$7

Vor. UMs 18, NUMssR 4 OcvoasR, 1946

:V. :icmwave .& ..ec1:renics

J. C. Sr.AvER
3Iassachusetts Institute of TecItnology, Cambridge, Massachusetts

Y microwave electronics one can refer to the highly perfected. The magnetron oscillator was
study of electromagnetic fields in regions of improved, first in England and then in this
the order of a wave-length in dimensions, bounded country, to the point where it was. a generator of
by reflecting walls, and of the interaction of these microwave power of very great capabilities. All
fields with electrons, ions, or other forms of space of these developments, particularly as they were
charge. This includes the whole high frequency carried on at the M. I. T. Radiation Laboratory,
side of microwave radar; the' nature of wave will be discussed at length in the great series of
guides and resonant cavities; and the nature of books to be issued from that laboratory, and
electronic tubes, such as klystrons and magne- published by the McGraw-Hill Book Company,
trons, in which transit time is of fundamental Inc. Other version~ of the same information have
importance. It also includes such devices as been, or will be, contained in publications from
cyclotrons, synchrotrons, linear accelerators, and the industrial laboratories in various periodicals,
other devices for the acceleration of charged such as the Bell System Technical Journal. Even
particles. This article presents some aspects of these publications, extensive as they will be,
this field, but is far from a complete treatment. represent but a small fraction of the great litera-
It represents essentially a set of lecture notes for ture which exists in the form of classified, or
a series of seminars delivered by the writer during formerly classified, reports prepared during the
the winter. of 1945— 46. It is hoped that the war. With this large program of projected publi-
material eventually will be expanded greatly, cation, it is worth asking why the present review
into a full-sized book, to be published by D. Van article, and its future enlargement into a book,
Nostrand Company, in the series of publications are necessary.
from the Bell Telephone Laboratories. Since, The answer would be that the author has
however, this publication will not appear for tried to introduce into the field a correlation and
some time, it was felt worth while presenting this unity which are perhaps lacking in most of the
abridged and incomplete version, on account of other work. During the war his work was largely
the great present lack of material on the subject. on magnetrons, both at the M. I. T. Radiation
During the war there was, of course, a very Laboratory and at the Bell Telephone Labora-
great development in the knowledge of micro- tories, of which he was for a time a staff member.
waves. The study of wave guides and resonant It became clear in the early stages of this work
cavities, originated before the war by Barrow, that a study merely of the electronics of the
Chu, Schelkunoff, and others, was carried to a magnetron was incomplete and unsatisfactory;
point of great advancement. The klystron, de- it was necessary in addition to take into account
veloped before the war by Hansen, the Varians, the resonant circuit, consisting of resonant cavi-
and others, became much better known and ties and the attached loads, and to consider the
41
 J ~ C. SLATE R
reaction of this circuit back on the electronic and others, though its formulation in terms of
motions. This in turn led to the development of the general theory of resonant cavities, and a
a circuit theory of resonant cavities, and of the number of the specific applications as well,
wave guides which form the leads of these cavi- originated with the author. The fifth chapter,
ties, based on the theory of orthogonal functions, application to the electronics of the reflex klys-
and op the expansion of Maxwell's equations in tron and the magnetron, likewise represents the
a closed region in terms of such orthogonal ideas of many, both in this country and in
functions. This development gives a logically England, though the author was associated with
satisfying foundation for the whole of microwave all parts of the subject, and wrote extensive
electronics, and at the same time proves to be of reports, much fuller than is suggested here, on
great practical value i.n the design and develop- both the electronic and circuit phases.
ment of magnetrons. Later application to reflex It seems worth while pointing out that the
klystrons has shown that the principles are of theory of expansion of solutions of Maxwell's
wide importance. Work since the war has con- equations in orthogonal functions, which is pre-
vinced the writer that in such problems as the sented here, can have useful applications in the
linear accelerator the methods are of just as theory of quantum electrodynamics, as well as
much value. The main purpose of this review in microwaves. The scalar and vector potentials,
article is to present this unified point of view, . as well as the fields, can be expanded in these
carrying the application to problems such as the orthogonal functions, and the Lagrangian and
klystron and the magneton only far enough to Hamiltonian of the field can be set up in terms
illustrate the general method. The later amplifi- of these expansion coefficients. By this means we
cation in book form will carry these applications can set up a classical relativistic Hamiltonian
much farther, and to a wider variety of problems. theory of the interaction of fields and matter in
The work presented in the present article, of an arbitrary cavity, which can then be carried
course, represents contributions from a variety over into quantum theory in a manner similar to
of workers, and no attempt is made to assign that of Dirac's radiation theory. This frees that
credit for it. The second chapter, on wave guides, theory from the requirement that the field be
is to some extent familiar. The orthogonal func- expanded in plane waves, and provides a much
tion development was worked out, not only by more general form of expansion. While this does
the author, but by Bethe in the Radiation Labo- not seem to remove any of the outstanding
ratory, and presumably by others. The material difficulties in quantum electrodynamics, it yields
of the third chapter, on resonant cavities, was a new point of view which may be useful. The
suggested to the author by a treatment given by author hopes to develop this application in a
Condon in 1941, but in a, much more incomplete later paper.
form. At about the same time that the writer Several books have appeared in the last few
was working it out in the Radiation Laboratory, years, treating the pre-war status of the subject
Schwinger was also carrying out very similar satisfactorily. Among these we list the following:
expansions in orthogonal functions, and using
J. G. Brainerd, G. Koehler, H. J. Reich, and L. F.
them for similar . purposes, though the two Woodruff, Ultra-High Frequency Techniques (D. Van
developments were largely independent. Fesh- Nostrand Company, New York, 1942).
bach, of the Massachusetts Institute of Tech- S. Ramo and J. R. Whinnery, Fields and Waves in
nology, was also working along similar lines. Modern Radio (John Wiley and Sons, Inc. , New York,
Much of the material of the fourth chapter, on 1944).
R. I. Sarbacher and. W. A. Edson, Hyper and Ultrahigh
the applications of the theory of resonant cavi- Frequency Engineering (John Wiley and Sons, Inc. , New
ties, was common knowledge at the Radiation York, 1943).
Laboratory, the result of work of Lawson, Rieke, S. A. Schelkunoff, Electrornagneti c Waves (D. Van
,

NostrandCompany, New York, 1943).

.

J. C. Slater,
Microwave Transmission (McGraw-Hill
Book Company, Inc. , New York, 1942).
(V, J. A. Stratton, Electromagnetic Theory (McGpaw-Hill
Fro. 1. A four-terminal network. Book Company, Inc. , New York, 1941.).
 M IC ROWA VE ELECTRON I CS

0-
J" G—

n+)
FrG. 2. A transmission line.

The book of Sarbacher and Edson contains a them. Nevertheless, the results are so similar to
very complete and excellent bibliography of the those of ordinary circuits that a knowledge of the
pre-war literature of the whole field. Since it is fundamentals of circuit theory is an excellent
so complete, it seems unnecessary to give refer- background for understanding microwave cir-
ences to that literature here. Various review and cuits. For that reason we start by discussing a
other articles have appeared since the books number of theorems in circuit theory. First we
mentioned above. Among these may be men- consider a fundamental starting point, the theory
tioned the following: of the four-terminal network. This is a collection
E. U. Condon, "Principles of Microwave Radio, " Rev. of circuit elements whose nature we do not have
Mod. Phys. I4, 341 (1942). This article presents a point to inquire into, except that it is provided with
of view similar to that of the present one in the matter of two input and two output terminals. Let V~, V2
the oscillations of cavity resonators, but does not carry be the voltages across the two sets of terminals,
the application to self-excited oscillators as far as we do
in the present article. and z~, z2 the currents Howing, the currents being
J. B. Fisk, H. D. Hagstrnm, and P. L. Hartman, "The positive when flowing in the direction of the
Magnetron as a Generator of Centimeter Waves, " Bell arrows in Fig. 1, and the voltages positive when
Sys. Tech. J. 25, 167 (1946). This is an excellent review the arrows point in the direction of decreasing
of the rnagnetron development during the war, treating
voltage. Then if the network is linear, the
both the theoretical and the practical side, and can well
be used to supplement the discussion of magnetrons in voltages will be linear functions of the currents:
the present article. &t =Zitst+Zt222,
J. R. Pierce, "Retlex Oscillators, " Proc. I.R.E. 33, 112
(1945). V2 = Z2121+Z2222i
E. L. Ginzton and A. E. Harrison, "Reflex-Klystron
"
Oscillators, Proc. I.R.E. 34, 97 (1946). These two articles where it can be proved that Z~2 = Z2~, the so-
present information on reflex klystrons which supplements called reciprocity relation. The Z s are quantities
the treatment of the present article, but they are not of the nature of impedances. By a transmission
complete reviews in the sense that the article of Fisk,
line we mean a set of many identical four-
Hagstrum, and Hartman is in its fie1d.
terminal networks, connected together. Each one
The references which we have just enumerated is governed by equations like (I.1). For sym-
will be sufficient to allow the reader rather easily metry, however, it is more convenient to choose
to become acquainted with the published litera- the convention of signs differently. As in Fig. 2,
ture t, n the field. The main references, however, we choose all voltages to be positive when the
should properly be to the unpublished material
upper terminal is at higher voltage, all currents
which is scheduled to come out within a year, to be positive when they How to the right in the
principally the Radiation Laboratory Series, and upper terminals. That is, the sign of the voltage
they will contain material far more complete at the right-hand terminals of a network is
than anything that has appeared in print so far. .
reversed with respect to the convention of (I.1),
I. THE FOUR-TERMINAL NETWORK AND THE but all other quantities are unchanged. In this
TRANSMISSION LINE case, with the Z's meaning the same as above,
1. Definition of the Four-Terminal Network and we have the following relations between the
the Transmission Line currents and voltages at the nth and (n+1)st
The principles of microwave electronics can be terminals:
developed entirely without using the theory of I ii Zll'4+Z122n+1
(I.2)
lumped constant circuits, and we shall so develop V~+I. = —Zg2z„—Z2gz~+ . I.
 J. C. SLATER
Our main problem will be to discuss the solutions real part n positive, the other with a negative.
of (I.2); in the process of discussing the trans- Henceforth we shall denote only that solution
mission line, we shall find the properties of a whose n is positive as p, and shall explicitly call
four-terminal network as a special case of a line the other one — y. Then we see at once that the
of one element. solution involving e & represents the wave propa-
gated to the right, that involving e& the wave
2. Exponential Solution for Voltage and Current propagated to the left, since the wave must be
attenuated in the direction of propagation. The
The family of Eqs. (I.2), taken for all the
equations above determine not only y; they also
values of n concerned in the transmission line,
determine the ratio V„/i„ for either of the solu-
which we assume for the moment to be unlimited
tions we have found. Finding V„/i„ from (I.3)
in both directions, can be solved by a simple
and (I.4), we have two alternative expressions;
assumption: we let both V and i„vary expo-
adding and dividing by two we get a symmetrical
nentially with n. That is, we assume
form,
V„= Vpe&", i„=ipe&", (I.3)
~11 ~22
~Z&2 sinh y= i azo, (I.S)
where Up,ip, y, are constants. Substituting in 2
(I.2), we find at once
where we define
Uo —(Z»+Z»e&)io —— —(Z»e — +Z22)$Q (I.4)
~11 ~12
Eliminating io from (I.4), we have a quadratic Zo= —Zi2 sinh y.
for e~,
t'Zii+Z22)
e'&+) ~e'+1=0, (I.5) In these expressions, the upper sign goes with
z„ the solution e&, the lower sign with e &.
whose solutions are
3. The Terminated Line
Zii+Z22 t'Zii+Z22& In the preceding section, we have found two
(I 6)
2Zi2 solutions of our problem:

We readily find that the product of the two V =(I+Z )e &" i =e &

solutions of (I.6) is unity, so that the two are

reciprocals of each other, and the corresponding
V„= (I —Zg)e&", i„=e&" (I.10)
values of y are the negatives of each other.
In these expressions, we have seen that the first
Remembering that all our quantities of course
corresponds to a wave propagated toward the
must be multiplied by the time variation e&"', we
right, the second to a wave propagated toward
see that our two solutions, corresponding to the
the left. The general solution can be built up as
two signs for y, represent traveling waves, the
a linear combination of these two. With an
imaginary part of p corresponding to a propaga-
infinite line, any linear combination is a possible
tion constant, the real part to an attenuation
solution. On the other hand, if the line is only
constant; the two solutions represent waves
semi-infinite, consisting of sections stretching
propagated in opposite directions, so that one
indefinitely to the left from the 4th terminals,
may be considered the direct wave, the other the
but is terminated with an impedance ZI, across
reflected wave. If we write the complex number
those terminals, the wave traveling to the right
in terms of its real and imaginary parts,
can be treated as an incident wave, that traveling
v=~+iP, to the left as a reflected wave, and we find that
the amplitude of the reflected wave is fixed if
then we shall have two solutions, one with the that of the incident wave is known. Let us
 M I CROWAVE ELECTRONICS
assume a solution Solving (I.12), we have at once
V„=A (I +Zo) e 7"—
+ B(l o) e'",
Z—
Zo+I —Za
(I 11) g2yk (I.13)
i„=As &"+Be& A Zp —f+ Zy
where A, j3 are coefficients to be determined.
Their ratio can be found from the fact that
4. Impedance of the Terminated Line
Zq= Vq/iq T.hus we have We now ask the question, what is the im-
pedance V /i„at the eth terminals, if a line is
i yz, + (l —z,)(B/A)" ~

terminated by the impedance Z& at the kth P

Zk
1+ (B/A) e'&" — To find the answer, we need merely take the
1 (B/A ) e'&" ratio V„/i„ from (I.11), substituting for B/A
=t+Zo (I.12)
1+(B/A) s'&" from (I.13). We find easily, denoting V /i„by Z„,

sinh y(k —n) + (Z~ I ) c—

osh y(k —n) q
z„—I =Zo!(Zo (I.14)
y(k —n)+(Zq —t) sinh
~

(Zocosh
~

y(k n))—
There are two interesting special cases: Zk —f' = 0, Equation (I.17), which is equivalent to (I.14),
and Z~ — I = ~. If I =0, a case which we often is often more convenient, on account of its very
meet, the first corresponds to a short circuited simple dependence on n. It tells us that each
line, and the second to an open circuited line. section of the line which we go through multiplies
In these two cases we have the reHection coefficient by the complex factor
Z„—I'=Zo tanh y(k —n), e '&. It may be used, for instance to give a
or
Zo coth y(k —n), (I.15) simple solution of the following problem. In
Section 1, we have characterized a single four-
respectively. Equation (I.14) expresses the prop-
terminal network by three coefficients, Z», Z»,
erties of the section of line between the nth and
and Z~~. From (I.6), (I.9), we can write these
the kth terminals as a transformer: If an imped-
in terms of y, I, and Zo
ance Z~ is connected across the kth terminals,
we find that the line transforms it into an im-
p+ t, Z22 = Zp coth p —I,
I

Z] y ='Zo coth
pedance Z„across the nth terminals. This rela- (I.18)
tion can be expressed in another form which is Z~2 = —Zo csch y.
sometimes more convenient, by introducing a
quantity r„by the definition But now suppose we have, not a single network,
z —I —zo but s identical networks connected together.
n = ——8 ~ ) (I.16) They still form a four-terminal network, and
A Z„—I +Zg
can be characterized by new coeAicients, which
where the second form is derived as is (1.12). as (Z~~)„(Z~2)„(Z~~)., which,
we may write
The quantity r„ is the negative of the ratio of the
substituted into equations like (I.1), would give
reflected current to the incident current, at the
the relationship between the voltage and current
nth terminals. Its value for n=k is a measure of
the reflection coefficient at the terminal imped- at the input and output terminals of the compo-
ance Z~. It is convenient to denote r„as the site network. It would be a considerable task to
complex reHection coefficient, in general. |A'e find these coefficients by straightforward manipu-
then have lation. However, we note from (I.17) that the
rn only difference between the transformation of
e '~&~ "'
the complex reflection coeAicient produced by a
(I.17) single section of line, and by s sections, is that p
Z„—f —Zp ——
— — (Zg: I Zo) is multiplied by s in the second case. Thus, by
& ~~(a n)
Z„—I +Zo Ez~ —I+zo& retracing all our steps, this must be the only
 J. C. SLATER
change introduced into (I.18), so that we have has three parameters, which can be uniquely
determined from the impedance coeScients. We
(Z11), = Zp coth Vs+ I, observe that (I.22) expresses a transformation
(Z22) Zp coth —
ys f, (I.19) of the complex number s into the complex num-
ber w. If we set up a complex plane for s, another
(Z12), = — Zp csch ys.
for m, the transformation maps each point of the
We may then write equations, similar to (I.2), s plane onto the m plane, or vice versa. We can
for the relation between the currents in the nth, prove in a trivial manner that the inverse of a
arid the kth, terminals of our line bilinear transformation (that is, the solution for
s in terms of w) is again a bilinear transformation,
V„= (Z11) 2i„+(Z12) „1,2(„ so that the mapping of either plane onto the
(I.20) other is of the same nature. We can furthermore
Vg = — (Z12) n —2&n (Z22) n —222,
prove that the result of making two bilinear
where the Z's are given in (I.19), substituting transformations in succession is itself a bilinear
n— k for s. If the kth terminals are connected by transformation. We now notice that since the
an impedance Z~, we have V~=Zl, iI, . Making function (I.22) is analytic, the mapping must be
that substitution, and solving for the impedance conformal, by fundamental principles of the
Z„= V„/ 2see nat the nth terminals, we have theory of functions of a complex variable. That
is, the shape of a small figure in the s' plane is
(Z12)n-1
Z„= (Z11) „2— preserved in the corresponding figure in the m
+
Z1. (Z22) n-1. plane, though the scale in general will change,
and the figure will be rotated. As a result of
2' +
(Zll) n — (Zl1) n —1 (Z22) n 2(Z12) n 1c— —
. (I.21) this conformal nature of the transformation, if
+
Zg, (Z22) „g two lines cross at a given angle in the s plane,
the transformed lines will cross at the same angle
The two forms of (I.21) are often used for in the m
plane.
expressing the transformer properties of a four- We next prove a property peculiar to the
terminal network. They are equivalent to (I.14)
bilinear transformation: any circle in the s plane
and (1.17), as can be shown by straightforward
is mapped by the transformation snto a circle in
manipulation. the m plane, and vice versa. To prove this, we
note that there are two simple ways of expressing
S. Bilinear Transformations
the equation of a circle. First, an equation
In (I.14) and (I.21), we have two forms of the
relation between the terminal impedance ZI, of a sz+As+Az+BB = 0, (I 23)
—
finite line of n k sections, and the transformed
where a bar indicates the complex conjugate,
impedance which we see across the nth terminals.
represents a circle. For if we let z=R+ jX, this
Both equations express Z„as a bilinear function
1s
of Z~„ that is, as a function of the form
(I.22) R'+X'+ (A+A)R+ j(A A) X+BB , (1.24)
w= (as+5)/(cs+d), = 0—
where the complex number m stands for Z„, z for in which A+A, j(A —3), and BB are all real
ZI„and a, b, c, d are complex constants which numbers, and this is obviously the equation of a
have different meanings in the two cases. The circle in the s' plane, in which R and X are
properties of bilinear transformations are so coordinates. Secondly, an equation
important in our whole theory that we shall w= C+ pe'~, (I.25)
study them in some detail. We note in the first
place that in any bilinear function like (I.22), where C is a complex constant, p a real number,
we can divide numerator and denominator by c, g a real number, represents a circle in the w
so as to reduce the coefficient of s in the denomi- plane, with center at C, radius p, provided p
nator to unity, as in (I.21). Thus such a function takes on diA'erent values to represent parametri-
 M IC RO~VA VE ELE CTRON I CS
6. Graphical Discussion of Bilinear
Transformations
The easiest way to visualize a bilinear trans-
formation is to consider how certain lines in the
s plane are transformed into corresponding lines
in the m plane. Suppose we consider the two
characteristic impedances, and the family of
circles passing through them, in the s plane, as
in Fig. 3. The characteristic impedances trans-
form into themselves in the m plane, and any
FIG. 3. Circles passing through characteristic irnpedances, circle transforms into a circle. Thus each circle
in bilinear transformation. of this family must transform into another circle
passing through the characteristic impedances,
cally the v'arious points of the circle. Let us now or into another circle of the same family. Con-
represent m in this way, eliminate p, and show sider similarly the family of circles orthogonal
that the resulting equation in s represents a to these, in the s space. Each one of these must
circle. We have transform into another circle in the tv space,
which must by the conformal property of the
az+b ez+ b
pe&~ = ——C, pe '& = —C. (I.26) transformation also be orthogonal to the family
cz+d cz+d of circles passing through the characteristic
impedances. In other words, each circle of this
Multiplying these together, p cancels out; ration- second family transforms into another circle of
alizing the denominator, we find at once that the the same family. We can understand the exact
equation for s is of the type described above, so nature of the transformation better if we consider
that we have a circle in the s plane, resulting the complex reHection. coefficient, defined as in
from transforming a circle in the m plane. As (I.16). That is, if zi, s2 are the two characteristic
another mathematical theorem regarding the impedances, the two solutions of (I.27), we
bilinear transformation, we shall prove that there define
are two particular values of s which are trans- r = (s — si)/(s —z, ), (I.28)
formed into themselves, or remain invariant
under the transformation. To find these, we need or, considering the actual impedances, we define
only require tha. t w=z in (I.22). Then we have
cz'+ (d —a) z —b = 0, a quadratic for s, whose
(Z f Zo— „I
) /(Z— +Zo—
) (I.2. 9)
solutions are This is again a bilinear transformation. We note
that when s=zi, or Z„=I+Zo, r or r„ is zero,
(g d)2 b and when z=z~, or Z„=I — Zo, r or r„ is infinite.
~ I+-c . (I.27) The circles passing through s~ and s2 in the z
2c
i
0 2c )
plane then transform into straight lines through
The physical meaning of these values, in our the origin in the r plane, and the orthogonal
case, is simple. They represent the impedances family of circles transform i@to concentric circles
which must terminate the line, in order that the with the origin at the center. We may call such
impedance across the nth terminals should be a transform of the impedance space a circl'e
. the same as across the kth terminals. That is, diagram. The bilinear transformation (I.14) or
they must be just the values given by (I.8), (I.21), when exhibited in the circle diagram,
characteristic of having only a direct or only a reduces to (I.17), That is, it corresponds to a
reflected wave. We find, in fact, that if we multiplication of r& by the constant factor e '&'.
substitute the proper coeiticients into (I.27) from The medulus of this factor corresponds to a
either (I.14) or (I.21) we come out with just the change of scale along the radii, and the phase
values (I.8). These values are called the char- gives the rotation. Having found this interpreta-
acteristic, or iterated, impedances of the line. tion of the bilinear transformation in the circle
 J. C. SLATE R
diagram, we can go back to the impedance plot, y are real. In that case the transformation is the
and see that the change of scale, which arises type which expands the scale around one char-
from the real part of y, corresponds to a process acteristic impedance, contracts it around the
of expansion of scale around one characteristic other. The first of these two types of lossless
impedance, and a shrinking around the other, networks is that found for instance in an ordinary
while the' rotation corresponds to a process in section of wave guide or other transmission line,
which one of the circles surrounding one of the where the points in the circle diagram rotate on
characteristic impedances transforms into itself. passing through a length of line. The second is
I

-that met in a wave guide beyond cut-off, in

7. Special Types of Networks and Lines which all impedances in the Z~ plane tend to
Let us now consider several special cases of shrink toward the characteristic impedance in
networks and lines and their corresponding bi- the Z„plane.
linear transformations, both in the impedance The general case of a network with losses
plane and the circle diagram. First we consider cannot be handled in such a simple way. The
a lossless network, containing no resistances. In general bilinear transformation requires three
this case, a reactance connected across the termi- complex constants to describe it, which we may
nals k must lead to a reactance seen across the take to be the values of the two characteristic
terminals n We c. an draw certain conclusions at impedances, and the value of y. If the network
once from this fact, using (I.21). If Zi is infinite, is 'passive, that merely tells us that every value
or the network is. open circuited, Z„must be of Z& in the right half plane (that is, every
pure imaginary; thus (Zii), must be pure imagi- impedance with a real resistive component) must
nary. Similarly, open circuiting the terminals k, lie in the right half plane of the Z„space. In
we find that (Zg2). must be imaginary. Short general, however, it will transform into less than
circuiting the terminals k, Z„must be imaginary, the whole half plane. The axis of ordinates in
or (Z»)P/(Z2~), must be imaginary, from which the Z~ plane must transform into a circle in the
(Zi2) 2 must be real. An impedance Zi, with a right half plane in the Z„space, so that all
positive resistive component must transform to physically allowable values of Zj, will transform
a Z„with a positive resistive component; thus into the interior of this circle in the Z„space.
(Z»), ' must be negative, and (Zi2), is imaginary. We find, in general, that the greater the loss in
Ndw we consider the transformation geometri- the network, the smaller is the circle in the Z„
cally. Any point on the axis of ordinates in the plane. Thus an attenuator in a line results in
Zy plane must transform into another point on seeing almost the same impedance, regardless of
this axis in the Z„plane, for all pure reactances what is attached to its other end.
are found on the axis of ordinates. There are
only two ways in which the bilinear transforma- 8. Transformation of Resistance and Reactance
tion can carry a straight line into itself. The Coordinates
,

6rst is that in which the straight line is the In the general network with losses it is not
perpendicular bisector of the line joining the particularly convenient to use the family of
characteristic impedances, and in which the circles which we have so far considered, for the
transformation in the circle diagram is a pure line of zero resistance does not form one of these
rotation, without change of scale. That is, the circles. It is often convenient to visualize the
resistive parts of s~ and s2 are equal and opposite, transformation by drawing in the Z„space the
so that I in (I.8) is pure imaginary, and Zo is transforms of 'the lines A~=constant, Xi=con-
real. Equation (I.9) then tells us that sinh y 'is stant, or the rectangular coordinates in the k
pure imaginary, or that y is imaginary, equal to space. These form two orthogonal families of
jP, and the factor e '&' becomes a pure rotation straight lines; they then transform into two
in the cirde diagram. The other case arises when orthogonal families of circles, as we see in Fig. 4.
the two characteristic impedances are both The circle R~ = 0, bounding the region of positive
located on the imaginary axis. In that case both resistances, forms a member of one of the families.
I and Zo are pure imaginary, so that sinh y and The point Zy, = ~ must lie on this circle, and each
 M IC RO% A VE ELECTRON I CS

measured along the line, which would be the

case if we regarded each unit length of the line
as one-of the networks of our previous discussion.
In the reference quoted above, it is shown how
the characteristic impedance Zo and the propa-
gation constant y can be found from distributed
constants of the line, by equations which repre-
+ON$TAN7
sent the passage to the limit of (I.6) and (I.9),
but we shall not need to use those equations.
Xg & coN
With the small changes noted, the whole discus-
+ R„=o
sion of Sections 4—7 applies to continuous lines.
FIG. 4. Transforms of resistance and reactance coordinates, Thus, in particular, we can introduce a reHection
for general transformation.
coefficient, by (I.16), equal to (Z — Zo)/(Z+Zo),
of the circles X~ = constant must go through this where Z is the impedance at an arbitrary point
point, and must cut the circle 8~=0 orthogo- of the line, and as in (I.17), the reHection
nally. Each circle R& = constant must also go coefficient is multiplied by a factor e—'&' if we
through this point, and must be tangent to the travel backward a distance s along the line. We
circle R~=O there. Thus the plot of lines of can regard a length s of transmission line as a
constant R~ and constant XI„ in the Z„plane, transformer, with transformer coefficients given
must have the form shown in Fig. 4. This, it by (I.19), setting (=0.
must be emphasized, is perfectly general, holding We are often interested in the lossless line. In
for an arbitrary network with losses. The three that case, as we saw in Section 7, there are two
complex constants characterizing the network, possible cases. First, we can have Zo real, y pure
in this way of exhibiting it, may be taken to be imaginary. In that case there is real propagation
the coordinate of the center of the ciI'cle; the along the line, with no attenuation, and the
radius, and orientation of the point of infinite elfect of traveling along the line a certain distance
impedance Zq (the modulus and phase of a is to rotate through a certain angle in the circle
complex number); and the location, within the diagram, a rotation of 360' corresponding to a
circle, of that characteristic impedance which half-wave-length. The rotation is positive, or
lies within. counter-clockwise, as we go away from the
generator toward the load, or negative as we go
9. The Continuous Transmission Line toward the generator. In the impedance plane,
A continuous transmission line, such as a we may set up circles like those in Fig. 5. The
parallel wire line or a wave guide, can be con- circles surrounding the characteristic impedances
sidered as the limiting case of the type of line represent circles on which the magnitude of the
we have been considering, as the separate ele- reHection coe%cient r is constant, and the circles
ments become smaller and smaller, and closer passing through the characteristic impedances
and closer together. * We may get the relations are circles of constant phase of reHection coefh-
for this continuous line easily by passing to the cient. Traveling along the line toward the gener-
1imit from the various formulas which we have ator, we travel in a negative direction about
developed. In such a continuous line, the voltage the first family of circles. Once in a half-wave-
and current are given as functions of distance length the impedance goes through a maximum
along the line by formulas like (I.11), in which real value (we shall presently show that this
we need make only two . modifications: first, corresponds to the standing wave maximum),
such lines are symmetrical, one end being and a quarter-wave-length further along it goes
equivalent to the other, which means that f is through a minimum (the standing wave mini-
zero; secondly, we must let n refer to the distance mum). The axis of ordinates forms a limiting
case of these circles, corresponding to the case
For a discussion of this limit, see J. C. Slater, 3ficro- where the terminal impedance is purely reactive.
mave Transmission (McGraw-Hill Book Company, Inc. , '

New York, 1942), pp. 17-21. In that case, the maximum impedance is infinite,
450 J. C. SLATER
cut-off. There is only attenuation, no propagation,
down the line, and the effect of traveling along
the line is to shrink the scale in the circle dia-
gram, without rotation, so that after traveling
such a distance down such a line, no matter
what the terminal impedance may be, the im-
pedance seen looking into the guide is very close
to the characteristic impedance, which we re-
member is purely imaginary. In Fig. 7, showing
FiG. 5. Circles of constant reRection coe%cient and phase, the impedance plane, the e8ect of passing down
in the impedance plane. the line away from the load is to travel along
the circles passing through the characteristic
the minimum is zero. All real loads lie in the impedance, as shown by the arrows. In the circle
right half plane, corresponding to positive re- diagram, the role of R and X is interchanged
sistive components. Correspondingly, in the circle with respect to the case of propagation on 'the
diagram, as in Fig. 6, all real loads correspond to lossless line. The horizontal axis becomes the
reflection coefficients within a unit circle, the line R=O, the positive resistances lying in the
circle into which the axis of ordinates in the lower half plane, and the unit circle becomes the
impedance plane transforms. To see this, we line X=O, so that we need the area outside as
need only notice that if we set Z = 0 or Z = ~ in well as inside the unit circle to describe all real
the expression r = (Z — Zo)/(Z+Zq), we' find terminal impedances. We can again use the circle
r= —1 and +1, respectively, so that these are diagram to represent the effect of a length of
two points on the circle representing a purely line: traveling down the line a certain distance
reactive load. Corresponding to Fig. 4, we can away from the load shrinks the radius vector by
draw lines of constant resistance and constant an appropriate factor, without rotation, so that
reactance in the circle diagram. Since we have as we have mentioned before all impedances
just seen that the point Z = ~ transforms to the approach closer and closer to the characteristic
point r=1, the circles of constant X all pass impedance as the length of the attenuating line
through this point, and the circles of constant R becomes greater.
are tangent to unit circle there. This diagram
can be used as a very convenient way for finding
10. Standing Wave Ratios
the transformation produced by a length of The case of a lossless continuous line is a very
lossless line on a given terminal impedance. If important one practically, for wave guides are
the lines of constant R and X are properly used in practice for measuring standing wave
labeled, we can at once look up on the diagram ratios, and hence terminal impedances. We,
the point corresponding to a given terminal therefore, next consider the definition of standing
impedance. If we choose, we can find the wave ratios in a continuous lossless line in which
modulus and phase of the reflection coefficient,
by measuring the radius and phase angle of the
radius vector out to the corresponding point,
from the center of the circle. To travel a given
distance along the line, we then merely rotate
the radius vector through an appropriate angle, Z=o Z=Zo I .Z, c (gl
and read off the transformed values of R and X.
)

X=0
This construction is made the basis of a con-
venient rotating slide rule for calculating im-
R= CONSTAI&T
pedances.
A=g
The other case of a lossless transmission line Cb
is that in which Zo is imaginary„p real. This
FiG. 6. Circles of constant resistance and reactance, in
corresponds, for instance, to a wave guide beyond the reflection coe%cient plane.
 M IC ROWA VE ELE CTRON I CS
there is real propagation. The measurement of
standing wave ratio involves a measurement of
voltage as a function of distance along the line.
What essentially is done is to put a very high
impedance shunt across the line, and measure FIG. 8. Diagram for interpreting standing wave ratio.
the power generated in that shunt; the power
will be the square of the modulus of the voltage, points, as we have mentioned earlier, are the
multiplied by the conductance of the shunt. standing wave maxima, or points at which maxi-
From (I.11), the voltage as a function of distance mum power is delivered to the standing wave
n along the line will be detector; and the minimum length comes when
r is real and negative, when its value is 1 — r l,
V =Z (Ae & Be&— "
) =Zo(Ae 't'" ") (I.30)
Be'&— and these points are the standing wave minima.
l

Introducing the reHection coefficient We now define the standing wave ratio in voltage
as the ratio of maximum to minimum voltage
r= —(B/A) e "t'" (I.31)
1+lrl
=—
from (I.16), we see that the voltage can be SWR(voltage) (I 34)
written
V„=ZOAe» (1+r), (I.32) We furthermore define the standing wave ratio
in power as the ratio of maximum to minimum
so that the power delivered to the shunt is
power
proportional to
= (I+ lr I )'
lv. =z. lAl (1+r)
l
(I.33) SWR(power) l (I.35)
and the magnitude of the voltage is proportional ?t is often convenient also to define a standing
to l(1+r) l. In Fig. 8 we see the locus of the wave ratio in decibels, as the number of decibels
points 1+r as we go along the transmission line. by which the power delivered to the stariding
It is clearly a circle of radius r l, with center at
l
wave detector at standing wave maximum must
the point 1. The radius vector from the origin be attenuated to make it equal to the power
out to this circle then has a length proportional delivered at standing wave minimum. We h'ave
to the voltage, so that the square of its length i.s
proportional to the power delivered to the shunt, SWR(db) = 10 Iogyp SWR(power)
as in (I.33). It is now clear that the maximum = 20 log~o SWR(voltage). (I.36)
length of this vector comes when r is real and It is now clear that by a measurement of the
positive, when its value is 1+ r l, so that these
l
standing wave ratio we can find the magnitude
of the reHection coefficient, lrl. Also from the
position of the standing wave minima and max-
ima along the line, we can find the phase of the
reHection coefficient: in the circle diagram, r has
a phase angle of zero at standing wave maxima,
and of 180' at standing wave minima. Thus a
measurement of standing wave ratio and position
of standing wave minimum allows us to find the
impedance seen across any arbitrary plane of the
continuous transmission line. A given standing
wave ratio corresponds to a given radius in the
circle diagram, as we see from (I.34), and the
phase angle of rotation corresponding to a given
plane is simply found by measuring the distance
FIG. 7. Transformation produced by lossless line
from standing wave maximum to the given plane,
beyond cut-off'. and noting that a half-wave-length rotates by
 J. C. SLATE R

360'. It is interesting to note that the impedance

at standing wave maxima and minima is easily C J
found from the standing wave ratio. Solving the
equation r = (Z —Zo)/(Z+Zo) for Z/Zo, we have
FiG. 9. Four-terminal network with transmission lines.

respect to the characteristic impedance of the

At standing wave maximum, r = r ~; thus, com-
~
line to the right. That is, if Zp1, Z02 are the
paring with (I.34), we see. that Z/Zo at standing characteristic impedances of these two lines, we
wave maximum equals the standing wave ratio have
in voltage. At standing wave minimum, r = —r ~; ~
Z1 Zpl Z2 Z02
thus Z/Zo at standing wave minimum equals r2— f (I.38)
the reciprocal of the standing wave ratio in Zl+Zol Z2+Z02
voltage. We note, then, that the characteristic where Z1, Z2 are the impedances at the two
impedance Zp is the mean proportional between
points. We now know that there must be a
the values of impedance at standing wave mini- bilinear transformation conriecting Z1 and Z2.
mum and maximum.
Since r1 and Z1, arid r2 and Z2, are connected by
the bilinear transformation (I.38), we know that
11. Transformers between Transmission Lines
r1 and r2 are connected by a bilinear transforma-
In many microwave applications, two sections tion. Furthermore, the transformer is lossless;
of wave guide are connected together by a lossless thus a point on the unit circle in the r2 space
section of some sort, acting as a transformer must transform into a point on unit circle in the
between the two guides. It can be merely an r1 space, since the unit circle in each case repre-
iris or a change of cross section, or a much more sents a pure reactance. A lossless transformer is
complicated object, such as a resonant cavity characterized by three parameters, as we saw for
with two separate wave guide outlets. The corre- instance in Section 7, where for the lossless case
sponding circuit is a four-terminal network, with- the three parameters Z11, Z12, and Z22, which in
out losses, with a continuous transmission line the general case are complex and hence amount
connected to each of its pairs'of terminals, as to six independent numbers, are pure imaginary.
shown in Fig. 9. We can get interesting and We shall now set up a bilinear transformation
valuable results by considering the impedance between r1 and r2 which satisfies all these condi-
transformation between points 1, 2, arbitrary tions, and which thus is a general expression of
points of the two guides. The whole circuit the transformation. This is
between 1 and 2 is, of course, itself a four-
terminal network, which can be handled by the
same methods we have already employed. We
(I.39)
shall, however, write the impedance transforma-
tion in a different form from what we have used where p~, p2, and D are three real constants. Let
before. In the first place, we shall make a trans- us check that this transformation has the re-
formation, not of the impedance, but of the quired properties. First, it is clearly a bilinear
reHection coefficient. We shall set this up, how- transformation, with three independent param-
ever, in a way which differs essentially from that eters. Next we must show that it transforms
used, for instance, in Eq. (I.17). In that case, unit circle into unit circle. If r2 is a point on
we defined the reHection coefficient with respect unit circle (that is, if its magnitude is unity),
to the characteristic impedances of the whole then clearly r2e &», which simply amounts to the
network between terminals 1 and 2. Here, on same vector rotated through angle — @2, is also
the contrary, we shall define r1 as the reHection on unit circle; similarly if r1 is on unit circle,
coefficient with respect to the characteristic im- r1e &» is on unit circle. We must then merely
pedance of the transmission line to the left of the prove that if r2 is on unit circle, (D+r2)/(1+Dr2)
network, and r2 as the reHection coefficient with is also on unit circle. We note first that the
 M I CROWAVE ELECTRONICS
points ~1 are invariant under this transforma- But by (I.37), this is equivalent to
tion: if r2= &i, rl= &i also. Thus the transform
Zi' 1+D Zg' Zoi 1+D )
of unit circle in the r2 space, into the r~ space,
, or Zi' —f IZ2'. (I.42)
passes through these two points. Furthermore, 1 D— D)—
~

Zoi Zou KZo2 1

the transformed circle in the r& space is' sym-
metrical about the axis of abscissae. We can see In other wordk, between the points i' and 2', the
this by noting that the transforms of the points transformation is one of multiplying the imped-
j
& in the r2 space are (D&j)/(1&jD) .in the ri ance by a real constant. We have then the
space, which are complex conjugates of each valuable general theorem: if we have any type
other, and hence are located symmetrically with of lossless network, with two outputs in the form
respect to the real axis. This fixes the transform of transmission lines, we can find sets of points
of unit circle into the r~ space uniquely, and on each of the lines, such that the impedance
shows that it is again unit circle. The transfor- across one of the points is a real factor times the
mation (D+r2)/(1+Dr2) is in fact the sort in impedance across the other.
which there is a motion along those circles
passing through the fixed points of the transfor- 12. Determination of Transformer Constants
mation (which in this case are &1), without
rotation, as shown in Fig. i0; such a transforma- It is often very important to determine experi-
tion must always carry the circle passing through mentally the three constants characterizing a
the two fixed points, and with its center at the lossless transformer connecting two transmission
point midway between the two fixed points, into lines, such as we have just been discussing.
itself. Now that we have justified the form There are several ways to do this. For example,
(I.39) for the transformation, we can give it a we may place the characteristic impedance Z02
very interesting physical interpretation. In the across the terminals 2, or match the right-hand
first place, r~e '&1 represents the reHection coeffi- line. Then r2 will be zero, and we shall have
cient, not across the point 1, but across another ri'=De~'&1, so that measuring the standing wave
point of the same transmission line; for we re- ratio and position of standing wave minimum in
member that the reHection coefficient is multi- the left-hand line, we measure D and pi. Simi-
plied by the factor e '&&' in moving a distance s larly we can match the left-hand line and measure
toward the generator. We may call the reHection the standing wave ratio and position of standing
coef6cient across this point r&', which is equal to wave minimum in the right-hand line. We must
r ie &&1. Similarly r~e &» is the reHection coefficient be careful how we apply the formulas to this
r2' across another point in the other transmission case, for we have set them up only for the case
line. The transformation between these quanti- where the power is Howing to the right; but it
ties is then turns out, as we should expect, that from this
= (D+r2')/(1+Drm'). measurement we can determine D and @2. An
&i' (I.40) experimental check is to see whether the values
Ke can easily prove from this that of D obtained in the two cases agree. This
I

method of finding the constants, though some-

1+ri' 1+D 1+r2' times convenient, is not ideal, in that both
(1.41)
i —ri' i —D 1 —r2' transmission lines have to be equipped so that
we can insert matched loads into them, and so
that we can measure standing waves in them. .
Often a more convenient measurement is that of
inserting a short circuit (that is, using wave
guides, a plunger) at an adjustable point along
the right-hand transmission line, and measuring
the position of standing wave minimum in the
.

FK'. 10. Mu)tiplication left-hand line as a function of the plunger posi-

of impedance by real constant, in
reRection coefficient plane. tion. Let us analyze this experiment. We assume
 J. C. SLATE R
the short circuit to be located a distance d2 to transformer equations in either of the forms
the right of the point 2. Then we shall have
1 —r1e —r2e
&» 1 &&'

2
—
e 2 jP2d2 (I.43) 1+r,e «1'+r2e
& &
«'+is (I.47)
where P2 is the value appropriate to the right-
1+r,e '«1+r~e &
«'
hand line. In this case we can show easily that +jx (I.48)
r1e '» 1 —r2e &&'

Z2'

Z02
= j t» —+P2d~
I I (I 44)
These relations are both bilinear transformations
(we can see this easily if we solve them for r~ in
terms of r2); they have three arbitrary constants,
In the left-hand line, there will be an infinite
if p&, p2, and y or x are real. Furthermore, they
standing wave ratio, or a reHection coefficient of
transform reactances into reactances. To show
unit magnitude, since we shall be seeing a purely
this, we let the magnitude of r1 be unity. That
reactive impedance. Thus at a standing wave
is, we let r2=e &'~2, where 02 is real. Then the
minimum the impedance will be zero, or there
expression on the right of (I.47) becomes
will be an eAective short circuit. Let a standing
wave minimum be located a distance d1 to the j
tan (82+&2)/2, which is pure imaginary. Add-
ing jy, we sti11 have a pure imaginary. Reversing
right of the point 1. Then by an argument just
the argument, this shows that r1 has unit magni-
like that used above we have
tude, or that the impedance Z1 is a pure react-
Zl (4i ance. The meaning of (I.47) or (I.48) is simple.
=jt» —
Z„&2 +Pidi I.
I
)
(I.45) The quantity on the left of (I.48) is the ratio of
the impedance, to the characteristic impedance,
at a certain point on the line, as we see from
We then have, as a result of (I.42),
(I.37). The quantity on the right has a similar
l 1+D interpretation. Thus we see that there is a
(A —
t» — +p~d~ =
(42
t» +P2d2 (I.46) certain point on the right-hand line, and a corre-
1— i
I I I I

D &2 sponding point on the left-hand line, such that

the impedance seen at the point on the left line
This equation is in a convenient form for practi- is a series combination of the impedance at the
cal use. If we plot P~d~ as a function of P2d2, we
point on the right, and a fixed series reactance jx.
get a curve of the form shown in Fig. 11, some- Similarly the quantities in (I.47), being the
times called an 5 curve on account of its shape. reciprocals of the impedances, are admittances:
The coordinates of the points of maximum and there are corresponding points on the two lines
minimum slope determine g~ and @g, as shown, such that the admittance across the left-hand
and the slope at these points determines D: the point is the shunt combination of the admittance
maximum slope is (1+D)j(1 D), and the mini-— across the right-hand
point, and a fixed shunt
mum is its reciprocal. admittance jy. It is to be understood that the
and p~ are in general different in
13. A Lossless Transformer as a Shunt or phase angles &I
Series Reactance
(I. 47) and (I..48), and in turn different from
those in (I.39); that is, there are different sets
In Section 11 we have shown that a lossless of corresponding points in the two lines, with
transformer between two transm'ission lines can respect to which the transformer is an ideal
be described as an impedance multiplication by transformer (that is, the impedance is multiplied
a real transformer ratio between certain definite by a given ratio, as in (I.41) ); or a shunt reactance
points in the two lines. There is another equally (as in (I.47)) or a series reactance (as in (I.48)).
legitimate way of considering an arbitrary loss- It would be possible to get the relations between
less transformer, which is sometimes useful. the various sets of constants, but it is not very
Using the same diagram of a transformer that profitable to do so. The expression of a lossless
we have given in Fig. 9, we may write the transformer as a shunt reactance is particularly
 M IC ROWAVE ELECTRON I CS
Let us first find the formula for power flow into
a pair of terminals between which there is a
voltage U, and a current i, the sign convention
being the one we have been continually using.
We must remember that when we write V as
the voltage, we really mean that the voltage is
the real part of Ve&'"'. That is, if V is written in
terms of its real and imaginary parts, as V, + V;,
the voltage is V„cos cot —V; sin oit. Similarly the
j
current is i„cos ~t — i, sin ~t, writing in a corre-
sponding way. The power How is the product of
voltage and current; that is, it is
FiG. 11. An 8 curve. Power = V„i„cos' &et+ Vp, sin' art
—( V„i,+ V,i„) sin ~t cos cot. (I.49)
appropriate in such cases as the transformer
produced by an iris in a diaphragm in a wave This is the instantaneous power, which as we
guide. In such a case, provided the hole in the see involves squares and products of sin ~f and
iris is small, we can show that, if points 1 and 2 cos cot. Ordinarily we are interested only in the
are immediately on the two sides of the dia-, time average of the power. Averaging, the
phragm, the angles gi and p2 are practically zero. average of cos'cot and of sin'cot is 2, and of
In other words, the diaphragm acts just like a sin cot cos cot is zero. Thus if I' denotes the
shunt reactance, without any correction terms. average power, we have
I~ more general cases, however, a transformer
will act like a combination of a shunt reactance,
P = ', (V„i,-+ Vp, ). (I.SO)
plus an effective lengthening or shortening of the This can, however, be written in another way.
line on each side. It is interesting to realize, as a If we take the product Us, we have
result of the discussion of the last three sections,
that any desired impedance transformation can (V„+ jV,) (i„—ji;) = (V„i„+V,~, +j(V,i„V,i;)). . —
be secured in many ways, including shunt and The real part of this quantity leads to the power
ser'ies reactances plus appropriate change of as in (I.SO). Thus we have
length of lines. For instance, suppose we have a
change of characteristic impedance between two I' = -', Re V~. (I.51)
guides. We can always insert an iris in such a
This is the standard formula for average power,
way as to match the guides into each other, in in terms of complex amplitudes of voltage and
the sense that a matched load (r2 —0) in one
current. If V=Zi =(R+jX)i, we then have
guide transforms into a matched load (ri —0) in
the other. (I 52)
14. Power Flow in Networks and Transmission where ~i is the magnitude
~
of the current.
Lines The formula (I.51) or (1.52) can be applied at
-
So far, our discussion has dealt almost entirely once to the flow of power in a transmission line.
with the impedance, the ratio of voltage to Let us assume the solutions (I.11) for voltage
current. Of almost equal importance is the power, and current at the nth terminals of a line, taken .

or product of voltage and current. From the together with the definition (I.16) for the com-
two together, we can work backward and find plex reflection coefficient. We may then write
voltage and current separately. We shall find,
in our microwave work, that it is more con-
venient to deal with impedance and with power
voltage and current in the form
V = t'i„+Zo(Ae «" Be&"
= gi
)— —
-
flow than with voltage and current; for imped- +ZOAe «"(1+r„),
ance and power How are more readily measurable. i„=Ae «"+Be«"=Ae «"(1 r). '
—
(1.53)
 J. C. SLATE R
If we let Zp=Rp+jXp, we then have internal reflection in optics. The totally inter-
nally reflected wave results in an exponentially
I'=-', ~~„~'Rei+ ,'Ro-(AAe '~" B—
Be'~")
damped wave in the rare medium. If that
+-', Xoj (A Be '» Be"~"),
A— medium extends to infinity, the exponential wave
carries no energy, and all the energy is reflected.
= ' ~~„~'Ref'+ ', R -ApAe '~"(1 r„r„—
—, ) If, on the other hand, there is only a thin sheet
+-'XpAAe ' "j (r„r„)—
, (I.54) of rare Medium, then another dense medium,
we shall have to have both sorts of exponential
where y=n+jP, as before. Let us see what waves to satisfy the boundary conditions, and
(I.54) becomes in certain simple cases. In a we shall find that power is transmitted through
network without losses, which is propagating the rare medium to the adjacent dense medium.
the wave, we have seen that 1 is imaginary, and Another interesting case is the guide or other
Zo is real, so that XO=O. Furthermore 0. =0. transmission line with a real characteristic im-
Thus in this case we have pedance, or Xp=0, but with a slight attenuation,
P =-', ZpAA. (1 r„r„) =— Zp(AA
-', BB). —(I.55) so that a is different from zero. If (=0, as it is
in a guide, then we have
Thus in this case the power How is simply the
difference between the flows of power in the P =-', Zo(AAe '"—BBe'~"). (I.58)
direct and reflected waves. The quantity r„r„ is This has a simple interpretation: the first term
independent of I;
for if we let is the power How in the direct wave, which varies
r„= —'&1'", as e ' " on account of the decrease of amplitude
pe (I.56)
of this wave as n increases; the other term is the
where p is the magnitude of the reHection coeffi- flow backward in the rejected wave, which
cient, we see that r„r„=p'. On the other hand, decreases as n decreases. In this case the magni-
in a network without losses in which there is tude p of the reflection coefficient depends on n.
attenuation, we have seen that t is again imagi-
15. Power Flow from a Lossless Line into a
nary, Zo is imaginary, and p is real, equal to n.
Terminal Impedance
Thus in this case we have
I' = ,'Xpj (AB A—B). —(1.57) If we have power Howing from a lossless trans-
mission line into a terminal irhpedance, we have
The power How is again independent of ri„but seen in (I.55) that the power How is -', ZpAA (1 rr). —
in this case there is no How unless A and 8 are By (I.38), r=(Z — Zp)/(Z+Zp). If Z=R+jX,
both different from zero. In other words, a we find easily that
purely attenuated wave, in for instance a wave I' (R —Zp+ jX) (R —Zp —jX)
guide beyond cut-off, carries no power. The
reason for this is simple. At one end of the guide, -', ZoAA (R+Zo+ jX) (R+Zo —jX)
at infinite distance, the amplitude is attenuated 4RZO.
to zero, and obviously no power can be flowing (I 59)
there. But since there are no losses along the (R+Z )'+X'
guide, all the power that flows in one end must The expression (I.59) tells how the power ab-
flow out the other. Thus there can be no power sorbed in the terminal impedance varies with R
flow anywhere. On the other hand, if the guide and X, provided AA remains constant. This
is only of finite length, both A and 8 must be means that the power flow in the direct wave is
different from zero, to satisfy boundary condi- constant, independent of the power flow in the
tions, as indicated for instance by Eq. (I.13). reHected wave (for of course that changes, as the
Power will then flow; and this is reasonable load changes). We can secure this situation in
physically, since a certain amount can pass practice by taking a generator of power, then
through such a guide beyond cut-off into a inserting an attenuator of high attenuation, then
terminal impedance. The problem mathemati- our transmission line and load. This is called
cally and physically is like that met in total padding the generator with an attenuator. The
 M I CROWA VE ELECTRON I CS

object of the attenuator is to absorb practically and correspond to generators, rather than passive
all of the reflected wave before it gets back to loads, so that in the presence of such a terminal
'

the power source, so that the reflected wave will impedance the power would How to the left, or
not react back on the source, and change its (I.59) should be negative. The function is nega-
power output. Put another way, we have found tively infinite at the left-hand characteristic im-
that if there is an attenuating section in a line, pedance; the meaning of this is that at that
the input impedance seen looking into the at- characteristic impedance the wave is Howing
tenuating section is almost independent of the entirely to the left, so that the wave traveling
output impedance. That is, the impedance seen to the left is infinite if A, the amplitude of the
by the generator is almost independent of the wave traveling to the right, is finite. These
load impedance; so much of the power is ab- features of the left-hand half plane are not of
sorbed by the attenuator that the small remain- physical interest in ordinary applications. In the
ing power absorbed by the load is negligible. reHection coe%cient plane, the contours of con-
Since the power output of a generator depends stant power are of course circles concentric with
on the load which it sees, this means that the the origin, the power being a maximum at the
power output is practically constant as Z is origin, going down to zero at unit circle, and to
varied, so that the incident amplitude A is a negatively infinite value at infinity.
practically, constant. We must note at the same
time, however, that A represents the amplitude
16. Circuit EfHciency and Insertion Loss of a
of the power delivered by the generator through Resistive Network
a high attenuation, so that this procedure is very If we have a network of the general type
wasteful of power. When we have this condition, characterized by transformer coefficients Z», Z»,'
which is often used for measurement, we then Z» in the manner of Eq. (I.2), and terminated
have an output power proportional to (I.59). by an. impedance Z, we shall find that some of
It is interesting to consider this function as it the power fed into the left-hand terminals of the
depends on R and X. Obviously as far as X is network (which we shall call terminals 1) is
concerned, the power is a maximum when X = 0, absorbed in the network, and some in the im-
or when the load is purely resistive. If further- pedance Z. If we let P~ be the power flowing
more we vary R to make the power a maximum, across the terminals 1, and p2 the power across
we find by differentiating with respect to R and the terminals 2 (the terminals connecting the
setting the derivative equal to zero that the right-hand side of the network to Z), then the
maximum comes when R=ZO, or when the load fraction P2/Pi of the input power will be de-
impedance equals the characteristic impedance livered to the load. If the object of the network
of the line, or is a matched load; in this case the is to deliver power, as it is in some practical
function equals unity. It is interesting to consider cases, we may call this ratio the circuit efficiency,
the contours of constant power in a Z plane, and denote it by q, . Thus we have
in which E. and X are plotted as variables. Since P2 Rt, ( U2z~)
the power depends on the magnitude of the (I.60)
reflection coefficient, the power will be constant Pi Re( Uizi)
on a circle of constant magnitude of reHection The circuit efficiency will always be less than
coefficient, or constant p. In Fig.- 5 we showed a unity if there are losses in the network; it will
diagram of these circles, a family of circles sur- be zero if the output load is reactive, and can
rounding the characteristic impedances of the absorb no power; and it is a measure of the
line. The function (I.59) is unity at the right- effectiveness of the network as a carrier of power.
band characteristic impedance, and decreases as Sometimes we are interested in a network as an
we go away from that point, becoming zero attenuator, and in that case we are interested in
along the imaginary axis, or for a reactive load, the amount by which it decreases the power
which of course can absorb no power. In the passing through it. In that case we define an
left-hand half plane, the function is negative; insertion loss, a measure of the decrease of power
such loads have negative resigtivg compont:nts„ in passing. through the network, measured in
 J. C. SLATE R

decibels. That is, we have Writing the output impedance as Z=. R+jX,
and writing Z11 —R11+jX11, etc. , and remember-
Insertion loss = 10 log12 P1/P2
= 10 log12 ing that V2=~2Z, we have
(1/21, ). (I.61)
Ke shall now calculate the circuit efficiency as a Riz, '
function of the terminal impedance Z; from it
the insertion loss can be found from (I.61).
From (I.2) and (I.21)'we have
21C

«l Z» — — —
z+z. ,&
I
fz11'
(I.63)

V1
—t' Z11 — Z12 I lz1
l
E Z+Z22)
z2
—.
z1
= ——
—Z12
Z+Z22
(I 62) Taking the values of the complex
we have
quantities,

( (R12 X12 ) (R+R22) +2R12X12(X+X22)

Z+Z22& (R+R22) '+ (X+X22) '
(R12'+X 12')
(I.64)
l
z1l' (R+R22) '+ (X+X22)'

Substituting these values in (I.63), and combining terms, we finally have

R(R12 +X12 )/R11

QC (I.65)
R12 X12 R12X12 R12 +X12
R+R22 —— + X+X22—
2R11 R11 2R11

While this is a rather formidable expression, it represents a function not greatly diff'erent in its prop-
erties from (1.59). We can show without trouble that the contours of constant circuit efficiency, in
a Z plane, are circles, all orthogonal to the family of circles passing through the two points

R= ~
R12
R22 ——
2R] 1
X12
—2
R122+X
2R11
.
2- 2

X = —X22+ ——.
R12X12

R11
(I.66)

The circuit eAiciency has a maximum at the emphasize that our results How merely from the
point corresponding to the positive sign, and assumption (I.1) or (I.2) that the various volt-
decreases everywhere from there until it is zero ages are linear functions of the corresponding
on the axis of ordinates, or for a reactive load. currents. We have such relations for lumped
We shall have occasion later to consider circuit constant networks, but we shall find as well that
efficiency, in connection with the power output we have such relations for the oscillations of
of magnetrons, klystrons, and other microwave resonant cavities. We shall now proceed to derive
generators, and shall postpone more detailed the properties of wave guides and of resonant
discussion of these formulas until then, when we cavities from Maxwell's equations, and to show
shall put them in simpler form. that we can define quantities analogous to volt-
age and current, which satisfy these same rela-
17'. Resume of Network Theory tions. We shall then be able to apply all the
analysis of this chapter to the results of our
In this chapter we have given a discussion of theory. We shall not be merely using analogies
some phases of network theory; we have omitted with ordinary circuit theory; we shall be deriving
many important points, but have taken up those results directly from the mathematical nature
which we shall most particularly want to use in of our solutions. We turn in the next chapter to
our later work. Before going on, we should wave guides, and show that they form a perfect
 MICROWAVE ELECTRONICS
analogy to the continuous transmission lines of We shall use rationalized m. k. s. units; that is, E
Section 9.
II. WAVE GUIDES
I
is measured in volts per meter, D in coulombs
per square meter, . in ampere-turns per meter,
B in webers per square meter (one weber per
1. The Electromagnetic Field in a Wave Guide square meter = 10' gauss), in amperes per J
By a wave guide we mean a cylindrical pipe, square meter, p in coulombs per cubic meter,
bounded by a conductor (of high conductivity), distances in meters, and times in seconds. The
filled with a dielectric (of low loss). It may have quantities 6p and pp are
arbitrary cross section; the two commonest cases
oo=8.85X10 "farad per meter
in practice are the rectangular and circular guide.
p=4x&(10 ' henry per meter
p,
lt may have more than one bounding surface,
as the coaxial line, which consists of the annular and satisfy the relations
space between two concentric circular cylindrical
conductors. Its object is to transmit electro- (po/eo)' = 3'76. 6 ohms,
magnetic power. We shall first consider the (oooo)
*'
=c =3.00 X 10' meter/sec.
guide without losses, either in the conducting
From Maxwell's equations, in the usual way,
walls or the dielectric, and shall show that in
we can derive the wave equations,
that case a disturbance can be propagated down
the guide, closely analogous to the disturbance 1 O'E 1 O'H
in a lossless continuous transmission line. Our 7'E —— = 0 7'H — — = 0. (I I.6)
c2 l9t2 g2 8
problem is to find E and H, solutions of Max-
well s equations, within the guide, satisfying
We shall first ask under what conditions the
suitable boundary conditions at the surface of
fields (I I. 1) will satisfy the wave equations.
the guide. Within a perfect conductor, no fields,
Substituting, we have at once
electric or magnetic, can exist. Thus, since the
normal component of B and the tangential 8 Eo (co
+ gyo +i ——P'
l9 Eo
component of E must be continuous at a surface, )Eo=O (II.7)
the boundary conditions are that E is normal,
H tangential, to the surface. We shall now show with the same equation for Ho. We shall rewrite
that we can set up solutions of the problem 'of this in the form
the form
E=Eo exp Lj(sit —Ps) j,
j
H = Ho exp L (oot —Ps) j, (II 1)
BoEo BoEo
+ ay' +I — Eo=O
(2~) o
(II.8)
ax' &li, ) f

where Ep, Hp are vector functions of x and y, the

coordinates in the plane of the cross section of where we have
the guide, which themselves satisfy the condition oi (2m'p (2n g (2xP (2T't
that Eo is normal, Ho tangential, to the surface.
To prove this statement, we must show that the co &X, i LX, ] &Z, ) &~, i
expressions (II.1) satisfy Maxwell's equations or

curl E+
8B
=0, B=0,
—= —+- 1
(II.9)
div Ap' X2) g~
Bt
(II.2) Here we have introduced a quantity ) p, which is
curl H— =J, div D=p clearly the free space wave-length, or the wave-
length which a disturbance with angular fre-
quency co would have if propagated in free space.
where we assume that the guide is empty, so that
We have also introduced X„equal to 2m/P. This
within it we have
quantity will be called the guide wave-length;
D = &oE B = poH, J=0, p=0. (II.3) from (II.1) we see that it is the wave-length with
460 J. C. SLATER
which the wave is propagated down the guide. transverse components of E and H, as follows:
The quantity X. will be called the cut-off wave-
length. The reason for this is simple. If the free AXE)
space wave-length is smaller than the cut-off
wave-length, then 1/Xo' is greater than 1/X, 2, so (I I.11)
that (II.9) tells us that 1/X, ' is positive, and X, = — —for T2lI.
(yo) * &O

is real. On the other hand, if the free space f /

wave-length becomes larger than the cut-off

wave-length, 1/X, 2 becomes negative, and the This means, since k and Eg are at right angles to
,
guide wave-length A, is imaginary. Inserting an each other, that Hi is at right angles to Ei in
imaginary wave-length into (II.1), we see that the xy plane, and is equal in magnitude to the
the solution represents an attenuated rather than magnitude of E&, divided by the quantity ZQ.
a propagated wave. This corresponds to the case Using (II.10) and (II.11), we can find all the
of the wave guide beyond cut-off mentioned in components of E and H from H, (in the '. TZ
Section 9, Chapter I, and leads to no transfer of case) or from E, (in the TcV case). These quan-
power down the line. Thus the guide acts like a tities, like the transverse components, satisfy the
high pass filter, only wave-lengths shorter (or wave equation (II.8), and are scalar solutions of
with higher frequency) than the cut-off wave- that differential equation. Furthermore, Z, is
length being propagated. zero on the boundary of the guide (since E must
have no tangential component on the surface);
2. Transverse Electric and Transverse while II, has a vanishing normal derivative on
Magnetic Nodes the boundary (since H must have no normal
The fields E and H must satisfy not only the component on the surface, and H, is proportional
wave equation, but Maxwell's equations as well. to the gradient of H, ). We may draw lines of
When we write these equations down, we find at II.=constant (in the TZ case) or of Z, =constant
once that two separate types of solutions are (in the TM case) in the xy plane. Then by (II.10)
possible: solutions for which Z, = 0, and solutions the orthogonal trajectories-of these lines will be
for which II, =0. The first type has E transverse along the direction of Hi (in the TE case) or of
to the guide, and is called a transverse electric E& (in the TM case). Finally, since by (II.11)

mode (abbreviated TE), and the second has H the direction of Hi is perpendicular to that of
transverse, and is called transverse magnetic E~, the lines of constant H, will be along the
(TcV). We shall let Ei, Hi, refer to the transverse direction of Ei (in the TZ case), and the lines of
components of E and H, and E, and II, be the constant Z, will be along the direction of Hi (in
components along the guide. Furthermore we the TM case). Proceeding in this way, we may
let k be unit vector along the s direction, or draw lines of force, for the transverse components
along the guide. Then from Maxwell's equations of E and H, in the xy plane, finding of course
we find directly that these components must that the electric lines of force meet the surfaces
satisfy the following relations: at right angles, while the magnetic lines of force
are tangential to the surface.
TE: grad JI, =2m j—
)g
H&,
For every scalar solution of the two-dimen-
sional wave equation satisfying the condition
(II.10) that it vanishes on the boundary, we get a TM
j— wave, and for every solution whose normal
A, g
TM: grad E, =2m E~.
derivative vanishes we get a TE wave, as we
have seen above. There will be an infinite num-
In these expressions, II, and E, represent the ber of solutions of each type, each corresponding
magnitudes of the corresponding vectors. Since to a particular cut-off wave-length. These wave-
they are functions of x and y, their gradients are lengths may be arranged in order of decreasing
in the xy plane, as is proper for Hi or Ei. We magnitude; they start with a largest cut-off
find that there is a relationship between the wave-length, associated with the lowest mode of
 M I CROWAVE E LE C TRON I CS

oscillation, and extend inde6nitely toward shorter real wave propagation occurs. For a principal
and shorter wave-lengths, so that we have an mode, the cut-off wave-length is infinite, so that
infinite number of modes of oscillation. For (II.10) tells us that grad H„or grad Z„must
instance, in a rectangular guide of dimensions be zero. That is, the longitudinal components of
a, b, the cut-off wave-lengths are given by the both B and H are zero, and such a wave is
formula simultaneously transverse electric and transverse.
'+ magnetic.
), = ((m/2a) (n/2b) '$ '*, (I I.12) electromagnetic It is sometimes called a transverse
wave (TE3f) for.. this reason.
where m, n are integers. If the dimension a of Furthermore for such a wave, as we see from
the guide is greater than b, the longest cut-off (II.9), the guide wave-length becomes equal to
wave-length is given by m = 1, n =0, and is equal the free space wave-length, so that the velocity
to 2a. In general, it is more convenient to describe of propagation becomes c. Finally, from (II.11),
the modes of a guide by a single index n, which the quantity Zo for a principal wave becomes
we shall take to be the number of the mode (po/eo)~, which can be shown to be the ratio of
when arranged in order of descending wave- the magnitudes of B and H in a plane wave in
length. We shall denote the functions E&, B„H&, free space. Thus a principal wave has many of
H„ZO, X., X„ for the nth mode by an additional the properties of a wave in free space.
subscript n. At any given frequency, or free space wave-
In some cases, the first mode has an infinite length )0, a given wave guide will in general
cut-off wave-length; in this case we call it a support disturbances corresponding to all the
principal mode. We should have such a case in modes. We now see, however, that some of these
(II.12) if m and 5 were both zero, but it turns disturbances will be really propagated, but others
out that this mode does not exist in this case, will be attenuated. In fact, the only disturbances
for the field described by these integers becomes which are propagated will be those for the finite
equal to zero identically. Such a situation does number of modes whose cut-off wave-lengths are
not occur always, however. We find that a greater than the free space wave-length; the
principal mode exists if the wall of the wave remaining infinite number of modes with cut-off
guide consists of at least two separated con- wave-lengths shorter than the free space wave-
ductors, as for instance in the coaxial line. When length will be attenuated. There will be a certain
a mode of infinite cut-off wave-length, or princi- range of free space wave-lengths, between the
pal mode, exists, it has great practical impor- longest and the next longest cut-off wave-lengths,
tance, because it can be used to propagate any in which only the lowest mode, often referred to
wave-length, no matter how long. The commonly as the dominant mode, will be propagated, and
used mode of the coaxial line is a principal mode, wave guides are generally used in this range, so
and the familiar parallel wire transmission line, as to avoid the difficulty of having many modes
ordinarily used for low frequencies, can also be simultaneously present. Guides, in other words,
considered as a wave guide with a principal mode. are ordinarily used only over' a rather limited
It is proved quite generally, on the other hand, range of wave-lengths. A coaxial line, on the
that any wave guide whose wall consists of only contrary, is used in its principal mode, or for
one conductor has no principal mode. The wave-lengths greater than the next cut-off wave-
physical reason for this is quite clear: we can length beyond that of the principal mode. Thus
put a very low frequency, or direct current, into the coaxial line is used as a low pass device, but
a transmission line consisting of two or more the ordinary wave guide as a band pass device.
conductors, and they will be insulated from each
other, and suited to conduct the current. If there 3. Standing Vfaves and Reaction CoeRcients
is only one conductor, however, as in an ordinary In addition to the solution (II.1) of Maxwell's
hollow pipe, there would clearly be a short equations, corresponding to a wave traveling
circuit for a low frequency or direct current, and along the positive s direction, there is of course
no propagation is possible until we get to a a wave traveling along the negative s direction,
wave-length short enough so that something like characterized by the opposite sign for P. Formally
462 J. C. SLATE R
this brings about a change in the sign of ) „and in the wall of the guide, the slot being located at
hence, in (II.10) and (II.11), a change in the sign a point in the cross section where no current
in the relation between E( and H(, and in the must flow, so that the field inside the guide is
relation between II, and H~, or between Z, and not disturbed appreciably by the slot. The probe
E(. If we choose to keep the signs of (II.10) and is connected to a coaxial line or other type of
(I I.11), we must then change signs appropriately line terminated by a power measuring device.
in writing the formulas for E and H. Doing this, The amplitude of the wave set up in this coaxial
and superposing waves traveling in both direc- line is proportional to the component of E along
tions, with appropriate amplitudes, we may the direction of the probe, or transverse to the
write the Geld as guide, and the power absorbed by the power
measuring device is proportional to the square
of the transverse component of E. Thus Et, plays
+E (A e (( e'')—+g e ( (+P )) the same part in finding the power measured in
standing wave measurements with a guide that
H =H(„(A„e'&"' e"*)+8 e'("'+e".)) the voltage does in the transmission line of
—g ej(e(en*)) Chapter I, as discussed in Section 10,
(A eg(~t pn~)
(II 13)
L

4. Impedance and Power Flow

where the relations between the various quanti-
ties are just as in (II.10) and (II.11). By com- In Chapter I, for instance in Eq. (1.37), we
parison for instance with (I.11) we recognize have found the relation between impedance and
the behavior of the transverse components of E reflection coefficient:
and H as being analogous to the voltage and
current in a transmission line. Just as in (I.31)
Z/Zp (1+r) /(1
—— r)— (I I.15)
we can introduce a reflection coefficient This defines the ratio of impedance to character-
istic impedance uniquely in terms of reHection
B„ coefficient. Using this definition, we can intro-
2 jPnz
(II.14) duce a ratio of impedance to characteristic im-
A„
pedance uniquely for a wave guide. We shall
(We must note. that the subscript n here refers refer to this ratio as the reduced impedance. The
to the nth mode of the wave guide, a completely properties of it are of course just as in Chapter
different meaning from that in Chapter I, where I, since it is defined in the same way. There is,
we were using it to refer to the nth set of termi- however, no unique way of defining the char-
nals in a transmission line made of discrete four- acteristic impedance for a wave guide, and hence
terminal networks. ) This reHection coefficient no unique way of defining the impedance itself.
has the same properties as that of Chapter I: From (II.13), we may take the ratio of the
its magnitude represents the ratio of reflected magnitude of E~ to the magnitude of H, (where
amplitude to incident amplitude, and its phase, we mean magnitude in the sense of three-dimen-
as we go along the line, rotates through an sional vectors in space, not in the sense of com-
angle of 2w in a half-wave-length, assuming that plex vectors representing the sinusoidal time
P is real, or that we are dealing with real propa- variation). Using (II.11), we have
gation. If on the other hand we have attenuation
the transformation of going along the line corre-
sponds to a shrinking of all points toward the where Zo now is as de6ned in (II.11). We see, in
origin in the reHection coefficient space. Mathe- other words, that if we choose to interpret that
matically, we can set up standing wave ratios in quantity as a characteristic impedance, the ratio
voltage, in power, and in decibels, as in (I.34), of magnitudes of tangential components of E
(I.35), and (I.36). We must ask how we measure and H will play the part of an impedance. On
these standing wave ratios, however. In the the other hand, if we choose to use any multiple
experimental measurement of standing waves in of the quantity Zo of (II.11) as the characteristic
a guide, a small probe is inserted th~ough a slot impedance, and the corresponding multiple of the
 M IC ROWAUE ELECTRON I CS 463

ratio of E& to as impedance, we shall equally

H& We have so far left the values of the integrals of
well have agreement with (II.15). We shall use ~
E)„~' or H, „~ ' over the guide arbitrary; obvi-
~

the quantity (II.11) as a characteristic imped- ously in (II.13) we can multiply the quantities
ance; for as we shall show shortly, there is no Z&„and H& by an arbitrary constant, and divide
other definition of characteristic impedance of a A„and B„by the same constant, without
wave guide which is more universally sensible. changing B and H, which alone have physical
For comparison with the results of Chapter I, significance. The value which we choose for these
we should consider not merely the impedance, constants is purely a matter of convenience.
but the power flow as well. This is of course Since we have already decided to make Z0
uniquely determined, since it can be directly analogous to a characteristic impedance, com-
measured, by terminating the guide by a power parison with (I.55) suggests that we make the
measuring device, as a bolometer or thermistor A„'s and 8„'s analogous to the A's and 8's
or water load, whose temperature rise indicates introduced in Chapter I, which by (I.53) are
the total power absorbed by it, and by assuming current amplitudes. To accomplish this, we may
that all power flowing down the guide is absorbed assume
in the power measurer. Mathematically, we can
find the power flow by integrating the normal
(II.20)
component of Poynting's vector over a cross so that
section of the guide. Poynting's vector is EXH, P= ', Zp (A„A —
—B„B„). (I I.21)
and its normal component, or s component, is
k (EXH). The time average is easily seen to be It is now clear that our values of impedance
computed as the time average power was in and power flow in a wave guide are analogous to
Chapter I, Section 14: it is Rek (EXH). In
—', the corresponding quantities for a transmission
Poynting's vector we encounter the vector quan- line, if we determine the magnitude of H&„by
tity k (E~XH~): the other quantities k (E, XH)), (II.20) (called a normalization condition), and
k. (E~XH, ), and k (E.XH, ) are all automati- if we postulate a voltage V„and current i„
cally zero. We notice that as a result of (II.11) given by
we have V„=Zo„A„e&'"' &n' —8 e~("'+~n
(I I.22)
~ j(&o5 Pn~) +B &j(iat+Pn~)—
ZQ
so that V„ is proportional to the transverse E,
Using these relations, we may then write the f„ to the transverse H, in the nth mode. Since
time average Poynting's vector, 5, in the form these equations are entirely analogous to those
appropriate for the case of real propagation, of Chapter I, we have mathematica11y justified
the results of that chapter, as applied to the
S=— —
ZQn
(A „A„BB„)— wave
can
guides. By entirely similar methods, we
justify. formulas like (1.57) with correspond-
ing interpretation of voltage and current, for the
= -', Zo„i H, '(A„A„B„B„). (II.1—
i
8) case where the guide is beyond cut-off, so that
there is only attenuation, not propagation, and
To get the total power fiowing through the guide,
where the characteristic impedance is pure
we must integrate this quantity over the area of
imaginary.
the guide, so that we have
While we have suggested a particular way of
setting up a voltage and current, this is by no
means the only possible way. In fact, it is
P=— (A „A„B„B„), .
— obvious that we can assume that the voltage is

)—
2 Zon any constant times the transverse E, and the
(II.19) current any constant times the transverse H.
= —Zp„' iH)„i'da(A„A B„B This gives two arbitrary constants in the inter-
2 pretation of the behavior of the guide as a
transmission line. On the other hand, if the power excited. The discussion is made possible by the
is to be determined by the relation P = — 'Re
, Ui, proof of several theorems related to the orthogo-
this imposes a relation between the definitions nality of the normal modes. We shall .first prove
of voltage and currents, so that only one arbi- these theorems. They are as follows:
trary constant is left. We may still use this (I) The integral over the cross section of the
arbitrary constant to make the characteristic guide of the scalar product E~„E~, or H&„H&,
impedance, or the voltage, or the current, any- where n and m are different, is zero.
thing we please, but if one of these quantities is (II) The integral over the cross section of the
determined, the others are also. In a few cases guide of the product B,„B, , or H, „II, , where
there is an obvious way to define voltage and n and re are different, is zero.
current. For instance, in a coaxial line, or other (III) The integral over the cross section of the
line possessing a principal mode, the voltage is guide of the quantity Ir (Ei~ XHi„), where n and
uniquely defined: for in that case (11.7) shows m are different, is zero.
that the transverse E obeys Laplace's equation,
The proofs follow easily from two-dimensional
so that its integral from one conductor to the
forms of Green's theorem. We first use Green's
other, being independent of path, forms a unique
theorem in the form
voltage. Similarly there is a unique current, the
actual current Howing in either conductor. The
voltage and current so defined do not agree with I (gV'/+grad p grad f)« = p —ds, (II.23)
an
our value (II.22). Again, in the lowest mode of
a rectangular wave guide, reasonable definitions where the integral on the left is over an area (in
of voltage and current can be given. These this case the area of the cross section), the inte-
definitions again do not agree with (II.22), but gral on the right. is over the perimeter, and 8$/Bn
neither do they have a simple relation to those is the normal derivative in the direction of the
used for coaxial lines. In fact, it seems to be outer normal to the area, and where p, f, are two
impossible to set up any general definition of scalar functions of position. We let p be E, , P
voltage and current which reduces in a reasonable be E, , and remember that on account of (II.8)
way to the natural values met in these simple we have
cases. For that reason we adopt our definitions „+
V'E, (2ir/X, „)'E, 0. „= (I I.24)
(II.22), which are the simplest ones. It is a
fortunate fact that this ambiguity in the defini- Thus we have, using (II.10),
tion of current, voltage, and -impedance really 2s )' r zg.xg
—E&„E& da
E,~E,„da 4''
.
I
does not affect us at all; for the quantities which
actually are important are the ratio of impedance Ey, ) a
to characteristic impedance, or reduced imped-
ance, and the power How, which are uniquely E,„(n grad E, )ds. (II.25)
determined, quite apart from this ambiguity.
Since E,„=o on the perimeter, the line integral
S. Expansion of the Field in Normal Modes on the right is zero, and we thus see that-. if
So far, we have assumed that the field consists J'E,„E, «=O, J'E~„.Ei da=O as well. That is,
of a single normal mode only, but of course on we show that theorems (I) and (II) are equiva-
account of the linear nature of Maxwell's equa- lent, as far as E is concerned. To show the same
tions the general solution of our problem is a thing for the H's, we proceed in the same way
superposition of all normal modes, each with its but now the line integral in (II.25) vanishes
appropriate amplitude and phase; that is, E and because grad II, is parallel to the surface, since
H are given by summations over n of the quan- H~ has no component normal to the surface.
tities given in (II.13), rather than just the nth Next we use Green's theorem in the form
term. In the present section we sha11 take up
those properties of the field that depend on the
fact that ordinarily all modes are simultaneously
 MiCROWWVE ELECTRONICS
and let p=E, , Q=Z. T. he right-hand side boundary conditions. We may well ask, how
vanishes as before, and the left-hand side gives much information about the field in a guide is
necessary to determine it uniquely? The answer
) 1 1

x, .
') ~ B,„B,
dc=0, (II.27)

so that, if n and m are different, and the first fac-

is simple: we must know the tangential compo-
nents of E and H as functions of x and y across
a single cross section. We shall prove this by
setting up the expansion coef6cients uniquely in
tor is not zero, we have J'E,„Z,„da =0, proving
terms of that information. For the sake of
theorem (II), and hence theorem (I). The corre-
simplicity, we shall choose the plane on which
sponding proof for the H's follows similarly. To
we know the tangential values of E and H as the
prove (III), we need only note that, by (II.11),
plane s= 0. In that plane, then, using (II. 13),
Hi~ = (k XEi~)/Zo~, from which we show at once
the tangential components of E and H are
that
k (E&„XH& ) =(E&„'E& )/Zo, (II.28) E,e&"' = P„E&„(A„+B„)e&
(I I.29)
so that (III) reduces to (I). H, e&"' = P„H,„(A„—
B„)e&'"'.
We shall first use these theorems to prove that
when we consider the flow of energy, or the total In this expression, Ei and Hi are assumed to be
energy in the guide, we may handle the various known functions of x and y, the factors e&"'
modes separately, the total energy flow or energy expressing their variation with time. If the field
being the sum of the corresponding terms for the is not sinusoidal with time, we first make a
various modes, without cross terms. For the Fourier integral analysis in time, and apply our
energy flow, we must compute the integral over discussion to a single sinusoidal component. We
the cross section of the quantity Rek (EXH),
—', now take the first equation of (II.29), take its
where E and H are given by summations over n scalar product with one of the E& 's, and inte-
of the terms as in (II.13). This will involve grate ever the cross section of the guide. Using
terms k (E,~XH&~), whose integrals over the our theorem (I), all terms except the mth will
cross section are zero, by (III), if n and m are drop out. We may then use the normalization
different. The only non-vanishing terms are then condition
those for e = m, and these a,re simply the terms
computed for the various modes separately, as (I I.30)
in Section 4. Thus the total flow of power is the
sum of the flows of the various modes. Similarly
the total energy is the integral over the volume which follows from (II.17) and (I-I.20), and find
of 4iRe(ooE E+ poH. H). The quantity E E or
H H is again the product of two sums from Ei Ei„dpi
~

(II.13), and we again write it as a double sum.

The terms involved are of the form E~„E~, =A„+B„. (I I.31)
E,„E, , H&„.H&, or H, „H, , all of which inte- ZOn

grate to zero over the cross section of the guide, Similarly we multiply the second equation of
as we see from theorems (I) and (II), if n and m (II.29) by one of the H&~'s and integrate. Using
are different. The only remaining terms again the normalization condition (II.20), we have
are those for n =nz, which are the terms relating
to the individual modes, so that the total energy
is the sum of the energies of the modes, without H, H]„du=A„—8„. (II.32)
cross terms. The superposition of modes, in other
words, brings about no complications when we From (II.31) and (II.32) we can find A„and B„
consider energy and its flow. in terms of integrals of the known functions E&
The more interesting application of our theo- and Hi. Thus we can set up the summation of
rems I to III comes in setting up the expansion terms (II.13), and hence the complete field within
of the field in the guide, subject to certain the guide, showing that the tangential compo-
 J. C. SLATE R

nent of field over a single cross section deter- with the boundary conditions (I I.34) at the
mines the complete field. surface of the guide, the solutions for E and H
which we have found in the preceding sections
6. Losses in the Wave Guide of this chapter are not correct. However, when
In our treatment so far, we have assumed that we insert numerical values for ordinary metallic
the walls of the wave guide were perfect con- conductors, we find that the tangential E re-
ductors, so that E had to be normal to the walls. quired at the surface is so small that the correct
In that case there is no power fIow into the walls, field is a very small perturbation of the field we
and the guide is a lossless transmission line. If have calculated earlier. Thus we are justified to
the walls have a finite conductivity, however, a first approximation in assuming that the tan-
power will be dissipated in them, and the guide gential IIwhich we have found is that actually
will show attenuation. From the theory of the present, and that it can be used in computing
skin effect, which we shall not go into, * we find the energy loss (II.35).
that a disturbance of angular frequency co, in a We may now, by simple calculation of power,
good conductor of conductivity o-, with the same find an approximate value of the attenuation
magnetic permeability p, o as free space, penetrates constant n„which is present in the nth mode, on
only a short distance into the conductor; the account of the losses in the walls. In an attenu-
amplitude of both electric and magnetic fields ated wave, traveling along the positive s direc-
falls to 1/e of the value on the surface in a tion, the magnetic field H, by analogy with
distance 8 equal to (II.13), will be
8= (2/gypsy)' (I I.33) H = (H(„+H,„)A„e'~"' &"'e ~"* (II.36)
We notice that' as the conductivity becomes By analogy with (II.19), the power flow will be
infinite, the distance 6, sometimes called the skin
depth, becomes zero, so that we approach the P=-,'ZO„A„A„e ' "') (II.37)
case we have treated earlier. We also find that
in which we have also used (II.20). The loss of
there is a small tangential component of E at
power into the walls, in unit length of the guide,
the surface, proportional to the tangential com-
will be —(dP/ds). By (II.35), this will be
ponent of H, which measures the surface current,
and at right angles to it, or in the same direction dP 1 (pp&u) '
If n is the outer normal
as the surface current. (gapa&/2o) LIII~ I'+ l~ I'1d~
to the guide, we find that the tangential compo-
ds 2&2~) &

nent of E at the surface is given in terms of the X A„A„e—'~"' (I I.38)

tangential H by the equation
in which the line integral is to be taken around
E = (H X n) l(1+j), (I I.34) the perimeter of the cross section of the guide.
so that as the conductivity becomes infinite, the At the same time, using (II.37), we have
tangential E goes to zero. As a result of (II.34), dP/dh = 2n„P. (I I.39)
there is a How of energy into the surface, which
we can find by computing the normal component Using (II.37), (II.38), and (II.39), we then find
of Poynting's. vector. The time average is = p (1/~p
~ ) (u p pp/2 &) '
'Ren. (EXH) = &(ppcv/20)'lHl'. (II 35)
—,

For a given tangential H, this How of energy x L l

~.l'+ l
Il. .
l']&~. (I1.40)
goes to zero as the conductivity becomes infinite.
We can now use this discussion to give an This gives us a formula for a in terms of an
approximate treatment of the effect of surface integral of the square of the tangential compo-
losses on the behavior of a wave guide. Clearly, nent of H„around the surface of the guide.
* See Slater, 3IIicromave Transmission, This, it should be repeated, is correct only if we
Section 12, for a
detailed discussion. assume that the tangential H in the presence of
 M I CROWA VE ELECTRON I CS

absorption is almost equal to that in the lossless formers and transducers, oscillators, and in fact
case, for which it is assumed that it is calculated. all types of microwave problems.
We shall not give detailed examples of the Our first problem will be to solve Maxwell's
application of this formula, but such examples equations in a hollow cavity, subject to certain
are worked out in the various texts on micro- boundary conditions around the surface. As with
waves. The one fact is obvious, that a guide in the problem of the wave guide which we have
which the tangential II (and therefore the surface taken up in Chapter II, this solution will be in
current) rises to high values at some points in terms of a summation over certain normal modes,
the metallic surface will have high losses. which possess orthogonality properties. The de-
In thi:s calculation of n„ from the power flow, tails of the process are quite different from the
we have assumed that one mode only was case of Chapter II, however, and we shall start
excited, and that there was only a direct wave, from the beginning with our discussion. We shall
not a reflected wave. Ke may not assume here, start by postulating the properties of the normal
however, that if we have a number of modes functions, shall expand the electric and magnetic
coexisting, the losses are simply a sum of the fields in terms of them, and shall then find what=
losses in the various modes. The integral of the conditions must be satisfied to solve Maxwell's
square of the tangential H over the-surface of equations. We wish to solve Maxwell's equations
the guide, which we And in (II.40), has no within a volume bounded by a certain surface
orthogonality property, and there is no way of (nothing in our treatment will prevent this
disentangling the effects of the various modes. surface consisting of several parts, as the inner
The losses, being a quadratic rather than a and outer surface of a hollow spherical shell).
linear function of the amplitudes, have no princi- We shall find that we can set up orthogonal
ple of superposibility, and the presence of one functions for two types of boundary conditions:
mode can affect the losses experienced by another the 6rst, which we may call short circuited
mode propagated through the same guide with boundary conditions, requires that the tangential
the same frequency. The detailed study of this component of E, and the normal component of H,
situation would be necessary to find the behavior be zero on the surface, while the second, which
in any particular case. we call open circuited boundary conditions, re-
quires that the normal component of E, and the
III. RESONANT CAVITIES tangential component of H, be zero on the
surface. The reason for the names is simple: a
I. Orthogonal Functions for a Hollow Cavity
perfect conductor has zero tangential component
Just as a wave guide forms the microwave of E, and forms the analog of a short circuit;
analogy for the transmission line of ordinary while a perfect insulator carries no surface cur-
circuit theory, so a hollow cavity forms the rent, and hence, if H is zero within it, it demands
analogy for a circuit element. A cavity can be a zero tangential component of H, and forms the
provided with one or more output leads, in the analog of an open circuit. We shall discuss these
form of wave guides of some type: rectangular points more in detail later. We shall find that
guides, coaxial lines, etc. If it has only one lead, we can use mixed boundary conditions, and that
it serves as an impedance terminating that lead; on occasion we shall want to: over part of the
if it has two, it serves as a transformer or surface (which we denote by S) we shall have
transducer, allowing powder to flow into one short circuited boundary conditions, while over
lead, out the other. In the present chapter we the rest of the surface (which we denote by 5')
take up the general theory uf resonant cavities we shall have open circuited boundary condi-
and the electromagnetic fields within them. We tions. Our object is now to set up orthogonal
consider the case of an arbitrary number of functions within the volume V bounded by S
output leads, and find the relation between the and S', suitable for expanding our fields within
electromagnetic fields in these various leads, each the volume.
of the form taken up in Chapter I. We shall be Our first step is to notice that by general
able to use these results later in discussing trans- principles of vector analysis, any vector field can
 J. C. SLATE R
be broken up into two fields, one of which is component of E around this contour, which is
solenoidal, or has zero divergence, and the other zero since the tangential component of E is zero
of which is irrotational, or has zero curl. We on S according to (III.2). By Stokes' theorem
shall consequently. set up two sets of orthogonal this line integral equals the surface integral of
functions, one set solenoidal, the other irrota- the normal component of curl E„or
of k, H,
tional, using the solenoidal functions to expand which is then zero, which is impossible, since we
the solenoidal part of any vector field, and the are dealing with an arbitrary contour, unless the
irrota. tional functions to expand the irrotational normal component of H, n-H„ is zero on 5.
part of a field. As a matter of fact, we go further: The second statement of (III.3) is proved in a
we set up two independent sets of solenoidal similar way.
functions, one adapted for expanding the sole- We may easily set up separate differential
noidal part of E, the other adapted for expanding equations for E and H, instead of having them
H (which is itself solenoidal, if we assume, as we defined in terms of each other as in (III.1). We
shall, that the magnetic permeability po is a see at once that these equations are
'constant). We use only one type of irrotational
function, used for expanding the irrotational curl curl E, = k, 'E„curl curl H, = k, 'H, . (II I.4)
part of E, and do not need another type only Using the vector identity that curl curl A
because H has no irrotational part. We shall =grad div A —V'A, and that div E =0, div H,
denote the solenoidal functions used in expanding = 0, these become the familiar wave equation
E by the symbol E, and the solenoidal functions
used in expanding H by the symbol H . Similarly V'E. +k,'E, = 0, V'H, +k 'H, = 0. (III.5)
we shall denote the irrotational functions used
in expanding E by the symbol F,. We shall now These equations may be assumed to have an
set up the equations used in defining these infinite set of solutions, corresponding to different
functions. values of k„subject to the boundary conditions
The functions E and H, having no divergence, (III.2). It is clear that corresponding to each k,
must be the curls of certain other vector func- we have both a function E, and an H, ; both
tions, and we assume that they satisfy the sets of functions correspond to the same set of
equations characteristic numbers.
We shall now prove that the functions E, and
k, E, = curl H„k,H, = curl E„(III.1) H have orthogonality properties of the form

where k is a constant, which will later prove to ~ E. Etdv=0 if a~b,

be the propagation constant (2ir divided by the "v
wave-length) associated with the uth mode. We (III.6)
assume that E and H, furthermore satisfy the H Hbde =0 if a/b.
following boundary conditions:

nXE, =O on S, nXH, =O on S' (III.2) To prove the first, we use the vector identity

where n is. the outer normal to the surface; that div (E|,Xcurl E,) —div (E~Xcurl Et, )
is, E has no tangential component over S, and =curl E curl Eb —Eb. curl curl E,
H has no tangential component over S'. From
—curl Et, curl E +E, curl curl Ei, . (ill. 7)
(III.1) and (III.2), and Stokes' theorem, we can Cancelling terms, and using (I I I.4), we can
then prove that rewrite (III.7) in the form
n H, =O on S, n E, =O on S'. (III.3) div (Eb X curl E,) —div (E~ X curl Et, )
= (kP —k, ')E, Et, . (III;8)
To prove the first of these, we take a small
closed curve lying in the plane of the surface 5, Integrating over the volume V, and using
and integrate the line integral of the tangential Green's theorem to convert the left side into a
 M I CROWAVE ELECTRON I CS
surface integral, we have at once that Ji H'dv=tyZ, that each
'dv, so
can be set equal to unity.
n (k.E»XH. —k»E. XH )da We shall next set up our functions F„which
have zero curl. On account of that property,
they can be set equal to the gradient of a scalar
= (k, 2 —k 2) E E»dv. (I I I.9) function. We write
Jy kP', =grad P,. (I I I.13)
We may now easily show that the surface integral We assume that the scalar P, satisfies the wave
on the left vanishes on account of the boundary equation
conditions. Over S, we may rewrite the integrand 'PP, +k '$, =0, (III.14)
in the form k,H, (nXE») — k»H» (nXE.), which
is zero on account of (III.2), while over 5' we from which we immediately see that F also
satisfies the wave equation. As boundary condi-
may rewrite it in the form k»E, (n XH»)
—k,E» (nXH. ), which is likewise zero. Thus, if tions we assume
k~' — k ' is different from zero, which will be the $, =0 on Sand 5',
case if a&b except in case of degeneracy, we have nXF =0 on S and S'. (I I I.15)
These conditions, in which the second obviously
E, Egdv=0 if a&b, (I I I.10) follows from the first, are not the most general
boundary conditions which can be applied to
these functions, but they will be suf6cient for
which we wished to prove. In the case of de- our purposes.
generacy we can prove, as in quantum mechanics, We can now prove that the functions F and
that we can always introduce normal functions P, have orthogonality properties of the form
E and E& in such a way as to secure the orthogo-
nality we desire, though it is no longer necessary F, .Fgdv =0 if a~b,
that the functions have that property. The proof ~v
of the orthogonality of the H 's, stated in (II I.16)
(III.6), follows in an entirely analogous manner. P,P»dv=0 if a&b.
In addition to the orthogonality, we shall ~v
assume that the E 's and H 's are normalized in
such a way that J'B,'dv and J'H, 'dv equal unity, To prove this, we have
so that the normalization and orthogonality div (f. grad P»)
conditions can be written in the form =f 7'P»+grad f, grad f»
= —k»'f, f»+grad f, grad P» (III.17.)
E, Ebdv=b g, H, H»dv = 8,», (III.11) Interchanging the order of a and b, we set up
4y
another equation like (I I I. 17), and subtract.
where as usual h, ~ is unity if a=b, zero if aWb. We then integrate over V, obtaining
Since E, and H are related by (III.1), we may
not simultaneously assume that E and H are
normalized in this way, without proof that the an an)
two conditions are consistent with each other.
To prove this consistency, we have
div (E, Xcurl E,)
= (curl E )' —E, curl curl E, The surface integral vanishes on account of the
= k, '(H~' B,') . — condition (III.15), so that, if k. Wk», the volume
integral must vanish, leading to the orthogonality
Integrating over V, the left side again transforms condition (III.16) for the f's If we integra. te
into a surface integral which vanishes, showing (III.17) itself over V, the surface integral again
470 J. C. SLATE R
1

vanishes, so that, if Jr
p, fbdv=0, we must also function which we wish to expand has boundary
have Ji grad p, .grad fbdv =0, so that we prove conditions which more closely resemble the
the orthogonality of the F's. Finally we assume boundary conditions satisfied by the E,'s than
normalization of the form Jr
I','dv P,'dv=1, =Jr by the H 's, then its expansion in terms of the
so that we can write the normalization and E,'s will converge better than the expansion in
orthogonality in the form terms of the H 's, and vice versa. This is analo-
gous to the situation according to which a func-
= tion in the range 0 to m can be expanded either
F Fbdv p, pbdv = 0 b (III.19)
v ~v in a sine or a cosine series of period 2m, but if
the function we are expanding satisfies the same
To prove the consistency of these two conditions, boundary conditions as the sines (function zero
we take (III.17), set a = b, and integrate over V. at 0 and ~), the expansion in sines will converge
Using (III.13), we find at once that Ji P,'dv better than that in cosines, in the sense that the
= Jv F,'dv, so that we are justified in setting derivative of the sine series will also converge,
them each equal to unity. while the derivative of the cosine series will
As a final step in setting up our orthogonal diverge. The other point which we wish to make
functions, we prove that one of the F,'s is about the expansion is that the characteristic
orthogonal to one of the E,'s: numbers k, for the solenoidal functions are not
the same as for the irrotational functions. In
F~'Eydv = 0. (II I.20) each case, for instance in the summations over
6 p. a which we shall soon set up, we are to under-
stand that we use the k appropriate for the type
To prove this, we note that of function under consideration.
div (P,Eb) = f, div Eb+grad f, Eb If we may expand an arbitrary function, say
= kZ Eb, (I I I.21) A, in terms of the E,'s and F,'s (or the H, 's and
F 's), it is then an easy matter from the orthogo-
since div E~ —0. Integrating over V, the surface nality and normalization conditions to find the
integral vanishes, proving our result (I I I.20). expansion coefficients. Let us assume that
We have now set up two orthogonal families
of solenoidal functions, the E 's and H, 's, and
A=A, +A„ (I I I.22)
one family of irrotational functions, the F,'s. It where A~ is the solenoidal, A2 the irrotational,
seems intuitively reasonable to suppose that the part of A. Then A~ can be expanded in series in
E 's and the F 's, or the H, 's and the F 's, form the E,'s (or the H, 's), and A2 in the F,'s:
complete sets of functions, in the mathematical
A = Pa eaEa+ ga faFa~ (III.23)
'sense, such that any artibrary vector function
of position within V, satisfying certain not very where the e, 's and f,'s are coeKcients, and where
stringent conditions of continuity, must be the first summation is over the normal functions
capable of being expanded in a series in the of the solenoidal, the second of the irrotational,
functions. It is a problem for the mathematician type. If we multiply A by one of the E 's, and
to prove rigorously the completeness of these integrate over V, then on account of the orthogo-
sets of functions, and we shall not attempt it. nality relations the integral of the product of
There are two points which are clear about this this E with every other term of the summation
expansion. In the first place, we certainly do not (II I.23) except itself will be zero, and the integral
need both the E,'s and the H 's for any given of its square will he unity on account of normal-
expansion; an E and an H~ are not orthogonal ization. Thus we have
to each other, but on the contrary one of the
E's can be expanded in series in the H's, or
vice versa We choose .either the E 's or the H, 's,
e, =) A E.dv. (I II.24)
in any given case, according to convenience. The
general situation to be expected is that if the There will be a similar result for f,. Hence we
 M I CROWAUE ELECTRON I CS
may rewrite (III.23) in the form method. We have

A=+, E,
v»v
r
,
A. E.dv+F, ,

There will be an analogous formula in terms of

A F,dv. (III.25)
div (E Xcurl E)
=curl E.curl E, —E-curl curl E,
=k H, curl E —k, 'E. E,. (III.28)
Integrating over V, and transforming the term
the H 's. Our method of expansion, which we on the left into a surface integral,
take up in the next section, is based on this
result. In expressing the volume integrals in the
future, we shall omit the subscript V for sim-
n (EXH.)da
plicity, assuming that all volume integrals are
~ver V unless it is stated to the contrary. curl E H, dv —k, E.K,dv. (III.29)
2. Maxwe11's Equations in a Hollow Cavity
In the surface integral, we may write the inte-
Our object in the present section will be to grand in the form H, (n XE) or E (H, Xn). The
expand the electric and magnetic fields, and second form shows that it vanishes over S'.
other related quantities, inside a hollow cavity Then, substituting in (III.27), we have
in terms of the orthogonal functions set up in
the preceding section, and by means of Maxwell's
equations to find relations between the expansion
curl E=Q, H, ~
k, I E.E.dv
coefficients. In Maxwell's equations, as given in
(II.2) and (II.3), we shall expand E in series in + (nXE). H da ~. (III.30)
the E,'s and F,'s; H in series in the H, 's; curl E
in the H, 's; curl H in the E,'s; J in the E,'s and
f,
F,'s; div D and p in the 's Thus we. have In other words, in addition to the sum of terms
involving volume integrals, which we got by
r taking the curl of the series for E, we also have
E» ) a sum of terms involving surface integrals. These
integrals will be zero if E is normal to the surface
S, but we shall later find cases where E has a
H=Q H H Hdv,
tangential component over S, so that the surface
(III.26) integral does not vanish. We shall shortly con-
sider the significanc of these surface integrals.
J=P. ~
E. J E.dv+F. J F.dv ~, In a corresponding way we find that

p = Qa 4'a p4'adv curl H=P. E. ~

k. H. H.dv
)
The other expansions take a little more thought.
For curl E, we might think at first sight that we
could take the expansion (III.26) for E, and take
Similarly for div D, using the relation
its curl directly, obtaining Pk, H, J'E E,dv. On
the other hand, we might consider the function div g, D) = f, div D+ D grad P,
curl E, and expand it directly in terms of the =f. div D+k, D F„(III.32)
functions H, . We then have we have

div D=Q, P, —k D F,dv

E=Q
~

curl H curl E H, dv. (III.27)

(D n)P.da I. (III.33)
We may evaluate the integral by the following
 J. C. SLATE R
The surface integral in (II I.33) vanishes because (III.38), at once leads to
P, is zero on S and 5'.
We have now set up the series for.the various div Jf.dv = —
dt ~
pf. dv. (I I I.40)
quantities concerned in Maxwell's equations. We
next substitute these series, and equate coefh- But (III.40) is true on account of the equation
cients, so as to get the differential equations of continuity,
satisfied by the various coe%cients. From curl E div J+Bp/Bt =0. (III.41)
+BB/Bt =0 we have
Thus we need use only (III.36) or (III.37), not
both.
k, ~K. E,dv+ po —H H, dv In discussing Maxwell's equations, we have
~ dt
two very distinct problems. First there are the
Eqs. (III.34) and (III.35). These will concern us
(n X E) H.da. (I I I.34) most in the future. They are the equations
"s determining the solenoidal part of E, and'H. It
is this part of the field that shows properties of
From curl H —D = J we have
wave propagation, and that is usually regarded
as the radiation field. These equations determine
d
k. "H.H.dv —eo— E E.dv
dt ~
the coefficients J'E. E,dv and f
H H, dv as func-
tions of time; they are determined in terms of
the integrals which appear on the right side
J E.dv —, (n XH) E,da, (III.35) of the equations. The quantity J' J'E dv
—J8 (nXH) E,da appearing on the right side
of (III.35) is easy to interpret. It is one of the
d
—co— E F,dv= 3J F dv. (III.36)
i
components of J (the volume integral), supple-
dt~ mented by the surface integra1. In this surface
integral —(nXH) is the tangential component
The equation div B = 0 is automatically satisfied. of surface current connected with the discon-
The equation div D = p gives tinuity in the tangential component of H at the
surface, so that the surface integral gives us the
contribution of surface currents. Similarly the
—k, eo ~E F.dv = f pP, dv. (I II.37) quantity —J's (nXE) H, da would have to be
~
interpreted as the integral of —(n XE), a surface
We may show from the equation of continuity density of fictitious magnetic current, which
that (III.36) and (II I.37) are equivalent. Taking appears at a .surface of discontinuity of the
the time derivative of (III.37), and multiplying tangential component of E. Thus we have the
(III.36) by k„ the left sides of the equations are interpretation of the surface integrals appearing
in (III.30) and (III.31).
identical. The two equations will then lead to
the same result if we have We can easily combine (II I.34) and (III.35), to
get separate equations for J'E E.dv, J'H H dv,
d as in the conventional derivation of the wave
k. J F.dv = —
dt
pP.dv. (III.38) equation. Thus we have

The integral on the left may be rewritten by

aopp — E Egdv+kg
dt~ ~
' E 'Egdv
using the relation
d(
= —po — p
div (P,J) = f, div J+J grad P, . (I II.39) dt E~
(
'

J E dv (nXH) 'E+dg

Integrating (III.39) over V, the surface integral —k, (nXE) H.da, (III.42)
,

is zero, and the volume integral, substituted in

 M IC Rog/A V E ELEC I'RON I CS

ootio —
dt'~
H H.dv+k, ' H
J
H„dv
is taken in the problem of retardation, or of
finite velocity of propagation of the disturbance.

3. Free and Damped OsciHations of a

=k, ( J E.dv — (nXH) E.da
~

Resonant Cavity
~

~s, ) ~

As a first example of the use of Eqs. (III.42)

—oo— (nXE) H, d~t. (III.43) and (III.43), let us consider the free oscillations
dt ~s of a resonant cavity. We assume that the cavity
contains no current density, so that J = 0.
These two equations will form the basis of most Furthermore we assume that over the part of
of our later treatment of resonant cavities. They the surface S the tangential component of E is
show that J'E. E,dv and J'H H, dv are deter- zero (which would be the case if that part of the
mined as functions of time by the type of differ- surface were formed of a perfect conductor, or
ential equations encountered in elementary prob- formed a short circuit) and that over S' the
lems of simple harmonic motion. The terms on tangential component of H is zero (which would
the right-hand sides take the place of the external be the case if that part of the surface were
force in the problem of simple harmonic motion, formed of a perfect insulator, in which no surface
so that we may get solutions of (III.42) and current could flow, so that it formed an open
(III.43) showing the properties of forced oscilla- circuit). Then all integrals on the right side of
tions and resonance, the forced motion resulting (III.42) and (III.43) would be zero, and the
from the currents within the cavity, or from equations would have solutions
disturbances propagated from the walls. We
shall take up these motions in later sections. E E dv =constant s'"~' =0 (III.45)
oi o ti
The remaining one of Eqs. (III.36) or (III.37)
refers to the irrotationa1 part of E, the part
whose curl is zero, or which is derivable from a with similar solutions for J'H H, dv. Thus the
scalar potential. To understand its interpreta- ~, 's are the angular frequencies of the resonant
tion, let us go back to Maxwell's equations, modes, and the general solution of the problem
writing the vectors E and J as sums of a sole- of free oscillations would be a superposition 6f
noidal part (Ei and Ji) and an irrotational part the various normal modes, each oscillating with
(Eo and Jo). We have
arbitrary amplitude at its resonant frequency.
From Maxwell's Eqs. (III.34) and (III.35) we
curl Ei+B =0, div 8=0, can find the relation between the coefficients
curl H —Di —Ji, div Pi —0, J'E E,dv and J'H H, dv for the normal modes.
—Do= Jo,
4

dlv D2=p By substituting, and using (III.45), we find

easily that
in which the last two equations are equivalent
as a result of the equation of continuity. Re- E E,dv
membering that curl K~=0, we may write
i (p, o/oo) '— (II I.46)
Eo = —grad g, Pg = —p joo. (I I I.44) H. H, dv
That is, E2 is derivable from a scalar potential,
which satisfies Poisson's equation. The problem That is, the magnitudes of the coefficients
of finding E2 is then identical with the electro-
static problem of finding the field of a known
fE E dv and J'H H, dv are in the same ratios
to each other as the values of E and H in a plane
distribution of charge, subject to a condition wave in empty space, but the electric and mag-
that the potential is zero on the boundary. The netic fields are 90' apart in phase, a character-
only difference between this and an electrostatic istic of standing electromagnetic waves. We
problem is that the charge distribution, and readily find that as a result of (III.46) the time
hence the field, varies with time; but no account average electrical energy, integrated through the
 J. C. SLATE R

cavity, equals the time average magrietic energy. The decrease in energy per unit time is pip/Q
The phase difference, however, results in the times the energy, or we may write
magnetic energy being large when the electric
2x g total energy
energy is small, and vice verse, just as with the
(I I I.51)
kinetic and potential energy in simple harmonic decrease of energy per period
motion, with the result that the total energy
remains constant. where the period concerned is 2m. /&op, as deter-
The solution for free oscillation which we have mined from the angular frequency which would
just found is analogous to the free oscillation of exist if the damping were absent.
a simple L-C series circuit. 9/e shall now look The description of a rate of damping of a
for the analog to damped oscillation, which circuit by means of a Q is one which is con-
occurs when the circuit contains resistance as venient in microwave work as well as with ordi-
well as inductance and capacity. The equation nary oscillating circuits. We shall adopt for Q a
for a series circuit containing inductance I, definition which is equivalent to (III.48): an
resistance R, and capacity C is oscillation whose angular frequency is determined
by the equation
I. +R—
dg
+—=0,
Q dg
(I I I.47)
dt dt C (III.52)
if we use the charge g on the condenser as the
will be referred to as having a given Q. This
variable. If we assume an exponential solution,
definition is equivalent to that of (III.51). In
q varying as e&"', this becomes
most of our applications, Q will be large enough
so that the distinction between &up and the cor-
rected angular frequency a&pt'1 —(1/2Q)']l can be
neglected. The advantage of introducing Q and
which may be rewritten in the form a&p is that Q and pi/pip are dimensionless quanti-

————j R
03

Mo
Mo

6)
—
L coo
= 0, where cop' —1/L C. (I I I, 48)
ties, easily measured, and easy to transfer to
microwave problems in which I., R, and C have
only a rather uncertain or ambiguous signifi-
cance. In many cases we shall find that the
Equation (III.48) is a quadratic for the fre- angular frequency coo is nearly equal to one of
quency, whose solution is found to be
the resonant frequencies pp, defined in (III.45);
this will be the case if the motion differs from a
free oscillation only by a small perturbation. In
such a case, we may let
where (III.49) pip = p~e+&p~a (I I I.53)
where her is a small quantity. In this case,
(III.52) may be rewritten in several forms, cor-
Using the value (III.49) for the frequency, we rect to the first order of small quantities, as
see that the charge varies as follows:
exp ( —(~p/2Q)~) exp (~j~pLI —(I/2Q)'j'~)
(I II.50)
ji
EM~
—) —
— R~r I+ 1
(pp

CO
2j
COg
=0,

07~ 2.
This represents a damped oscillation, with angu- =J
lar frequency equal to p&pal —(1/2Q)'j', and such
pi Gpp + 2ppg, kppg, ,

that the energy, which is proportional to the

+j—.
M~
square of the amplitude, decreases with time pi=&a +hp~, (I II.54)
according to the expenential function e ' 0'~&'. 2
 M I CROWAVE ELECTRONICS
We see from (III.54) that the real part of the 4. The Unloaded Q
quantity cv' —~ ' leads to a shift of resonant fre-
I

First we consider the losses in the cavity

quency, and the imaginary part to a damping
resulting from the finite conductivity of the
of the oscillation.
walls. In (II.34) we have found that there is a
We now ask what sort of perturbation of our
tangential component of E over the surface 5,
problem of the free oscillation of a cavity is
in case of finite conductivity. From that equation
necessary to produce damped oscillations. Three
we find at once that
types of perturbation are most common. First,
the walls of the cavity, instead of being perfect nXE=H(goo~/2o)l(1+ j)
conductors, may have only finite conductivity,
resulting in resistive losses. In this case, the
=H-,'biioid(1+ j) (I I I.56)
tangential component of E over the surface S at the surface, Thus the first term on the right
will not vanish, and the integral J's (nXE) H side of (III.55) becomes .

in (III.42) will be different from zero. Secondly,

the cavity may have certain windows, or wave
guide outputs, which allow the escape of energy,
~ .(n X E) H.da
S
with consequent decrease of the total energy. In
this case, the tangential component of H over oi~(&ohio) '
the surface S' will not be'zero, and the integral E E,dv
JB (nXH) E~da will not vanish. Thirdly, the
cavity may contain currents in phase with the
'8H H da
—,
voltages producing them, as if they contained ~8
resistive mater'ial obeying Ohm's law. In that = —j(u o/«) '(1+j) (II I.57)
case, the integral J'J E,dv will not vanish. In
all these cases, in order to have damped oscilla- ~
EE,dv
tions, the
gaia
various integrals must be proportional
to the amplitude J'E. E,dv of the field compo- If the oscillation is taking place with almost the
nent, and must oscillate with the same period. frequency of the ath mode, we may assume that
In such a case, we may assume a solution of to a first approximation the field distribution
(III.42) varying as e&"', where o& will in general will be that of the ath mode as well. Thus we
be complex. Substituting this time variation, shall approximately have
and using (III.54), we find
H=H, H Hdv
~ (nXE) H, da
—— her,
22
= j(«/~o) 'H. E E.d~ (I II.58)
Q ~a(&ohio) '
~E E,dv in which used (III.46) for the ratio
we have
between J'E E,dv
and J'H H.dv; a value which
f' strictly holds only for the free oscillation, but
(nXH) E.da J E.dv which would be a good approximation for
~s damped oscillation. . Substituting, we have
. (III.55) h

6{)M~
~E E,dv E E,dv —2 j(A&u = (1+j)
1/Q /o~„) , 6II,'da.
—'
(I II.59)

We shall now consider the various terms on the We see from (III.59) that the surface losses
right-hand side, showing how they lead to damp- result in a shift of wave-length, as well as a
ing and to displacement of the resonant fre- contribution to Q. The value of Q given in
quency. (III.59) is generally ca,lied the unloaded Q; we
476 J. C. SLATE R
shall denote it by Q, . We note that it really impedance or admittance across S' has a certain
should also have a subscript a to denote the ath definite value, so that the voltage is proportional
mode, but we shall omit this when it is not to the current (or E is proportional to H), the
necessary. We may get an idea of the order of result will be a contribution to Q and the fre-
magnitude of the unloaded Q as follows. We quency shift. The more general case, however,
remember that from the normalization condition is one in which there are arbitrary impressed
(III.11) the integral J'H, 'dv is equal to unity. voltages or currents across 5'. In such a case,
That is, if U is the volume of the cavity, and if power can How in as well as out through the
(H, ')A„ is the mean value of H, ', we have (H, ')All guide, so that we have the possibility of forced
= 1/ V. If we assume, to get orders of magnitude, oscillation of the cavity as well as damped
that b is constant over the surface, and that the oscillation. We take up this general case, later
value of II,' on the surface equals 'its average obtaining the case of damping as a special case
over the volume and that furthermore the surface of our general treatment.
area is A, then we should have 1/Q, =SR/2V. The general outline of our derivation will be
That is, Q, would be the ratio of the volume, to as follows. We assume a given distribution of H
the volume of a thin shell of thickness 8/2 over the surface S', or a given current Rowing in
surrounding the volume. It is interesting to see the guide. We can then calculate the integral
how the Q, of a cavity will change with the J'(n&&H) E,du over S'. From (III.42) we can
wave-length. Of two cavities of the same shape compute the electric field everywhere within the
but different sizes, the wave-length will be pro- cavity, and in particular within the guide, and
portional to the linear dimensions, so that the at the surface 5'. From this electric field we can
volume will vary as X', and the area as V. The find the voltage at S', and can take the ratio of
skin depth, by (II.33), is proportional to (X)*. voltage to current, and hence the impedance at
Thus if the conductivity is independent of wave- the plane 5', the input impedance leoking into
length, Q, will be proportional to ('A)', decreasing the cavity. First we must consider the nature of
as we go to shorter wave-lengths. The actual the held in the wave guide. In the guide, the
magnitude of Q, will of course depend on the function II must have the general form given
shape of the cavity, and the material of which by a summation over the modes n of the guide
it is made. of terms as given in (II.13); for any solution of
Maxwell's equations in the guide must have
S. The Input Impedance of a Cavity that form. Thus at the surfaceD' the transverse
We next take up the effect of coupling the cavity component of E, must be expressible in the form
to an outside system by an output lead, which
will be assumed to take the form of a wave guide
E. = Q„E,„(v. /Zp„) (II I.60)
or coaxial line. We assume that there is such a whele Zo is the characteristic impedance of the
line attached to the cavity, and that the surface guide in the nth mode, for an angular frequency
5' is a surface at a cross section of the line. The ~„and where the v, „'s are coeScients (inde-
volume in which we are solving Maxwell's equa- pendent of time, as E, is) which as we see from
tions then includes not merely the cavity, but (II.22) represent the voltage at plane S' and in
the part of the output lead out to the surface S'. the nth wave guide mode set up by the resonance
The normal resonant mode is that which arises E in the cavity. The transverse component of
when there is an open circuit at 5', that is, H is zero at 5', by hypothesis; that is, as we
when there is an infinite standing wave ratio in mentioned earlier, the corresponding current
the output line, with a standing wave maximum components are zero at this plane. If the trans-
at 5'. This would correspond to a voltage maxi- verse component of H, is zero, then by (II.13)
mum, and current node, at this surface. Our the normal component of E, will automatically
problem is now to substitute other boundary be zero on 5'. It is interesting to consider the
conditions at 5', and to find what effect that effect on the v, „'s of choosing the surface 5' at
has on the oscillations in the cavity. If the different points along the output line, If we go
boundary condition consists of stating that the a half guide wave-length along the guide, for
 MICROWAVE ELECTRONICS 477

J'E.E,dv

Q„.
any of the propagated modes, the disturbance (III.42), and assume that varies as
will come back to its initial value. Thus v, „, e&"' we find at once that
though it will vary with the position of 5', will
be a periodic function. We shall discuss the
implications of this periodicity at a later point. E E.dv = (III.63)
For the attenuated modes, on the contrary, the iL(~/~. ) —(~./~)j
disturbance will generally fall off exponentially
as we go along the guide away from the cavity The voltage corresponding to the nth guide mode,
(there is normally no reason to expect the other if E=E„or if J'E is v, „. Thus the
E de=1,
exponential term, which increases exponentially whole voltage corresponding to the nth guide
as we go away from the cavity, to be present). mode is v, „J'E E,dv. If we call this voltage
P,
Thus if we take S' some little distance away from U, we have
the cavity, the values of the v, „'s for the attenu-
ated waves will have fallen to very low values, Vn = Pm &mZam,

and may be neglected. As a rule we shall assume (III.64)

that our surface 5' is far enough from the cavity Z„„=Q.
+adam/&0&a
for us to have this situation. Then our summation iE(~/~-) —(~./~) j
over n really has non-vanishing terms only for
the propagated waves. In case we are considering That is, there are linear relationships between
an angular frequency co, for which only the the V„'s and the i 's, and we have been able to
dominant mode in the guide will be propagated,
write down the coefficients, of the nature of
the summation will reduce to a single term.
impedances, in an explicit manner.
We shall now assume that over the surface S'
There are a number of remarks which can be
we impose a tangential magnetic field which in
made about this important result. In the first
accordance with (I I. 13) and (I I.22) we may
write place, suppose that we are applying it in a region
of wave-lengths where the guide can propagate
H=&„H,„i„ (I II.61)
only in its dominant mode. Then in the summa-
where the i„'s can be interpreted, as in Chapter tion over m, we shall have only one term, that
II, as the currents associated with the various related to the dominant mode; all the other Z„'s
modes of the guide. AVe assume that H varies will vanish. If this mode is called the 1st mode,
as e'"', each of the i„'s having this variation. We we shall have
may now compute the integral over 5' of
(nXH) E„which appears on the right side of Ug = ~iZgg)
(III.42). We shall take the positive direction in (I I I.65)
the guide (the direction of the unit vector Ir of V &a& /&0&a
Chapter II) as being into the cavity, so that the Z 11
ir (~/~. ) —(~./~) 3
unit vector n will be — k. From (II.11), we then
have nXH = p„ i„E,„/Z0„. Multiplying this by
This then represents the impedance seen looking
the value of E, from (III.60), and integrating
into a cavity through a wave guide which propa-
over the surface S', the cross terms in the double
sum integrate to zero on account of the orthogo- gates only in the dominant mode. We observe
nality properties of Section 5, Chapter II, and that it is a sum of resonant terms, the impedance
the result, using the normalization condition becoming infinite when the frequency equals the
(II.30), is resonant frequency of any one of the modes. At
frequencies near co„ the term in this frequency is
much greater than any other, and varies rapidly
(nXH) E,da = Q„ i v, „. (III.62) with frequency, whereas the other terms are
S'
small and slowly varying. It is often convenient
If now we use this value of the integral in to lump these other terms together, using a
 J. C. SLATER
,

notation Z l to represent them, so that (III.65) in the form

2/ ~a&a Zll
Zll — — ~al /
—+Z.l. (III.66)
1/Qext, al
—
jL(M/M. ) —(M./M) ] Z01 j DM/M. ) (M /M) j
1/Qext, al Zal
We may now ask what will be the damping if + . (I I I.71)
the guide is terminated by an impedance Zl. The jL(M/M. ) —(M. /M) j
quantity Zll in (I II.66) is the impedance looking
into the cavity, or is the negative of Zl, the In the treatment we have just given, we have
impedance looking out. Making this substitution, neglected the other terms on the right side of
we have from (III.66) (III.42), coming from losses in the walls and
other forms of losses. If however we have the
Mal &at /&pMa situation of Section 4, we may easily take account
+ =o. (II I.67) of the losses in the walls. When the external
(Ma M J Zl+Zal frequency is near the resonant frequency of the
ath mode, the term in a in the summation
This is in the form of the first of Eqs. (III.54),
and shows that we have
(II1.65) will be much greater than any other,
which means that the field is almost like E„ the
assumption underlying Section 4. In setting up
—
1
—2j AMa Val /CPMa 1 ZP1
(III.63), we then see that we must put additional
Q Ma Zl+Zal Qext, al Zl+Zal terms as given by (III.59) in the denominator.
We then have in place of (III.71)
where
l

&al
2

(III.68) 1/Q. t, .
i
(I I I.72)
jL(MIM. ) —(M.'/M)1+ (1/Q. )
Qext, al &PMaZP1
Zol

In this formula, we have introduced a quantity, where we have used co ' to refer to the resonant
,
Q, t, l, which we may call the external Q of the frequency as modified by the correction term
ath mode of the cavity, and the 1st mode of the hM, derived from (III.59). We see that the'input
guide. Its meaning is simple; if we make Zl+Z, l impedance is no longer purely reactive, but that
=Zpl, it is the Q of the resonant mode. If we it has a resistive term, the resonance term be-
neglect Z, l, this means that there is to be a coming purely resistive at resonance, just like
matched load in the guide. We shall see presently the input impedance of a parallel resonant circuit
that if we choose the position of 5' properly, we in ordinary circuit theory. Proceeding as in the
can make Z l equal to zero, so that in this case derivation of (III.68), we now find that if there
Q, t is exactly the Q which we should have with is loss in the walls, the quantity 1/Q — 2jhM /M
a matched load. The external Q is clearly a is the sum of the quantities (III.59) and (III.68)
measure of the coupling of the ath mode to the arising from the losses in the walls, and from
output, through the j.st mode of the guide. In the effect of the output lead. The resulting Q is
case v, l is very small, the external Q is large, called the loaded Q:
or there is very small coupling. It is convenient
'

to define an admittance g+j b by the relation —

i
=—+- 1 g
(III.73)
ext al
g+ jb = Z pl/(Zl+Z.
L a
l) (II I.69)
We next consider the case where there are
In this case, we have several output leads for the cavity. In (III.62),

.1. g

ext, al
t
———.
—2 Aced, =
~a
b

ext, al
(I I I.70)
the integration must be carried over each of the
surfaces 5' closing the various leads, and the
summation- over n will include terms for each of
the propagated modes in each of the leads. This
In terms of the external Q, we may rewrite same situation will carry through to (III.64), in
 ELEC I RON I CS
which we can formally use the same expressions for any value of I, including m. The impedance
we have already derived, but in which we must looking into the eth mode is
now understand that the summation over rn
involves a summation over each mode of each V„v,„ E E.dv. (I I I./5)
lead. We see then that a res'onant cavity acts i„ i„ ~
like a network with as many pairs of terminals
as there are propagation modes of the various As in (III.64) and (III.65), if we are in the
leads. A cavity with two leads, each propagating neighborhood of the ath resonant frequency, the
only the dominant mode, acts like a four-terminal term in a in (III.75) will be large and rapidly
network, such as we have discussed in Chapter I. varying with frequency, while the other terms
Furthermore, the impedance coefficients are as will lump together to a slowly varying term. Let
found in (III.64). We note that the denominator us assume that we are near the ath resonant
in this expression is to be modified as in (III.72) frequency, and as in (III.66) lump together all
in case we consider the losses in the walls. Our these slowly varying terms, rewriting'$(111. 75)
present result proves the existence of linear rela- in the form
tions between the various voltages and currents, V„v „
and hence justifies the whole treatment of E E,dv+Z„„ (III.76)
Chapter I in its application to problems of reso- &n &n ~
nant cavities. Beyond that, however, we now
see how the impedance coefficients vary with
where Z, „ is simply defined as the sum of all
terms of (III.75) except that with index a. We
frequency, a feature which we omitted from our
now have, since the output impedance at the
discussions completely in Chapter I.
In case there are a number of output leads, we nth mode is Z„, for n&m, the relations
can treat the problem, as we have just seen, like
a network with an appropriate number of pairs —Z = ~an — E Edv+Z
of terminals. If power is being fed in only through
one. lead, and in only one mode, however, and if
(I I I. /7)
&an
all the other leads and modes are terminated with
passive impedances, they will contribute merely Z„+Z.„
to the Q and to the displacement of the resonant
frequency of the cavity, and the problem may be
handled as that of a cavity with one output. Ke then wish to find
Following back over the argument, we see that
we can handle these terms as in the last para-
(II I.78)
graph. Suppose that a particular mode of the
particular lead in which power is being fed in is
denoted by the mth, and that each other mode and to get it we must know an accurate value of
of each lead, say the nth, is terminated by an J'E E,dv. To find this, we use (III.42) aga, in, as
impedance Z„. We wish to find the input imped- in the derivation of (III.55). Replacing the inte-
ance V /i =Z looking into the mth mode.
We have
gral f
(nXE) .H, da by the value found in Sec-
tion 4, for the case where B is almost exactly
proportional to B„and replacing the integral
V„=g, v E E dv, (I I I.74) J'(n XH) E,da by the value (III.62), rewriting
the i„'s in that expression from (III.77), we have

(o& cg i f6 (v „/eoM)
i+(I+/) & —H
(o)
~i

2
2da+ Q
~ Z +Z. ) . E E dv+
COGOg ~
J E,dv= i „,v, „. (I I 1.79)

Introducing the unloaded Q from (III.59), the modified resonant frequency +,' from (II I.72),
 J. C. SLATER
and the external Q from (III.68), we then have
&mvam/popish
E E,dv= (I I I.80)
)
J E.dv
(0

( pi. '
M )
/+ —
1
+2 —
1/Q
+
1 i,

cp ) Q. ~a~ Z„+Z,„popi
E Ed@
and, substituting in (III./8),
1/ ext, am

(I I I.81)
( op pr, ')
) I+ —
1
—+—2
"&'"
—
1/Q, o,' „1
+-
,
,
t'J E.d.

(Ma pi Q~ Zn, +Zgn &opia r'

E 'Egdv
J
Equation (III.81) exhibits the input iinped- breadths of the two resonant peaks are great
ance in a mode of the output leads as a resonance enough so that they overlap. In such a case,
term, plus a slowly varying term. In the reso- more elaborate methods than we have used are
nance term, we see the eBect on the resonance necessary to get valid approximations to the
frequency and the Q of all the types of perturba- input impedance. If all resonances are well sepa-
tion which we met in (III.55): the losses in the rated, however, (III.81) should be accurate
walls, the losses in the windows, and the losses enough for ordinary purposes.
on account of current in the cavity. We see that
each mode of each output lead furnishes a 6. Currents within the Cavity
correction to the frequency, and a contribution Ke have not so far considered the contributl011
to Q, just like the value given in (III.68) for a of currents in the cavity to the losses and fre-
single mode of a single lead. We sha11 discuss the quency displacement. From either (II I.55) or
term involving currents in the next section. In (III.81) we see that this contribution is
Eq. (III.81), we have a formula correct near
the ath resonance frequency, when the resonance J E.dv
term becomes large. We may reasonably assume
that the slowly varying term Z „may be approx-
——22 (II I.82)
imated, as for instance we saw by comparison of t E E.dv
(III.65) and (III.66), by a sum over a of terms
like the resonance term of (III.81). Such a sum The simplest application of these terms is the
will have the correct behavior. near each .reso- case where the cavity is uniformly filled with a
nance. Furthermore, far from resonance the first material which conducts according to Ohm's
term in a resonance denominator will be large law, with a conductivity |T, and which also has a
compared to the others, and this erst term is just dielectric constant e, rather than the value 6p
as given in (III.65), in which we showed that characteristic of empty space. From the con-
the sum of all the resonance terms except the ductivity, there is a current density J=o.E, and
ath was equal to Z, ~. In other words, we are from the polarization P = (p —pp) E there is a
justi6ed in assuming that Z is approximately current density given by the time rate of change
given by a sum, over all values of a except a of P, or by jop(p —pp)E, if there is a sinusoidal field.
itself, of resonance terms like that in (III.81). In that case, from (III.82), we see that we have
The only case in which we may assume that this
formula is not very accurate is that in which two
1
'
p Api 1(p —ppq
(111.85)
resonant modes are close together, so that the Q pppi. pi. 2 ( pp
 MICROWAVE ELECTRONICS
The first of these indicates greater and greater same basis. This equation furnishes the funda-
losses as the conductivity becomes larger, and mental basis for all discussions of microwave
the second gives the first-order correction -to the oscillators. There are important cases in which
frequency if the cavity is filled with a dielectric; the left side can be transformed into a form
the correct formula for the change of frequency, suggesting lumped constants. In various types
as we can see by substituting e for eo in (III.45), of oscillators, such as the magnetron and the
is that the frequency is proportional to 1/ge, klystron, the current is largely confined to a
from which the frequency change of (III.83) region smal1 compared to a wave-length, and
follows at once. the electric field is likewise confined to such a
As a next more complicated case, we may small region. For instance, in the klystron, both
imagine that there is conducting or dielectric electric field and current are in the region be-
material distributed through part of the cavity, tween the grids, which acts almost exactly like a
but not everywhere. In that case, the distribution lumped capacity. Thus there is a definite vo1tage
of E and H might be violently altered, and in this U between the grids. Similarly there will be a
case our assumption that the field is nearly E, current Iflowing in the direction opposing the
in the neighborhood of the ath resonant fre- field (if the current is generating power). If we
quency, on which parts of our deduction are treat the condenser as having an area A, distance
based, would not hold. If the effect of the con- of separation between the plates d, then we shall
ductor or dielectric material in the field distribu- have 8=V/d, I/A, if—
J= positive current is
tion is not great, however, so that we can in the same direction as positive voltage. We
approximately replace E by E, J'E E,dv, (II I.82) shall furthermore have E a constant over the
becomes volume of the condenser, where alone E and J
are different from zero. Thus we shall be able
1
1
2j —
Aw,
=— (o+j u(e eo))E.'dv.—
to rewrite the left side of (III.84) in the form

b.
07~ CpGDg ~

That is, the effect of a conductor or dielectric is J E.dv I

1 g+j
large at places where E is large, small where it (II I.85)
is small. ep(o, p (eoA/d) o), V C(u,
IE E,dv
A different type of problem is that in which
the cavity acts like an oscillator, the currents
leading to the generation of power. In such a in which g+jb is written for I/V, the ratio of.
case, in (III.81), the impedance Z would corre- current to voltage, and C= eoA/d is the capacity
spond to having a load impedance in the mth of the condenser. This same formula can be
mode. That is, we could change the sign of the justi6ed in another perhaps more general form,
right side of (I I I.81), and regard Z as an by considering the definition of Q in terms of
external impedance. A simple transformation energy loss. Thus suppose the current, instead
then leads to the result of acting as a generator, is acting as a load, and
is contributing to the energy loss in the cavity.
In this case the term on the left of (III.84) should
1 ~
J E.dv have a real part which is the contribution of these
losses to 1/Q. Let us suppose that the complex
fE.E dg voltage is V, the complex current I=
(g+jb) V.
Then the decrease of energy per unit time is

= jl ——] — ~u' i 1
I+ +Z-
Q.
1/Qext,

z„+z.„
an
(«I 84)
~ReIV = ~ g U ', and the stored energy is 2 C V ',
~ ~

where C is the capacity, so defined as to give

the stored energy properly in terms of the
~ ~

voltage. The contribution to 1/Q is then given

In this expression, the summation over n includes by (III.51), from which we see at once that it is
all modes of the output leads; they are all on the g/C&u„as we should deduce from (III.85), This
 J. C. SLATE R
formula then holds, so long as it is possible to desired above, since n is the inner normal into
define a voltage, current, and capacity, satisfying the volume enclosed between the two surfaces.
the correct energy relations. Using (I I 1.85), then, Thus we have
the final result is

g+jb
C(a,
=&~ —— —
((o ~ ') 1
(a) ' co )
~+ +g.
Q,
1/Q ii
Z„+Z,„
. (III.86)
'

(nXH) E,dc=k
~
(II,' Z'—
)dv, (III.88)

where the surface integral is over the perturbed

We shall use this formula in the next chapter surface, the volume integral over the volume
for giving a general discussion of the properties between the surfaces. If the field H, instead of
of microwave oscillators. being H„ is a constant times this, or H, J'H H, dv,
then we must multiply the right side of (III.88)
7'. Perturbation of Boundaries
by J'H. H, dv. We may then introduce this inte-
One more use of (III.42) is frequently very gral into (III.43), finding
valuable: the problem of the change of frequency
of a resonent cavity, when its boundaries are B' )/&
perturbed, as by pushing a small part of the
Eg4 M +k = k (II
wall in or out. Thus let us start with a cavity, (III.89)
entirely enclosed by a perfect conductor, in
E,
H, . Now let
'i 1+ (H' E')dv —l.
which we have found functions )
the mall be pushed into the cavity by a small
amount, and let us consider the solution of the This very valuable formula gives the perturbed
problem with the perturbed walls. In the small frequency co, resulting from a mode of reso-
volume between the original wall and the per- nant frequency co„ in terms of an integral
turbed one, the final functions E and H will be J (II,' —Z,')dv over the volume which is re-
zero. Thus there will be a discontinuity of the moved from the volume of the resonant cavity
tangential component of H at the perturbed by the perturbation of'the surface. We see that
wall. This corresponds to a surface current, and when the surface is pushed in, the frequency
hence, in (III.42) and (III.43), we must include increases if the magnetic 6eld is strong at the
an integral J'(nXH) E,da over such a surface, part of the surface perturbed, and is decreased
even though it is not part of a surface 5'. The if the electric field is strong there. As a simple
perturbed field H will be very nearly equal to illustration of this formula, let us start with a
the unperturbed H over the surface, so that the cylindrical cavity, in the mode in which E runs
term can be approximated as J'(nXH ) E d&, axially from one face to the other, and let us
integrated over the perturbed surface. This may consider the charge in frequency when posts are
be rewritten J' — n (E, XH, )da. But we may extended down from the two faces toward each
transform this by using other, as in constructing a klystron. The electric
field is strong at the place where the posts are
div (E, XH, ) =H, curl E, —E, curl H introduced; thus th'e effect is to decrease the
= k, (H, ' —Z '). (I I I.87)
frequency. As another illustration, consider the
We integrate this quantity over the small volume effect of making a hole in the cylindrical face of
between the original and perturbed boundaries. such a cavity. Here the magnetic field is strong,
The integral of the left side, by the divergence the electric field zero; pushing in the wall would
theorem, equals the surface integral of the normal increase the frequency, whereas pushing it out
component of E, )&H, over the boundary. Over decreases the frequency. Finally, there can be
the original surface, we may replace the quantity simple cases where the effects cancel; thus in the
n (E XH, ) by (nXE, ) H„ in which nXE, is same cylindrical cavity, pushing down the whole
zero over the surface, and hence the contribution top face of the cavity decreases the frequency on
vanishes. Over the perturbed surface, the contri- account of the center part, where the electric
bution is just J' — n. (ENXHa)da, the quantity field is large, but increases it on account of the
 M IC ROWA VE ELECTRON I CS

outer part, where the magnetic field is large. A or

calculation of J'(H, ' B— , ~)dv over the cylindrical
face would show that these eRects just balance,
and actually such a change of dimension, with
that particular mode of the cavity, does not m = integer. (IV.2)
change the frequency at a11. We shall meet other
examples of the use of this theorem later on. It This equation determines the wave-length (&u,

must be remembered that it is strictly correct and X„which is of course related to it), as a
only for a'n infinitesimal distortion of the surface. function of d, the plunger position, or more
directly d as a function of the wave-length. The
IV. APPLICATIONS OF THE THEORY OF curves determined by (IV.2) are similar to those
RESONANT CAVITIES given in Fig. 12. Far from a resonant frequency,
I. The Tuning of Resonant Cavities , the summation over a is small (equal to Z„, of
(II I.71)), so that d is approximately nag/2, shown
As a first and very instructive example of the by a set of straight lines through the origin in
application of our general theory of resonant the figure. As the frequency goes through a reso-
cavities, we consider the following problem: a nant frequency, however, the summation goes to
resonant cavity is provided with 5 wave guide infinity, changes sign, and again becomes small
output of some form, and the guide is closed with far from the resonance on the other side; the
a movable short circuiting plunger. We ask, how inverse tangent in the process increases by x, so
do the frequencies of the various resonant modes ),
that d increases by /2, or the curve crosses
vary with plunger position? This is the problem from one of the straight lines to the next. Just
met when we try to tune a cavity by connecting at resonance, d= (e+~~)X,/2. In the figure, only
with a tunable wave guide, and it furnishes a two resonances are shown; but actually there
simple example of the general problem of two will be an infinite set, stretching down to shorter
coupled cavities, one of which can be tuned and shorter wave-lengths without limit.
through the resonant frequency of the other. At A number of observations can be made about
the same time, as we shall see later, it has a very this tuning curve, as we can call it (since it
close relationship to the problem of determining shows how the frequency of the cavity is tuned
the input impedance of a cavity, as a function by moving the plunger). In the first place, a
of frequency, by measuring the standing wave position of the plunger can be found by which
ratio and position of the standing wave minimum the resonant frequency has any desired value.
as functions of frequency. We shall assume for A mode which has, for instance, the frequency ~&
the present that there are no losses in the system, for one position of the plunger, will tune con-
and no damping of any sort. We consider the tinuously into a mode with frequency ~2, simply
surface S' of the preceding chapter as being a by moving the plunger. The resonant frequencies
definite cross section of the wave guide output, are, of course, periodic with plunger position,
between the cavity and the plunger. Let the increase of d by a half guide wave-length bringing
distance from S' out to the plunger be d. Then the whole set of frequencies back to their original
the impedance looking out across S', assuming values. We observe next that the resonant be-
that the guide will propagate only in its dominant haviors are essentially tied up with the inter-
mode, is jZo tan 2md/X„where Zo is the char- sections of the dotted lines d=n), /2, and the
acteristic impedance of the guide, X, the guide dotted lines representing the various resonant
wave-length. This must be the negative of the frequencies. That is, they come when a resonant
input impedance seen looking into the cavity frequency of the wave guide itself coincides with
across the same plane, which is given by (I I I.71). a resonance of the cavity, so that we have
Thus we have essentially a coupled system. We notice -further-
more that a resonance with a small external Q,
(IV.1) or a tight coupling, has the eRect of pushing the
tuning curve far from the intersection of the
 J. C. SLATE R
dotted lines (as with the resonance ~i in the
figure) while a resonance with large external Q d,
or loose coupling, like ~2 in the figure, has the
effect of letting the tuning curves approach each
other very closely at the resonance. In the case
of a loose coupling, then, as we move the plunger,
we find that over wide ranges of plunger position,
the resonant frequency is almost independent of I
plunger position; we find definitely non-tuning I/
resonances of this type, which are resonances of /I]
I/] /
the cavity, and also other resonances, which tune ]l~ /
greatly, following the dotted lines nX, /2, which // //
///~
are the resonances of the output wave guide, , ///~ g
and do not a6'ect the cavity at all. Only when
these frequencies coincide do we get appreciable
Flr. . 12. Tuning curve of a resonant cavity.
tuning of the cavity resonance. Thus this situa-
tion is not suitable for an actual tuner for a
open circuit at S', that is, these resonant fre-
cavity resonance. With the tight coupling, how-
quencies are just the resonances co . Furthermore,
ever, the cavity resonance tunes strongly over a
from the slope of the tuning curve at these
wide range of tuner positions, so that this is the
intersections we can find the external Q's. If we
situation actually desired for a tuner for a cavity.
differentiate d, in (IV. 2), with respect to 'A„and
An interesting insight into the nature of the
set the frequency at such a value that co=co,
tuning curves is found from the relation (III.89)
so that the ath term of the summation becomes
of the last chapter, in which we studied the
infinite, we easily find
change of resonant frequency of a cavity when
we push in a section of mall. From that equation, dd (n+1/2) ,
Q, „», d In &d

(IV.3)
we find that if the field is large near the movable
d)g 2 d In X,
section of wall (which in this case may be taken
to be the plunger) a small displacement will make which shows at once, as we have already pointed
a large frequency change, while if the field is out, that the vertical part of the tuning curve is
small it will make a small frequency change. We very steep for a case of loose coupling, or large
notice that in the vertical part of the tuning external Q, but which also shows that from this
curve, a large displacement makes only a very slope we can find the external Q directly. Un-
small frequency change; that means that the fortunately, though this gives in principle a way
field at the plunger, and in the wave guide in of finding the external Q, it is not accurate in
general, is very small in this case. In other words, practice, on account of the very large slope,
that corresponds to a resonance of the cavity, which is hard to measure.
the field being such that only a small amount of We may now consider a question which has
it is located in the wave guide output. On the been disregarded until the present: how has the
other hand, in the approximately horizontal surface S' been chosen? We have stated merely
parts of the curve, a small displacement makes that it is an arbitrary surface in the wave guide
a large frequency change, showing that the field output. We could, then, equally well have chosen
is strong in the wave guide, as we should expect another surface. If we had done this, however,
if it is the guide that is resonating, rather than the distances d to given plunger positions, meas-
the cavity. ured from the new surface S', would have been
Suppose we take the intersection of the tuning different. In other words, the tuning curve would
curve with the straight line d =X,/4. That is, we have been moved up or down in the figure, by
ask for the resonant frequencies when there is a the amount of displacement in S' from its original
short circuit a quarter-wave down the line from position. The straight line d=X, /4 would then
the surface S'. In this case, there must be an intersect the tuning curve at different places, so
 M I CROWAVE ELECTRON I CS
that with the new S' the resonant frequencies purely reactive load), and S' is the plane of the
would be different; and the slope of the tuning standing wave minimum. We may, then, choose
curve at the intersection would be different, so S' to be the plane of the standing wave minimum,
that the external Q would be diferent. This is when the resonance is tuned away fr'om the
the most direct way of seeing the fact, which we frequency at which we are working; then Z, will
mentioned in Chapter III, Section 5, that the be zero, and if we tune the resonance back to
external Q's were a function of the position of S' the frequency at which we are making our meas-
along the line. urements, the input impedance will consist of
It is now natural to ask, is there some particu- just one resonance term. We can make this even
larly correct way of choosing S', which leads to more correct, if we notice that Z, will always be
more sensible results than any other? We have a function of frequency, by choosing S' to be a
drawn the figure in a particular way: with one plane whose position depends on frequency,
of the branches of the tuning curve lying close taking at each frequency the position of the
to 2 =0, and others close to d =nX, /2. Clearly a, standing wave minimum looking into the cavity,
change in S' would change this situation. At when the resonance is tuned out of the way.
least in the neighborhood of one of the reso-
nances, we can always secure this situation, 2. Measurement of the Properties of a
though the same choice of S' will not always be Cavity Resonance
correct for different resonances. Suppose we are The input impedance looking into a cavity,
considering a resonance of high Q. Near the in which we can no longer neglect the losses, can
resonance frequency, by (111.71), the sumination be written
can be replaced by its ath term, plus the slowly
varying quantity Z . If we entirely omitted the Z I/Q. „, . Z
+—, (IV.4)
resonance term, then the tuning curve would go
right through the resonance frequency without
. .
iL( / ) —( / ) j+(&/Q. )
showing a resonance phenomenon. If we wished in the neighborhood of the ath resonance. We
this tuning curve to coincide with the line d =0 shall now ask how the parameters describing
in the neighborhood of the resonance, we should this resonance can be found by measurement of
then have tan 2nd/X~=Z =0. We can do this, the standing wave ratio and position of standing
on account of the fact that Z is a function of wave minimum looking into the cavity, 'as a
the choice of S'. We can give a physical meaning function of frequency. In the first place, if the
to the process of omitting the resonance term. losses are negligible, so that Q, is infinite, the
Suppose that our cavity is tunable, as for instance standing wave ratio will be infinite at all fre-
a wave-meter cavity. Then by tuning it, the quencies, and the position of the standing wave
resonance frequency ~ can be shifted around at minimum in the line will represent a plane of
will. The process of tuning, however, will have zero impedance, which could be closed by a short
relatively small effect on Z, made up as this is circuiting plunger without change of conditions.
of contributions from an infinite number of reso- In other words, the curve of position of standing
nant modes. We may then tune the resonance wave minimum as a function of guide wave-
away from the frequency co in whose neighbor- length is just the same as the tuning curve
hood we are considering Z„and the remaining which we have already shown in Fig. 12. As we
input impedance of the cavity will be Z . But go through a cavity resonance frequency, the
now this impedance, as seen across an arbitrary position of the standing wave minimum rather
plane, will vary, just as any impedance takes on suddenly. shifts by a half wave-length, 'while.
different values across diR'erent planes in the between resonances the positions of the standing
transmission line. Thus if we find that Z =0 wave minima move along gradually and regu-
across the plane S', we should then find across larly.
a plane distant d from this plane that Z, In case the losses must be considered, the
j
=.— tan 2nd/X, . In other words, there is an
infinite standing wave ratio (since we have a
problem is more involved. Let us first consider
the form of the curve of impedance versus fre-
 J. C. SLATE R
each term of Z varies with frequency. Most of
the terms come from high resonance frequencies
so that the terms of (III.71) can be approxi-
,
mately written in the form (j/Q, „&, )(ao/ra, ), so
that, since each of these varies proportionally to
the frequency, the same is the case with Z .
Thus as the frequency increases, we not only
traverse the circle in the impedance plane, in
the clockwise direction as we readily verify, but
the circle also moves bodily upward. Thus the
curve of impedance is similar to that shown in
Fig. 13. As the frequency continues to increase,
FIG. 13. Impedance of a resonant cavity, for frequencies the point representing the impedance will travel
near resonance. upward. from the loop shown, and at the next-
resonance will traverse another loop, and so on
quency in an impedance plane. If the frequency indefinitely. We now notice that the procedure
takes on all values, the first term of the right of the end of the last section, in which we
side of (IV.4) is a circle in the impedance plane; measured the impedance, not across a fixed
for Z/Zo is a bilinear function of the quantity plane, but across a plane S' which varied with
j(~/~, —M /s&), which takes on only pure imagi- frequency, such that it always formed a standing
nary values, and hence traces out a straight line, wave minimum when the resonance was tuned
the imaginary axis, in its own complex plane, out of the way, corresponds to disregarding the
so that its transform into the impedance plane gradual vertical motion of the circle in the figure
must be a circle. Corresponding to frequencies above, replacing (IV.4) by the first term, repre-
far from resonance, where the denominator be- sented by a circle with center on the axis of
comes very large, the impedance goes to zero, abscissas, as we first described it.
so that the circle passes through the origin; at To find the standing wave ratio and position
resonance, the impedance is again real, and we of standing wave minimum, we wish the plot of
see immediately that Z/ZO=Q, /Q. ~, at reso- the impedance as a function of frequency, 'not in
nance. Thus the two intersections of the circle the impedance plane as in Fig. 13, but in the
with the real axis are determined, and since the reHection coefficient plane. This will be as in
circle is clearly symmetrical with respect to the Fig. 14. From p, the magnitude of the reHection
real axis, it is completely fixed in position by coefficient, we can find the standing wave ratio
these two conditions. We see that there is a in voltage, power, or decibels, from (I.34), '(1.35),
special case when the loaded Q equals the ex- or (I.36); it can be calculated by using the
ternal Q: the cavity forms a matched load at relation r = (Z — Zo) /(Z+Zo), in combination
resonance, so that all power fed down the line is with (IV.4). 'We see that the standing wave ratio
absorbed by the cavity. If the unloaded Q is
greater than the external Q, the impedance at
resonance will be greater than the characteristic
impedance, and if Q, is less than Q, „~ the imped-
ance wi11 be less than the characteristic imped-
ance. When now we add the quantity Z, /Zo, as
a first. approximation we simply shift the circle;
if Z, /Zo is purely reactive, we shift it vertically,
but Z mill actually have a small resistive compo-
nent, so that we shift it slightly away from the
imaginary axis. On the other hand, Z mill actu-
ally be a function of frequency, though a slow'ly
FiG. 14. Impedance of resonant cavity, in reflection
varying one; as we see from (II I.71), for instance, coefficient plane.
 M IC ROWA VE ELECTRON I CS 487

goes from a very large value oE resonance, down

to a minimum value at resonance. At the same
time the distance d from the plane S' (which is
fixed in the form of diagram shown above) out
to the position of the standing wave minimum
is given by
4ird/X, = y, (IV.S)
where p is the angle in the figure. Thus we see
that, as we go through the resonance, p increases
by 2s, so that d increases by X,/2. In other
words, d, the position of the standing wave
minimum, as a function of )„behaves in a FIG. 15. Position of standing wave minimum. as function
manner similar to that of Fig. 12, which repre- of wave-length, for resonant cavity with large losses.
sents the limiting case of no losses. The situation
is quite different, however, in case Q, /Q. „» is There is one qualification to be made to this
less than unity. For in that case, the loop does definition. If we have the case shown in Fig.. 14,
not encircle the origin in the reflection coefficient where Q, /Q, „» is greater than unity (we may call
plane, so that P, instead of incr'easing by 2ir as we this Case I), then 00 will be greater than unity
go through a resonance, merely goes through a in (IV. 6), and it will be the standing wave ratio.
maximum and minimum, but ends up at almost On the other hand, if we have the other case,
the same value at which it started. In such a wher'e Q, /Q, „, is less than unity (which we may
case, the behavior of d, the distance from S' to call Case II), 00 will be less than unity, and will
the standing wave minima, behaves as in Fig. 15. be the reciprocal of the standing wave ratio in
For the accurate determination of the con- voltage at resonance. We may distinguish experi-
stants of the resonance, measurement of d, the mentally between the two cases by examining
position of the standing wave minimum, is not the position of the standing wave minimum d as
nearly as good as a measurement of the magni- a function of frequency; if it resembles Fig. 12,
tude of the standing wave ratio, as a function of we have Case I, while if it resembles Fig. 15, it
frequency. A formulation of the equation of this is Case II.
curve, which has been found convenient in With the definitions of 0.0 and 0~, we can then
practice, is the following. First, we tune the rewrite (IV.4) in the form
resonance we are interested in away from the
frequency range we are considering, and. measure z
+ (rii (IU. 7)
the standing wave ratio as a function of fre- Zo i&+ (1/(~o —~i) ]
quency. We choose the plane S' as the plane of where
the standing wave minimum under these circum-
stances, so that S' varies with frequency. The
8 = Q.„, ———-2Q, „, (~ —~0)
((d N
—. (IV.8)
i
( (dg, (4 . )
)

impedance as observed across S' is then purely

resistive. Remembering, as in (1.37), that the Defining the reflection coefficient from (IV. 7),
standing wave ratio in voltage is the reciprocal we may solve for the magnitude of the reflection
of the value of Z/Zo at standing wave minimum, coefficient, and find without difficulty
we may define Z /Zo as 0 i, the reciprocal of the
(0 i —1) '((ro —o'i) 'Ii'+
(ao- 1)'
standing wave ratio in voltage off resonance. (IV.9)
Next, we tune the resonance back to the point (~i+1)'(~0 —~i)'&'+(~o+1)'
we are interested in, and measure the standing
wave ratio on resonance. From (IV.4), this is From this quantity, we find the standing wave
ratio in voltage by the equation
Q~/Q, ~»+oi. We define this standing wave ratio
in voltage as ao, so that we have
= SWR(volt) = 1+)r[ (IV.10)
&0 a ext &l. (IV.6)
488 J. C. SLATER
and the standing wave ratio in decibels from it 3. Power Flow Through a Cavity
by (I.36). The most accurate way to use this
Another way to investigate the resonant
equation, in practice, is to observe a complete
properties of a cavity is to use it as a transmission
curve of standing wave ratio as a function of
device, allowing power to fiow in through one
frequency; this curve will resemble Fig. 16. From
lead, out through another, and measuring the
the frequency of the minimum we find or, , from
transmission as a function of frequency, as the
the standing wave ratio at the minimum, we
frequency is varied through resonance. We shall
find op. The asymptotic value which o(db) ap-
then show that the breadth of the resonant peak
proaches at some distance from the resonance,
is a measure of the loaded Q of the cavity. To
or the value which it has at the resonant fre-
carry out this process, we must feed power into
quency when the resonance itself is tuned out of
the cavity from an oscillator through a line with
the way, gives pi. From these values, and (IV.9),
much padding or attenuation, as described in
we may plot a curve of SWR(db) as a function
Section 15, Chapter I; for only in this case will
of b. Then we choose a horizontal scale, deter-
the output of the oscillator be independent of
mining frequency in terms of 8, and hence from
the load. Ke note that in this case the input line
(IV.S) determining Q, „t, which leads to the best . from the oscillator to the
cavity will be practi-
agreement with experiment. For rapid testing,
cally matched, so that, as we see from Section 5,
we can observe the breadth of the resonance
Chapter III, the contribution of the input lead
curve at a definite height above the minimum; to the loaded Q will be 1/Q, t. As we see from
one such measurement will give Q, „t. Various that section, the contribution of the output lead
types of charts have been set up, giving conveni- to the loaded Q will be a similar term computed
ent points at which to determine this breadth. for that lead, if the output is matched as well,
One convenient chart gives the height of which we shall assume that it is. The input
SWR(db) at which the breadth should be deter- impedance into the cavity can be written in the
mined, in order that the frequency width will form (IV. 7), but we shall neglect the correction
give Q, t directly, by the relation Dtp/&p =1/Q. t. term o&, which is generally small, and we shall
By the type of analysis of resonance curves lump the loading by the output lead in with the
which we have indicated in this section, we can unloaded Q in the term 1/op. Thus we sha. ll
find values of the three parameters Q, t, Q„and assume that the input impedance is 1/(jr+1/op).
co, characterizing each resonance of a cavity. It We now find, from (1.59), that the power flowing
has proved in practice to be a very powerful into the cavity will be given by
method of analyzing resonant oscillations, first
to use our general theory to indicate the form of 4R/Zp
(IV.11)
the input impedance, in terms of these constants -,'ZpA A (R/Zp+1) '+(X/Zp) '
for each resonance; and then to use experimental
measurement of input impedance to determine where R, X, are the resistive and reactive compo-
the constants experimentally for the resonances nents of input impedance. From the assumption
which are actually of interest. Of course, we above, we have
have given in the preceding chapter general
directions from which the values of these con-
R+ jX
(IV. 12)
stants can be computed purely. theoretically, by ZQ jr+ 1/o'p
suitable solutions of Maxwell's equations, and
Substituting from (IV. 12) in (IU. 11), and carry-
integrations over these solutions. Such calcula-
ing out a little algebraic manipulation, we find
tions have been made in a very few cases, with
straightforwardly that

f1'
agreement with experiments. The semi-experi-
mental approach suggested here is very valuable 4/Q. Q-t. i (IV.13)
in the much greater variety of problems in
2
which exact calculation is too difFicult to ~2ZpAA ( tp pp, q
attempt. Ma Cp I ( Qa Qsxt1Qsxt2), ,
 M I CROWA VE ELECTRON I CS 489

swR (db) as a function of frequency will have the same

form as (IV. 13), so that a determination of its
width will lead to the loaded Q of the cavity.
If the degree of coupling to both leads can be
adjusted, as it can if the coupling is through
irises of variable sizes, or through a coupling
loop whose intrusion into the cavity can be
varied, then we may reduce the coupling as far
as possible, so that the contribution of the
loading of the leads to the loaded Q is negligible;
in this limit, the resonance curve will approach
FIG. 16. Standing wave ratio as function of frequency,
through resonance. a limiting breadth, determined now by the un-
loaded Q, which can thus be measured by
where Q, „», ~ is the external Q of the input lead, transmission.
Q.„», 2 of the output lead. This equation shows 4. Properties of a Self-Excited Oscillator
that the power flowing into the cavity follows
a resonance curve, the power flow at resonance One of the most important uses of the theory
being of resonant cavities which we have developed is
in discussing the properties of self-excited micro-
I' 4Qz, ' wave oscillators, such as klystrons and magne-
trons. These all operate on essentially the same
2ZpAA Q.Q.„», g
principles, which we can discuss in general terms
where without detailed study of the electronic flow
within the cavities, which of course leads to
—= —+ 1 1
+
1
(IV.14) their operation. We shall give the general dis-
L e ext, 1 ext, 2 cussion of their circuit properties here, post-
poning until later a treatment of the electronic
The maximum value which this factor can have
motions underlying them. In very simple lan-
is of course unity, when the cavity is matched to
guage, a self-excited oscillator consists of an
the line; this is the case when Q, », 2 is infinite,
electronic discharge and a resonant cavity, which
and when Q, „», ~ ——Q, = 2QI, . We can find the
plays the part of the "tank circuit" of an ordi-
width of the resonance curve at once; the half-
nary oscillating circuit. The electronic discharge
power point is found when
consists of certain currents flowing within the
»0»0, 2(a) —(u, ) 1 cavity. These currents, by the principles of
(IV. 15) Chapter III, set up voltages. The voltages in
(Og CO COg
turn are what maintain the discharge. The rela-
That is to say, the frequency difference between tion between current and voltage is set by the
the two half-power points is given by hen/&u properties of the discharge, which is always non-
= 1/Ql. . This calculation of power represents linear, so that for only one amplitude will there
that flowing into the cavity. On the other hand, be a given ratio between current and voltage.
this power wi11 be divided up between the losses On the. other hand, this ratio between current
in the cavity, and the 'power flowing out the and voltage must also be set up by the resonant
output lead, in proportion to their contributions circuit. On account of its resonant properties,
to the loaded Q. That is, a fraction the correct ratio can be set up only at one
frequency, near resonance, so that the resonant
1/Q. », 2 cavity has the effect of stabilizing the frequency
(IV. 16) of oscillation at a definite value.
(1/Q. )+ (1/Q-», 2)
The theory of the oscillator is essentially
of the input power wi11 flow out through the contained in Eqs. (III.84) or (III.86). We re-
output lead. Thus a curve of transmitted power write (III.86) with the following changes. First,
490 J. C. SLATE R
we assume that there is only one mode of the of oscillation as the applied voltage in which we
output in which power can escape from the are interested; there will in general be other
cavity; secondly, we assume that the surface in, Fourier components of current, with all multi-
the output line across which the impedance is ples of this fundamental frequency, including a
measured is such that Z, is zero, as we have constant component, and for the present we can
explained in the earlier part of this chapter. disregard these other Fourier components. The
Finally, we write the load in terms of its ad- result of the electronic discussion is then a curve
mittance G+j B, which is equal by definition to giving the electronic current as a function of
1/Z. Then we have voltage. Since the discharge is not in general

=P ——~+—G+jB
(to —
a
&u p 1

a
+, ext
(IV. 17)
linear, this curve will not in general be a straight
line through the origin; and since there is in
general a phase difference between voltage and
current, there will be both a real and an imagi-
which may be written in two parts, the real and nary component of current, assuming the voltage
imaginary parts of the equation: amplitude to be real. The quantity g is then the
ratio of the component in phase with the voltage,
g
= —
1
+
G
to the voltage; b is the ratio of the component
Q. Qt out of phase, to the voltage. We may then find
(IV. 18) as a result of the electronic computations curves
ta Caa 2 (ta —(da ) B for g and b as functions of voltage. No restriction
I /
on the shape of these curves is imposed by
GOa CO GOa &&a Qext
general considerations; if the problem is linear
In this equation, we are assuming a steady for very small voltages, g and b will approach
oscillation, and a real cv,. the case of an expo- constants as the voltage approaches zero, but
nentially increasing or decreasing amplitude, even this situation does not always hold, notably
with a complex ~, will be taken up later, in in magnetrons. We shall find that in important
connection with the problem of starting of oscil- cases g decreases with increasing voltage, the
lations. In (IV. 18), we have the relation between current increasing less rapidly than the voltage.
the electronic admittance g+jb and the load We may well have a situation like that shown in
admittance G+jB and the frequency which is Flg. 17.
demanded by the circuit properties. In addition Given these curves for g and b as functions of
to this, however, we have to know certain V, which we may arrive at as a result of electronic
information about the electronic behavior. theory, we may then combine them with (IV. 18).
Looking back to Chapter III, the relations The first of those may be rewritten g/Car = 1/Ql. ,
(IV. 18) give a calculation of the voltages arising where the loaded Q includes the loading resulting
from the currents present in the electronic dis- from the conductance G in the output lead. We
charge. On the other hand, the purely electronic now see that if G is determined, and hence Qq,
part of the problem is that which predicts the the value of g is fixed, and hence of the voltage V.
currents arising from .certain voltages; we find For the type of curve of g vs. V which is shown
this, as we shall show in later chapters, by in Fig. 17, we see that an increase of loaded Q
apphcation of the mechanical laws to the motions increases the voltage at which the oscillator
of electrons in the assumed 6elds. If there is a operates. If the curve of g es. V crosses the axis
well-defined voltage in the regional where the for a certain 6nite voltage, as it often does, a
electrons are Howing, as is assumed in the deriva- sufficiently great decrease of loading will tend to
tion of (IV. 18), then we may regard the ampli- approach this condition. We can never reach it,
tude of this voltage as an independent variable , however, as we see from (IV. 18); for with a
in solving the electronic problem; and we shall given resonant cavity, we may decrease G by
6nd the electronic current as a function of the changing the loading, but we can never get rid
voltage. We must note in this that it is the of the term 1/Q, , so that we can never make g
component of current having the same frequency approach zero: In the other direction, by in-
 M I CROWAVE ELECTRON I CS

g, b furnish the foundation for a method of frequency

modulation.
Ke have seen how from the external load we
can find the voltage V, and the conductance g,
of the electronic discharge. We know, however,
that the power produced by the electrons, which
we may call P,&, is equal to -,'ReiV, where i is
V the current, V the voltage; or is ~gV'. Multi-
FIG. 17. Electronic conductance and susceptance as plying g as given in the curve of Fig. 1. 7
functions of voltage, for self-excited oscillator.
by V', . this gives a curve starting off as a par-
abola for small U, but falling again to zero as
creasing the tightness of coupling, which amounts g approaches zero. The curve will not have a
to increasing G or decreasing Q, i so as to de- simple analytic form in general, unless we can
crease the loaded Q, g increases, and if the derive a simple theory (as we shall find that we
situation is as in Fig. 17, a quite finite value of can for the reflex klystron) for g as a function of
g corresponds to zero voltage. Thus a given V. We see, however, that as G is changed, varying
quite deFinite tightness of coupling suffices to V, the power will change: as G is increased from
load the cavity down so that it ceases to oscillate, a small to a large value, the power will rise from
and at any higher loading than that it cannot a small value (on account of the low g value at
operate. For a reHex klystron, this, happens at high voltage) through a maximum, and down to
quite ordinary loading; for a magnetron, on the a small value again (as the voltage becomes small
other hand, the curve of g vs. V rises much for high G). It is to be noted that the position
higher, and ordinarily this limit is not reached. of maximum power output cannot be found from
From the conductance 6, then, we can fix g, any simple rules, such as are used for instance
and hence the voltage U. In turn, Fig. 17 shows with constant voltage generators; a generator in
that b is determined in terms of V. The second general does not satisfy the simple postulates of
equation of (IV. 18) then suffices to define the a cog.stant voltage generator. It is not hard to
frequency at which the oscillator must operate, see how a constant voltage generator would
for a given load susceptance B. With a given behave, however. The voltage across the termi-
value of voltage, and hence of b, we note that nals of such a generator equals its e, m. f. minus
(IV. 18) states that the frequency is displaced by the iR drop in the generator. That is, we have
an amount proportional to B. This phenomenon V=E — iR, so that
is the so-called frequency pulling, resulting from
load susceptance. We note that its magnitude is
inversely proportional to the external Q; thus
(IV. 19)
the method of decreasing frequency pulling is to
decrease the coupling, or increase Q, i. We shall This is a hyperbola, rising to infinite values when
see presently, however, that this has compen- U=O (corresponding to the fact that i is a finite
sating disadvantages. In some cases the quantity value, 8/R, when V=O), a, nd crossing the axis
b can be controlled by certain subsidiary condi- when V =B. It forms, in fact, not a bad approxi-
tions of the discharge. Thus in a reHex klystron mation to the curve for the magnetron, where i
it depends on the reHector voltage, and in a has large values near V=O, and where g rises to
magnetron it depends on the anode current. very high values in this neighborhood; it j.s a
Changing such a parameter will change the poor approximation for the reHex klystron, how-
frequency, with a constant load, as we see from ever. For this simple case, the power is given by
(IV. 18). In the case of a reflex klystron, this is
the phenomenon of electronic tuning and in a P, i — (1/R) (Z V —V')
-', (IV.20)
magnetron it is the frequency pushing. These
phenomena can be used to modulate the fre- a parabola with maximum at V=8/2, or half
quency of an oscillator electrically, and hence the voltage at which the value of g goes to zero.
 J. C. SLATER
The power produced by the electrons is not (on account of the vanishing of circuit efficiency),
all delivered to the external load, on account of through a maximum, and down to zero again
the losses in the cavity. We can discuss the losses (on account of the vanishing of voltage), this
by the general method given in Section 16 of second zero value being reached generally at a
Chapter I, finding the circuit efficiency, repre- finite rather than infinite value of G. For any
senting the fraction of the electronic power value of G, the frequency in turn is determined
delivered to the load. In terms of it, the output by 8, as we saw in (IV. 18). The value of G fixes
power is then b, and this determines the frequency when 8 = 0,
gcI el- (IV.21) or for a purely resistive load; changes in 8 from
this value then displace the frequency, by an
On account of our simple circuit, however, we amount inversely proportional to Q, &. We may
can give an elementary derivation of the circuit then draw curves of constant power, and of
efficiency, at least in case we can neglect the constant frequency, in an admittance plane, in
quantity Z, . The power P, must all be dissipated
~
which G and 8 are plotted as abscissa and
in the various losses in the circuit, by conserva- ordinate. The constant power contours are verti-
tion of energy. These losses are of two sorts, cal lines. The constant frequency contours are
those arising from resistance in the cavity, and all obtained from the contour corresponding to
from the external load. The power delivered to co = co
'
by vertical displacement. That contour is
each of these 1osses is proportional to its contri- determined by the simultaneous solution of
bution to 1/Qz„by the fundamental definition the two equations g/C~, =1/Q, +G/Q, „&, b/C&d,
of Q. Since these contributions are 1/Q, and 8/Q, i, =— 0. Each value of G leads to a value
G/Q, &, respectively, the circuit efficiency is given of g; by Fig. 17, this leads to V, hence to b, and
by hence to J3. We may write the equations
G/Q-~
(IV.22) G=Q, „i(g/Ccd. —1/Q ), B=Q, i(b/Cid. ). (IV.23)
1/Q. +G/Q-~ G+Q-~/Q.
-
Thus the curve of 8 vs. G is like that of b vs. g,
This quantity rises from a value of zero when except for a change of scale given by the factor
G=0 (corresponding to the maximum voltage at Q, „i/C&0, and a horizontal displacement given
which the oscillator can be made to operate) to
by the term —Q, „,/Q, . Often in practical cases
unity when G= ~, a value which can ordinarily a good enough approximation to this relation is
not be reached; at the value corresponding to a straight line, which is not in general horizontal.
V=O, the circuit efficiency will still be less than The frequency contours are then such a family
unity. Clearly the circuit efficiency will become of parallel straight lines.
greater, as Q, becomes greater (that is, as the A plot of contours of constant power and
losses in the cavity are d'iminished) or as Q,„i constant frequency in the reflection coefficient
becomes smaller (that is, as the coupling between plane is generally called a Rieke diagram. Since
generator and load increases). This bears out the transformation from admittance plane to
our statement that increasing Q,„i to decrease . reflection coe%cient plane is a bilinear transfor-
the frequency pulling of a generator has a com- mation, the contours will be circles. The power
pensating disadvantage: it decreases the circuit contours all pass through the point of infinite
eAiciency, and hence the output power. admittance, or zero impedance, and are all
parallel to the line of zero resistance at that
5. Output of Oscillator as a Function of Load
point; thus in the reflection coeScient plane they
We have seen in the preceding section how the will be circles tangent to unit circle at the point
behavior of an oscillator depends on the imped- of zero impedance. The frequency contours also
ance of the load. The value of G, the load pass through this point, but at a different angle.
conductance, determines g, and in turn the Thus the Rieke diagram has the appearance
voltage, the electronic power, and the circuit shown in Fig. 18. It should be remembered that
efficiency, and hence the output power. As G goes this is on the assumption that Z, =O; that is,
from zero to infinity, the power goes from zero that we are measuring load admittance across a
 M I CROWAUE ELECTRON I CS

K. lKY =
CON S~ANy

STANT

FIG. 18. Idealized Rieke diagram, FIG. 19. Actual Rieke diagram.

'
plane in the output line across which the standing coincide. If the load is a resonant cavity with a
wave minimum is located when the resonance is resonant frequency near that of the oscillator,
removed, as by tuning out of the way; this plane, its input impedance will be like Fig. 13, and the
we remember, is a function of frequency. It is a corresponding reHection coefficient like that of
plane which is electrically a whole number of Fig. 14. The corresponding admittance will be
half wave-lengths away from the electronic dis- like that shown in Fig. 20. We see that there
charge, so to speak. If we take a Rieke diagram can well be more than one point of coincidence
across a fixed plane at-a distance from the tube, of frequencies on the two curves. In other words,
there are various complications which we shall it can well happen that the oscillator has a
not go into; they result in having the frequency choice of operating in two different frequencies,
contours cross in the neighborhood of a point with a single load. This is closely related to the
inside the unit circle, as shown in Fig. 19. We existence of two resonant frequencies when two
must also remember that we cannot in general circuits of about the same resonant frequency,
choose a plane across which the real as well as the tank circuit of the oscillator and the external
the imaginary part of Z, vanishes; there are load, are coupled together. The weaker the
inevitably losses included in this term. When we coupling, or the larger the external Q, the smaller
analyze the effect of these losses, we find that will be the frequency separation of the two modes
they affect the dependence of both electronic of oscillation.
power and circuit efficiency on G and B. The In Fig. 20, the straight lines represent the
result is that the output power no longer depends curves of 8 vs. G for constant frequency, as
on G alone, but also on 8, in such a way that described in the preceding paragraphs. The fre-
the power contours are deformed in the way quencies are numbered from 1 to 10. The curve
shown in Fig. 19. represents an admittance curve for the external
The Rieke diagram, or corresponding plot in resonant load, as a function of frequency, again
the admittance plane, determines the power and with the frequencies numbered. It is clear that
frequency of operation of an oscillator when the two sets of frequencies coincide, so that a
operating into a given load. Often, however, the resonance exists, for the frequencies labeled 4,
load may have an admittance which is not con- 5, and 6. It can be shown that of these resonances,
stant, but which is a function of the frequency. one is unstable, and would not really exist,
In such a case, we cannot specify the load
admittance in advance. To find the operating 6. Starting of an Oscillator
point, we may draw the locus of the load admit- We have spoken of the stable operation of an
tances for different frequencies in the admittance oscillator, in which the frequency is real. It is of
plane, draw the contours of constant frequency interest to ask, however, how the oscillations
as before, and ask for what frequency the two build up from zero amplitude. For a short time
 J. C. SLATE R
positive, so that even if we start with a finite
voltage amplitude, the amplitude will decrease
exponentially, and the electronic discharge will
act like a load, not a generator.
Clearly the voltage as a function of time is not
a simple exponential, but we can give an analytic
evaluation of the relationship. We note that if
instantaneously the voltage is increasing as
J e"2', we have a&2=d ln U/dt. Substituting from
2
3 (IV.24), and integrating, we have
6 dV
8 (IV.25)
&0
~ ~, ~.U )R(U) 1 l
FIG. 20. Admittance plane, for operation of self-excited
oscillator into resonant load.
2 ( C&o, Ql. )
where we assume that U= Uo when t =to. Over
interval during the build-up, we may assume that a range of voltage for which g may be treated as
the amplitude is increasing exponentially with a 'constant, this shows that t varies logarithmi-
the time, so that formally we may treat the cally with voltage, or voltage exponentially with
frequency as being complex, the imaginary term time, as we have already seen. This has a bearing
representing the exponential increase, and thus on the initial process of build-up. With a curve
having the opposite sign to that which we have like that of Fig. 21, in which g approaches a
previously used in discussing exponential damp- finite value when U=O, we have this situation
ing, and the dissipation of power in a cavity. of exponential build-up for small voltages. In
Thus suppose co=co& — j~2, where co& is the real other words, it would take an infinite time to
frequency, and where the amplitude is building build up from zero voltage; all we can do in a
up according to the exponential e"2'. Substituting finite time is to start with an already existing
this value in (IV. 17), we see that voltage, and amplify that. In practice, in oscil-
lators having such a characteristic, like the reflex
C02 g klystron, the oscillation starts from the noise, or
2 (IV.24) fluctuation, voltage always present in an elec-
M~ Cld~ Qg
tronic discharge. It is interesting to note that
We can see the interpretation of this equation if the rate of build-up is greater, the greater the
we use a diagram like that of Fig. 17. Plotting loaded Q, or the greater the final voltage. Quite
g/C~, against U, we see that the difference a different case is that of Eq. (IV. 19), in which
between this curve and the straight line at height the current is a linear function of the voltage,
1/Q~ determines the rate of increase of the so that the conductance becomes infinite at zero
voltage. Starting at a given voltage, in the case voltage. In that case, the term in 1/U in g(U)
shown in Fig. 21, where the curve of g/Ca&, lies cancels the factor U in the denominator . of
above that of 1/Ql„ the voltage will then increase, (IV. 25), so that for small values of U the inte-
more or less exponentially, with the time. As grand is constant, and U increases linearly with
the curves then draw closer together (assuming time. In this case it is not necessary to start from
that the curve of g/C~, slopes downward), the noise. The characteristic of having a finite current
rate of increase will Hatten off, until finally a for zero voltage is obviously discontinuous; for
steady state will be reached when the two curves it is also naturally possible to have zero current
intersect. Clearly, if the curve of g/C&o, slopes for zero voltage. Such a situation is found ap-
upward, the amplitude will increase without proximately in the magnetron, where the state
limit, and no stable operation will be possible. of zero current is inherently unstable, and breaks
Also clearly if the curve of 1/Ql, lies entirely at the slightest provocation into a state of large
above that of g/Cco„cv2 will be negative instead of current. This then builds up linearly, finally
 M I CROWAVE ELECTRONICS

approaching asymptotically to its limiting value. the oscillations. Unfortunately comparatively

In fact, if we insert the value (IV.19) for g, and little use has been made of this method for
integrate (IV.25), setting U=O when t=0, we investigating the characteristics of microwave
find at once that oscillators. Rieke and his collaborators have used
it to some extent for magnetrons, but no thorough
study has yet been made of the characteristics
RCcv. (1/RCa), +1/Qr) of any type of microwave oscillator as a function
of the various parameters, such as d. c. currents,
1 —exp ——Cd@,

(1/RCco + 1/Q') t . (IV.26) voltages, etc. , which control the oscillation. In

2 the next chapter we shall, however, find the
general form of behavior to be expected, for the
That is, the voltage increases in a way like that reOex klystron and the magnetron.
of the voltage in a condenser being charged
through a resistance in ordinary circuit theory. V. ELECTRONICS GF THE REFLEX KLYSTRON
AND MAGNETRON
The time of build-up is related to the loaded Q,
as modified by the term 1/RC"„which acts like 1. The Re6ex Klystron
an additional form of loading as far as the The resonant cavity of the reflex klystron, or
mathematics is concerned. Clearly, then, a mag- refiex oscillator, is shaped fundamentally like a
netron with a high loaded Q, or with a low short length of coaxial line, with a gap in the inner
frequency, will take a relatively long time to conductor (Fig. 22). This gap acts like a lumped
reach its full voltage. capacity, so that in the mode in which it operates,
the electric lines of force run between the faces
7. Experimental Investigation of Electronic of the gap, and the magnetic lines are in circles
Admittance
surrounding the axis of the cavity. As a result
In Fig. 17 we postulated a form for the curves of the loading by the capacity, the length of the
of g and b as functions of U, and have asked how cavity is much less than the half wave-length
the oscillator would behave with such values. In which it would be if the inner conductor were
the next chapter w e shall see that in simple continuous. On the other hand, the diameter of
cases, like the reHex klystron, we can actually the cavity is also much less than it would be
calculate these curves, and obtain a relatively for a cylindrical cavity without inner conductor,
complete theory of their operation. %ith more operating in the corresponding mode, with E
complex cases like the magnetron, however, such along the axis, H in circles around it. This may
calculation is extremely difficult, if not impos- be easily seen by considering how the frequency
sible. In these cases, we can work backward from would change if we started distorting such a
observed operating behavior, and find the curves cylinder by allowing posts to protrude from the
of g and 5-from experiment. We ordinarily meas- opposite circular faces, and approach each other.
ure two quantities, power and frequency, as The posts would push into the cavity at points
functions of the load admittance G+jB. The of high electric field; thus, by Eq. (III.89), they
output power is of course g, (~g U'). Knowing the, would have the effect of decreasing the resonant
load admittance and the circuit constants, we frequency, and to restore it to its original value
can find g from (IV. 18), and q, from (IV.22). we should have to decrease the diameter of the
Measuring the power we can then compute U,
the vo1tage. Hence we get a curve of g as a calcu.
function of voltage. Next we measure the lines
of constant frequency in the admittance plane.
From these, as in Section 5, we can find b as a
function of g, and hence as a function of voltage.
Thus the complete information can be found
experimentally. The curves so found can then U
be used in (IV.25) to investigate the build-up of FIG. 21. Conductance curve for starting of oscillator.
 J. C. SLATE R
cavity, thereby pushing in the walls in a region I'
of high magnetic field. The whole cavity is then I

I
rather small compared to a half wave-length, I
particularly so if the gap between posts is narrow I

and its capacity is high, so that even more the I

gap itself is small compared to a half wave- AXIS OF

length, and hence the field in its immediate RoTPTION
neighborhood can be treated rather accurately FK;. 22. Scheiuatic diagram of klystron cavity.
by electrostatics. It forms, in other words, a
condenser, whose capacity C can be easily found. with suitable distances, and phase separations
For a small gap and high capacity, the resonant between, it is possible to have the bunches of
frequency of the cavity can be rather accurately electrons 'reach each grid system in such phase
computed by a simple approximation of assuming as to absorb energy, and thereby to gain more
a length of coaxial line to be terminated by a and more energy with each cavity which they
capacity. traverse. This device is the linear accelerator,
The two condenser plates, instead of being by which r-f power is converted into d. c. power,
solid, are made of grids, so that the effect of the in the production of high energy particles. Our
oscillation is to impress an r-f voltage between present interest, however, is the inverse of this,
these grids. Thus the resonator forms the sim- or the oscillator, in which d. c. power is converted
plest form of diode, in the microwave region. An into r-f.
electron gun then shoots a beam of electrons, of In the reHex oscillator, one cavity and grid
a fixed energy, at the pair of grids. These elec- system is made to fulfill the function of both
trons are speeded up or slowed down in passing buncher and catcher, by having an electrode,
through the grid system, depending on the phase called the reHector or repeller, negatively charged,
of the r-f field. The resulting velocity modulation which reverses the electrons after traversing the
results, after passing through a certain drift grids once, and causes them to traverse them
distance on the other side of the gap, in a again in the opposite direction. The time of
bunching of the electrons: the electrons which transit from the grids back to the grids again
have gained energy speed ahead, those which depends not only on the dimensions, but also on
have lost energy lag behind, until at a certain the repeller or reflector voltage. If this is adjusted
distance they meet and form periodic bunches. properly, the electrons re-enter the grid in such
In the ordinary klystron, a second resonant phase as to deliver power, and the oscillations
cavity and grid system, called the catcher, is are maintained, the tube operating as a micro-
located at such a distance from the bunching wave generator. On the other hand, for other
cavity that the bunching is formed approxi- adjustments, the electrons enter in such phase as
mately at the grid system of the catcher. If an to act as a load, and the oscillation cannot main-
r-f voltage is impressed on the catcher, the tain itself; such operation can be observed only
bunches of electrons will either deliver energy, by feeding power into the cavity from outside.
or absorb it, from the cavity oscillations. If they We find that there are a number of different
deliver energy, they will maintain the oscillation, values of reHector voltage for which oscillation
and by a proper feedback system between the is possible; these differ by such amounts that
two cavities the device will act like an oscillator, the transit times of the electrons in the drift
much as in an ordinary triode oscillator, the space differ by whole periods of the r-f. The
buncher fulfilling the function of the grid circuit, power shows a maximum for each of these re-
the catcher of the plate circuit. On the other flector voltages; they are called different elec-
hand, if the phase is such that the electrons tronic modes of the oscillator. As the voltage is
absorb energy from the catcher, the oscillation changed from each of these values, the power
will not be maintained, but the electrons will decreases, and in between them oscillation does
absorb energy on the average in the catcher. By not occur. At the same time that the power
arranging a number of cavities in succession, changes, the phase of the bunched electrons
 M IC ROWA VE ELE CT RON I Cs 497

a~sacvoR of simplifying assumptions. We assume a one-

dimensional problem, neglecting all transverse
motions of the electrons, or departures of the
fields from the one-dimensional case. We assume
~ ~ ~ ~
the r-f voltage amplitude U to be small compared
Cgog to the incident voltage Vp of the electrons, so
Vp that we can neglect powers of the ratio U/Uo.
We assume the grid separation to be so small that
CATHQQF
w r
the transit time through the grids can be neg-
Fro. 23. Schematic diagram of reflex klystron electrodes. lected (it is not hard to remove this restriction,
and investigate the effect of transit time).
introduces a reactive component of r-f current, Furthermore, we neglect the possibility of colli-
which results in a change of frequency of opera- sions of the electrons with the grids, with their
tion of the tube. This is the electronic tuning consequent loss from the beam; and we neglect
which makes the reflex oscillator such a con- multiple transit electrons, which are not caught
venient power source for any purpose requiring by the electrodes a'fter their first transit through
frequency modulation. the grids.
We shall now examine the details of this Let us now consider the motion of an incident
electronic process. Our fundamental object, as is electron which strikes the grids at time to. If e
clear from the preceding chapter, must be to is the magnitude of the electronic charge ( = 1.60
determine r-f current, as a function of r-f voltage. "
X 10 coulomb), and m is its mass ( = 9.0)& 10 "
The voltage is simply that which is impressed on kgm), its incident' velocity vo (in meters per
the grids. To find the r-f current, we shall ex- second) is given by
amine the current carried by the electrons which I =eVp.
—,mvp (V. 1)
have been bunched by traveling through the
drift space. We shall find that superposed on After emerging from the grids it will have a
their d. c. current is an r-f Huctuation. It is this kinetic energy e(V0 —Vcos &uto), so that its ve-
which we must compute, and which forms the r-f locity v& will be given by
current which we must use in computing g and b. -,'nwi2 = e( Vo —V cos s&to) (V.2)
We shall find that we can get explicit formulas
for those quantities, in a certain limiting case of Solving for v&, expanding the square root by
small amplitude operation, so that we can set up binomial expansion, and rejecting higher powers
a theory of the operation of the reflex klystron, of U/ Uo than the first, we have
along the lines of that sketched in the preceding vi = vpL1 —~ ( V/ Vo) cos Nfo j ~
'

(V.3)
chapter.
The electron now emerges, with velocity v&, into
2. Electronic Operation of the ReQex Klystron the drift space, where it is acted on by a deceler-
In Fig. 23 we show the various electrodes of ating force of eV„/d„. That is, its equation of
motion in the drift space is
the reflex klystron, in a schematic manner: the
cathode, from which the electrons are emitted,
being accelerated by a d. c. voltage Up before
reaching the grid system; the grids, on which the
m —= —e V„
dv

dt
—, d,
v = vi — —
m
—(t —to).
e V„

d„
(V.4)

r-f voltage Ve&' ' is impressed, by the oscillations The electron will return to the grids at the time
in the resonant cavity; .and the reflector, which t& at which its velocity has reversed, or has
repels the electrons, the retarding voltage be- become — v~. That is, we have
tween grid system and reflector being U„. We 2mv~d
ti = to+
&

shall treat Vp and V„as positive numbers, in

case the voltages have the usual sign, and shall
take V as positive when it slows down the elec- =to+
2mvod, p
(
1
1 V
—— —cos &oto (. (V.5)
trons between the grids. We shall make a number eV„& 2 Vp
498 J. C. SLATE R
We shall de6ne the average transit angle of the returning current is
electrons, from the grid back to the grid again, —
as 8. This is co times the time required for this
I) I—
()/— (1+', (8 U-/ Up) sin pptp). (V. 10)
transit, for the case V=O, and is given by On account of the sinusoidal function in the
denominator of (V. 10), we see that I) has an r-f
8 = 2m p) vpd, /e U, . (V. 6) component of frequency (p. Equation (V. 10) is
not, however, a convenient formula from which
In terms of 8, we may rewrite (V.S) in the form
to determine the r-f component, for it expresses
(z)ty = p)tp+ 8 —p(8 U/:Up) cos Mtp. (V 7) I» in terms of tp rather than in terms of t», the
time at which the electrons return. We can
We are now ready to find the current as it easily get around this difhculty as follows. To
returns to the grids. Let the cathode current be And the r-f component of I», the standard pro-
Ip, which we take to be positive when electrons cedure is to multiply by e &"'&, and average over
flow from the cathode, or positive current flows a complete period. That is, we have
to it. Then the number of electrons per' second
striking the grids from the cathode is i= — e-& "I)(t))dt)
2~ ~
dn/dt Ip/e. — (V. S)
. period

—
p
ao p dtp
Ip e 7""dt» (V. 11)
These same eIectrons strike the grid in the re- 2x ~ dt»
verse direction on their return from the reflector 'period

in a time interval dt». Thus the returning current 07Ip

e &"'&dtp.
ls 2m.
dn I» . dn dtp —Ip dtp period
(V.9)
dt» e dtp dt» e dt» 1
In (V. 11) we have converted our integral into
one over tp instead of t~ Inser. ting t) from (V. 7),
Determining dt~/dtp from (V. 7), we see that the we have

p=
2'
Ip '

exp [——
j(cotp+8 —p(8U/Up—) cos Gptp)jdtp. (V. 12)
period

We shall introduce the abbreviations where n is integral, we then have

0 V i=Ipe-&(P-' tP)
Jg(s). (V. 17)
s= ——p)tp=P. (V. 13)
2 Vo In Eq. (V. 17) we have the formula for the r-f
current amplitude; the complex current is ie&"',
Then (V. 1. 2) becomes associated with the complex voltage Ve&"'. The
sign is so chosen that when the current opposes
i= ——
Ip —
e &' e-z'($ —z ooz p)dQ (V. 14) the voltage, or when the electrons are acting as
2m
period
a generator, i and V are in the same phase. We
recall that (V. 17) involves a number of approxi-
Using Sommerfeld's integral relation* mations, as we have enumerated above.
In addition to the r-f current p of (V. 17), we
eg'z ooz shall want the r-f admittance introduced across
pe jn@dp (V. 15)
the grids by the electrons. This is
and
I--(&) = (-1)"I-(s) (V. 16) Ip 8 — — J&(s)
* See Jahnke-Emde, g+p e j(p pz/2) (V 18)
p. . 149. Vo 2
 MICROWAVE ELECTRONICS
a function of 8, the transit angle. This in turn is
a function of the reflector voltage, as we see
from (V.6).

g, (zl 3. Power and Frequency of Re6ex Klystrons

From the curves of g and b as functions of r-f
voltage, which we have just derived, we can
carry out a discussion of the power delivered by
the tube as a function of load, and of the fre-
quency of operation, as in the preceding chapter.
The curve of 6 vs. g, which determines the nature
I 2 3 z.
of the frequency contours in the admittance
FIG. 24. Dimensionless representation of current, ad- plane, or in the Rieke diagram, is a straight line,
mittance, and power of reRex klystron, as function of as we have just seen in (V.21); furthermore, the
voltage.
slope varies with 8, or with reflector voltage.
We shall now show that we can summarize the
where we have used (V. 13). Separating real and
information about both power and frequency,
imaginary parts, we have
for all values of 0, in a single simple diagram.

g= —-cos ( ——
Ip 8 3~q Jg(s)
2) s
)
8 i
This is a figure in which g is plotted as abscissa,
b as ordinate. First we can draw a line of constant
Vp2 &
8 in this space. By (V.21), it is a straight line
(V 19)
through the origin, with a slope of —(8 —
b= — —sin —
Ip 8
I
( 8 ——
3~1 A(s)
( 2) s f
3s/2).
We notice that a horizontal line corresponds to
(8 —3m. /2) =27r times an integer, or to
Vp2

We shall also need the electronic power I', ~, 8/2m =n+-,', (V. 22)
which is -', g V', or is
where n is an integer. We shall see in a moment
( 3m/
that these values of 8 correspond to power
2i maxima of the various electronic modes, and we
el IO Up s J,(s) . (V.20) , shall refer to the corresponding modes as the
1~, 24, etc. , modes. We observe that as the
Remembering that IpV0 is the input power, we reflector voltage increases, 8 decreases, or the
may write the electronic efficiency as the quan- line rotates counterclockwise, or in a positive
tity (V. 20) with the factors IpUp left out. The direction.
three functions Ji(s), J&(s)/s, and sJ~(s) then Next we can draw contours of constant power.
represent in a dimensionless way the r-f current, First let us consider the curve corresponding to
admittance, and power, as functions of the r-f z = 0, or vanishing r-f voltage. This curve is
voltage, which is represented dimensionlessly by obviously a contour of zero electronic power, and
z. We show these three functions in the curves may be called the small signal curve. Since it
of Fig. 24, and observe that the middle one, can be shown that J~(s)/s approaches —' , as s
proportional to g, is similar to the schematic approaches zero, the equation of this curve is
curve which we gave in Fig. 17. We notice from —3m-/2),
g = (Ip8/4 Vp) cos (8
(V. 19) that the curves of g and b as functions of (V.23)
U are both of the same form; for we have b = —(Ip8/4Vp) sin (8 —3s /2).
J

That is, if we let r = (g'+b')', the length of the

(
b= —g tan 8 — l. —3sp
(V.21) radius vector, we have
I

2i
r = (I,8/4V, ) (V.24)
That is, the two quantities are proportional to
each other, the constant of proportionality being as the equation of the small signal curve, in
500 J. C. SLATER
efficiency, for the 14, 2~, and 34 modes, computed
for a case similar to those met in practice. The
contours are simple to interpret. As we approach
either the vertical line of zero circuit efficiency,
or the small signal spiral, the efficiency goes to
zero, and it reaches a single maximum in the
middle of the operating range, approximately on
the horizontal axis. For the modes of small n
values, the maximum efficiency is low because
FIG. 25. Dimensionless representation of power as function
of admittance, for reflex klystron. the maximum comes for low g, where the circuit
efficiency is low; for large n values, the maximum
polar coordinates. It is then a spiral, the larger efficiency is low because of the factor 0 in the
loops corresponding to larger 0, or smaller re- denominator of (V.20). In between, there is an n
Hector voltage. We may now consider how the value for which the peak efficiency has its largest
electronic ek. ciency varies as we go along a value, in this case the 24 mode. Clearly the
radius, from the origin to the small signal spiral. question of which mode has the highest power
We note that the radius vector is proportional to will depend on the value of Ccv /Q; the. smaller
Ji(s)/s, and that the electronic efficiency is pro- it is, the higher the efficiency of all modes, but
portional to sJi(s). To show the relationship the greatest improvement will come about in
between these, we plot sJi(s) as a func'tion of the modes of small n value.
Ji(s)/s, in Fig. 25. We see that as we go out We may now use these figures to discuss the
from the origin to the small signal spiral, the operation of the oscillator. First we consider
electronic efficiency increases from zero to a oper'ation into a matched load, and ask what
maximum, which it reaches at a value equal to will be the power and frequency as functions of
about 43 percent of the radius of the spiral, and reflector voltage Co.mbining Eqs. (IV. 18),
then decreases to zero again at the spiral. (V. 19), and (V. 21), we have
Further, on account of the factor cos (8 —3m/2)
in (V. 20), the electronic power decreases as 8
goes in either direction from the value (V. 22)
g= —
Io
—cos ( ——
tt 3iri Ji(s)
i
i s
0 =Ceo, —
( +
i i
1 G )
VO2 & 2 (Q. Q. ~i
representing a horizontal line; and the maximum
(V. 26)
power decreases as 0, or n increases. The power
is negative for negative g; that is, in the left half
plane the tube is a passive load, not an active
2(~ —~.')
co,
=— tan(
t
(
e —
2i
—+
3~y p1
))
(Q.
G

Q,
y

ii
8
Q.„,
oscillator.
Before we can draw contours of constant For a matched load, we have G= 1, 8 =0. Thus
power, we must consider the circuit efficiency. by the first equation of (V.26) the operation will
We saw in (IV.22) that this depends on G, or be at points of a vertical line, g=constant. As
on g, It is zero for 6 =0, or for the reHector voltage changes, 0 changes, and the
intersection of the radial line corresponding to
g = C~0,/Q . (V. 25)
the reflector voltage, and the vertical line corre-
As G increases, the circuit efficiency increases, sponding to the value 6= I, will give the oper-
becoming unity for large G. The vertical line ating point. As the reflector voltage changes, the
denoted by (V.25) will then be a contour of zero power will go from the maximum value corre-
power, and tube operation can only occur to the sponding to the horizontal axis in the figure,
right of this line, between the line and the spiral. down to zero, and the tube will go out of oscilla-
We can now compute the power at any value of tion, starting up again when the reHector voltage
g and b, using (V. 19) to find Ji(s)/s, from this reaches the value at which the next mode starts
finding sJi(s), and getting the electronic power up. Correspondingly, from the second equation
from (V.20), and the circuit efFiciency from of (V.26), the relation between frequency and 0
(IV.22). In Fig. 26 we show contours of constant will take the form of a tangent curve. This is
 M IC ROWA VE ELECTRON I CS 50i

the relation describing the electronic tuning of coefficient plane grows until it finally includes
the oscillator. We notice that, the smaller the the whole region of loads with positive resistive
external Q, or the tighter the coupling to the components. Outside that region the oscillator
load, the greater is the g value corresponding to cannot operate. Our formulas for g and 6 continue
a matched load. For a very tight coupling, the to be valid, however; the only difference is that
vertical line in the figure below on this page g has the opposite sign, so that the tube acts as
may well be so far to the right that it does not a non-linear load rather than an oscillator. We
intersect the spiral at all, for lower modes. These may feed power into a reflex klystron whose
modes then do not operate with tight coupling. reflector voltage is in this non-operating region,
We should realize that not only does this set a and may measure its input impedance. From
limit on the lowest mode which will operate, but Eq. (III.81) we see that the electronic term
there is also a limit set on the highest mode, by (which may be rewritten by (III.85)) has the
the condition that the reflector voltage U„must effect of modifying the apparent loaded Q, and
clearly be greater than the beam voltage Vp, or resonant frequency of the cavity. The term g+j b
the electrons will strike the reflector and be has such a sign that in the non-oscillatory region,
absorbed, rather than being reflected. From where the electrons act as a load, they increase
(V.6), this means that the maximum value of 0 the value of 1/Qi„or add to the loading; in
is definitely determined, by the condition actual cases they reduce the loaded Q to a low
value. As the reflector voltage is varied toward
2m
Q, „. = eVp ~52(e/m) V, jld, the operating range, however, the effect of the
electrons on the loading decreases, becomes zero
(m$ ' cud,
when g =0, then changes sign, and helps to cancel
(V.2~) the losses resulting from the unloaded Q of the
hei (V)
cavity. As the operating range is reached, the
showing that the lower the beam voltage Up, the apparent Q of the cavity increases without limit,
higher is the maximum usable mode. until 6nally oscillation occurs. Along with this
'
In addition to studying the operation of the change of the apparent Q with reflector voltage,
oscillator when. looking into a matched load, we there is also a change in the apparent resonant
'

may study the operation as a function of load, frequency of the cavity, as we can at once
with fixed reflector voltage. This leads us to the compute. If only a small amount of power is
Rieke diagram, as in the preceding chapter. being fed into the cavity to make the impedance
There is nothing unusual about the individual measurement, we may assume that we have a
diagrams, but it is interesting to see how they
change from one reflector voltage to another.
The slope of the lines of constant frequency in
the admittance plane is the same as the slope of
the corresponding radius in Fig. 26, being hori-
zontal at the position of maximum power, and
rapidly becoming steeper and steeper as the
reflector voltage is varied on either side of this
value. We note that G can vary from zero only
to a certain. maximum value, related to the g
value of the small signal spiral; thus operation
in the admittance plane is possible only out to a
certain horizontal line, and in the reflection
~~/4 HOOE
coefficient plane in to a certain circle. As the
reflector voltage is shifted away from the value
for maximum power, over to the edge of the
mode, the maximum value of G decreases toward FIG. 26. Ef6ciency contours in admittance for
plane,
zero, so that the forbidden circle in the reflection reflex klystron.
502 J. C. SLAYER
concentric cylinders, with a constant magnetic
field along the axis of the cylinders. The anode,
which is outside the cathode, is split into an
even number, of segments, say X, and forms part
of a resonant cavity such that, in the resonant
mode in which the magnetron is operated, suc-
Fig, 27. Schematic diagram of' two-anode magnetron. cessive segments are positively and negatively
charged. The electrons move, after emission from
small signal, and may compute the value of b, the cathode, under the action of the magnetic
and hence of the frequency, from the small signal field; an impressed d. c. electric field accelerating
values (V.23). We have then an electronic means them from cathode to anode; and the r-f field
of tuning a resonant cavity, the non-operating between oppositely charged segments of the
klystron. An electronic reactor, a reactance which 'anode. This r-f field leads to the r-f voltage V
can be tuned by electrical means, is a very useful which appears in our theory, as in the preceding
device, and in a sense the reflex klystron in its chapter. . As a result of the combination of these
non-operating range forms such a reactor. On fields, the electrons move in a complicated way
account of the very large losses which are associ- which we shall describe. They eventually reach
ated with the g, however, this use as a reactor is the anode, but not with the kinetic energy which
not of practical importance. Other types of tubes they would have acquired if they had fallen
can be designed, however, in which there is an directly from cathode to anode under the d. c.
electrically controlled reactance, without accom- difference of potential. Instead, on account of
panying large resistive losses. the interaction with the magnetic and r-f fields,
they have very small kinetic energy on reaching
4. The Magnetron the anode, so that they dissipate only a small
The multi-segment magnetron oscillator is a fraction of the input power at the anode. The
much more complicated problem than the reflex rest of the input power goes into sustaining the
klystron, for two principa1 reasons. First, its r-f oscillation, and is the electronic power, I', ~,
oscillating circuit, or resonant cavity, is of a about which we have previously spoken. We
more complicated shape, and it is considerably can compute the r-f current amplitude i, which
harder to make approximately correct theories we need in the theory; it is here not at all clear
explaining its resonant modes. Furthermore, at first sight how we are to find this current, and
there is a difficulty arising in the circuit, on it is necessary to go 'back to the definition in
account of other resonant modes than the desired terms of the integral J'J. E,dv to find how to
one, which come at nearly the same frequency, compute it. When we find i as a function of V,
unless special means are taken to remove them. we find that the general situation is similar to
The second reason for the complication of the that of the preceding chapter, so that we can
magnetron is the electronic motion. The presence compute output power and frequency as func-
of the magnetic field makes a one-dimensional tions of load as in that chapter. We also can
treatment, such as we used for the klystron, consider the dependence of power output on the
impossible, and the existence of large space d. c. parameters, the d. c. voltage between cathode
charge makes any analytical treatment of the and anode, the magnetic field, and the d. c.
motion a very difficult thing. For these reasons, current which flows. We shall now give a short
we shall not be able to give a very complete discussion of,the nature of the cavity resonator
treatment of the theory of, the . magnetron. which produces the r-f oscillations in the magne-
Nevertheless we shall be able to go far enough tron, and shall then consider the electronic
to indicate the reasons for the high efficiency and motions which lead to its operation.
high power characteristic of this type of oscil-
lator. S. The Resonant Circuit of the Magnetron
The magnetron oscillator, in its present form, For a good many years the split anode magne-
consists of a cathode and anode in the form of tron, having an anode of two segments, has been
 M I CROWA VE ELECTRON I CS
I

used as an oscillator. Treated from the stand-

point of lumped constants, the two anode seg-
ments, charged to opposite potentials, form a
condenser, and they must be connected by an
inductance, as shown in Fi:g. 27, to make them
into a resonant circuit. To increase the power of
such an oscillator, it is natural to increase the
number of circuits, having many anode segments. FIG. 28. Anode of multi-segment magnetron.
By connecting each pair of adjacent segments
by an inductance, such a system can be made to in actual cases, is much' less than the free space
resonate. The geometry then allows a large wave-length, so that, the effective velocity of
cylindrical cathode, instead of the linear cathode propagation of this wave along the z axis is much
of the split anode magnetron, and this permits less than the velocity of light. We may regard.
the How of a large anode current, with conse- the field drawn in the figure as a standing wave,
quent high power. If the magnetron is to be used a superposition of two traveling waves moving
in the microwave range, it is natural to use in opposite directions along the +s axis. We
'
resonant cavities instead of lumped constants can show in a very simple manner that if the
for the inductances, and the problem of heat effective wave-length along z is less than the free
dissipation requires that the anode be made of space wave-length, the field must fall off expo-
solid metal. The simplest structure embodying nentially as we go away from the anode; this
these principles is shown in Fig. 28, and was the expresses the way in which the fringing field falls
type of magnetron which, at the beginning of off with distance. To see this most simply, we
the war, first showed the possibility of really may regard the interaction space as a wave
large power generating capacity. The larger part guide, with propagation along the s axis. Using
of the capacity between adjacent anode segments the expression (I I.9) for the relation between free
is here concentrated in a slot between the seg- space wave-length, guide wave-length, and cut-
ments, and a hole in a solid copper block forms off wave-length, we see that if the guide wave-
the inductances. Magnetic lines of force from the length is smaller than the free space wave-length,
inductances thread through the spaces above and as it is here, the cut-off wave-length must be
below the block, to complete the magnetic imaginary, corresponding to an exponential
circuit. rather than sinusoidal variation of the field at
The problem of the nature of the field in such right angles to the direction of propagation. Such
a cavity may be divided into two parts: the a, variation of course cannot satisfy the boundary
field in the interaction space (that is, the cy- conditions in a closed wave guide, for if the field
lindrical region between cathode and anode, in iricreases exponentially as we approach the anode
which the electronic discharge is located) and the surface, there will be a tangential electric field
field in the separate hole and slot resonant ele- at that surface, which would be impossible if we
ments. The first of these problems can be handled had a conducting surface for the anode. In our
by solving Maxwell's equations in cylindrical case, however, the anode surface is broken by
coordinates. We may get an approximate physi- the slots leading to the resonators, a,nd there
cal idea of the problem, however, by imagining can be a tangential component of electric field
the anode and cathode flattened out into planes, across imaginary surfaces closing these slots. We
as shown in Fig. 29. Then, for the mode in can, in fact, get. a simple and fairly accurate
which we wish to operate, the electric lines of approximation to the actual solution by com-
force will run as in Fig. 29: the field will be the puting the ratio of tangential B to tangential H,
fringing field of the slots, regarded as condensers. or the impedance, along the anode surface, from
The field will be periodic along the direction the solution holding in the interaction space, and
parallel to the anode and cathode faces (the s equating this to the corresponding input imped-
direction in the 6gure), the effective wave-length ance of the resonant cavities.
being twice the spacing of the segments. This, For a wave traveling in the +s direction in
504 J. C. SLATER
nent of E in each slot, while there is no tangential

QD component along the metallic segment. The

average field is then d/D times the field in a slot.
The current, and hence the magnetic field, how-
ever, are continuous along the face of the anode,
so that they are not affected by the slots; current
flows along an anode segment, into a resonator,
FIG. 29. Electric lines of force in linear magnetron. out again, and along the next segment, as if the
slot were not there. Thus the average impedance
the interaction space, we find easily that of the surface with the slots is (d/D)Z„(where
solution of Maxwell's equations is to make this expression comparable to (V. 29) we
;7 must assume that the height of the anode is
8 =Ae e &"' &') 8 =0 Z = —j-Z unity; if it is not, it is simple to correct for it).
We may then equate this quantity with the
value of (V.29). Writing Z, = jX„ this gives
easily
where
— =—
j. 1
11+(&-d/D)'(«/Po) )"' (V 31)
&'/c' = P' —&' a/c = 2~/go, P = 2x/Xg. (U. 28) Xg Xp

The standing wave formed by superposing two Remembering that the input reactance X„of a
traveling waves like (V. 28) in opposite directions resonator acts like an inductance at long wave-
has much the form of that shown in Fig. 29. At lengths, or is proportional to the frequency, we
the cathode, the field is small, if it is large at the see that, for large values of Xo, (V.31) shows us
anode, on account of the factor e&, which in- that), = A, p. As the frequency increases, how'ever,
creases rapidly as we approach the anode; by and we approach the first resonant frequency of
superposing a similar wave in e ~', we can make the resonator, X„ increases to infinity, so that
the tangential field exactly zero at the cathode, 1/X, becomes infinite. It. is this first resonance
without making app'reciable change in the field which concerns us at present, so that we need
near the anode. At the anode, there is clearly a not consider further resonances. We may then
tangential component of B, Z, . The wave could easily plot 1/X, as a function of 1/XO. . For our
be propagated in a guide, if the anode surface purposes, as we shall see presently, it is better
had a surface impedance, given by the ratio to plot 1/Xo as a function of 1/X, . We show this
E,/II„, or— function in Fig. 30. It is clear that as the fre-
quency approaches the resonant frequency, and
1/X, becomes large, or the guide wave-length
as we find easily from (V. 28). The series of slots, becomes small, we approach the situation which
with their attached resonators, can simulate a is actually present in the interaction space of the
surface with this impedance. Looking into one magnetron.
of the slots, a single resonator will have an input A formula of the type of (V.31) takes no
impedance which may be written account of the fact that the anode is made of

Z„=Zp Q
1/Q-»,
iT(~/~. ) (~./~)
.
as we have seen in previous chapters, where Zp
j— (V.30)
segments of finite size, and is a periodic structure.
When we consider this fact, the theory becomes
much more involved. We must represent the
field in the interaction space, not by a single
wave like (V. 28), but by a superposition of. an
is the characteristic impedance of the slot, re- infinite number. of waves, satisfying certain
garded as a transmission line. If we now have periodicity relations, so that a superposition of
resonators spaced a distance D apart, and if each them, with appropriate coefficients, can actually
slot has a width d, we shall find in the anode satisfy the 'boundary conditions at the anode
surface that actually there is a tangential compo- surface, with zero tangential E, along the seg-
 M I CRovVA VE ELF cTRQN I CS
ments. Carrying out such a calculation, we find discrete values. Such values are shown in Fig. 31,
two differences in the results, one minor, the for the case of eight segments. For this case,
other fundamental. The minor result is a small 2+R=SD, where D is the width of a segment.
change in the curve of 1/Xo as a function of 1/X„ For the maximum value of n, we have 1/2D
for all values of 1/X, . The fundamental result is =e/8D, or n=4; in general, as we see from this
that the curve now becomes periodic in 1/X, . example, the maximum value of n is X/2. We
We find that there is a minimum value of the now see from the figure that each value of n
guide wave-length for which we can have a gives a different frequency. Thus we have made
solution: twice the distance between slots, or 2D, an approximate calculation of several of the
using our notation. If the guide wave-length has resonant frequencies of the cavity. Experimental
this value, we have just the solution shown in measurement, or more accurate theory, shows
Fig. 29. If we try to make the guide wave-length that this type of theory is qualitatively correct.
less than this, or 1/X, greater, we find that we It is now obvious from the curve that the modes
merely repeat the solution already found for a n = 3 and n = 4 lie close together. Since it is n = 4
smaller value of 1/X, . The situation is identical in which we wish to operate, this means the
with that met in the theory of the weighted presence of a disturbing mode, and for this reason
string, or of electric filters composed of a succes- this type of anode is not very satisfactory for
sion of identical four-terminal networks. As a actual magnetron operation.
result of this, we find that the true curve of 1/P o It would take too long to go into the theory of
as a function of 1/X, has the form shown in the various means which have been used for
Fig. 31, resembling the curve of Fig. 30 for small separating the resonant modes of the magnetron
values of 1/X„but then becoming periodic. The anode, so that the desired mode (which is often
value of 1/P, corresponding to the maximum called the x mode, since in it the phase of each
1/Xo, or the maximum frequency, is then the segment differs by x from that of the adjacent
type of oscillation in which we are interested. segment) shall have no other modes very close
In the solution of the type shown, we can to it. The simplest method is called strapping,
compute the frequency for a wave of arbitrary and may be described in one of its forms as
guide wave-length. The fact that the curve follows. In Fig. 29, we imagine two parallel wires
connecting frequency with 1/X, is not a straight or straps located over the segments of the anode,
line shows that there is dispersion; the velocity parallel to the anode 'surface. One of these straps
of propagation is a function of frequency. For a is connected, by short wires or posts, to all the
linear structure, like that of Fig. 29, any wave- segments marked + in the figure, and the other
length and any frequency would be possible. strap to all segments marked —.It is then clear
For the cylindrical magnetron structure, as that, in the mode shown, one strap will be
shown in Fig. 28, however, the situation is quite positively charged, the other negatively. Thus
different. The anode now closes on itself, and is the strap will act like an added capacity in shunt
of finite length; since the field must be continuous with the capacities of the slots, hence increasing
in going completely around the anode, we see the capacity, and decreasing the resonant fre-
that the circumference must be a whole number quency of the m mode. For the other modes,
of wave-lengths. This can be automatically
handled by using the correct solution of the
problem, in terms of Bessel's functions, but we
can treat it approximately, from our linear
model, merely by demanding that the circumfer-
ence 2xR, divided by the wave-length X„should
be an integer n. That is, we have,
1/) 0 —
n/27rR— (V.32)
In other words, we do not have all values of
FIG. 30. Frequency as function of reciprocal wave-length,
1/X~ allowed,but only a set of equally spaced for linear magnetron.
r
 J. C. SLATER
6. Electron Motions in the Magnetron
Now that we have a fairly correct picture of
the r-f field inside the interaction space of the
magnetron, we can ask how electrons move, in
the combined d. c. electric field, along the x axis
i/io
of Fig. 29, the magnetic field along the y axis,
lli4
n» 2. 3 and the r-f field. Here, as in discussing the
FI(y. 31. Frequency as function of reciprocal wave-length, resonant oscillations, it is convenient to use the
for' eight segment magnetron.
linear model with a plane cathode, rather than
the actual cylindrica case. As a first step, we
however, each strap is charged positively over consider the motion of an electron in the d. c.
part of its length, negatively over the rest, and fields, without the presence of the r-f field.
thus the system of straps is less effective as a An electron of charge — e, moving in a con-
condenser. When we 'examine the theory, or stant field of magnitude E. accelerating it along
measure the position of the modes experimen- the +x axis, and a magnetic induction B along
tally, we find that the displacement of modes to the y axis, has an equation of motion
lower frequency, or longer wave-length, decreases
regularly as we go from the x mode to modes of mS' = eE+eBe, my = 0, me = eBi. —
(V.33)
lower n value. This effect is much greater than
We may satisfy the second by setting y=0, and
the mode separations present in an unstrapped '
the third by setting nsi= — eBx, which is con-
magnetron, which are shown in Fig. 31, and
sistent with the initial condition that the electron
which are in the opposite direction. Thus the
order of modes is reversed, and in a strapped
i
starts from rest, or has =0, i =0, when it leaves
the cathode, which we take to be x=0. Substi-
magnetron the m. mode has the lowest frequency-,
tuting in the first equation, we then have
or longest wave-length, the frequencies of the
other modes coming at higher and higher fre- eB
quencies as n decreases, and being separated far
e
i+(a~'x = E, wh—
ere (u~ —— —. (V.34)
m m
enough from each other so that there is no
interference between them. Solutions of these equations, satisfying the
The reason why other modes are undesirable initial conditions, are
in a magnetron is mainly that with two modes
nearby, the electronic discharge may be unable mE
to decide which of the two modes to operate in, x= (1 —cos (oII/),
eB'
and there may be power emitted in two or more (V.35)
frequencies. It is unlikely that simultaneous Bt mE
e= + sill GDHf,
operation in two modes is possible; more often, 8 8B
in a pulsed magnetron, either successive pulses
are in different modes, or the magnetron shifts where t is the time measured from the instant
mode during a pulse from one mode to another. when the electron leaves the cathode. We see
Either of these phenomena decreases the power that the electron describes a circle, of radius
in the mode in which operation is desired. These mE/eB', with angular velocity &o~, about a point
difhculties of modes have been among the most x=mE/eB', s= Et/B. We readily se— e that the
troublesome and least understood features of resulting path is a cycloid. For we remember
magnetron operation and construction. We have that a cycloid is the path made by a point on the
not time to go into them further, and shall rim of a rolling wheel. The wheel is rotating
assume in our discussion of the electronic motions about a point whose height is its radius, and
in the magnetron that the resonant cavity has whose linear velocity. is the product of the radius
only one mode, the x mode. The simple theory and the angular velocity, as our value E/B is the
of Eq. (V.28), and Fig. 29, give a fairly satis- product of (mE/eB') and (eB/m) Thus the path.
factory description of this mode. is as shown in Fig. 32. The maximum height of
 M IC RO% A VE ELECTRON I CS 507

the orbit above the cathode is 2mE/eB'. If this

distance is less than the distance to' the anode,
the electrons will never reach the anode, and the
magnetron will not pass a d. c. current. If how-
ever the height is more than this, all electrons
leaving the cathode will reach the anode, so that FIG. 32. Cycloidal orbit of electron in magnetic and
electric 6elds.
there will be a d. c. current. The value of voltage
for a given magnetic field, or magnetic field for
We can, as a matter of fact, satisfy resonant
a given voltage, for which the electrons just conditions similar to this for the other modes of
reach the anode are called the cut-oR voltage the resonant cavity. From (V.32) we see that
and cut-oR magnetic field respectively. The
the wave-length of the disturbance is inversely
range of values of E, and 8 for which the magne-
proportional to n, the index of the mode; thus
tron can act as an oscillator are those for which
the velocity varies in the same way, and we have
the electron would fail to reach the anode in the
static case; in fact, ordinarily those for which it 8 n, 2D
fails by a large margin. That is, it acts as an C) (V.37)
oscillator for magnetic fields large compared to
8 n 'Ap

the cut-o8 value. where n, is the maximum value of n, which we

With this knowledge of the motion of an have for the x mode. This relation. is not quite
electron in the static electric and magnetic fields, as simple as it seems, for we must remember
it is not hard. to see how electrons move in the that A, p, the free space wave-length connected
actual field of the magnetron. In the type of with the frequency of oscillation, is itself a
operation which leads to an efficient oscillator, function of n. For the unstrapped tube, as shown
there proves in the first place to be a resonance in Fig. 31, 1/Xo is proportional to n for small n
relation occurring between the average drift values, so that E/B is constant for these values,
velocity of the electrons, on account of the and in fact is approximately equal to the velocity
motion of the center of the cycloid, and the of light; it is only for the higher n values that
velocity of the electromagnetic wave in the inter- 1/Xo becomes small enough, and neo big enough,
action space. We have seen in the preceding to reduce the velocity well below the velocity of
section that this wave consists of a superposition light. On the other hand, with strapped tubes,
of two traveling waves, traveling in opposite we have seen that Xo becomes considerably
directions. Each of these waves travels with a greater than for unstrapped tubes, so that the
velocity less than the normal velocity of light; velocity becomes considerably less than the
in fact, in practical cases, much less, perhaps a velocity of light.
tenth of the velocity of light or less. The drift We have stated that the resonance condition
velocity of the electron, by (V.35), is 2/B. The must be satisfied, according to which the drift
resonance condition we have mentioned is that. velocity of the electrons equals the velocity of
these two velocities should agree. That is, there the traveling wave. Now we must ask why this
is a linear relation between P and 8 required for is so. If the condition is satisfied, then as the
good operation of the magnetron. We can readily electron moves along, the r-f field which travels
find the constant of proportionality. If D is the along with it will appear to it to be a constant
spacing of the segments, the effective wave- field. Then we can immediately find its motion
length of the disturbance in the interaction space in this field. We need merely take the vector
must be 2D, so that the ratio of wave-length to sum of the d. c. field, and of the r-f field which
the free space wave-length is 2D/Xo, and this appears to be constant, and find the motion of
must equal the ratio of the velocity, to the an electron in this constant field. If the phase of
velocity of light. Thus we must have the electron is such that the r-f field following
along with it is in the x direction, or from the
E 2D
anode to the cathode or vice versa, then this r-f
c (V.36)
B )p field will merely add in magnitude to the d. c.
 J. C. SLATE R
field, and the result will be that the drift velocity for the r-f force will have worked on them during
of the electron will be increased or decreased. the one cycle of their motion, and they will have
On the other hand, if the r-f field is along s, or accumulated some kinetic energy, instead of
at right angles to the d. c. field, then the vector returning to the cathode with exactly no kinetic
sum of the r-f and d. c. field will be in a different energy, as in the static case of Fig. 32. This
direction from the d. c. field. The motion will dissipated power is far less, however, than that
then be similar to the cycloid of Fig. 32, but the delivered to the r-f field by all the other elec-
direction will be different. Depending on the trons, which finally reach the anode.
phase, the path will either carry the electron We can make a simple approximate calculation
down into the cathode, in which case it will of the efficiency to be expected from the magne-
strike the cathode after its first revo'lution, and tron, as a function of the d. c. voltage and mag-
be lost; or it will carry it away from the cathode, netic field. Let the distance from cathode to
in which case the electron will move in a diagonal anode be X. Then an electron gains potential
path toward the anode. It is electrons of this energy of eEX, or of eVO, where VO=BX is the
type which act as generators. For they move d. c. voltage, in falling from cathode to anode.
from cathode to anode, and yet at the anode Part of this appears as kinetic energy of the
they do not have a very large kinetic energy. electron, which is dissipated in the collision with
Since they would have acquired a large kinetic the anode, and lost. The rest is delivered to the
energy falling from cathode' to anode in the r-f field. We can easily find the average kinetic
absence of an r-f field, it must be that the energy of the electrons. The electrons are most
remaining energy was used up in working against likely to strike the anode when at the top of
the r f field, and therefore must have been their cycloidal path. At such a point, we can
delivered to the resonant circuit as electronic find easily, from (V.35), that the kinetic energy
power. Thus it is not surprising that the magne- is 2m(Z/B)'. Thus the energy put into an elec-
tron can have high efficiency. tron by the d. c. field is e Vo, the energy dissipated
Let us first consider electrons leaving the is 2m(Z/B)~, and the useful energy delivered to
cathode in different phases, and see how they the r-f circuit is the diHerence of these quantities.
act. Referring to Fig. 29, we see first that those The electronic efficiency, or ratio of electronic
electrons for which the r-f field is in the direction poorer to input power, is thus (e Vo — 2m(B/B)')/
of the d. c. electric field, and which therefore e Vo. This may be rewritten in the form
drift faster than the velocity of the field, will
then catch up with the field, and enter a region g, i ——1 —(2m/e) (Z/B'X). (V.38)
where the r-f force is opposite to the drift
velocity. This combines with the d. c. force to We may put this in a convenient form in terms
of the cut-off voltage and magnetic field. At
give a net force such that the electron drifts
toward the anode, in the manner described cut-off, the maximum height of the cycloid,
above. On the other hand, those electrons which 2mB/eB', equals the distance X from anode to
have the r-f field opposite to the direction of the cathode. Thus we have
d. c. field will tend to be slowed down, and will
likewise enter a region where the r-f force is
(2m/e) (8,/B, ') = X, (V.39)
I
opposite to the drift velocity. There is, in other
tvords, a tendency toward bunching of the elec-
where E„B, are cut-off electric and magnetic
fields. Substituting this value for X in (V. 38),
trons, into the phase in which they can operate and using the fact that the cut-off voltage is
as efficient generators of power. The only elec- B,X, we have
trons which do not work in this way are those
which start out with a component of r-f force in
the direction of their velocity. They will immedi-
ately start to drift further toward the cathode, This very simple formula is surprisingly accurate
and will be lost at the end of their first cycle. for giving the general form of the electronic
They will dissipate some power at the cathode; efficiency of a magnetron. It shows that the
 M IC ROKA V E ELE CT RON I CS Sos

efficiency is zero at cut-oR, but increases as the the orbit, and hence the drift velocity of the
magnetic field is increa, sed or the voltage de- electrons, will be roughly proportional to the
creased, so that we are farther and farther from r-f voltage. The d. c. current will then be propor-
cut-off. Furthermore, no upper limit is indicated tional to the density of space charge times the
for the efFiciency; it appears from (V.40) that drift velocity. If the discharge is space charge
it can become arbitrarily close to unity, for limited, a'more elaborate discussion than we can
sufficiently high magnetic fields. As far as is give here indicates that the total space charge
known experimentally, there is nothing to contra- density is roughly independent of the r-f voltage.
dict this feature of the theory. Magnetrons have Thus we conclude that the d. c. current will be
been operated with electronic efficiencies in the proportional to the r-f voltage. Furthermore,
neighborhood of 90 percent, in striking contrast the nearer the d. c. voltage is to cut-oR, the
to reflex klystrons, in which efficiencies of a few larger will be the d. c. current. This is partly on
percent are common. account of the natural increase of space charge
density as the voltage increases, since the space
7. Operating Characteristics of the Magnetron charge limited current would be proportional to
the 3/2 power of the voltage in the absence of a
We have seen in the preceding section some- magnetic field. It is partly also on account of
thing of the type of electronic motion in the the fact that as we approach cut-off, the cycloids
magnetron, and the physical reason for its high become nearly as large as the distance between
efficiency. To proceed further, however, we wish cathode and anode, and a small drift of the
to find the r-f current as a function of r-f voltage, cycloid toward the anode will cause the electron
to find the quantities g and b, and to discuss to strike the anode. Using these principles, we
operation as a function of load. Furthermore, may then deduce that the curves of d. c. current
we wish to understand the particular sort of as a function of d. c. voltage, for a variety of r-f
operating curves convenient for practical dis- voltages, and a given magnetic field, have the
cussion of magnetron operation, in which we are form shown in Fig. 33. At zero r-f voltage, no
concerned with the relations between d. c. cur- current theoretically will flow until the voltage
rent, voltage, and magnetic field. The ordinary reaches its cut-off value. At this point the current
experimental test of a magnetron is conducted will suddenly jump to a value determined by
with a matched output, and consists of an space charge theory, as influenced by the mag-
observation of the relation between d. c. current netic field. It will then increase rapidly with
and voltage at diferent values of magnetic field increasing voltage, not just according to the 3/2
as a parameter. The power, or efficiency, are power law given by space charge theory, but
observed as functions of current and voltage. somewhat more rapidly, as one can find by
The commonest type of plot, the performance considering the effect of magnetic field. With a
chart, has d. c. current as abscissa, d. c. voltage finite r-f voltage, current will begin to How at a
as ordinate, and consists of lines of constant smaller d. c. voltage, and will increase rapidly
magnetic field and of constant power or effici- with d. c. voltage. The exact form of the curves
ency, in such a plane. We shall now try to is not known either from detailed theory or from
understand the theory underlying such a per- experiment, but it presumably is much as shown
formance chart. in the figure. We can see clearly from the figure
First let us inquire what is the d. c. current that at constant d. c. voltage the current increases
Howing to the anode, with a given d. c. voltage, with r-f voltage, as it should.
magnetic field, and r-f voltage. It is clear from A graph like that above gives only part of the
the discussion of the preceding pages that the information we need to interpret magnetron
r-f voltage makes electrons drift across from operation. We need also a curve giving r-f
cathode to anode, even when the magnetic field current as a function of r-f voltage. By knowing
is beyond cut-oR, or when the d. c. electric field that, and using the principles of the preceding
is so small that it by itself would not carry work, we can find the r-f admittance g+j b
electrons to the anode. Furthermore, the tilt of coming from the electronic discharge. Then if we
510 J. C. SLATER
voltage, the electrons tend more and more to
b. C. C LIAME gPPCE CHARCK
QINITA'I ION
drift straight across to the anode, the angle of
tilt of the orbit getting greater and greater. They
are then less favorably inclined to the r-f field,
LARg FF and as a matter of fact when calculations are
VOLTA'
made the r-f current proves to decrease, appar-
ently approaching zero for r-f voltages suffici-
ently great. This applies only to the component
of r-f current in phase with the r-f field; the
L R.F. VOLTAQP component out of phase, the reactive component,
D. C. VOLTA@ E
is large at all values of r-f voltage, and results in
'
FIG. 33. D.c. current as function of d. c. voltage, for a reactive tuning of the magnetron. The problem
magnetron with various r-f voltages. of determining the Rieke diagram of a magne-
tron, then, is not greatly diferent from the
other cases we have previously considered, and
know the load, we can find the r-f voltage. we need not give a separate discussion for it.
Going back to the curve of Fig. 33, this locates The curve of r-f current vs. r-f voltage of
us on a definite curve of d. c. current vs. d. c. course will depend on the d. c. parameters. As
voltage, so that if we know one of these quanti- the d. c. ; voltage incr'eases, the space charge
ties, we can find the other. We must then con- density will. increase, and the r-f current will
sider the r-f current Rowing in the magnetron. increase, as does the d. c. current. Thus we find
This is less simple to understand than in the that curves of r-f current vs. r-f voltage, for a
reflex klystron; we must really go back to the set of d. c. voltages or of d. c. currents, have the
fundamental formula involving J'J
E,dv to find appearance of the curve of Fig. 34. We can read
what it is. The quantity Z is the field in the off from these curves the decrease of r-f current
mode we are interested in; that is, it is the with increasing r-f voltage at a constant d. c.
fringing field as shown in Fig. 29. The current voltage, the r-f current approaching a constant
density in the neighborhood of this fringing field, value at small r-f voltages; and also the rapid
as we have seen in our earlier discussion, is decrease of r-f current, at a given r-f voltage,
located in bunches, in the part of the 6eld with decrease of d. c. voltage. Furthermore, we
opposing the direction of drift of the electrons. see the corresponding decrease of d. c. current,
The component of current density in the direc- and as a result see that the lines of r-f current as
tion of this field is parallel to s. That is, it is a function of r-f voltage, at constant d. c. current,
proportional to the product of component of are much more sloped than those at constant
drift velocity in that direction (which is relatively d. c. voltage. The ordinary Rieke diagrams of
independent of r-f voltage) and charge density magnetrons are usually taken at constant d. c.
(which is also roughly independent of r-f voltage) . current, and it is accordingly these strongly
Thus it arises that there can be a large r-f sloping curves which must be used in discussing
current even for very small r-f voltages. As long them.
as the r-f voltage is large enough to produce
bunching, which may require only a relatively
few volts, the bunches will have as much charge
density, and will be moving with the same drift
velocity, as if the r-f voltage had been much
greater. This is the situation which was men-
SN LAQQE D. C.
tioned in earlier sections as the peculiar property VO CuRRE, NT

of the current in the magnetron, in which a small

r-f voltage almost discontinuously produces a
large r-f current. As the r-f voltage increases, Frc. 34. R-f current as function of r-f voltage, for
however, and becomes comparable to the d. c. magnetron with various d. c. voltages and currents.
 M IC ROQlA VE ELECTRONICS
From these curves of r-f current as a function HIGQ HhgNKTC
FIELO
of r-f voltage, we can now find the r-f voltage
with which the tube will operate, with a given
load and given d. c. conditions. If we know the
ER
load admittance, then by methods that we have
often used we can find the ratio g+jb of r-f
current to r-f voltage. The curves of Fig. 34
show the component of r-f current in phase with
the voltage; thus a line of constant g is a straight
line through the origin. Knowing the load ad-
mittance, we draw such a straight line. Its inter- EQ

section with the line of constant d. c. voltage, or

constant d. c. current, appropriate to the circum- LOw NACiNQTlc
I=IELQ

stances, shows the r-f voltage and current at 0- C. C URBE.NT

which the tube operates. One corollary of this is
FIG. 36. Magnetron performance chart.
clear. For a given load, we are always operating
on a given straight line through the origin. We
then see, from the way the curves are arranged, r-f voltage is large, a,nd the second for tight
'
that as either the d. c. voltage or the d. c. current, coupling). We see that the increase of g decreases
increases, the r-f voltage and r-f current will the r-f voltage at constant d. c. current, and
increase. Since the electronic power is propor- hence increases the d. c. voltage. A curve of
tional to the product of r-f current and r-f constant load, and constant magnetic field, is
voltage, the power will also increase with increase then, as this figure shows, approximately
of d. c. voltage or current. All these considerations
straight line, starting at small d. c. currents from
assume that the magnetic field is constant during
a value of d. c. voltage somewhat below cut-off,
the discussion.
and with the d. c. voltage increasing slightly with
We can now return to Fig. 33, and find the
increase of d. c. current. Of course, as the mag-
curve of d. c. current vs. d. c. voltage, not for
netic field is changed, the d. c. cut-off voltage
constant r-f voltage, but for constant load, as
changes, increasing as the magnetic field in-
in an ordinary test of performance. Since the
creases, so that the line representing operation
d. c. current increases rapidly with r-f voltage,
the curve will cut across the curves of constant at constant load shifts to higher d. c. voltage
r-f voltage, falling at higher and higher r-f with increase of magnetic field. We thus see the
voltage, and higher d. c. voltage, as the d. c. origin of the performance chart of the magnetron,
current increases. That is, the curves of constant which takes the form shown in Fig. 36. It is
load will look as in Fig. 35, where we give two usual to interchange the axes, plotting d. c. volt-
curves, one for small g, one for large g (that is, . age as a function of d. c. current, and drawing
the first one is for a loose coupling, for which the lines of constant magnetic field (and constant
load) on the chart. In addition, the lines of
constant power are plotted on the chart (and
O. C CURREN f sometimes also the lines of constant efFiciency).
We can find the power from the information
0 CO%OFF already stated. Since we have seen that, at
constant magnetic field and load, the power
continuously increases with increasing d. c. volt-.
~PJ'S RF. VQ
-SrWLL R, F VOLTAGE
age or current, the lines have the general form
Sl1AI L LARt"E
shown in the figure. To find the ef6ciency, we
'D. C VOLTA% K
must divide the power (which will be the elec-
FIG. 35. D c. current
tronic power times the circuit efFiciency) by the
as function of d c. voltage, for
magnetron with various load conduqtances. input power, the product of the d. c. current and
 J. C. SLATER
d. c. voltage. When we do this, we find that the come into the operation, in particular the inter-
efficiency has its maximum values at low d. c. ference of other modes. These interferences have
currents, but high voltages, as shown in the the eRect of decreasing the power and efficiency,
figure. The curves shown above agree with the so that as a rule the eSciency, rather than
actual performance charts of magnetrons, except having its maximum value for low current, is
in the very low current region. In that range, as low at that point, rapidly increases as the current
we see from the preceding page, the r-f voltage increases, and then decreases again with still
is very small, and many disturbing features can further increase of current.