3/19/2009 4_5 Signal Flow Graphs.

doc 1/2

4.5 – Signal Flow Graphs

Reading Assignment: pp. 189-197

Q: Using individual device scattering parameters to analyze a

complex microwave network results in a lot of messy math!
Isn’t there an easier way?

A: Yes! We can represent a microwave network with its signal

flow graph.

HO: SIGNAL FLOW GRAPHS

Then, we can decompose this graph using a set of standard

rules.

HO: SERIES RULE

HO: PARALLEL RULE

HO: SELF-LOOP RULE

HO: SPLITTING RULE

It’s sort of a graphical way to do algebra! Let’s do some

examples:
EXAMPLE: DECOMPOSITION OF SIGNAL FLOW GRAPHS

EXAMPLE: SIGNAL FLOW GRAPH ANALYSIS

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 4_5 Signal Flow Graphs.doc 2/2

Signal Flow graphs can likewise help us understand the

fundamental physical behavior of a network or device. It can
even help us approximate the network in a way that makes it
simpler to analyze and/or design!

HO: THE PROPAGATION SERIES

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Signal Flow Graphs 1/12

Signal Flow Graphs

Consider a complex 3-port microwave network, constructed of
5 simpler microwave devices:

S1 S2 S3

S
S4 S5

where Sn is the scattering matrix of each device, and S is

the overall scattering matrix of the entire 3-port network.

Q: Is there any way to determine this overall network

scattering matrix S from the individual device scattering
matrices Sn ?

A: Definitely! Note the wave exiting one port of a device is a

wave entering (i.e., incident on) another (and vice versa). This
is a boundary condition at the port connection between
devices.

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Signal Flow Graphs 2/12

Add to this the scattering parameter equations from each

individual device, and we have a sufficient amount of math to
determine the relationship between the incident and exiting
waves of the remaining three ports—in other words, the
scattering matrix of the 3-port network!

Q: Yikes! Wouldn’t that require a lot of tedious algebra!

A: It sure would! We might use a computer to assist us, or

we might use a tool employed since the early days of
microwave engineering—the signal flow graph.

Signal flow graphs are helpful in (count em’) three ways!

Way 1 - Signal flow graphs provide us with a graphical

means of solving large systems of simultaneous equations.

Way 2 – We’ll see the a signal flow graph

can provide us with a road map of the wave
propagation paths throughout a microwave
device or network. If we’re paying
attention, we can glean great physical
insight as to the inner working of the
microwave device represented by the graph.

Way 3 - Signal flow graphs provide us with a

quick and accurate method for approximating
a network or device. We will find that we can
often replace a rather complex graph with a
much simpler one that is almost equivalent.

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Signal Flow Graphs 3/12

We find this to be very helpful when designing microwave

components. From the analysis of these approximate graphs,
we can often determine design rules or equations that are
tractable, and allow us to design components with (near)
optimal performance.

Q: But what is a signal flow graph?

A: First, some definitions!

Every signal flow graph consists of a set of nodes. These

nodes are connected by branches, which are simply contours
with a specified direction. Each branch likewise has an
associated complex value.

j
0.5
-j0.2
0.1
0.7 -0.1

Q: What could this possibly have to do with microwave

engineering?

A: Each port of a microwave device is represented by two

nodes—the “a ” node and the “b ” node. The “a ” node simply
represents the value of the normalized amplitude of the wave
incident on that port, evaluated at the plane of that port:

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Signal Flow Graphs 4/12

Vn + ( zn = znP )
an 
Z 0n

Likewise, the “b” node simply represents the normalized

amplitude of the wave exiting that port, evaluated at the
plane of that port:
Vn −( zn = znP )
bn 
Z 0n

Note then that the total voltage at a port is simply:

Vn ( zn = znP ) = ( an + bn ) Z 0n

The value of the branch connecting two nodes is simply the

value of the scattering parameter relating these two voltage
values:

Vn + ( zn = znP ) Smn
an  Vm−( z m = z mP )
Z 0n bm 
Z 0m

The signal flow graph above is simply a graphical

representation of the equation:

bm = Smn an

Moreover, if multiple branches enter a node, then the voltage

represented by that node is the sum of the values from each
branch. For example, the signal flow graph:

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Signal Flow Graphs 5/12

b1
S11 S13
a1 S12
a3
a2
is a graphical representation of the equation:

b1 = S11 a1 + S12 a2 + S13 a3

Now, consider a two-port device with a scattering matrix S :

⎡S S ⎤
S = ⎢ 11 12
⎥
⎣S21 S22 ⎦
So that:

b1 = S11 a1 + S12 a2

b2 = S21 a1 + S22 a2

We can thus graphically represent a two-port device as:

a1 b2
S21

S11 S22

S12
b1 a2

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Signal Flow Graphs 6/12

Now, consider a case where the second port is terminated by

some load ΓL :

S ΓL

We now have yet another equation:

V2+ ( z 2 = z 2P ) = ΓL V2− ( z 2 = z 2P )
a2 = ΓL b2

Therefore, the signal flow graph of this terminated network

is:
a1
S21 b2

S11 S22

S12 ΓL
b1
a2

Now let’s cascade two different two-port networks

Sx Sy ΓL

Here, the output port of the first device is directly

connected to the input port of the second device. We
describe this mathematically as:

Jim Stiles The Univ. of Kansas Dept. of EECS

 3/13/2009 Signal Flow Graphs 7/12

a1y = b2x and b1y = a2x

Thus, the signal flow graph of this network is:

b2x a1y b2y

a1
x

S21x 1 S21y

S11x S22x S11y S22y

S12x 1 S12y ΓL
b1 x

a2x b1y a2y

Q: But what happens if the networks are connected with
transmission lines?

A: Recall that a length A of transmission line with

characteristic impedance Z0 is likewise a two-port device. Its
scattering matrix is:

⎡ 0 e −j βA ⎤
S = ⎢ −j βA ⎥
⎢⎣e 0 ⎥⎦

Thus, if the two devices are connected by a length of

transmission line:

a1x b2x b1y b2y

Sx Z0 Sy ΓL
b1 x a2x a1
y
a2 y

Jim Stiles The Univ. of Kansas Dept. of EECS

 3/13/2009 Signal Flow Graphs 8/12

a1y = e − j β A b2x a2x = e − j β A b1y

so the signal flow graph is:

b2x a1y b2y

a1
x

S21x e −j βA S21y

S11x S22x S11y S22y

S12x e −j βA S12y ΓL
b1 x

a2x b1y a2y

Note that there is one (and only one) independent variable in
this representation.

This independent variable is node a1x .

This is the only node of the sfg that does not have any
incoming branches. As a result, its value depends on no other
node values in the sfg.

Æ From the standpoint of a sfg, independent nodes are

essentially sources!

Of course, this likewise makes sense physically (do you see

why?). The node value a1x represents the complex amplitude
of the wave incident on the one-port network. If this value is
zero, then no power is incident on the network—the rest of
the nodes (i.e., wave amplitudes) will likewise be zero!

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Signal Flow Graphs 9/12

Now, say we wish to determine, for example:

1. The reflection coefficient Γin of the one-port device.

2. The total current at port 1 of second network (i.e.,

network y).

3. The power absorbed by the load at port 2 of the

second (y) network.

In the first case, we need to determine the value of

dependent node b1x :
b1x
Γin = x
a1

For the second case, we must determine the value of wave

amplitudes a1y and b1y :
a1y − b1y
I1 =
y

Z0

And for the third and final case, the values of nodes a2y and
b2y are required:
2 2
b2y − a2y
Pabs =
2

Q: But just how the heck do we determine the values of

these wave amplitude “nodes”?

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Signal Flow Graphs 10/12

A: One way, of course, is to solve the simultaneous equations

that describe this network.

From network x and network y:

b1x = S11x a1x + S12x a2x b1y = S11y a1y + S12y a2y

b2x = S21x a1x + S22x a2x b2y = S21y a1y + S22y a2y

From the transmission line:

a1y = e − j β A b2x a2x = e − j β A b1y

And finally from the load:

a2 = ΓL b2

But another, EVEN BETTER way to determine these values is

to decompose (reduce) the signal flow graph!

Q: Huh?

A: Signal flow graph reduction is a method for simplifying

the complex paths of that signal flow graph into a more direct
(but equivalent!) form.

Reduction is really just a graphical method of decoupling the

simultaneous equations that are described by the sfg.

For instance, in the example we are considering, the sfg :

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Signal Flow Graphs 11/12
b2x y
a1 b2 y
a1x
S21x e −j βA S21y

S11x S22x S11y S22y

S12x e −j βA S12y ΓL
b1 x

a2x b1y a2y

might reduce to: 0.1 e − j
π
8

−0.6 b2x a1y b2y

a1 x

0. 3
− j 0.2
−0.2 j 0.1 0.05
b1x
a2x b1y a2y
From this graph, we can directly determine the value of each
node (i.e., the value of each wave amplitude), in terms of the
one independent variable a1x .

b1x = −0. 2 a1x

b2x = −0. 6 a1x a2x = j 0. 1 a1x
b1y = 0. 05 a1x a1y = 0. 1 e − j a1x
π
8

b2y = 0. 3 a1x a2y = −0. 2 a1x

And of course, we can then determine values like:

b1x −0. 2 a1x

1. Γin = x = = −0. 2
a1 a1x

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Signal Flow Graphs 12/12

a1y − b1y 0. 1 e
−j π
− 0. 05 x 8

2. I1y
= = a1
Z0 Z0

2 2
b2y − a2y ( 0. 3)
2 2
− ( 0. 2 ) x 2
3. Pabs = = a1
2 2

Q: But how do we reduce the sfg to its simplified state?

Just what is the procedure?

A: Signal flow graphs can be reduced by sequentially applying

one of four simple rules.

Q: Can these rules be applied in any order?

A: No! The rules can only be applied when/where the

structure of the sfg allows. You must search the sfg for
structures that allow a rule to be applied, and the sfg will
then be (a little bit) reduced. You then search for the next
valid structure where a rule can be applied.
Eventually, the sfg will be completely reduced!

Q: ????

A: It’s a bit like solving a puzzle. Every sfg is

different, and so each will require a different
reduction procedure. It requires a little thought,
but with a little practice, the reduction procedure is
easily mastered.

You may even find that it’s kind of fun!

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Series Rule 1/2

Series Rule
Consider these two complex equations:

b1 = α a1 a2 = β b1

where α and β are arbitrary complex constants. Using the

associative property of multiplication, these two equations
can combined to form an equivalent set of equations:

b1 = α a1 a2 = β b1 = β (α a1 ) = (αβ ) a1

Now let’s express these two sets of equations as signal flow

graphs!

The first set provides:

a1 α b1 β a2 b1 = α a1
a2 = β b1
While the second is:

α b1 a2
a1 b1 = α a1
a2 = αβ a1
αβ
Q: Hey wait! If the two sets of equations are equivalent,
shouldn’t the two resulting signal flow graphs likewise be
equivalent?

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Series Rule 2/2

A: Absolutely! The two signal flow graphs are indeed

equivalent.

This leads us to our first signal flow graph reduction rule:

Rule 1 - Series Rule

If a node has one (and only one!) incoming branch, and

one (and only one!) outgoing branch, the node can be
eliminated and the two branches can be combined, with
the new branch having a value equal to the product of
the original two.

For example, the graph:

a1 b1 a2
b1 = 0.3 a1
0. 3 −j a2 = − j b1

can be reduced to:

0.3 b1 a2
a1
b1 = 0.3 a1
a2 = − j 0.3 a1
− j 0. 3

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Parallel Rule 1/5

Parallel Rule
Consider the complex equation:

b1 = α a1 + β a1

where α and β are arbitrary complex constants. Using the

distributive property, the equation can equivalently be
expressed as:
b1 = (α + β ) a1

Now let’s express these two equations as signal flow graphs!

The first is:

α b1
a1 b1 = α a1 + β a1

With the second:

b1
a1 b1 = (α + β ) a1
α +β

Q: Hey wait! If the two equations are equivalent, shouldn’t

the two resulting signal flow graphs likewise be equivalent?

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Parallel Rule 2/5

A: Absolutely! The two signal flow graphs are indeed

equivalent.

This leads us to our second signal flow graph reduction rule:

Rule 2 - Parallel Rule

If two nodes are connected by parallel branches—and

the branches have the same direction—the branches
can be combined into a single branch, with a value equal
to the sum of each two original branches.

For example, the graph:

0.2 b1
a1 b1 = 0.3 a1 + 0.2 a1

0. 3

Can be reduced to:

0.5 b1
b1 = ( 0.3 + 0.2 ) a1
a1
= 0.5 a1

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Parallel Rule 3/5

Q: What about this signal flow graph?

note
direction!

0.2 b1
a1

0.3

Can I rewrite this as:

- 0.2 b1
a1

0.3

so that (since 0.3-0.2=0.1):

0.1 b1
a1 ???

A: Absolutely not! NEVER DO THIS!!

Q: Maybe I made a mistake. Perhaps I should have rewritten:

note
direction!

0.2 b1
a1

0.3
Jim Stiles The Univ. of Kansas Dept. of EECS
3/13/2009 Parallel Rule 4/5

as this:

5 = 1 0.2 b1
a1

0. 3

so that (since 5.0+0.3=5.3):

5.3 b1
???
a1

A: Absolutely not! NEVER DO THIS EITHER!!

From the signal flow graph below, we can only conclude that
b1 = 0.3 a1 and a1 = 0.2 b1 .
0.2 b1
a1

0. 3
Using the series rule (or little bit of algebra), we can conclude
that an equivalent signal flow graph to this is:

a1 = 0.06 a1
b1 = 0.3 a1 0.06
b1
a1

0. 3
Q: Yikes! What kind of goofy branch begins and ends at
the same node?

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Parallel Rule 5/5

A: Branches that begin and end at the same node

are called self-loops.

Q: Do these self-loops actually appear in signal flow graphs?

A: Yes, but the self-loop node will always have at least one
other incoming branch. For example:

a1 = 0.06 a1 − j b2
b1 = 0.3 a1 0.06
b1
a1

b2 −j 0. 3

Q: But how do we reduce a signal flow graph containing a

self-loop?

A: See rule 3 !

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Self Loop Rule 1/4

Self-Loop Rule
Now consider the equation:

b1 = α a1 + β a2 + γ b1

A little dab of algebra allows us to determine the value of

node b1 :
b1 = α a1 + β a2 + γ b1

b1 − γ b1 = α a1 + β a2

(1 − γ ) b1 = α a1 + β a2

α β
b1 = a1 + a2
1−γ 1−γ

The signal flow graph of the first equation is:

γ
b1 = α a1 + β a2 + γ b1
α

b1 a2
a1
β

Jim Stiles The Univ. of Kansas Dept. of EECS

 3/13/2009 Self Loop Rule 2/4

While the signal flow graph of the second is:

b1 β
α β α 1−γ
b1 = a1 + a2 a2
1−γ 1−γ a1 1−γ

These two signal flow graphs are equivalent!

Note the self-loop has been “removed” in the second graph.

Thus, we now have a method for removing self-loops. This
method is rule 3.

Rule 3 – Self-Loop Rule

A self-loop can be eliminate by multiplying all of the

branches “feeding” the self-loop node by 1 (1 − Ssl ) ,
where Ssl is the value of the self loop branch.

For example:
0.2
0. 6 b1 b1 = 0.6 a1 + j 0.4a2 + 0.2b1
a1 a2
j 0.4

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Self Loop Rule 3/4

can be simplified by eliminating the self-loop. We multiply

both of the two branches feeding the self-loop node by:

1 1
= = 1.25
1 − Ssl 1 − 0.2

Therefore:
0.6 (1.25 )
b1
a1 a2

j 0.4 (1.25 )

And thus:

b1 = 0.75 a1 + j 0.5 a2
0.75
b1
a1 a2

j 0.5
Or another example:

a1 = 0.06 a1 − j b2
b1 = 0.3 a1
0.06
b1
a1

b2 −j 0.3

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Self Loop Rule 4/4

becomes after reduction using rule 3:

b1
−j a1
a1 = b2
0.94
−j 0. 3
b1 = 0.3 a1 b2
0.94

Q: Wait a minute! I think you forgot something. Shouldn’t

you also divide the 0.3 branch value by 1 − 0.06 = 0.94 ??

A: Nope! The 0.3 branch is exiting the self-loop node a1.

Only incoming branches (e.g., the –j branch) to the self-
loop node are modified by the self-loop rule!

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Splitting Rule 1/5

Splitting Rule
Now consider these three equations:

b1 = α a1
a2 = β b1
a3 = γ b1

Using the associative property, we can likewise write an

equivalent set of equations:

b1 = α a1
a2 = αβ a1
a3 = α b1

The signal flow graph of the first set of equations is:

β a2

a1 α b1

a3
γ
While the signal flow graph of the second is:
αβ
a2
b1
a1
α γ
a3

Jim Stiles The Univ. of Kansas Dept. of EECS

 3/13/2009 Splitting Rule 2/5

Rule 4 – Splitting Rule

If a node has one (and only one!) incoming branch, and

one (or more) exiting branches, the incoming branch
can be “split”, and directly combined with each of the
exiting branches.

For example:
0.3 a2
b1 = − j a1
a1 −j b1
a2 = 0.3 b1
a3 a3 = −0.2 b1

−0.2
can be rewritten as:
− j 0.3
a2
b1 = − j a1
b1
a1 a2 = − j 0.3 a1
−j − 0. 2 a3 = −0.2 b1
a3

Jim Stiles The Univ. of Kansas Dept. of EECS

 3/13/2009 Splitting Rule 3/5

Of course, from rule 1 (or from rule 4!), this graph can be
further simplified as:
− j 0. 3
a2
−j b1 = − j a1
a1 b1 a2 = − j 0.3 a1
a3 a3 = j 0.2 a1

j 0. 2
The splitting rule is particularly useful when we encounter
signal flow graphs of the kind:
0.3
−j j 0.1

Note this node has two −0.2 Note this node has only
incoming branches !! one incoming branch !!

We can split the -0.2 branch, and rewrite the graph as:

−0.2 ( 0.3)

−j j 0.1
−0.2

Note we now have a self-loop, which can be eliminated using

rule #3:
−j
1 + 0.06 −0.2 j 0.1

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Splitting Rule 4/5

Note that this graph can be further simplified using rule #1.

j 0.1
− j 0.94
− j 1.89

Q: Can we split the other branch of the loop? Is this signal

flow graph:

−0.2 ( 0.3)

−j j 0.1
−0.2
Likewise equivalent to this one ??:

0.3
−j j 0.1

− 0. 2
A: NO!! Do not make this mistake! We cannot split
the 0.3 branch because it terminates in a node with
two incoming branches (i.e., -j and 0.3). This is a
violation of rule 4.

Moreover, the equations represented by the two signal flow

graphs are not equivalent—they two graphs describe two
different sets of equations!

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Splitting Rule 5/5

It is important to remember that there is no “magic” behind

signal flow graphs. They are simply a graphical method of
representing—and then solving—a set of linear equations.

As such, the four basic rules of analyzing a signal flow graph

represent basic algebraic operations. In fact, signal flow
graphs can be applied to the analysis of any linear system, not
just microwave networks.

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Example Decomposition of Signal Flow Graph 1/3

Example: Decomposition of
Signal Flow Graphs
Consider the basic 2-port network, terminated with load ΓL .
a1 b2
S21

1 S 2 ΓL S11 S22

S12 ΓL
b1
a2
Say we want to determine the value:

V1 − ( z = z1P ) b1
Γ1  + = ??
V1 ( z = z1P ) a1

In other words, what is the reflection coefficient of the

resulting one-port device?

Q: Isn’t this simply S11 ?

A: Only if ΓL = 0 (and it’s not)!!

So let’s decompose (simplify) the signal flow graph and find out!

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Example Decomposition of Signal Flow Graph 2/3

Step 1: Use rule #4 on node a2

a1 b2
S21

S22ΓL
S11

S12 ΓL
b1
a2
Step 2: Use rule #3 on node b2
S21
1 − S22ΓL b2
a1

S11

S12 ΓL
b1
a2
Step 3: And then using rule #1:
S21
a1 1 − S22ΓL b2

S21ΓL
S11 1 − S22ΓL

ΓL S21 S12
b1 1 − S22ΓL
a2

Jim Stiles The Univ. of Kansas Dept. of EECS

3/13/2009 Example Decomposition of Signal Flow Graph 3/3

Step 4: Use rule 2 on nodes a1 and b1

a1 b2
S21
1 − S22ΓL
Γ L S21 S12
S11 + S21ΓL
1 − S22ΓL
1 − S22ΓL
b1 a2

Therefore:

b1 Γ S S
Γ1 = = S11 + L 21 12
a1 1 − S22ΓL

b1
Note if ΓL = 0 , then a1 = S11 !

Jim Stiles The Univ. of Kansas Dept. of EECS

3/25/2009 Example Signal Flow Graph Analysis 1/6

Example: Analysis Using

Signal Flow Graphs
Below is a single-port device (with input at port 1a) constructed
with two two-port devices ( Sx and Sy ), a quarter wavelength
transmission line, and a load impedance.

j2
A=λ 4

Z0 Sx Z0 Sy Γ L = 0 .5

port 1x port 2x port 1y port 2y

(input)

Where Z 0 = 50Ω .

The scattering matrices of the two-port devices are:

⎡0.35 0.5 ⎤ ⎡ 0 0.8 ⎤

Sx = ⎢ Sy = ⎢ ⎥
⎣ 0.5 0 ⎥⎦ ⎣0.8 0.4 ⎦

Likewise, we know that the value of the voltage wave incident on

port 1 of device Sx is:

V01+x ( z1x = z1xP ) j2 j 2

a1x  = = V
Z0 50 5

Jim Stiles The Univ. of Kansas Dept. of EECS

 3/25/2009 Example Signal Flow Graph Analysis 2/6

Now, let’s draw the complete signal flow graph of this circuit,
and then reduce the graph to determine:

a) The total current through load ΓL .

b) The power delivered to (i.e., absorbed by ) port 1x.

The signal flow graph describing this network is:

S21x b2x e −j βA a1y S21y

a1x b2y

S11x S22x S11y S22y

ΓL
S12x S12y
e −j βA
b1x
a2x b1y a2y

Inserting the numeric values of branches:

a1x = j 2
5 0.5 b2x −j a1y 0.8 b2y

0.35 0.0 0.0 0.4

0.5
0.5 −j 0. 8
b1x
a2x b1y a2y

Jim Stiles The Univ. of Kansas Dept. of EECS

 3/25/2009 Example Signal Flow Graph Analysis 3/6

Removing the zero valued branches:

a1x = j 2 0.5 b2x −j a1y 0.8

5 b2y

0.35 0.4

0.5
0.5 −j 0.8
b1x
a2x b1y a2y
And now applying “splitting” rule 4:

a1x = j 2 0.5 b2x −j a1y 0.8 b2y

5

0.35
( 0.4 ) 0.5 = 0.2
0.5
0.5 −j 0.8
b1x
a2x b1y a2y
Followed by the “self-loop” rule 3:
0. 8
−j a1y = 1. 0
a1x = j 2
5
0.5 b2x 1 − 0.2 b2y

0.35

0.5
0.5 −j 0.8
b1x
a2x b1y a2y

Jim Stiles The Univ. of Kansas Dept. of EECS

3/25/2009 Example Signal Flow Graph Analysis 4/6

Now, let’s used this simplified signal flow graph to find the
solutions to our questions!

a) The total current through load ΓL .

The total current through the load is:

I L = −I ( z 2y = z 2yP )
V02+ y ( z 2y = z 2yP ) −V02− y ( z2y = z2yP )
=−
Z0
a2y − b2y
=−
Z0
b2y − a2y
=
50

Thus, we need to determine the value of nodes a2y and b2y. Using
the “series” rule 1 on our signal flow graph:

a1x = j 2
5
− j 0.5 b2y
Note we’ve simply
ignored (i.e.,
neglected to plot)
0.35
the node for
which we have no
0.5
− j 0.4 interest!
b1x
a2y
From this graph we can conclude:

Jim Stiles The Univ. of Kansas Dept. of EECS

3/25/2009 Example Signal Flow Graph Analysis 5/6

⎛j 2⎞
b2y = − j 0.5 a1x = − j 0.5 ⎜⎜ ⎟⎟ = 0.1 2
⎝ 5 ⎠
and:
a2y = 0.5 b2y = 0.5 0.1 2 = 0.05 2( )
Therefore:

b2y − a2y ( 0.1 − 0.05 ) 2 0.05

IL = = = = 10.0 mA
50 50 5

b) The power delivered to (i.e., absorbed by ) port 1x.

The power delivered to port 1x is:

Pabs = P + − P −
2 2
V1x+ ( z1x = z1xP ) V1x− ( z1x = z1xP )
= −
2Z 0 2Z 0
2 2
a1x − b1x
=
2
Thus, we need determine the values of nodes a1x and b1x. Again
using the series rule 1 on our signal flow graph:

a1x = j 2
5
Again we’ve
simply ignored
(i.e., neglected to
0.35 plot) the node for
−0.1 which we have no
interest!
b1x

Jim Stiles The Univ. of Kansas Dept. of EECS

3/25/2009 Example Signal Flow Graph Analysis 6/6

And then using the “parallel” rule 2:

a1x = j 2
5

0.25 = 0.35 − 0.1

b1x

Therefore:
b1x = 0.25 a1x = 0.25 ( j 2
5 ) = j 0.05 2

and:

2 2
j 2
5 − j 0.05 2 0.08 − 0.005
Pabs = = = 37.5 mW
2 2

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 1/18

The Propagation Series

Q: You earlier stated that signal flow graphs are helpful in
(count em’) three ways. I now understand the first way:

Way 1 - Signal flow graphs provide us with a graphical

means of solving large systems of simultaneous equations.

But what about ways 2 and 3 ??

“Way 2 – We’ll see the a signal flow graph

can provide us with a road map of the wave
propagation paths throughout a microwave
device or network.”

“Way 3 - Signal flow graphs provide us with

a quick and accurate method for
approximating a network or device.”

A: Consider the sfg below:

a1 0.5 b2 a3 0.8 b4
j
0.144 0.35 0.4 0.5
j
b1 0.5 a2 b3 0.8 a4

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 2/18

Note that node a1 is the only independent node. This signal

flow graph is for a rather complex single-port (port 1) device.

Say we wish to determine the wave amplitude exiting port 1.

In other words, we seek:
b1 = Γin a1

Using our four reduction rules, the signal flow graph above is
simplified to:

a1 j 0 .5
0.5
b2 a3 b4

j 0 .4

−0.36
a2 j 0.36 b3 j 0.2 a4
b1

Q: Hey, node b1 is not connected to anything. What does

this mean?

A: It means that b1 = 0 —regardless of the value of incident

wave a1 . I.E.,:
b1
Γin = =0
a1

In other words, port 1 is a matched load!

Q: But look at the original signal flow graph; it doesn’t look

like a matched load. How can the exiting wave at port 1 be
zero?

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 3/18

A: A signal flow graph provides a bit of a propagation road

map through the device or network. It allows us to
understand—often in a very physical way—the propagation of
an incident wave once it enters a device.

We accomplish this by identifying from the sfg propagation

paths from an independent node to some other node (e.g., an
exiting node). These paths are simply a sequence of branches
(pointing in the correct direction!) that lead from the
independent node to this other node.

Each path has value that is equal to the product of each

branch of the path.

Perhaps this is best explained with some examples.

One path between independent (incident wave) node a1 and

(exiting wave) node b1 is shown below:

a1 0.5 0.8
j
0.144 0.35 0.4 0.5
j
b1 0.5 0.8

We’ll arbitrarily call this path 2, and its value:

p2 = ( 0.5 ) j ( 0.4 ) j ( 0.5 ) = −0.1

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 4/18

Another propagation path (path 5, say) is:

a1 0.5 0.8
j
0.144 0.35 0.4 0.5
j
b1 0.5 0.8

p5 = ( 0.5 ) j ( 0.4 ) j ( 0.35 ) j ( 0.8 )( 0.5 )( 0.8 ) j ( 0.5 )

= j 4 ( 0.35 ) ( 0.4 ) ( 0.8 ) ( 0.5 )
2 3

= 0.0112

Q: Why are we doing this?

A: The exiting wave at port 1 (wave amplitude b1 ) is simply the

superposition of all the propagation paths from incident node
a1 ! Mathematically speaking:

b1
b1 = a1 ∑ pn ⇒ Γin = = ∑ pn
n a1 n

Q: Won’t there be an awful lot of propagation paths?

A: Yes! As a matter of fact there are an infinite number of

paths that connect node a1 and b1 . Therefore:

∞
b1 ∞
b1 = a1 ∑ pn ⇒ Γin = = ∑ pn
n a1 n

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 5/18

Q: Yikes! Does this infinite series converge?

A: Note that the series represents a finite physical value

(e.g., Γin ), so that the infinite series must converge to the
correct finite value.

Q: In this example we found that Γin = 0 . This means that

the infinite propagation series is likewise zero:

∞
Γin = ∑ pn = 0
n

Do we conclude from this that all propagation paths are zero:

pn = 0 ?????

A: Absolutely not! Remember, we have already determined

that p2 = −0.1 and p4 = 0.0112 —definitely not zero-valued! In
fact for this example, none of the propagation paths pn are
precisely equal to zero!

Q: But then why is:

∞

∑n pn = 0 ???

A: Remember, the path values pn are complex. A sum of non-

zero complex values can equal zero (as it apparently does in
this case!).

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 6/18

Thus, a perfectly rational way of viewing this network is to

conclude that there are an infinite number of non-zero
waves exiting port 1:

∞
Γin = ∑ pn where pn ≠ 0
n

It just so happens that these waves coherently add together

to zero:
∞
Γin = ∑ pn = 0
n

—they essentially cancel each other out !

Q: So, I now appreciate the fact that signal flow graphs: 1)

provides a graphical method for solving linear equations and
2) also provides a method for physically evaluating the wave
propagation paths through a network/device.

But what about helpful Way 3:

“Way 3 - Signal flow graphs provide us with

a quick and accurate method for
approximating a network or device.” ??

A: The propagation series of a microwave network is very

analogous to a Taylor Series expansion:

d nf ( x )
∞
f (x ) = ∑
n
( x − a )
n =0 d x
n
x =a

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 7/18

Note that there likewise is a infinite number of terms, yet

the Taylor Series is quite helpful in engineering.

Often, we engineers simply truncate this infinite series,

making it a finite one:

N
d nf ( x )
f (x ) ≈ ∑
n
( x − a )
n =0 d x
n
x =a

Q: Yikes! Doesn’t this result in error?

A: Absolutely! The truncated series is an approximation.

We have less error if more terms are retained; more error if

fewer terms are retained.

The trick is to retain the “significant” terms of the infinite

series, and truncate those less important “insignificant”
terms. In this way, we seek to form an accurate
approximation, using the fewest number of terms.

Q: But how do we know which terms are significant, and

which are not?

A: For a Taylor Series, we find that as the order n

increases, the significance of the term generally (but not
always!) decreases.

Q: But what about our propagation series? How can we

determine which paths are “significant” in the series?

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 8/18

A: Almost always, the most significant paths in a propagation

series are the forward paths of a signal flow graph.

forward  path ‐ \ˈfȯr‐wərdˈ päth\  ‐noun   
 
A path through a signal flow graph that passes through any given node 
no more than once.  A path that passes through any node two times (or 
more) is therefore not a forward path.  
 

In our example, path 2 is a forward path. It passes through

four nodes as it travels from node a1 to node b1 , but it passes
through each of these nodes only once:

a1 0.5 0. 8
j
0.144 0.35 0.4 0.5
j
b1 0.5 0.8

Alternatively, path 5 is not a forward path:

a1 0.5 0.8
j
0.144 0.35 0.4 0.5
j
b1 0.5 0. 8

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 9/18

We see that path 5 passes through six different nodes as it

travels from node a1 to node b1 . However, it twice passes
through four of these nodes.

The good news about forward paths is that there are always a
finite number of them. Again, these paths are typically the
most significant in the propagation series, so we can
determine an approximate value for sfg nodes by considering
only these forward paths in the propagation series:

∞ N

∑n pn ≈ ∑
n
pnfp
=1

where pnfp represents the value of one of the N forward

paths.

Q: Is path 2 the only forward path in our example sfg ?

A: No, there are three. Path 1 is the most direct:

p1 = 0.144
a1 0.5 0.8
j
0.144 0.35 0.4 0.5
j
b1 0.5 0. 8

Of course we already have identified path 2:

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 10/18

p2 = −0.1
a1 0.5 0. 8
j
0.144 0.35 0.4 0.5
j
b1 0.5 0.8

And finally, path 3 is the longest forward path:

p3 = ( 0.5 ) j ( 0.8 )( 0.5 )( 0.8 ) j ( 0.5 )

2 3
= j 2 ( 0.8 ) ( 0.5 )
= −0.08
a1 0.5 0. 8
j
0.144 0.35 0.4 0.5
j
b1 0.5 0.8

Thus, an approximate value of Γin is:

b1
Γin =
a1
3
≈ ∑ pnfp
n =1
= p1 + p2 + p3
= 0.144 − 0.1 − 0.08
= −0.036

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 11/18

Q: Hey wait! We determined earlier that Γin = 0 , but now

your saying that Γin = − 0.036 . Which is correct??

A: The correct answer is Γin = 0 . It was determined using

the four sfg reduction rules—no approximations were
involved!

Conversely, the value Γin = − 0.036 was determined using a

truncated form of the propagation series—the series was
limited to just the three most significant terms (i.e., the
forward paths). The result is easier to obtain, but it is just
an approximation (the answers will contain error!).

For example, consider the reduced signal flow graph (no

approximation error):

a1 j 0 .4
0.4
b2 a3 b4

0.0 j 0.32 Exact SFG

−0.288
b1 a2 j 0.288 b3 j 0.16 a4

Compare this to the same sfg, computed using only the

forward paths: j 0. 5
a1 0.5
b2 a3 b4

−0.036 j 0. 4

Approx. SFG −0.36

a2 j 0.36 b3 j 0.2 a4
b1

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 12/18

No surprise, the approximate sfg (using forward paths only) is

not the same as the exact sfg (using reduction rules).

The approximate sfg contains error, but note this error is not
too bad. The values of the approximate sfg are certainly
close to that of the exact sfg.

Q: Is there any way to improve the accuracy of this

approximation?

A: Certainly. The error is a result of truncating the infinite

propagation series. Note we severely truncated the series—
out of an infinite number of terms, we retained only three
(the forward paths). If we retain more terms, we will likely
get a more accurate answer.

Q: So why did these approximate answers turn out so well,

given that we only used three terms?

A: We retained the three most significant terms, we will find

that the forward paths typically have the largest magnitudes
of all propagation paths.

Q: Any idea what the next most significant terms are?

A: Yup. The forward paths are all those propagation paths

that pass through any node no more than one time. The next
most significant paths are almost certainly those paths that
pass through any node no more than two times.

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 13/18

Path 4 is an example of such a path:

a1 0.5 0. 8
j
0.144 0.35 0.4 0.5
j
b1 0.5 0.8

There are three more of these paths (passing through a node

no more than two times)—see if you can find them!

After determining the values for these paths, we can add 4

more terms to our summation (now we have seven terms!):

b1
Γin =
a1
7
≈ ∑ pn
n =1

= ( p1 + p2 + p3 ) + ( p4 + p5 + p6 + p7 )
= ( −0.036 ) + ( 0.014 + 0.0112 + 0.0112 + 0.0090 )
= 0.0094

Note this value is closer to the correct value of zero than

was our previous (using only three terms) answer of -0.036.

As we add more terms to the summation, this approximate

answer will get closer and closer to the correct value of zero.
However, it will be exactly zero (to an infinite number of
decimal points) only if we sum an infinite number of terms!

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 14/18

Q: The significance of a given path seem to be inversely

proportional to the number of times it passes through any
node. Is this true? If so, then why is it true?

A: It is true (generally speaking)! A propagation path that

travels though a node ten times is much less likely to be
significant to the propagation series (i.e., summation) than a
path that passes through any node no more than (say) four
times.

The reason for this is that the significance of a given term in

a summation is dependent on its magnitude (i.e., pn ). If the
magnitude of a term is small, it will have far less affect (i.e.,
significance) on the sum than will a term whose magnitude is
large.

Q: You seem to be saying that paths traveling through fewer

nodes have larger magnitudes than those traveling through
many nodes. Is that true? If so why?

A: Keep in mind that a microwave sfg relates wave

amplitudes. The branch values are therefore always
scattering parameters. One important thing about scattering
parameters, their magnitudes (for passive devices) are always
less than or equal to one!

Smn ≤ 1

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 15/18

Recall the value of a path is simply the product of each branch

that forms the path. The more branches (and thus nodes),
the more terms in this product.

Since each term has a magnitude less than one, the magnitude
of a product of many terms is much smaller than a product of
a few terms. For example:

3 10
− j 0.7 = 0.343 and − j 0.7 = 0.028

Æ In other words, paths with more branches (i.e., more

nodes) will typically have smaller magnitudes and so are
less significant in the propagation series.

Note path 1 in our example traveled along one branch only:

p1 = 0.144

Path 2 has five branches:

p2 = −0.1

Path 3 seven branches:

p3 = −0.08

Path 4 nine branches:

p4 = 0.014

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 16/18

Path 5 eleven branches:

p5 = 0.0112

Path 6 eleven branches:

p6 = 0.0112

Path 7 thirteen branches:

p7 = 0.009

Hopefully it is evident that the magnitude diminishes as the

path “length” increases.

Q: So, does this mean that we should abandon our four

reduction rules, and instead use a truncated propagation
series to evaluate signal flow graphs??

A: Absolutely not!

Remember, truncating the propagation series always results in

some error. This error might be sufficiently small if we
retain enough terms, but knowing precisely how many terms to
retain is problematic.

We find that in most cases it is simply not worth the

effort—use the four reduction rules instead (it’s not like
they’re particularly difficult!).

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 17/18

Q: You say that in “most cases” it is not worth the effort.

Are there some cases where this approximation is actually
useful??

A: Yes. A truncated propagation series (typically using only

the forward paths) is used when these three things are true:

1. The network or device is complex (lots of nodes and

branches).

2. We can conclude from our knowledge of the device

that the forward paths are sufficient for an accurate
approximation (i.e., the magnitudes of all other paths in
the series are almost certainly very small).

3. The branch values are not numeric, but instead are

variables that are dependent on the physical parameters
of the device (e.g., a characteristic impedance or line
length).

The result is typically a tractable mathematical equation that

relates the design variables (e.g., Z 0 or A ) of a complex
device to a specific device parameter.

For example, we might use a truncated propagation series to

approximately determine some function:

Γin ( Z 01 , A 1 , Z 02 , A 2 )

Jim Stiles The Univ. of Kansas Dept. of EECS

3/19/2009 The Propagation Series.doc 18/18

If we desire a matched input (i.e., Γin ( Z 01 , A 1 , Z 02 , A 2 ) = 0 ) we

can solve this tractable design equation for the (nearly)
proper values of Z 01 , A 1 , Z 02 , A 2 .

We will use this technique to great effect for designing

multi-section matching networks and multi-section coupled
line couplers.

e − jθ e − jθ e − jθ
a1 b2
jc1 sin θ e − jθ
jc2 sin θ e − jθ
jc3 sin θ e − jθ

jc1 sin θ e − jθ
jc2 sin θ e − jθ
jc3 sin θ e − j θ
b1 a2
e − jθ
e − jθ e − jθ

e − jθ e − jθ e − jθ
b3 a4
jc1 sin θ e − j θ jc2 sin θ e − j θ jc3 sin θ e − j θ
jc1 sin θ e − j θ jc2 sin θ e − j θ jc3 sin θ e − j θ

a3 b4
e − jθ e − jθ e − jθ

The signal flow graph of a three-section coupled-line coupler.

Jim Stiles The Univ. of Kansas Dept. of EECS