Dividing network is another name for
a crossover network used to separate and direct parts of the audio frequency band to specific
speakers  bass, midrange, and tweeter, for example. It is the audio equivalent of a diplexer
or multiplexer filter in the RF and microwave realm. As with many other aspects of consumer
electronics, there was a time when many audiophiles chose to design and/or build their own
crossover networks for homebuilt speakers. Although never really a bona fide audiophile,
I did design and build my own custom speaker cabinets and the crossover network. The woodwork
was done at the
woodshop at Robins Air Force Base while I was serving
in the USAF as an Air Traffic Control Radar Repairman
(not an air traffic controller). I don't know if is still the case, but back then most large
bases had hobby and craft establishments for woodworking and metalworking, ceramics, artistic
painting, and even electronics benches. To the left is the only photo I have of one of the
speakers, which have since been sold (my son, Philip, is retrieving
his Easter basket which was hidden behind it  circa 1988  while attending
U. of Vermont and living in a trailer
park). My units each had three speakers and a homemade LC crossover network. Admittedly,
no real science went into the electrical design; I adjusted the inductor and capacitor values
until everything sounded 'about right.' All the components, including the individual speakers
and speaker cloth, were purchased at the downtown Radio Shack.
Dividing Networks
By Harold Renne
Technical Editor. Radio & Television News
This is an interior view of an RCA type of dividing network for theater
use. The crossover frequency is 400 c.p.s.
A twospeaker system using a tweeter and a woofer has many advantages over a single speaker
system.
Because of the difficulty in manufacturing a speaker with a single cone assembly which
will satisfactorily reproduce both the extreme low and extreme high audio frequencies, it
has become rather common practice to use a dual speaker system for highquality installations.
In such installations, a lowfrequency speaker, or woofer, is used which is designed primarily
to satisfactorily reproduce low frequencies, and a high frequency speaker, or tweeter, is
used which is designed for the high frequencies.
It is desirable in systems of this nature to "sort out" the low frequencies from the high
frequencies and apply each to the proper speaker. A network for performing this function is
called a dividing network, and the point at which the frequency division takes place is called
the crossover frequency. This is the frequency at which the two speakers receive equal amounts
of energy.
Fig. 1  Schematic of a simple dividing network which will operate quite
satisfactorily.
Experience has indicated that a dividing network should have an attenuation beyond the
crossover frequency of from 6 to 12 db. per octave. This may be accomplished with fairly simple
networks. Two types of circuits may be used: the filter network and the constantresistance
network. Both have advantages and disadvantages, and both will be discussed.
Fig. 2 shows parallel and series filter dividing networks, made up of halfsection elements
of the socalled mderived type. Full section elements could be used, giving an attenuation
of 18 db. per octave beyond the crossover frequency, but the slight improvement in operation
is not worth the additional cost and losses involved. With extreme care in design, a dividing
network may introduce as little as 0.5 db. power loss, but even this can become appreciable
when high powers are involved. For example, at a power level of 100 watts, a 0.5 db. loss
represents a power loss of about 11 watts.
The equations in Fig. 2 indicate how the condenser and inductance values may be determined.
The value of R_{o} indicates the impedance of each of the speakers and the input impedance
to the network. The crossover frequency f_{c} is determined by the speakers
used and should be as low as possible. Each speaker should be able to contribute appreciably
to the sound output at least onehalf octave beyond the crossover frequency.
The filtertype dividing network is somewhat more versatile than the constantresistance
type and has slightly better transmission characteristics in both the transmission and attenuation
bands. It does not lend itself readily to massproduction techniques, however, as two different
values of inductance and capacity are required.
Some typical commercial dividing networks. (A) RCA network with 400 cycle
crossover for theater use. (B) Brociner Electronics constantresistance network with 500 cycle
crossover and an attenuation of 12 db. per octave. (C) Circuit used by Stephens Mfg. Co. to
give an attenuation of 12 db. per octave and a crossover of 600 cycles. (D) University Loudspeakers.
Inc., network with 600 cycle crossover frequency.
Both the parallel and series type of networks are effective, but listening tests seem to
favor the series type (Fig. 2B).
It might be instructive to calculate a typical dividing network on the basis of the circuit
and equations given in Fig. 2. For the reasons given before, we will choose the series type
for our calculations. A typical value for the speaker and input impedances would be 8 ohms,
and the crossover frequency will be chosen as 800 cycles. Substituting these values in the
proper equations gives the following constants for the network:
L_{2} = 1.6 mhy.
C_{3} = 40 μfd.
L_{3} = 1.0 mhy.
C_{1} = 25 μfd.
The inductances should be wound with fairly heavy wire on a nonmagnetic coil form, such
as wood. An inductance bridge is very helpful in obtaining the correct values. The coils should
be mounted with their axes perpendicular to avoid mutual coupling. The condensers must not
be of the electrolytic type, but they may be paper or oilfilled. Some condensers available
on the surplus market would be suitable. Observers report that the calculated values of inductance
and capacity may be varied as much as 25% without any appreciable effect on reproduction as
judged by listening tests.
The constantresistance type of dividing network is shown in Fig. 3. It will be noted that
for a given network, the values of the two inductances and the two condensers are the same,
making this unit easier and cheaper to build on a production basis. When properly designed,
this network is equally as effective as the filter type. It has the theoretical advantage
of presenting a constant load to the source at all frequencies, but the wide variation in
impedance of the voice coils with frequency tends to defeat this advantage.
Either of the networks shown will give an attenuation of about 12 db. per octave beyond
the crossover frequency and will introduce a power loss of between 0.5 and 1 db. An attenuation
of 18 db. per octave may be obtained by the use of π or T sections instead of L sections,
but such attenuation is not essential, and, as with the filter type, the additional power
loss resulting from introducing the additional components more than outweighs any advantage
that might be obtained.
Fig. 2  (A) Parallel and (B) series filter dividing networks made up of
halfsection elements.
As an example of a typical seriestype constant resistance dividing network (Fig. 3B) let
us take the same conditions as before, namely, R_{o} = 8 ohms, f_{c}
= 800 cycles. This gives values for L_{2} of 1.1. mhy. and C_{2} of 31 μfd.
A much simpler dividing network than those described above is frequently used. This network
is shown in Fig. 1 and consists simply of a 2 μfd. condenser in series with the voice coil
of the tweeter. The inductance of the woofer voice coil is appreciable, and its impedance
rises with frequency. The inductance of the tweeter voice coil is relatively small, so the
impedance of the condenservoicecoil series combination decreases as the frequency increases.
These two effects tend to cancel each other, giving a fairly constant impedance.
In the above discussion, we have ignored the fact that the tweeter speaker is, in general,
more efficient than the woofer. This would tend to give an unbalanced output with excessive
high frequencies. For this reason, an attenuator is usually placed in the tweeter circuit
to compensate for this increased efficiency.
We have assumed in our dividing network calculations that the voice coil impedance of the
tweeter and of the woofer were' the same. If such is not the case, the problem is considerably
more complicated. If the tweeter voice coil impedance is higher, it may be shunted by a resistor,
within limits, to bring the impedance down to the correct value. For example, if the woofer
has an 8 ohm impedance and the tweeter is rated at 16 ohms, a 16 ohm resistor may be connected
in parallel with the tweeter to bring the total impedance down to 8 ohms. Half of the highfrequency
power is lost in this resistor, but the higher tweeter efficiency may make up for this loss.
Fig. 3  (A) Parallel and (B) series constant resistance dividing networks
made up of L sections.
The above discussion has assumed that the dividing network is placed between the output
transformer and the speakers. It is entirely possible to place this network in the center
of the amplifier and then use separate power amplifier stages for the low and high frequencies.
The network may be somewhat simplified since the value of R_{o} may be greatly increased
and since power loss in the network is no longer a vital consideration. These advantages may
be overcome by the additional cost and complexity of the two separate power amplifier channels,
but this system has been used with marked success in a commercial amplifier manufactured in
England.
Another possibility is to place the dividing network in the plate circuit of the output
stage, and then to use separate matching transformers for the high and low frequency speakers.
This permits matching the network to any speaker impedance, and because the network is in
a high impedance portion of the circuit, more convenient values of inductance and capacity
are possible. At least one company using this latter system reports highly satisfactory results.
With good woofer and tweeter speakers, proper enclosures, and a suitable dividing network,
frequency response from 50 to 15,000 cycles can be readily achieved.
Posted January 11, 2017
