TECHNICAL SECTION

SIMPLIFIED SIGNAL FLOW

GRAPH ANALYSIS

NICHOLAS KUHN
HEWLETT-PACKARD COMPANY

Palo Alto • California

The use of signal flow graphs as a means of Wrttll1g ing the scattering matrix method of writing the network
and solving linear microwave network equations is direct, equations.5 Figure 1 (a) shows a two-port network with
simple and should be of value to all who work with wave 0/ entering POrt 1 and wave a, entering port 2. The
transmission line problems. Mason'·2 developed flow emerging waves from rhe corresponding ports are repre­
graphs mainly for linear circuit analysis with some appli­ sented by b/ and b�. The relationship of the emergen'
cation to transmission lines.2 Hunton3 has .described the waves to the incident waves is written as the linear equa­
usefulness of signal flow graphs as applied to the scatter­ tions
ing matrix to solve transmission line problems, especial­
ly microwave measurement problems. Signal flow analy­
sis, however, has not been widely applied to transmission ( 1)
line problems. The trouble appears to be not in formu­
lating the problem but in arriving at a solution via the
nontouching loop rule. To exercise this rule, the micro­ These are the scattering equations and the Smn's are the
wave engineer must observe the various paths a signal scattering coefficients.
takes from the source to the sink and then observe all
Figure 1 (b) shows a signal flow graph representation
of Figure l(a) and E qua tion
the possible reflections (in a transmission line system)
(1).
which occur and then insert these in a gain formula (the
nontouching loop rule).1.2.3.4 The occasional user of this
rule must review the notation of the formula and he is
not always confident that he has considered all the possi­

b2
ble paths from source to sink or all the re-reflections.
The purpose of this paper is to make known, in a con­

HL---_----lH =
cise form to the microwave engineer, methods of manipu­
lating the flow graph to arrive at the desired answer
without applying the troublesome nontouching loop rule.
°2
The topological manipulations or reductions here des­
cribed are simple and clear enough to be remembered or Figure l(a) - General microwave two-port network showing
kept for easy reference and to earn the confidence of incoming and outgoing waves.
even the occasional user. The sequences to be followed
keep the physical picture in the user's mind as he manipu­
lates the flow graph. The topological rules are present
in the works of Mason1,2 and of Chow and Cassignol.4
This article will discuss the construction of flow graphs,
give the flow graphs of some common transmission line
elements, discuss the topological rules and give some
example problems.

Construction of Signal Flow Graphs

The use of signal flow graphs for the solution of crans­

mission line problems is most easily shown by consider- Figure l(b) - Signal flow graph of the same network.

November, 1963 59
 a a
�
L
t PLej.p
)

b
r-CJ � :
b

( a)
01 b2 01 e-je b2
� - ) • )' • :-

�
bl· e-jB
- --- '" • E • '"
bl 02 °2

( b)
Figure 2 - Cascaded flow graph of a two-port nern'ork and n a
three-port network. -

b
�
b

(c)
01 I+ ej.p b 2
b2 p
On this graph each port is represented by two nodes.
01

:I i.
Node an represents the wave coming into the device from
�
I
- �
another device at port 1Z and node bn represents the j
� pe pe
wave
-
,
-

b l I+pej.p °2
leaving the device at POrt n. A directed branch runs �

from each a node to each b node within the device. Each bl °2

of these branches has a certain scattering coefficient from -y
Equation ( 1) associated with it. This coefficient shows
p ej.p
Y+2
=

how an incoming wave gets changed to become an out­

going wave at the b node. Because they act on both the (d)
phase and amplitude of incoming waves, the scattering co­
efficients are complex numbers. The value of a wave at a
b node when waves are coming in both a nodes is the 01
�
superposition of the individual waves arriving at b from
each of the a nodes considered separately. Because the
�

bl
value of the wave at � relies on superposition the tech­
nique described: here pertains only to linear systems or

(e )
even, perhaps, to time varying systems where superposi­
tion still applies.
By comparing Equation (1) and Figure l(b) it is seen
b

when Port 2 is terminated in a match ( al = 0 ) . Similarly

that Sl1 is the reflection coefficient looking into port 1

S22 is the reflection coefficient looking into port 2 when : :( �

-
port 1 is terminated in a match (al = 0). S21 is the trans·
mission coefficient from port 1 to port 2 when port 2 is
-
o

terminated in a match (t1l = 0 ) . S12 is the transmission (0

coefficient from port 2 to port 1 when port 1 is termina­
ted in a match (a) = 0) .

Networks are cascaded by joining their individual flow

graphs as in Figure 2. Notice how tI} is synonomous
with b2 and how b'} is synonomous with al. This can be
shown by a connecting branch of value 1. Figure 2 also
shows how a three·port network may be represented. In
a similar manner the constructions may be extended to Figure 3 - Block diagram and flow graph representation of
a device with n ports. Examples of cascade flow graph (a) termination, (b) length of transmission line, (c) detec·
tor, (d) shunt admittance, (e) series impedance, (f) genera­
construction are given in the last section of this article. tor, (g) Jossless, bilateral two-port network.

Flow Graphs of Common Transmission Line Elements

Some transmission line elements which are frequently cient. Figure 3 (b) shows a length of loss less transmission
encountered have their flow graphs shown in Figure 3. line which expectedly has no reflection coefficient. When
Figures 3(a) to 3(f) are taken from Hunton's3 Figure l. compared to the flow graph of Figure 1 (b) the S11 and
A few comments about Figure 3 are necessary here. In S22 branches have the value zero or can be left out en­
this figure, p represents the magnitude of reflection coeffi- tirely.

60 'the microwave journal

 apparent. The successive manipulations are based o�
Figure 3(c) shows a detector and k d�notes the scal�r
conv ersion efficiency relating the incommg·wave amph­
four rules which are simp le, easy to remember and, If

tude to a meter reading. The meter reading M is assume�

forgotten, easy to reconstruct.

calibrated to take into account the detector law �o � IS RULE I: Two branches, whose common node has
in dependent of level. Since detector outputs are lfl�IC�­ only one incoming and one outgoing branch (branches in
tive of the power absorbed by (rather than power I no­
. series), may be combined to form a single branch whose
dent on) the detector, the calibration of M takes Into coefficient is the product of the coefficients of the origi­
account (1-plP/2 as part of the detector law. nal branches. Th,ts the common node is eliminated.
Figure 3(f) shows a flow graph representation of a
This topological manipulation is applied in Figure 4 (a).
generator. Here there is a departure from circuit theory
The left hand side of Figure 4 ( a) is a representation of
where a generator is thought of as a constant voltage
the two equations.
generator t with a series impedance (Thevenin) or a con­
stant current generator with a shunt admittance (Nor­
ton). In the microwave spectrum it is usually more con­ (6)
venient to think of a generator as a constant source of

(7)
outward traveling waves with a reflection coefficient look­
ing back into the generator output. Both Mason and
Zimmerman2 and Hunton3 use this representation. This
representation can be shown equivalent to Thevenin's or The value of E3 in terms of El is, from these equation�
Norton's representation (see Appendix).
Figure 3 (g) shows the flow graph of a lossless, bi­ (8)
lateral two-port network. Being bilateral' implies that
512 =521, For a lossless network the scattering matrix
This is the same equation which the flow graph on the
is unitary5 or
right side of Figure 4(a) represents. Since the two graphs
lead to the same equation the graphs can be considered
equal.
� /I (2) RULE II: Two branches pointit1>g from a common node
to another common node (branches in parallel) may be
where 5*mn is the complex conjugate of 5mn. From this combined into a single bra1tch whose coefficient is the sum
matrix identity, the following relationships can be ex­ of the coefficients of the original branches. This rule is
tracted: demonstrated in Figure 4 (b). Verification of this rule is
also demonstrated by writing the equations represented
by the two signal flow graphs claimed to be equal and
(3) then showing that the equations lead to the same result.
The equation for the left side of Figure 4 (b) is

(4) )
9
(

(5)
EI S2 1 E2 S32 E3 EI E3

In Equation 5, n is any odd integer (positive or nega-

• � • ;) • • ) •

tive), <P12 is the phase angle of S12 and <PI and <P2 are the
S 31= S21 S 32

angles of the input and output reflection coefficients re- (a)

spectively. </>12 is the transmission phase shift between
the two reference planes from which </>1 and </>2 are SA
measured.
EI C=> E2
EI
• �
S31= SA+ Sa
E2
•

Topological Manipulation of Flow Graphs Sa

Construction of the flow graph is only part of the ( b)

problem. There remains the problem of finding the
value of a wave at a certain node. The solutions for
S2 2
linear transmission line problems may be arrived at by
S 21
a series of topological manipulations which reduce a flow
graph to simpler and simpler forms until the answer is
EI
•
S
!, 0 E2
S 32
�
E3
• •
EI
I-S22
•
E2
S32
>- •
E3

Egen=Einc+Eref
(b) branch­
t By constant voltage is mean! net 1Jo!tage, i.e.,
(vector slim).
Figure 4 -
Reduction rules: (a) branches in series;
es in parallel; (c) reduction of a feedback loop.

November, 19 63 61
 =
E
.
1..-_-+--0..:.::.....
�
EA S2 A E2 S32 E3
$4 3
� •
E4

(a) (0)
EI 54
1
Ea
554 $23 E 3'

�
E5
S32 $43
J •
(b) EA S2 A E2 S32 E3 " E4

(b)
E 2A

E3

Figure 5 - Node du-

plication rule: (a)
node with single
E2a
::,,(S"
EA S2A E2 S32
)0 •
E3"
S43
..- •
E4

(c)
input branch; (b)
node with single = EI E3
o � ut branch; (c) S 23532
n e with neither
single input nor EB
5 2B
( el
single output bran-
ches;

52A S43

EA E 2' S32 E3" E4

� (d)
That of the right side i s

(10)

That these lead to the same result is obvious.

RULE III: When node 1Z possesses a relf loop (a branch

which begins and ends at n) of coefficient S'l1I the self
�:�;�
EA E 2' S 32 E 3"
S 43
•
E4

loop may be eliminated by dividing the coefficient of (e)

every other branch entering node n by 1-S"". Figure 4(c)
shows the elimination of a self loop using this rule. The Figure 6 - Node duplication with a feedback loop: (a) original
equations for the left hand side of Figure 4(c) may be graph; (b) duplication of node E3; (c) elimination of node
E'3 to form a self loop; (d) duplication of node E2 with
written
self loop; (e) elimination of self loops.

(1J.)

(12 )
RULE IV: A node may be duplicated, i.e., split into
Solving for E2 and E3 in terms of El yields two ,wdes which may be subsequently treated as two
separate nodes, sO long as the resulting signal flow graph
contains, once and only once, each combination of seP­

(13)
arate (not a bra?zch which forms a self loop) inp1tt and
output branches which· connected to the original node.
Any self loop attached to the origi1wl node must also be
attached to each of nodes resulting from duplication.
(14) This rule may be easily understood by referring to
Figures 5 and 6. In Figure 5 (a) node E2 has been dupli­
cated into E'2 and E"2- Rule I may now be applied, if
�quations (13) and (14) are also the equations of the desired, to eliminate node E'2' Then node E"2 may be
flow graph on the right side of Figure 4(c). Note the duplicated again and one of these eliminated by Rule I,
correspondence between l-Snn in the denominator of etc., until E2 is eventually removed from the flow graph.
each incoming branch and the 1- fLf3 which occurs in In Figure 5 (b) node E4 is duplicated into E'4 and E"4'
feedback circuits. Preserving combinations of input and output branches is
62 the microwave journal
demonstrated in Figure 5 (c). Figures 6 (a) to 6 (c) show and for Figure 7 (d)
how a self loop turns up in flow graph reductions. The
self loop has the coefficient S23 S32 in Figure 6(c). Fig­ (l6�
ure 6 ( d) shows the duplication of node Ez which posses­
ses a self loop and Figure 6(e) shows the elimination of and the insertion loss is then
the self loops at nodes E'2 and E"2. Alternatively the self
.oop at node E2 of Figure 6(c) could have been elimina­
ted by application of Rule III before node E 2 was dupli­
Ins. loss = 20 log �: = 20 10g \ �\
cated.
Rule IV may be verified by writing the equations of and all conditions of the definition are satisfied.
each flow graph in Figures 5 and 6 and showing the
indicated equalities are indeed true. The usual case, however, is pg¥=O, Pd¥=O. The calcula­
tion of the signal now apparent at the detector for Fig­
ure 7(b) can be found by first applying Rule I to make
Examples of Solution by Reduction of Flow Graphs a self loop of value pgPd and then eliminate the self loop
The first example is concerned with the measurement by changing the generacor scattering coefficient from 1 to
of insertion loss of a bilateral device and the errors due to
source and detector mismatch. Insertion loss is defined 1
I-pgPd
--- in accordance with Rule III.
as ratio of power to a detector with and without the test
device inserted in a perfectly matched transmission line.
Figures 7(a) and (c) show a block diagram of the sys­ Thus
tem witham and with the test device inserted. Figure
7(b) shows a flow graph representation of Figure 7(a), (18)
and Figure 7 (d) shows a flow graph representation of
Figure 7(c). It might be noted that T in Figure 7(d) is
Fi�ure 8 sho�s the topological manipulations applied
loss due co power absorbed in the device and power re­
flected from the device. Here pg, Pd, Pi> P2 and T are
to Figure 7(d) In order to find the signal apparent at the
complex numbers.
For a perfectly matched transmission line pg=Pd=O,
so for Figure 7 (b)
Eg (I) T (3) k
M,=k E� (15 )

(2") T (4")
(0)
GEN DET
f T

(a)
Eg (I) (3)
• •

E9 k MI
• � ,. •

P9 0 Pd

( b)
E•g
1 T
I-PgPI (I) 1- PdP2 (3')
• •
GEN ? DET

(e)

Eg 1 (I) T (3) k M2
• T

E.�
.
• .

pg C)P I C:)P
P2 d __ �____ .( ! )
___ d ��
I -_P�� __ �{�')�
________ k
____ ��
(2) T (4)

(d) (d)
Figure 7 - Block diagrams and signal flow graphs for measure- Figur<: 8 - Reduction procedure for the signal flow graph of
ment of insertion loss. FIgure 7(d).
November, 1963 63
detector. In Figure Sea) nodes (2) and (4) have been
duplicated into nodes (2'), (2"), (4') and (4") by Rule
�eflected �ave . from the load. A simplified flow graph
. IS shown 1O Figure 9(b). For simplicity and since this
e �am�le .IS only concerned with directivity errors, the
of maO!pulatlOns : (a) eliminate node (4') by Rule I
IV. Fl?Ure �(b) results from the following sequence
dlfectlOnal couplers are assumed to have no internal re­
gi�ing a self loop at node (3) of value Pd(l2; (b) eliminate �ections �nd the detectors also do not have any reflec­
thiS self loop by Rule III changing the value of the t�ons whICh could get back into the main transmission
branch from node (1) to node (3) from T to T line. �ach directional coupl�r has a forward coupling
; coeffiCient Cn, and back couplIng CnDn or directivity Dn.
. . I- P P2
d
(c) elimlOate node (2') by Rule I giving a self loop at M2
node (1) of v�lue PgPI ; (d) eliminate this self loop by It is desired to know how the ratio compares to
MI
Rule III changlOg the value of the branch leading from
1 Ipd. All the branches in Figure 9(b) may have complex
the generator to and also changing the branch values.
1-PgPI
from node (2") to node (1) to pg Figure 9(c) shows a simplification of 9(b) for evalu­
; (e) eliminate
I-pgpi ating MI' Since waves down branches C2 and C2D. do not
nodes (2") and (4") giving a branch from node (3) to re-enter the main line to interact with waves already there
T pgPd they can �e le!t o�t of Figure 9 (c). Rule I has alreadi
node (1) of value . Figure S(c) shows the dupli- been applIed 10 Figure 9(c) where possible. Node (2)
I-pgpi
may be duplicated by Rule IV and then each of those
cation of node (3) into nodes (3') and (3"). Figure
nodes eliminated by Rule I. The resulting signal flow
Sed) shows the signal flow graph which results from graph is shown in Figure 9(d). By application of Rule
eliminating node (3") by Rule I and then eliminating III to eliminate the self loop and then applying Rule II
the resulting self loop at node (1) by Rule III. I n Fig­ for branches in parallel, Ml may be found
ure 8(d) there now exists only a single path from the
Eg to M2 and nodes (I) and (3') can be eliminated by
Rule I yielding

( 1 9)

Using Equations (18) and (19) the apparent insertion

loss is then

(20)

Comparing this with the measurements made in a per­

fectly matched transmission line (Equation 17) the error
for the imperfectly matched line in db is

(21)

If only the magnitudes of each of the reflections in Equa­ To evaluate M2, branches CI and CIDI can be elimina­
tion (21) are known then the maximum error can be ted and Rule I applied where possible. The resulting
calculated by setting the numerator to a maximum and signal flow graph is shown in Figure 9(e). Node (3)
the denominator to a minimum and vice versa for the may be duplicated by Rule IV and each of those nodes
minimum error. eliminated by Rule I yielding the signal flow graph of
Figure 9(f). Node (2) may be duplicated into nodes
It should be noted that Equation (18) gives the vol­
(2') and (2") as shown in Figure 9 (g). Eliminating
tage incident on an arbitrarily matched load from an
node (2") by Rule I gives a self loop at node (1) which
arbitrarily matched generator. The extremes of variation
can be eliminated by Rule III. M� can then be found
of generator output with the phase relationship between
by Rules I and II to be
pg and pd is available in nomogram form.6
= The second example is concerned with directivity
errors in a reflectometer. Figure 9(a) shows the block
diagram of a reflectometer. Directional coupler #1 is
supposed to take a sample of the incident wave and direc­
tional coupler #2 is supposed to take a sample of the

64
the miCrOWa1Je journal
 Dividing Equation (23) by Equation (22)

� �� J l ) M2 k2C2 T1(D2+T2pd (24)

0
(0)
I
X
OIR COUP#I
X
OIR COUP*2
I
M1 k1C1(1+T1 T/ pLDd

M,
To evaluate the approximate sensitivity of �: to D1 and

to D2, let us make the simplifying assumptions that

Eg kl C1 =k2 C2, T1 =T2 = 1 (for small C1 and C2 the
PL
magni tude of T 1 and T2 is close to 1) and let us further­
more not consider the phase angles of various terms.
(b) T, Then Equation (24) becomes

M, (25)

E . ..ty 0f M2 to
The senSltlVl D1 may be statcd as
M1
(I)
(C)
(26)
Mr

kl
Eg TIT22PLCI DI

(d) 1j2 T22 Pg h (27)

M"
Comparing thcse equations it is seen that for small pr, that

Eg (I) T,
D2 causes a larger variation in �: than D1 of the same size.

T2PL The solution of the above problems was arrived at by

(e) (3)
considering only the block diagram and the flow graph
T r T2 P g
derived therefrom. Network equations were never writ­
ten and cascading networks did not involve changing
Mz forms of network equations, Re-reflections, which usual­
ly make error analysis of measurement systems quite in­
k2 volved, now b eco me a 1 - P1P 2 denominator term and
become analagous to the feedback of circuit theory.
Eg (I) T, (2) APPENDIX
A Thevenin equivalent generator is shown in Figure
10 driving a load ZL. E1 is defined as the voltage which

T, TiPgh appears across the output terminals when ZL= 00. Zl is

the impedance seen looking into the generator when
M2 E1 =0. According to Thevenin's Theorem, the E1 and Z1
so found may be used to drive any load Zr •.

k2 The signal flow graph of a microwave generator is

shown in Figure 11 driving a load of complex reflection
(2)
coefficient PL' Eg is defined as the voltage which appears
Eg (I) Tr across the OUCput terminals when PL= O. pg is the reflec­
C2TzPL tion coefficient seen looking into the generator terminals
(2"') when Eg =0. Eg and pg so found may be used to drive
any load PL' Using the rules for flow graphs set forth
Z in this article the voltage incident upon the load PL is
TI T2 Pg PL

Figure 9- Block diagram and signal flow graphs for evalua­ (28)
tion of directivity errors in a reflectometer.

November, 1963 65
 I, - 2 E.,
E 1--- (31)
I-pI':

Figure 10 - Theve­ Then by setting Z L=ZO in Figure 10 (PL=O), the vol.

nin equivalent gen­ tage across Z o is the voltage E g to be used for the tnicro­
erator.
wave generator.

(32)

Substituting Equation (31) into Equation (32) the ex­

Eg pected relationship is found

O------�-- --p-g- �. <:::Y PL

Figure 11 - Signal
Bow graph for a
mi crowave equiva­
lent generator. pg=
z, +Zo
or- = --
Z,-z" Z, l+p.
Zo 1- p" . (3=3)

Although Equations (31) and (33) relate E1 and Zl to

E� and pg for the specific cases mentioned in the two defi­
nitions they can be shown to hold in the general case.
The total voltage across the load of Figure 10 is

The total voltage across the load is composed of both the (34)
incident and reflected waves.
Substituting into this from Equations (30), (31) and
(29)
(33), it can be shown

In order to find relationships between the twO genera­ (35)

tors let us find first E1 from Figure 11 and then Eg from
Figure 10. Using the definition This is the same result found in Equation (29) so the
Thevenin equivalent generator yields the same load vol­
---
1 + pr,
(30)
Zr,-Zo Zr, tage as the microwave generator.
pr,= ZI,+ZO or-=
Z" I-pI, Both generator representations are subject to the sam...
Iinelrity restrictions. E, and Z, as well as Eg and pg must
and by letting PT,=1 in Figure 11 (Zr,= 00 ) the voltage be independent of the load placed on the generator (e.g.,
across PL is the voltage E1 to be used in the Thevenin the generators cannot be pulled in frequency by load­
equivalent. Using Equation (29) to solve for this voltage ing).

References

1. Mason, S. J., "Feedback Theory - Some Properties of Signal

Flow Graphs," Proc. IRE, Vol. 41, pp. 1144-1156, Sept. 1953;

,
also Mason, S. J., "Feedback Theory - Further Properties of
Signal Flow Graphs," Proc. IRE Vol. 44, pp. 920-926, July
1956.

2. Mason and Zimmerman, Electr01zic Circuits, Signals and Systems.

John Wiley & Sons, Inc., New York, N. Y., 1960.

3. Hunton, J. K., "Analysis of Microwave Measurement Tech­

niques by Means of Signal Flow Graphs," Trans. IRE, Vol.
MTT-S, pp. 206-212, March 1960.

,
4. Chow, Y. and E. Cassignol, Linear Signal-flow GrafJh.f ,.md A/I"
plicatio11S John Wiley & Sons, Inc., New York, N. Y., 1962.

5. Montgomery, et aI., Principles of MicfOUJatle Ci.-mits, McGraw­

Hill Book Co., Inc., New York, N. Y., 1945.
6. Microula,'e Engineers' Handbook and Bu),er.r' Guide, Horizon
House, Inc., Brookline, Mass., p. T-15, 1963.

66 the microwave journal