It's quite simple to make a filter that goes up with a first-order slope from 3 kHz and then stops rising somewhere in the ultrasonic range. Two resistors and one capacitor (with buffer op-amps if needed) can do the trick. It doesn't look much like the curve on the bottom left of post #17 with the vertical axis reversed, though.
Thanks MarcelvdG!
That's exactly what I was hoping for and direction I'm heading after these responses so far. Also, just realized my horn is the MR931-12 not the MR944, but I suspect it's not that different.
Incidentally, I have a number of op amps handy and would love to make use of one of these. (opa1612, muse02, lme 42970, burson vivid/classic, etc from when I was op amp rolling my DAC). I do have another balanced 15v output on the power supply to power said circuit, if need be, but would need to purchase a regulator, which are readily available)
Would you (or anyone) have any recommendations on how to figure out what this circuit might look like and R/C values? I recall there's a circuit designer/simulator app somewhere, but don't know much about that area.
Perhaps also recommended op amps?
And where in the circuit would be the best place to insert it into the circuit? (just after the input xlr connector? Just before output xlr? - hoping it's that easy!).
I do have another 15v output on the power supply to power said circuit, if need be, but would need to purchase a regulator.
Thank you to everyone for their responses so far!
That's exactly what I was hoping for and direction I'm heading after these responses so far. Also, just realized my horn is the MR931-12 not the MR944, but I suspect it's not that different.
Incidentally, I have a number of op amps handy and would love to make use of one of these. (opa1612, muse02, lme 42970, burson vivid/classic, etc from when I was op amp rolling my DAC). I do have another balanced 15v output on the power supply to power said circuit, if need be, but would need to purchase a regulator, which are readily available)
Would you (or anyone) have any recommendations on how to figure out what this circuit might look like and R/C values? I recall there's a circuit designer/simulator app somewhere, but don't know much about that area.
Perhaps also recommended op amps?
And where in the circuit would be the best place to insert it into the circuit? (just after the input xlr connector? Just before output xlr? - hoping it's that easy!).
I do have another 15v output on the power supply to power said circuit, if need be, but would need to purchase a regulator.
Thank you to everyone for their responses so far!
ok... found a circuit example and will research the simulation tool but appreciate any help if this is a cakewalk for someone who is an expert.
So open questions are:
1. Where in the circuit to add?
2. If seperate power is needed or can be integrated into existing circuit and how.
3. Does anything have to change, since it's a balanced crossover? (ie dual op amp instead of single?)
So open questions are:
1. Where in the circuit to add?
2. If seperate power is needed or can be integrated into existing circuit and how.
3. Does anything have to change, since it's a balanced crossover? (ie dual op amp instead of single?)
Linkwitz Lab has pretty much all the info you need. Not sure about balanced xo's. Quite often I've read that the xo is done non balanced and then there is a circuit after to create a balanced signal.
Rob.
Rob.
I've tried to mimic the response of the MR944A from post #17 in a program that (among other things) can calculate the magnitude and phase plots for given poles and zeros. I got this:
The upper graph is the magnitude plot, vertical axis 5 dB per division (well, per dot really), horizontal from 1 kHz to 20 kHz. The poles and zeros were
Poles and zeros in hertz rather than in rad/s.
It is very well possible to make a filter that has a response that is the mirror image (mirrored in the dB axis) of this up to some ultrasonic frequency, but it is more complicated than a first-order correction from 3 kHz onward.
The upper graph is the magnitude plot, vertical axis 5 dB per division (well, per dot really), horizontal from 1 kHz to 20 kHz. The poles and zeros were
Poles and zeros in hertz rather than in rad/s.
It is very well possible to make a filter that has a response that is the mirror image (mirrored in the dB axis) of this up to some ultrasonic frequency, but it is more complicated than a first-order correction from 3 kHz onward.
Thanks, this does have some good info and I'll have to dig in.
Incidentally, I did also look at the Nelson Pass's 6-24 crossover and the Linkwitz diy ASP boards, but the 6-24 is 2 way and trying to modify one of the LX ASP boards seemed daunting, but did seem to have options for delay and some eq shaping. Perhaps after I have a bit of experience with this simple mod.
Regarding the first-order correction from 3 kHz, these networks should do the trick. The response stops rising around 33 kHz. The circuit without op-amp attenuates 11 times at low frequencies and has unity gain at high frequencies and needs to be driven from a low and loaded by a high impedance. The one with op-amp has unity gain at low frequencies and amplifies 11 times at high frequencies.
Thanks agian MarcelvdG! I truly appreciate your time and effort!
I'll start ordering parts/modeling and building this out!
I'll start ordering parts/modeling and building this out!
Adding my starting point as a frame of reference. 1/24 smoothing. Boosted 3k to try and help diminish the dip, so that can be flattened by eq.
Last edited:
figured out the snapshots.
1/3 smoothing - both speakers - seating position
Coorrection to above... Boosted 1k, 2.5k and 3k on the EQ to try and help diminish the dip, so I should be able to flatten those humps before and after the 1-3k range.
1/3 smoothing - both speakers - seating position
Coorrection to above... Boosted 1k, 2.5k and 3k on the EQ to try and help diminish the dip, so I should be able to flatten those humps before and after the 1-3k range.
Last edited:
Exactly. Additionally the drivers should also be measured simultaneously without any filters in order to establish the relative acoustic centres: otherwise the summing will not be correct. It seems you have the REW curves ultra smoothened. You need more resolution and 5dB/div. in order to properly "map" bigger peaks and valleys: these are now largely hidden and thus invisible. We have some insight on the basis of the Altec data, but only a measurement of your drivers will tell whether yours behave identically. The Altecs alsp are 10dB/div. which is pretty rough for x/o design.
Online: only with the same baffle size and dimensions will online measurements be representative.
Thanks Bowden, very helpful and will make some measurements and post for each driver separately with no smoothing and 5db / div.
Some clarifications.
Should I send a full 0 to 20k hz sweep to each driver or do I limit the frequency for the tweeter as not to damage it? Perhaps I should set the sweep to start at 400hz for that one?
Incidentally, the system sounds quite good and nicely balanced, in spite of the dip and I was wondering why, compared to the dsp, which has almost continuous eq bands and a more linear frequency response. Looking into this, I learned that the dip seems to align fairly close with the “BBC dip”, by mere coincidence.
Some clarifications.
Should I send a full 0 to 20k hz sweep to each driver or do I limit the frequency for the tweeter as not to damage it? Perhaps I should set the sweep to start at 400hz for that one?
Incidentally, the system sounds quite good and nicely balanced, in spite of the dip and I was wondering why, compared to the dsp, which has almost continuous eq bands and a more linear frequency response. Looking into this, I learned that the dip seems to align fairly close with the “BBC dip”, by mere coincidence.
Last edited:
Tonight I will try to find time to mimic the passive filter you posted in an active opamp version. No time now unfortunately.
Boden,
Apologies for misspelling your name previously. Just noticed my iPad autocorrect adds a “w“ every time.
Apologies for misspelling your name previously. Just noticed my iPad autocorrect adds a “w“ every time.
In post #25, I tried to find a set of poles and zeros that give about the same response as your tweeter. Assuming the response matches the real thing sufficiently accurately, correcting for it can be done as follows:
A. Choose the poles and zeros of the correction filter such that there is a correction filter zero on each pole and a correction filter pole on each zero. That is,
Correction filter poles:
-3 kHz +9.5j kHz
-3 kHz -9.5j kHz
Correction filter zeros:
-1.8 kHz +5.4j kHz
-1.8 kHz -5.4j kHz
-3.5 kHz +12.2j kHz
-3.5 kHz -12.2j kHz
B. The correction filter now has more zeros than poles, meaning its response has to keep increasing indefinitely with increasing frequency. As that is impossible, add poles to level off its response at some ultrasonic frequency. For example, the poles of a 30 kHz second-order Butterworth filter:
Extra correction filter poles:
-15 sqrt(2) kHz +15 sqrt(2) j kHz
-15 sqrt(2) kHz -15 sqrt(2) j kHz
C. We now have four poles and four zeros. A Linkwitz transform filter (biquad) can realize two poles and two zeros, so we need two of them.
When you look at https://www.linkwitzlab.com/filters.htm#9 and the links it refers to, you see that the pole positions are usually expressed in f and Q rather than in real and imaginary parts. Converted into those terms:
First correction filter pole pair:
fp1,corr = sqrt(3000^2 + 9500^2) Hz ~= 9962.429423 Hz
Qp1,corr = sqrt(3000^2 + 9500^2)/(2*3000) ~= 1.660404904
Second correction filter pole pair:
fp2,corr = sqrt((15000 sqrt(2))^2 + (15000 sqrt(2))^2) Hz = 30 000 Hz
Qp2,corr = sqrt((15000 sqrt(2))^2 + (15000 sqrt(2))^2)/(2*15000 sqrt(2)) ~= 0.707106781
First correction filter zero pair:
f01,corr = sqrt(1800^2 + 5400^2) Hz ~= 5692.099788 Hz
Q01,corr = sqrt(1800^2 + 5400^2)/(2*1800) ~= 1.58113883
Second correction filter zero pair:
f02,corr = sqrt(3500^2 + 12200^2) Hz ~= 12692.12354 Hz
Q02,corr = sqrt(3500^2 + 12200^2)/(2*3500) ~= 1.813160506
You can plug these values into Linkwitz's pz-eql.xls to find suitable component values for the two cascaded Linkwitz transform circuits.
A. Choose the poles and zeros of the correction filter such that there is a correction filter zero on each pole and a correction filter pole on each zero. That is,
Correction filter poles:
-3 kHz +9.5j kHz
-3 kHz -9.5j kHz
Correction filter zeros:
-1.8 kHz +5.4j kHz
-1.8 kHz -5.4j kHz
-3.5 kHz +12.2j kHz
-3.5 kHz -12.2j kHz
B. The correction filter now has more zeros than poles, meaning its response has to keep increasing indefinitely with increasing frequency. As that is impossible, add poles to level off its response at some ultrasonic frequency. For example, the poles of a 30 kHz second-order Butterworth filter:
Extra correction filter poles:
-15 sqrt(2) kHz +15 sqrt(2) j kHz
-15 sqrt(2) kHz -15 sqrt(2) j kHz
C. We now have four poles and four zeros. A Linkwitz transform filter (biquad) can realize two poles and two zeros, so we need two of them.
When you look at https://www.linkwitzlab.com/filters.htm#9 and the links it refers to, you see that the pole positions are usually expressed in f and Q rather than in real and imaginary parts. Converted into those terms:
First correction filter pole pair:
fp1,corr = sqrt(3000^2 + 9500^2) Hz ~= 9962.429423 Hz
Qp1,corr = sqrt(3000^2 + 9500^2)/(2*3000) ~= 1.660404904
Second correction filter pole pair:
fp2,corr = sqrt((15000 sqrt(2))^2 + (15000 sqrt(2))^2) Hz = 30 000 Hz
Qp2,corr = sqrt((15000 sqrt(2))^2 + (15000 sqrt(2))^2)/(2*15000 sqrt(2)) ~= 0.707106781
First correction filter zero pair:
f01,corr = sqrt(1800^2 + 5400^2) Hz ~= 5692.099788 Hz
Q01,corr = sqrt(1800^2 + 5400^2)/(2*1800) ~= 1.58113883
Second correction filter zero pair:
f02,corr = sqrt(3500^2 + 12200^2) Hz ~= 12692.12354 Hz
Q02,corr = sqrt(3500^2 + 12200^2)/(2*3500) ~= 1.813160506
You can plug these values into Linkwitz's pz-eql.xls to find suitable component values for the two cascaded Linkwitz transform circuits.