019 - Network Synthesis, Vol - XIX (1957) (Prof. E. A. Guillemin, Prof. P. M. Lewis II, Dr. M. V. Cerrillo)

XIX.
NETWORK SYNTHESIS
Prof. E. A. Guillemin Dr. M. V. Cerrillo J. C. Pinson

Prof. P. M. Lewis II E. F. Bolinder T. G. Stockham
A. SYNTHESIS OF SAMPLED SIGNAL NETWORKS
A sampled signal network is defined as a network that contains a finite number of

linear, passive circuit elements and which, when excited by a signal consisting of a
series of impulses occurring at fixed, equally spaced time intervals, gives a response
that consists only of a series of impulses occurring at these same time intervals. Such
networks might find use in sampled-data systems or in radar systems.
A general synthesis procedure for such networks, which includes the basic realiza-
bility conditions and general realization methods, was developed. These networks are
composed of resistors and open- and short-circuited transmission lines all of which
have the same time delay. Synthesis with such transmission lines has also been studied
by Richards (1).
The approximation problem for these networks was also studied. A general approxi-
mation procedure was developed, as well as an approximate lowpass-to-sampled-signal
transformation. A paper is being prepared for publication which will give the main
results of this research. These results will be summarized here.
1. Realizability Conditions
It is not difficult to show that a system function of this type must be a rational func-
tion of e-Ts
2
a +a e -TS +a2e - TS
.. .,ae
-nTs
H(s) =o 1 2 n
b O + be -TS ++b.e
b- TS
,b e -nTS
The realizability conditions become apparent immediately when the behavior of this
function is considered in the complex s-plane. These conditions are exactly the same
as those for lumped networks; this becomes clear when we remember that the condi-
tions were derived from the standpoint of energy considerations and do not depend on
the network being lumped. These conditions are well known and concern the analyticity
of the function in the right-half plane and certain constraints on the j-axis real part.
In the case of rational functions of e -TS , these conditions imply the additional constraint
that (a) bo 0, and (b) a 0 * 0 for driving-point functions. These conditions can be
immediately transformed to the z-plane, in which z = eTS
It is convenient to introduce another transformation
--
TS 1 - wl - e
e 1 w w = = tanh Ts/2
l+e
156
(XIX. NETWORK SYNTHESIS)
(Note that this is the negative of the usual w-transformation.) This transformation has
the property that it transforms the j-axis of the s-plane into the j-axis of the w-plane
with the left- and right-half planes transforming into the left- and right-half planes,
respectively. Thus the system function, when it is expressed in terms of w, represents
a realizable lumped network, and all of the well-known realizability tests for these net-
works can be used.
2. Realization Methods
The system function, expressed in terms of w, represents a realizable lumped

network and can be realized by using the ordinary methods of lumped-network synthesis.
Then we have only to perform some operation on this network so as to "undo" the change
of variable. This operation is easily performed by using the transformation in the form
w = tanh Ts/2.
The function tanh Ts/2 represents the input impedance of an ideal shorted trans-
mission line with a characteristic impedance of one ohm and a one-way time delay
of T/2 sec (or the input admittance of an open-circuited line). Therefore, the
inverse transformation can be accomplished by substituting a short-circuited line with
Z = L ohms for each inductor of value L henrys, and an open-circuited line with
Y o = C mhos for each capacitor - all of these lines having a one-way time delay of T/Z
seconds. This completes the synthesis.
3. The Approximation Problem
Two approximation methods will be discussed; one is an adaptation of a well-known

method (2), and the other is an approximate lowpass-to-sampled-signal transformation.
a. Guillemin's method
This method can be used to approximate any arbitrarily given magnitude character-
istic plotted against jw. The given magnitude function will be an even function of w and
will be periodic, with a period wo = 2r/T. It can therefore be approximated by the first
n terms of a Fourier series of cosines
n
F(w) 2 = ak cos k Tw
0
The Tchebycheff multiple-angle formulas can then be used to transform this expression
into a polynomial in cos k Tw:
F(w) 2 Zbk cos k TW
157
Finally, making use of the fact that the transformation
-Ts +w
e -w
implies the existence of the relation
1 - 0
COS TO
1 +w
where s = o + jw and w = F + jo, and we can write, for the magnitude function,
This sum can be combined into a rational function of Z2 which is the approximation to
the required function. Then standard methods can be used to find the realizable function
of w that has this magnitude.
b. Lowpass -to-sampled-signal transformation
Another method would be to construct first, by any of the standard methods, a

lumped lowpass network, which approximates the required characteristic and is all
"contained" in the interval -Tr/T < Z < Tr/T, and then to make a suitable change of variable
similar to the lowpass-to-bandpass transformation which will transform the network into
an equivalent sampled-signal filter. Such a transformation is given by M ---- 2/T tan T/2.
The factor 2/T is included so that the two variables will have the same slope at the
origin. The transformation can be interpreted physically as substituting a short-
circuited ideal transmission line with characteristic impedance 2/T L ohms and one-way
time delay of T/2 sec for each inductor of value L henrys in the lowpass network; and
an open-circuited line with characteristic admittance of 2/T C mhos for each capacitor
of value C farads.
This is not an exact transformation (neither is the usual lowpass-to-bandpass trans-
formation), but it is quite good for frequency values near the origin. Moreover, since
the transformation affects only the abscissa (w-axis) and not the ordinate, the absolute
tolerances on the error are not affected; for example, an equiripple filter would still
be equiripple.
As far as time response is concerned, it can be shown (by computing derivatives at
the origin) that, for any sample time T, the average value and the first two moments of
the impulse response of the lumped and sampled filters are identical, and if these two
moments should be zero, then the first two nonzero moments are equal. In general, the
smaller the sample time, the closer the higher-order moments agree.
Because the average values of the time response of the lumped and sampled networks
158
are equal, the envelope of the impulses produced by the sampled network will be a con-
stant multiplied by the time response of the lumped network - the constant depending
upon the sample time. If it is desirable to have the envelopes of the lumped and sampled
signals equal, then an impedance-level change can be made on one or the other of the
networks.
P. M. Lewis II
References
1. P. I. Richards, Resistor transmission line circuits, Proc. IRE 36, 217-220 (1948).
2. E. A. Guillemin, Synthesis of RC networks, J. Math. Phys. 28, 22-42 (1949).
159

019 - Network Synthesis, Vol - XIX (1957) (Prof. E. A. Guillemin, Prof. P. M. Lewis II, Dr. M. V. Cerrillo)

Uploaded by

Copyright:

Available Formats

019 - Network Synthesis, Vol - XIX (1957) (Prof. E. A. Guillemin, Prof. P. M. Lewis II, Dr. M. V. Cerrillo)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

019 - Network Synthesis, Vol - XIX (1957) (Prof. E. A. Guillemin, Prof. P. M. Lewis II, Dr. M. V. Cerrillo)

Uploaded by

Copyright:

Available Formats

XIX.

Prof. E. A. Guillemin Dr. M. V. Cerrillo J. C. Pinson

A. SYNTHESIS OF SAMPLED SIGNAL NETWORKS

A sampled signal network is defined as a network that contains a finite number of

The system function, expressed in terms of w, represents a realizable lumped

seconds. This completes the synthesis.

3. The Approximation Problem

Two approximation methods will be discussed; one is an adaptation of a well-known

F(w) 2 Zbk cos k TW

Finally, making use of the fact that the transformation

implies the existence of the relation

b. Lowpass -to-sampled-signal transformation

Another method would be to construct first, by any of the standard methods, a

You might also like