Three Way Evil Design Study

[Evil Twin]

Great link on the duelund filter. I am absolutly fascinated. It has been along time since I have done any s-domain math.

Is this a typical process for designing a 3-way type crossover?

Any more references to this type of work? I would really love to become more well versed.

There are more slight variations on the theme, but nothing I've come across substantially different.

This method is interesting and sound because it treats the filter design for three way as a whole, not as an attempt to glue a pair of two way crossovers together. This would be the way to do it in the old days, as Steen describes.

Now, one might implement the desired transfer function in MathCAD, import the acoustical targets into LspCAD, then develop and fit the crossover using your measured driver data.

[ssabripo]

Calrissian, you must have been down in the Bespin mines one time too many times without your respirator... always carry a spare.

W12CY: 85 dB sensitivity, with a moderately horrendous breakup mode between 8 and 18 kHz. Possessed of adequate Xmax (3 mm), the breakup mode would be very difficult to suppress adequate and sill contribute to raised 3rd hamronic distortion from 3 kHz and up.

W15CY: 3 mm Xmax OK, big break up mode starting at 5 kHz. Makes the C79 look pristinely clean in comparison.

W12CY Nextel: Promising SPL plot, but what is that impedance wiggle at 800 Hz? Get's a little rough at 4 kHz and above, but looks manageable. VERY large magnet assembly almost complete masks rear wave. Why did they not use the W15 Nextel neodymium motor? Paper cone? Paper? no self respecting Sith speaker designer... mumble mumble.

M15CH002: Fairly smooth, but a curious glitch in impedance curve and response at 1100 Hz. Some narrow peaks in 4-5 kHz region- odd, with the incredibly small Neodymium magnet assembly, it's not plausibe to blame those on a reflection from the magnet assembly. From the impedance curve and SPL data I expect some minor energy storge issues. Potentially usable... but I beleive I recall others lamenting linear or nonlinear distoriton issues for these- will have to search the hypernet to confirm. Cone size and dispersion might be concerns.

Scanspeak 15M/4532K: Ugly impedance ripple at 1 kHz. It doesn't feel right in the Force...

Scanspeak 15W/4531G: Ugly impedance ripple at 1 kHz....

I looked at other 5" SS- similar issues.

This is not to say others couldn't try these- the M15CH002 would be easiest to fit to the requirements of the crossover, with fairly well controlled response to 10 kHz. Does any one have test data on the Nextel?

Yes indeed Dark Lord....the white substances around Floridan known as Erythroxylum coca have caused some blurring in my thought mechanism.


I will monitor this thread for more information regarding the proposed drivers with curiosity.

Have you narrowed down the topology and cabinet design/layout?

signed,

[Inu_Yasha]

I'm kinda intrigued by this Aurasound NS12-513A. How much does one driver cost and where are they sold at? I'm thinking of using this driver as a woofer for the Nat P's (bass bin), would this work in this type of application? I was also considering one of the 15" dayton HF drivers as a bass bin.

[Dennis H]

There are more slight variations on the theme, but nothing I've come across bestially different.

This method is interesting and sound because it treats the filter design for three way as a whole, not as an attempt to glue a pair of two way crossovers together. This would be the way to do it in the old days. Now, one might implement the desired transfer function in MathCAD, import the acoustical targets into LspCAD, then develop and fit the crossover using your measured driver data.

I thought it was interesting that SL was able to pretty well replicate the 8th order Duelund curves (at least for some "a" value) with LR2 filters by adjusting the Fc's: woofer = two ganged LR2, midrange = single LR2 highpass and lowpass with the driver inverted, tweeter = two ganged LR2.

[ThomasW]

I'm kinda intrigued by this Aurasound NS12-513A. How much does one driver cost and where are they sold at? I'm thinking of using this driver as a woofer for the Nat P's (bass bin), would this work in this type of application? I was also considering one of the 15" dayton HF drivers as a bass bin.

They're from Madisound, $184 ea. For a bass bin for the NatP's I think you'd be fine with the RS315's.

The Aura would the better choice if you had a situation where you needed a woofer to play into higher frequencies like those in a real 3-way. There the woofer needs to reach up to a midrange. That's not the situation you're in with the NatP's.

[mmoeller]

I thought it was interesting that SL was able to pretty well replicate the 8th order Duelund curves (at least for some "a" value) with LR2 filters by adjusting the Fc's: woofer = two ganged LR2, midrange = single LR2 highpass and lowpass with the driver inverted, tweeter = two ganged LR2.

Reading through, the "a" value was a term used as the denomonator of the damping factor. His first choice being a=2 or a .5 damping factor, critically damped.

I noticed earlier in the thread that the dark master suggest another value, a=2.82. Damping of about .35. Now I'm sure there are some other considerations that will change the overall response of the system. However, I'm not sure what those would be. I'm sure the volume and driver mechanical Q would effect this??

I wasn't sure about Fc, his examples use only one angular variable w. I assume that it can't be only 1 frequency for the entire network, but how do the substitutions work. The numerator of the equation is evidently a term for each w of the x-way system, but which w goes in the deonmonator for D1 and D2 equations?

[mmoeller]

There are more slight variations on the theme, but nothing I've come across bestially different.

This method is interesting and sound because it treats the filter design for three way as a whole, not as an attempt to glue a pair of two way crossovers together. This would be the way to do it in the old days. Now, one might implement the desired transfer function in MathCAD, import the acoustical targets into LspCAD, then develop and fit the crossover using your measured driver data.

Thanks for the insight my master. It's always nice to know where you come from though. Takes the understanding to a new level. I like a line from the Duelund paper, "...what is gained in practice must be expressed mathematically, otherwise you have gained nothing." I wouldn't take this as an absolute, but in terms of the level of understanding that I am comfortable with, math tells no lies. (unless its some M.C.Esher kinda stuff, weird?)

[CraigJ]

Oh Evil Twin,

Will the picture below give us a glimpse of what you are intending on building? Ah yes, she has changed since you were last with her

[Jed]

This is not to say others couldn't try these- the M15CH002 would be easiest to fit to the requirements of the crossover, with fairly well controlled response to 10 kHz. Does any one have test data on the Nextel?

I think I do- I'll dig around for it when I get to work tomorrow. All my data is on my laptop. IIRC, it wasn't that good. Funky energy storage dip at 1K and 2nd order harmonics above 1% at modest levels.... 85DB? Like I said, I'll confirm tomorrow but I don't think it fits the emperor's mantra of pistonic behavior and low distortion.

Jed

[Evil Twin]

Yes indeed Dark Lord....the white substances around Floridan known as Erythroxylum coca have caused some blurring in my thought mechanism.


I will monitor this thread for more information regarding the proposed drivers with curiosity.

Have you narrowed down the topology and cabinet design/layout?

signed,

I am not ready to repsond completely and explicitly to this inquiry- I'm in the middle of "delicate nogotiations" with the Trade Federation here in Austrianus.

They're going to feel very delicate if they don't negotiate to my terms....

The concept residing in the Splinter of my minds eye will most likely be realized in a two cabinet configuration, a base cabinet 14" wide and 18" deep and 36-38" tall with a tilt back of about 10 degrees, with a tapered width vertical front tapered back top module sufficient to hold the midrange in the center/base area and the tweeter at the top. Narrowing the cabinet width forces the baffle step up higher in frequency for the upper range drivers, increasing the off axis dispersion, but increasing also the demands on the driver and crossover.

After return to the Terran system from this "business" trip a concept illustration will be in prepration, and final analytical analysis will commence before committing to the project.

That's why it's called a "design study".

There is another project which may be completed first as a test example using the Duelund approach, much simpler in construction execution, which could happen in the following two weeks- the Emperor insists on a test case for correlation of the subjective acoustic properties of an all pass network with the drivers always in relative phase at each frequency... this could even be evaluated in a preliminary manner using the crossover emulation tools in LspCAD professional.

[Evil Twin]

I think I do- I'll dig around for it when I get to work tomorrow. All my data is on my laptop. IIRC, it wasn't that good, just like the W15LY. Funky energy storage dip at 1K and 2nd order harmonics above 1% at modest levels.... 85DB? Like I said, I'll confirm tomorrow but I don't think it fits the emperor's mantra of pistonic behavior and low distortion.

Jed

Thank you for the results of your investigation- your comments mirror my intuition in the Force.

[Evil Twin]

Oh Evil Twin,

Will the picture below give us a glimpse of what you are intending on building? Ah yes, she has changed since you were last with her

Yes... memories...

In response, see above. Conceptually in loose terms, think an Isis with no slope forward on the upper module, and smaller, LF section size as described above. The custom Eton woofers used in the Isis are also neodymium motors with underhung geometry; I wonder what their Xmax is?

So, to summarize loosely and very briefly, a more compact Isis with a single 12" woofer.

[Jed]

Thank you for the results of your investigation- your comments mirror my intuition in the Force.

Here are some results....MCH002

[TacoD]

I am told that the neodym hexa motor system of the Seas mid is not that great. It simply resonates, the build quality of Seas Excel is in my experience not comparable with, e.g., Scanspeak. Some time ago, when working with the Seas Excel W18/ W22 it occured to me that the centering of the phase plug wasn't that precise. Other (budget series) Seas woofers had glue around the former.

[capslock]

Force must be strong with me. I just won a pair of black OEM W22s on ebay. They'll make beautiful "midranges" but not the way his Lordship envisions. So I'll do something M8a-like in Frequency Allocator, and if I like it, might try my hand at an M8a-like passive XO.

Can't complain about Seas build quality. Taco, the phase centering is something you can change with a screwdriver, and it is only of cosmetic concern. Scan Speak has a design philosophy I don't agree with -- too much cone flex at 800 Hz already.

I hava a hunch the higher distortion of the hexadyme might be due to flux modulation in those six iron bars that conduct the flux to the top plate. I was going to wind a pick-up coils around them to check. If that is the reason, copper rings around those bars will cure the problem.

Greetings

Eric

[cjd]

Why haven't we just designed our own drivers again? Clearly the knowledge exists here...

C

[Evil Twin]

Reading through, the "a" value was a term used as the denomonator of the damping factor. His first choice being a=2 or a .5 damping factor, critically damped.

I noticed earlier in the thread that the dark master suggest another value, a=2.82. Damping of about .35. Now I'm sure there are some other considerations that will change the overall response of the system. However, I'm not sure what those would be. I'm sure the volume and driver mechanical Q would effect this??

I wasn't sure about Fc, his examples use only one angular variable w. I assume that it can't be only 1 frequency for the entire network, but how do the substitutions work. The numerator of the equation is evidently a term for each w of the x-way system, but which w goes in the deonmonator for D1 and D2 equations?

jw (actually, omega, not w), is the frequency variable in radians in the S plane; to calcuate the transfer function in MathCAD or Maple, you'd define a range variable to index a vector, I usualy use a simple ratio function with exponential terms to get a certain number of data points spaced in each octave. I'll post this after I get back from my business trip to Austria, after the Christmas holidays. I want to transfer this to Maple, but maybe I'll just do this in MathCAD again.

[Evil Twin]

Here are some results....MCH002

I believe the one of my old nemesis from Dagobah would say something along the lines of...

"Do not try, it is doing you must.."

It is apparent that Seas does not yet understand that concept.

[Paul H]

Why haven't we just designed our own drivers again? Clearly the knowledge exists here...

C

Because we couldn't order quantities to get them built by anyone for a reasonable price?

[cjd]

Because we couldn't order quantities to get them built by anyone for a reasonable price?

Since when has that been an issue? :P

Reasonableness is in the eye of the beholder.

When reason dictates the ultimate... it's still less than the stuff we're "competing" against.

C

[chasw98]

jw (actually, omega, not w), is the frequency variable in radians in the S plane; to calcuate the transfer function in MathCAD or Maple, you'd define a range variable to index a vector, I usualy use a simple ratio function with exponential terms to get a certain number of data points spaced in each octave. I'll post this after I get back from my business trip to Austria, after the Christmas holidays. I want to transfer this to Maple, but maybe I'll just do this in MathCAD again.

Yeah, and I stayed at a Holiday Inn last night!

[Paul H]

Since when has that been an issue? :P

Reasonableness is in the eye of the beholder.

When reason dictates the ultimate... it's still less than the stuff we're "competing" against.

C

All very true ....

[jkrueger]

The speaker I have been waiting for

This whole thread just really floats my boat. 8O Please keep going. If I ever built a plan from here that was the be all and end all for my system, this one looks like it might be it. I love the two cabinet idea, and the all out performance stance. ;x( ;x( ;x( ;x( I can hardly wait to see how this turns out.


Jon

[oxcartdriver]

This whole thread just really floats my boat. 8O Please keep going. If I ever built a plan from here that was the be all and end all for my system, this one looks like it might be it. I love the two cabinet idea, and the all out performance stance. ;x( ;x( ;x( ;x( I can hardly wait to see how this turns out.


Jon

+1 on the all out performance stance. It will be interesting to see if I choose the 3-way, Isiris, or possibly both. Nothing like comparing speakers on your own audio equipment in your own home.

[diyDreamer]

Master of the Dark can i suggest this? Usher Beryllium Midrange



Price? Uch 560 euro!!! here
______________________________
Peace is a lie; there is only passion.

Roland.

[mmoeller]

jw (actually, omega, not w), is the frequency variable in radians in the S plane; to calcuate the transfer function in MathCAD or Maple, you'd define a range variable to index a vector, I usualy use a simple ratio function with exponential terms to get a certain number of data points spaced in each octave. I'll post this after I get back from my business trip to Austria, after the Christmas holidays. I want to transfer this to Maple, but maybe I'll just do this in MathCAD again.

Yest dark one. I realize the use of j(omega), just used w as a substitue for the curvy greek numeral.

The real question is if the omega is the same variable throughout the transfer function. It was mentioned that the Fc's would change, and omega being (2piFc), would change. So the question is, What is Fc for a three way polynomial numerator if each term is meant to represent T, M, or W? Also what is used for omega in the denominator? If there is only 1 Fc for the whole system then again what does Fc mean? Is it the split between the T and M because they are the only numerators with omega terms? or is it the middle of the midrange coverage w/ the filters spread a certain distance apart based on "a"?

I'm sure I'm missing somthing obvious.

edit for learning

I went into MathCAD and figured out some of the equations, along with the selection of omega steps in the array calculations. Fc looks to be the center of the midrange. So there is only 1 Fc. I think I've cleared it up somewhat. If the plots look like what should be expected, I'd be happy.

[JonMarsh]

Fc is the center frequency- every thing else derives from that and the gain/damping coefficient "a". "a" sets the corner frequency spread, Q, and midrange level. Sort of like a unified field theory, really, as the mathematical relationship of the HP and LP are all with respect to the center frequency and "a".

[mmoeller]

Good deal I got it now I must have been editing when you posted. Thanks Jon.

[Evil Twin]

Master of the Dark can i suggest this? Usher Beryllium Midrange



Price? Uch 560 euro!!! here
______________________________
Peace is a lie; there is only passion.

Roland.

Interesting technology, I have a lot of respect for Berylium driver technology (NS1000 Yamaha's).

At twice the price I would hope they would extend to the same 10 kHz top end as the Accuton C79.


C79-6rear.pdf

Regrettably that is NOT the case. They have suspicious impedance curve ripples just above 3 kHz, and "die" above 6 kHz.

I trust you will redouble your efforts to find a comparable driver with a better value proposition....

[Tommythecat]

jw (actually, omega, not w)

It's iw, not jw


lol

[kravi4ka]

+1 on the all out performance stance. It will be interesting to see if I choose the 3-way, Isiris, or possibly both. Nothing like comparing speakers on your own audio equipment in your own home.

Both...(salivating heavily). Kidney anyone?
Definitely +1 on the all out approach...

[mmoeller]

It's iw, not jw


lol

:rofl: ;x(

[ssabripo]

It's iw, not jw


lol

uh oh.......

I believe the Dark Lord will soon be talking to you about:

Image not available

:rofl: :lol:

[Evil Twin]

It's iw, not jw


lol

this is clearly an imposter....

My bionics and life support add 80 kg to my body mass, and I don't see that Shakespeare quoting Captain Picard lifting that combined weight into the air with a single blow...

[chasw98]

this is clearly an imposter....

My bionics and life support add 80 kg to my body mass, and I don't see that Shakespeare quoting Captain Picard lifting that combined weight into the air with a single blow...

Ummm, maybe you have lost some weight traveling around the galaxy, Dark Lord? :rofl:

[ssabripo]

Ummm, maybe you have lost some weight traveling around the galaxy, Dark Lord? :rofl:

Image not available

"Public Transportation has failed me for the last time..."

:lol:

[Evil Twin]

Yest dark one. I realize the use of j(omega), just used w as a substitue for the curvy greek numeral.

The real question is if the omega is the same variable throughout the transfer function. It was mentioned that the Fc's would change, and omega being (2piFc), would change. So the question is, What is Fc for a three way polynomial numerator if each term is meant to represent T, M, or W? Also what is used for omega in the denominator? If there is only 1 Fc for the whole system then again what does Fc mean? Is it the split between the T and M because they are the only numerators with omega terms? or is it the middle of the midrange coverage w/ the filters spread a certain distance apart based on "a"?

I'm sure I'm missing somthing obvious.

edit for learning

I went into MathCAD and figured out some of the equations, along with the selection of omega steps in the array calculations. Fc looks to be the center of the midrange. So there is only 1 Fc. I think I've cleared it up somewhat. If the plots look like what should be expected, I'd be happy.

Yes, now you appear to have it. Just adjust the commeon filter coefficient a until you get a transfer function which is realizable with the drivers you're considering- tradeoffs, of course, but with considerable elegance.

[jkrueger]

Just Wonderin'

I know that this is not a complete design yet, (You don't even understand how excited I am about this one) but if this is something that will be finished one day should it not be a sticky thread? I vote lets go with the Accuton C-79. :T

edit-Evil Twin is my hero. ;x( ;x( ;x( ;x( ;x( ;x( ;x( ;x( ;x( ;x( ;x(

[TacoD]

I think it's a great effort, but I hope it's not the old C-79. It's a nice driver but I do not like such small driver coupled to a 12". If it has to be Accution it needs to be the C2-90.

[Evil Twin]

I think it's a great effort, but I hope it's not the old C-79. It's a nice driver but I do not like such small driver coupled to a 12". If it has to be Accution it needs to be the C2-90.

There are tradeoffs at either end of the spectrum with driver selecction.

I am not sure exactly which driver you are refering to above- in the new part numbering scheme from Accuton, there is the C90-6-79, the C90-T6-89, the C173N-T6-90 (the old C90-T6), etc.

The C90-T6-89 and C152-T8-82 offer only marginal improvements in LF SPL extension over the C79, while having pronounced drawbacks above 5 kHz.

I don't have a set of C89's yet, but I have some C88's, which apart from grille revisions and a Titanium former and different surround are quite similar in fundamental contruction.





The nearfield plot will show a roll off above 2 kHz due to path length differences. The first real problems in linear distortion with this driver appear in the 4-5 kHz area.


The C173N-T6-90 has a first internal cone mode up pretty high for it's size, but there's no getting around the fact that that BECAUSE of it's size, 2.5 ~ 3 kHz is an upper limit for opeartion in the audible range.



The inescapable conclusion I have is that I can not make a conclusion at this point, and the first upper module will have a removable front baffle using furniture connector bolts for testing several drivers, prototyping transfer function proposals with MathCAD and initial evaluations using a DCX2496 to prototype transfer functions.

The Emperor has expressed his considerable displeasure at the ballooning budget for this "Design Study", "advising" me that FUBAR's like the last Battle Station project will not be suffered without dire consequences in the future.

[TacoD]

Yes I was talking about the C173N-T6-90, for the required spl levels you need a bigger mid. Maybe you need to do a 4-way... (8" lower mid). Every small driver has higher distortion, whatever brand you're using. If Duelund is a design parameter you need also look further than those pistonic drivers.

[jdybnis]

I don't have a set of C89's yet...

You are becoming the Noah's ark of Accuton.

[Evil Twin]

Let us consider the details of this Duelund transfer function calculation.

The Sith still use the old ways, revering the original creation of MathCAD under the primitive Disk Operating System used prior to the development of windowing operating environments.

The basic transfer function can be calculated normalized to one radian; first, an index and indexed range variable for frequency must be defined:



The individual transfer functions H(s) share a common denominator but different numerators:



The derivation is explained in Steen Duelunds papers available on the Hypernet. Note that a wide range of variation in net transfer functions for a three way are obtained with varying values of the damping coefficient aleph; if this value is reduced to the square root of 2, the midrange section disappears completely and what is left is an LR-4 two way.

Next, using the absolue magnitude function and converting to dB (taking the log of the transfer function and multiplying by 20), the individual transfer functions are plotted, in this case using an aleph value = 2X the square root of 2. Note that the LP and HP networks show an ultimate slope of 24dB/octave, while the midrange is 12 dB/octave.




This displays the transfer function curves well, but would be more interesting normlized to a range likely for the center of the midrange bandpass, perhaps 1 kHz. Multiplying the s frequency variable by 1000 sets a 1 kHz center frequency.




In this plot, markers were added for the -3 dB point corresponding with the peak midrange level, as well as the -18 dB level where the HP and LP cross, and which defines some possible limits for required driver response in the midrange as well as woofer and tweeter.

Perhaps a midrange bandpass from 175 Hz to 6 kHz seems excessive- the tuning coefficient aleph can be reduced, relaxing the requirement on extension and level for the midrange. But this moves the corner frequencies for the HP and LP sections closer, and increases the demands on those drivers, as shown below.




With some experimentation with the damping coefficient aleph and the center frequency, a proposed 3 way system transfer function can be developed, which should then be checked for intended playback levels using tools such as SL's SPLMax which calculate the maximum "linear" SPL possible for a given driver Sd and Xmax and operating frequency.

[augerpro]

There seems to be alot driver FR overlap with this method. Is combing a problem with the mid and tweet? Also what is the benefit of this method?

[Evil Twin]

The interesting properties of the Duelund approach to synthesizing the three way filter at once are several-

• The combined transfer function sums to unity across the frequency spectrum as does a simpler two way crossover, and starts from only one equation
• The transfer function is set by the center frequency Fc and the coefficient damping factor, aleph, and can realize a range of transfer functions with wider or narrower spread of the corner frequencies for LP and HP and midrange level just by adjusting aleph
• If aleph is reduced down to the square root of 2 (~1.414), the midrange portion by that point dissappears, leaving an LR-4 two way.
• For aleph = 2 or more, the step response has no ringing- fast transient settling.
• For aleph = square root of 3, the transfer function is a three way Bessel with the HP, LP, and peak of midrange combining at - 9 dB; midrange driver efficiency and output level requirements are the least, making the smaller wide range drivers (Tangband, HiVi, etc) suitable for a midrange filler while keeping frequency requirements reasonable for woofer and tweeter with Fc of 1400-1800 Hz.
• The complete transefer function from low to high exhibits only one phase rotation through the entire three way function, with lower group delay peaking than a single LR-4 two way.
• The midrange is wired out of phase to the woofer and tweeter (a minus sign in the complete expression of the numerator for the midrange)
• All three drivers are always in matching relative phase at any frequency, so the possibility of destructive interference or off axis lobing is minimized. This results in a reported sense of integration and not being able to identify sounds as coming from one driver or another

Note that there is a greater sense of driver overlap because the graphs are posted with a scale extending to -60 dB. I will post an example later with a more typical Y axis scale.

This is for another project in consideration with a dome midrangea and small form factor dome tweeter.



Note that the drivers are -6dB at the crossover frequency, not -3 dB as in most conventional crossovers. This may be adjusted to fill in the power response; it matters whether a BBC dip is preferred (better for most commercial program material), or something closer to flat power response, which could justify elevating the midragne level a couple more dB.

This suggests a two or three way switch control- "Presense".