C H A P T E R

7
Linear Systems in the Complex
Frequency Domain: The Laplace
Transform
7.1 GOALS OF THIS CHAPTER
We now have two techniques to find the response of a system to any input. well, almost
any input: it has to be either periodic or aperiodic. In Chapter 5, we divided the input into
short time segments and used convolution to sum up the resultant impulse responses. In
Chapter 6, we used the Fourier transform to divide the input into sinusoids, found the output
to each sinusoid by multiplying it with the transfer function, and then summed the output
sinusoids using the inverse Fourier transform. We should always keep in mind that both
of these methods involve the principle of superposition and time invariance so they apply
only to linear, time invariant (LTI) systems. Even so, neither of these approaches can handle
a third class of signals: waveforms that suddenly change and never return to a baseline level
(recall Section 1.4.2). A classic example is the “step function,” which changes from one value
to another at one instant of time (often taken as t ¼ 0) and remains at the new value for all
eternity, Figure 7.1.

FIGURE 7.1 Time plot of a step function that changes from 0 to value V1 at t ¼ 0.0 s. It remains at this value for all
time.

Circuits, Signals, and Systems for Bioengineers

https://doi.org/10.1016/B978-0-12-809395-5.00007-2 295 Copyright © 2018 Elsevier Inc. All rights reserved.
296 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

One-time changes, or changes that do not return to some baseline level, are common in
nature and even in everyday life. For example, this text should leave the reader with a lasting
change in his or her knowledge of biosystems and biosignals, one that does not return to the
baseline, precourse level. However, unlike a step function, this change is not expected to
occur instantaneously. An input signal could change a number of times, either upward or
downward, and might even show a partial return to its original starting point. But if the
signal never returns to some original level, it cannot be either periodic or aperiodic and we
do not have the tools to analyze how a system responds to these signals. As noted in Chapter
1, signals that change and never return to baseline are sometimes referred to as “transient”
signals. Of course all signals vary in time and therefore could technically be called transient
signals, but this term is often reserved for signals that have one-time or step-like changes.
This is another linguistic issue in engineering where context can vary the meaning.
The goal of this chapter is to master the Laplace transform, a technique that allows us to
analyze the response of LTI systems to transient or step-like inputs. It also enables us to
analyze systems that have initial conditions. The downside is that this is a purely analytical
technique and cannot be applied to discrete signals. So this rules out computer applications.
Moreover, think back on the difficulty of working out the convolution integral analytically as
in Example 5.3. Maybe we should avoid the Laplace transform and try working around the
signal limitations of our previous methods. Could we not approximate a step function as a
pulse having a very long pulse width? You could, and that is very likely what you would
do in many situations. But we still cannot avoid the Laplace transform; it is just too important
a concept in systems analysis.
The importance of the Laplace transform lies in its theoretical implications, although it
does have some practical applications. Like the Fourier transform, it converts signals or func-
tions from the time domain to something we call the Laplace domain. The Laplace domain is
actually more general than the Fourier transform and, since it is easy to go from the Laplace
domain to the frequency domain (but not vice versa), it is the preferred domain for transfer
function equations. So most systems are described in the Laplace domain even if they do not
use Laplace transforms for evaluation. MATLAB’s powerful simulation language SimulinkÒ
described in Chapter 9 uses the Laplace notation to define system elements. If the input
signals are periodic or aperiodic, we can easy slip back into the frequency domain and apply
all the computer-friendly methods described in Chapters 5 and 6.
Topics in this chapter reflect many of those in Chapter 6 and include:
• The development of the Laplace transform as an extension of the Fourier transform and
the introduction of the concept of complex frequency;
• How to determine the Laplace transform of a signal;
• The transfer function in the Laplace domain;
• How to find the output to any input in the Laplace domain (the output signal is found
in the same way as in Chapter 6, by multiplying the signal by the transfer function, but
this time in the Laplace domain);
• How to take inverse Laplace transform to get the time domain signal (this may involve
additional algebraic tools such as partial fraction expansion);
• The relationships between the various methods for defining systems and describing
their behavior.

II. SYSTEMS
 7.2 THE LAPLACE TRANSFORM 297

7.2 THE LAPLACE TRANSFORM

The Laplace transform is used to define systems using transfer functions similar to the
frequency domain transfer functions used in the last chapter.1 But the Laplace domain can
also represent signals that do not return to a baseline level. With this tool, we can extend
all of the techniques developed in the last chapter to systems exposed to this wider class
of signals. The Laplace domain can also be used to analyze systems that begin with nonzero
initial conditions, a situation we have managed to avoid thus far, but one that does occur in
the real world.

7.2.1 Definition of the Laplace Transform

Transfer functions written in Laplace notation differ only slightly from the frequency
domain transfer function used in the last chapter. The reason that frequency domain transfer
functions cannot be used when signals change in a step-like manner is simply that these
functions cannot be decomposed into sinusoidal components using either Fourier series
analysis or the Fourier transform. Consider a function similar to that shown in Figure 7.1,
a function that is 0.0 for a t  0 and 1.0 for all t > 0:


0 t0
xðtÞ ¼ (7.1)
1 t>0

The function defined in Equation 7.1 is known as the “unit step function” since it begins at
zero and jumps to 1.0 at t ¼ 0, but a generic step function could begin at any level and jump
to any other level. Try to find the Fourier transform of this function and you get:
Z N Z N
FTðuÞ ¼ xðtÞejut dt ¼ 1ejut dt0N (7.2)
0 0

The problem is that because x(t) does not return to its baseline level (zero in this case), the
limits of the integration must be infinite. Since the sinusoidal function ejut has nonzero
values out to infinity, the integral becomes infinite. In the past, our input signals had a finite
life span and the Fourier transform integral need only be taken over that finite time period.
But for transient signals this integral cannot be computed.
The trick used to solve this infinite integral problem is to modify the exponential function
so that it converges to zero at large values of t even if the signal, x(t), does not. This can be
accomplished by multiplying the sinusoidal term, ejut, by a decaying exponential such as
est where s is some positive real variable. In this case the sinusoid term in the Fourier trans-
form becomes a complex sinusoid, or rather a sinusoid with a complex frequency:

ejut est ¼ eðsþjuÞt ¼ est (7.3)

1
In fact, some purists believe that only transfer functions written in Laplace notation are worthy of the term
“transfer function.”

II. SYSTEMS
298 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

where s ¼ s þ ju and is termed the “complex frequency” because it is a complex variable, but
has the same role as frequency, u, in the Fourier transform exponential. The complex variable,
s, is also known as the “Laplace variable” since it plays a critical role in the Laplace transform
(another example of multiple names for the same thing). A modified version of the Fourier
transform can now be constructed using complex frequency in place of regular frequency;
that is, s (which is s þ ju) instead of just ju. This modified transform is termed the Laplace
transform:
Z N
XðsÞ ¼ LxðtÞ ¼ xðtÞest dt (7.4)
0

where the script L indicates the Laplace transformation.

So the trick is to use complex frequency, with its decaying exponential component, to
cause convergence for functions that would not otherwise converge. For any general signal,
x(t), the product of x(t)est ¼ x(t)e(sþju)t may not necessarily converge to zero as t / N, in
which case the Laplace transform does not exist.2 Some advanced signal processing gurus
have spent a lot of time worrying about such functions: which functions converge, their
ranges of convergence, or how to get them to converge. Fortunately, such matters need
not concern us since most common real-world signals, including the step function, do
converge for some values of s and so have a Laplace transform. The range of s’s over which
a given product of x(t) and e(sþju)t converges is another occupation of signal processing
theoreticians, but again is not something bioengineering signal processors need worry about.
If a signal has a Laplace transform (and all the signals we use do), then the product x(t)est
converges as t / N.
The Laplace transform is a purely analytical technique so it cannot be solved on a com-
puter and also has two other issues: it cannot be applied to functions for negative values
of t, and it is difficult to evaluate using Equation 7.4 for any but the simplest of functions.
The restriction against representing functions for negative time values comes from the fact
that est becomes a positive exponential and will go to infinity as t goes to large negative
values. (For negative t, the real part of the exponential becomes eþst and does exactly the
opposite of what we want it to do: it forces divergence rather than convergence.) The only
way around this is to limit our analyses to t > 0, but this is usually not a serious restriction.
The other problem, the difficulty in evaluating Equation 7.4, stems from the fact that s is com-
plex, so although the integral in Equation 7.4 does not look so complicated, the complex inte-
gration becomes very involved for all but a few simple functions of x(t). To get around this
problem, we use tables that give us the Laplace transforms of frequently used functions. Such
a table is given in Appendix B, and a more extensive list can easily be found on the internet. A
Laplace transform table is used both to determine the Laplace transform of a signal and,
using it in reverse, the inverse Laplace transform. The only difficulty in finding the inverse
Laplace transform is rearranging the output Laplace function into one of the formats found
in the table.

For example, the function xðtÞ ¼ et will not converge as t / N even when multiplied by est, so it does not
2 2

have a Laplace transform.

II. SYSTEMS
 7.2 THE LAPLACE TRANSFORM 299

EXAMPLE 7.1
Find the Laplace transform of the step function in Equation 7.1.
Solution: The step function is one of the few functions that can be evaluated easily using the basic
defining equation of the Laplace transform, Equation 7.4.
Z N Z N N  
est  1
XðsÞ ¼ xðtÞest dt ¼ 1est dt ¼   ¼ 0  
0 0 s 0 s
1
XðsÞ ¼
s

7.2.2 Calculus Operations in the Laplace Domain

Just as in the frequency domain, the calculus operations of differentiation and integration
can be reduced to algebraic operations in the Laplace domain. The Laplace transform of the
derivative operation can be determined from the defining equation, Equation 7.4.
Z N
dxðtÞ dxðtÞ st
L ¼ e dt
dt 0 dt

Integrating by parts:
Z N
dxðtÞ N
L ¼ xðtÞest j0 þ s xðtÞest dt
dt 0
R 
N st
From the definition of the Laplace transform, 0 xðtÞe dt ¼ XðsÞ , the right-most term
in the summation is sX(s), and the equation becomes:

dxðtÞ
L ¼ xðNÞeN  xð0Þe0 þ sXðsÞ
dt

dxðtÞ
L ¼ sXðsÞ  xð0  Þ (7.5)
dt

Equation 7.5 shows that in the Laplace domain, differentiation becomes multiplication by
the Laplace variable s with the additional subtraction of the value of the function at t ¼ 0.
The value of the function at t ¼ 0 is known as the “initial condition.” This value can be
used, in effect, to account for all negative time history of x(t). In other words, all of the
behavior of x(t) when t was negative can be lumped together as a single initial value at
t ¼ 0. This trick allows us to include some aspects of the system’s behavior over negative
values of t even if the Laplace transform does not itself apply to negative time. If the initial
condition is zero, as is frequently the case, then differentiation in the Laplace domain is sim-
ply multiplication by s.

II. SYSTEMS
300 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

Multiple derivatives can also be taken in the Laplace domain, although this is not such a
common operation. Multiple derivatives involve multiplication by s n-times, where n is the
number of derivative operations, and taking the derivatives of the initial conditions:

dn xðtÞ dxð0  Þ dn1 xð0  Þ

L n
¼ sn XðsÞ  sn1 xð0  Þ  sn2 . (7.6)
dt dt dt

Again, if there are no initial conditions, taking n derivatives becomes just a matter of
multiplying x(t) by sn.
If differentiation is multiplication by s in the Laplace domain, then integration is simply a
matter of dividing by s. Again, a second term account for the initial conditions:
2 3
Z T Z
4 5 1 1 0
L xðtÞdt ¼ XðsÞ þ xðtÞdt (7.7)
0 s s N

If there are no initial conditions, then integration is accomplished by simply dividing by s.

The second term of Equation 7.7 is a direct integral that accounts for the initial conditions and
is again a way of accounting for the negative time history of the system.

7.2.3 SourcesdCommon Signals in the Laplace Domain

In the Laplace domain, both signals and systems are represented by functions of s. As
mentioned earlier, the Laplace domain representation of signals is determined from a table.
Although it might seem that there could be a large variety of such signals, which would
require a very large table, in practice only a few signal types commonly occur in systems anal-
ysis. The signal most frequently encountered in Laplace analysis is the step function shown in
Figure 7.1, or its more constrained version, the unit step function given in Equation 7.1 and
repeated here:


0 t0
xðtÞ ¼ uðtÞ ¼ (7.8)
1 t>0

The symbol u is used frequently to represent the unit step function. The Laplace transform
of the step function was found in Example 7.1 and repeated here:
1
XðxÞ ¼ UðsÞ ¼ LuðtÞ ¼ (7.9)
s

As with the Fourier transform, it is common to use capital letters to represent the Laplace
transform of a time function. Two functions closely related to the unit step function are the
ramp and impulse functions, Figure 7.2.
These functions are related to the step function by differentiation and integration. The unit
ramp function is a straight line with slope of 1.0.


t t>0
rðtÞ ¼ (7.10)
0 t0

II. SYSTEMS
 7.2 THE LAPLACE TRANSFORM 301

FIGURE 7.2 The ramp and impulse are two signals related to the step function and are commonly encountered in
Laplace analysis. As described in Chapter 5, an ideal impulse occurs at t ¼ 0 and is infinitely narrow and infinitely
tall. Real-world impulse signals are approximated by short pulses (see Example 5.1).

Since the unit ramp function is the integral of the unit step function, its Laplace transform
will be that of the step function divided by s:
 
1 1 1
RðsÞ ¼ LrðtÞ ¼ ¼ 2 (7.11)
s s s

The impulse function is the derivative of a unit step, which leads to one of those mathe-
matical fantasies: a function that becomes infinitely short, but as it does its amplitude
becomes infinite so the area under the function remains 1.0 as shown in Equation 7.12.

1 a a
xðtÞ ¼ dðtÞ ¼ lim t (7.12)
a/0 a 2 2

In practice, a short pulse is used as an impulse function and an approach to finding the
appropriate pulse width for any given system is described in Example 5.1.
Since the impulse response is the derivative of the unit step function, its Laplace transfer
function is that of a unit step multiplied by s:
 
1
DðsÞ ¼ LdðtÞ ¼ s ¼ 1 (7.13)
s

Hence the Laplace transform of an impulse function is a constant, and if it is a unit impulse
(the derivative of a unit step) then that constant is 1. As you might guess, this fact will be
especially useful in the analysis of Laplace transfer functions. The Laplace transforms of other
common signal functions are given in a table in Appendix B.

7.2.4 Converting the Laplace Transform to the Frequency Domain

Remember that s is a complex frequency and equals a real term plus the standard imagi-
nary frequency term: s ¼ a þ ju. To convert a Laplace function to the frequency domain, we
simply substitute ju for s. When we do this, we are agreeing to restrict ourselves to sinusoidal
steady-state signals, or those that can be decomposed into such signals, so the real component
of s with its convergence properties is no longer needed. Converting the Laplace transform
function to the frequency domain allows us to determine the frequency characteristics of
that function using Bode plot techniques or computer-based methods. This approach is
used in the next section to evaluate the frequency characteristics of a time delay process.

II. SYSTEMS
302 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

TABLE 7.1 Transformations Between the Time, Frequency, and

Laplace Domains.
From To Transformation

Time Frequency Fourier transform

Frequency Time Inverse Fourier transform

Time Laplace Laplace transform

Laplace Time Inverse Laplace transform
Laplace Frequency s / ju
Assumes sinusoidal steady state
Frequency Laplace ju / s
Assumes no initial conditions

7.2.5 The Inverse Laplace Transform

Working in the Laplace domain is pretty much the same as working in the frequency
domain: the math is still algebra. The transfer function of system elements will be in Laplace
notation as described later and summarized in Table 7.1. First, the input signal is converted to
the Laplace domain using the table in Appendix B. Then multiplying the input signal with the
transfer function provides the output response, but again as a function of s, not ju. Sometimes
the Laplace representation of the solution or even just the Laplace transfer function is
sufficient, but if a time domain solution is desired, then the inverse Laplace transform
must be determined.
The equation for the inverse Laplace transform is given as:
Z sþjN
1 1
xðtÞ ¼ L XðxÞ ¼ XðsÞest ds (7.14)
2p sjN

Unlike the inverse Fourier transform, this equation is quite difficult to solve even for
simple functions. So to evaluate the inverse Laplace transform, we use the Laplace transform
table in Appendix B in the reverse direction: find a function (on the right side of the table) that
matches your Laplace output and convert it to the equivalent time domain signal. The diffi-
culty is usually in rearranging the Laplace output function to conform to one of the formats
given in the table. Methods for doing this are described next and specific examples given.

7.3 THE LAPLACE DOMAIN TRANSFER FUNCTION

The analysis of systems using the Laplace transform is no more difficult than in the
frequency domain, except that there may be the added task of accounting for initial condi-
tions. In addition, to go from the Laplace domain back to the time domain may require
some algebraic manipulation of the Laplace output function.

II. SYSTEMS
 7.3 THE LAPLACE DOMAIN TRANSFER FUNCTION 303
The transfer function introduced in Chapter 6 is ideally suited to Laplace domain analysis,
particularly when there are no initial conditions. In the frequency domain, the transfer func-
tion is used primarily to determine the spectrum of a system or system element, but it can
also be used to determine the system’s response to any input provided that input can be
expressed as a sinusoidal series or is aperiodic. In the Laplace domain, the transfer function
can be used to determine a system’s output to a broader class of input signals. The Laplace
domain transfer function is similar to its cousin in the frequency domain, except the
frequency variable, u, is replaced by the complex frequency variable, s:

OutputðsÞ
TFðsÞ ¼ (7.15)
InputðsÞ

Like its frequency domain cousin, the general Laplace domain transfer function consists of
a series of polynomials. The general transfer function in the frequency domain was given in
Equation 6.50 and repeated here:
   2 !
u u 2d1 u
Gju 1 þ j 1 þj .
u1 un1 un1
TFðuÞ ¼    2 ! (7.16)
u u 2d2 u
ju 1 þ j 1 þj .
u2 un2 un2

In the Laplace domain, the general form of the transfer function differs in three ways: (1)
the frequency variable ju is replaced by the complex frequency variable s; (2) the polynomials
are normalized so that the highest order of the frequency variable is normalized to 1.0 (as
opposed to the constant term as in Equation 7.16); and (3) the order of the frequency terms
is reversed with the highest-order term (now normalized to 1) coming first:

Gsðs þ u1 Þ s2 þ 2d1 un1 s þ u2n1 .
TFðsÞ ¼ (7.17)
sðs þ u2 Þðs2 þ 2d2 un2 s þ u2n2 Þ.

The constant terms, d, u, un, etc. have the same meaning as in Equation 6.50. Even the ter-
1
minology used to describe the elements is the same. For example, sþu , which is related to 1 ,
1 ju
1þ
u1
is called a first-order element, and s2 þ2du1 sþu2 is a second-order element. Note that in
n n
Equation 7.17, the constants d and un have a more orderly arrangement in the transfer function
equation.
As described in Section 7.1.4, it is possible to go from the Laplace domain transfer function
to the frequency representation simply by substituting ju for s, but you should also rearrange
the coefficients and frequency variable to fit the frequency domain format. As in the
frequency domain, the assumption is that any higher-order polynomial (third-order or above)
can be factored into the first- and second-order terms.
As with the frequency domain transfer function, we cover each of the element types sepa-
rately: gain and first-order elements followed by the intriguing behavior of the second-order
element. But first we introduce a new element, the “time delay” element, commonly found in
physiological systems.

II. SYSTEMS
304 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

7.3.1 Time Delay Element: The Time Delay Theorem

Many physiological processes have a delay before they begin to respond to a stimulus.
Such physiological delays are termed “reaction time,” or “response latency,” or simply
“response delay.” In these processes, there is a period of time between the onset of the stim-
ulus and the beginning of the response. In systems involving neurological control, this delay
is due to processing delays in the brain. The Laplace domain has an element to represent such
delays.
The Time Delay theorem can be derived from the defining equation of the Laplace trans-
form, Equation 7.4. Assume a signal x(t) that is zero for negative time, but normally changes
at t ¼ 0 s. If this signal is delayed by T seconds, then the delayed function would be x(t  T).
From the defining equation (Equation 7.4), the Laplace transform of such a delayed function
would be:
Z N
L½xðt  TÞ ¼ xðt  TÞest dt
0

Defining a new variable: g ¼ t  T

Z N Z N Z N
L½xðt  TÞ ¼ xðgÞesðTþgÞ dg ¼ xðgÞesT esg dg ¼ esT xðgÞesg dg
0 0 0

But the integral in the right-hand term is the same as Laplace transform of the function,
x(t), it just has a different variable for time. So the right-hand integral is the Laplace transform
of an unshifted x(t); i.e., it is L½xðtÞ. Hence the Laplace transform of the shifted function
becomes:

L½xðt  TÞ ¼ esT L½xðtÞ (7.18)

Equation 7.18 is the Time Delay theorem, which can also be used to construct an element
that represents a pure time delay, an element with a transfer function:

TFðsÞ ¼ esT (7.19)

where T equals the delay, usually in seconds. So an element with a transfer function TF(s) ¼ esT
is a time delay of T seconds. This element is often found in models of neurological control
where it represents neural processing delays.

EXAMPLE 7.2
Find the spectrum of a system time delay element for two values of delay: 0.5 and 2.0 s. Use
MATLAB to plot the spectral characteristics of this element.
Solution: The Laplace transfer function of a pure time delay of 2.0 s would be:

TFðsÞ ¼ e2s

II. SYSTEMS
 7.3 THE LAPLACE DOMAIN TRANSFER FUNCTION 305
To determine the spectrum of this element, we need to convert the Laplace transfer function to
the frequency domain. This is done simply by substituting ju for s in the transfer function equation:

TFðuÞ ¼ ej2u
The magnitude of this frequency domain transfer function is:
 
jTFðuÞj ¼ ej2u  ¼ 1 (7.20)

for all values of u since changing u (or the constant 2 for that matter) only changes the angle of
the imaginary portion of the exponential, not its magnitude. The phase of this transfer function is:

:TFðuÞ ¼ :ej2u ¼ 2u rad: (7.21)

Since the magnitude of the transfer function is 1.0, the input and output signals have the same
magnitude at all frequencies. The time delay process only changes the phase of the output signal by
making it more negative with increasing frequency (i.e., adding 2u to the phase of the input signal).
For a 0.5-s delay, :TFðuÞ ¼ :e0:5ju , so the phase curve would be 0.5u. A pure time delay
decreases the phase component of a signal’s spectrum in a linear manner, as shown in Figure 7.3. A
time delay process is also explored in one of the problems.

FIGURE 7.3 The magnitude and phase spectrum of a time delay element with two time delays: 2 and 0.5 s. In
both cases the magnitude spectrum is a constant 0 dB (output equal input) as expected and the phase spectrum
decreases linearly with frequency. Note that when u ¼ 100 rad/s the phase angle is 12,000 degrees for the 2-s time
delay as predicted by Equation 7.21 (2*100*360/2p ¼ 12,000 degrees).

II. SYSTEMS
306 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

The MATLAB code used to generate the magnitude and phase spectrum is a minor variation of
Example 6.7. The plotting is done using linear frequency rather than log, to emphasize that the
phase spectrum is a linear function of frequency.

% Example 7.2 Use MATLAB to plot the transfer function of a time delay
%
T ¼ 2; % Time delay in sec.
w ¼ .1:1:100; % Frequency vector
TF ¼ exp(-j*T*w); % Transfer function
Mag ¼ 20*log10(abs(TF)); % Calculate magnitude spectrum
Phase ¼ unwrap(angle(TF))*360/(2*pi); % Calculate phase spectrum
... Repeat for T ¼ 0.5 and plot and label..

Results: The time delay element increases the phase linearly with frequency as shown in
Figure 7.3.

7.3.2 Constant Gain Element

The gain element is not a function of frequency, so its transfer function is the same irrespec-
tive of whether standard or complex frequency is involved:

TFðsÞ ¼ G (7.22)

The system representation for a gain element is the same as in the frequency domain
representation, Figure 6.9.

7.3.3 Derivative Element

Equation 7.5 shows that the derivative operation in the Laplace domain is implemented
simply by multiplication with the Laplace variable s. This is analogous to this operation in
the frequency domain, where differentiation is accomplished by multiplying by ju. So in
the absence of initial conditions, the Laplace transfer function for a derivative element is:

TFðsÞ ¼ s (7.23)

The Laplace system representation of this element is shown in Figure 7.4A

FIGURE 7.4 (A) Laplace system representation of a derivative element. (B) Laplace system representation of an
integrator element.

II. SYSTEMS
 7.3 THE LAPLACE DOMAIN TRANSFER FUNCTION 307

7.3.4 Integrator Element

As shown in Equation 7.7, integration in the Laplace domain is accomplished by dividing
by s. So in the absence of initial conditions, the system representation of integration in the
Laplace domain has a transfer function of

TFðsÞ ¼ 1=s (7.24)

This parallels the representation in the frequency domain where 1/s becomes 1/ju. The
representation of this element is shown in Figure 7.4B.

EXAMPLE 7.3
Find the Laplace and frequency domain transfer function of the system in Figure 7.5. The gain
term k is a constant.
Solution: The approach to finding the transfer function of a system in the Laplace domain is
exactly the same as in the frequency domain used in Chapter 6. Here we are asked to determine
both the Laplace and frequency domain transfer function. First we find the Laplace transfer func-
tion, then substitute ju for s, and rearrange the format for the frequency domain transfer function.
We could solve this several different ways, but this is clearly a feedback system so the easiest
solution is to use the feedback equation, Equation 6.7. All we need to do is find the equivalent
feedforward gain function, G(s), and the equivalent feedback function, H(s). The feedback function
is H(s) ¼ 1 since all of the output feeds back to the input in this system.3 The feedforward gain is the
product of the two elements:
1
GðsÞ ¼ k
s
Substituting G(s) and H(s) into the feedback equation:
k
GðsÞ s k
TFðsÞ ¼ ¼ ¼
1 þ GðsÞHðsÞ k sþk
1þ
s
Result: Again, the Laplace domain transfer function has the highest coefficient of complex fre-
quency, in this case s, normalized to 1. Substituting s ¼ ju and rearranging the normalization:
k 1
TFðuÞ ¼ ¼ u
ju þ k 1þj
k

FIGURE 7.5 Laplace representation of a system used in Example 7.3.

II. SYSTEMS
308 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

This has the same form as a first-order transfer function given in Equation 6.31 and shown in
Figure 6.12B, where the constant term u1 is k.
3
Such systems are known as “unity gain feedback systems” since the feedback gain, H(s), is 1.0 (unity).

7.3.5 First-Order Element

First-order processes contain an s þ u1 term in the denominator (or possibly only an s).
Later we find that these systems contain a single energy storage device. The Laplace transfer
function for a first-order process is:
1
1
TFðsÞ ¼ ¼ s (7.25)
s þ u1 1
sþ
s

In this transfer function equation, a new variable is introduced, s, which is termed the
“time constant” for reasons shown in Example 7.4. As seen in Equation 7.25, this time con-
stant variable, s, is simply the inverse of the frequency constant, u1:
1
s ¼ (7.26)
u1

In Chapter 6, we found that the frequency constant u1 is where the magnitude spectrum
is 3 dB (Equation 6.34) and the phase spectrum is 45 degrees (Equation 6.37). In the next
example, we show that the time constant s has a direct relationship to the time behavior of a
first-order element.
Now that the transfer function is in the Laplace domain, we can explore the response of
these two similarly behaving systems using input signals other than sinusoids. Two popular
signals that are used are the step function, shown in Figure 7.1, and the impulse function,
shown in Figure 7.2. Typical impulse and step responses of a first-order system are found
in the next example.

EXAMPLE 7.4
Find the arterial pressure response of the linearized model of the GuytoneColeman body fluid
balance system (Ridout, 1991) to a step increase of fluid intake of 0.5 mL/min. The frequency
domain version of this model is shown in Figure 6.4. Also find the pressure response to fluid intake
of 250 mL administered as an impulse.4 This is equivalent to drinking approximately 1 cup of fluid
quickly. After finding the analytical solution, use MATLAB to plot the outputs in the time domain.
Solution, step response: The first step is to convert to Laplace notation from the frequency notation

used in Figure 6.4. The feedforward gain becomes GðsÞ ¼ 16:67 0:06 s ¼ 1s and the feedback gain

becomes H(s) ¼ 0.05. Again applying the feedback equation (Equation 6.7), the transfer function is:
1
PA ðxÞ GðsÞ s 1
TFðsÞ ¼ ¼ ¼ ¼
FIN ðsÞ 1 þ GðsÞHðsÞ 1 s þ 0:05
1 þ 0:05
s

II. SYSTEMS
 7.3 THE LAPLACE DOMAIN TRANSFER FUNCTION 309
where PA is arterial blood pressure in mmHg and FIN is a small change in fluid intake in
mL/min.
To find the step response, multiply the Laplace input signal by the transfer function:
1
PA ðsÞ ¼ FIN ðsÞTFðsÞ ¼ TFðsÞ (7.27)
s
Substituting in for TF(s), Equation 7.27 becomes:
 
1 1 1
PA ðsÞ ¼ ¼ (7.28)
ðs þ 0:05Þ s sðs þ 0:05Þ
To take the inverse Laplace transform and get the time domain output, we need to rearrange the
resulting Laplace output function so that it has the same form as one of the functions in the Laplace
Transform Table of Appendix B. Often this can be the most difficult part of the problem. Fortu-
nately, in this problem, we see that the right-hand term of Equation 7.28 matches the Laplace
function in entry number 4 of the table.
1 a
has the form : 5ð1  eat Þ=a
sðs þ 0:05Þ sðs þ aÞ
where a ¼ 0.05. Hence the step response in the time domain for this system is an exponential:
1  
pA ðtÞ ¼ 1  e0:05 ¼ 20 1  e0:05 mmHg
0:05
Solution, impulse response: Solving for the impulse response is even easier since for the impulse
function, FIN(s) ¼ 1. So the impulse response of a system in the Laplace domain is the transfer
function itself. Therefore, the impulse response in the time domain is the inverse Laplace transform
of the transfer function:
In this example the impulse input was 250 mL, so FIN(s) ¼ 250, and PA(s) ¼ 250 TF(s):
250
PA ðsÞ ¼ mmHg
s þ 0:05
The Laplace output function is matched by entry 3 in the Laplace Transform Table except for the
constant term. From the definition of the Laplace transform in Equation 7.4, any constant term can
be removed from the integral, so the Laplace transform of a constant times a function is the constant
times the Laplace transform of the function. Similarly, the inverse Laplace transform of a constant
times a Laplace function is the constant times the inverse Laplace transform. Stating these two
characteristics formally:
L½kxðtÞ ¼ kLxðtÞ (7.29)
1 1
L ½kXðsÞ ¼ kL ½XðsÞ (7.30)
So the time domain solution for PA(s) is obtained:
 
1 1
Vout ðsÞ ¼ 250 which has the same form as : k 5keat
s þ 0:05 sþa
vout ðtÞ ¼ 250e0:05 mmHg

II. SYSTEMS
310 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

Plotting this result with the help of MATLAB is easy as shown.

% Example 7.4 First-order system impulse and step response for two time constants
%
t ¼ 0:.1:100; % Define time vector: 0 to 100 min
x ¼ 250*exp(-0.05); % Impulse response
plot(t,x,'k'); % Plot impulse response
....labels, title, and other text..
x ¼ 20*(1 e exp(-0.05)); % Step response.
plot(t,x,'k'); % Plot step response
....plot, labels, title, and other text..

Results: The system responses are shown in Figure 7.6. Both the impulse and step responses are
exponential with time constants of 1/0.05 ¼ 20 min. When time equals the time constant (i.e., t ¼ s),
the value of the exponential becomes: et/s ¼ e1 ¼ 0.37. Hence at time t ¼ s, the exponential is
within 37% of its final value. In other words, the exponential has attained 73% of its final value after
one time constant.
The time constant makes a good measure of the relative speed of an exponential response. As a
rule of thumb, an exponential is considered to have reached its final value when t > 5s, although in
theory the final value of an exponential is reached only when t ¼ N.
4
Logically, the response to a step change in input is called the “step response,” just as the response due to an
impulse is termed the “impulse response.”

FIGURE 7.6 Response of a linearized model of the GuytoneColeman body fluid balance system. The graph
shows the change in arterial pressure to an increase of fluid intake, either taken in a step-like manner (right) or as an
impulse (left).

II. SYSTEMS
 7.3 THE LAPLACE DOMAIN TRANSFER FUNCTION 311
There are many other possible input waveforms in addition to the step and impulse
function. As long as the waveform has a Laplace representation, the response to any signal
can be found from the transfer function.5 However, the step and/or impulse responses
usually provide the most insight into the general behavior of the system.

7.3.5.1 The Characteristic Equation

First-order transfer functions can be slightly different than that of Equation 7.25. There
can be a single s or even an (s þ 1/s) in the numerator. What bonds all first-order systems
is the denominator term, (s þ 1/s), or equivalently (s þ u1). All first-order systems will
contain a s in the denominator irrespective of what is in the numerator. In the next section,
we find that all second-order systems contain an s2 in the denominator of the transfer func-
tion, usually as a quadratic polynomial of s. Again the numerator can be a variety of poly-
nomials of s.
The characteristics of the transfer function are primarily determined by the denomina-
tor, which explains why the denominator is called the “characteristic equation.” The
characteristic equation describes the general behavior of the system. Often, only the
characteristic equation is required to determine the salient features of a response: a
time domain solution is not needed. For example, the characteristic equation s þ 3 tells
us that the system’s response will include an exponential having a time constant of 1/3
or 0.33 s. Second-order characteristic equations are even more informative, as the next
section shows.

EXAMPLE 7.5
Find the transfer function of the system shown in Figure 7.7.
Solution: As in Example 7.3 and previous examples, we use the feedback equation. We just need
to find G(s) and H(s). Again H(s) ¼ 1: another unity gain feedback system. The feedforward gain
function is just the product of the two feedforward elements:
 
1 5 5
GðsÞ ¼ ¼ 2
sþ7 s s þ 7s

FIGURE 7.7 System used in Example 7.5. The transfer function of this system is to be determined.

5
Even a sine wave has a Laplace transform, but it is complicated. It is much easier to use the frequency domain
techniques developed in Chapter 6 for sine waves as long as they are in steady state. If the sine wave is not in
steady state, but starts at some particular time, say t ¼ 0, then Laplace techniques and the Laplace transform of a
sine wave must be used.

II. SYSTEMS
312 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

Result: Substituting into the feedback equation gives the transfer function:
5
TFðsÞ ¼
GðsÞ
¼ s2 þ 7s ¼ 5
1 þ GðsÞHðsÞ 5 s2 þ 7s þ 5
1þ 2
s þ 7s
The s2 in the characteristic equation indicates that this is a second-order system. These elements
are covered in the next section.

7.3.6 Second-Order Element

Second-order processes have an s2 in the denominator of the transfer function, usually as a
quadratic polynomial:

1 1
TFðsÞ ¼ ¼ 2 (7.31)
s2 þ 2dun s þ u2n s þ bs þ c

The denominator of the right-hand term is the familiar notation for a standard quadratic
equation and can be factored using Equation 7.32. As we find later, second-order systems
must contain two energy storage devices.
One method for dealing with second-order terms would be to factor them into first-order
terms. We could then use partial fraction expansion to expand the two factors into two terms
of the form s þ a. This approach is perfectly satisfactory if the factors, the roots of the
quadratic equation, are real. Examination of the classic quadratic equation demonstrates
when this approach will work. Since the coefficient of the s2 term is always normalized to
1.0, the a coefficient in the quadratic is always 1.0 and the roots of the quadratic equation
become:

b 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
r1 ; r2 ¼  b2  4c (7.32)
2 2

If b2  4c then the roots will be real and the quadratic can be factored into two first-order
terms: s  r1 and s  r2. However, if b2 < 4c, both roots will be complex and have real and
imaginary parts:

b 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffi2ffi b 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffi2ffi
r1 ¼ þj 4c  b and r2 ¼ j 4c  b (7.33)
2 2 2 2

If the roots are complex, they both have the same real part (b/2), whereas the imaginary
parts also have the same values but with opposite signs. Complex number pairs that feature
this relationship, the same real part but oppositely signed imaginary parts, are called “com-
plex conjugates.”
Whether or not the roots of a second-order characteristic equation are real or complex has
important consequences in the behavior of the system. Sometimes all we need to know about
a second-order system is whether the roots in the characteristic equation are real or imagi-
nary. This saves the effort of finding the inverse Laplace transform.

II. SYSTEMS
 7.3 THE LAPLACE DOMAIN TRANSFER FUNCTION 313
The second-order transfer function variables d and un were introduced in Chapter 6. Recall
that the parameter d is called the “damping factor,” whereas un is called the “undamped nat-
ural frequency.” As with the term “time constant,” applied to first-order systems, these
names relate to the step and impulse response behavior of the second-order systems. As is
shown later, second-order systems with low damping factors will respond with an exponen-
tially decaying oscillation, and the smaller the damping factor, the slower the decay. The rate
of oscillation is related to un. Specifically, the rate of oscillation, ud, of these underdamped
systems is:
pffiffiffiffiffiffiffiffiffiffiffiffiffi
ud ¼ un 1  d2 (7.34)

So as d becomes smaller and smaller, the oscillation frequency ud approaches un. When
d equals 0.0, the system is “undamped” and the oscillation continues forever at frequency
un. Hence the term “undamped natural frequency” for un: it is the frequency at which the
system oscillates if it has no damping; i.e., d / 0.
Here we equate the variables un and d to coefficients a and b in Equation 7.31 and insert
them into the solution to the quadratic equation:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2dun 4d2 u2n  4u2n pffiffiffiffiffiffiffiffiffiffiffiffiffi
r1 ; r2 ¼  ¼ dun  un d2  1 ¼ dun  ud (7.35)
2 2

From this equation, we see that the damping factor d alone determines if the roots will be
real or complex. Specifically, if d > 1 then the constant under the square root is positive and
the roots will be real. Conversely, if d < 1 the square root will be a negative number and the
roots will be complex. If d ¼ 1, the two roots are also real, but are the same: both roots
equal un.
Again, the behavior of the system is quite different if the roots are real or imaginary, and
the form of the inverse Laplace transform is also different. Accordingly, it is best to examine
the behavior of a second-order system with real roots and complex roots separately: they act
as two different animals.

7.3.6.1 Second-Order Elements With Real Roots

If the roots are real (i.e., d > 1), the system is said to be “overdamped” because it does not
exhibit oscillatory behavior in response to a step or impulse input. Such systems have re-
sponses consisting of double exponentials.
The best way to analyze overdamped second-order systems is to factor the quadratic equa-
tion into two first-order terms each with its own time constant, s1 and s2:

1 1
TFðsÞ ¼ ¼    (7.36)
s2 þ 2dun s þ un
2 1 1
sþ sþ
s1 s2

These two time constants are the negative of the inverted roots of the quadratic equation as
given by Equation 7.32. In other words, s1 ¼ 1/r1 and s2 ¼ 1/r2. The second-order transfer

II. SYSTEMS
314 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

function can also have numerator terms other than 1, but the essential behavior does not
change and the analysis strategy does not change. Typical overdamped impulse and step
responses will be shown in the following example.
After the quadratic equation is factored, the next step is either to separate this function into
two individual first-order terms sþ1=s k1
1
þ sþ1=s
k2
2
using partial fraction expansion (see later
discussion), or find a Laplace transform that generally matches the unfactored equation in
Equation 7.36. If the numerator is a constant, then entry 9 in the Laplace Transform
Table (Appendix B) matches the unfactored form. If the numerator is more complicated, a
match may not be found and partial fraction expansion will be necessary.
Figure 7.8 shows the response of a second-order system to a unit step input. The step
responses shown are for an element with four different combinations of the two time con-
stants: s1, s2 ¼ 1, 0.2; 1, 2; 4, 0.2; and 4, 2 s. The time it takes for the output of this element
to reach its final value depends on both time constants, but is primarily a factor of the slowest
time constant. The next example illustrates the use of Laplace transform analysis to determine
the impulse and step responses of a typical second-order overdamped system.

7.3.6.2 Partial Fraction ExpansiondManual Methods

Partial fraction expansion is the opposite of finding the common denominator: instead of
trying to combine fractions into a single fraction, we are trying to break them apart. We want
to determine what series of simple fractions will add up to the fraction we are trying to
decompose. The technique described here is a simplified version of partial fraction expansion
that deals with distinct linear factors, that is, denominator components of the form (s  p).
Moreover, this analysis will be concerned only with single components, not multiple compo-
nents such as (s  p)2.

FIGURE 7.8 Typical step responses of a second-order system with real roots. These responses are termed
overdamped relating to the exponential-like behavior of the response. Four different combinations of s1 and s2 are
shown. The speed of the response depends on both time constants, but is dominated by the slower of the two.

II. SYSTEMS
 7.3 THE LAPLACE DOMAIN TRANSFER FUNCTION 315
Under these restrictions: the partial fraction expansion can be defined as:

NðsÞ k1 k2 k3
TFðsÞ ¼ ¼ þ þ þ. (7.40)
ðs  p1 Þðs  p2 Þðs  p3 Þ. s  p1 s  p2 s  p31

kn ¼ ðs  pn ÞTFðsÞjs¼pn (7.41)

Since the constants in the Laplace equation denominator will always be positive, the
values of p will always be negative. The next example uses partial fraction expansion to
find the step response of a second-order system.

EXAMPLE 7.6
Find the impulse and step responses of the following second-order transfer function. Use Laplace
analysis to find the time functions and MATLAB to plot the two time responses.
25
TFðsÞ ¼ (7.37)
s2 þ 12:5s þ 25
Solution, General: First find the values of d and un by equating coefficients with the basic
equation shown in Equation 7.31:

u2n ¼ 25; un ¼ 5
12:5 12:5
2dun ¼ 12:5; d ¼ ¼ ¼ 1:25
2un 10
Since d > 1, the roots are real. In this case, the next step is to factor the denominator using the
quadratic equation, Equation 7.326:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
b 1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi 12:5 1
r1 ; r2 ¼  b2  4c ¼  12:52  4ð25Þ ¼ 6:25  3:75
2 2 2 2
r1 ; r2 ¼ 10:0 and  2:5
Solution, Impulse Response: For an impulse function input, the output Vout(s) is the same as the
transfer function since the Laplace transform of the input is Vin(s) ¼ 1:
 
25 25 7:5
Vout ðsÞ ¼ ¼
ðs þ 10Þðs þ 2:5Þ 7:5 ðs þ 10Þðs þ 2:5Þ
The right hand term of Vout(s) has been rearranged to match entry #9 in the Laplace Transform
Table:

ga
5eat  egt where a ¼ 2:5 and g ¼ 10
ðs þ aÞðs þ gÞ
So vout(t) becomes:
 
25  2:5t 
vout ðtÞ ¼ e  e10t ¼ 3:33 e2:5t  e10t (7.38)
7:5

II. SYSTEMS
316 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

Solution, Step response: To find Vout(s), multiply the transfer function by the Laplace transform
of the step function, 1/s:
 
1 25 25
Vout ðsÞ ¼ ¼ (7.39)
s ðs þ 2:5Þðs þ 10Þ sðs þ 2:5Þðs þ 10Þ
With the extra s added to the denominator, this function no longer matches any in the Laplace
Transform Table. However, we can expand this function using partial fraction expansion into the
form:
k1 k2 k3
Vout ðsÞ ¼ þ þ
s s þ 2:5 s þ 10
Applying partial fraction expansion to the Laplace function of Equation 7.39, the values for p1, p2,
and p3 are, respectively: 0, 2.5, and 10, which produces the numerator terms k1, k2, and k3:

25  25
k1 ¼ ðs þ 0Þ  ¼ ¼ 1:0
sðs þ 2:5Þðs þ 10ÞS¼0 2:5ð10Þ

25  25
k2 ¼ ðs þ 2:5Þ  ¼ ¼ 1:33
sðs þ 2:5Þðs þ 10ÞS¼2:5 2:5ð2:5 þ 10Þ

25  25
k3 ¼ ðs þ 10Þ  ¼ ¼ 0:33
sðs þ 2:5Þðs þ 10ÞS¼10 10ð2:5  10Þ

This gives rise to the expanded version of Equation 7.39:

25 1 1:33 0:33
Vout ðsÞ ¼ ¼  þ (7.42)
sðs þ 2:5Þðs þ 10Þ s s þ 2:5 s þ 10
Each of the terms in Equation 7.42 has an entry in the Laplace Table. Taking the inverse Laplace
transform of each of these terms separately gives:

vout ðtÞ ¼ 1:0  1:33e2:5t þ 0:33e10t (7.43)

MATLAB is used as in the Example 7.4 to plot the two resulting time domain equations,
Equations 7.38 and 7.43 in Figure 7.9.
The next example reiterates the role of partial fraction expansion in the solution of the inverse
Laplace transform.
6
If you are challenged by basic arithmetic you can easily factor any polynomial using the MATLAB roots routine.
An example of using this routine to factor a fourth-order polynomial is given in Example 7.9.

7.3.6.3 Partial Fraction ExpansiondMATLAB

Not surprisingly, MATLAB has a routine to perform partial fraction expansion. The
routine also finds the roots of the denominator. (A MATLAB routine that only calculates
the roots of a polynomial is described in the next section.) To expand a ratio of polynomials,
you first define the numerator and denominator polynomials using two vectors: one that
specifies the numerator coefficients, the other the denominator coefficients. For example,
for the transfer function:
s2 þ 5s þ 15 s2 þ 5s þ 15
TFðsÞ ¼ ¼ 3
sðs2þ 12:5s þ 25Þ s þ 12:5s2 þ 25s

II. SYSTEMS
 7.3 THE LAPLACE DOMAIN TRANSFER FUNCTION 317

FIGURE 7.9 The impulse (dashed curve) and step (solid curve) responses of the transfer function used in
Example 7.7.

The vector defining the numerator would be b ¼ [1, 5, 15];, whereas the denominator
would be defined as a ¼ [1, 12.5, 25 0];. Note that since the constant term is missing in the
denominator, a zero must be added to the vector definition. Once these vectors are defined,
the partial fraction expansion terms are found using:

[numerators, roots, const] ¼ residue(b,a) % Partial fraction expansion

This routine gives the outputs:

numerators ¼
0.8667
-0.4667
0.6000

roots ¼
-10.0000
-2.5000
0

const ¼
[ ]

II. SYSTEMS
318 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

and leads to the expanded transfer function.

0:6 0:47 0:87

TFðsÞ ¼  þ
s s þ 2:5 s þ 10

The roots are the same as in Example 7.9 because the denominator equation is the same.
The basic components of the time domain solution found from the inverse Laplace transform
will also be the same. The numerator influences only the amplitude of these components.
This again shows why the denominator polynomial is called the “characteristic equation”
(Section 7.2.5.1).

EXAMPLE 7.7
Find the step response of the system having the following transfer function. Determine the
solution analytically and also use MATLAB’s residue routine.
s
TFðsÞ ¼ 2 4
s þ 15s þ 50
Solution, Analytical: Since In(s) ¼ 1/s, the output in Laplace notation becomes:

  s
1 4 :25
OutðsÞ ¼ ¼ 2 (7.44)
s s þ 15s þ 50
2 s þ 15s þ 50
Next, we evaluate the value of d by equating coefficients:

15 15
2dun ¼ 15; d ¼ ¼ pffiffiffiffiffi ¼ 1:06
2un 2 50
Since d ¼ 1.07, the roots will be real and the system will be overdamped. The next step is to factor
the roots using the quadratic equation, Equation 7.30:

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
15 1
r1 ; r2 ¼  152  4ð50Þ ¼ 7:5  2:5 ¼ 10:0; 5:0
2 2
:25
and the output becomes: OutðsÞ ¼
ðs þ 10Þðs þ 5Þ
The inverse Laplace transform for this equation can be found in Appendix B (#9) where g ¼ 10
and a ¼ 5. To get the numerator constants to match, multiply top and bottom by 10  5 ¼ 5:
 
0:25 5 5
OutðsÞ ¼ ¼ ð0:05Þ
5 ðs þ 10Þðs þ 5Þ ðs þ 10Þðs þ 5Þ
From the Laplace Transform Table we get the time function:

outðtÞ ¼ 0:05 e5t  e10t (7.45)

II. SYSTEMS
 7.3 THE LAPLACE DOMAIN TRANSFER FUNCTION 319
Solution, MATLAB: For those of us who are arithmetically challenged, this problem could be
solved using the MATLAB residue routine described in Section 7.2.6.3. First define the numerator
and denominator vectors: b ¼ 0.25; and a ¼ [1, 15, 50];. The MATLAB command:

[numerators, roots, const] ¼ residue(b,a) % Partial fraction expansion

Gives:

Numerators ¼
-0.0500
0.0500

Roots ¼
-10
-5

Const ¼
[ ]

This gives the expanded equation:

0:05 0:05
OutðsÞ ¼ 
s þ 5 s þ 10
Result: Taking the inverse Laplace transform of the two first-order terms leads to the same time
function found using the unfactored transfer function. Figure 7.10 shows out(t) plotted using
MATLAB code.

0.015

0.01
out(t)

0.005

0
–0.5 0 0.5 1 1.5 2 2.5 3
Time (sec)

FIGURE 7.10 The impulse response of the second-order system in Example 7.7.

II. SYSTEMS
320 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

7.3.6.4 Second-Order Processes With Complex Roots

If the damping factor, d, of a second-order transfer function is <1, then the roots of the
characteristic (i.e., denominator) equation are complex and the step and pulse responses
have the behavior of a damped sinusoid: a sinusoid that decreases in amplitude exponentially
with time (i.e., of the general form eat sin unt). Second-order systems that have complex
roots are said to be “underdamped,” a term relating to the oscillatory behavior of such
systems. This terminology can be a bit confusing. Underdamped systems respond with
damped sinusoidal behavior, whereas overdamped responses have double exponential
responses and no oscillatory behavior.
If the roots are complex, the quadratic equation is not factored as the inverse Laplace trans-
form can usually be determined directly from the Table. Inverse Laplace transforms for
second-order underdamped responses are provided in the Table in terms of un and d and
in terms of general coefficients (Transforms #13e17). Usually, the only difficulty in finding
the inverse Laplace transform to these systems is in matching coefficients and scaling the
transfer function to match the constants in the Table. The next example demonstrates the
solution of a second-order underdamped system.

EXAMPLE 7.8
In a simple biomechanics experiment, a subject grips a rotating handle that is connected through
an opaque screen to a mechanical system that can generate an impulse of torque. To create the
torque impulse, a pendulum behind the screen strikes a lever arm attached to the handle. The
subject is unaware of when the strike and the torque impulse it generates will occur. The resulting
rotation of the wrist in degrees is measured under relaxed conditions. The rotation can be repre-
sented by the following second-order equation (Equation 7.46). As shown in Chapter 14, such an
equation is typical of mechanical systems that contains a mass, elasticity, and friction or viscosity. In
this example, we solve for the angular rotation, q(t).
qðsÞ 150
TFðsÞ ¼ ¼ 2 (7.46)
TðsÞ s þ 6s þ 310
where T(s) is the torque input and q(s) is the rotation of the wrist in radians. The impulse
function has a value of s 2  102 newton-meters.
Solutions: The values for d and un are found from the quadratic equation’s coefficients.
pffiffiffiffiffiffiffiffi 6 6
un ¼ 310 ¼ 17:6 and d is : 2dun ¼ 6; d ¼ ¼ ¼ 0:17
2un 2ð17:6Þ
Since d < 1, the roots are complex.
The input T(s) was an impulse of torque with a value of 2  102 newton-meters, so in the
Laplace domain: T(s) ¼ 2  102. q(s) becomes:
 150 3
qðsÞ ¼ 2  102 ¼ 2 (7.47)
s2 þ 6s þ 310 s þ 6s þ 310
Two of the transforms given in the Laplace tables will work (#13 and #15): entry #13 is used here
with b ¼ 0 (although #15 is more direct):
 
at c  ba bs þ c
e sinðbtÞ þ b cosðbtÞ 5 2
b s þ 2as þ a2 þ b2

II. SYSTEMS
 7.3 THE LAPLACE DOMAIN TRANSFER FUNCTION 321
Wrist Response
8

2
θ(t) (deg)

–2

–4

–6

–8
0 0.5 1 1.5 2
Time (sec)

FIGURE 7.11 The response of the relaxed human wrist to an impulse of torque. This system is of second order,
underdamped so the impulse response is a damped sinusoid.

Equating coefficients with Equation 7.47 we get:

6 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
c ¼ 3; b ¼ 0; a ¼ ¼ 3; a2 þ b2 ¼ 310; b ¼ 310  a2 ¼ 310  9 ¼ 17:3
2
Substituting these values into the time domain equivalent on the left side of the Transform
Table gives q(t):

qðtÞ ¼ e3t ð0:173 sinð17:3tÞÞ ¼ 10e3t ðsinð17:3tÞÞ rad

Result: This response, scaled to degrees, is plotted in Figure 7.11 using MATLAB code.

7.3.7 Higher-Order Transfer Functions

Transfer functions that are higher than second-order can usually be factored into multiple
second-order terms, or a combination of second- and first-order terms. The math would be
tedious, so we turn to MATLAB for assistance. The MATLAB routine roots computes
the roots of a polynomial. The polynomial equation is defined by a vector, c, that contains
the polynomial’s coefficients beginning with the highest power in the polynomial and
continuing down to the constant. For example, the roots of the polynomial:

s4 þ 14s3 þ 25s2 þ 10s þ 35

are found using: roots([1, 14, 25, 10, 35]);.

II. SYSTEMS
322 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

If a root is real, it is represented by a first-order term. Complex roots are given as complex
conjugates and should be combined into second-order terms. This is achieved by multiplying
the roots together. For example, the polynomial above has four roots: two are real and two
are complex conjugates, 0.1431  j1.1122. We could combine these by multiplying
(s þ 0.1431 þ j1.1122) (s þ 0.1431  j1.1122). If we want to make our life easier and avoid
doing the complex arithmetic, we could use the MATLAB routine poly, which performs
the inverse of roots. So the command:

poly([0.1431 þ j*1.1122, 0.1431 e j*1.1122]);

gives the polynomial coefficients: 1.0, 0.2862, 1.2575 corresponding to the second-order
polynomial s2  0.2862s þ 1.2575.
The use of roots and poly to factor a fourth-order transfer function is illustrated in the next
example. After factoring, we find the magnitude and phase spectrum. This example uses
MATLAB to generate the plot, but the plotting could also have been done using the graphical
Bode plot methods developed in Chapter 6.

EXAMPLE 7.9
Factor the transfer function shown below into first- and second-order terms. Write out the
factored transfer function and plot the Bode plot (magnitude and phase). Use roots to factor the
numerator and denominator and use poly to rearrange complex congregate roots into second-order
terms. Generate the Bode plot from the factored transfer function.

s3 þ 5s2 þ 3s þ 10
TFðsÞ ¼
s4 þ 12s3 þ 20s2 þ 14s þ 10
Solution: The numerator contains a third-order term and the denominator contains a fourth-
order term. We use the MATLAB roots routine to factor these two polynomials and rearrange
them into first- and second-order terms, then substitute s ¼ ju and plot. Of course, we could plot the
transfer function directly without factoring, but the factors provide some insight into the system.
The first part of the code factors the numerator and denominator.

% Example 7.9 Find the Bode plot of a higher order transfer function.
%
num ¼ [1 5 3 10]; % Define numerator polynomial
den ¼ [1 12 20 14 10]; % Define denominator polynomial
n_root ¼ roots(num) % Factor numerator and display
d_root ¼ roots(den) % Factor denominator and display

The results of this program are:

n_root ¼ -4.8087
-0.0957 þ 1.4389i
-0.0957 - 1.4389i

II. SYSTEMS
 7.3 THE LAPLACE DOMAIN TRANSFER FUNCTION 323

d_root ¼ -10.1571
-1.4283
-0.2073 þ 0.8039i
-0.2073 - 0.8039i

The numerator factors into a single root and a complex pair indicating a second-order term. The
denominator factors into two roots and one complex pair. To combine two complex roots into a
single second-order term we use poly:

nun_sec_ord ¼ poly([-0.0957þj*1.4389,-0.0957-j*1.4389]) % 2nd-order numerator

den_sec_order ¼ poly([-0.2073þj*0.8039,-0.2073-j*0.8039]) % 2nd-order denominator

This gives the output:

nun_sec_ord ¼ 1.0000 0.1914 2.0797

den_sec_order ¼ 1.0000 0.4147 0.7893

Results: Combining these first- and second-order roots, the factored transfer function is written
as:

ðs þ 4:81Þðs2 þ 0:19s þ 2:08Þ

TFðsÞ ¼
ðs þ 10:2Þðs þ 1:43Þðs2 þ 0:41s þ 0:69Þ
The Bode plot can be obtained by substituting ju for s in this equation and evaluating over u
where now each of the terms is identifiable as one of the Bode plot elements presented in Chapter 6.
As mentioned, although we could use MATLAB to plot the unfactored transfer function, factoring
shows us the Bode plot primitives (see Table 6.2), providing insight into what the elements make up
the system. In this example, we use MATLAB to plot this function so there is no need to rearrange
this equation into the standard Bode plot format (i.e., normalized so that the constant terms are
equal to 1).

% Plot the Bode Plot

w ¼ 0.1:.05:100; % Define the frequency vector
TF ¼ (j*w þ 4.81).*(-w.^2þj*0.19*w þ 2.08)...
./((j*w þ10.2).*(j*wþ1.43).*(-w.^2þj*0.41*wþ0.79)); % Define TF
subplot(2,1,1);
semilogx(w,20*log10(abs(TF))); % Plot magnitude spectrum in dB
...labels...
subplot(2,1,2);
semilogx(w,angle(TF)*360/(2*pi)); % Plot the phase spectrum
...labels....

The resulting spectrum plot is shown in Figure 7.12.

II. SYSTEMS
324 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

FIGURE 7.12 The magnitude and phase frequency characteristic (i.e., Bode plot) of the higher-order transfer
function used in Example 7.9.

7.4 NONZERO INITIAL CONDITIONSdINITIAL

AND FINAL VALUE THEOREMS

7.4.1 Nonzero Initial Conditions

Although the Laplace analysis method cannot deal with negative values of time, it can
handle systems that have nonzero conditions at t ¼ 0. So one way of dealing with systems
that have a history for t < 0 is to summarize that history as an initial condition at t ¼ 0. To
evaluate systems with initial conditions, the full Laplace domain equations for differentiation
and integration must be used. These equations, Equations 7.5 and 7.7 are repeated here.

dxðtÞ
L ¼ sXðsÞ  xð0  Þ (7.48)
dt
2 3
Z T Z 0
1 1
L4 xðtÞdt5 ¼ XðsÞ þ xðtÞdt (7.49)
0 s s N

II. SYSTEMS
 7.4 NONZERO INITIAL CONDITIONSdINITIAL AND FINAL VALUE THEOREMS 325
In both cases, we add a term to the standard Laplace operator. In taking the derivative, the
term is a constant: the negative of the initial value of the variable; that is, x(0). With inte-
gration, the term is a constant divided by s where the constant is the integral of the past
history of the variable from minus infinity to zero. This is because the current state of an
integrative process is the result of integration over all past values.
So for systems with nonzero initial conditions, it is only necessary to add the appropriate
terms in Equations 7.48 and 7.49. An example is given in the solution of a one-compartment
diffusion model. This model is a simplification of diffusion in biological compartments such
as the cardiovascular system. The one-compartment model would apply to large molecules in
the blood that cannot diffuse into tissue and are only slowly eliminated. The compartment is
assumed to provide perfect mixing; that is, the mixing of blood and substance is instanta-
neous and complete.

EXAMPLE 7.10
Find the concentration, c(t), of a large molecule solute delivered as a step input to the blood
compartment. Assume an initial concentration of c(0) and a diffusion coefficient, K.
Solution: The kinetics of a one-compartment system with no outflow is given by the differential
equation:
dcðtÞ
V ¼ Fin  KcðtÞ
dt
where V is the volume of the compartment, K is a diffusion coefficient, and Fin is the input,
which is assumed to be a step change of a given amount A: Fin ¼ Am(t). Converting this to the
Laplace domain, Fin(s) ¼ A/s. Next we assume that the initial concentration of solute is c(0).
Substituting A/s for Fin(s), converting to Laplace notation, and applying the derivative operation
with the initial condition (Equation 7.48):
A
VðsCðsÞ  cð0ÞÞ ¼  KCðsÞ
s
Solving for the concentration, C(s):
A
CðsÞðVs þ KÞ ¼ Vcð0Þ þ
s
A
A Vcð0Þ V cð0Þ
CðsÞ ¼ þ ¼  þ
sðVs þ KÞ Vs þ K K K
s sþ sþ
V V

Taking the inverse Laplace transform:

A  
V  A 
cðtÞ ¼ 1  etK=V þ cð0ÞetK=V ¼ 1  etK=V þ cð0ÞetK=V
K K
V
Results: We can interpret the solution as consisting of two components: the exponentially
increasing concentration from an initial value of c(0) to A/K resulting from the step input and an

II. SYSTEMS
326 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

60

50

40 Step Response

C(t)
30

20
Initial Condition
Response
10

0
0 200 400 600 800 1000 1200
Time (sec)

FIGURE 7.13 The diffusion of a large molecule solute delivered as a step input into the blood compartment. The
molecule had an initial concentration of 25 g/mL. The solution (dark curve) can be viewed as having two components
(light curves): the exponentially increasing concentration due to the step input and the exponential decay of the initial
concentration. Perfect mixing in the blood compartment is assumed.

exponential decay of the initial value to zero. This second component is what we would see if the step
input is not present. Using typical values from Rideout (1991) of A ¼ 0.7 g, V ¼ 3.0 mL, K ¼ 0.01,
c(0) ¼ 25.0 g/mL, leads to the results for both components and their sum as plotted in Figure 7.13.

Nonzero initial conditions also occur in circuits where preexisting voltages exist or in
mechanical systems that have nonzero initial velocities, so you can look forward to more
examples of systems with nonzero initial conditions in Chapter 13.

7.4.2 Initial and Final Value Theorems

The time representation of a Laplace function is obtained by taking the inverse Laplace
transform using tables such as that found in Appendix B. But sometimes we are only looking
for the value of the function at the very beginning of the stimulus, t ¼ 0, or at its very end,
t / N. Two useful theorems can supply us with this information without the need to take
the inverse Laplace transform: the “Initial and Final Value Theorems.” These theorems
give us the initial and the final output values directly from the Laplace transform equation.
Recall that frequency and time are inversely related: f ¼ 1/T. For example, as the period of
a sine wave increases, its frequency decreases and vice versa. The Laplace variable, s, is com-
plex frequency so we might expect to find the value of a Laplace function at t ¼ 0 by letting
s / N. The Initial Value Theorem supports this idea, but we must first multiply X(s) by s,
then let s / N:

xð0þÞ ¼ lim xðtÞ ¼ lim sXðsÞ (7.50)

t/0 s/N

II. SYSTEMS
 7.4 NONZERO INITIAL CONDITIONSdINITIAL AND FINAL VALUE THEOREMS 327
The Final Value Theorem follows the same logic, but now since t / N, it is s that goes to 0.
The Final Value Theorem states:
xðNÞ ¼ lim xðtÞ ¼ lim sXðsÞ (7.51)
t/N s/0

The application of either theorem is straightforward as is shown in the following example.

EXAMPLE 7.11
Use the Final Value Theorem to find the final value of x(t) to a step input for the system whose
transfer function is given below. Also find the final value the hard way: by determining x(t) from the
inverse Laplace transform, then letting t / N.
:28s þ :23
TFðsÞ ¼ (7.52)
s2 þ 0:3s þ 2
Solution, Inverse Laplace Transform: First find the output Laplace function X(s) by multiplying
TF(s) by the step input function in the Laplace domain:
1 :28s þ 0:92
XðsÞ ¼ TFðsÞ ¼ (7.53)
s sðs2 þ 0:03s þ 2Þ
Next we find the full expression for x(t) from the inverse Laplace transform. We need to examine
d to determine if the roots are real or complex:
0:3 0:3
2dun ¼ 0:3; d ¼ ¼ pffiffiffi ¼ 0:106 < 1.
2un 2 2
So the system is underdamped and one of the transfer functions #13e17 in the Laplace Trans-
form Table should be used to find the inverse. The function X(s) matches entry #14 in the Laplace
Transform Table, but requires some rescaling to match the numerator.
 
ab bs þ a2 þ b2
1  eat sin bt þ b cos bt5  2
b s s þ 2as þ a2 þ b2

Considering only the denominator, the sum a2 þ b2 should equal 2, but then the numerator
needs to be rescaled so the numerator constant is also 2 (i.e., a2 þ b2 ¼ 2). To make the numerator
constants match we need to multiply the numerator in Equation 7.53 by: 2/0.92 ¼ 2.17. Multiplying
top and bottom by 2.17, the rescaled Laplace function becomes:
 
1 0:61s þ 2 0:46ð0:61s þ 2Þ
XðsÞ ¼ ¼
2:17 sðs2 þ 0:3s þ 2Þ sðs2 þ 0:3s þ 2Þ
Now equating coefficients with entry #14:
0:3 pffiffiffiffiffiffiffiffiffi
b ¼ :61; a ¼ ¼ 0:15; a2 þ b2 ¼ 2; b2 ¼ 2  0:152 ; b ¼ 1:98 ¼ 1:41
2
The inverse Laplace Transform becomes:
 
:15  :61
xðtÞ ¼ 0:46 1  e:15t sinð1:41tÞ þ cosð1:41tÞ
1:41

xðtÞ ¼ 0:46 1  e:15t ð  0:33 sinð1:41tÞ þ cosð1:41tÞÞ

II. SYSTEMS
328 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

Now letting t / N, the exponential term goes to zero and the final value becomes:
xðtÞ ¼ 0:46
Solution, Final Value Theorem: This approach is much easier. Substitute the output Laplace
function given in Equation 7.52 into Equation 7.51:

0:281s þ 0:92 0:28s þ 0:92 0:921

lim sXðsÞ ¼ lim s ¼ lim 2 ¼ ¼ 0:46
s/0 s/0 sðs2 þ 0:3s þ 2Þ s/0 s þ 0:3s þ 2 2
This is the same value that is obtained when letting t / N in the time solution x(t).
An example of the application of the Initial Value Theorem is found in the problems.

7.5 THE LAPLACE DOMAIN, THE FREQUENCY

DOMAIN, AND THE TIME DOMAIN

We know how to move between the Laplace domain and the time domain: take the Lap-
lace transform or its inverse. To move between the frequency domain and the time domain:
take the Fourier transform or its inverse. To move from the Laplace domain to the frequency
domain, we note that s is complex frequency, s þ ju, so we just do away with the real part, s,
and substitute ju for s. (We should also renormalize the coefficients so the constant term
equals 1, particularly if we are planning to use Bode plot techniques.) To make this transfor-
mation, we must assume that we are dealing with periodic steady-state signals only. Since we
usually work with systems in the Laplace domain, converting the other way, from the fre-
quency domain to the Laplace domain ( ju / s), is not common and we would also need
to assume zero initial conditions. These transformations are summarized in Table 7.1.
Some of the relationships between the Laplace transfer function and the frequency domain
characteristics have already been mentioned, and these depend largely on the characteristic
equation. A first-order characteristic equation gives rise to first-order frequency characteris-
tics such as those shown in Figures 6.13 and 6.14.
Second-order frequency characteristics, like second-order time responses, are highly
dependent on the value of the damping coefficient, d. As shown in Figure 6.17, the frequency
curve shows a peak for values of d < 1, and the height of that peak increases as d decreases.
This peak occurs at the undamped natural frequency, un.
The peaks in the frequency domain have dramatic correlates in the time domain. As illus-
trated in the next example, when d < 1 the response overshoots the final value, oscillating
around this value at frequency ud. This oscillation frequency is not quite the same as the un-
damped frequency, un, although for small values of d the oscillation frequency approaches un
(Equation 7.34). The undamped natural frequency, un, is the frequency that the system would
like to oscillate at if there was no damping, that is, if it could oscillate unimpeded. But the
decay in the oscillation caused by the damping lowers the actual oscillation frequency to
ud. As the damping factor, d, decreases, its influence on oscillation frequency, ud, is reduced,
so it approaches the undamped natural frequency, un, as described in Equation 7.34.
The next example uses MATLAB to compare the frequency and time characteristics of a
second-order system for various values of damping.

II. SYSTEMS
 7.5 THE LAPLACE DOMAIN, THE FREQUENCY DOMAIN, AND THE TIME DOMAIN 329

EXAMPLE 7.12
A system with the following transfer function has an undamped natural frequency of
10,000 rad/s and can have three different values of the damping factor: 0.05, 0.1, and 0.5.
1
TFðsÞ ¼
s2 þ 200ds þ 104
Plot the frequency spectrum (i.e., Bode plot) of the time domain transfer function and the step
response of the system for the three damping factors. Use the Laplace transform to solve for the time
response and MATLAB for calculation and plotting.
Solution, Time Domain: The step response can be obtained by multiplying the transfer function
by 1/s, then determining the inverse Laplace transform leaving d as a variable:
1
Vout ðsÞ ¼
sðs2 þ 2  102 ds þ 104 Þ
This matches entry #17 for d < 1, although a minor rescaling is required:
 
1 104
Vout ðsÞ ¼
10 4 sðs þ 200ds þ 104 Þ
2

The time domain solution is:

  pffiffiffiffiffiffiffiffiffiffiffiffiffi!
1 edun t  pffiffiffiffiffiffiffiffiffiffiffiffiffi  1  d2
1
vout ðtÞ ¼ 1  pffiffiffiffiffiffiffiffiffiffiffiffi2ffi sin un 1  d t þ q ; q ¼ tan
2
104 1d d

where u2n ¼ 104, un ¼ 100.

The equivalent time response will be programmed directly into the MATLAB code.
Solution: Frequency Domain: To find the frequency response, convert the Laplace transfer
function to a frequency domain transfer function by substituting ju for s and rearranging into
frequency domain format where the constant is normalized to 1.0:
1 1
TFðuÞ ¼ ¼
2
ð juÞ þ 200dju þ 104 1  104 u2 þ 0:02dju

This equation can also be programmed directly into MATLAB. The following is the resulting
program.

% Example 7.12 Comparison of time and frequency characteristics of a second-order

system
% with different damping factors.
%
w ¼ 1:1:10000; % Frequency vector: 1 to 10,000 rad/sec
t ¼ 0:.0001:.4; % Time vector: 0 to 0.4 sec.
delta ¼ [0.5 0.1 0.05]; % Damping factors
% Calculate spectra
hold on; % Plot superimposed
for k ¼ 1:length(delta) % Repeat for each damping factor

II. SYSTEMS
330 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

TF ¼ 1./(1 - 10.^-4*w.^2 þ j*delta(k)*.02*w); % Transfer function

Mag ¼ 20*log10(abs(TF)); % Take magnitude in dB
semilogx(w,Mag); % Plot semilog (log frequency)
end
...labels and title....
% Calculate and plot time characteristics
figure ; hold on; % Plot superimposed
wn ¼ 100; % Undamped natural frequency
for k ¼ 1:length(delta)
c ¼ sqrt(1-delta(k)^2); % Useful constant
theta ¼ atan(c/delta(k)); % Calculate theta
xt ¼ (1 - (1/c)*(exp(-delta(k)*wn*t)).*(sin(wn*c*t þ theta)))/1000; % Time
function
plot(t,xt);
end
...label, title, and text...

Results: The results are shown in Figures 7.14 and 7.15. The correspondence between the fre-
quency characteristics and the time responses is evident. When a system’s spectrum has a peak, the
system’s time response has an overshoot. Note that the frequency peak associated with a d of 0.5 is
modest, but the time domain response still has some overshoot. As shown here and in one of the
problems, the larger the spectral peak, the greater the overshoot. The minimum d for no overshoot
in the response is left as an exercise in one of the problems.

FIGURE 7.14 Comparison of magnitude spectra of the second-order system used in Example 7.12 having three
different damping coefficients: d ¼ 0.05, 0.1, and 0.05.

II. SYSTEMS
 7.6 SYSTEM IDENTIFICATION 331

x 10–3
2

1.8
δ = 0.5
1.6 δ = 0.1
δ = 0.05
1.4

1.2
Vout(t)

0.8

0.6

0.4

0.2

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Time (sec)
FIGURE 7.15 Comparison of time domain step responses for a second-order system with three different damping
factors. The magnitude spectrum for this system is shown in Figure 7.14.

7.6 SYSTEM IDENTIFICATION

Bioengineers often face complex systems with unknown internal components. In such
cases, it is usually impossible to develop equations for system behavior to construct the fre-
quency characteristics. However, if you can control the stimulus to the system, and measure
its response, you should be able to determine the system’s spectrum experimentally, pro-
vided that the system is linear or can be taken as linear. Once you determine the system’s
spectrum, its transfer function can be estimated using Bode plot methods from Chapter 6.
Then the system’s response to any input can easily be computed. Finding a system’s transfer
function from external behavior is called “system identification.”
The are several approaches to identifying a system if we can control, or at least have access
to, its input. If we can generate an impulse input, we can determine its spectrum from the im-
pulse response by taking the Fourier transform of the impulse response (see Section 5.3.2).
If the inputs are sinusoidal, or are decomposed into sinusoids, then we can estimate the
frequency characteristics by taking the ratio of output amplitude to input amplitude at each
frequency. That is, we transform the input (if needed) and output signals to the frequency
domain and divide Output(u) by Input(u) to get TF(u), as shown in Equation 6.23. This
approach works if we can measure the input signal as long as that signal contains energy
over the frequency range of the system.
Another powerful method that works as long as we can measure both input and output
signals is model-based simulation. In this approach, we construct a model we believe repre-
sents the system, then adjust the model parameters (or elements) until the inputs and outputs

II. SYSTEMS
332 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

match our data. The result is not just a transfer function that matches our system, but a
representation of possible internal biological elements. Simulation approaches are explored
in Chapter 9.
Here we use frequency-based methods to find the spectrum of several unknown systems,
then apply Bode plot primitives to estimate the transfer function. In the following three
examples of system identification, the first uses Fourier decomposition, the second uses
sinusoidal and impulse inputs, and the last example uses measurements of the input and
output signals.

EXAMPLE 7.13
Use white noise to estimate the magnitude spectrum of a system represented by MATLAB
routine y ¼ unknown_sys7_1(x).m. The input argument, x, is taken as the input signal and the
output argument, y, is the output signal. Plot the magnitude spectrum, then apply Bode plot
primitives to estimate the transfer function.
Solution: Since we have complete control of the input signal (not often the case in dealing with
biological systems), we use an input signal that contains energy over a broad range of frequencies.
As discussed in Chapter 1, a random white noise signal has equal energy over all frequencies or, for
digital signals, equal energy up to fs/2 (see Figure 1.13). With a random signal as our input, we take
the Fourier transform of both this input signal and the system’s output. We divide the two
frequency domain signals and plot the magnitude (in dB) as the spectrum of our transfer function.
As always, we only plot the meaningful spectral points below fs/2.
We initially select a sampling frequency of 1000 Hz. This gives us a spectrum that ranges be-
tween 0 and 500 Hz. As we have no knowledge of the system’s transfer function, we do not know if
this range is sufficient to cover the frequencies of interest. We should be prepared to adjust fs to
cover other frequencies if they are needed to define the transfer function. We use a signal with a
large number of points, N ¼ 10,000, to improve the resolution.

% Example 7.13 Use MATLAB to find the transfer function of an unknown

% system from its input and output signals.
%
fs ¼ 1000; % Assumed sample rate
N ¼ 10000; % Number of points
nf ¼ round(N/2); % Number of valid spectral points
f ¼ (0:N-1)*fs/N; % Frequency vector for plotting
t ¼ (1:N)/fs; % Time vector for plotting
x ¼ rand(1,N); % Input signal
y ¼ unknown_sys7_1(x); % Unknown system
X ¼ fft(x); % Convert input to freq. domain
Y ¼ fft(y); % Convert output to freq. domain
TF ¼ Y./X; % Find TF
Mag ¼ 20*log10(abs(TF)); % TF in dB
semilogx(f(1:nf),Mag(1:nf),'k'); % Plot magnitude, valid points only

Results: The spectrum produced by this code is shown in Figure 7.16. To convert these frequency
characteristics into transfer functions, we need to rely on the skills developed in the last chapter. The
spectrum looks like a combination of three Bode plot primitives: a second-order underdamped

II. SYSTEMS
 7.6 SYSTEM IDENTIFICATION 333
35
30

25
20
15
TF(ω) (dB)
10

5
0
–5

–10

–15
10–1 100 101 102 103
Frequency (Hz)

FIGURE 7.16 The spectrum of a signal from the unknown system used in Example 7.13 when the input is white
noise. Since white noise has energy at all frequencies, this spectrum is the same as the system’s spectrum and can be
used to estimate the transfer function.

element, an integrator (i.e., 1/ju), and an inverted low-pass filter (a numerator 1 þ ju/u1). In fact, it
looks a lot like the spectrum in Figure 6.25, which was derived from a transfer function having these
three elements. Thus the transfer function of this system has the general form:
u
1þj
u1 !
TFðuÞ ¼ K  2
u u
ju 1 þ þ j2d
un un

We can get the parameters for u1, un, and d from the spectral plot. The peak occurs at 50 Hz,
so un ¼ 2pf ¼ 2p(50) ¼ 314 rad sec. and the descending low frequency slope is within 3 dB of
leveling off at about 2 Hz so u1 ¼ 2pf ¼ 2p(2) ¼ 12.6 rad/sec. To find the baseline gain, K, note that
at f ¼ 0 Hz, the integrator element has a gain of 0.0 (Figure 6.11) as do the other elements, but the
spectrum has a gain of 20 dB. So K must be 10 (i.e., 20 dB). To find d, note that the peak is about 14
dB above the baseline, so:

20 logð2dÞ ¼ 14dB; logð2dÞ ¼ 0:7; 1=ð2dÞ ¼ 100:7 ; d ¼ :1999=2 ¼ 0:1

This gives the complete transfer function as:

1 þ j:08u
TFðuÞ ¼ 10  
ju 1  ð0:0032uÞ2 þ j0:00064u

If we ignore initial conditions: we can rewrite this transfer function in Laplace notation:
 
1
0:8s s þ
0:08 78125ðs þ 12:6Þ
TFðsÞ ¼   ¼ 2 þ 62:5s þ 97656Þ
0:00064s 1 sðs
0:00322 s s2 þ þ
0:00322 0:00322

II. SYSTEMS
334 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

White noise is not an easy stimulus to induce in most biological systems. Another way to
determine frequency response experimentally is to take advantage of the fact that a sinusoi-
dal stimulus into a linear system will produce a sinusoidal response at the same frequency.
By stimulating the biological system with sinusoids over a range of frequencies and
measuring the change in amplitude and phase of the response, we can construct a plot of
the frequency characteristics by simply combining all the individual measurements. This
approach is illustrated in the next example.

EXAMPLE 7.14
Find the magnitude of spectral characteristics of the process represented by
unknown_sys7_2(x).m. Use sinusoids to identify the spectrum and Bode plot primitives to estimate
the transfer function. Also determine the system spectrum from the impulse response and compare.
The actual magnitude spectrum in dB can be found as the second output argument of
unknown_sys7_2(x).m.
Solution: Generate a sinusoid with an RMS value of 1.0. This requires the amplitude to be 1.414.
Input this sinusoid to the unknown process, and measure the RMS value of the output. The RMS
value is usually a more accurate measurement of a signal value than the peak-to-peak amplitude as
it is less susceptible to noise-induced error. Repeat this protocol for increasing frequencies until the
output falls to very low levels (in this case up to 400 Hz). Plot the results in dB against log frequency.
Also construct an impulse signal and input it to the system. Take the Fourier transform of the
impulse response and plot in dB against log frequency. Finally, plot the actual system spectrum and
compare it with the two experimentally obtained spectra.
To find the transfer function from the spectrum, use the sinusoidal responses as they are likely to
be more accurate. Estimate asymptotes if need be and apply Bode techniques to determine the
transfer function.

% Example 7.14 Identify an unknown system unknown_sys7_2.

%
N ¼ 1000; % Input signal length
t ¼ (0:N-1)/N; % Time vector ( 1 sec)
for k ¼ 1:400 % Try frequencies up to 400 Hz
f(k) ¼ k; % Freq ¼ k ¼ 1-400Hz
x ¼ 1.414*cos(2*pi*f(k)*t); % Sinusoid with RMS ¼ 1.0
y ¼ unknown_sys7_2(x); % Input stimulus to process
Y(k) ¼ sqrt(mean(y.^2)); % Calculate RMS value as magnitude
end
Y ¼ 20*log10(Y); % Convert to dB
% Now use the impulse response for system identification
x1 ¼ [1.0, zeros(1,N-1)]; % Generate impulse response
[y1, spec] ¼ unknown_sys7_2(x1); % Get impulse response and true spectrum
Mag ¼ abs(fft(y1)); % Get magnitude spectrum
Y1 ¼ 20*log10(Mag(1:400)); % Convert to dB
semilogx(f,Y,'k','LineWidth',1); hold on; % Plot magnitude as dB/log f
semilogx(f,Y1,':k','LineWidth',2); % Plot as dB/log f
semilogx(f,spec(1:400),'--k','LineWidth',2); % Plot actual spectrum
...labels...

II. SYSTEMS
 7.6 SYSTEM IDENTIFICATION 335

FIGURE 7.17 Three spectra generated in Example 7.14: one from sinusoidal simulation (solid line), another from
the impulse response (dashed line), and a third showing the true spectrum (dotted line). Except at the lowest
frequencies, the three spectra overlap and are hard to distinguish.

Results: The three spectra are plotted superimposed in Figure 7.17 with labels. Note that the
spectrum generated from sinusoids (solid line) closely matches the true spectrum (dotted line). The
spectrum determined from the impulse response (dashed line) deviates slightly at the higher
frequencies. This is likely due to computational error at the low output amplitudes.
Based on the spectrum from sinusoidal stimulation, the process appears overdamped, as there
are no spectral peaks (Figure 7.18). At the higher frequencies, the slope is 40 dB/decade, indicating a
second-order system. The lower frequencies show a slope of 20 dB/decade. Applying Bode plot
techniques, we fit the curve with lines of 20 and 40 dB/decade (Figure 7.18 dashed lines).
From the spectrum of Figure 7.18, it looks like this system consists of two first-order elements
with cutoff frequencies somewhere around 10 and 80 Hz (or 63 and 503 rad/sec). The gain is around
0 dB or 1.0. So an estimate of the frequency domain transfer function of this system would be:
1:0 1:0
TFðuÞ ¼    ¼
ju ju ð1 þ j0:0159uÞð1 þ j0:002juÞ
1þ 1þ
63 503
Assuming no initial conditions, the transfer function in Laplace notation becomes:

1:0 3:14  104

TFðsÞ ¼    ¼
1 1 ðs þ 63Þðs þ 503Þ
ð0:0159Þð0:002Þ s þ sþ
0:0159 0:002

II. SYSTEMS
336 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

FIGURE 7.18 The magnitude spectrum of an unknown system that is represented in the routine
unknown_sys7_2.m. The spectrum was determined by stimulating the system with sine waves ranging in fre-
quency from 1 to 400 Hz. The stimulus sine waves all had root mean square (RMS) values of 1.0, so the RMS values of
the output indicate the magnitude spectrum of the unknown system.

When it is not practical to simulate biological systems, the system’s natural input can be
used as long as it contains energy covering the range of the system’s spectrum. To estimate
the Fourier transform, you take the Fourier transform of the output signal and divide it by the
Fourier transform of the input signal. The next example embodies this approach to identify a
biological system.

EXAMPLE 7.15
The data file bio_sys.mat contains the input and output signals of a biological system in var-
iables x and y, respectively. fs ¼ 150 Hz. This file also contains the true spectrum of the system in
variable spec. Determine if the input signal can be used to accurately determine the system’s
spectrum. If so, estimate that spectrum and then use Bode plot primitives to find the related transfer
function.
Solution: Use the Fourier transform to calculate both the input and output magnitude spectra,
then plot. Check to see if the input spectrum has energies out to a frequency range where the output
spectrum is considerably attenuated. In other words, ensure that a decrease in the output spectrum
is due to the system and not insufficient energy in the input spectrum. Then divide the magnitude
spectrum of the output by the input to get an estimate of the system’s spectrum. Use Bode plot
methods to estimate the transfer function.

II. SYSTEMS
 7.6 SYSTEM IDENTIFICATION 337
% Example 7.15 Identify a biological system from input/output data.
%
load bio_sys;
fs ¼ 150; % Sample frequency
N ¼ length(x); % Signal length
N_2 ¼ round(N/2); % Half signal length for fft
f ¼ (1:N)*fs/N; % Frequency vector for plotting
X ¼ abs(fft(x)); % Fourier transform of the input signal
Y ¼ abs(fft(y)); % Fourier transform of the output signal
.......linear plot of x and y, label, new figure.........
TF ¼ Y./X; % Calculate magnitude transfer function
TF_dB ¼ 20*log10(TF); % in dB
.......semilog plot, label.........

Results: Figure 7.19 shows the input and output spectra plotted as linear functions. Although the
energy in the input spectrum falls off at the higher frequencies, it does appear to have energy over
the range of system output frequencies except possibly at 30 and 60 Hz.
Since the input spectrum appears to contain sufficient energy over the frequency range of
interest, the ratio of output spectrum to input spectrum should give a reasonable estimate of the
system’s spectrum. The result of dividing the output spectrum by the input spectrum produces the
system’s spectrum estimate shown in Figure 7.20 (solid line). This estimated spectrum closely fol-
lows the actual spectrum (dashed line) except at the two frequency extremes. Despite the deviations
at the high and low frequencies, the estimated spectrum is sufficient to determine the transfer
function using Bode plot methods.

4.5
Input
4 Output
3.5

3
Magnitude

2.5

1.5

0.5

0
0 10 20 30 40 50 60 70 80
Frequency(Hz)

FIGURE 7.19 The magnitude spectra of input and output signals associated with an unknown biological system
in Example 7.15. The input spectrum is seen to decrease with increasing frequency, but still exceeds that of the output
spectrum except possibly at 30 and 60 Hz. This indicates that the attenuation seen in the output spectrum is not a
result of insufficient energy in the input spectrum.

II. SYSTEMS
338 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

0 Estimated
spectrum

Magnitude (dB)
–5

–10

–15 Actual spectrum

–20
10–2 10–1 100 101 102
Frequency(Hz)
FIGURE 7.20 An estimate of system spectrum obtained by dividing the output spectrum by the input spectrum
(solid line) and the system’s actual spectrum (dotted line).

The system spectrum in Figure 7.20 has the shape of a band-pass filter with cutoff frequencies
around 0.2 and 5 Hz corresponding to 1.26 and 31.4 rad/sec. The slope on either side appears to be
20 dB/decade and the midrange gain is near 0 dB or 1.0. Applying Bode plot techniques with this
type of curve (see Example 6.11 and Equation 6.56) with these parameters leads to an estimated
transfer function.
ju ju ju
TFðuÞ ¼    ¼    ¼
ju ju ju ju ð1 þ j0:794uÞð1 þ j0:032uÞ
1þ 1þ 1þ 1þ
u1 u2 1:26 31:4
Again assuming no initial conditions, the Laplace transfer function is:
s 39:4 s
TFðsÞ ¼ ¼
ð0:794Þð0:032Þðs þ 1:26Þðs þ 31:4Þ ðs þ 1:26Þðs þ 31:4Þ

If you can control, or at least measure, the stimulus and response of a system, the ap-
proaches used here can be very useful. Variations of both impulse and frequency response
methods have been used to estimate the transfer function of the extraocular muscles, the
iris, and lens muscles in the eye, and the response of chemoreceptors in the respiratory system
and numerous other biosystems.

7.7 SUMMARY
With the Laplace transform, all of the analysis tools developed in Chapter 6 can be applied
to systems exposed to a broader class of signals. Transfer functions written in terms of the

II. SYSTEMS
 PROBLEMS 339
Laplace variable s (i.e., complex frequency) serve the same function as frequency domain
transfer functions, but now include transient signals such as the step function. Here only
the response to the step and impulse signals is used in examples because these are the two
stimuli that are most commonly used in practice. Their popularity stems from the fact that
they provide a great deal of insight into system behavior, and they are usually easy to
generate in practical situations. However, responses to other signals such as ramps or expo-
nentials, or any signal that has a Laplace transform, can be analyzed using these techniques.
Laplace transform methods can also be extended to systems with nonzero initial conditions, a
useful feature explored later.
The Laplace transform can be viewed as an extension of the Fourier transform where com-
plex frequency, s, is used instead of imaginary frequency, ju. With this in mind, it is easy to
convert from the Laplace domain to the frequency domain by substituting ju for s in the Lap-
lace transfer functions. Bode plot techniques can be applied to these converted transforms to
construct the magnitude and phase spectra. Thus the Laplace transform serves as a gateway
into both the frequency domain and the time domain through the inverse Laplace transform.
Determining a system’s transfer function from external behavior is a broad and expanding
area of signal processing known as systems identification. If it is practical to generate a sinu-
soidal or impulse input to a system and measure the response, we should be able to deter-
mine the spectral characteristics of that system. Alternatively, we can use the system’s own
natural stimulus as long as we can measure that stimulus and it contains energy of the spec-
tral frequencies of interest. From the system’s spectrum, we can use Bode plot methods devel-
oped in Chapter 6 to reconstruct the system’s transfer function.

PROBLEMS

1. Find the Laplace transform of the following time functions:

a.
20
(A)
18

16

14

12
Volts

10

0
0 1 2 3 4 5
Time (sec)

II. SYSTEMS
340 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

b.

(B)
1

0.8

Volts 0.63
0.6

0.4

0.2

0
0 1 2 3 4 5
Time (sec)

c. (e2t  e5t)
d. 2e3t  4e6t
e. 5 þ 3e10t
2. Find the inverse Laplace transform of the following Laplace functions:
a. s 10
þ5
b. 10
sðs þ 5Þ

c. 5s þ 4 (Hint: Check roots.)

s2 þ5sþ20
d.  2 5s þ 4
s s þ 5s þ 20
3. Use partial fraction expansion to find the inverse Laplace transform of these functions:
a. 2s þ 4
s þ10sþ10
b.  2 10
s s þ 4s þ 3
4. Find the time response of the following system if the input is a step from 0 to 5:

In + 2 Out
Σ s
−

0.5

II. SYSTEMS
 PROBLEMS 341
5. Find the time response of the following system to a unit step function. Use Laplace
methods to solve for the time response as a function of k. Then use MATLAB to plot
the time function for k ¼ 0.1, 1, and 10.

In + 1 Out
Σ k
s
−

6. Find the time response of the following two systems if the input is a step from 0 to 8.
Use MATLAB to plot the time responses. Plot superimposed.

(A) (B)

In + Out In + Out
k
Σ S Σ S
− −

7. Solve for the Laplace transfer function of the following system where k ¼ 1. Find the
time response to a step from 0 to 4 and an impulse having a value of 4. (Hint: You can
apply the feedback equation twice.)
+ +
In 1 10 Out
Σ Σ
S S
− −

8. Find the time response of the system with the following transfer function if the input
is a unit step. Also find the response to a unit impulse function. Assume k ¼ 5.
+
1 5
In Σ s
Out
s+k
−

II. SYSTEMS
342 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

9. Repeat Problem 8 assuming k ¼ 0.2.

10. Find the unit impulse response to the system in Problem 7 for k ¼ 1 as on Problem
7 and k ¼ 0.1. Use MATLAB to plot the two impulse responses. How does decreasing
k change the output behavior?
11. Find the impulse response to the following system for k ¼ 1 and k ¼ 0.1. Use MATLAB
to plot the two impulse responses. How does decreasing k change the output behavior?
Compare with the results in Problem 10 to the same values of k.
+
In 10 k Out
Σ s+1 s
−

12. Use MATLAB to plot the magnitude spectra to the two systems shown in Problem 6.
Plot superimposed.
13. Use MATLAB to plot the magnitude spectra of the system in Problem 10 to the two
values of k. Repeat for the system in Problem 11.
14. Use Laplace analysis to find the transfer function of the following system that contains
a time delay of 0.01 sec (e0.01s). Find the Laplace domain response of a step from 0 to
10. Then convert to the frequency domain and use MATLAB to find the magnitude
and phase spectrum of the response. Note that such a time delay is typical in biolog-
ical systems. Plot the system spectrum over a frequency range of 1 to 200 rad/sec.
Owing to the delay, the phase curve will exceed -180 deg and will wraparound, so use
the MATLAB unwrap routine.
+
In k Out
Σ e−.01s s
−

15. Demonstrate the effect of a 0.2-sec delay on the frequency characteristics of a second-
order system. The system should have an un of 10 rad/sec and a d of 0.7. Plot the
magnitude and phase with and without the delay. (Hint: Use MATLAB to plot the
spectrum of the second-order system by substituting ju for s. Then replot adding
an e0.2s (¼ej0.2u) to the transfer function.) Plot for a frequency range of 1 to 100
rad/sec. Again the unwrap routine should be used since the phase plot will
exceed -180 deg.

II. SYSTEMS
 PROBLEMS 343
16. Find the time function of the following higher-order Laplace function. Use roots to fac-
tor the denominator (and poly if needed). Then apply partial fraction expansion to
separate out the denominator terms and find the inverse Laplace transform.

s3 þ 17s2 þ 80s þ 100

TFðsÞ ¼
s4 þ 10s3 þ 45s2 þ 110s þ 104

17. Find the time function of the following higher-order Laplace function. Use roots to
factor the denominator (and poly if needed). Then apply partial fraction expansion to
separate out the denominator terms and find the inverse Laplace transform. Alterna-
tively, use MATLAB’s residue to find the partial fractions directly,

3ðs þ 5Þ
TFðsÞ ¼
s3 þ 6s2 þ 11s þ 6

18. The impulse response of a first-order system is:

0 1
 
1 BB 1 C
C
Vout ðsÞ ¼
s @ 1A
sþ
s

Use the Initial Value Theorem to find the filter output’s value at t ¼ 0 (i.e. vout(0) )
for the filter.
19. The transfer function of an electronic system has been determined as:

5s þ 4
TFðsÞ ¼
s2 þ 5s þ 20

Use the Final Value Theorem to find the value of this system’s output for t / N if
the input is a step function that jumps from 0 to 5 at t ¼ 0.
20. The MATLAB file unknown_sys7_4(x).m found on the associated files represents a linear
system as in Examples 7.13 and 7.14. The input is x and the output is the output argu-
ment, i.e., y ¼ unknown_sys7_4(x);. Assume a sampling frequency of 1.0 kHz and use
40,000 points to get good spectral resolution. Determine the magnitude spectrum for
this unknown process using a random input as in Example 7.13. Estimate the transfer
function of this system based on Bode plot primitives.
21. Find the magnitude spectrum of the unknown system as represented by
unknown_sys7_5.musing sinusoids as in Example 7.14. Vary the range of frequencies of
the sine wave between 0 and 400 Hz in 1.0 Hz intervals. Make N ¼ 1000 and assume
fs ¼ 1 kHz. Estimate the transfer function for this system based on Bode plot
primitives.

II. SYSTEMS
344 7. LINEAR SYSTEMS IN THE COMPLEX FREQUENCY DOMAIN: THE LAPLACE TRANSFORM

22. Use the impulse response to find the magnitude spectrum of the system represented
by unknown_sys7_6(x).m. Estimate the transfer function for this based on Bode plot
primitives.
Use an impulse input of 1000 points and assume they are spaced 1.0 msec apart. As
always, be sure to plot only valid spectral points.
23. The file bio_sys1.mat contains the input and output signals of a biological system
sampled at 1000 Hz. Follow the approach used in Example 7.15 to find the magnitude
transfer function, then use Bode plot primitives to estimate the transfer function. Ignore
obvious artifacts.

II. SYSTEMS