some practical tips for speakerbuilders

I am going to make a lot of enemies here,  because crossover design is supposed to be SCIENCE and contain a lot of theories, phisics and math. I have been there and done that, and  - I did read books, I did use softwares, I followed Thiele-Small parameters and Linkwitz, Bessel, Chebychev formulas. I wasted 10 years for all that stuff fanatically. You can see the proof in P0 to P10 designs.

In short (maybe I am simply too stupid, who knows?) all the science has brought me to a corner, where I have done everything I could but achieved nothing. No music. Just well measuring speakers.
The most deceiving part is chasing those measurement anomalities like bumps, dips and peaks. Like cutting table's legs. You cut and equalize and cut and you are left with a table without legs. With speakers - you equalize and flatten the wrinkles and at the end everything just does not sound like music at all.
Commercial makers MUST make flat speakers because A) they are afraid of magazines catching them with abnormal responses and B) because they use shitty plastic drivers for 1 euro from China.

There is only one way out of this corner. You throw away all these theories and start building music making machines. LIVE sounding speakers. I do NOT mean abnormal measuring distortion generators but on contrary - speakers made with XO being in tune with the driver own characted and intelligenly addressing the issues to acheve speaker that both measures well and sounds great.  The theorical model, XO calculations and  and computer modelling are the WORST starting point. Listening to naked drivers in their real room position is the BEST STARTING POINT.

First - get rid of all software. Best way is to format C:

I AM DEAD SERIOUS ABOUT IT. But I do not mean measurement software but modeling and calculating software that produces ready recipes.

Second - choose drivers which are good. (As described in the Nirvana chapter)

I will repeat again: starting point is broadband midbass. Size 20 cm, (or oval close to 20 cm) paper cone, alnico magnet , or ferrite if alnico is not possible. Best source of such drivers is old german made radios like Saba, Telefunken, Siemens, Philips and so on. They usually have sensitivity or efficiency of circa 100 dB/W/m

The Bass

The bass foundation must be done by a large driver. Minimum 12 inch, better 15 inch, best - of course - you guessed right - 18 inch. The drivers over 18 inch are too rare to mention. Japanese freaks bought them all already and mounted in their attics.
The "quality" of bass driver  is not as important as many people believe. The bass driver is a big piston moving the most air volume relative to the volume of the room. Cubic  centimetres versus cubic meters. Of course heavy cast baskets are good, but even stamped sheet baskets will do. Do not spend 2000 euro on super bass. The "quality" factor is important only for commercial speakers which use 4 " or 6 inch bass driver imitations.

The TWEETER

The top end must be reproduced by a driver which has sensitivity similar to the midrange - and this is a difficult part. Most silk domes are max. 93 dB. Thats too little. We need 97-98 dB. Paper alnicos have that, some ribbons have that too. Paper cone tweeters should have at the back of the cone an open basket not sealed.
Top end response of the broadband drivers , even the best ones like like Lowthers or Fostex's is NOT ENOUGH. Our loudspeaker needs a real tweeter - or call it - a super tweeter.

Now comes my philosophy - how to get the crossover (XO) from scratch, without even fully understanding whats going on.

If we agree that the main part of the music is reproduced by the broadband midbass, the two other drivers only augument the sound in botom-most and top-most octaves.
We must LET THE MIDBASS PLAY. Not limit it, not flatten it, not use notch filters or any manipulation by XO. Let's make it play the way it plays. If you do not like the way it sounds "naked", please change it.

In order to protect it from damage from excursion, and audible distortions from overloading and doppler's effect, we will restrict slightly the bass signals. The antique alnico 20 cm drivers handle not more that 1 - 2 watts of musical signal without filters. Thats not enough headroom. But thats enough volume !!! 106 dB  of pure silk.
So I use a capacitor in series just to make sure that at 40 Hz I don't get too much excursion - no more than 5 mm maximum. That's at the high volume listening level.

In 99 % of cases a 100 uF MKP cap works great. Any non-electrolytic is good. There are for example good MKP caps for passive power compensation in home appliances.
If I listen to a programme via such cap and without (material is vocal without bass content) I hear absolutely no sound degradation due to this cap. It is transparent to the midrange content. I cant detect when this cap is shorted in bypass. It is completely transparent. DONT EVER SPEND $$$ on the  exotic caps here !! Please !


So we have done the first element of the crossover (XO) :  a first order electrical high-pass  filter superimposed over the natural rolloff of the driver itself and again superimposed over the open baffle effect (too narrow baffle compared to infinite). These three elements add up to circa second order accoustical rolloff at around 300-400 Hz. (If the baffle is 40-50 cm wide).

The bass driver must only support the midrange where it is weak. The two response curves must "meet" in the exact point of 400 Hz. Better some overlap than a dip.
To do a bass response that crosses at say 400 Hz we must act strongly. Otherwise the natural upper end  rolloff of the bass happens at 2 or even 4 kHz. Too high. So we need a second order filter - symmetrical accousticly to the midbass response after filtration.
Because changing inductor coils is not practical (they are bulky, expenceive and hard to get) - we choose a given size that must do. Any size from 3 to 5 mH. DCR must be low (below 1 Ohm, that translates (for air-cores) to the wire of 1,4 mm or up to 2 mm diameter.  0,5 Ohm is a beautiful coil. Again - no flatfoil ribbon coils please.

We have the coils for bass. That is one kilo of copper. With a dust iron core - we may get away with smaller physical sizes. I believe we can use a core and get away with that compromise because the core distortions happen only at 10 x or even 20 x  the current that is practically usable. ( we VERY rarely exceed 1 ampere in peak demand).

BASS Low Pass filter

After adding the coil in series with the bass driver, we add a paralell capacitor after it. 100 uF is a good ballpark value. This cap directly shorts the driver terminals. Then we measure the whole thing with a microphone callibrated in dB. And a true RMS meter or better - laptop with TRUE RTA software (www.trueaudio.com)  (note that this is NOT the software that designs speakers, it just changes a laptop to a measuring device)  If the bass curve reaches too far - it is time to increase the cap. Otherwise - decrease it. Make incremental changes of 50 % not less. The bass and mid drivers (with the mid filter installed)  must be measured first separately and then - together. If the "together" response has a dip in the XO region compared to the individual components - it means we must reverse one driver's polarity (any one of the two). Then we measure again - the polarity with higher response in XO region is the correct one.


me and my true spectrum analyser and pink noise

That means we completed the second stage - the bass to mid XO.

Below the picture illustrates "my mother's little helper" - universal XO board. I used a piece of particleboard with clip-spring terminals which can form any 1st, 2nd or 3rd order XO, with zobels, filters, resistors, polarity swap and so on. I can compare results in seconds just by changing the parts. And I can also compare parts with same value but different brand or technology.
If we use a good test signal and we have big component drawer like I do, we can create a perfect XO in less than an hour. Like on the picture - after 2 years of pain - I finally found a great XO for the Raven2 ribbon. Without the "board" it wouldn't be possible at all!!

To do the mid to tweeter XO we act similarily.

First - we measure the midrange in its top region - 2 to 10 kHz.

Normally - it has a rising response - a peak near 5-8 K before the sharp rolloff at 10KHz.  Now if we add a small series coil like 0,2 - to 0,5 mH - it will smoothen out the peak and roll off nicely at the top. Just what we need - eliminate the peak AND the uppermost octave. Two birds in one shot and by one component only.

This will result with a rolloff at around 6 kHz or something similar.  We are perfectly happy because the top octave from the midbass sounds very dirty anyway.

This is a good moment to mention the capacitor quality dilemma.

Is the cap quality important in this circuit? 100 uF of foil is VERY COSTLY in premium brand.
Well - forget good caps. The capacitor "quality" in audiophile terms means low inductance and low resistance and low leakage current. And low tangent of current to voltage phase shift.  So in this case, the cap will be in series with a coil which has 10 000 times bigger inductance than the cap. And resistance too. So who cares if the capacitor is a Hovland, Jensen or no name. I tried them all and they are inaudible for midrange if they are working in series with the coul or even resistor.

Now the last part is the tweeter - first we add a cap in series to protect it. Start with 3 uF MKP and work from here. Make the measurements "meet" and match the midbass. If 3 uF is too big, - reduce it to 2 or 1,5. If 1,5 is too big - go for second order with a paralell coil after the cap. 0,5 mH air core with 1 mm wire is ok.
In the parallel tweeter application the coil quality and the capacitor quality are both very important. The capacitor works ONLY with the tweeter without series parts. And the coil directly affect the treble signal being parallel to the tweeter. So here I say YES, Go and spend but not too much.


OHM's law is very handy in making cross-over circuits.
The simple formulas for XO part choice is that caps in series increase impedance with frequency the smaller they get. The rule of thumb is that  a 4,7 uF equals 8 Ohm  at 3 kHz. At 6 kHz it will double, at 1,5 kHz it will halve. And from Ohm's law we can see that at 3kHz the cap equals the driver impedance (8 Ohms)  So the cap AT THIS FREQUENCY is simply an 8 Ohms resistor. Then it is easier to understand it's role.  It will absorb half the energy. And so on. Simple, isn't it ? Coils are the opposite - series impedance will increase with value. X=1/jwL remember ?

At the end we measure tweeter plus midbass in the XO region from 2 to 10 kHz, then we change tweeter polarity and choose the higher curve as the correct one.

Important - we place the speaker in its future position, mike in the listening position at ear's height. We DO NOT MEASURE CLOSE TO THE DRIVER'S CONE OR AT 1m DISTANCE. That produces "shop" speakers that sound like shit.

At the end we add bi- or tri-wiring terminals, wire everything with a decent wire (not necessarily all runs with the same one - use solid core for bass and stranded individually insulated for tweeter, any good one for midrange). Same wire theory in bi-wiring, in triwiring or inside speakers is a stupid myth. Please forget it.

Place the speakers symmetrically in the room at a perfect unilateral triangle from listening chair, move speaker's base line 1 m from back wall, toe in 10-15 degrees, sit back and enjoy.

This approach produces the best sound achievable from any set of 3 drivers in 100 % of cases out of all I ever tried. And no maths is required because ALL elements are already built in the measuring process. The driver spacing, the baffle shift, the wall refflections, the carpet, the sofa, the cones, the domes, the interferences, the refflections, diffractions, and the wire influence. The part quality, the flowers in the corner, the MDF vibrations and early floor reflections. ITS ALL IN THE RESULT. Just imagine a mathematical or PC software model which envelopes ALL THESE FACTORS.  DO YOU GET IT !?!?!


Please note that I do not mention resistors. I believe in resistorless crossovers. I try to match the drivers so I don't need resistors which consume the signal energy and reduce the pleasure. Resistors are an evil necessity like a condom. If the drivers sensitivity does not match significantly,  sometimes I'll use a series resistor to the midrange or a tweeter, never to the bass. The value should not be higher than single ohms. Four- I would say - is the maximum value.
If I only can, I "build-in" the resistance to the coils. I just choose thinner wire and it increases DCR just rightly. And the coil is cheaper and smaller. What a stupidity is to put a resistor in series and then buy EXPENCEIVE thick wire coil with very macho ribbon or something. This is like driving with the handbrake applied.

BACK


Links:
Saba driver catalog with measurements      http://home.arcor.de/pfaue/klangkue/gruenwunder/gruenwunder.htm

4 years later:

Time flies, it is 2013 now. It 's been 4 years since I devoted myself to the audio business, no longer a hobby. Looking back - here is what I think about the article written above:

LampizatOr shop with pars and kits for DIY projects

Lampizator finished products (not for DIY)

Amplifiers page

DAC pages

Music server and transport page

SILK AC filter