6.2.

SIGNAL-PLOW GRAPHS 93

6.2 Signal-Plow Graphs 1 - - X z ( . r)

0 (l)~C,)
- - X1(1)
Ci )0 (l) ~ (I )
- - CC,)
G (I (l)G (S) ~t6)
G1<t> 01(1} G,t,)

Norman S. Nise, California State Polytu:h-

nfc University, Pomon,a Figure 6.19 C,asaded subsystems.
X 1V) R(1)G1(1)
G1CI)
6.2..1 Introduction
I(,)
t
Signal-flow gnpht are an alternate system representation. Un- Gz(I)
X2(1) "(,t)G1(1) :i: AC>I C(1) (:tG1V> t~(I) tG3(1))RCs)
~
like block diagrams oflinear ay,tems, which consist of blocks, :i:
signals, summing junctions, and piclcoff points, signal-flow X,(1)RC,)G3(J)
graphs of linear systems consist of only braocbes, which repre- a,<>
sent systems, and ooclea, which represent signals. The&e elements
are shown in Flgure6.18aand Figure 6.18b ttspectively. A system Fipre 6.20 Parallel subsyaterns.
(Figure 6.18a) is represented by a line with an arrow showin.g the Plant and
C.ontroller
R(s) + (s} C(.r)
G(s)
Input _ Actuating
+ signal ...__ _

R z(S)
(error)
X(I)
H (s)
G(1)
0 R.i(.I)

(e) .
X(s)
, (&') figure6.ll Feedback oontrol system.
lflphC 6.J,I Slpal-flow papb component putl: (a.) l)'Strm; (b,} s.ig-
system. Next. we intc.rconncct the signal nodes wi1h Sf5{em
tnll: (c) imereon.nection of systmu ed ,1l'gnals. branches. Figura 6.22a, 6.2'2c, and 6,.22e show the signal n,odes
(or the cascaded. paralld, and fffllhaclc forms. re:sptively. Nm,
direction of signal flow through the system. Adjaan't to the line interconnect the nodes with branches ithat rcpresc:.nt the subsys-
we wriie the tnmlcr fwlction. A s[gnal (figure 6.18b) b a node tmis. This ii dQqe in fig\lm 6.22b, 6.22d, and 6.22f for the
with the sjgnal nanie written adjacent to the node. cascaded. parallel, and fttdba.ck forms, respectively. For the par-
!Figure 6.lk sh,ows the mttrconnection o( die systmis and allel form, positive signs ,are Wlllme.d at all iop uts to the summing
the aignall. Ba.ch lignall is the: sum of signal$ ff owing into jt, junction; for the fudbacl< form, 11 neptl'w sign ii mWlled for
for c:amplc, in Figure 6.18c the signal X (s) = R1(.r)G1(.r} - the feedback.
R2(.t)G2(.r)+,R,(s)G:,(.1). l'hesignal.C:,(s) = - X(s)G6(:r} = Im 'the nm example we start with a mo11t complicated bloclc
"" R1(.s)G1(.Y)G6(s) + R2(.r)G2(.s)G6(.s) - R, C.s)G J (.r) G6(.r), diagram and end with the equivalent si:gnal-flaw graph. Convert
Notathat rhe1ummin1 ofneptiw,s:ignals is handled bywo- the block diagram of Figure 6.2.3, to a signal-flow gniph. Begin
dating the n.-tM li,gn with lite system and not with a summing by drawing the signal nodes as sh,own in IPigure 6.24a.
junction as in tthe eut of block diqrams. Nat. intcriconnect the nodes,showing; the direction of :signal
flow md idetitifying each ttanlfer function. The result is 1hown
6.2..2 Reladomhip e,erween Block Diagrams io Figure 6.24b. Notice that !the negative signs at the summimg
ud Sign1J-Plow Graphs junctions of the block diagnm arc represented by the neg,ativc
transfer Cuna.ions of the signal-Bow graph.
To show th, panlld bdweeD blodc diagrams and sigpal-f1~ Fin:al]y, iif desired, simplify the signal-flow graph to the one
pap~ v.ee c:onwrt some block diagram forms to signal-flow shown in Figure 6.24c by diminating signals that have a single
igrapba. In am case, we :first comat the signals to nodes amd How in and a single flaw out sud1 as V2{s). V6(s), V1(.s), and
'then mmwnneet the node. with syneim,, V8(s ).
Let us conwrt th,e catcaded, pualld, and fecclback fo.rms of
the blodc diagnm, ahown in Figura 6.19 through 6.21, rap~-
~ . !into signal-flow graph& 6.1.3 Mason's Rule
In each cue. 'W!! ltilrt by drawing the signal nodes1for that Biodc diiagram reduction ttqUiru sua:asive application of fun
damcn.tal rdationsbipa :io on:lcr to uriw il l the system trmsfc:r
functiiom. On the othtr hand, Mason's. Rule for the reduction of
l Adapted &om Omtr.ol s,::,ra,u &finMrini. sea,nd edilion. by Ner,,
signal-flow igrapm to a trans.fer function relating thtt output to
mm S. NiR. Copyright t(O) 1.995 by The Benjamin/Cummings Pub- the input ffl[uirc:s the application of ai single formula. The for'-
lkhing Company. 'lleprbrted by pffllliuion. mula wa,1derived! by S. J. Muon when he related the signal-flow
94 THE CONTROL HANDBOOK
01(1) GJ(I) GJ() direction of signul flow. Examples of forward-path
RtJO 0 0 O c t,> R(IO 0 O 0C(d
X1tll x,u> X1ti) X1(1> gains arc also shown in Figure 6.25. There are two
I> (t )
forward-path gains as follows:

0
X 10)
G 1(s )G2.s)G3(s)G, (s)Gs(s)Cr,(s) (6. l2a)
R(J,Q 0
X1CII
O C<,> ~,,, C(I) Gt (s)G2,s)G3(s)G4 (s)G5(.J)G7(s)(6.12b)

0 Nontoucbing I.oops:: ?-1ontoucbingloops are loops that do

x,,,, x,u> not have any nodl!S in common. ln FigUtt 6.25, loop
(C) (I) G2(.J)H1 (s) doo. not touch loop G,(s)H2(s), loop
G4(s)Gs(s)H3(!), or loop G,4(s)G6(s) H3(s).
Ncmtoacbing-1.oop Gam: Nontouc:hing-loop gain is the
Rb)Q
(0
0 QCt.,) I
I (1~ :
G(1)
:JC{)) product of 1001= gains from nontoucbing loops
taken two, three, four, etc. at. a time. In Figure 6.25,
-H(I)
the product oflo:>p gain G2(.r)H1(s) and loop gain
(t ) (/)
G4(s)H2(s) is a nontoucbing-loop gain taken two
at a time. In summary, all th.rec of the nontoucbing-
Figure 6.22 Building signal-ff ow graphs: (a) cascaded systffl\! nodes
loop gains taken :wo at a time arc
(from Figure 6. 19); (b) cascaded '}'Stem: signal-flow graph; (c) parallel
system: nodes (from Figure 6.20}; (d) paralld syitem: signal-flow graph; 1. [G21s)H1(s) )( G4 (s)H2(s)J (6.13a)
(c) feedback system: nodes (from Figu.re 6.21); (0 feedback system:
signal-flow graph.
2. IG21.r)H1(s) 11 G,e (s) Gs(s)H3(~.13b)
3. IG2cs)H1 (s)l(G4(s)G6(s)H3(sM.I3c)
graphs to the simultaneous equations that can be written from
the graph ([ 1J, 121). Theproduct ofloupgains (G4(s)G5(s)H3(.s) Jl G4(s)
In general, it can be complicated to implement the formula G6(s)H3(s)J is not a nontouching-Joop gain since
without making mistakes. Specuically, the existence of what we th~ nm loops have nodes in common. ln our a-
later c.all non touching loops increases the complexity of the for- ample there are no nontouching-loop gains taken
mula. However, many syst.ems do not have nontoucbing loops three at a time since three nontouching loops do not
and thus lend themselves to the e.asy application of Mason's gain exist in the cxamole.
formula. For these systems, you may find Mason's Rule easier to We are now re.adv to state M.ason's Rule.
use than block diagram reduction. Maaolis Jbalc The transfer function, C (s)/ R(s), of a sys-
The formula has several component parts that must be evalu- tem represented :,ya signal-flow graph is
ated. We must first be sure that the definitions of the component
parts are well understood. Then, we must exert care in evaluating
the component parts of Mason's formula. To that end. we now
discuss some basic definitions applicable to signal-flow graphs.
Later we will state Mason's Rule and show an example.
where E dcnotca iummation
k number of forward paths
Definitions
Loop Gain! Loop gain is the product ofbranch gains found
T.1: = the kth forward,path pin
A 1- E loop gains + E nontoucblng
by traversing a path that starts at a node and ends loop pins taken two t a time - E
at the same node without passing through any other nontouching-loop gains tu.en thrtt at a
node more than.once and following the direction of dme + E nontouc:hing-loop pint cm:n
the signal Oow. For examples of loop gains, look at four at a time -
figure 6.25. There are four loop gains as follows: A- E loop gain terms in A toching the
ktb forward pa.th. In other words, A1c ia
1. G2(s)H1 (s) (6.1 la)
formed by tliminating &om A th!* loop
2. G,(s)H2(s) (6.llb) pins that touch the kth forward path.
3. G4(s)G5(s)H3(s) (6.1lc)
4. G4(s)G5(s)H3(s) (6.lld)
Notice the alternating signs for each component part of A. The
Forward-Path Gain: Forward-path gain is the product of following example will nelp clarify Masons Rule.
gains found by travming a path from the input node Find the transfer function, C(s)/ R(s), for the signal-flow
to the output node of the signal-flow graph in the graph in Figure 6.26. Pint, identify the forwud-,-da pm,
6.2. SIGNAL-PLOW GRAPHS 95

R(s) + v,cs> C(s>

V7(s) V8(s)

Mo) 0 0 0 0 0 OCln
VzCJI v,u1 v,v1 v,u1
0 0 0
\ ',CJ) Vt(r) ,,c.,1
CI
I

Fipn 6.25 Sample signal-flow graph for demonstrating Mason's

Rule.

07(1)

PJgurc 6.26 Signal-flow graph.

-H;,, does not touch loop 3. Notice that loops 1, 2, and 3 all touch loop
4. Thus, the combinations of nontouching-Joop gains taken two
at a time are as follows:

Loop 1 and loop 2: G2(s) H1 (s)G4 (s)H2 (s) (6.17a)

(c)
Loop 1 and loop 3: G2(s)H1 (s)G1(s)l4 (s) (6.17b)
Pigm 6.24 5'gnalflow graph dew.lopment: (a) signal nodes; (b) Loop 2 and loop 3: G4(s)H2(.s)G1(s)H4(s) (6.17c)
sigml.flow graph: (c) mnpti&d sign,1-flow graph.
Finally, the nontonddng-loop pio, taken three ate time are as
In this aarnple, there is only one: follows:

G1 (s)02(s)G:,(.r)G4(.r)Gs(s) (6.15) Loops 1, 2, and 3: G2 (s)H1 (s)G,(.s)H2(s)G1(s)H4(s)

(6.18)
Second. identify the loop pJm. There are four loops as follows:
Now, from Equation 6.14 and its definitions. we form I::. and l::.k :
J. G2(.r)H1 (s) {6.16a)
1 G4(.r)H2(s) (6.16b) I::. = 1 - [G2(s)H1{s) + G4(s)H2(s) + G1(s)H4(s)
3. G7(1)H4(s) (6.16c) + G2(s)G:,(s)G4(s)Gs(s)G6(s)G1(s)Ga(s) I
4. G2(1)Gs(s)G4(.s)Gs(s)G6(.s)G7(s)Ga(s) (6.16d) + (G2(s)H1(s)G4(s)H2(s)
+ G2(s)H1 (s)G7(s)H4(s)
Third. identify the aontoachhtg-loop pto fekm two at a time.
Prom Equations 6.16a to 6.16d and Figure 6.26, ~ see that loop + G4(1)H2(s)G7(s)H4(s) I
1 does not touch loop 2, loop 1 does not touch loop 3, and loop 2 - IG2(s)H1(s)G4(s)H2(1)G1 (s)H4(s)I (6.19)
96 THE CONTROL HANDBOOK
We form A1t by eliminating from 6. those loop gains that touch !
the kth forward path: rO 0 0 C 01 rO 0 o-!...o 01
x,
6) .r2 Z) .IJ
'
(6.20)
(1 ) <>
l !s I
Equations 6. 15, 6.19, and 6.20 are substituted into Equation 6.14 ,o I
C 0 1 r ;r
0 I 0 I

yielding the transfer function .r, ,') .\ .

.r,

T16.1
G (s) -
6.
[G 1(s)G2 (s)G3(s)G4(s)Gs(s)J( 1 - G1(s)H4(s)J (C)
= 6.
(6.21) Figure 6.27 Stagu in th" da'Clopment ofa signal-flow graph in pbue-
variable form for the system of Equation 6.22 or Equations 6.23.
Since there is only one forward path, G (s) consists only of one
term rather than the sum of terms each coming from a forward
path. signal-flow graph tbcn iollmvs from the state equations as in the
preceding section. Con:.ider the transfer function
6.2.4 Signal-Flow Graphs of Differential
Equations and State Equations G ( ) _ 1 ~1.r+2
- s ++261+24
(6.2-4)
lo this section \\'e show how to convert a differential equation or
state-space representation to a signal-Oow graph. Consider the Converting to the phase-variable representation in state-space3
differential equation

(6.22) i1 - .12 (6.25a)

Converting to the phase-variable representation in state-space 2

i2 - X3 (6.25b)
i3 - -24.xt - 26x2 - 9.x3 + 24r (6.25c)
i1 - Xl (6.23a) y - 2x, + 7x2 + x 1 (6.25d)
i2 - XJ (6.23b)
i3 - -24x1 - 26x2 - 9.x3 + 24r (6.23c) Following the same prcccdurc used to obtain Figure 6.27d, we
arrive at Figure 6.28. Notice that the denominator of the transfer
y - x1 {6.23d)
function is represented :,y the feedback paths. while the numer-
To &aw the associated sigoal-6ow graph, first identify three ator of the transfer function is represented by the linear annbi-
nodes. as in Figure 6.27a, to be the three state variables, xi , x2, nation of state variabl~ forming the output.
and x3. Also identify a node as the input, r, and another node as
the output, y. The first of the three state equations, it = x2, is
modeled in Figure 6.27b by realizing that the derivative of state 1
variable .x1, which is x2, would appear to the left at the input to
an integrator. Remember, division bys in the frequency domain
is equivalent to integration in the time domain. Similarly, the
second equation. i2 = .x3 , is added in Figure 6.27c. The la.st 1
of the st.ate equations, :c3 = -24x1 - 26x2 - 9x3 + 24r, is R<s) C<s>
added in Figure 6.27d by formingx3 at the input of an integrator
whose output is x3. Finally, the output, y = .x1 . is also added. in
Figure 6.27d completing the signal-flow graph. Notice that the
state variables are outputs of the integrators.
-24
6.2.5 Signal-Flow Graphs of Transfer Functions Figure 6.28 Sjgnal-flow graph in phase.-variable form for Equa,
To convert ttansfer functions to signal-flow graphs we first con- tion 6.24.
vert the transfer function to a state-space representation. The

2sec [3) for description of bow to convert differential equations into 3 sec (3) fi>r description of how to c:onvut transfer functions into
the phase-variable rcpresc:ntatlon in state space. the phase-variable rq>rcsct1tation in state 1pKC.