0.

1 Multiple Time Constant Circuit

0.1.1 System Theory approach
Consider the circuit we are studying (example with 2 capacitors only):

C1

ic1
vc1

iin iout

passive load
LTI
generator

vin No energy storage vout

elements
No independent
generators

vc2
ic2

C2
Since it is Linear Time Independent (LTI) we may study it with a system approach.

According to what is taught in the course of Automatic Control, we may identify the following
types of variables:

• inputs ui the variables we are able to control

• outputs yi the variables whose behavior we want to observe

• states xi the variables that appear under the sign of derivative inside the characteristic
equations
 
d dx
Notice that, if we have a double derivative we may consider it as: . Therefore, both
dt dt
dx
and x appear under signs of derivative!
dt
dx
So both x and are states for the system!
dt
In particular, since the characteristic equations for capacitors and inductors are:
dvC
• Capacitor: iC = C
dt
diL
• Inductor: vL = L
dt

1
So, the best choice for the states is:
• Voltages across the terminals of capacitors
• Currents through the inductors
We may collect all the inputs, the outputs and the states respectively into an input vector, an
output vector and a state vector:
     
x1 u1 y1
 ..   ..   .. 
x= .  u= .  y=.
xn up yq
The system can be then represented as:

u1 LTI y1
u2 x1 y2
u= ... x2 y= ...
x= ...
up yq
xn

Notice that we can consider the states as ” internal ” parameters of the LTI network.

Since the states are the only quantities under the signs of derivative we can write a matrix
equation:
d
x=Ax+Bu
dt
Considering now the LTI memoryless element we can obtain a matrix equation that relates the
outputs to the inputs and the states (the states can be considered as inputs at the ports of the
memoryless element to which capacitors and inductors are connected).

So, the outputs are linear combinations of the inputs and the states:

y =C x+Du

So, the relation between the outputs and the inputs of a LTI network is fully described by the
set of matrix equations:

d
 x=Ax+Bu

dt

y = C x + D u

We can now apply the Laplace transform to both sides of both of the equations to get:
  
L d   
s X − x (0+ ) = A X + B U
 x =L Ax+Bu
dt ⇒
   Y =C X +DU
   
L y =L C x+Du

2
We can obtain an expression in the Laplace domain for the state vector X:

s X − x (0+ ) = A X + B U

s X − A X = B U + x (0+ )
s I − A X = B U + x (0+ )

−1 
B U + x (0+ )

X = sI −A

−1
−1
X = sI −A BU + sI −A x (0+ )
We can obtain an expression in the Laplace domain for the output vector X:

Y =C X +DU
h
−1
−1 i
Y =C sI −A BU + sI −A x (0+ ) + D U

−1
−1
Y =C sI −A BU +C sI −A x (0+ ) + D U
Adding the previously omitted s-dependencies (omitted to have a lighted notation) we get:
h
−1 i
−1
Y = C sI −A B + D U (s) + C s I − A x (0+ )
| {z } | {z }
transfer function H(s) W (s)

Let’s now recall the inversion formula for matrices:

adj(A)
A−1 =
det(A)

With adj(A) the adjoint matrix of A (transpose of its matrix of cofactors). (See Linear Algebra)

So, we may rewrite the last equation as:

"
#

adj s I − A adj s I − A
Y = C
B + D U (s) + C
x (0+ )
det s I − A det s I − A
| {z } | {z }
transfer function H(s) W (s)

Notice now that the entries of adj s I − A are polynomials in s and all have degree strictly


smaller than the one of det s I − A .

Compare now the two expressions:

"
# "

#
adj s I − A C adj s I − A B + D det s I − A
H(s) = C
B+D =

det s I − A det s I − A


adj s I − A
W (s) = C

det s I − A


Notice that the two expressions both have denominator det s I − A .

The poles of the transfer function H(s) are a subset of the poles of the transfer function W (s)
since they have the same denominator and, therefore, we can just have zero-pole cancellations

3
(since the numerators are different) that lead to the disappearance of some of the poles of W (s)
from H(s).

So, in order to determine the poles of the transfer function, instead of studying the zero-state
output response we could study the zero-input output response!
Moreover, since the zero-input response is:

Yzi (s) = C Xzi (s)

and there is a proportionality relation between the zero-input output response Yzi (s) and the
zero-input state response Xzi (s), the two share the exact same poles!

So, we can study the poles of the zero-input state response Xzi (s) to find the ones of Yzi (s)
and, consequently, argue that the system poles are a subset of these.

0.1.2 Circuit Theory approach

Since we know that, in order to determine the poles of the transfer function, it is enough to
study the poles of the zero-input state response and argue that they contain the system poles
(it is then easy to determine whether some poles cancel out with the zeroes).

We can thus switch off the source, by replacing:

• a voltage source with an open-circuit

• a current source with a short-circuit

We can then consider the circuit as a LTI memoryless multiport element with just capacitors
attached to the ports (there is no need to consider an outout port, so will keep it ”lumped”
inside the LTI memoryless element, along with the load:

ic1 ic2

C1 LTI C2
vc1 No energy storage vc2
elements
Independent generators
removed

Let’s now suppose the LTI memoryless element admits a conductance matrix representation:
 
g11 g12
G=
g21 g22

And we have the following relation for the LTI memoryless element:

i (t) = G v (t)

4
     
ic1 (t) g11 g12 vc1 (t)
=
ic2 (t) g21 g22 vc2 (t)
We can then write down the two characteristic equations of the two capacitors:

d
ic1 = −C1 vc1 (t)


dt
ic2 = −C2 d vc2 (t)


dt
We can then substitute the two characteristic equations of the two capacitors into the LTI
memoryless element matrix equation:
 
d
−C v
 1 dt c1 (t)   
  = g11 g12 vc1 (t)
 d  g21 g22 vc2 (t)
−C2 vc2 (t)
dt
We can then reorder it to get:
 g11 g12 
   
d vc1 (t) C C1  vc1 (t)
− = g 1 g22  vc2 (t)
dt vc2 (t) 21
C2 C2
We can then apply the Laplace transform to both sides of the equation to get:
 g g12  
11
      
d vc1 (t) 
 C1 C1  vc1 (t)

L − = L g g22
dt vc2 (t)  21 vc2 (t) 

 
C2 C2
 g11 g12 
+
      
Vc1 (s) vc1 (0 )  C1 C1  Vc1 (s)
− s − = g g22  Vc2 (s)
Vc2 (s) vc2 (0+ ) 21
C2 C2
 g11 g12 
vc1 (0+ )
     
Vc1 (s)  C1 C1  Vc1 (s)
−sI + = g g22  Vc2 (s)
Vc2 (s) vc2 (0+ ) 21
C2 C2
 g11 g12 
vc1 (0+ )
     
 C1 C1  Vc1 (s) Vc1 (s)
= g g22  Vc2 (s) + s I Vc2 (s)
vc2 (0+ ) 21
C2 C2
 g11 g12 
+ s
vc1 (0+ )
   
 C1 C1  Vc1 (s)
= g g22
vc2 (0+ ) 21 Vc2 (s)

+s
C2 C2
 g11 g12 −1
+s
vc1 (0+ )
   
Vc1 (s)  C1 C1 
= g g22
Vc2 (s) 21 vc2 (0+ )

+s
C2 C2

5
By the inverse matrix formula the denominators of the components Vc1 (s) and Vc2 (s) of the
voltage vector are given by:  g11 g12 
+s
C C1 
det  1g g
21 22

+s
C2 C2
We can compute this determinant:
 g11 g12 
+s    
 C1 C1  g11 g22 g12 g21
det  g g = +s · +s − ·
21 22 C1 C2 C1 C2

+s
C2 C2
   
g11 g22 g11 g22 g12 g21
= s2 + + s+ · − ·
C1 C2 C1 C2 C1 C2
We can now multiply this characteristic polynomial by C1 · C2 without affecting its roots, which
are the poles of the zero-input state response:
 g11 g12 
+s
C C1 
C1 · C2 det  1g g22
2
 = C1 C2 s + (g11 C2 + g22 C1 ) s + (g11 · g22 − g12 · g21 )
21
+s | {z
∆G
}
C2 C2
With ∆G = g11 ·g22 −g12 ·g21 the determinant of the conductance matrix of the LTI memoryless
element.

Let’s call ∆1 (s) the obtained polynomial:

∆1 (s) = C1 C2 s2 + (g11 C2 + g22 C1 ) s + (g11 · g22 − g12 · g21 )
We can now consider the resistance matrix R of the LTI memoryless element:
 
r r
R = 11 12
r21 r22
We have the following two relations:
v (t) = R i (t)
R = G−1
Since the resistance matrix is the inverse of the conductance matrix we can apply the formula
for the inverse matrix to get the following four results:
g22 g22
r11 = =
∆G g11 g22 − g12 g21
g11 g11
r22 = =
∆G g11 g22 − g12 g21
r22 r22
g11 = =
∆R r11 r22 − r12 r21
r11 r11
g22 = =
∆R r11 r22 − r12 r21
Moreover, we recall that the determinant of the inverse of a matrix is the reciprocal of the
determinant of the matrix.

Therefore:
1
∆R =
∆G

6
We can now divide again the polynomial ∆1 (s) by ∆G (since the roots do not change), obtaining
∆2 (s):
1 g
11 g22 
∆2 (s) = C1 C2 s2 + C2 + C1 s + 1
∆G ∆G ∆G
We can also rewrite it in terms of the elements of the resistance matrix:
∆3 (s) = ∆R C1 C2 s2 + (r11 C1 + r22 C2 ) s + 1
We can also obtain simpler expressions for ∆R and ∆G:
g22 1 r11
r11 = ⇔ =
∆G ∆G g22
r22 r22
g11 = ⇔ ∆R =
∆R g11
g11 1 r22
r22 = ⇔ =
∆G ∆G g11
r11 r11
g22 = ⇔ ∆R =
∆R g22
Therefore, we may substitute these eight relationships into the expressions for ∆2 (s) and ∆3 (s)
to obtain:
r11
∆4 (s) = C1 C2 s2 + (r11 C1 + r22 C2 ) s + 1
g22
r22
∆5 (s) = C1 C2 s2 + (r11 C1 + r22 C2 ) s + 1
g11
Moreover, we can use the two relationships:
r22 r11
∆R = ∆R =
g11 g22
and compare them to get:
r22 r11
=
g11 g22
(it will be useful later)

General case

We may find the poles of the system by finding the roots of any of this polynomials:
∆1 (s) = C1 C2 s2 + (g11 C2 + g22 C1 ) s + (g11 · g22 − g12 · g21 )
1 g
11 g22 
∆2 (s) = C1 C2 s2 + C2 + C1 s + 1
∆G ∆G ∆G
2
∆3 (s) = ∆R C1 C2 s + (r11 C1 + r22 C2 ) s + 1
r11
∆4 (s) = C1 C2 s2 + (r11 C1 + r22 C2 ) s + 1
g22
r22
∆5 (s) = C1 C2 s2 + (r11 C1 + r22 C2 ) s + 1
g11
However, despite being easily solvable second-degree equations, determining the set of parame-
ters required for writing them, which is either one of the following:
• (g11 , g22 , g12 , g21 )
• (r11 , r22 , r12 , r21 )
• (r11 , r22 , g11 , g22 )
can be a quite long task, which makes the use of this method not convenient at all.

7
Moreover, this general method has no design usefulness, since we cannot foresee the effect of a
change in the topology or in the values of the components on the position of the poles.

This is why the next two special cases will be very important in the study of amplifiers.

Special case 1: TOPOLOGICALLY SEPARATED CAPACITORS

Let’s consider the special case in which we have both r12 = 0 and r21 = 0.

The equation for ∆3 (s) can be divided by C1 C2 ∆R, obtaining:

 
2 r11 r22 1
∆6 (s) = s + C1 + C2 s +
∆R C1 C2 ∆R C1 C2 ∆R C1 C2
We can notice that the determinant ∆R becomes:

∆R = r11 r22 − r12 r21 = r11 r22

∆R = r11 r22
Substituting it into the equation for ∆6 (s) we get:
 
2 r11 r22 1
∆6 (s) = s + C1 + C2 s +
r11 r22 C1 C2 r11 r22 C1 C2 r11 r22 C1 C2
    
1 1 1 1
∆6 (s) = s2 + + s+
r11 C1 r22 C2 r11 C1 r22 C2
We recall that, whenever we have a second-degree polynomial in the variable z of the kind:

z2 + S · z + P

with: (
S =a+b
P =a·b
it can be rewritten as:
(z + a) · (z + b)
So, the equation for ∆6 (s) can be rewritten as:
    
2 1 1 1 1
∆6 (s) = s + + s+
r11 C1 r22 C2 r11 C1 r22 C2
  
1 1
= s+ s+
r11 C1 r11 C1
The positions of the two poles are, therefore:
1 1
sp1 = − ⇒ fp1 = −
r11 C1 2πr11 C1
1 1
sp2 = − ⇒ fp2 = −
r22 C2 2πr22 C2
Notice that r12 = 0 and r21 = 0 means that each of the capacitor ports is not influenced by
what is connected to the other.

8
Moreover, let’s consider the new resistance matrix for the LTI memoryless element seen from
the two ports of the capacitors:  
r 0
R = 11
0 r22
We can consider the matrix equation:
v (t) = R i (t)
It becomes:  
r11 0
v (t) = i (t)
0 r22
the associated system of equations is:
(
vc1 = r11 ic1
vc2 = r22 ic2
From these equations we can see that:
• the equivalent impedance seen by the first capacitor C1 is r11
• the equivalent impedance seen by the second capacitor C2 is r22
So, the equivalent impedance seen by the first capacitor is real (it’s a resistance) and does not
depend on the second capacitor and viceversa.

Therefore, we can see that the two coefficients r12 and r21 are zero from this fact that the
equivalent impedance seen by each capacitor does not involve the other capacitor.

So, the two capacitors do not interact, which is the reason why two such capacitors are called
topologically separated.

The computation of the system poles is then nothing but the computation of two pole frequen-
cies using the same method derived for the Single Time Constant circuits:

• Compute the equivalent impedance seen by each of the two capacitors (switching off all
independent sources)
• If such impedances are real, the two capacitors are topologically separated
• Derive the pole frequencies as:
1
fp1 = −
2πReq (C1 ) C1
1
fp2 = −
2πReq (C2 ) C2

Note: In general and in cases with more than 2 capacitors we have that a capacitor Ci is
topologically separated from the others if the equivalent impedance Zeq (Ci ) seen by it is real
and does not depend on any of the other capacitors (it is a resistance Req (Ci )). In this case the
pole frequency associated to it can be computed using the formula:
1
fpi = −
2πReq (Ci ) Ci

9
Special case 2: NUMERICALLY SEPARATED CAPACITORS

In general, capacitors are not topologically separated but we can have good numerical approx-
imations.

Let’s now see how we can evaluate the parameters g11 ,g22 ,r11 ,r22 directly from the circuit.

CASE 1
Let’s start by studying the equivalent resistance seen by capacitor 1 when capacitor 2 is substi-
tuted with an open circuit:

ic1 ic2
LTI memoryless element:

LTI   
vc1
vc2
=
 
r11 r12 ic1
r21 r22 ic2
vc1 No energy storage vc2
elements open circuit at port 2:
Independent generators
removed ic2 = 0

The equivalent resistance seen by capacitor 1 when capacitor 2 is an open circuit will be denoted
OC (C1) and is obtained in the following way:
by Req

vc1 = r11 ic1 + r12 ic2

vc1 = r11 ic1 + r12 0 = r11 ic1

vc1 = r11 ic1
So, we have that:
OC
Req (C1) = r11
CASE 2
Let’s study the equivalent resistance seen by capacitor 2 when capacitor 1 is substituted with
an open circuit:

ic1 ic2
LTI memoryless element:

LTI   
vc1
vc2
=
 
r11 r12 ic1
r21 r22 ic2
vc1 No energy storage vc2
elements open circuit at port 1:
Independent generators
removed ic1 = 0

The equivalent resistance seen by capacitor 2 when capacitor 1 is an open circuit will be denoted
OC (C2) and is obtained in the following way:
by Req

vc2 = r21 ic1 + r22 ic2

10
 vc2 = r21 i0 + r22 ic2 = r22 ic2
vc2 = r22 ic2
So, we have that:
OC
Req (C2) = r22
CASE 3
Let’s study the equivalent resistance seen by capacitor 1 when capacitor 2 is substituted with
an short circuit:

ic1 ic2
LTI memoryless element:

LTI   
ic1
ic2
=
 
g11 g12 vc1
g21 g22 vc2
vc1 No energy storage vc2
elements short circuit at port 2:
Independent generators
removed vc2 = 0

The equivalent resistance seen by capacitor 1 when capacitor 2 is a short circuit will be denoted
SC (C1) and is obtained in the following way:
by Req

ic1 = g11 vc1 + g12 vc2

ic1 = g11 vc1 + g12 0 = g11 vc1

ic1 = g11 vc1
So, we have that:
SC 1
Req (C1) =
g11
CASE 4
Let’s study the equivalent resistance seen by capacitor 2 when capacitor 1 is substituted with
an short circuit:

ic1 ic2
LTI memoryless element:

LTI   
ic1
ic2
=
 
g11 g12 vc1
g21 g22 vc2
vc1 No energy storage vc2
elements short circuit at port 1:
Independent generators
removed vc1 = 0

The equivalent resistance seen by capacitor 1 when capacitor 2 is a short circuit will be denoted
SC (C1) and is obtained in the following way:
by Req

ic2 = g21 vc1 + g22 vc2

11
 ic2 = g21 0 + g22 vc2 = g22 vc2
ic2 = g22 vc2
So, we have that:
SC 1
Req (C2) =
g22
SUMMARIZING:

OC
Req (C1) = r11
OC
Req (C2) = r22
SC 1
Req (C1) =
g11
SC 1
Req (C2) =
g22
we can now compute four time constants using these four equivalent resistances and the corre-
sponding capacitors:
τ1OC = r11 C1
τ2OC = r22 C2
1
τ1SC = C1
g11
1
τ2SC = C2
g22
Let’s now recall the two expressions we previously derived:
r11
∆4 (s) = C1 C2 s2 + ( r11 C1 + r22 C2 ) s + 1
g22
r22
∆5 (s) = C1 C2 s2 + (r11 C1 + r22 C2 ) s + 1
g11
We can rewrite them substituting the derived time constants:

∆4 (s) = τ1OC τ2SC s2 + τ1OC + τ2OC s + 1

∆5 (s) = τ1SC τ2OC s2 + τ1OC + τ2OC s + 1

Let’s now suppose that the time constants with subscript 1 are both smaller than each of the
time constants with subscript 2. (when this is not true we can just relabel the two ports and
capacitors to make it true)

The two equations are the same, from which we derive the following constraint:

τ1OC τ2SC = τ1SC τ2OC

τ1OC τ2OC
=
τ1SC τ2SC
So, the relative orders of the short-circuit and open-circuit constants for both capacitors is the
same.

12
Consider now the case:
0 < τ1SC < τ1OC  τ2SC < τ2OC
which is the case for most resistive networks and a significant part of the active networks.

(the order or the subscripts 1,2 can be reversed, which is solvable by relabeling the capacitors
but not the order of the superscripts OC,SC )
(one reason for requesting the time constants to be greater than zero is that, even if the next
computations worked for some negative values of the time constants, at least one of the poles
would be positive because of the results, so there would be no point in applying the method to
an unstable system)

We now want the two poles of the system to be real!

This can be achieved by checking the discriminant of the second-order polynomial for ∆4 (s):
∆4 (s) = τ1OC τ2SC s2 + τ1OC + τ2OC s + |{z}


1
| {z }
c
| {z }
a b

2
b2 − 4 a c = τ1OC + τ2OC − 4 τ1OC τ2SC
= (τ1OC )2 + (τ2OC )2 + 2 τ1OC τ2OC − 4 τ1OC τ2SC
SC
 
OC 2 OC OC OC OC τ2
= (τ2 ) + τ2 τ2 + 2 τ1 − 4 τ1
τ2OC
| {z }
>0 for 0<τ1SC <τ1OC τ2SC <τ2OC

>0 for (0 < τ1SC < τ1OC  τ2SC ) < τ2OC

So, if the separation between τ1OC and τ2SC is large enough (a decade will do) the poles of the
system are real.

Moreover, since the time constants are all four positive (by hypothesis), we have:
a>0 b>0 c>0
so the two poles will both be real negative, by Descartes’ rule of signs (no changes in the signs).

Since ∆4 (s) = ∆5 (s) is a second-degree equation it is possible to associate it to a parabola,

which must have intersections with the abscissa axis, since we know that real negative roots
must exist:

Δ4=Δ5

- 1/τ1sc - 1/τ2oc
sp1 sp2
- 1/τ1oc - 1/τ2sc
s

13
this parabola will always be oriented upward as in the figure, since the coefficient a of the
highest-degree term is positive.

So, it is enough to show that the points of the parabola corresponding to the opposite of the
reciprocal of the time constants (which are generally correspondent to the angular frequencies
1
of the poles since: fp = − in single time constant circuits) lie below the abscissa
2π Req (C)C
| {z }
τ
axis to show that the solutions are external to all the intervals defined by the opposites of the
reciprocals of all the time constants, because of the continuity of the parabola.

In order to do this, let’s start from the left extremum τ2OC and evaluate ∆4 (s) at the pole
corresponding to the time constant τ2OC . We obtain:
   2  
1 1 1
∆4 − OC = τ1OC τ2SC − OC + τ1OC + τ2OC


− OC + 1
τ2 τ2 τ2
τ1OC τ2SC τ1OC τ2OC
= − − +1
τ2OC τ2OC τ2OC τ2OC
τ1OC τ2SC τ1OC
= − −1+1
τ2OC τ2OC τ2OC
τ OC
 SC 
τ2
= 1OC − 1 <0
τ2 τ2OC
| {z } | {z }
>0 <0

We now start from the right extremum τ1SC and evaluate ∆5 (s) at the pole corresponding to
the time constant τ1SC . We obtain:

1 2
     
1 1
τ1SC τ2OC OC OC


∆5 − = − SC + τ1 + τ2 − SC + 1
τ1SC τ1 τ1
τ1SC τ2OC τ1OC τ2OC
= − − +1
τ1SC τ1SC τ1SC τ1SC
τ1OC
=1− <0
τ1SC
| {z }
>1

So, since the poles corresponding to the outermost time constants τ2OC and τ1SC correspond to
negative values of an upward oriented parabola, we can conclude, by continuity of the parabola,
1
that the roots of its associated second order equation are one at the right of − OC and one at
τ2
1
the left of − SC .
τ1
Now, since what we have found is the angular frequency ( 2πf ) of the poles we may say that
the pole frequencies are external (one to the right and one to the left) to the interval defined
by:    
1 1 1 1 1 1 1 1
min − SC , − OC , − SC , − OC ÷ max − SC , − OC , − SC , − OC
τ1 τ1 τ2 τ2 τ1 τ1 τ2 τ2

14
Remember that this holds just if we have:
• τ1SC < τ1OC
• τ2SC < τ2OC
• Either τ1SC , τ1OC  τ2SC , τ2OC or τ1SC , τ1OC  τ2SC , τ2OC
Since the two poles will lie to the opposite sides of such interval, we will have that the two poles
will be far away from each other, that is:
either fp1  fp2 or fp1  fp2
Time constant method

This method can be generalized to circuits with more than two reactive elements, even though
expressions will look much more complicated!

Let’s suppose all the previous hypotheses hold true and consider one of the forms of the char-
acteristic polynomial:
∆5 (s) = τ1SC τ2OC s2 + τ1OC + τ2OC s + 1

Moreover, suppose that τ1SC , τ1OC  τ2SC , τ2OC (relabel the capacitors if it isn’t true).

Now, we can divide the polynomial by τ1SC τ2OC without affecting its roots (the system poles)
to get:
τ OC + τ2OC 1
∆6 (s) = s2 + 1 SC OC s + SC OC
τ1 τ2 τ1 τ2
We have already proven that the characteristic polynomial has two real negative roots, so it
will admit the factorization:
∆6 (s) = (s − sp1 ) · (s − sp2 )
If we carry out all the products we get:
∆6 (s) = s2 + (−sp1 − sp2 )s + sp1 sp2
By comparison of the coefficients we get the following two relationships:
τ1OC + τ2OC
−sp1 − sp2 =
τ1SC τ2OC
1
sp1 sp2 = SC OC
τ1 τ2
Because of our hypotheses we have that the two poles will be far apart, so we have that one is
much smaller in modulus than the other one. If we denote such pole with sp1 we have:
|sp1 |  |sp2 |
Which we can combine with the previous hypothesis:
τ1SC , τ1OC  τ2SC , τ2OC
We can manipulate the first equation to get the following:
τ1OC + τ2OC τ2OC 1
−sp1 − sp2 ≈ −sp2 = SC OC
≈ SC OC
= SC
τ1 τ2 τ1 τ2 τ1
So:
1
sp2 ≈ −
τ1SC

15
From the second equation we can get, by substituting this last result into it:
1
sp1 sp2 ≈ −sp2
τ2OC

Which gives:
1
sp1 ≈ −
τ2OC
So, the two poles can be approximated with the extrema of the interval defined by the two
outermost of the four poles!
τ SC
It is possible to show that the error in this approximation is of the order of 1OC .
τ2
Determination of transmission zeroes

Transmission zeroes are complex values of the variable s that make the transfer function of the
system equal to zero.

Some of them can be easily found by observing the shape of the transfer function and looking
for the values of the impedances that make it zero.

Once those are found, finding the gains in the different bands is a straightforward task!

Zero-Pole-Gains dependence

Consider a transfer function having a Bode diagram like the following:

|H(jf)|dB

HLF
-2
0
db
/d
ec

HLF

0dB
fpi fzi
log10(f/f0)

in which we can isolate a portion of the magnitude Bode diagram that looks like the Bode
diagram of a single time constant circuit.

Notice that we can have either of the two situations:

• The zero precedes the pole

• The pole precedes the zero

16
Let’s consider now the general shape of a transfer function with multiple zeroes and multiple
poles:
(s + sz,1 )(s + sz,2 ) · · · (s + sz,i ) · · · (s + sz,m−1 )(s + sz,m )
H(s) = k sp
(s + sp,1 )(s + sp,2 ) · · · (s + sp,i ) · · · (s + sp,n−1 )(s + sp,n )
We may distinguish three terms:

a. Containing the poles and zeroes before the zero and pole we are observing (”LF part”)

b. Containing the zero and pole we are observing

c. Containing the poles and zeroes after the zero and pole we are observing (”HF part”)

That is:
(s + sz,f +1 ) · · · (s + sz,i−1 ) (s + sz,i ) (s + sz,i+1 ) · · · (s + sz,m )
H(s) = k sf
(s + sp,1 ) · · · (s + sp,i−1 ) (s + sp,i ) (s + sp,i+1 ) · · · (s + sp,n )
| {z } | {z } | {z }
(a) (b) (c)

We can now rewrite:

• the ”low-frequency part” in high-frequency canonical form

• the ”high-frequency part” in low-frequency canonical form

Therefore, we obtain:
s s
(1 + ) · · · (1 + )
(s + s z,f +1 ) · · · (s + sz,i−1 ) (s + s z,i ) sz,i+1 sz,m
H(s) = k sf w s s
(s + sp,1 ) · · · (s + sp,i−1 ) (s + sp,i ) (1 + ) · · · (1 + )
| {z } | {z } sp,i+1 sp,n
(a) (b) | {z }
(c)

Since we have the same number of poles and zeroes at the left of fpi and fzi we have that:

lim H(a) (j2πf ) = 1

f →∞

And the limit of the part to the right of fpi and fzi as f → 0 is:

lim H(c) (j2πf ) = k

f →∞

If we focus now on the frequencies around the pole fpi and the zero fzi we have that:

f  fz,i+1 , · · · , fz,m

f  fz,f +1 , · · · , fz,i−1
So, we can approximate:

• Part (a) with limf →∞ H(a) (j2πf ) = 1

• Part (c) with limf →∞ H(c) (j2πf ) = k

near the interval defined by the pole fpi and the zero fzi .

17
Therefore, we are left with:
(s + sz,i )
H(s) ≈ k w
(s + sp,i )
Since the one at the right is a high-frequency canonical form we can say that the product kw is
about HHF :
HHF ≈ kw
Therefore:
(s + sz,i )
H(s) ≈ HHF
(s + sp,i )
(near the pole fpi and the zero fzi )

We can now rewrite it in LF canonical form:

s
(1 + )
sz,i
H(s) ≈ HLF s
(1 + )
sp,i
Notice that we can derive the LF canonical form from the HF one by factoring out sz,i and sp,i :
s s
(1 + ) (1 + )
(s + sz,i ) sz,i sz,i sz,i
H(s) ≈ HHF ≈ HHF s ≈ HLF s
(s + sp,i ) sp,i
| {z } (1 + sp,i ) (1 +
sp,i
)
HLF

By comparison between the two we can obtain the relationship:

sz,i
HLF = HHF
sp,i
remembering that sz,i = 2πfz,i and sp,i = 2πfp,i we can rewrite it as:
2πfz,i fz,i
HLF ≈ HHF ≈ HHF
2πfp,i fp,i
Which can be reordered in the more useful form:
HLF fz,i
≈
HHF fp,i

Dominant pole approximation

Sometimes, we can approximate the behavior at the extrema of the band of an amplifier with
dominant-pole approximations.

This is a very useful approximation when the poles that start and end the band are distant
enough from the ones that lie outside the band.

We can distinguish two different kinds of dominant-pole approximations:

• the high-frequency dominant pole approximation

• the low-frequency dominant pole approximation

18
Let’s start from the HIGH-FREQUENCY dominant pole approximation.

When the Bode diagram at the high-frequency extremum of the amplifying band resembles
enough the one of a single-pole system with zero at infinity, we can use such approximation.

For example, look at the following Bode diagram:

|hfe(jf)|dB=|β(jf)|dB

HLF

0dB fp fT
log10(f/f0)

At frequencies about the pole frequency fp we may effectively approximate the transfer function
with a system having low-frequency gain HLF and a single pole fp (with HLF the frequency
right before the end of the band).

The resulting Bode diagram is:

|H(jf)|dB

HLF
-2
0
db
/d
ec

0dB
fp fT
log10(f/f0)

The transfer function for this kind of Bode diagram is:

1
H(s) ≈ HLF   with sp = 2πfp
s
1+
sp

19
We can now introduce the concept of upper cutoff frequency, which is the frequency after
the end of the band at which the gain is unitary (0 dB).

Once we identify the band of the amplifier (if there is more than one), we can define the band-
width as the range of frequencies at which the gain is higher than 3 dB less than the in-band
gain ( > Hband,dB − 3 dB ).

In a dominant high-frequency pole approximation the bandwidth goes all the way from 0 Hz to
the pole frequency fp (remember that the non-asymptotic bode diagram is 3dB lower at the
pole frequency).

Therefore, we may identify the bandwidth in a high-frequency dominant pole approximation

with the frequency of the dominant pole ( BW = fp ).

We would now like to find a good enough approximation for the upper cutoff frequency. In
order to do this, we write down the definition of cutoff frequency:
 
 
 
 1 
|H(jfT )| ≈ HLF
   =1
jfT 
 1+
 f p


Which can be rewritten with the definition of modulus as:

1
|H(jfT )| ≈ |HLF | s  2 = 1
fT
1+
fp
If the low-frequency gain is high enough to allow a good enough separation between the pole
and cutoff frequency, that is: fp  fT (one decade will do it), we may approximate it in the
following way:
1 1
|H(jfT )| ≈ |HLF | s  2 ≈ |HLF | s 2 = 1
fT fT
1+
fp fp
Therefore:
1
|H(jfT )| ≈ |HLF |   =1
fT
fp
We can conclude that:
1
|HLF |   ≈1
fT
fp
fT ≈ |HLF | · fp
If we consider the low-frequency gain as positive, we get:
fT ≈ HLF · fp
Which shows that, in a dominant high-frequency pole approximation, the upper cutoff frequency
is the gain-bandwidth product, which is used as an alternative name for the upper cutoff
frequency.

20
We can now study the LOW-FREQUENCY dominant pole approximation.

When the Bode diagram at the low-frequency extremum of the amplifying band resembles
enough the one of a single-pole system with zero at the origin, we can use such approximation.

For example, look at the following Bode diagram:

|hfe(jf)|dB=|β(jf)|dB

HHF

log10(f/f0)
fT fp 0dB

At frequencies about the pole frequency fp we may effectively approximate the transfer function
with a system having high-frequency gain HHF and a single pole fp (with HHF the frequency
right after the beginning of the band).

The resulting Bode diagram is:

|H(jf)|dB

HHF
ce
/d
db
0
+2

0dB
fT fp
log10(f/f0)

The transfer function for this kind of Bode diagram is:

s
H(s) ≈ HHF with sp = 2πfp
(s + sp )
We can now introduce the concept of lower cutoff frequency, which is the frequency before
the beginning of the band at which the gain is unitary (0 dB).

21
In a dominant low-frequency pole approximation the bandwidth goes all the way from the pole
frequency fp to infinity (remember that the non-asymptotic bode diagram is 3dB lower than
the asymptotic one at the pole frequency).

Therefore, we may identify the bandwidth in a low-frequency dominant pole approximation

with the frequency of the dominant pole ( BW = fp ).

We would now like to find a good enough approximation for the lower cutoff frequency. In order
to do this, we write down the definition of cutoff frequency:
 
 jfT 
|H(jfT )| ≈ HHF
 =1
(jfT + fp ) 

Which can be rewritten with the definition of modulus as:

fT
|H(jfT )| ≈ |HHF | q =1
fT2 + fp2

We can now bring the cutoff frequency inside the square root to get:
1
|H(jfT )| ≈ |HHF | s  2 = 1
fp
1+
fT

We can now rearrange it to obtain an estimate for the lower cutoff frequency.
Moreover, let’s consider HHF as positive. We get:
 2
2 fp
HHF =1+
fT
 2
fp 2
= HHF −1
fT
fp
q
= 2 −1
HHF
fT
fp
fT = q
2 −1
HHF

Which allows us to compute the lower cutoff frequency in a dominant low-frequency pole ap-
proximation.

In GENERAL we may extend these two approximations also to systems with multiple dominant-
pole approximation (dominant pole approximation with more than one pole). The algebraic
steps and reasoning are similar to the ones we have for the single pole.

22