Colorado School of Mines CHEN403

Higher Order Systems

Higher Order Systems ......................................................................................................................................... 1

Higher Order Systems ......................................................................................................................................... 1
Multicapacity Processes in Series................................................................................................................... 1
Two Non-Interacting 1st Order Systems.................................................................................................. 2
Two Interacting 1st Order Systems............................................................................................................ 2
N Identical Processes in Series ........................................................................................................................ 6
Calculation of the Step Disturbance Coefficients ................................................................................. 9
Processes With Dead Time............................................................................................................................. 11
Approximate Reduced Order Transfer Functions FOPDT .............................................................. 15
FOPDT Example ............................................................................................................................................. 16
Processes With Inverse Response............................................................................................................... 17
Inverse Response from the Sum of Two 1st Order Transfer Functions.................................... 18
Inverse Response from the Difference Between Two 1st Order Transfer Functions.......... 18
Inverse Response from the Difference Between a 2nd & 1st Order Transfer Functions ..... 19
Inverse Response from the Difference Between Two 2nd Order Transfer Functions......... 21

Higher Order Systems

Three types of common systems with higher than 2nd order dynamics are:

Multicapacity processes in series.

Processes with dead time.
Processes with inverse response.

Multicapacity Processes in Series

When we introduced transfer functions, we talked about the overall transfer function for a
series of non-interacting processes.

y1 ( s ) y2 ( s ) y N 1 ( s ) yN ( s )
f (s) G1 G2 GN

If there are a series of transfer functions, then:

yN ( s ) = GN ( s ) yN 1 ( s )
= GN ( s ) GN 1 ( s ) yN 2 ( s )
= GN ( s ) GN 1 ( s ) G1 ( s ) f ( s )

Higher Order Systems -1- December 21, 2008

Colorado School of Mines CHEN403

yN ( s ) N
= Gi ( s )
f (s) i =1

Two Non-Interacting 1st Order Systems

For a series of two non-interacting 1st order systems with different time constants 1 & 2 ,
then the overall transfer function is 2nd order:

y2 ( s ) K1 K2
= .
f (s) 1 s + 1 2 s + 1

Since the denominator of the transfer function can be factored into linear terms each with
real coefficients, the overall transfer function will give an overdamped response to a
disturbance. Also, since the poles of the transfer function, 1/ 1 & 1/ 2 , are real and
negative, then the overall transfer function will give a stable overdamped response to a
disturbance.

We could also analyze the expected response knowing something about 2nd order systems.
Putting the overall transfer function into the standard form for a 2nd order system gives:

y2 ( s ) K1 K2 K 1K 2 K p*
= =
f ( s ) 1 s + 1 2 s + 1 1 2 s + ( 1 + 2 ) s + 1 *2
p s + 2 p p s + 1
* *

where:

K p* = K 1K 2
*2
p = 1 2 *p = 1 2
1 + 2 1
2 *p *p = 1 + 2 *p = = 1 + 2
2 1 2 2 2 1

For positive 1 & 2 that are not equal, then the damping factor *p will always be greater
than one (meaning an overdamped system). In the event that 1 = 2 , then *p = 1 and the
system will be critically damped. There are no combinations of 1 & 2 so that the system
would be underdamped this makes sense since an underdamped system requires linear
factors with complex coefficients.

Two Interacting 1st Order Systems

Higher Order Systems -2- December 21, 2008

Colorado School of Mines CHEN403

But what if the processes interact? Can we get an underdamped system then? The answer
is it depends how they interact. Lets look at a simple liquid flow system where the
interaction is due to back-pressure from the second tank. Assume the liquid density is
constant. Further assume that the tanks are open to the atmosphere, have constant cross-
sectional areas A1 and A2 , and that the flow through each valve is proportional to the liquid
level providing the hydrostatic pressure (i.e., Fi = Ci ( h) ).
F 0,

h 1, A 1, h 2, A 2,

F 1, F 2,

The material balance for Tank 1 leads to:

dh1
A1 = F0 C1 ( h1 h2 )
dt
dh
A1 1 + C1h1 = F0 + C1h2
dt
A1 dh1 1
+ h1 = F0 + h2
C1 dt C1

Since a linear system, can immediately use deviation variables:

A1 dh1 1
+ h1 = F0 + h2
C1 dt C1
A1 1
s + 1 h1 = F0 + h2
C1 C1
K1 1 A 1
h1 = F0 + h2 where 1 = 1 and K 1 = .
1 s + 1 1 s + 1 C1 C1

The material balance for Tank 2 leads to:

dh2
A2 = C1 ( h1 h2 ) C2h2
dt
dh
A2 2 + ( C1 + C2 ) h2 = C1h1
dt

Higher Order Systems -3- December 21, 2008

Colorado School of Mines CHEN403

Again, since a linear system, we can immediately use deviation variables:

dh2
A2 + ( C1 + C2 ) h2 = C1h1
dt
( A2s + C1 + C2 ) h2 = C1h1

We would normally define the 2 time constant by dividing by the sum of the valve
coefficients, C1 + C2 . However, instead, lets define the time constant in a manner similar to
that for the 1st tank, i.e., 2 = A2 / C2 . This leads to:

( 2C2s + C1 + C2 ) h2 = C1h1

C1 C1
2 s + 1 + h2 = h1
C2 C2

K2 C
so: h2 = h1 where K 2 = 1 .
2 s + 1 + K 2 C2

Now, the transfer function between F0' and h2' will be:

K2 K1 1
h2 = F0 + h2
2 s + 1 + K 2 1 s + 1 1 s + 1

K2 1 ' K1 K2
1 h2 = F0'
2 s + 1 + K 2 1 s + 1 1 s + 1 2 s + 1 + K 2

( 1 s + 1)( 2 s + 1 + K 2 ) K 2 h ' = K 1K 2
F'
( 1 s + 1)( 2 s + 1 + K 2 ) 2
( 1 s + 1)( 2 s + 1 + K 2 ) 0
K 1K 2
h2' = F0'
( 1 s + 1)( 2 s + 1 + K 2 ) K 2
K 1K 2
h2' = F0'
1 2 s + 1 (1 + K 2 ) + 2 s + 1
2

Notice the extra K 2 term in the denominator because of the interaction of the tanks. The
coefficients in the denominator are all positive & real, so the resulting response to a
disturbance will be stable. But will it be overdamped, critically damped, or underdamped?

Higher Order Systems -4- December 21, 2008

Colorado School of Mines CHEN403

To answer this, lets look at the damping factor. In the standard form for a 2nd order system:

*2
p = 1 2 *p = 1 2
1 (1 + K 2 ) + 2
2 *p *p = 1 (1 + K 2 ) + 2 *p =
2 1 2
1 + 2 1K 2
= +
2 1 2 2 1 2
1 K 1
= 1 + 2 + 2
2 2 1 2 2

Note that the first term for *p will always be greater or equal to one (only equal to one if
1 = 2 ). The second term will always be positive. Together, this shows that the overall
damping factor will always be greater than one this type of system will always be
overdamped.

We can do an analysis of the poles, too. The poles are given by:

1 (1 + K 2 ) + 2 1 (1 + K 2 ) + 2 41 2
2

p1 , p2 =
21 2
1 (1 + K 2 ) + 2 12 (1 + K 2 ) + 21 2 (1 + K 2 ) + 22 41 2
2

p1 , p2 =
21 2

p1 , p2 =
(
1 (1 + K 2 ) + 2 12 21 2 + 22 + 12K 22 + 212K 2 + 2K 21 2 )
21 2
1 (1 + K 2 ) + 2 ( 1 2 ) + ( 12K 22 + 212K 2 + 2K 212 )
2

p1 , p2 =
21 2

Notice that the resulting system is still 2nd order, stable, and overdamped. As an example,
what happens to the poles when the tanks are identical? Here, 1 = 2 = and K 2 = 1 .
Then:

(1 + 1) + ( ) + ( 2 + 22 + 22 )
2

p1 , p2 =
22
3 52 1 3 5 0.382 2.618
p1 , p2 = = = ,
2 2
2

Higher Order Systems -5- December 21, 2008

Colorado School of Mines CHEN403

N Identical Processes in Series

When all of the processes in series are identical then the overall transfer function will
become:

yN ( s ) N
= G ( s ) = GN .
f (s) i =1

This could lead to a very simple solution depending upon the forcing function. For example,
if the processes are all first order and the forcing function is an impulse,
f = C ( t ) f = C , then the output response will be:

N
K
yN = C
s + 1
K
=
N
1 ( N 1 )! C
( N 1)! 1 N
s +

N
N 1 t /
K t e
yN ( t ) = C .
( N 1 )!

The following figure shows the response curves for an impulse disturbance to a series of
identical 1st order processes (with C = 1 , K = 1 & = 2 ). Notice that the response from the
1st process is an exponential & the responses from the rest of the processes look like spread
pulses. The peak or maximum value of the disturbance occurs later for each additional
process & (in this case) the peak is a little smaller for each additional process. In fact, we
can determine when the peak will occur it will be at that time t m where t m > 0 and the
first time derivative of the response curve will be zero:

N
dyN C K t N 1e t /
= ( N 1 ) t N 2 t /
e
dt ( N 1)!

N
C K t mN 1e tm /
( N 1)tm e
N 2 t m /
=0
( N 1)!
N
C K t m N 2 t m /
( N 1 ) t m e =0
( N 1 )!
t
( N 1) m = 0

Higher Order Systems -6- December 21, 2008

Colorado School of Mines CHEN403

tm = ( N 1) .

The corresponding value of the peak will be:

N 1 ( N 1 )
K ( N 1) e
( N 1 )
CK N ( N 1) e
N N 1

yN ( t m ) = C = .
( N 1)! ( N 1)!

0.50

0.45

0.40

0.35
Response Value

0.30
N= 1
0.25

0.20
N= 2

0.15 N= 3
N= 4
0.10
N= 5
0.05

0.00
0 2 4 6 8 10 12 14 16 18 20
Time (min)

Impulse Disturbance to Identical 1st Order Processes in Series

A more complicated forcing function will lead to more complicated Laplace functions that
need to be inverted. However, with some ingenuity, it may be fairly easy to work with these
more complicated functions. For example, if the forcing function is a step function,
f = A S ( t ) f = A / s , then:

N
K A
yN =
s + 1 s

and the partial fraction expansion must have all of the powers of the s + 1 term up to the
N 1 power:

Higher Order Systems -7- December 21, 2008

Colorado School of Mines CHEN403

C N
Ci N
Ci
yN = AK N 0 + y = AK N
C + e t /
t i 1

s i =1 ( s + 1) i =1 ( i 1 )!
i N 0 i

where the C i coefficients must be determined to be consistent with the original overall
transfer function. Ones first thought is that these coefficients will be very difficult &
tedious to determine. On the contrary, using the Heaviside method we can find:

1 i =0
Ci =
i = 1,2, , N

so:

1 N N
1
yN = AK N y = AK N
1 e t /
t i 1
.
s i =1 ( s + 1)
i N
i =1
i 1
( i 1)!

The following figure shows the response curves for a step disturbance to a series of
identical 1st order processes (with A = 1 , K = 1 & = 2 ). Notice that the higher order
response curves look very much like the 2nd order response (i.e., that the initial response
has a zero slope) but the time period in which there is very little response is much longer.

Higher Order Systems -8- December 21, 2008

Colorado School of Mines CHEN403

1.0

0.9
N= 1
0.8
N= 2
0.7
N= 3
Response Value

0.6
N= 4
0.5
N= 5
0.4

0.3

0.2

0.1

0.0
0 2 4 6 8 10 12 14 16 18 20
Time (min)

Step Disturbance to Identical 1st Order Processes in Series

Calculation of the Step Disturbance Coefficients

The transfer function to be split apart should be put into the following working
form:

1 1 C0 N Bi
= +
s ( s + 1) N
s i =1 1 i
s +

where the Bi coefficients are related to the C i coefficients by:

C i = Bi i .

The first coefficient, C0 , is very easily calculated with the Heaviside method:

1 1
C0 = lim s =1.
s ( s + 1)
s 0 N

The coefficient for the highest order term is also very easily calculated:

Higher Order Systems -9- December 21, 2008

Colorado School of Mines CHEN403

1 N 1 1 1 1
BN = lim s + = lim N = N C N = .
s 1/
s ( s + 1 )
N s 1/
s

The other coefficients are calculated using the formula:

1 N
s +
1 d k 1
BN + 1 k = lim k = 2,3, , N
s 1/ ( k 1 )! ds k 1
s ( s + 1)
N

1 1 d k 1 1
BN +1k = lim N k 1
s 1/ ( k 1 ) ! ds s

The derivatives of 1/ s have a fairly simple form:

dn 1 1
= ( 1) n! n+1
n
n
ds s s

So:

1 1 ( k 1) 1
BN +1k = lim N ( 1 ) ( k 1)! k
s 1/ ( k 1 )!
s
( 1 )( k 1)
= lim N k
s 1/
s
.
( k 1 )
= ( 1 ) ( ) ( )
k N

( k 1 )
= ( 1 ) ( 1 ) k N
k

(2k 1) k N
= ( 1 )
= k N
( 2 k 1 )
Note that since 2k 1 will always be odd then ( 1) will always be -1. Finally:

C N +1k = BN +1 k N +1k = k N N +1k = .

So this shows that:

Higher Order Systems - 10 - December 21, 2008

Colorado School of Mines CHEN403

1 1 N
= .
s ( s + 1 ) s i =1 ( s + 1)i
N

Processes With Dead Time

So far we have assumed that whenever a change occurs, it reaches the measurement device
instantaneously. However, there can be dead time between the process & the
measurement device and we get transportation lag. One can picture the situation as being
similar to plug flow through a pipe the material does not change in the pipe, but there is
a time difference between when it enters and when it reappears at the other end.

Since:

y (t + td ) = f (t ) y (t ) = f (t td )

then the transfer function is immediately:

y ( s ) = e td s f ( s )

We can develop this Laplace formula be thinking of dead time as an infinite number
of identical non-interacting processes in series. Image a series of N tanks, each
with the same volume, with a total volume of V . Into the first tank we make a
change to the composition of component A. The component balance around the first
tank will be:

dC A1
V1 F0C A1
= F0C A0
dt

and the corresponding transfer function will be:

C A1 1 1
= = .

C A0 V1 V
s +1 s +1
F0 NF0

Out of the Nth process:

N
1 N N
C '
C A' ,0 Vs td s
A ,N
= V = 1 + = 1 +
C ' s + 1 C A' , N NF0 N
NF0
A ,0

Higher Order Systems - 11 - December 21, 2008

Colorado School of Mines CHEN403

where t d V / F0 . It is known from advanced calculus that:

n/ r n
r r
lim 1 + = e lim 1 + = e r
n
n n
n

so as N then:

N
C A' ,0 t s
= 1 + d exp ( t d s )
C A' ,N N

and:
C A' , N
G(s) = exp ( t d s ) .
C A' ,0

It is not always convenient (or even possible) to work with the dead-time term as it is
defined. Remember that to invert the Laplace functions in the presence of a time delay
term, all of the time delay terms must be factored out of & away from the rest of the Laplace
expression. For example, the time delay term in the following expression can be factored
out & the remain expression inverted:

Ke s K K K t
y = = e s L1 = exp
s + 1 s + 1 s + 1
.
1 K s K t
L e = exp S (t )
s + 1

However, the following expression comes about when we consider systems with a feedback
of information:

Ke s
s + 1 K
y = s
= s
e s .
Ke s + 1 + K c Ke
1 + Kc
s + 1

Weve factored the e s time delay term out of the numerator but it is still in the
denominator & it cannot be removed form here. We cannot use the standard methods to
invert this Laplace function.

When we have a situation like that above we will often approximate the exponential time
delay term with a ratio of polynomials. Once we do this the exponential term is gone & we
can use standard methods to break apart the transfer function into partial fractions and

Higher Order Systems - 12 - December 21, 2008

Colorado School of Mines CHEN403

then invert. Approximate methods for dealing with time delays are based upon
approximating the exponential term with some type of polynomial. Remember that the
Taylor series expansion formulas is:

1 dn f
f (x) = ( x x0 )
n
n
i =0 n! dx x = x0

so the exponential function expanded around x = 0 is:

( x ) + ( x ) +
2 3

n
e +s
= x n = 1 + x +
i =0 n! 2 6

and for the actual time delay term:

( x ) ( x ) +
2 3

n n
e s
=
i =0
( n
1)n!
x = 1 x +
2 6

One could just truncate the polynomial after a prescribed number of terms. So we could use
linear forms of either:

e s 1 x

or:

1 1
e s = +s
.
e 1 + x

And, depending upon how the approximation will be used, this may be perfectly acceptable.
However, it is more typical that the exponential term is approximated with a ratio of
polynomials. The most typical set of functions used for the time delay term are the Pad
approximations. There are approximations of various order. The most common one used
(especially for hand calculations) is the 1st order Pad approximation:

1
1 s
e s 2 = 2 s
1 + s 2 + s
1
2

Higher order approximations are:

Higher Order Systems - 13 - December 21, 2008

Colorado School of Mines CHEN403

1 1
( s ) + ( s ) 12 6 ( s ) + ( s )2
2
1
e s 2 12 =
1 + ( s ) + ( s ) 12 + 6 ( s ) + ( s )
2
1 1 2

2 12
1 1 1
1 ( s ) + ( s ) ( s ) 120 60 ( s ) + 12( s )2 ( s )3
2 3

e s 2 10 120 =
( s ) 120 + 60 ( s ) + 12( s ) + ( s )
2 3
1 1 1
1 + ( s ) + ( s ) +
2 3

2 10 120

and so on.

Pad approximations match terms to the exponential functions Taylor series expansion
one order higher than the functions used in the Pad approximation. The 1st order
approximation has a Taylor series expansion:

1
1 s ( s ) ( s )
2 3

e s 2 = 1 ( s ) + +
1 2 4
1 + s
2

and so it actually matches terms up to the 2nd power, the 2nd order approximation has a
Taylor series expansion:

1 1
( s ) + ( s )
2
1 ( s ) ( s ) ( s )
2 3 4

e s 2 12 == 1 ( s ) + +
1 1 2 6 18
1 + ( s ) + ( s )
2

2 12

and matches terms up to the 3rd power, and so forth. You should get higher order accuracy
using lower order polynomials.

The response curves using Pad approximations tend to be least accurate at short times
(especially on the order of the time delay ) and become more accurate with higher order
approximations. However, the difficulty of working with the expressions becomes much
greater with higher order approximations. For example, using a 1st order approximation,
the example transfer function is approximated as:

2 s
Ke s K
2 + s K ( 2 s )
y = = .
s + 1 + K c Ke s 2 s ( 2 + s )( s + 1) + K c K ( 2 s )
s + 1 + K c K
2 + s

Higher Order Systems - 14 - December 21, 2008

Colorado School of Mines CHEN403

This is a transfer function for a 2nd order process and the inverse can be determined using
standard forms. Using a 2nd order approximation, the example transfer function is:

12 6 ( s ) + ( s )
2

K
12 + 6 ( s ) + ( s )
2
Ke s
y =
s + 1 + K c Ke s 12 6 ( s ) + ( s )
2

s + 1 + K c K
12 + 6 ( s ) + ( s ) .
2

K 12 6 ( s ) + ( s )
2

=
12 + 6 ( s ) + ( s ) ( s + 1 ) + K c K 12 6 ( s ) + ( s )2
2

This is a transfer function for a 3rd order process and it must be split apart using partial
fractions into the sum of a 1st order response & a 2nd order response before standard forms
can be used to determine the inverse.

Approximate Reduced Order Transfer Functions FOPDT

Even though weve spent a lot of time developing transfer functions of high order, it may be
convenient to work with an approximate low order transfer function that will give
essentially the same type of dynamic response. The slow response at small times (due to
the smallest time constants) can be approximated with a time delay term. Keeping only one
time constant gives rise to an FOPDT transfer function (First Order Process with Time
Delay).

One way to reduce the order of a transfer function is approximate the small time constants
with corresponding time delay terms. For example, if 1 > 2 > 3 > 4 is the largest time
constant, the following 4th order transfer function can be approximated with the following
FOPDT model:

Kp Kp 1 1 1
=
( 1 s + 1)( 2s + 1)( 3 s + 1)( 4 s + 1) 1 s + 1 2 s + 1 3 s + 1 4 s + 1
Kp 1 1 1

1 s + 1 2 s + 1 3 s + 1 4 s + 1
Kp
= e 2s e 3s e 4 s
1 s + 1
( 2 +3 +4 ) s
K pe
=
1 s + 1

Higher Order Systems - 15 - December 21, 2008

Colorado School of Mines CHEN403

An approximate SOPDT model (Second Order Process with Dead Time) would be:

( 3 +4 ) s
Kp K pe
.
( 1 s + 1)( 2s + 1)( 3 s + 1)( 4 s + 1) ( 1 s + 1)( 2s + 1)

Skogestad has proposed a related method. He suggests that the largest neglected time
constant should be split between the smallest retained time constant and the time delay.
So, Skogestads FOPDT model would be:

( 2 /2+3 +4 ) s
Kp K pe

( 1 s + 1)( 2s + 1)( 3s + 1)( 4 s + 1) ( 1 + 2 /2) s + 1

and the SOPDT model would be:

( 3 /2+4 ) s
Kp K pe
.
( 1 s + 1)( 2s + 1)( 3 s + 1)( 4 s + 1) ( 1 s + 1) ( 2 + 3 /2) s + 1

These methods are best applied when you already have the coefficients for the high order
transfer functions. If you must first derive the coefficients of the high order model from
empirical data, it is better to derive the coefficients of the FOPDT or SOPDT model directly.

FOPDT Example

The 4th order transfer function:

1
G(s) =
(10s + 1)( s + 1)(0.1s + 1)(0.05s + 1)

can be approximated as the following FOPDT model using 1st order Taylor series
expansions:

e 1.15s
G(s)
10s + 1

or the following FOPDT model using the Skogestad method:

e 0.65 s
G(s) .
10.5s + 1

Higher Order Systems - 16 - December 21, 2008

Colorado School of Mines CHEN403

1.0

0.9

0.8

0.7

0.6

0.5

0.4
4th Order Transfer Function Solution
0.3
Taylor Series FOPDT

0.2 Skogestad FOPDT

0.1

0.0
0 2 4 6 8 10 12 14 16 18 20
Time (min)

The figure above compares the response curves for a step change forcing function for the
original transfer function & the two FOPDT models. Notice that both FOPDT models match
the response from the original transfer function once past t 8 min .

Processes With Inverse Response

When the overall transfer function is found as the difference of individual transfer
functions, then the initial response in a system may be in the opposite direction than where
it will end up. One example is adding cold feedwater to a drum reboiler to control the
liquid level. The 1st bit of cold feedwater will decrease the temperature, causing the volume
of the entrained vapor to decrease. The may be the overwhelming effect at first, causing
the overall liquid level to drop. The liquid level will then increase again once the heat is
added.

The overall transfer function for a process that shows inverse response will look something
like:

G ( s ) = G1 ( s ) G2 ( s )

where G2 ( s ) dominates at early times and G1 ( s ) dominates at long times. Expressing the
overall transfer function as a ratio of two polynomials:

Higher Order Systems - 17 - December 21, 2008

Colorado School of Mines CHEN403

Nm ( s )bm s m + bm1 s m1 + + b1 s + b0
G(s) = =
Dn ( s ) an s n + an1 s n1 + + a1 s + a0

then there will be an inverse response when there is at least one positive zero in the
transfer function (i.e., there is at least one positive root to the polynomial in the numerator
Nm ( s ) ).
Lets look at the necessary conditions for inverse response when combining several types
of common transfer functions.

Inverse Response from the Sum of Two 1st Order Transfer Functions

The transfer function formed from the sum of two 1st order transfer functions is:

G(s) =
K1
+
K2 ( K + K 21 ) s + ( K 1 + K 2 )
= 1 2
1 s + 1 2 s + 1 ( 1 s + 1)( 2s + 1)

and the root of the numerator will be:

K1 + K2 K1 + K2
s= implies an inverse response for < 0.
K 1 2 + K 2 1 K 1 2 + K 2 1

If we only consider positive gains & time constants then there will not be an inverse
response the combination of positive terms cannot give a negative number.

Inverse Response from the Difference Between Two 1st Order Transfer Functions

The transfer function formed from the difference of two 1st order transfer functions is:

G(s) =
K1

K2 ( K K 21 ) s + ( K 1 K 2 )
= 1 2
1 s + 1 2 s + 1 ( 1 s + 1)( 2s + 1)

and the root of the numerator will be:

K1 K2 K1 K2
s= implies an inverse response for < 0.
K 1 2 K 2 1 K 1 2 K 2 1

Inverse response will occur if the numerator & denominator are of opposite sign so that
overall the fraction is negative. So, there are two possibilities. If:

Higher Order Systems - 18 - December 21, 2008

Colorado School of Mines CHEN403

K1
K1 K 2 > 0 K1 > K 2 >1
K2

then:

1 K 1
K 1 2 K 2 1 < 0 K 2 1 > K 1 2 > >1.
2 K 2

However, we can also have:

K1
K1 K 2 < 0 K1 < K 2 <1
K2

then:

1 K 1
K 1 2 K 2 1 > 0 K 21 < K 1 2 < <1.
2 K 2

This shows that just having the difference of two 1st order transfer functions is not a
sufficient condition for an inverse response.

We can also look at many more combinations of transfer functions to determine

when we might get an inverse response.

Inverse Response from the Difference Between a 2nd & 1st Order Transfer
Functions

Now, lets look at when the overall transfer function is the difference between 2nd
order & 1st order transfer functions. Then:

G(s) = 2 2
K1

K2
=
( )
K 212 s 2 + ( K 1 2 2K 21 1 ) s + ( K 1 K 2 )
1 s + 21 1 s + 1 2 s + 1 ( 2s + 1) 12 s2 + 211 s + 1( )
The roots to the numerator are:

r1 , r2 =
( K 1 2 2K 21 1 ) ( K 12 2K 211 )
2
( )
4 K 212 ( K 1 K 2 )
(
2 K 2 12 )
( K 12 2K 211 ) + 4K 212 ( K 1 K 2 )
2
K 1 2 2K 21 1
r1 , r2 =
2K 2 12

Higher Order Systems - 19 - December 21, 2008

Colorado School of Mines CHEN403

The largest root will be with the positive sign, so to guarantee an inverse response
we only need to restrict this root:

( K 12 2K 211 ) + 4K 212 ( K 1 K 2 ) > 0

2
K 1 2 2K 21 1 +

( K 12 2K 211 ) + 4K 212 ( K 1 K 2 ) > ( K 12 2K 21 1 )

2

Note that the 1st term under the radical is:

( K 12 2K 211 )
2
> K 12 2K 21 1

If ( K 1 2 2K 211 ) > 0 , then we only need the 2nd term under the radical to be
positive for the inequality to hold:

4K 212 ( K 1 K 2 ) > 0
K1 K 2 > 0
K1 > K 2
K1
>1
K2

which is what is given in the text. However, we really need to combine this with the
1st assumption to get:

( K 1 2 2K 211 ) > 0
2K 21 1 > K 12
211 K 1
>
2 K2
211 K 1
and: > >1
2 K2

But what if ( K 1 2 2K 211 ) < 0 ? Now the 2nd term under the radical can be
negative and the inequality may still hold. Lets multiply the inequality by itself
the direction of the inequality must change since we are essentially multiplying it by
a negative value. Then:

( K 1 2 2K 21 1 ) + 4K 2 12 ( K 1 K 2 ) < ( K 1 2 2K 21 1 )
2 2

K1 K 2 < 0

Higher Order Systems - 20 - December 21, 2008

Colorado School of Mines CHEN403

K1 < K 2
K1
<1.
K2

Now when we combine this with the 1st assumption we get:

( K 1 2 2K 21 1 ) < 0
2K 21 1 < K 1 2
21 1 K 1
<
2 K2
21 1 K 1
and: < <1
2 K2

This allows for the possibility that the ultimate response is controlled by the 2nd
(negative) transfer function, not the 1st (positive) one.

Inverse Response from the Difference Between Two 2nd Order Transfer
Functions

Finally, lets look at the case when both of the individual transfer functions are 2nd
order. Then:

K1 K2
G(s) = 2 2
s + 21 1 s + 1 2 s + 2 22 s + 1
2 2
1

=
(K 2
1 2 )
K 2 12 s 2 + 2( K 122 K 21 1 ) s + ( K 1 K 2 )
( s
2 2
1 )(
+ 21 1 s + 1 22 s 2 + 2 22 s + 1 )
Again there are two zeros to the 2nd order numerator given by the quadratic
formula:

r1 , r2 =
( )
2( K 1 2 2 K 21 1 ) 4 ( K 1 22 K 211 ) 4 K 122 K 212 ( K 1 K 2 )
2

(
2 K K 2
1 2
2
2 1 )
( K 1 2 2 K 21 1 ) ( K 122 K 211 ) ( K 122 K 212 ) ( K 1 K 2 )
2

r1 , r2 = .
K 1 22 K 2 12

When the denominator is positive (i.e., K 1 22 K 212 > 0 K 1 22 > K 212 ) then the
largest root is given by the plus sign on the radical that is all that is necessary
for an inverse response. Then:

Higher Order Systems - 21 - December 21, 2008

Colorado School of Mines CHEN403

( K 1 2 2 K 211 ) + ( K 122 K 211 ) ( K 122 K 212 ) ( K 1 K 2 ) > 0 .

2

( K 122 K 211 ) ( K 122 K 212 ) ( K 1 K 2 ) > ( K 122 K 211 ) .

2

As we did for the previous set of transfer functions, we will multiply the inequality
by itself. What happens to the inequality sign depends upon the sign of what is on
the right hand side (since the radical on the left hand side has to be positive). If
( K 122 K 211 ) > 0 then:

( K 122 K 211 ) ( K 122 K 212 ) ( K 1 K 2 ) > ( K 122 K 211 )

2 2

( K 1 22 K 2 12 ) ( K 1 K 2 ) > 0
( K 1 K 2 ) > 0 since weve already assumed K 1 22 K 212 > 0
K1 < K 2
K1
<1
K2

Now taking all of the conditions into account:

K 1 12
K 1 22 K 2 12 > 0 1 > >
K 2 22
K
and: ( K 122 K 211 ) > 0 1 > 1 > 1 1
K 2 22

However if ( K 1 22 K 21 1 ) < 0 then we must flip the inequality:

( K 122 K 211 ) ( K 122 K 212 ) ( K 1 K 2 ) < ( K 122 K 211 )

2 2

( K 1 22 K 2 12 ) ( K 1 K 2 ) < 0
( K 1 K 2 ) < 0 since weve already assumed K 1 22 K 212 > 0
K1 > K 2
K1
>1
K2

Now when taking the other conditions into account we cannot compare them to 1:

K 1 12
K 1 22 K 212 > 0 >
K 2 22

Higher Order Systems - 22 - December 21, 2008

Colorado School of Mines CHEN403

K 1 1 1
and: ( K 122 K 211 ) > 0 >
K 2 2 2

We still have another pair of cases to consider, that when the denominator is
negative (i.e., K 1 22 K 2 12 < 0 ). Now, the largest root is given by the minus sign on
the radical. Then:

( K 1 2 2 K 211 ) ( K 122 K 211 ) ( K 122 K 212 ) ( K 1 K 2 ) < 0 .

2

( K 122 K 211 ) ( K 122 K 212 ) ( K 1 K 2 ) > ( K 122 K 211 ) .

2

Again, what happens to the inequality sign depends upon the sign of what is on the
right hand side. If ( K 1 22 K 21 1 ) > 0 then we must flip the inequality:

( K 122 K 211 ) ( K 122 K 212 ) ( K 1 K 2 ) < ( K 122 K 211 )

2 2

( K 1 22 K 2 12 ) ( K 1 K 2 ) < 0
( K 1 K 2 ) > 0 since weve already assumed K 1 22 K 212 < 0
K1 < K 2
K1
<1
K2

Now taking all of the conditions into account:

K 1 12
K 1 22 K 2 12 < 0 <
K 2 K22
and: ( K 122 K 211 ) < 0 1 < 1 1 .
K 2 2 2

Finally, if ( K 1 2 2 K 21 1 ) < 0 then we keep the inequality the same:

( K 122 K 211 ) ( K 122 K 212 ) ( K 1 K 2 ) > ( K 122 K 211 )

2 2

( K 1 22 K 2 12 ) ( K 1 K 2 ) > 0
( K 1 K 2 ) < 0 since weve already assumed K 1 22 K 212 < 0
K1 > K 2
K1
>1
K2

Now when taking the other conditions into account we cannot compare them to 1:

Higher Order Systems - 23 - December 21, 2008

Colorado School of Mines CHEN403

K 1 12
K K <0 1<
2 2
<
K 2 22
1 2 2 1

K
and: K 1 22 K 21 1 < 0 1 < 1 < 1 1 .
K 2 2 2

Higher Order Systems - 24 - December 21, 2008