4/14/2010 5_6 Binomial Multisection Matching Transformers 1/1

5.6 – Binomial Multi-section

Matching Transformer
Reading Assignment: pp. 246-250

One way to maximize bandwidth is to construct a multisection

matching network with a function Γ (f ) that is maximally flat.

Q: Maximally flat? What kind of function is maximally flat?

This function maximizes bandwidth by providing a solution

that is maximally flat.

A: HO: MAXIMALLY FLAT FUNCTIONS

1. We can build a multisection matching network such

that the function Γ (f ) is a binomial function.

2. The binomial function is maximally flat.

Q: Meaning?

A: Meaning the function Γ (f ) is maximally flatÆ a wideband

solution!

HO: THE BINOMIAL MULTI-SECTION MATCHING TRANSFORMER

Jim Stiles The Univ. of Kansas Dept. of EECS

4/14/2010 Maximally Flat Transformer Functions present 1/7

Maximally Flat Functions

Consider some function f ( x ) . Say that we know the value of the function at x =1 is 5:

f ( x = 1) = 5

This of course says something about the functionf ( x ) , but it doesn’t tell us much!

We can additionally determine the first derivative of this function, and likewise evaluate
this derivative at x =1. Say that this value turns out to be zero:

d f (x )
=0
d x x =1

Note that this does not mean that the derivative of f ( x ) is equal to zero, it merely
means that the derivative of f ( x ) is zero at the value x = 1 . Presumably, d f ( x ) dx is
non-zero at other values of x.

Jim Stiles The Univ. of Kansas Dept. of EECS

4/14/2010 Maximally Flat Transformer Functions present 2/7

Taking derivates: way too fun to stop!

So, we now have two pieces of information about the function f ( x ) . We can add to this
list by continuing to take higher-order derivatives and evaluating them at the single point
x =1.

Let’s say that the values of all the derivatives (at x =1) turn out to have a zero value:

d f n (x )
= 0 for n = 1, 2, 3, " , ∞
d x n x =1

We say that this function is completely flat at the point x=1.

Because all the derivatives are zero at x =1, it means that the function cannot change in
value from that at x =1.

In other words, if the function has a value of 5 at x =1, (i.e., f ( x = 1 ) = 5 ), then the
function must have a value of 5 at all x !

The functionf ( x ) thus must be the constant function:

f (x ) = 5

Jim Stiles The Univ. of Kansas Dept. of EECS

4/14/2010 Maximally Flat Transformer Functions present 3/7

A more realistic function

Now let’s consider the following problem—say some function f ( x ) has the following form:

f (x ) = a x 3 + b x 2 + c x

We wish to determine the values a, b, and c so that:

f ( x = 1) = 5

and that the value of the function f ( x ) is as close to a value of 5 as possible in the
region where x =1.

In other words, we want the function to have the value of 5 at x =1, and to change from
that value as slowly as possible as we “move” from x =1.

Jim Stiles The Univ. of Kansas Dept. of EECS

4/14/2010 Maximally Flat Transformer Functions present 4/7

Completely flat in not possible!

Q: Don’t we simply want the completely flat function f ( x ) = 5 ?

A: That would be the ideal function for this case, but notice that solution is not an option.
Note there are no values of a, b, and c that will make:

a x3 +b x2 +c x = 5

for all values x.

Q: So what do we do?

A: Instead of the completely flat solution, we can find the maximally flat solution!

The maximally flat solution comes from determining the values a, b, and c so that as many
derivatives as possible are zero at the point x=1.

Jim Stiles The Univ. of Kansas Dept. of EECS

4/14/2010 Maximally Flat Transformer Functions present 5/7

How many derivatives can be zero?

For example, we wish to make the first derivate equal to zero at x =1:

d f (x )
0=
d x x =1
= (3ax 2 + 2bx + c )
x =1

= 3a + 2b + c

Likewise, we wish to make the second derivative equal to zero at x =1:

d 2 f (x )
0=
d x 2 x =1
= ( 6ax + 2b ) x =1
= 6a + 2b

Here we must stop taking derivatives, as our solution only has three degrees of design
freedom (i.e., 3 unknowns a, b, c).

Jim Stiles The Univ. of Kansas Dept. of EECS

4/14/2010 Maximally Flat Transformer Functions present 6/7

We’re out of degrees of design freedom

Q: But we only have taken two derivatives, can’t we take one more?

A: No! We already have a third “design” equation: the value of the function must be 5 at
x =1:
5 = f ( x = 1)
3 2
= a ( 1 ) + b (1 ) + c (1 )
= a +b +c

So, we have used the maximally flat criterion at x =1 to generate three equations and
three unknowns:
5 = a +b +c

0 = 3a + 2b + c

0 = 6a + 2b

Solving, we find:
a =5
b = −15
c = 15

Jim Stiles The Univ. of Kansas Dept. of EECS

4/14/2010 Maximally Flat Transformer Functions present 7/7

Look! The function is maximally flat at x=1!

Therefore, the maximally flat function (at x =1) is:

f ( x ) = 5x 3 − 15x 2 + 15x

10

7.5
f(x)

2.5

0
0 0.5 1 1.5 2

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 1/24

The Binomial Multi-Section

Transformer
Recall that a multi-section matching network can be described using the theory of small
reflections as:

Γin (ω ) = Γ 0 + Γ1e − j 2ωT + Γ2e − j 4ωT + " + ΓN e − j 2N ωT

N
= ∑ Γn e − j 2n ωT
n =0

where:
A
T  = propagation time through 1 section
vp

Note that for a multi-section transformer, we have N degrees of design freedom,

corresponding to the N characteristic impedance values Zn .

Z0 Zin Z1 Z2 " ZN RL

A A A
Jim Stiles The Univ. of Kansas Dept. of EECS
4/15/2010 The Binomial Multisection Matching Transformer present.doc 2/24

Behold the Binomial Function!

Q: What should the values of Γn (i.e., Zn ) be?

A: We need to define N independent design equations, which we can then use to

solve for the N values of characteristic impedance Zn .

First, we start with a single design frequency ω0 , where we wish to achieve a perfect
match:
Γin (ω = ω0 ) = 0

That’s just one design equation: we need N -1 more!

These addition equations can be selected using many criteria—one such criterion is to
make the function Γin (ω ) maximally flat at the point ω = ω0 .

To accomplish this, we first consider the Binomial Function:

( )
N
Γ (θ ) = A 1 + e − j 2θ

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 3/24

What’s so special about the Binomial Function!

The Binomial Function has the desirable properties that:

Γ (θ = π 2 ) = A (1 + e − j π )
N

N
= A (1 − 1 )
=0
and that:

d n Γ (θ )
= 0 for n = 1, 2, 3, " , N − 1
dθ n
θ =π 2

In other words, this Binomial Function is maximally flat at the point θ = π 2 , where it has
a value of Γ (θ = π 2 ) = 0 .

Q: So? What does this have to do with our multi-section matching network?

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 4/24

A: Plenty!
Let’s expand (multiply out the N identical product terms) of the Binomial Function:

( )
N
Γ (θ ) = A 1 + e − j 2θ

= A (C N
0 + C1N e − j 2θ + C 2N e − j 4θ + C 3N e − j 6θ + " + C NN e − j 2N θ )
where:
N!
CnN 
(N − n ) ! n !

Compare this to an N-section transformer function:

Γin (ω ) = Γ 0 + Γ1e − j 2ωT + Γ2e − j 4ωT + " + ΓN e − j 2N ωT

and it is obvious the two functions have identical forms, provided that:

Γn = A CnN and ωT = θ

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 5/24

See, the Binomial Function is very useful!

Moreover, we find that this function is very desirable from the standpoint of a matching
network. Recall that Γ (θ ) = 0 at θ = π 2 --a perfect match!

Additionally, the function is maximally flat at θ = π 2 , therefore Γ (θ ) ≈ 0 over a wide

range around θ = π 2 --a wide bandwidth!

Q: But how does θ = π 2 relate to frequency ω ?

A: Remember that ωT = θ , so the value θ = π 2 corresponds to the frequency:

1 π vp π
ω0 = =
T 2 A 2

This frequency ( ω0 ) is therefore our design frequency—the frequency where we have a

perfect match.

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 6/24

What about the length of each section?

Note that the section-length A has an interesting relationship with this frequency:

vp π 1 π λ0 π λ0
A= = = =
ω0 2 β 0 2 2π 2 4

In other words, a Binomial Multi-section matching network will have a perfect match at
the frequency where the section lengths A are a quarter wavelength!

Thus, we have our first design rule:

Set section lengths A so that they are a quarter-wavelength ( λ0 4 ) at the design

frequency ω0 .

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 7/24

And that pesky constant A ?

Q: I see! And then we select all the values Zn such that Γn = A CnN . But wait! What
is the value of A ??

A: We can determine this value by evaluating a boundary condition!

Specifically, we can easily determine the value of Γ (ω ) at ω = 0 .

Z0 Zin Z1 Z2 " ZN RL

A A A
Note as ω approaches zero, the electrical length β A of each section will likewise approach
zero. Thus, the input impedance Zin will simply be equal to RL as ω → 0 .

As a result, the input reflection coefficient Γ (ω = 0 ) must be:

Zin (ω = 0 ) − Z 0 RL − Z 0
Γ (ω = 0 ) = =
Zin (ω = 0 ) + Z 0 RL + Z 0

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 8/24

Aren’t boundary conditions great ?

However, we likewise know that:

( )
N
Γ (0) = A 1 + e − j 2( 0 )

N
= A (1 + 1 )
= A 2N

Equating the two expressions:

RL − Z 0
Γ ( 0 ) = A 2N =
RL + Z 0
And therefore:

RL − Z 0
A = 2− N (A can be negative!)
RL + Z 0

We now have a form to calculate the required marginal reflection coefficients Γn :

AN!
Γn = ACnN =
(N − n ) !n !

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 9/24

How do I determine characteristic impedance?

Of course, we also know that these marginal reflection coefficients are physically related
to the characteristic impedances of each section as:

Z n +1 − Z n
Γn =
Z n +1 + Z n

Equating the two and solving, we find that that the section characteristic impedances
must satisfy:

1 + Γn 1 + ACnN
Z n +1 = Zn = Zn
1 − Γn 1 − ACnN

Note this is an iterative result—we determine Z1 from Z0, Z2 from Z1, and so forth.

Q: This result appears to be our second design equation. Is there some reason why
you didn’t draw a big blue box around it?

A: Alas, there is a big problem with this result.

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 10/24

The BIG problem with this result!

Note that there are N +1 coefficients Γn (i.e., n ∈ { 0,1, ", N } ) in the Binomial series, yet
there are only N design degrees of freedom (i.e., there are only N transmission line
sections!).

Thus, our design is a bit over constrained, a result that manifests itself the finally
marginal reflection coefficient ΓN .

Note from the iterative solution above, the last transmission line impedance Z N is
selected to satisfy the mathematical requirement of the penultimate reflection
coefficient ΓN −1 :

Z N − Z N −1
ΓN −1 = = A C NN−1
Z N + Z N −1

Thus the last impedance must be:

1 + A C NN−1
Z N = Z N −1
1 − AC NN−1

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 11/24

Our degrees of freedom have run out!

But there is one more mathematical requirement! The last marginal reflection coefficient
must likewise satisfy:
R − Z0
ΓN = A C NN = 2−N L
RL + Z 0

where we have used the fact that C NN = 1 .

But, we just selected Z N to satisfy the requirement for ΓN −1 ,—we have no physical design
parameter to satisfy this last mathematical requirement!

As a result, we find to our great consternation that the last requirement is not satisfied:

RL − Z N
ΓN = ≠ A C NN !!!!!!
RL + Z N

Q: Yikes! Does this mean that the resulting matching network will not have the
desired Binomial frequency response?

A: That’s exactly what it means!

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 12/24

&*#@*!&!!!!!
Q: You big #%@#$%&!!!! Why did you waste all my time by discussing an over-
constrained design problem that can’t be built?

A: Relax; there is a solution to our dilemma—albeit an approximate one.

You undoubtedly have previously used the approximation:

y −x 1 ⎛y ⎞
≈ ln ⎜ ⎟
y +x 2 ⎝x ⎠

y
An approximation that is especially accurate when y − x is small (i.e., when x
 1 ).

1.0 1 ⎛y ⎞
ln ⎜ ⎟
2 ⎝x ⎠
y −x
0.5 y +x

2 4 6 8 10
y
- 0.5 x

- 1.0

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 13/24

Use this approximation for value A!

Now, we know that the values of Zn +1 and Zn in a multi-section matching network are
typically very close, such that Zn +1 − Zn is small.

Thus, we use the approximation:

Z n +1 − Z n 1 ⎛ Z n + 1 ⎞
Γn = ≈ ln ⎜ ⎟
Z n +1 + Z n 2 ⎝ Z n ⎠

Likewise, we can also apply this approximation (although not as accurately) to the value of
A:

RL − Z 0 − ( N +1 ) ⎛ RL ⎞
A =2 −N
≈2 ln ⎜ ⎟
RL + Z 0 Z
⎝ 0⎠

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 14/24

Let’s try this again—with approximations!

So, let’s start over, only this time we’ll use these approximations. First, determine A :

⎛R ⎞
A ≈ 2 ( ) ln ⎜ L ⎟
− N +1
(A can be negative!)
⎝ Z0 ⎠

Now use this result to calculate the mathematically required marginal reflection
coefficients Γn :

AN!
Γn = A CnN =
(N − n ) !n !

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 15/24

Here’s (finally) our second design rule!

Of course, we also know that these marginal reflection coefficients are physically related
to the characteristic impedances of each section as:

1 ⎛ Z n +1 ⎞
Γn ≈ ln ⎜ ⎟
2 ⎝ Zn ⎠

Equating the two and solving, we find that that the section characteristic impedances
must satisfy:

Zn +1 = Zn exp ⎡⎣2 Γn ⎤⎦

Now this is our second design rule. Note it is an iterative rule—we determine Z1 from Z0,
Z2 from Z1, and so forth.

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 16/24

I don’t understand what just happened

Q: Huh? How is this any better? How does applying approximate math lead to a better
design result??

A: Applying these approximations help resolve our over-constrained problem. Recall that
the over-constraint resulted in:

RL − Z N
ΓN = ≠ A C NN
RL + Z N

But, as it turns out, these approximations leads to the happy situation where:

1 ⎛ RL ⎞
ln ⎜ ⎟ = A CN
N
ΓN ≈ Å A Sanity check!!
2 ⎝ ZN ⎠

provided that the value A is likewise the approximation given above.

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 17/24

I still don’t understand what just happened

Effectively, these approximations couple the results, such that each value of
characteristic impedance Zn approximately satisfies both Γn and Γn +1 . Summarizing:

* If you use the “exact” design

equations to determine the
characteristic impedances Zn , the
last value ΓN will exhibit a significant
numeric error, and your design will
not appear to be maximally flat.

* If you instead use the “approximate”

design equations to determine the
characteristic impedances Zn , all
values Γn will exhibit a slight error,
but the resulting design will appear to
be maximally flat, Binomial
reflection coefficient function Γ (ω ) !
Note that as we increase the number of
sections, the matching bandwidth increases.

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 18/24

Bandwidth: How do we define it?

band·width (band′widt̸h′, -witt̸h′) - noun

1. the range of frequencies within a ….

Q: Can we determine the value of this bandwidth?

A: Sure! But we first must define what we mean by bandwidth.

As we move from the design (perfect match) frequency f0 the value Γ (f ) will increase.
At some frequency (fm, say) the magnitude of the reflection coefficient will increase to
some unacceptably high value ( Γm , say). At that point, we no longer consider the device
to be matched.

Γ (f )

Δf
Γm
f

fm1 f0 fm2

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 19/24

Bandwidth: How do we calculate it?

Note there are two values of frequency fm —one value less than design frequency f0, and
one value greater than design frequency f0.

These two values define the bandwidth Δf of the matching network:

Δf = fm 2 − fm 1 = 2 (f0 − fm 1 ) = 2 (fm 2 − f0 )

Q: So what is the numerical value of Γm ?

A: I don’t know—it’s up to you to decide!

Every engineer must determine what they consider to be an acceptable match (i.e., decide
Γm ).

This decision depends on the application involved, and the specifications of the overall
microwave system being designed.

However, we typically set Γm to be 0.2 or less.

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 20/24

We get to perform some Algebra!!

Q: OK, after we have selected Γm , can we determine the two frequencies
fm ?

A: Sure! We just have to do a little algebra.

We start by rewriting the Binomial function:

( )
N
Γ (θ ) = A 1 + e − j 2θ

= Ae − jN θ (e + j θ + e − j θ )
N

= Ae − jN θ (e + j θ + e − j θ )
N

N
= Ae − jN θ (2 cos θ )

Now, we take the magnitude of this function:

N
Γ (θ ) = 2N A e − jN θ cos θ
N
= 2N A cos θ

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 21/24

It gets better—even more algebra!!

Now, we define the values θ where Γ (θ ) = Γm as θm . I.E., :

Γm = Γ (θ = θm )
N
= 2N A cos θm

We can now solve for θm (in radians!) in terms of Γm :

⎡ ⎛ Γ ⎞ 1N ⎤ ⎡ 1
⎛ Γm ⎞ N ⎤
1 −1 ⎢ 1
θm 1 = cos −1 ⎢ ⎜ m ⎟ ⎥ θm 2 = cos − ⎜ ⎟ ⎥
⎢2 ⎝ A ⎠ ⎥ ⎢ 2⎝ A ⎠ ⎥
⎣ ⎦ ⎣ ⎦

Note that there are two solutions to the above equation (one less that π 2 and one
greater than π 2 )!

Now, we can convert the values of θm into specific frequencies.

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 22/24

Converting θm to fm
Recall that ωT = θ , therefore:
1 vp
ωm = θm = θm
T A

But recall also that A = λ0 4 , where λ0 is the wavelength at the design frequency f0 (not
fm !), and where λ0 = v p f0 .

Thus we can conclude:

vp 4v p
ωm = θm = θm = ( 4f0 ) θm
A λ0

or:

fm =
1 vp
θm =
( 4f0 ) θm = (2f0 ) θm
2π A 2π π

where θm is expressed in radians.

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 23/24

And thus the bandwidth is…

Therefore:

1 1
⎡ ⎤ ⎡ ⎛ Γm ⎞ N ⎤
2f0 −1 ⎢ 1 ⎛ Γm ⎞ N ⎥ 2f0 −1 ⎢ 1
fm 1 = cos + ⎜ ⎟ fm 2 = cos − ⎜ ⎟ ⎥
π ⎢ 2⎝ A ⎠ ⎥ π ⎢ 2⎝ A ⎠ ⎥
⎣ ⎦ ⎣ ⎦

Thus, the bandwidth of the binomial matching network can be determined as:

Δf = 2 (f0 − fm 1 )
1
⎡ ⎤
4f0 −1 ⎢ 1 Γm N ⎥
⎛ ⎞
= 2f0 − cos + ⎜ ⎟
π ⎢ 2⎝ A ⎠ ⎥
⎣ ⎦

Note that this equation can be used to determine the bandwidth of a binomial matching
network, given Γm and number of sections N.

However, it can likewise be used to determine the number of sections N required to

meet a specific bandwidth requirement!

Jim Stiles The Univ. of Kansas Dept. of EECS

4/15/2010 The Binomial Multisection Matching Transformer present.doc 24/24

In summary, our design steps

Finally, we can list the design steps for a binomial matching network:

1. Determine the value N required to meet the bandwidth ( Δf and Γm )

requirements.

2. Determine the approximate value A from Z 0 , RL and N.

3. Determine the marginal reflection coefficients Γn = A CnN required by the

binomial function.

4. Determine the characteristic impedance of each section using the iterative

approximation:
Zn +1 = Zn exp ⎡⎣2 Γn ⎤⎦
5. Perform the sanity check:

1 ⎛ RL ⎞
ln ⎜ ⎟ = A CN
N
ΓN ≈
2 ⎝ ZN ⎠

6. Determine section length A = λ0 4 for design frequency f0.

Jim Stiles The Univ. of Kansas Dept. of EECS