimum phase" behaviors so that the FIR filters can perfectly unwrap the phase response. In a 3 way with all 12dB/octave slopes, with the midrange wired in reverse polarity, all the surfaces are basically moving the same direction at all frequencies.
Since the MiniDSP Flex Eight only has 2 FIR filters on the inputs, rather than separate FIRs for each driver output, I have to take care to keep the entire speaker minimum phase, which is rarely done.
In the "FIR Phase Adjust" screen I've built 3 filters which delicately unwrap the phase rotations. Maxphase 1st order at 150Hz; Maxphase 1st order at 3500 Hz; and Maxphase 2nd order filter at 20KHz with a bandwidth of 0.25 octave.
These push the phase the exact opposite direction as you can see in the upper right.
Image above, Lower right: Blue is the frequency response magnitude (still unchanged).
Image above, Faint red is the uncorrected phase response.
Image above, Bright red is the corrected phase response. Nearly flat +/- 45 degrees from 150Hz to about 15KHz. For a speaker this is exemplary.
~
Below - the grey dot at 480Hz indicates where magnitude correction kicks in. Above that frequency, the system is being EQ'd to 1/48th octave precision. The upper right GREEN curve is the correction curve.
The faint blue indicates the uncorrected magnitude response of the system. The Dark Blue indicates the flattened response.
The light red indicates the uncorrected phase response of the system. The Dark Red indicates the phase correction curve.
~
Below:
The upper right is the 2000 samples of the FIR filter, with windowing.
Center right is the red phase correction curve of the filter; and the blue magnitude correction of the filter.
Lower right is the error of the FIR filter compared to the ideal that it's trying to achieve. Fractions of a dB and a few degrees phase.
Note: It is possible to skip the "FIR Phase Adjust" step in the FIR Designer software and just have the "Auto Phase" do all the correction in one step.
However because the "Auto Mag" and "Auto Phase" curves are specific to a very specific individual set of drivers; and since your mileage may vary with whatever you get from the factory - I can export just this generic FIR curve as you see here, and it will work with anyone's system. You get to decide if you want to use DIRAC or your own FIR to smooth out the 1/48th octave bumps.
This is the export screen with no 1/48th octave amplitude and phase correction. Any Bitches Brew owner should be able to use this FIR filter to correct phase and you're on your own for amplitude correction using the other tools that any DSP system offers you. This post contains a ZIP file with the .bin FIR filters for this.
So what's the punch line of all this? Measurements that exceed 99.9% of speakers at any price. Most speakers butcher phase, impulse and step response. Have all kinds of timing errors.
I insist you can hear the superiority of a phase-accurate speaker in the form of better imaging and very fine high definition resolution and deep texture.
To put it another way: The Bitches Brews have the qualities that people love in high end headphones, planars, electrostatic speakers like the Quads and Martin Logans, and full range drivers - the transparency and "open window" resolution - but the Bitches Brews also achieve SLAM, outstanding bass and 20-30dB more dynamic range.
It's no accident many electrostatics and planars have an inviting transparent quality. It's because they are phase coherent.
It's no accident some audiophiles prefer shallow slope crossovers. But this design achieves the promise of those shallow slopes. Most speakers attempt but fail.
~
Below, 1/24th octave frequency response in a real room with no gating. Includes room reflections which is why the curve is bumpy below 1KHz (and not particularlly accurate below 50Hz). For a real room this is quite good. Notice it has completely smoothed the breakup modes above 12KHz.
~
Below, the phase response is flat from 70Hz-20KHz:
Below, the impulse response looks more like an amplifier or CD player or DAC than a speaker:
An ideal measurement will look like a perfect upside-down T. The above graph literally means the speaker can start and stop on a dime. Most of the energy of the pulse is contained in a 1/10,000 of a second window. Speakers with impulse response this slender and accurate are almost unheard of.
And the step response is nearly perfect as well:
(This month's Stereophile mag features a digital speaker, the Grimm LS1C, which gives a comparably good step response. That model with subs costs $45,000.)
Practically speaking what accurate phase and time information means is incredible microdyamics and resolution.
Imaging is incredible. Very large and deep sound stage. Separate of instruments is extreme. The dynamics are vibrant and colorful.
The accurate phase and impulse response extends far off-axis because of the Constant Directivity design. Owing to the coax, there are no lobing problems such as you get with separate drivers. Reflected energy coming from the rest of the room is very coherent.
The Bitches Brews hold their own with the most respected and expensive designs in the world, as others have attested.
Last edited:
Perry, nice writeup, nice work!
Isn't Fir designer a great tool ! I've been using it since its startup, when it began as "free to beta-testers" about 10 years ago.
It's come a long way since then, but the old original will still do about all you can ask. (I'm using non-multi subscription now.)
( I was a fan of Dunlavy too)
Which or course needs driver sections with very wide frequency response having natural first-order roll offs.
And then, to have any driver protection and thermal bandwidth reduction at all, first order electrical becomes the least order possible.
So, first order driver natural + first order electrical =2nd order acoustic. That simple, huh? 🙂
One thing I've learned to keep in mind when using FirD's auto-mag smoothing, is the need to keep the frequency resolution of the FIR filter in mind, compared to the chosen smoothing in FirD and the relevant frequency range.
If I'm interpreting your graphs correctly, the 2000 tap filter @96kHz with impulse centering being used, has a freq resolution of 96Hz.
Which stays constant across the spectrum. As the top octave spans about 10kHz, 96Hz resolution more than handles FirD's 1/48th setting.
From the bottom octave, 480Hz to 960Hz, 96Hz res is what, effectively about 1/5th octave....
Make sense to you?
(And don't forget it will automatically match whatever acoustic xover target curve is desired for each driver section.)
I have to believe the manual phase correction you generated for your IIR crossovers is also specific to the individual set of drivers you used and their IIR tunings
It would be pretty remarkable I think for different sets of drivers to end up with the same set of acoustic xover frequencies, and fully matching summed phase traces. Maybe I'm sissing something....
I can see how IIR xovers only, the pure electrical component, could be generically phase corrected if used exactly with different driver sets.
But past that, I think phase correction of summed drivers sections is necessarily a measured fit.
Fwiw, When I use auto-phase, or manual phase corrections, I've come to rely on the step function (beyond looking at summed phase) to show how well the correction fits. I realize you do too.
If step dips below zero anywhere in any fashion before impulse peak, I think I've put in overcorrections.
Another make sense to you?
Isn't Fir designer a great tool ! I've been using it since its startup, when it began as "free to beta-testers" about 10 years ago.
It's come a long way since then, but the old original will still do about all you can ask. (I'm using non-multi subscription now.)
Totally agree that 12dB/oct, second order acoustic, is the most realistic goal for achieving the least phase rotation possible, when only using IIR.
( I was a fan of Dunlavy too)
Which or course needs driver sections with very wide frequency response having natural first-order roll offs.
And then, to have any driver protection and thermal bandwidth reduction at all, first order electrical becomes the least order possible.
So, first order driver natural + first order electrical =2nd order acoustic. That simple, huh? 🙂
Below - the grey dot at 480Hz indicates where magnitude correction kicks in. Above that frequency, the system is being EQ'd to 1/48th octave precision. The upper right GREEN curve is the correction curve.
The faint blue indicates the uncorrected magnitude response of the system. The Dark Blue indicates the flattened response.
The light red indicates the uncorrected phase response of the system. The Dark Red indicates the phase correction curve.
One thing I've learned to keep in mind when using FirD's auto-mag smoothing, is the need to keep the frequency resolution of the FIR filter in mind, compared to the chosen smoothing in FirD and the relevant frequency range.
If I'm interpreting your graphs correctly, the 2000 tap filter @96kHz with impulse centering being used, has a freq resolution of 96Hz.
Which stays constant across the spectrum. As the top octave spans about 10kHz, 96Hz resolution more than handles FirD's 1/48th setting.
From the bottom octave, 480Hz to 960Hz, 96Hz res is what, effectively about 1/5th octave....
Make sense to you?
Hell yes it is !!! And super highly recommended if you go full monte, and put FIR on all individual outputs !!! Sorry, couldn't resist😀
(And don't forget it will automatically match whatever acoustic xover target curve is desired for each driver section.)
I have to believe the manual phase correction you generated for your IIR crossovers is also specific to the individual set of drivers you used and their IIR tunings
It would be pretty remarkable I think for different sets of drivers to end up with the same set of acoustic xover frequencies, and fully matching summed phase traces. Maybe I'm sissing something....
I can see how IIR xovers only, the pure electrical component, could be generically phase corrected if used exactly with different driver sets.
But past that, I think phase correction of summed drivers sections is necessarily a measured fit.
Fwiw, When I use auto-phase, or manual phase corrections, I've come to rely on the step function (beyond looking at summed phase) to show how well the correction fits. I realize you do too.
If step dips below zero anywhere in any fashion before impulse peak, I think I've put in overcorrections.
Another make sense to you?
Yes, you are right that at lower frequencies the resolution is far from 1/48th dB/octave. Probably 1/5 or so.
I think getting 6dB final acoustic slopes is well nigh impossible but you can get pretty close if you use shelf and notch filters wherever possible instead of high and low pass filters (see the next post I'm about to make in a few minutes below, it's how I designed the Bitches Brews). If you do that then the FIR has very little work to do.
I regard FIR as "nuclear weapons" and I always try to get as far as I can with a "bow and arrow" before I resort to the high power tools. I guess my philosophy is that very complex tools usually have unintended consequences and they are never cure-alls. I still think you need to design the system as conservatively and elegantly as you can.
Re "which specific drivers" I mean unit-to-unit variations of, say, a 15CXN88. The macro level stuff will be identical but I doubt the 14KHz resonance I see in my compression driver will exactly match the one in yours, and in the 1/48th octave smoothing that's the stuff where your mileage may vary.
Earlier versions of the Bitches Brews the phase compensation inputs to the FIR had been generated by the phase curve from FuzzMeasure, and I get more accurate results when I generate the phase curve from an FFT. It gives me sharper edges on the bottom half of the step response and less ringing on the top. Basically it's because I was able to see the issues introduced by the phase shift between 15-25KHz where I added that 2nd order maxphase filter.
The difference is clearly audible. The resolution and texture with the new tuning is almost scary.
I think getting 6dB final acoustic slopes is well nigh impossible but you can get pretty close if you use shelf and notch filters wherever possible instead of high and low pass filters (see the next post I'm about to make in a few minutes below, it's how I designed the Bitches Brews). If you do that then the FIR has very little work to do.
I regard FIR as "nuclear weapons" and I always try to get as far as I can with a "bow and arrow" before I resort to the high power tools. I guess my philosophy is that very complex tools usually have unintended consequences and they are never cure-alls. I still think you need to design the system as conservatively and elegantly as you can.
Re "which specific drivers" I mean unit-to-unit variations of, say, a 15CXN88. The macro level stuff will be identical but I doubt the 14KHz resonance I see in my compression driver will exactly match the one in yours, and in the 1/48th octave smoothing that's the stuff where your mileage may vary.
Earlier versions of the Bitches Brews the phase compensation inputs to the FIR had been generated by the phase curve from FuzzMeasure, and I get more accurate results when I generate the phase curve from an FFT. It gives me sharper edges on the bottom half of the step response and less ringing on the top. Basically it's because I was able to see the issues introduced by the phase shift between 15-25KHz where I added that 2nd order maxphase filter.
The difference is clearly audible. The resolution and texture with the new tuning is almost scary.
Bitches Brew Latest Flex Eight files
In the attached ZIP file is the most recent Bitches Brew XML configuration file for the MiniDSP Flex Eight. This has IIR settings for the standard DSP and a "generic" phase-correction-only FIR filter that should work with any set of stock drivers, not just the specific drivers I got from the factory with their own particular squiggles above 12KHz.
Notes:
This is the current EQ for the LF section of 2x15OB350+15CXN88 bass mid with passive series LF crossover. It's all parametric. There are no high or low pass filters on the LF section at all. You'll notice there is no rolloff above 3KHz, the natural acoustic response of the woofer takes care of that. This gives an absolute minimum of phase shift so acoustically we get gentle, Low-Q 12dB/octave slopes.
You will find greyed-out inactive notches at 39Hz and 75Hz in the file. Those are correction factors for my room. Your room will be different.
Below is the only traditional crossover in the DSP - a 15KHz 6dB/octave high pass on the HF section:
It might seem very odd to have a 15KHz crossover in a system that actually crosses at 3K, but the 15CXN88 compression driver has a downward sloping response from about 1KHz all the way to the top, and a 15KHz 6dB slope, with the gain turned up on the HF section, is just about perfect.
The parametric filters below round out the HF section to make it very close to flat:
As you can see, the driver only needs a modest amount of correction, mostly a gentle shelf filter below 2000 Hz. This gives the very well-behaved acoustic crossover below.
First the phase response. The LF sections and HF sections are 150-180 degrees apart from 500Hz to 8KHz, so after the HF phase is inverted, it means they are in lock step within 0-30 degrees. This change noticeably added to the transparency. The LF and HF sections of the coaxial are not fighting with each other at all.
Below is magntitude. Notice the overlap of the two drivers runs from 1K to 7K. Because it's a coaxial and there are no lobing errors, this is about as seamless as a crossover can get.
Again a zip file for the MiniDSP Flex Eight is attached. For other MiniDSP units you can hand copy the parametric and crossover settings.
I've also attached just the FIR phase correction files which you can use separately.
In the attached ZIP file is the most recent Bitches Brew XML configuration file for the MiniDSP Flex Eight. This has IIR settings for the standard DSP and a "generic" phase-correction-only FIR filter that should work with any set of stock drivers, not just the specific drivers I got from the factory with their own particular squiggles above 12KHz.
Notes:
This is the current EQ for the LF section of 2x15OB350+15CXN88 bass mid with passive series LF crossover. It's all parametric. There are no high or low pass filters on the LF section at all. You'll notice there is no rolloff above 3KHz, the natural acoustic response of the woofer takes care of that. This gives an absolute minimum of phase shift so acoustically we get gentle, Low-Q 12dB/octave slopes.
You will find greyed-out inactive notches at 39Hz and 75Hz in the file. Those are correction factors for my room. Your room will be different.
Below is the only traditional crossover in the DSP - a 15KHz 6dB/octave high pass on the HF section:
It might seem very odd to have a 15KHz crossover in a system that actually crosses at 3K, but the 15CXN88 compression driver has a downward sloping response from about 1KHz all the way to the top, and a 15KHz 6dB slope, with the gain turned up on the HF section, is just about perfect.
The parametric filters below round out the HF section to make it very close to flat:
As you can see, the driver only needs a modest amount of correction, mostly a gentle shelf filter below 2000 Hz. This gives the very well-behaved acoustic crossover below.
First the phase response. The LF sections and HF sections are 150-180 degrees apart from 500Hz to 8KHz, so after the HF phase is inverted, it means they are in lock step within 0-30 degrees. This change noticeably added to the transparency. The LF and HF sections of the coaxial are not fighting with each other at all.
Below is magntitude. Notice the overlap of the two drivers runs from 1K to 7K. Because it's a coaxial and there are no lobing errors, this is about as seamless as a crossover can get.
Again a zip file for the MiniDSP Flex Eight is attached. For other MiniDSP units you can hand copy the parametric and crossover settings.
I've also attached just the FIR phase correction files which you can use separately.
Last edited:
Agree. 6dB seems pretty impossible. Assuming wide bandwidth first order drivers are the best we have to work with for pure IIR, to me that means 6dB xover slopes effectively equal no xover at all, no electrical attenuation applied. That's OK if we are willing to keep SPL low enough to stay safe, both for excursion and thermal. But I think that would be a big design handicap.
I think SPL capability is also one of limitations of all second order for the same reason. You did a nice job of driver selection to overcome that imo.
Aaah, gotcha. I thought you meant completely different drivers...my bad. Yep, if drivers and IIR xovers are matched up, I can't see how a little in band filtering choice differences could keep your global phase correction from working pretty well.
Interesting. It shouldn't matter, the FFT to IFT back and forth, but I've also been finding (thought the instruction of a mentor) that starting with time domain measurements, and deriving the frequency domain gives more accurate phase info. (I think most folks don't realize phase is a frequency domain, not time domain, component.)
I appreciate the goal of simple elegance. For me, FIR on all individual output channels is that simple elegance.
A two stage technique, use IIR first on individual channels, then global FIR on top of that , has proven to me at least, to be more complex.
Once I've used FIR at all, either global or per channel, I feel I've gone nuclear 🙂, and might as well go all in and do it as best I can.
One really cool thing about FIR I've learned only in the last year (and again due to a mentor) is to look deeper than measured acoustic response.
I've been taught to look closely at the FIR filters electrical impulse response. The cleaner the FIR filters impulse, that still achieves the desired acoustic target, the better. Cleaning up the electrical impulse leads to lower taps counts/shorter filters being applied. And begs for frequency dependent tap counts (another thing mulit-way output FIR solves) It's likes some kind of noise floor gets lowered ....at the same SPL, sounds jump out clearer and more vivid.
Anyway, ironically, I worked nearly 10 years trying to acquire hardware DSP's with higher taps counts.
Nowadays, 4k per ch at 48kHz is all I need (although I need 5 channels of output per speaker)
Damn shame miniDSP effectively ditched out of the DIY FIR market (a rather long time ago when Dirac emerged).
The OpenDRC DA-8 with digital I/O would be a great product to still have around.
That does look very good.If you were asked what the crossover frequency is, however, what would you say? (3 kHz ?)
That's very interesting. In the pdf you published the crossover point was 1.1kHz, I think. That's a big shift upwards! On the face of it 3 kHz is pretty high to cross a 15", but you're obviously happy with it. How would you describe the effect on the sound?
I agree, 3K seems awfully high for a 15. It wasn't even a deliberate attempt to have a high crossover frequency, it's more like after I played with the settings to find the most natural, gradual transition between drivers, that's where it fell.
Audibly what I noticed was after I got the LF and HF phase aligned to within 30 degrees (just using IIR, before the FIR) the depth and transparency went up. It was like the last little bit of "drivers fighting with each other" (which I wasn't even aware of) vanished. It's the best sound I've gotten from them to date.
Audibly what I noticed was after I got the LF and HF phase aligned to within 30 degrees (just using IIR, before the FIR) the depth and transparency went up. It was like the last little bit of "drivers fighting with each other" (which I wasn't even aware of) vanished. It's the best sound I've gotten from them to date.
1) I'm in the process of building a version of Bitches Brew with all drivers as stated. I need to go the tri-amp route because my amp for mid and high are 10W SE amp. I believe they should be fine for that purpose but will run out of power for bass boost. I have seen the update Perry posted on 2025-02-22 which have tri-amp settings. Is there any addition caveat that I need to pay attention to?
2) Another question is on the tilt. Understand it is updated to 5 degree. I'm trying to understand if the listening distance has anything to do with this tilt. Usually, you want to listen to co-ax slightly off-axis. This is commonly adjusted by the degree of toe-in. Because of the height of the speaker, the tweeter axis is already above the listening position. Is there a specific angle and/or distance to achieve the time domain performance as stated?
3) If the high pass is only a first order at 15kHz, would it be simpler to implement it as a passive filter at the power amp?
I'm still debating if I should get a MiniDSP Flex 8, or use it with a multi-channel DAC with Audiolense XO (which I already have in my setup).
2) Another question is on the tilt. Understand it is updated to 5 degree. I'm trying to understand if the listening distance has anything to do with this tilt. Usually, you want to listen to co-ax slightly off-axis. This is commonly adjusted by the degree of toe-in. Because of the height of the speaker, the tweeter axis is already above the listening position. Is there a specific angle and/or distance to achieve the time domain performance as stated?
3) If the high pass is only a first order at 15kHz, would it be simpler to implement it as a passive filter at the power amp?
I'm still debating if I should get a MiniDSP Flex 8, or use it with a multi-channel DAC with Audiolense XO (which I already have in my setup).
@JoeLam; I think you should try both the Flex 8 and AudioLense and see if there is a significant difference between them. This is entirely for selfish reasons as I’m keen to try the latter and it would be good to know that its worth the money !🤣
On paper Audiolense ought to wipe the floor with the Flex 8, bit it would be very interesting to have that confirmed based on user experience.
On paper Audiolense ought to wipe the floor with the Flex 8, bit it would be very interesting to have that confirmed based on user experience.
1- No caveats. Personally, I would use solid state for the bass section. The high efficiency is perfect for small amps for the mids and HF.
2- 5 degree tilt is about right for most listener positions. I don’t think a 0 degree tilt aesthetically looks right.
3- Do not implement tweeter crossover passively, there is much more to the HF filter than a 6dB slope.
@Studley, On paper, AudioLense has better precision and a lot more powerful than Flex8, but it doesn't allow fine grain control. You cannot specify behavior of individual filters. It's kind of like Dirac in the sense that you give it a target curve and specify how tight you want to follow it and it will do the rest. For the XO, you can specify the crossover frequency and slope. Not that I'm an expert in designing FIR filter. Just that I like the idea of using the minimum amount of filter (and processing) to get the job done.
1) Yes, I have a pair of ICEPower modules for the bass.
2&3) well received.
Thanks.
Mitch is also an advocate of minimising correction. I’m surprised by your comment about it not having fine grain control. I’d got the impression from Mitch’s articles and video that if you apply correction driver by driver rather than the speaker as a whole then you had a lot of flexibility available to you, but then I have never used it so maybe I have misunderstood.