Cross-Over Frequency Selection


I am designing a two way system. The tweeter and the mid-bass drivers have a good 3khz+ overlap region between them. There are no nasty cone break-ups in the mid-bass driver at high frequencies and the roll off is very smooth and the resonance frequency of the tweeter is very low (600Hz) again with a smooth roll-off towards low frequencies. I have a XTA active cross-over so I am playing with different topologies and frequencies both symmetric/asymetric with a nice freedom.

When I measure a flat response in my reasonably dead room, I listen to the combination, however, as you would expect, every on-axis flat response sound different because of the off axis radiation (power response) of the chosen filter type. I even try to chose a low order high pass for the tweeter and a higher order low-pass for the mid-bass to help the time alignment a bit, but they do not sound any better when compared to some symmetric combinations that I've made.

I can either go on and on with playing this game till I say "enough" and sattle for a combination, or learn if there is a way to define an optimum cross-over frequency. Learning is my first choice at the moment ;)

Without any special considerations.... the normal way is to find the bandwidth of the each driver. This gives an low and high frequency limit for each driver, F(low) and F(high). Multiply FL and FH times each other and take the square root. This provides the center (Fcenter) of the bandwidth for the particular driver. Do this for both drivers. Now take the two values for F(center) and multiply and take the square root. That will be the correct crossover frequency.

If there are no other considerations this will be the optimal frequency.

i did many 2-Ways and imo 3Khz is way too high.

Given the tweeter has low distortion down to 2Khz
i would tend to go 2,0 ... 2,3 Khz if possible, maybe even

Why ?

1) Avoid excessive beaming of the woofer and avoid
excessive discontinuity in power response due to
different directivity of woofer and tweeter.
(It is different enough at 2Khz believe me ...)

2) Choose a frequency where the ear is less sensitive,
you will have disturbances at crossover regardless of
the type of filter you use.

3) This is to be considered imo:

Blauertsche Bänder ? Wikipedia

Directional bands correspond to the angle dependent
filter function of our pinna and head related transfer function.
There are some ranges, where peaks and notches are evaluated
in a way to influence/disturb spatial imageing more than
other ranges.

If the tweeter can do, go lower ...

Keep the directivity of woofer and tweeter as similar
as possible at crossover frequency.

That is the key for a speaker to sound homogenouusly.


Then you can tweak the filter slopes due to amplitude
and phase and also do group delay compensation by
mounting the tweeter recessed ... but the above
part is the more important.

Choosing an XO frequency is not a "compare on axis
FR of drivers task". There is more to it and it is one of
the essential and first decisions, when planning a 2-Way.

In fact the drivers should be selected a f t e r
that decision was made thoroughly.

If recessed mounting the tweeter, avoid protruding
and sharp edges on the baffle near the tweeter,
it will spoil the effect and make things worse than


Kind Regards
Last edited:
Confusing part

Without any special considerations.... the normal way is to find the bandwidth of the each driver. This gives an low and high frequency limit for each driver, F(low) and F(high). Multiply FL and FH times each other and take the square root. This provides the center (Fcenter) of the bandwidth for the particular driver. Do this for both drivers. Now take the two values for F(center) and multiply and take the square root. That will be the correct crossover frequency.

If there are no other considerations this will be the optimal frequency.

Interesting point, I don't remember reading this before anywhere, do you mind if you share your reference or how did you come to this conclusion.

A little input from my side:
My low frequency driver works between 50-4000Hz roughly and the high frequency driver is around 1000-22000Hz - roughly again. These figures suggest a cross-over frequency about 1.5KHz.

I am looking in my notes now and I see some successful combinations near this frequency and they are as follows:

1- 1.47KHz, Bessel 2nd order, symetric
2- 1.6Khz, 2nd order Butterworth for high pass, 4th order LR for LPass
3- 1.8Khz Bessel 2nd order high pass, 1.4Khz Butterworth 2nd order low pass

They all sound nice to my ear in one way to another and in a reverberant space (my garrage :) ) when I measure the power response (or something close), they all measure without any disturbence in the pass-band.

This is confusing for me. Isn't one of the combinations must be better than all the others?

Well- that reference is difficult to pin down. Been doing this to long. The concept is simply center the crossover between the two drivers. As frequency is a "log" relationship multiplication and square roots find centers and not adding and divide by two as done when averaging. Pretty standard filter theory stuff actually simply applied to loudspeakers. :)
Hello sum,

to me the question arises whether to use
the drivers on axis response or energy response
to estimate bandwith for this purpose ...

I must admit that i estimate energy response to
be the more suitable one or at least a mixed

Choosing XO frequencies without taking
the physiological side of hearing into account and
also polar dispersion seems a bit too
"straight forward" to me.

Kind Regards
Got the phase in mind-

Well yes the approach is the simplest. I actually measure the energy bandwidth of drivers which have already been shown to maintain a certain angle of radiation. So only "previously qualified" drivers are used. Next the energy bandwidth is measured and the center of that bandwidth is assigned a zero phase condition. After knowing this must be the zero phase frequency the phase response of the driver is measured. This leads to an "in constant phase" bandwidth result. Around here the constant phase bandwidth is the one used to determine the crossover frequency. Assigning the center of the energy band zero phase seems to make a lot of sense because the center of the energy bandwidth should be where the driver is working correctly. That is the assumption.

It often turns out drives fail to stay in phase for wide bandwidths with something around a decade being a common limit. The result is also often the "ends of the bands" between two drivers do not overlap and stay in phase so the crossover is set where the drivers do not maintain constant phase anyway.

The simplest way to measure phase is to use a delay and the test signal as applied to the speaker. The amplifiers test signal output is delayed to match up with the signal from the microphone. Again this is done at the energy center frequency. Delay the test signal until it is exactly in phase with the microphone signal using a tone burst or some other test tone where beginning and end may be accurately identified so it is easy to tell when the delay adjustment is not a full or multiple full cycles off. After this is done change frequencies and check the phase shift by either changing the delay until a match is again achieved (noting the new delay time) or using Lissajous figure on a scope or however you like. Phase will vary with frequency eventually becoming 180 degrees or more at both ends of the in phase bandwidth. The "in phase" bandwidth is the one that sounds good and makes sense to the brain. As a note, the phase response from all the impulse systems are invalid because the assumptions made to determine phase are incorrect. What is written above is the easiest straight forward and simple way to measure phase requiring very little equipment.

If a tone burst set at the center frequency is used the direction of the first half cycle of the burst goes will be the same as the detected signal if everything is in correct phase. If speaker wires are reversed then the initial positive going burst will result in the detection of an initial burst going negative. This allows correct determination of the absolute phase of the driver. This is very important when using horns as absolute phase is not necessarily positive when the diaphragm moves forward. All open dome tweeters (without lens or waveguide) will reverse phase in relation to their size. A 1" will flip about 6700Hz and 20mm will flip at 8400Hz and a 1.25" will flip at 5400Hz. I have never seen an open dome tweeter that does not flip phase in spite of what all the people with impulse test systems claim with their invalid phase measurements. Of course, those same people (to the date of this post) refuse to measure the phase response by the above method and compare that accurate value to the impulse test system.

If the result can be a speaker which maintains a constant phase over a wide bandwidth then clarity and stereo image will be superb. Once one has heard an in phase speaker it is very difficult to return to the muddy sound of all the rest.

Also note, almost all speakers made have horrible phase response or if they do have a good area of flat frequency, tonal balance, and relatively constant phase that "sweet spot" is very small, less than 15cm. A marvelous speaker will have a large sweet spot and likewise mitigate the concerns of polar response and many other forms of miladies.

Good luck-:)

choosing a crossover point

I have come across a few rules of thumb over the years. Many of which I have lost the original reference, but here they are anyhow:

For a two-way speaker:
high pass tweeter at 2-3 times Fs, never lower than 2*Fs
low pass the woof where off-axis response difference between 0 degrees and 30 degrees is less than 3-4 db difference

The center to center distance should of the drivers should be less than about 13578/Fxover. So, for example, with a driver crossover at 1250 Hz, the driver spacing center to center should be no more than about 11 inches, for example. This insures a good vertical window and good power response integration in the crossover region.

Hope this helps to some extent...

Yes, the final analysis! Everyone here "calibrates" a system at the final install with speakers and amplifiers in place. First measures all the drivers using the amps of the installed system then design the crossover as described. As suggested changing anything tends to change lots of things. This is one reason an "off the shelf" speaker system without calibration in the actual install might not "always be what you hope for." With active crossovers, DSP, and plenty of testing a system can be far better than any non-calibrated system.


Paid Member
2004-10-07 6:05 am

At this point you've gotten more into the craft of crossover design than the hard science. And you've heard the difference. Room response becomes so difficult to measure that you really have to tune by ear in the final stages.

If you can spread the frequencies some (15-30%) that can often help. Try that if you can. You have already done something similar with asymmetric slopes.
I have used active 3 way crossovers (DG13R) since 1972 from the audio amateur mag,in it was the first ones I used,I found the approximate center freqs,and set the crossover freq. about 20% under it's suggested range,they supplied a bell labs chart to figure the exact freq=R=C,most was already figured out ,what I did do was adjust these points as they sounded,I also had six tiger 207a amps with volume controls,i would suggest you put a heavy diodes( if I remember correctly they were back to back) across your tweeters I did fry a few speaker lab products,but mostly how did they sound is where I adjusted them ,I had a halfler dh 101 pre amp into a BSR EQ into DBX sound processors 3 of them expansion,active dynamics,and bass processor,I DROVE IT HARD TOO I rattled plates two or three houses away,lol.People would call and request songs,lol.I had 8 ribbon tweeters,4 soft dome mids with a speaker labs horn,2 12's and 2 15's for bass all tuned by ear was the best in the long haul,I used the eq to boost or cut and sometime let that guide me to move a crossover point when I had finished I took the eq out ,Just my 2 cents!!
Last edited: