INTRODUCTION: The design of a frequency-dividing network to feed the individual drivers in a
loudspeaker system is a difficult task if optimum results are desired. In the case of a passive
dividing network between a single amplifier and several drivers, problems arise because the
impedance of these drivers is frequency dependent. Most likely this impedance is not purely real
over the frequency range where the transition in acoustic output from one driver to the-other has
to be made. Thus simple filter theory cannot be used. Furthermore the drivers are likely to have
different efficiencies so that attenuation has to be incorporated into the dividing network. This
increases the source impedance which the driver sees and reduces the damping of its mechanical
resonance.
Because maximum damping is needed for the woofer to control its low-frequency behavior, no
attenuation can be allowed in the low-pass filter which drives the woofer. This then means that
the tweeter has to have efficiency equal to or higher than the woofer and restricts the choice of
units that could be combined with a particular woofer or necessitares the use of matching
transformers.
Finally there is the problem of designing the dividing network so that it presents a tolerable
impedance to the power amplifier driving it. The active crossover network gets around all of these
problems and offers some additional benefits. Each driver is driven directly from its own power
amplifier whose gain may be adjusted to equalize differences in efficiencies between the drivers.
This gives complete freedom in selecting drivers.
The output impedances of the power amplifiers directly control the damping of the drivers
without intervening crossover networks, and each can be optimized for its particular driver if
necessary.
The load presented to each amplifier is less frequency dependent. System distortion is reduced
because the woofer channel may be driven into clipping without also affecting the midrange or
tweeter channel.
Because the signal is divided into the separate frequency ranges at the input to the power
amplifier where one deals with low-level signals and a well-defined interface between the
electronics, it now becomes possible to choose from a wide variety of dividing networks which can
be convenient[y realized with operational amplifiers.
The difficulty now becomes to decide which network is optimum. This paper is an attempt to
classify the different networks which have been used, to point out some of their properties which
have not been considered before, and to propose a unique class of filters which appears to be the
best engineering compromise. This compromise is based upon a consideration of the radiation
patterns produced by different networks operating with noncoincident drivers as well as the
transient responses of those networks.
THE PROBLEM
The design problem is illustrated in Fig. 1. Two sound sources H and L contribute to the sound
pressure at point Pi in space. This point is "on axis" of the cabinet C-C, but at unequal distances
from the drivers because the effective plane of radiation for driver L is offset by the distance d2
from that of driver H.
The objective is to apply such an input signal to H and L that the sound pressure at P1 is
independent of frequency.
To simplify the analysis, the sound pressure at point P0 will be considered first. This point has
equal distance from H and L and is assumed to be located on the axis of both drivers.
The signal arriving at this point has the transfer function:
Fo = Fu + FL
FH is the transfer function of the driver H and its associated high-pass filter, while FL is the
corresponding low-pass function. The driver axis can be made to coincide with the cabinet axis
either by physically mounting driver H by a distance d2 behind the plane C-C or by electrically
delay-ing the signal which is applied to H by the time which it takes a signal to propagate the
distance d2 in air. A simple circuit for producing such electrical delay will be show later.
The problem is thus reduced to that of Fig. 2. Of interest is the sound pressure not only on axis but
also at angles a off axis. This is modified by the separation of the drivers d1 which causes different
path lengths to point P and thus changes the phase with which the sound pressures from H and L
combine at P. It is also influenced by the relative phase of the electrical signals driving H and L.
First the sound pressure for the on-axis point Po will be considered. The combined acoustic output
should be as frequency independent as possible. It will be described by the transfer function Fo.
Later on the effect of the transfer function Fo upon the radiation pattern will be investigated.
THREE FILTER TYPES
The transfer function
Fo = Fu + FL
will meet the requirement of frequency independence to varying degrees, depending on the
crossover network design. Fo can be classified to belong to one of the following three types of
functions:
