Introduction to designing crossovers without measurement

On the topic of using a resistor to bring down the level of a tweeter and whether it is best to place it before or after a second order filter. Either method can work but each has it's own effect. Placing it before the filter may be more predictable when not using measurements and since tweaking by ear requires uncomplicated variations to keep track of progress, this is what to expect...

The first plot shows the difference in response when placing a few different values of resistance after the filter, including using no resistor. The treble level is reduced but the damping of the filter is also reduced. This raises the response near resonance. It could be fixed by using a full L-pad rather than a single resistor since the added parallel resistor tends to restore the impedance the filter sees. It would then be near to an ideal solution.

However, having to change two interdependent components at a time while trying to keep track of the sound can increase the time between tests. It could be much easier to make the level change followed by an adjustment near the cross, as two separate operations.

Thus, the second plot shows placing the resistances before the filter. This gives a cleaner reduction in level. What is not as easy to see is that the crossover frequency gets lower as the level is reduced, which may call for reducing the value of capacitance or inductance (next post).

w1.png


w2.png
 
What if you try to raise the filter frequency by reducing the value of only the capacitor or only the inductor (in this case by 20%).

Compared to the original red trace, the blue and yellow traces each end together down at 55dB, meaning they are each able to raise the frequency. The difference is in the output at the 'knee' of the filter. The blue trace is when reducing the inductance, where the Q factor of the filter increases, and yellow is when reducing the capacitance.

w3.png
 
Last edited:
Hello Gents,

I recently started to develop a higher interest in cross-overing. I've ready lots of articles including this thread ofcourse but I fairly new to this world so forgive me if I'm saying something stupid.
I've refurbished my Kef Reference Three's from 1996. Replaced all the bennic and alcap caps with electrolit mundorfs and a couple of alcaps. I've measured all the drivers in free air with a dayton V3 but lost the data so Idon't have the individual driver impedance files or T/S parameters. The tweeters were re-fluid and I've replaced all the foam sealings. All the surrounds appeared to be in quit a good state. The speakers definetly sound better to my ears after the cap replacement so that's one step forward....
Besides the general increase of interest in how passive XO's work I wander whether or not it could be usefull (besides the fun) to have closer look at the scheme taking into account I have to dismantle the speakers (again) in order to obtain the driver data.
If I look at the scheme I only recognise a third order tweeter with some L-pad attenuation. 2 second order band-pass filters (upper and lower mid). Upper with L-padding and the lower which appears to be a notch filter without seperate resistor (didn't measure the inductors.......).
The subs appear to have "high" first order combined with a second order low-pass including a noth filter. Again without seperate resistor.

I'm interested in any opinion.....and whether or not it might be worth having a go to improve the XO's.

Kef Ref refurbish pics

Screenshot 2025-02-28 103821.png


Mesured values of the old components with the dayton V3
Screenshot 2025-02-28 104117.png


Only electrical response with nonimal 4 oh impedance for what it's worth.....
Screenshot 2025-02-28 103953.png
 
  • Like
Reactions: AllenB
That's an interesting design. Is that a coax, and are those lower woofers coupled for force cancelling?

One thing about crossing a coax is that it's a little like a synergy horn. This means there's some commonality in their radiation and it's not as difficult to fine tune their cross. All they want is some cooperation in phase and a good overall tone. For this reason, you could learn the voicing you need from them by equalising.

The three sections in parallel shown for S2 can be drawn in any order and still be the same. It is 2nd order bandpass with one notch filter. These notch filters are mostly limited in depth by the resistance of the coil.
 
That's an interesting design. Is that a coax, and are those lower woofers coupled for force cancelling?
What do you mean with coax? (besides the TV signal carrier). Each speaker has 2 physically coupled 8 ohm subs in paralell so nominal 4 ohms with each their own cavity and their own internal reflex "horn" (don't know what it's called). I haven't dived deep into any of the cabinet designs yet. First things first🙂
 
Btw...is it correct to first look at the RC zobelling part for each driver killing or flattening the rising frequency impedance and then look at the attenuation using L-pads as then the impedance used for calculating the L-pad resistors becomes more predictable........Not sure though...🙂
 
I've refurbished my Kef Reference Three's from 1996.
Those are very well-regarded loudspeaker system. I expect that your refurbishment will have brought them up to their original performance levels when they left the KEF factory.
I'm interested in any opinion.....and whether or not it might be worth having a go to improve the XO's.
I assume that the original crossovers were very carefully designed by KEF. As best as I can determine, the nominal crossover frequencies were 140Hz, 400Hz, and 3kHz.

Doing your own modifications may lead to changes that you consider to be improvements. However, you will need to undertake very careful measurements so that you can correlate any audible differences with the changes in the crossover circuit. With respect to the loudspeaker system's frequency response, a slight boost here or a slight cut there can really alter your perception of the loudspeaker's subjectively-appraised "goodness".

If you do embark on such a project, I think that you will need to very carefully collect together a detailed set of reference measurements of the filtered responses of each of the drivers, as well as the overall frequency response. Once you have those, you would then have to add measurements of the raw driver responses to your database. Now, you are in a position to use the crossover simulator to add all the various components. Then, you could compare your simulations with the measurements. Once things are in good agreement, you could work on some crossover component changes and predict their effects with some degree of certainty.

Another approach that could be taken is to go active and use DSP to create a completely new active crossover design, using the ability of DSP to do some very delicate tuning of the new design. The use of DSP offers the potential to make changes that would be very difficult to achieve with a passive crossover.
 
  • Like
Reactions: wkloppen
If you do embark on such a project,
Well I certainly do consider it and I'm aware more or less whatI'm doing. There is a charming element in a passive crossover I guess. It's analog over digital. It has more feel knowing how the the actual components physically try the best they can to control the signal. I've build the wolverine amp 2 years ago. If I would go DSP, I will be needing 4 amplifiers...or not? Not something I'm looking forward to I think.....
 
  • Like
Reactions: witwald
are read as so many thousand and not as decimal places
Hello Wolf.....the values of the inductors in the schematic were values from another X-O of the same speaker and confirmed by Kef. In the excel the values measured of the caps were the actual values of 1 of the X-O's of my own. The values of the caps in the schematic are indeed the values labeld on the caps. Al in all not a 100% representative of the actual situation.🙂
 
I don't think you understand me. Over here, numbers of 100,000 are one hundred thousand. 100.000 is 100. Over there on your side of the pond, 100,000 is 100. Being that you used what appears to be Xsim, and is a USA made xover Modeler, I feel your capacitor values are erroneously huge.

While I feel you might actually have a clue what you are doing, I feel the Modeler does not and will not reflect the circuit as you have drawn it from your European perspective.
 
  • Like
Reactions: stv