View Full Version : A novel cross-over design method (by Oddesit)

11-09-2005, 01:31 AM
Hi, all,

I have a friend (his nick is Odessit) who is very good in speaker design. His method is based on two principles. This works well especially for tube amplifiers. The point is that typical modern networks cannot have the constant impedance. Consider tube amplifier having 2-3 ohms at the output. Than even simple 1st order network will have changes in impedance say from 5 to 8 ohms or even more, which will change the tonal balance on more than 3 dB. Besides the tonal balance the correct phase of the signal will be lost due to interaction of amplifier with complex impedance.

The method of Odessit is based on two principles:

1. First, he connects a pair of similar drivers for future stereo speakers consequently by wires and connect them to one output of the amp. The input of amplifier is connected to sine generator.

Now, one changes the frequency litlle bit up and down, arround the testing frequency, which makes a "wo-oo-u" sound. The point is to listen and observe the sound image between two drivers. The ideal image must concentrate in one point. The idea is that the human ear cannot tell much about constant frequency signal, but is very sencitive to frequency changes. If the image jumps left or right or in other directions at some frequency this pair is considered as having potential problem at given frequency and might be not good to work together. If one follows the method, he finds that for instance the woofer works well up to some frequency, then he observes the frequency points where the image is detoriated. It is up to You - to accept these frequencies and go further up in the range, or to stop. Increasing the frequency several "potential problem" can be met, and at last You fing the frequency where You decide that You don't want to go anymore up. Finally, one finds the working range of the woofer (and same for the midrange). Now the cross-point is obtained as the geometrial mean value of two boundaries for the woofer and midrange.

The important is that the cross-over point is obtained not only from the driver specs (this is of course still useful), but from the subjective test for a given pair of the drivers.

My friend starting designing new speakers usually finds many similar drivers, and makes the cross-matrix puting in table the test results. Finally, only two units are selected which hold stereo image in the maximum range. :applaud:

2. Point two is how to design the cross-over. He only uses 1st order filters which don't detoriate the phase, and compensate impedances of each driver. Only after compensation - this is important- the woofer is added with the coil, and the midrange with capacitor. The accuracy of the filter tuning is about 1 percent or so, same for active resistances (after compensation) of each drivers. For instance, if LE123-A has 5.5 ohm coil, that means that all drivers must be adjusted to the same 5.5 ohm with serial and parallel resistors (like L-pad). That means all drivers have only real impedances (no complex values), and only after that are crossed. This finally means that entire speaker has constant and only real impedance, with no imaginary part reacting with the amplifier. The tuning must be very accurate.

I am now doing alsmost the same (skiping p.1 with the sound image but I want to try it) for JBL LE123 and LE5-2 and I am very much impressed by the first results.


11-09-2005, 03:04 AM
In particular, what I have now for L100 is:

LE123-A1 - serial coil 0.92 mH, Zobel network 20 uF + 6.8 Ohm resistor in parallel to LE123-A1, driver wired in phase,

LE5-2 - Capacitor 36.6 uf (33 uF Audin MQP Cap + 3.3 uF Fostex CS cooper-tin foil cap), and resistive attenuator to adust level and impedance arround 5 ohm, driver also in phase.

LE5-2 is not yet compensated but I will try too.

This is arround 900 Hz cross-over point. The measument showed that 123-A1 has break-dance mode around 2000 Hz, while gowing lower is not also desireable: the resonance frequency of LE5-2 is abourt 280 Hz. I'd like also to try Odessit's method with "sound image".

The overall impedance is constant arround 5 ohm from 100 Hz and higher. I am using SE 2a3 amp.

No tweeter yet as at first I'd like to work out the cross-over for these two units.

11-09-2005, 07:31 AM
BUT, "JBL professional frequency dividing networks are intended for use with high and low frequency driver combinations The networks use 12 dB per octave parallel L-C circuits with additional conjugate elements to cancel the inductive reactance of the low frequency loudspeaker. Highest quality electronic components are used throughout-non-inductive, non-polarized capacitors having high AC current capacity built expressly for use in dividing networks, individually calibrated low-loss inductors, and oversize switches and resistors High frequency shelving of networks crossing over is accomplished with tapped autotransformers rather than conventional pads." J/K, If the "Oddiset" sits well with your ears, you're there, that is if you can stop experimenting. To me, it's like finding the best "What-ever", and knowing that indeed, it is the best it can be. Within the world of capacitors alone, is many thoughts/opinions. Throw in conductor selection to add to the "Mix". Let us know if/when you find the "Right" sound. :)

11-10-2005, 03:04 AM
That's great about JBL, but i.e. 4311 are considered as professional studio monitors, and have simple 6 db/oct X-over. How about their Z-curve?

It is an interesting question - how old JBL speaker models, including professional ones, hold constant impedance? Was it still the tube era at that time, or already transistor?

I am sure that many models even with L-C 12 db/oct also have variable impedance. BTW, can anyone post here Z-curve of L100 century earlier model (which uses 2nd order X-over at 2500 and 7500 Hz)? I dont have that network, but it will be very interesting for me, as I use tube amp. Thanks.


11-10-2005, 12:32 PM
hmmm, I think a lot has been said about this [some of it by me] in other threads about 6dB crossovers

the best thing is to start with drivers that have little impedance rise for higher frequencies anyway in the first place and use simple crossovers

I don't like this exact tuning of compensation networks. Drivers often change due to ageing and running in and that can affect the compensated frequency so that the cure can become worse than the disease

you would have to periodically review zobel networks working at resonance frequencies for instance, I think


11-10-2005, 11:47 PM
I believe that foam and cone material can change with time, as well as magnet can loose its strength. But what will happen to

1) Ohmiс resistance of the wire? The metal will vapour?
2) Impedance of the coil, determined by the number of wire turns? This number cannot change :blink:


11-11-2005, 11:22 AM
mostly the spider will age, the surround I think is less of influence

my kevlar bas/mids sounded very bad for the first year I had them, I even thought of selling them

now, the resin has set, the spider has set, I don't know, but they sound totally different then when new

all suspensions age, also the aluminium and titanium diaphragms of compression drivers, so the resonance frequencies will shift lightly with time

of course the metal will not vaporize and coil windings will not unwind...

11-11-2005, 12:35 PM
i belive 300W + a dick of a loud singer will vaporise wire in a k-120 but only the first 1/4 inch

11-11-2005, 03:47 PM
Consider this:

SPeakers make noise and that noise should be as constant throughout the range in terms of amplitude and dispersion. Harmonic resonances are a byproduct , however and can contribute in both positive and deterimental ways to the final sound, not all harmonics are bad.

Drivers disperse sound well up to a certain point determined by cone or dome diameter, shape, depth and in some cases material.

Crossover networks tailor the transistion so that the two drivers do not either cancel each other out ( causing a null in response) or augment too much (causing a sharp peak) while taking into consideration the distance between them for imaging purposes. Crossovers also tailor responses so that the output is relatively smooth, a difficult task to do by driver design alone.
Matching the different efficiencies of LF, MF, and HF drivers is done through the network as well.

Last thing to consider is the impedance of the system. It takes a back seat to the other conditions as long as the impedance does not fall too low. Only electrostatic magnetic drivers exhibit flat impedance over their entire range, due to the driver having low moving mass and a uniform motor system. Very flat, accurate system have had wild impedance ranges, the tannoy monitors are an example and they use 6 db/oct crossovers in their 12 inch coaxial flagship DM-12 ( I think that is the model, not sure).

I prefer to use sound level meters to check the output, not woo-uu-oh sounding sweep tones. Your friends speakers may sound quite good and your obviously happy with the method, but it goes against the grain of what companies like JBL have done for 50 years to a great deal of commercial success