Sec. 1.

2 State-Determined Systems 5

STATE-DETERMINED SYSTEMS
System Dynamics
Fundamental to system dynamics is the interaction between a system and its environment. In
the broadest context of system dynamics, a system and its environment are defined as abstract
entities:
ISystem:
prepared these of
A collection notes onthoughts,
matter, systemordynamics, basedwithin
concepts contained a real or 2.1
primarily First law of thermodynamics mass: kinetic energy stored as massive translating ele-
on Rowell and Wormleys System Dynamics: An Introduction
imaginary boundary.
ments,
[1], for my own use and share it with others with the under- Defining power P as the positive time rate of change of the total spring: potential energy stored as elastic deformation of
Environment: All that is external to the system.
standing that the contents are not meant to be original work energy stored in the system E, springlike elements, and
of my own,betweenbut merely damper :energy dissipated through friction to heat.
The interaction a system unofficial notes istaken
and its environment from inRowell
characterized terms ofand
a set
Wormley. That means the overall structure, some phrases, and
of system variables, as illustrated in Fig. 1.3, which in engineering systems may be time P(t) = dE/dt. (1) Note that the typical linear relationships that we will see below
most figures are copied directly from [1]. There is no guarantee
varying physical quantities such as forces. voltages, or pressures or mathematical variables are ideal, and that real springs and friction (damping) behave
that the contents are true to the book, however, so use it at your
with no direct physical context. These variables may be internal to the system and reflect We assume that this energy exchange (or power flow ) between differently.
own
lhe state risk. Pleaseforsend
of an element, anytheerrors
example, and ontypos
force acting toorpiconer@uw.edu.
a spring, they might express the the system and the environment occurs through a finite number Ideal sources (which can give infinite power) are:
lime variation of some quantity at the interface between the system and its environment. It of ports.
is useful to define two important classes of system variables: force source: in which supplied force Fs (t) is designated
1 Inputs:
Basics and the resulting velocity is a function of the system, and
An input is a system variable that is independently prescribed, or defined, by velocity source: in which supplied velocity Vs (t) is desig-
the system's environment. The value of an input at any instant is independent of the nated and the resulting force is a function of the system.
1.1systemState-determined systems[1,
behavior or response . Inputs define p 5]of the system and
the external excitation
can be quantities such as the external wind force acting on a tall building system or
System Dynamics
the rainfall can
fOmling the inputbe thought
flow of assystem.
into a reservoir the study ofmay
A system thehave
inter-
more
2.2.2 Rotational
action between
than one input. a system and its environment. System vari- Here the power flow P(t) is the product of the power flow vari-
ables characterize the system and can be considered inputs
Outputs: An output is defined as any system variable of interest. It may be a variable ables: angular velocity (t) and torque T (t) about a fixed axis,
(those arising from interaction with the environment) and out-
measured at the interface with the environment or a variable that is internal to the so
putssystem
(those of interest).
and does not directly interact with the environment. P(t) = T (t)(t). (5)
This power flow is a result of work W being performed on (+) Mechanisms for energy storage and dissipation are:
------------- or by () the system and heat energy H flowing in (+) or out rotational inertia: kinetic energy stored as massive rotat-
Environment
., .. .. .... .... () of the system: ing elements,
/
., .... , rotational spring: potential energy stored as elastic angu-
I
/ '\
\ Pdt = W + H (2) lar deformation of springlike elements, and
[
Inputs Outputs \ rotational damper : energy dissipated through friction to
Dynamic system
/
I heat.
/ Conservation of energy can be expressed as the sum of the n
.... .... .,
/
power flows into the system being equal to the time rate of Of interest here is the rotational moment of inertia (see [1, p
.... System bounda~ _..- ..- .. change of the sum of the energy stored in the m system ele- 34] for J of some shapes),
-- ------------ -- ments:
n m
X X dEj n Z
Figure 1.3: Schematic representation of a dynamic system. Pi (t) = . (3) X
A state-determined system model is a mathematical descrip- dt J= mi ri2 J = r2 dm. (6)
i=1 j=1 V
tionThe(state
identification of a system
equations) of aandsystem
inputs and outputsbehavior
whose may be illustrated
for allbyt consid
t0 i=1
canthebedesign
ering of an automobile
determined givensuspension. The conditions
the initial goal is to achieve both good
of the systemhandling
at This is valid for lumped-parameter systems comprised of lumped-
t = t0 and the inputs for all t t0 ( the system forgets). good
characteristics, to ensure safe operation during cornering and driving maneuvers, and parameter elements that represent the behavior of a region of Ideal sources (which can give infinite power) are:
ride comfort while traversing bumpy roads . Suspension design requires a trade-off in the se the system that is considered to have somewhat uniform behav-
lectionState variables
of the stiffness of the are a minimum
springs and damping set ofofvariables
effects that uniquely
the shock absorbers, to achieve ior (e.g. a cars velocity as opposed to the velocity of each point torque source: in which supplied torque Ts (t) is desig-
define the system response for all t. They must be independent nated and the resulting angular velocity is a function of
the good handling (relatively stiff suspensions) associated with high-performance cars, and
and complete. on the car), and can be described by ODEs. This is opposed to the system, and
the good ride quality (relatively soft suspensions) associated with more conventional cars. spatially continuous or distributed systems, which are described
Elements of systems are the primitive blocks that supply, angular velocity source: in which supplied angular velocity
store, and dissipate energy and together build systems. A state- by PDEs. s (t) is designated and the resulting torque is a function
determined system model is created by identifying proper ele- of the system.
ments and their interactions.
2.2 Mechanical system elements[1, p 21]
2.2.1 Translational 2.3 Electrical system elements[1, p 37]
2 System elements: one-port[1, p 19] Here the power flow P(t) is the product of the power flow vari- Here the power flow P(t) is the product of the power flow vari-
ables: velocity v(t) and collinear force F (t), so ables: current i(t) and voltage drop v(t), so
One-port elements represent the spatial locations (ports) where
energy is transfered. These can generate, store, or dissipate P(t) = F (t)v(t). (4) P(t) = i(t)v(t). (7)
energy in (but not between) the following energy domains: me-
chanical, electrical, fluid, and thermal. Mechanisms for energy storage and dissipation are: Mechanisms for energy storage and dissipation are:
 capacitor : electrical energy stored as charge q in an elec- 3.1 Through- and across-variables Variable units and definitions
trostatic field, v generalized across
inductor : magnetic energy stored in a magnetic field, and v m/s velocity difference

across
resistor : energy dissipated through material resistivity to rad/s angular velocity difference
heat. v V voltage drop
Through variables are continuous between the two terminals of
Ideal sources (which can produce infinite power) are: a system element, and can only be measured by breaking the P N/m2 pressure difference
current source: in which supplied current Is (t) is desig- system and inserting a sensor (e.g. force and current). T K temperature difference
nated and the resulting voltage is a function of the system, f generalized through
and F N force

through
Across variables are relative quantities because they are
voltage source: in which supplied voltage Vs (t) is desig- measured as the differences between values at each terminal of T Nm torque
nated and the resulting current is a function of the system. the element (e.g. velocity and voltage). i A current
Q m3 /s volume flow rate
2.4 Fluid system elements[1, p 44] Generalized through- and across-variables are defined as fol- q W heat flow rate
lows. x generalized integrated across
Here the power flow P(t) is the product of the power flow vari- x m linear displacement

across
ables: fluid volume flowrate Q(t) and pressure drop P (t) across rad angular displacement
the port, so Vs flux linkage
P(t) = P (t)Q(t). (8)

R
N s/m2 pressure difference momentum
Mechanisms for energy storage and dissipation are: h generalized integrated through

through
fluid capacitor : potential energy stored in the fluid, p Ns momentum
fluid inertance: kinetic energy stored in the fluid, and h Nms angular momentum
fluid resistor : energy dissipated through fluid work to q As charge
heat.
V m3 volume

R
Generalized variables
Ideal sources (which can give infinite power) are: H J heat
across v
flow source: in which supplied volume flow Qs (t) is desig-
x = 0t vdt + x(0)
R
nated and the resulting pressure is a function of the sys- integrated across
tem, and through f 3.2 A-, T-, and D-type elements
pressure source: in which supplied pressure Ps (t) is des-
h = 0t fdt + h(0)
R
ignated and the resulting volume flow is a function of the integrated through
3.2.1 A-type energy storage elements
system. power passing into (non-
P(t) = fv A-type elements are those in which the energy stored in the
thermal) element
element is a function of the across-variable and are called gen-
2.5 Thermal system elements[1, p 53] work done by system on ele- W = R T Pdt = R T fvdt eralized capacitances, which, when linear, are represented by C.
ment for 0 t T 0 0
Here the power flow P(t) is the not the product of the power See [1, p 71].
flow variables: temperature T (t) and heat flow rate q(t) (the
time derivative of heat or thermal energy H), but A-type elements elementary relationships
Element Constitutive Elemental Energy
dH(t) Equation Equation
P(t) = q(t) = . (9)
dt general- dv
ized h=Cv f = C dt E = 12 C v2
Mechanisms for energy storage and dissipation are: Across- and through-variables by energy domain
trans. p = mv F = m dv E = 12 mv 2
thermal capacitor : thermal energy stored as heat H, the
R R
System across through across through mass dt
thermal energy, and
resistor : energy dissipated through transferring heat by General v f x h rot. in- h = J T = J d E = 21 J2
conduction, convection, and radiation. Transla- ertia dt
tional v F x p elec.
Ideal sources are: cap. q = Cv i = C dv
dt
E = 12 Cv 2
heat flow source: in which supplied heat flow Qs (t) is des- Rota- T h
tional fluid V = Cf P Q = Cf dP E = 12 Cf P 2
ignated and the resulting temperature is a function of the cap. dt
system, and Electric v i q
temperature source: in which supplied temperature Ts (t) Fluid P Q V thermal H = Ct T q = Ct dT E = Ct T
is designated and the resulting heat flow is a function of cap. dt
the system. Thermal T q - H

3.2.2 T-type energy storage elements

3 Generalized one-port elements[1, p 66]
T-type elements are those in which the energy stored in the el-
Here we describe a generalization of the system elements of Sec- ement is a function of the through-variable and are called gen-
tion 2 in order to use a unified method (linear graphs) to analyze eralized inductances, which, when linear, are represented by L.
systems of the different energy domains. See [1, p 71].
2.4 Fluid system elements system dynamics page 2 of 26
 T-type elements elementary relationships or across- variable is defined, the other must be determined by 3.5.3 D-type element linearization
Constitutive Elemental the elemental equation.
Element Equation Equation Energy If the system-defined or given variable must be differenti- A D-type element has an algebraic relationship v = F (f), which
ated in the elemental equation to obtain the other unknown can be linearized and re-written in terms of perturbed variables
general- x = Lf df
v = L dt E = 12 L f2 variable, the element is in derivative causality, whereas if the v = v v0 and f = f f0 , where v0 and f0 are the equilibrium
ized system-defined or given variable must be integrated to obtain values about which the linearization is being performed, as
trans. 1 1 dF 1 the other unknown variable, the element is in integral causality.
spring x= K F v= K E = 2K F2 All energy storage elements have either derivative or integral
dt
causality, while all dissipative elements have algebraic causality, v = R f (14a)
tors. = K1 T = K1 dT E = 2K 1
T2 since the elemental equations are merely algebraic.
spring r r dt r where 
elec. in- di dF (f) 
duc. = Li v = L dt E = 12 Li2 R = . (14b)
3.5 Linearization of nonlinear elements df f=f0
fluid in- = If Q P = If dQ E = 12 If Q2
ert. dt To approximate the behavior of a nonlinear element about an So we have a linearized elemental equation
80 Summary of One-Port Primitive Elements operating point, perform a Taylor-series expansion of the consti-
tutive equation about the operating point, re-define the across-
3.2.3 D-type energy storage elements and through-variables as perturbed variables (e.g. v = v v0 ), v = R f (15)
3.3.4 Ideal Sources
D-type elements are those in which the energy dissipated and and through out second- and higher-order terms to obtain the
theenergy
In each element
domainhas
two an algebraic
general relationship
types of idealized sourcesbetween across and
may be defined: perturbed constitutive equation. [1, p 84] 4 System
Sec. 4.2
models: linear graphs[1, p 92]
Linear Graph Representation of One-Port Elements 93
through variables and are called generalized resistances, which,
when linear, are represented by R. See [1, p 71]. Linear graphs are representations r ] N O d e of lumped-parameter systems
3.5.1 A-type element linearization
The ideal across-variable source in which the generalized across-variable is
fied function of time f (r),
D-type elements elementary relationships
constructed from branches:
ments connect and define points
Across-

An A-type element has a single-valued monotonic constitutive or source system elements, and nodes:ofFigure
Through
energy
'''',bI,
where
"Or )
ports that represent passive
points where
4.1: Linear
adistinct
graph system
representation ele-
single passive across-variables
element as a directed
Element Elemental Vs(t) = f(l)
Equations Power Dissipated relationship h = F (v), which can be linearized and re-written Branch line segment.
can be measured w/respect to the systems reference node: a
general- in terms of perturbed variables h = h h0 and v = v v0 ,
ized
and f =
is independent of the
1
v v =
R through-variable
Rf P = R1 v2 = R f2 where h0 and v0 are the equilibrium values about which the node chosen to reference other across-variables to (e.g. for me-
ARepresentation
typical completeoflinear graph,Elements
representing a simple mechanical system with a single
linearization is being performed, as Sec. 4.2 Linear Graph
chanical systems One-Port
it is typically an inertial reference frame). 93
source and three one-port elements, is shown in Fig. 4.2. In this case there are two nodes
trans.
The ideal through-variable
F = Bv source vin=which 1 the generalized through-variablt
F P = Bv 2 = B 1 2
F representing points in the system at which distinct velocities may be measured. In practice it
damper B is common, but not necessary, to designate one of the nodes as a reference node and to draw
specified function of time h = C v (10a) this node as r ] N O line
a horizontal de (sometimes cross-hatched) as shown. In mechanical systems the
rot. 1
P = Br = B T 2
2 1
damper
elec.
and
T = Br

is independenti =
resist.
1
of the v
=
Fs(t) = f(t)
Br

v = Ri
R across-variable.
T
1 2
P = R v = Ri2
r
where 
dF (v) 
reference
"Or )'''',bI,
Across-
node is usually Through
selected to be the velocity of the inertial reference frame, while in
electric systems it commonly represents the system "ground" or zero-voltage point. In fluid
systems the reference node designates Figure 4.1: pressure
the reference Linear graph
of a singleApart
from which all system pressures are measured. passive
from
(oftenrepresentation
element
atmospheric pressure)
as a directed
this special interpretation the

C = . (10b) Branch line segment.
reference node behaves identically to all other nodes in the graph.
fluid Q = R1 P P = Rf Q P = R1 p2 = Rf Q2 dv v=v0
resist.
An example f
of a through-variable f
source is an idealized positive displacement pump i
v
thermal
fluid system, 1 rate is a prescribed function none:
in which qthe=flow T T = Rt q of time impedes
and heat
is independenl So we have a linearized elemental equation A typical complete linear graph, representing a simpleIdeal mechanical
source system Node
with a single
resist. Rt flow
the pressure required to maintain the flow, while an example of an across-variable soure. source and three one-port elements, is shown in Fig. 4.2. element
In this case there are two nodes
a regulated laboratory electric power supply in which the output voltage is independen1 dv representing points in the system at whichmdistinct velocities may be measured. Kim In practice it
3.3 drawn
the current Ideal
by the sources
circuit to which it is connected. These ideal sources are not pOll f = C . is common,(11)but not necessary, to designate one of the nodes as a reference /~node Branchesand to draw
representing
or energy-limited and theoretically may supply infinite power and energy.
dt , passi ve elements
this node as a horizontal line (sometimes cross-hatched) as shown. In mechanical systems the
Ideal sources provide the appropriate across- or through-variable
Theasymbols
as fortheof
function ideal sourceswhile
time, are shown Fig. 3.11,variable
thein other where in the through-varia
depends on reference node is usually selected to Figure
be the4.2:velocity ofrepresentation
Linear graph the inertial of areference frame,
simple mechanical while in
system.
sourcethe
the arrow
systemdesignates the assumed
to which positive direction
the source of through-variable
is connected. Note in flowthe
an( 3.5.2 T-type element linearization Noterepresents
electric systems it commonly that thetheelements and methods
system "ground" of linear
or zero-voltage point.graphs
In fluid do not
figure thatsource
the across-variable the for through-variables,
the arrow designates the assumedthedirection
arrow ofpoints in the
the across-varia
assume the elements to be linear. Each branch is associated with
assumed-positive
decrease direction
or drop. For each source type oneofvariable
through-variable flow
is an independently and funct
specified for A T-type element has a single-valued monotonicsystems the reference
constitutive annode designates
elemental
duceddifference
in Chap.
the reference
In a linearthrough-variable
graph pressure
one-port elements are
and (often
represented atmospheric
its across-variable pressure)
in the two-terminal form intro
is defined
are3. Each element generates a across
branch
thisin the graph and isat
drawn as a line segment
across-variables, the arrow points in the assumed direction of relationship x = F (f), which can be linearized and from
re-written in as the
which all system pressures between
measured. Apart thefrom variable
special each
interpretation
between the two appropriate nodes. Associated with each branch is an elemental through
node.
the
of time. reference node behaves identically to all other nodes in the graph.
across-variable decrease or drop. [1, p 80] terms of perturbed variables x = xx0 and f = ff0 , where x0 variable, assumed to pass through the line segment, and an elemental across-variable which
and f0 are the equilibrium values about which the linearization 4.1is the Linear graph
difference between representations
the across-variable values at the two nodes.ofEachone-port
linear graph
2 + '"' 2 is being performed, as branch thus represents the functional relationship between its across- and through-variables
elements v
as defined by the elemental equation. Linear graph segments may be used to represent pure
Ideal source Node
x = L f (12a) Theor generalized
ideal elements. element
ideal elemental equations are
t) F,(r) t) V,(r)
m A-type element capacitance KimC: dt ELEMENTS
LINEAR GRAPH REPRESENTATION OF ONE-PORT
= Cf dv 1
where T-type element inductance /~ df
L:Branches 1
= representing
v, and
 dt L
dF (f)  Graph branches that represent one-port elements
, are drawn
passi as oriented line
ve elements 1 segments with
L = . (12b) arrow
an D-type element
designating resistance
a sign convention R:forvthe=through-
adopted Rf orand f= v.
across-variables.
R Fig
(a) Through-variable source (b) Across-variable source
df f=f0 ure 4.3 shows branches for the generalized passive energy storage and dissipation ele
Figure 4.2: Linear graph representation of a simple mechanical system.
The
ments. branches
Each branch ishave
labeledthe
withfollowing sign
the generalized convention,
element type, and the designated
across- and
Figure 3.11: Idealized source elements. So we have a linearized elemental equation by the arrows: inthe
through-variables arrow
the branch are is drawn
related by the in the direction
elemental inthree
equation. For the which
gener
alized ideal (linear) elements the relationships are
3.4 Causality In a linear graph one-port elements are
the branchs represented in thev two-terminal
across-variable is assumedform intro
decreasing (in
duced in Chap. 3. Each element generates a branch in the graph and is drawn as a line segment
df the assumed direction of the across-variable drop), and
The elemental
The value equations relate
of the complementary the
variable ofthrough-
each sourceand across-variables
is determined by the sys' v = L . (13)
between the two appropriate nodes. Associated with feach branch isasanhaving
elemental through value.
andthe
to which must hold
source at all times
is connected. in an element,
A source may providesopower
if either
and the through-
energy to a SySl dt the through-variable is defined a positive
variable, assumed to pass through the line segment, and an elemental across-variable which
3.3absorb
or may Ideal
powersources
and energy, depending upon the sign of the complementary SOL
is the difference between the across-variable values at the two nodes. Each linearpage
system dynamics graph3 of 26
variable. Table 3.5 defines the source types in each of the energy domains.
branch thus represents the functional relationship between its across- and through-variables
 df I The sum of through-variables fiowing into any closed contour drawn on a linear graph
- =-v (4.2)
dt l is zero, that is,

For a generalized ideal D-type element (resistance) R, (4.7)

1
v = Rf or f - -v
-R (4 .3) for any N branches that intersect a closed contour on the graph.
Sec.4.3 Element Interconnection Laws 9S Continuity is applied by drawing a closed contour on the linear graph and summing the
where for energy storage elements the equations are expressed with the derivative on the
through-variables of branches that intersect the contour, as shown in Fig. 4.6. The arrow
left-hand side.
Ideal source elements are represented by linear graph segments containing a circle as direction on each branch is used to designate the sign of each term in the summation .

j~
shown in Fig. 4.4 . In all source elements one variable, either the across- or through-variable,

) )

is a prescribed independent function of time. For source elements the arrow associated with
4.4 Linear graph models of systems of one-
the branch designates the sign associated with the source variable:
110 port elements Fonnulation of System Models CI

\
\
I
wherefollowing
The v is the source across-variable
procedure fromwhen[1,it pis 101]
supplying
canthrough-variable
be used to fconstruct
to the
tem. The first form states that if f = 0, the across-variable is equal to Vs , and as f incre
1. For a through-variable//97;source the arrow designates the direction defined for positive 8 6/~ system graphs:
the output across-variable v decreases linearly. The second form states that if v = (
(a) A-type flow.
through-variable elements (b) T -type element (c) D-type element - -' Closed
define
output the system
through-variable f is equal boundary
to F" and as v and analyze
increases, the physical
the through-variable decn sys-
contour
(a) (b) linearly.tem features to be included in the model: (a) inputs, (b)
2. For an across-variable
Figure source
4.3: Linear graph the arrow designates
representation theone-port
of generalized direction defined
passive for. the across
elements
Notice that
variable drop.A-type elements (with the exception of electric ca- Figure 4.6: The definition of continuity conditions at (a) a single node on a linear outputs
The two formsofgenerate
interest, (c) energy
two possible models fordomains involved,
a power-limited source withand
a I (d)
characteristic:
required elements;
As described
pacitors), must in Chap.
have 3, A-type
their elements (with the exception
across-variable definedofwith capacitors) 4.2.3 graph.
electricrespect and (b) the extended principle of continuity applied 10 any closed contour on a
Series and parallel graph. connection of elements
must draw a schematic and assign a sign convention;
to have their across-variable
a constant referencedefined
value with respect dashed-line).
(denoted to a constant reference value . For
1. Equation
determine (4 .16) lumped-parameter
may be implemented by an ideal across-variable
elements: source of valu
(a) source, (b) en-
The arrow on an across-variable source branch is commonly drawn toward the reference
example, the velocity difference for a mass element is defined with respect to a constant Analogous to circuits, with a parallel
Thethatsign conventions forofsources aredrop
similar, and are shown For the special case in which a contour is drawnconnection
around a singleofnode,branches be-
the continuity in series with a resistance element with a value R as shown in Fig . 4.19a. This Sl
node since
velocity
in theinertial is usually frame.
reference
following the direction
figure. the assumed
The branches representing in anA-type
these across-variable
elements value.
therefore tween
law statestwo
that points,
the sum ofthe across-variable
through-variables flowingforintoeach parallel
any node branch
on a linear graph is
is
ergy storage, and (c) energy dissipative;
equivalent source model is known as a Thevenin equivalent source.
must have one end connected to the reference node . Some authors use a dotted line to identical (the
identically zero. Thethrough-variable splitsnode
law of continuity at a single between branches).
is illustrated in Fig. 4.6a. With
In this identify across-variables that define nodes and draw the
indicate this implicit connection to ground, as shown in Fig. 4.3. Apart from this notational a series
case fl - f2connection
- f3 = O. Theof branches
extended between
principle twofor
of continuity points,
a generalthe through-
contour may be nodes;(4.17) may be implemented by an ideal through-variable source of valu
2. Equation

~ ~
indetermine
parallel with a between which
resistance of value R as nodes each
shown in Fig element
. 4.19b. lies and
This configuratic
difference, A-type branches are treated identically to all other branches.
Direction of Direction of
variable
demonstratedfor each series
by considering branchcontaining
the example is identical
three nodes(the across-variable
shown in Fig. 4.6b. The
drawasthem;
known a Norton equivalent source model.
Each branch t across-variable
V, contains an arrow designating the t convention
F, sign continuity conditions at the three nodes are
associated with the varies).
through-variable select a sign convention for the passive elements and draw
across- and through-variables.
which
1
drop The
f flow
arrow on the graph element is drawn in the direction in
f, - f4 + fs = 0 at node A (4.8) These twoarrows
models(see
of realSection
sources are4.3); andand have identical characteristics as n
equivalent
100 Formulation
at node Bof System Models Cha(4.9) atselect
sured the sign
their terminals. conventions
Either may be used in the formodeling
sourceof elements to be
systems involving phy!con-
f2 - fs - f6 = 0
v, the across-variable associated with the branch is defined to be decreasing , that is,
(a) Across-variable source (b) Through-variable source
4.3 Sign conventions on one-port
-f3 + f4 + f6 = 0 \ at node C
system ele- (4.10)
sourcessistent
that may with the physical
be approximated model
by a linear (see Section 4.3) and draw
characteristic.
in the direction of the assumed across-variable "drop," and ments
Figure 4.9 shows a simple mechanical system consisting of a mass resting on a f them.
The load power P delivered by an equivalent source model depends on the acr
Figure 4.4: f is
the through-variable Linear graphas
defined representation of ideal source
having a positive valueelements.
. tionless plane and
For the contour moving all
enclosing under
threethenodes,
influence of anofexternal
the sum prescribed force
through-variables source.
into the F is
contour and through-variables at the terminals. For the Thevenin source the power is
In addition
possible assumedto to following
positive the conditions
force and velocity conventions from
are shown the with
together beginning
the COl
4.5 Physical source modeling
4.2
With Element
this convention , when interconnection laws
the elemental across- and through-variables have the same of
sponding linear graphs.
this section In each case
(Section the best
4) is upper to
nodefollow
represents
thethefollowing
velocity of the mas~
conven-
(4.11 ) P = vf = Vsf - Rf2 (4
direction (or sign) power, P = tv , is positive and flows into the element. the defined direction. An increase in the value of the across-variable indicates an increast
tions: Since sources are often non-ideal, the following can be useful
4.2.1 Compatibility
:LEIVIENT INTERCONNECTION LAWS The principle of continuity applied to any node states that there can be no accumulation
The choice of arrow direction for passive branches simply establishes a convention to velocity in that direction . The sign convention assigned to the force source defines whet source models:
and for the Norton source it is
of the through-variable atorthat node . the
If this principle did not Inhold, it would
and dimply fo that
define positive and
Quoting thenegative values of thelaw
compatibility through-
fromand[1,across-variables
p 95]: theand of theThe athepositive
is arbitrary.
sum force increases
define the positive decreases
modelvelocity
source of the mass.
direction Fig. 4.9a
compatible thewith
Thevenin equivalent (across-variable) source: an ideal
Linear
arrow graphs represent
direction the
does notdropsstructure
affect the of a system
onequation modelaround
formulation and specify
procedures theclosed
mannerinin
described which5 and integrated
Chap. that
through-variable
velocity directions
of are alignedis nonzero at the force
and a positive node,accelerates
and the node the would
mass ineither store or
the direct. I 2
across-variable the branches any loop thus acting as one of the primitive elements described in Chap. 3. sec-
the physical system (and per beginning of this P = vf = vFs - -v (4
elements
or a are
on subsequent
any connected.
linear systemThe
graph generaltheinterconnection
is identically
analyses; zero,
effect or laws anforarrow
of reversing lineardirection
graph elements
is simplyare
dissipate energy,
to of the applied force. across variable source Vs in series
R with a resistance R,
In
tion)
In each
practice
and
of the
it isenergy
often domains,
convenient thetoprinciple of continuity
adopt a convention corresponds
directing to theonfollowing
all arrows pass which can be described by v = Vs Rf.
derived in this section, with one set of laws relating across-variables and a second
reverse the sign of the defined across- and through-variable on the element. The choice of physical set
assume
elements away fromarrows
constraints: sources and always
toward point away
the reference from
node sources,
and then to assigntoward
a sou Themaximum
Norton equivalent (through-variable) source: an ideal
power an equivalent source can provide is found by differentiating Eq. (4
96relating through-variables, following the developments
sign convention is discussed more fully in Sec. 4.4. of several
Formulation of authors
System [1-3).
Models Chap. 4 grounds.
N convention that is compatible with the convention defined in the physical system. with respect to for Eq
through . (4.19) withsource
variable respect to F
v andin
equating
parallelthe derivative
with toa zero. In ei
resistance
X s
v =0 (16) case the maximum power is supplied when f = Vs/2R and v =1 RF.. /2 . The maxin
A4.3.1
compatibility equation may be written foriany closed loop on a graph , including inner
Compatibility
R, which can be
power supplied is Pm,x = Vs Fs/4.
described by f = F s R
v.
i=1
loops or outer loops, as shown in Fig. 4.5 . Because the arrows on the branches indicate Define positive velocity Define positive velocity
the direction of
compatibility
Thefor the across-variable
law represents drop,
a set of they are used on to assign the signontoa terms
graph in the ~v v~
summation;anyifNtheelements forming
loop traverses a constraints
a branch closed loop across-variables
onofthe graph. that
thein the direction of an arrow, the termIt in the

F"'--oJ
may be Ifrelated
a to physical
loop is laws governing
drawn as in the interconnection
following figures,lumped elements.
branches with may
summation
stated: ispointed
be arrows
is assigned
and those
positive, while if a branch is traversed against an arrow, the term in the sum
a negative
pointed
with the loop-arrow can be summed as positive,
value. against negative. \

The sum of the across-variable drops on the branches around any closed loop on a linear
graph is identically zero, or
Loop 2 B Loop
""--oJ variable
source
Through
variable
source
------
-----... F m = F(r) F", = -F(r) V,(I) FlO
(4.4)
Loop I
A m m
c (a) Thevenin equivalent source (b) Norton equivalent source

for any N elements forming a closed loop on the graph. /

Figure 4.19: Thevenin and Norton models of power-limited physical sources.
4 3
(a) (b)
5 State equation formulation[1, p 120]
(a) (b)
Define positive velocity Define positive velocity 5.1 State variable system representation
4.5: Compatibility equations defined from loops on a linear graph. (a) Some
4.2.2Figure
Continuity
possible loops on a graph, and (b) a loop containing four nodes and four branches. ~v v~ The state equations are a set of differential equations with state
variables as unknowns that completely describe the time evolu-
Quoting the continuity law from [1, p 97]: the sum of through-
Figure 4.5bflowing
variables shows a single
graphas isshown,
directions
into loop
anywith four branches
closed contouranddrawn
zero,theorcompatibility equation for this loop is
N
four nodes.
on With the arrow
a linear
F"'--oJ F"'--oJ tion of the system, given a set of initial conditions (expressed as
state variables). The system has order n, which corresponds to
the number state variables (minimum set of variables that com-
4 X
Fm=F(r) pletely describe the system behavior); n is equal to the number
LVi = VI - V2
fi +
= V3
0 - V4 = 0 (17)(4.5) Fm= -F(r)
of independent energy storage elements in the system.
;=1 i=1
m m

for demonstrate
We can any N branches that intersect
the compatibility law usinga the
closed contour
loop in Fig. 4.5b.onThe
the graph.
across-variable 5.1.1 State equations
drop on If
an the countour
element is drawnbetween
is the difference as in the following
the value figures, branches
of the across-variable at the two
with arrow The system state x(t) at any time t can be understood as a
nodes to which it ispointed
connected,into the closed
for example, contour are positive in the
V I = V A - VB is the drop associated with (c) (d)
sum, those pointed out negative. point in an n-dimensional state-space , and its time-evolution
element 1. If all the nodal values are substituted into Eq. (4 .5), then
4.2 Element interconnection laws system dynamics page 4 of 26
Figure 4.9: Possible force and velocity orientations for a simple translational mass.
4
LVi = (VA - VB) - (ve - VB) + (ve - VD) - (VA - VD) = 0 (4 .6)
as a trajectory on that state-space. The equation of state is All A-type elements that cannot be included in the normal (h) write B N +1ST independent compatibility equa-
tree and all T-type elements that are included in the normal tions involving only one secondary across-variable
x = f(x, u, t), (18) tree are dependent energy storage elements; all others are inde- (link across-variable) by using loops created by re-
pendent energy storage elements, elements whose stored energy
may be independently set and controlled. placing passive element links back into the tree, and
where u(t) is a vector of r system inputs. For a linear system, Two situations can occur when an excess of state variables write each equation explicitly in terms of the sec-
the Equation (18) becomes may be found by the previous procedure: ondary across-variable, then
1. when there are two or more A-type elements in direct se- 2. algebraically manipulate this set of n differential equations
x = Ax + Bu, (19) ries connection, and in the n state-variables and S specified source variables:
2. when there are two or more T-type elements in direct par-
where A is an n n matrix and B is an n r matrix. allel connection. (a) use the continuity and compatibility equations to
This excess can be mitigated by combining these elements as eliminate all secondary variables from the elemental
shown in the following figure with the equations equations,
5.1.2 Output equations (b) reduce the resulting B S equations in the primary
System outputs y(t), an m-vector quantity, are variables of in- 1 variables to n equations in the n state-variables and
Ceq = Pn and (21a) S source-variables, and
terest and can be expressed as a linear combination of state i=1 1/Ci
variables and inputs in the equation (c) write the resulting state equations in the standard
1 ~;;","" ,~ L - ".- Ie:
r
':'0 ' .('"~''' <
......
form.
y = Cx + Du, (20) Leq = Pn . - k'1 r oJ (21b)
134 i=1 1/L i
State Equation Formulation Chap.

where C is an m n matrix and D is an m r matrix. 5.4 Systems with non-standard state equa-
tions
5.2 Linear graph usage )c, 5.4.1 Input derivative form
If a system graph (oriented linear graph) is connected, meaning \
\
When the state equations have the forms
every point can be reached traveling along branches (usually t C; C =
eq
,,-
I
2, I/C;
. L, L" L =
cq
1/

2, IlL;
I
works for systems of one-port elements), the procedures in this I
;=1 i=1

)~
section (5) can be used. x =Ax + Bu + Eu, and (22)
If there are B branches in a system graph and of these S y =Cx + Du + Fu, (23)
are sources, there are 2B S unknowns and therefore required
equations. These will be a combination of elemental, continuity,
and compatibility equations. Figure 5.10: Combinalion of elemenlS in series and in parallel 10 eliminale excess stale
variables. which are typically due to either (1) a compatibility equation
5.2.1 Normal tree construction
5.3 State equation formulation includes the across-variable on a dependent A-type element and
5.3 STATE EQUATION FORMULATION an across-variable source term or (2) a continuity equation in-
A normal tree is constructed in order to define the system pri- The following procedure can be used to give a set of state equa- cludes the through-variable on a dependent T-type element and
mary variables, secondary variables, system order n, state vari- tions: The system normal tree may be used to generate a set of state equations in terms of the
energy storage variables on the n independent energy storage elements. In a system grapl a through-variable source term.
ables, continuity equations, and compatibility equations. From 1. derive B ofSwhich
with B branches, differential
S represent ideal and sourcealgebraic
elements, there equations
are 2(B - S) systen in terms These equations can be transformed into standard form by
these, a state model can be systematically constructed by elim- ofvariables
primary variables only by: one starting
across- and with passive 01elemen-
tal
associated
each equations
with the passive
branch. On each branchandoneusing
branches
B isSa primary
of these variables
one through-variable
compatibility
variable, while theand
othe conti-
a change-of-state-variables x0 = x Eu, which gives
inating all secondary variables from the n state equations. The is a secondary variable. The desired 11 state variables are a subset of the B - S primar~
procedure for constructing the normal tree is thus (starting with nuity equations to eliminate secondary variables (note:
variables. There are B - S elemental equations relating these primary and secondary vari
a system graph): S(#
abies forof thesources)
passive branches; =the S T (#
normal treeof through-variable
is used to generate B - S continuity sources)
an< + x0 =Ax0 + B0 u, and (24)
compatibility
(# equations that can be used
of across-variable to eliminate the secondary variables associate(
sources)):
1. draw the system graph nodes, with the passive elements.
(a) The generate y =Cx0 + D0 u + Fu (25)
2. select tree branches using the following rules: state equationsaarenormal
formulated intree (see above procedure),
two steps:
(a) no loops can be created, (b) identify primary variables as across-variables on tree-
1. Derivation of a set of B - S differential and algebraic equations in terms of prima~
(b) N 1 branches must be selected (N = number of branches and through-variables on tree-links,
variables only by starting with the passive elemental equations and using B - ~ where B0 = AE + B and D0 = CE + D.
nodes), and (c) compatibility
identifyandsecondarycontinuity equationsvariables
to eliminate allas through-variables
secondary variables. on
tree-branches
2. Algebraic manipulation ofand this setacross-variables
of B - S equations to produce on tree-links,
n differentia
3. the following order: (d) equations
identify in the nthe system order n source
as the number of inde-
(a) select all across-variable sources,
state variables and the S specified variables.
5.5 State equation generation using linear al-
pendent energy storage elements (number of A-type
Since sources have one variable independently specified, only B - S elemental equation: gebra
(b) select as many as possible A-type energy storage el- in normal
for the passive elements need tree plusItnumber
to be written. is convenient toof T-type
divide the numbernot in ~ normal
of sources
ements, tree), sources SA and through-variable sources Sr, and so S = SA + Sr. Thl
into across-variable
(c) select as many as possible D-type energy dissipative secondary variables may be eliminated from these equations by using a total of B - ~
If the combination of the elemental equations, after they have
(e) select
independent the and
compatibility state-variables
continuity equations fonned as from
across-variables
(I) N - I - SA continuit~ on A- been written in terms of primary variables, is difficult, linear al-
elements, and equations,type and (2)inB normal
- N + I - Sr tree and through-variables
compatibility on T-type
equations. (The secondary variable: gebra methods can be used to obtain the state equations. See [1,
(d) select as many as possible T-type energy storage el- associatedinwithtree-links,
sources do not enter directly into the state equation formulation; therefore
p 150] for the formulae.
ements. SA continuity and Sr compatibility equations do not need to be considered .)
(f) write B S elemental equations for passive (non-
All elements included in the normal tree are called branches, source) elements with primary variables explicitly on
and others called links, which, when connected to the tree form the left-hand-side, 5.6 Nonlinear systems
loops. (g) write N 1SA independent continuity equations in-
If all across-variables sources cannot be included in the nor- volving only one secondary through-variable (branch
mal tree, they must form a loop, and compatibility is violated. The same methods for finding the state equations used above
If at the end of the procedure an additional branch is required, it through-variable) by using contours that cut only one can be applied to nonlinear systems, but simplification of the
must be a through-source, which would violate continuity since passive branch element, and write each equation ex- elemental equations into the state equations can be trickier or
it cannot be independently specified. plicitly in terms of the secondary through-variable, impossible. See [1, p 152] for examples.
5.2 Linear graph usage system dynamics page 5 of 26
 Sec. 6.4 State Equation Formulation
5.7 Linearization of state equations 6.2.1 Normal tree
Some transformer ratio & gyrator modulus examples
By using Taylor series expansion in the usual way, we can lin- specify that if v I (or V2) is considered to be a primary variable, then V2
earize state equations into the form model TF/GY notes Transformers (top figure) require that one or the other, but not
secondary variable. Similarly. if fl (or f 2 ) is chosen as a primary variabl
rack & pinion r r: radius of pinion [1, p 176] both, branches be in the normal tree. Gyrators (bottom figure)
x = Ax + Bu (26) require definition
that botha or
secondary
neither variable.
branchesThebe transformer equations
in the normal tree. allow only on
N = r1 /r2 = n1 /n2 and one through-variable to be used as primary variables.
gear train 1/N
(r: radii, n: teeth) [1, p 175]

transformer TF
where A and B are the Jacobian matrices whose entries are
  DC motor 2N B`r see [1, p 179]
fi (x, u)  fi (x, u) 
aij = , and b ij = . slider-crank r r: radius of crank [1, p 171]
xj x=x ,u=u uj x=x ,u=u
0 0 0 0 block & tackle 2 see [1, p 172]
lever 1/L L = l2 /l1 [1, p 172]
6 Energy-transducing system elements[1, belt drive R R = r2 /r1 [1, p 172]
190 Energy-Transducing System Elemen
p 169] elec trans 1/N N = N2 /N1 (turns) [1, p 172]

6.1 Ideal transformers & gyrators

fluid trans
hydraulic ram
A
1/A
A = A2 /A1 (area) [1, p 172]
A: piston area [1, p 180]
Primary variables : VI' f2
~\
Primary variables:
"

I
fl ' v2 Figure 6.18: The two
figuration s for a lransf

GY
\ I

Secondary variables: fl' v2 Secondary\ variables: vI ' 12

I
shown as dOlled lines.
Whereas in the preceding, one-port elements were used to rep- disp pump 1/D D: vol disp/rad [1, p 171]
resent energy storage, dissipation, and storage in a single energy 1(W)2
, I
domain, we no introduce two-port elements called transducers
which represent energy transfer between two energy domains Ideal energy transduction models often need supplemented I
Since only one across-variable Imay\

be a primary variable for a tran

\

by passive one-port elements to take into account losses and

(e.g. motor, rack and pinion). Two-port elements are also used branch of a transformer may appearI as a \ tree branch. In other words. ei
in specific cases to represent energy transfer within a single en-
energy storage in the energy transducer.

ergy domain (e.g. levers, gears). 1. Primary

branchvariables:
I appears in the Primary
tree and branchI I'212is a link. in which case
Each port has a through- and across-variable associated with 6.1.1 Gear ratio v l ,v2 variables:
Figurevariables.
6.19: The or
lWO all
two primary
This requirement variables
changes theand V2 and flin
procedure areSection
secondary5.2.1 tofor a gyrator.
figurations T
it in its own energy domain. We require that power (P = f v) The notation N : 1 denotes that the input gear 1 rotates the following:Secondary variables: II ' 12 Secondary variables: VI. V2 shown as dotted lines.
flowing into each port to be positive: N times for every output gear 2 rotation, so 1 = N 2 or 2. branch 2 appears in the tree and branch I is a link, and so V2 and f I ar
1 = N 2 (note that N = 1/N ). variables and v Iofand are secondary
P1 + P2 = 0 (27) 1. draw6.4.3
the Derivation
system graph thef2Normal
nodes, Tree variables.
2. select tree branches using the following rules:
v1 f2 6.1.2 Mechanical levers The derivation of a normal tree for a system containing two-port elements is
= = TF (28) These two allowable
v2 f1 The transformer ratio TF for a lever can be found by assuming (a) of
nothe procedure
loops becausalities
can described
created, are
in Sec shown
. 5.2 in Fig.of6.one-port
for systems 18. elements. The s
v1 v2 small angles and recognizing that 1 = 2 : N
(b) tree for1 abranches
system graphmust be containing
model selected (N = number
two-port of should be
transducers
= = GY (29) The Gyrator:
nodes),
following steps: The generation of a set of independent compatib
f2 f1
x1 = `1 sin1 = `1 sin `1 (32) (c) one and only
ity equations from one of a
a tree transformers
structure twoabranches
containing can
gyrator requires a diffe
In order to satisfy this requirement, the two elemental equations x2 = `2 sin2 = `2 sin `2 . (33) be selected,
conditions.Step
For1:anDraw
ideal the systemsuch
gyrator. graphasnodes.
shown in Fig. 6.19, with element
defined by a two-port element must be of one of the two forms (d) both or neither of a gyrators two branches can be
that follow. Step 2:
selected, andInclude all across-variable sources as tree branches. (If all a
The transformer equations arise when across-variables relate Solving for x1 = f (x2 ), we see x1 = ``1 x2 = L
1
x2 = TFx2 . the =normal
to other across-variables and through-variables relate to other ., 2
sources cannot be included in VI
compatibility
GYf2tree, then the sources must fo
order: is violated.)
(G~)
through-variables as 3. the following
= -
6.1.3 Method for effective capacitances Step 3: Include as many as fl V2
possible of the A-type elements as tree
" # " #" # (a) select all across-variable sources,
v1 TF 0 v2 It is often that we wish to know the effective capacitance that the completion of the tree does not require the placement of both
= (30) (b) select as many as possible A-type energy storage el-
f1 0 1/TF f2 (mass, inertia, etc.) of some A-type element that is connected it canements,
be seen that or one branch of a gyrator in the tree. (Any A-type eleme
transformer
through one or more energy-transducing element to some source (c) select be correct
included in the normal tree
transducer is a dependent
branches, energy storage
minimizing the element.)
element that is driving it. This is done by 1.resulting
if VI is taken as a primary variable . thenstorage
V2 mustelements
also be considered a
where TF is called the transformer ratio. Step 4:number
Includeof one T-type
branch energy
of each transformer and both or neither b
The gyrator equations arise when across-variables relate writing down the elemental equation of the A-type element insince
tree,f2 is a secondary variable. or
to through-variables and through-variables relate to across- whose capacitance we would like to reflect, (d) select gyrator
as many in theastree so that the
possible maximum
D-type number
energy of T-type energy sto
dissipative
variables as substituting into this equation compatibility, continuity, remain
elements, out
and of the tree. If this step cannot
2. if f l is considered a primary variable. then because V2 isbecompleted , the system
then mo
by defin
and elemental (including transducer) equations as re- (e) select as 5:many as possible T-type
variable, f2 isAttempt
also a primary the energy
variable. storage el-
" # " #" #
v1 0 GY v2 quired to transform the original equation into an A-type Step to complete tree by including as many as po
ements.
= (31) elemental equation that is directly connected to the source dissipative elements in the tree . It may not be possible to include all D-t
f1 1/GY 0 f2 that is driving it, and
defining the effective capacitance Ce from this elemental To satisfyStep
the first
6: If case. withistwo
the tree not primary
completeacross-variables.
after the addition of both gyrator
D-type elem
where GY is called the gyrator modulus. relationship: dv = C1 f. pl aced in minimum
a tree . To number
satisfy the secondenergy
of T-type possibility.
storagewhere
elementsbothrequired
through-varia
to com
The following steps can be used to determine the TF (similar dt e 6.2.2 State-equation
T-type element generation
included a dependent
the two gyrator branches mustinboth
the tree
be attree
this links.
point isThe energy stor
two allowable tr
for GY): gyrators are illustrated in Fig. 6. 19.
6.2 State-equation formulation Step 7: can
The state equations Examine the tree
be found to determine
using the normalif any through-variable
tree cre- source
1. establish positive v1 , f1 , v2 , and f2 on a diagram, ation methods to of complete
Section it.6.2.1 and the proceduresource of Section 5.3
2. determine v1 -v2 relationship, If any through-variable can be inserted into the
The process of generating a state-equation for a system includ- with the following alteration:
then that source cannottwobe extra elemental
independently equations
specified and continuity is vi
3. define TF as v1 /v2 , and ing two-port elements is very close to that of Section 5.3. The for each transformer and gyrator will be found using Equa-
4. using Equation (28), define f1 -f2 relationship main difference is in the creation of the normal tree. tions 30 and/or
System 31.graphs and their normal trees for some simple systems containing on
5.7 Linearization of state equations storage elements and an ideal system dynamics
two-port transducerpageare 6illustrated
of 26 in Fig. 6
cases a choice of two-port cau sality exists in formulating the normal tree.
7 Operational methods for linear 8 System properties & solution the following Euler formulas are useful:
systems[1, p 205] techniques[1, p 244]
ejt = cos(t) + jsin(t) (42)
In this section we discuss time-domain operational mathemati- 8.1 System input function characterization e jt
= cos(t) jsin(t) (43)
cal and graphical methods often used to develop different system
representations. The four system representations in this section 8.1.1 Singularity input functions 1
are cos(t) = (ejt + ejt ) (44)
2
state equation form These functions are used to determine a transient (because ape- 1 jt
riodic) system response to discontinuous (or are so in their sin(t) = (e ejt ) (45)
x = Ax + Bu, derivative); they are identically zero for all t < 0. 2j
y = Cx + Du, 1. The unit pulse is defined as For more properties of exponentials see [1, p 250].
classical form (SISO) (
0 for t 0
T (t) = 1/T for 0 < t T (38)
8.2 Classical solution of linear differential
dn y dn1 y
dy 0 for t > T
equations
+ an1 + ... + a1
+ a0 y
dtn dtn1 dt Since we can re-write any SISO state equations in classical form
dm u dm1 u du where T is the pulse duration. In the limit as T 0 (or (see Section 7), we can use this formulation to solve for the out-
= bm m + bm1 m1 + ... + b1 + b0 u, (34) for T << ), we have the unit impulse, put. Defining the forcing function f (t) to be the right-hand-side
dt dt dt
of Equation 34, we have
time-domain operator form, where H{} is the dynamic

undefined for t = 0
transfer operator for SISO systems (t) = (39) dn y dn1 y dy
0 otherwise + an1 n1 + ... + a1 + a0 y = f (t). (46)
dtn dt dt
y(t) = H{u(t)} (35)248 R
with K(t)dt =System Properties
K where K and
is Solution Techniquesof the
the strength Chap. 8
The general solution y(t) to this equation is found by summing
bm S m + bm1 S m1 + ... + b1 S + b0 scaled impulse.
= {u(t)} (36) , the homogeneous solution yh (t), found when f (t) = 0, and the
2. The unit step is defined, as
U,(I)
S n + an1 S n1 + ... + a1 S + a0 , particular solution yp (t), found when f (t) is the specific input:
1.0

S 1

where S and are the differential and integral opera- 0 for t 0 y(t) = yh (t) + yp (t). (47)
tors; or, for MIMO systems, we define the matrix transfer us (t) = 1 for t > 0. (40)
operator
y(t) = H{u(t)}, and (37) 8.2.1 Homogeneous solutions
3. The unit ramp is defined as
operational block-diagram form (see [1, p 217,218,235] for The homogeneous solution, that of Equation 46 with f (t) = 0,
vector state equation, state equation, and classical forms o 1.0 can be found by the standard method of assuming a solution
Time 0 for t 0 yh (t) = Cet where C is a nonzero constant, and plugging into
of block-diagrams). ur (t) = t Figure 8 .3 :
for t > 0. (41)
The unit ramp function.
the homogeneous equation to get the characteristic equation
Relationships Among Singular Functions
7.1 Transformation from state-space equa-The ramp,Although not formally
step, and impulse functionsdifferentiable,
represent a familythese functions
of functions which,are re- in
as shown n + an1 n1 + ... + a1 + a0 = 0. (48)
tions to classical form Fig. lated
8.4 are by integrals
related and derivatives
by successive integrations. as shown in the figure.
For first-order systems, see Section 9.1; for second-order sys-
tems, see Section 9.2.
(1) u, (t) U,(I)

. By solving this equation for 1 , 2 , ..., n the solution with con-
stants C1 , C2 , ..., Cn . If there are n distinct roots to Equa-

1
1.0 r-l--- 1.0 tion 48,
For a quick and dirty second-order example, see [1, p 219, ex Integration Integration
7.5]. For higher-order or MIMO systems, a more formal method ..
Differentiation
..
Differentiation yh (t) = C1 e1 t + C2 e2 t + ... + Cn en t . (49)
using the transfer operator is useful, and described in [1, p 228].
For higher-order SISO systems see [1, p 231-3]. o 1 o
These methods essentially show how to find the state equa- Time Time Time If there are m repeated roots to Equation 48 and n m distinct
tions transfer operator H{} and use Equation 35 to back-out roots, then the unrepeated root solutions appear in the sum as
an equation of the form of Equation 34. 8.1.2 Sinusoidal
Figure 8.4: input functions
The relationship between singularity function s. before, but the repeated root solutions are multiplied by tk1
for the kth repetition (e.g. for 1 = 1 and 2 = 3 = = 2,
These
Time have
Shifting of the form u(t)
Singularity = Asin(t + ) and u(t) = Acos(t +
Functions yh (t) = C1 e1 t + C2 et + C3 tet ).
7.2 Transformation from classical form toThe singularity
), wherefunctions
A is themay
amplitude, is thetransient
be used to describe angular frequency
inputs that take (rad/s),
place at a time
state-space equations (state space real-otherand
than t=isO. the phase (rad).
The discontinuity The frequency
associated f is occurs
with each function foundwhenfrom thethe
function
izations) relationship
argument = 2f
is zero; therefore = 2/T
, a step where
that occurs at T is tothe
time mayperiod.
be written as usCt - to)
8.2.2 Particular solutions
since t - to = 0 at t = to. This property may be used to synthesize a transient functionThe particular solutions, that of Equation 46 with f (t) = f (t),
Although the state variables may not be physical, block dia-from a sum of singularity functions ; for example, Fig. 8.5 shows the function u(t) =
can be found by the method of undetermined coefficients, which
grams or canonical forms may be used to transform a classical 8.1.3 Exponential input functions
form differential equation into state-space equations. This real-u s (t) - 2u s (t - 1) + u,(t - 2) - u,(t - 3) . takes an educated guess based on the forcing function, checks,
then chooses proper coefficients to satisfy the differential equa-
ization (including the canonical forms, which I think easiest) is These have the form u(t) = est where s is complex in general tion. The following table has common forcing function with
8.2.2 Sinusoidal Inputs
described in [1, p 233-6]. (possibly real). When imaginary exponents are encountered, corresponding assumed solutions.
Sinusoidal input functions such as u(t) = A sin (WI + ) and u(t) = A cos (wt + ) , system dynamics page 7 of 26
shown in Fig. 8.6, are periodic with period T = 2rr /w seconds . These functions are de
scribed by three parameters: w , the angular frequency, (rad/s); , the phase (rad); and A,
 1. determine solution yIC (t) to the equation with no inputs 9 1st & 2nd order system response[1, p
yp (t) assumptions for method of undetermined coefficients and given nonzero IC, 276]
test 2. determine solution yu (t) to the equation with given in-
term in u(t) assumed form for yp (t) value puts and zero IC, and
3. add the two solutions for the total solution: Significance of 1st &2nd order to higher order systems
k K1 0 y(t) = yIC (t) + yu (t). While most systems are of order higher than two, the rough
ktn (n =1,2,...) Kn tn +Kn1 tn1 +...+K1 t+K0 0 dynamics of many systems can be approximated by a first or
If the ICs are zero, then step (1) is not required; if it is unforced, second order model, which are easy to derive and have useful
ket K1 et characteristics such as time constants, natural frequencies, and
kejt K1 ejt j then step (2) is not required; so only use this if it is both forced damping ratios. In addition, systems of higher order than two
and has nonzero IC, and then only when one or the other is can be considered to be comprised of first and second order com-
kcos(t) K1 cos(t) + K2 sin(t) j already known, or we only care about one or the other. ponents, which interact to form the higher order system. This
ksin(t) K1 cos(t) + K2 sin(t) j is easily understood in terms of Bode plot construction, since
real (first order) and complex conjugate pairs (second order) of
8.3.4 Differentiation & integration of LTI sys inputs poles and zeros, regardless of the order of the system, contribute
The test values in the above table are to test whether or not
the particular solution is a component of the homogeneous so- similar cutoff and asymptotic characteristics.
lution. If the test value is equal to any root of the characteristic If y(t) is the output of a LTI system with input u(t), and the This chapter deals with single-input systems (but typically
equation of multiplicity m, then the assumed solution must be input is differentiated du(t)/dt and re-applied to the system, SISO), whose state equations can be written in classical form of
multiplied by tm . the new output to this new input is dy(t)/dt. (Caviot: u(t) = 0 time-derivatives of output variables equal to time-derivatives of
for t < 0) input variables.
8.2.3 General (total) solutions If y(t) is the output of a LTI system with input u(t), and The solution method used in this section is based on super-
the input is integrated 0t u(t)dt and re-applied to the system,
R position and differentiability of LTI systems (see Section 8.3).
Equation 47 is the general solution to Equation 46. Typically, The solutions are found by (1) finding a solution yic (t): the
the new output to this new input is 0t y(t)dt. (Caviot: u(t) = 0
R
we apply the initial conditions (IC) after we have formed the homogeneous-IC (i.e. unforced) solution with nonzero ICs, (2)
general solution. This is not strictly required - see Section 8.3.3. for t < 0) finding another solution yf (t): the total forced solution with
zero ICs, and (3) adding solutions from (1) and (2) together:
8.3 System properties
8.4 Convolution & impulse response
8.3.1 Stability y(t) = yic (t) + yf (t) (53)
A system is said to be in equillibrium when its stave vector See [1, p 264] for a great discussion of impulse response and
does not change with time; it is said to be at rest when it is convolution.
where we call yic (t) the homogeneous-IC unforced response
in equillibrium and without inputs. Every system is at rest at
x = 0. and yf (t) the total forced response, which we will also call the
Assymptotic stability depends on the system response to a 8.4.1 Impulse response input-output response.
disturbance from an equillibrium state. If, when perturbed, it
assymptotically returns to its equillibrium state, it is assymp- The impulse response h(t) entirely characterizes a system by
totically stable. If it diverges, it is unstable. If it remains at its allowing the computation of the response to any other input 9.1 First order linear system response
perturbed state or oscillates about the equillibrium point, it is
neutrally or marginally stable. u(t). The impulse response is the response to the system to an
impulse (t). The classical form of this equation is
8.3.2 Time invariance
dy
Time-invariant systems have A, B, C, and D matrices that do 8.4.2 Convolution + y(t) = f (t) (54)
not depend on time. It is defined as time-invariant if in delaying dt
the input by T , the output is also delayed by T . Convolution is a function that maps a systems impulse response
and input to the output. It is instantiated in the following equa- where (seconds in English units for mechanical systems and in
8.3.3 Superposition of LTI systems tion:
SI units in general) is the time constant and f (t) is the forcing
If a system is both linear and time-invariant, it is abbreviated Z t function. If we would like to re-write a first-order state equation
LTI. These systems obey the principle of superposition, which y(t) = H{u(t)} u(t) ? h(t) = u( )h(t )d. (51) in this form, we can easily do so by the equation
states:
The response of an LTI system to a set of given ICs 1 dy d du ad bc
and an input consisting of several components, + y(t) = + u(t) (55)
For time-limited inputs (u(t) = 0 t < t1 & t2 < t; t1 is often a dt a dt a
0),
u(t) = u1 (t) + u2 (t) + ... + uk (t) (50)
where a, b, c, and d are the scalar state matrices.
t
R
u( )h(t )d for t < t2


may be found by determining the response to (1) the
t
ICs with zero input and (2) each of the k individual y(t) = t12 (52)
9.1.1 The homogeneous-IC (unforced) response yic (t)
input components and then summing all component
R
u( )h(t )d for t2 t.

responses to determine the total response.

t1 This is found by find the homogeneous response to Equa-
One use of superposition is as a shortcut to a solution when tion (54) and applying the initial condition yic (0) = y(0) to
part (either yIC (t) or yu (t) below) is already known. Then we Convolution is a linear operator and is commutative, asso- obtain
can solve a forced differential equation with nonzero IC in two ciative, and distributive. This has some interesting properties
for cascaded and parallel systems, see [1, p 268]. yic (t) = y(0)et/ . (56)
parts:
8.3 System properties system dynamics page 8 of 26
9.1.2 The characteristic response yu (t) So, with only the A matrix, we can determine the system prop- yic (t) for critically-damped system: = 1
erties The roots of the characteristic equation are equal:
The characteristic response is not just a particular solution, but
the general solution with zero ICs. The equation is n = a11 a22 a12 a21 (63)
1
yu (t) = yh (t) + yp (t) (57) = (a11 + a22 ) 1 = 2 = n (73)
2n
t/
= Ce + yp (t) (58) (a11 + a22 )
= . (64)
2 a11 a22 a12 a21 So from Equation (67), realizing we must include a factor
where C is found by the IC yu (0) = 0 and yp (t) is the particular of t, and the ICs yic (0) = y0 and yic (0) = 0, we have
solution for the input u(t) (note: not the forcing function f (t)). To assign to the output y a non-state-variable, see [1, p 298,
Input-output responses describe the response of a system to in- Eqs 9.54,55]. Of course, n and , and therefore the LHS of
puts of the form (all non-stochastic inputs) f (t) = q1 u + q0 u. Equation (59), are unchanged for any choice of output y. yic (t) = y0 en t + n ten t .


(74)
The following table has characteristic and input-output (total If we would like to develop homogeneous-IC solutions, we
forced) responses for the singularity inputs. must know ICs, typically output ICs y(0) and y(0). If we have
state equations, the output ICs can be found from the state and
output equations to be yic (t) for underdamped system: 0 < 1
Some Responses of System y + y = q1 u + q0 u The roots of the characteristic equation are
y(0) = c1 x1 (0) + c2 x2 (0) (65)
u(t) Characteristic re- Input-output (total forced)
sponse yu (t) response yf (t) for t 0 y(0) = c1 x1 (0) + c2 x2 (0) p
(t) 1 t/
e q1
(t) + q0
q12 et/ = c1 (a11 x1 (0) + a12 x2 (0)) 1 , 2 = n jn 1 2 = n jd . (75)

1 et/ q0 q1 et/ + c2 (a21 x1 (0) + a22 x2 (0)). (66)


us (t) q0
ur (t) t (1 et/ ) q0 t + (q1 q0 )(1 et/ ) p
where d = n 1 2 is the damped natural frequency.
9.2.2 The homogeneous-IC (unforced) response yic (t)
So from Equation (67), using Eulers formulas to write in
9.2 Second order linear system response The assumed homogeneous solution terms of trigonometric functions, and the ICs yic (0) = y0
and yic (0) = 0, we have
The classical form of this equation is yh (t) = C1 e1 t + C2 e2 t (67)
can be found by the characteristic equation
d2 y(t) dy(t) 2 en t
+ 2n + n y(t) = f (t) (59)
dt2 dt det [I A] = 2 + 2n + n
2
=0 (68) yic (t) = y0 p cos(d t ) (76)
1 2
where n is the undamped natural frequency, is the (dimen- to be p
sionless) damping ratio, and f (t) is a forcing function. 1 , 2 = n n 2 1. (69)
In the following list, homogeneous-IC solutions yic (t) will be where the phase angle
9.2.1 Transforming from state equations to classical developed for different situations arising from Equation (69).
form Note that in the following, it is always assumed that yic (0) = y0
and yic (0) = 0!
If we would like to re-write state equations in this classical form, = tan1 p . (77)
we must decide which output we desire. For an output of one yic (t) for overdamped system: > 1 1 2
or the other state variable, we can use Cramers rule The roots of the characteristic equation are
 p 
1 , 2 = n 2 1
" #
b1 a12 (70) Note that as increases, d decreases. It is some-
det times useful to know the amplitude Decay Ratio (DR)
b2 S a22 and with > 1, we will get real-valued 1 and 2 . So from
x1 = u (60) for yic (t) 6= 0:
det [SI A] Equation (67) and the ICs yic (0) = y0 and yic (0) = 0, we
" # have
S a11 b1
det + 2 1 2 1n t yic (t + Tp )
p
2
a21 b2 yic (t) = y0 e DR = = e2/ 1 (78)
x2 = u (61)
p
2 2 1 yic (t)
det [SI A]
2 1 + 2 1n t
p !
then re-arrange. The denominator det [SI A] is very impor- p e (71) where DR is the ratio of amplitude at time (t + Tp ) and
tant, since it determines the system response of every variable 2 2 1
time t, where Tp = 2/d .
in a second-order system. Essentially, it describes the dynamics
of the system without inputs (since it creates the LHS of the which is the sum of two decaying exponentials with yic (t) for unstable system: < 0
The roots of the characteristic equation have positive real
classical equation (59)), so we have the equality 1 1 parts, and result in an exponentially increasing, unstable
1 = and 2 = . (72) response. If 1 0, the response will oscillate with
1 2
d2 y(t) dy(t) 2
increasing amplitude; if < 1, the amplitude will grow
det [SI A] = + 2n + n y(t). (62) This response has neither overshoot nor oscillation. exponentially.
dt2 dt
9.2 Second order linear system response system dynamics page 9 of 26
9.2.3 The characteristic response yu (t) d2 y(t) dy(t) 2 y(t) = g(t) homogeneous-IC response yic (t) and the total forced
Responses of System dt2
+ 2n dt
+ n
response yf (t):
Damping g(t) Characteristic response yu (t)
ratio
ent y(t) = yic (t) + yf (t). (84)
0<1 (t) sin(d t)
The input-output (total forced) response yf (t) is the general n 1 2 
n t
solution to the equation us (t) 1
n2 1 e cos(d t + ) 10 General solution of the linear state
1 2 equations[1, p 331]

1 en t 
 10.1 State variable response of linear systems
2
t+ 2cos d t 10.1.1 Homogeneous state response
n n
ur (t)
The homogeneous state equation is
! !
2 2 1 2
+ p sin d t
1 2 n x = Ax (85)
d2 y(t) dy(t) 2 d2 u(t) du(t)
+ 2n + n y(t) = q2 + q1 + q0 u(t)
dt2 dt dt2 dt with an arbitrary set of ICs x(0).
=1 (t) ten t
(79) 1 t t

The assumed solution is
with ICs y(0) = y(0) = 0. State in another way, us (t) n2 1e n n te n

1
  xh (t) = (t)x(0) (86)
ur (t) n2 t 2 en t + ten t 2
n n
1 et/1 et/2


>1 (t) where (t) = eAt is the state transition matrix in terms of a
2n 2 1  matrix exponential. For more info about state transition matri-
1
1 n2 1 et/1 2 et/2


us (t) n2 ces (including how to compute them) and matrix exponentials,
2 1 see Section 10.3.

1

2 10.1.2 Forced state response
yf (t) = yh (t) + yp (t) (80) 2
t+
n n The forced state equation with input u(t) is
ur (t)
= C1 e1 t + C2 e2 t + yp (t) (81)
!
n 
2 t/1 2 t/2

x = Ax + Bu. (87)
p 1 e 2 e
2 2 1
with an arbitrary set of ICs x(0).
The solution is typically written in one of the following forms
Note: in the above table 1 = 1/1 and 2 = 1/2 as in Equations (72). Z t
x(t) = eAt x(0) + eAt eA Bu( )d (88)
9.2.4 The general (total) response y(t) 0
Z t
where C1 and C2 are found by applying the ICs y(0) = y(0) = 0, As mentioned above, we can find the total response to an input x(t) = eAt x(0) + eA(t ) Bu( )d (89)
and yp (t) is found by the method of undetermined coefficients u(t) by using the principles of superposition and differentiabil- 0
(see Section 8.2.2). ity/integrability of LTI systems. The input u(t) always enters
the equation in the form where is a dummy integration variable. Note that in Equa-
tion (89) the integral is the convolution integral.
Note that the RHS of Equation (79), the forcing function d2 u(t) du(t)
f (t) = q2 + q1 + q0 u(t). (82)
f (t) contains a linear combination of the input u(t) and its time dt2 dt 10.2 System output response
derivatives. From superposition and derivative/integral proper-
If we find the response of the system to the forcing function The output equation is
ties of LTI systems, we can find the input-output (total forced)
g(t) = u(t), yu (t), from the above table, then we can construct
response yf (t) by finding the characteristic response yu (t) to the
the total forced response to g(t) = f (t), yf (t), by the equation y = Cx + Du, (90)
input u(t) and differentiating, multiplying, and summing to get
the general (total) response (see Section 9.2.4 for more details). d2 yu (t) dyu (t) the homogeneous response is
yf (t) = q2 + q1 + q0 yu (t). (83)
dt2 dt
Since we can combine responses in this way, we xh (t) = (t)x(0), (91)
need only find the characteristic response yu (t) for u(t), If q1 = q2 = 0, the total forced response is just those solutions
in the table scaled by q0 . and the forced response is
then we can find the total forced response yf (t) for
any linear combination of u(t) that may arise in the The general (total) response y(t) to an input u(t) Z t
RHS of (79), f (t). The characteristic response yu (t) entering through a forcing function f (t) with ICs y(t) = CeAt x(0) + C eA(t ) Bu( )d + Du(t). (92)
is shown in the table below for the singularity inputs. y(0) = y0 and y(0) = 0 is the sum of the 0
system dynamics page 10 of 26
10.3 State transition matrix (t) Defining M in terms of column vectors, 10.3.3 Systems with complex eigenvalues
h i
10.3.1 Properties of (t) and eAt M = m1 | m2 | . . . | m n , (97)
The state transition matrix is defined as If some roots of the characteristic equation arise in complex
if we plug Equation (95) into Equation (85), we get conjugate pairs i,i+1 = j, the modal matrix will have
(t) = eAt (93) i mi = Ami i = 1, 2, . . . , n (98) corresponding terms e(j)t = et ej , which can be con-
verted into trigonometric form using the Euler relationships (44)
where the matrix exponential eAt
is defined in the table below which is the eigenvalue/eigenvector problem. The homogeneous and (45) to et sin t or et cos t.
as power series. Although this series always converges, rarely response is determined by the eigenvalues and eigenvectors of
does so quickly, so other methods (see Section ??) typically are A. Equation (98) can be re-written as
used.
Here are some properties of the state transition matrix: [i I A]mi = 0, (99)
1. (0) = I, for which a non-trivial solution requires
2. (t) = 1 (t) which gives x(t) = 1 (t)x(0),
3. (t1 )(t2 ) = (t1 ) + (t2 ) which gives (i ) = det[i I A] = 0 (100) 10.3.4 Systems with repeated eigenvalues
x(0) = (t0 )x(t0 ) and xh (t) = (t t0 )x(t0 ), and which is defined as the characteristic equation of A.
4. if A is a diagonal matrix, eAt is a diagonal matrix with Eigenvalues are the n roots of the characteristic
the diagonal elements eaii t . equation, i . These systems are not dealt with in [1], but a preliminary dis-
5. x(nT ) = [(T )]n x(0) n = 1, 2, . . . so if we know For a physical system, the eigenvalues are either real or occur cussion is presented here in Sec 10.3.9. The methods developed
(t) at some time T , we can easily find it for integer mul- in complex conjugate pairs. in 10.3.6 for finding the response do not apply to these systems.
tiples of T . For each eigenvalue i , there is an eigenvector mi found
by substituting into Equation (100). No unique solution exists,
since mi = mi m 6= 0.
Some matrix exponential properties The modal matrix M is defined as the matrix made of an
arbitrary set of corresponding eigenvectors mi ,
Description Matrix exponential property h i
2 2 3 3
definition eAt = I + At + A2!t + A3!t + ... M = m1 | m2 | . . . | m n (101) 10.3.5 Stability of linear systems
when t = 0 eA0 = I

1 Now we can write Equation (95) as
inverse eAt = eAt The definition of asymptotic stability is equivalent to stating
shifting in t eA(t1 +t2 ) = eAt1 eAt2 xh (t) = Met (102) that the homogeneous response of all state variables must de-
cay to zero in the absence of an input:
- e(A1 +A2 )t = eA1 t eA2 t only if A1 A2 = A2 A1 = 1 m1 e1 t + 2 m2 e2 t + ... + n mn en t (103)
d At
derivative e = AeAt = eAt A where
Rdtt At t
dt = A1 eAt I = eAt I A1 if



0 e
e 1 0 ... 0
integral
A1 exists; otherwise defined by the series 0 e2 t ... 0
et

=
.. .. .. .. (104)
lim xi (t) = 0 i = 1, 2, . . . , n. (107)
. . . . t
10.3.2 System Eigenvalues and Eigenvectors

0 0 ... en t
We are finding homogeneous solutions of the form
and (103) is a very useful form, showing that the solution is a
n
X linear combination of eigenvalue exponentials, in the directions
xi (t) = mij ej t (94) of the eigenvectors, weighted by constants determined by the
j=1 initial conditions. For distinct eigenvalues
= M1 x(0). (105) This leads to the summary:
where mij are constant coefficients that depend on the system
structure and initial conditions x(0), i.e. Note that the eigenvector columns of M must match the
t columns in which the corresponding eigenvalue appears in et .
e 1 This gives the state transition matrix for n distinct eigenvalues
A linear system described by state equations
e1 t
(since M1 does not exist otherwise) x = Ax + Bu is asymptotically stable iff all eigen-
xh (t) = M
..
(95) values of the matrix A have negative real parts.
. (t) = Met M1 . (106)
en t
This is itself a convenient way of computing (t) (note: the
where Voyage 200 calculator can compute eigenvalues and eigenvec-

tors) and leads to the important result: Asymptotic instability occurs when at least one eigenvalue
m11 m12 ... m1n has a positive real part. Marginal asymptotic stability occurs
m21 m22 ... m2n the homogeneous response of any state variable in when at least one eigenvalue has zero real part; if it also has a
zero imaginary part the system will get stuck at some con-

M= .. .. .. .
.. (96) the system from any ICs x(0) is a linear combina-
. stant nonzero state, if instead it also has a nonzero imaginary
. . . tion of n modal components ei t where i are the part the system will oscillate with constant amplitude about a
mn1 mn2 ... mnn eigenvalues of A. finite state.
10.3 State transition matrix system dynamics page 11 of 26
10.3.6 Transformation of state variables & modal de-
composition
For general linear transformations of the state variables
x = Pq
see [1, p 349] for a full discussion.
Using these methods, if a system has n distinct eigenvalues,
we can transform the state variable s.t. A becomes a diagonal
matrix which can provide insight into the internal structure of
a system. The transformed state equations are
q = q + B0 u
y = C0 q + Du
1 and lying in the invariant plane spanned by the eigenvectors of
where q = M x and M is the modal matrix of eigenvectors,
1 0 ... 0 q = Jq + B0 u
0 mode, which in this case was unstable (probably why we didnt
2 . . . 0
= .. . .. ,
.. (111)
. .
. . . real A-invariant subspace spanned by a complex conjugate pair
0 0 . . . n
B0 = M1 B, and C0 = CM.
This gives (t) = et as given by Equation 104 (the modal
matrix of eigenvectors for the decoupled system is simply the
identity matrix). This gives the homogeneous response in the
simple uncoupled form
qi (t) = qi (0)ei t .
Furthermore, in this form the solution can be written more
directly as a modal decomposition
x(t) = q1 (t)m1 + q2 (t)m2 + ... + qn (t)mn
where qi (t) is the i-th component of q. This equation also ap-
plies when the modal matrix M comprised of both eigenvectors
and generalized eigenvectors [2, p 312], so it is most certainly
true that any solution to the state equation can be written in
terms of its modal components. This is obvious when we think
geometrically: the set of eigenvectors and generalized eigenvec-
tors, since by definition linearly independent, constitutes a basis
for the state space . The fact that the input is mapped by B
to also makes the next point apparent: even the response
of a system to an input may be written in terms of its modal
components, and (113) still applies:
B(t)u(t) = 1 (t)m1 + 2 (t)m2 + ... + n (t)mn . (114)
The state equation can now be written as
q1 (t)m1 + q2 (t)m2 + ... + qn (t)mn =
q1 (t)Am1 + q2 (t)Am2 + ... + qn (t)Amn
+1 (t)Am1 + 2 (t)Am2 + ... + n (t)Amn .
For the special case of n eigenvectors, the state equation
becomes
(q1 (t) 1 q1 (t) 1 )m1 + (q2 (t) 2 q2 (t) 2 )m2 + ...
0 and required the excitation of the second mode, which in this
+ (qn (t) n qn (t) n )mn = 0.
10.3 State transition matrix
Since the set of eigenvectors {mi } is linearly independent,
qn (t) = n qn (t) + n for i = 1, 2, ..., n.
(108)
So for the case of a constant A matrix with a full set of eigenvec-
tors, the system is completely described by a set of n uncoupled
scalar equations whose solutions are of the form
Z t
qi (t) = e(tt0 )i qi (t0 ) + e(t )i i ( )d. (116)
t0
(109)
For systems without a full set of eigenvectors, a similar map-
(110) ping can be performed using the modal matrix M comprised of
both eigenvectors and generalized eigenvectors [2, p 312]. This
transformation into normal form results in a system
y = C0 q + Du
where q = M1 x, J = M1 AM is the Jordan canonical form
matrix (see Sec 10.3.9), B0 = M1 B, and C0 = CM. Note that
this system is as nearly decoupled as possible, but not fully so.
10.3.7 The invariant subspace
The invariant subspace of a linear map P : V V for a vector
space V is a subspace V for which P() is contained in .
This subspace can be termed P-invariant. The space spanned
(112) by eigenvectors of P is P-invariant. A subspace spanned by the
eigenvectors (assuming P has n distinct eigenvectors) will be
mapped by P back to the same subspace.
This is important in linear systems theory when we are inter-
ested in exciting specific modes of the system without exciting
(113)
others. To excite a specific mode or modes (which span an in-
variant subspace) with only initial conditions, choose an initial
condition x(0) that lies in the desired invariant subspace, i.e. se-
lect x(0) to be a linear combination of the eigenvectors spanning
the invariant subspace.
The case of real eigenvectors is easiest. We desire an initial
condition that lies in the invariant subspace of A, and this can
by achieved by selecting the coefficients of the linear combina-
(115)
tion of eigenvectors (which happen to be the initial conditions
of the diagonalized system),
x(0) = M (119)
= 1 m1 + 2 m2 + ... + n mn (120)
= Mq(0). (121)
So select i of only the desired modes to be nonzero, and the re-
sulting linear combination yields an A-invariant initial condition
x(0). The above figure shows two examples of initial conditions
for such a system: in blue, one selected by the preceding method
the desired modes (1 and 3); in purple, one selected that did not
lie in the that plane and required the excitation of the second
(117)
(118) want to wake it up).
The case of complex eigenvectors is a bit more subtle. The
of eigenvectors m1 and m2 is what we are concerned with,
since our initial conditions must be real. This real subspace
is spanned by
{Re[m1 ], Im[m1 ]} or {Re[m2 ], Im[m2 ]} . (122)
This arises from recognizing the requirement that, in order to
obtain a real initial condition, the linear combination coefficient
vector must contain complex conjugate pairs of complex num-
bers for complex eigenvectors.
The above figure shows two examples initial conditions for
such a system: in blue, one selected by the preceding method
and lying in the invariant plane spanned by the real and imag-
inary components of the eigenvectors of the desired modes (1
and 2); in purple, one selected that did not lie in the that plane
case was unstable.
system dynamics page 12 of 26
 C : "L~ uym
0: ou' ~ uy m

Figure 11 .1 symbolizes these relationships .

11.21. Controllability

Controllability is a property of the coupling between the input and the

10.3.8 Response of linear systems to singularity inputs 10.4 Controllability & observability 10.4.2 Dependence
involves the matriceson model
A and B.
Some special cases are worked out in [1, p 353] for singularity Two very important system properties, controllability and ob- Both controllability and observability are properties of the spe-
inputs (t), us (t), and ur (t). servability are briefly presented here (for more see [2, p 373]). cific systemDefinition 11.1. {A,
representation B, C,
A linear D}, which
system is said is
to not unique
be controllable at to if
Two similarity transforms are helpful in transforming the sys- for a given system. While this is true, a system lacking controlla-
find some input function (or sequence in the
bility can typically be granted it with the addition of actuators, discrete case ) u(t) , defin
tem into forms in which controllability and observability are
10.3.9 Systems with repeated eigenvalues apparent; they are that using the modal matrix, x = Mq, and just which will transfer
as a system lackingthe initial state can
observability x(to) typically
to the origin at some
be given it finite tim
the QR decomposition. The former leads to a Jordan form by adding
That issensors.
, there exists some input uI1 o, I,J , which gives x(t ) = 0 at a finite t
This topic is fully treated in [2, p 250], but not in [1]. Every of the state equations, the latter to the Kalmans controllable If there exists a system representation of order n that isl both
true for and
controllable all initial times to
observable, alland all initial
system states x(to), of
representations system is com
theorder
n n matrix has n eigenvalues, and for each distinct eigenvalue and/or Kalmans observable form.
i , a linear independent eigenvector mi exists. For every eigen- n arelable
both
. controllable and observable. If there exists a system
The system representation or realization {A, B, C, D} can representation of order n that is either not controllable or not
value i repeated i times (termed algebraic multiplicity of i ), be considered as the maps observable, no system representation exists that is both, some
any number qi (termed geometric multiplicity or degeneracy of Some authors
may be controllable butdefine another kind
not observable, othersof controllability
may be observ- involving the
i ) up to and including i of independent eigenvectors may ex- A: able The
but definition
not controllable, and still
given above others to
is re ferred may be neither.
as state controllability. It is the
ist: 1 qi i . qi is equal to the dimension of the null space definition , and is the only type used in this text , so the adjective " sta
of A Ii , B:Ur 10.4.3 LTI systems with isdistinct
Complete controllability obviousleigenvalues
y a very important property. If a
C:Ym In this case therecontrollable
is no need to refer
completely , then fortosome
any specific time no
initial states interval.
input exists whic
qi = n rank(A i I). (123) D:U Ym The system can be decomposed into its Jordan form as

This gives rise to three cases:
Fully degenerate: qi = i In this case, the eigenvalue
problem has qi = i independent solutions for mi . So, even
though there were not n distinct eigenvalues, n distinct eigen-
vectors still exist and we can diagonalize or decouple the system
as before.
Simple degeneracy: qi = 1 In this case, the eigenvalue
problem has qi = 1 independent solutions for mi . We would
still like to construct a basis set of n independent vectors, but
they can no longer be eigenvectors, and we will no longer be
able to fully diagonalize or decouple the system. There are mul-
tiple ways of doing this (e.g. Gram-Schmidt), but the typical
and most nearly diagonal way is to construct i qi generalized
eigenvectors (here also called mi ), which will be included in the
modal matrix M along with the eigenvectors. The generalized
eigenvectors are found by solving the usual eigenvalue/vector
problem for the first eigenvector m1i corresponding to i , then
solving it again with the following equations to find the gener-
alized eigenvectors
(A i )m2i = m1i
(A i )m3i = m2i
..
. is possible to find some input function (or sequence
This forms the modal matrix M . The block-diagonal Jordan
form matrix, analogous to the diagonal is
J = M1 AM, (124)
which gives the most-decoupled state transition matrix
(t) = MeJt M1 . (125)
General degeneracy: qi = 1 If 1 < qi < i , the preceding
method applies, but it may be ambiguous as to which eigenvec-
tor the generalized eigenvectors correspond (or how many for
each). This can be approached by trial and error or a system-
atic method presented in [2, p 255].
10.4 Controllability & observability
between the different sets U r , the r-dimensional input set; ,
the n-dimensional state-space; and Y m , the m-dimensional out-
put set (see figure below).
10.4.1 Definitions
Controllability [2, p 374] gives the following definition of con-
trollability.
A linear system is said to be controllable at t0 if it
in the discrete case) u(t), defined over t T , which
will transfer the initial state x(t0 ) to the origin at
some finite time t1 T , t1 > t0 . That is, there
exists some input u[t0 ,t1 ] , which gives x(t1 ) = 0 at
a finite time t1 T . If this is true for all initial
times t0 and all initial states x(t0 ), the system is
completely controllable.
Observability [2, p 375] gives the following definition of ob-
servability.
A linear system is said to be observable at t0 if x(t0 )
can be determined from the output function y[t0 ,t1 ]
(or output sequence) for t0 T and t0 t1 , where
t1 is some finite time belonging to T . If this is true
for all t0 and x(t0 ), the system is said to be com-
pletely observable.
q = q + B0 u (126)
y = C0 q + Du (127)
where q = M1 x, = M1 AM is the modal matrix of eigen-
vectors, B0 = M1 B, and C0 = CM. This fully decoupled form
allows us to easily ascertain the controllability and observability
of the system representation.
Controllability criterion 1
The constant coefficient system, for which A has dis-
Figure 11.
tinct eigenvalues, is completely controllable if and
only if there are no zero rows of B0 = M1 B.
Observability criterion 1
The constant coefficient system, for which A has
distinct eigenvalues, is completely observable if and
only if there are no zero columns of C0 = CM.
10.4.4 LTI systems with arbitrary eigenvalue
This case can be handled using a variation of the distinct eigen-
values criteria, but the method is lengthy, since it involves find-
ing the modal decomposition into Jordan form. The gist of it
is that for a decoupled system mode to be controllable, it must
either have a direct connection to the input (nonzero row of B0 )
or be coupled to another mode that has one. So nonzero rows
of B0 can be tolerated if they are not the last row associated
with a given Jordan block. Similarly, a system is observable
if the first column associated with a given Jordan block is not
identically zero.
The more useful criteria follow.
Controllability criterion 2
A constant coefficient linear system is completely
controllable if and only if the n rn matrix of
P = B|AB|A2 B| |An1 B ,
 
(128)
called the controllability matrix, has rank n. The
number of partitions of P required to achieve rank
n is termed the controllability index, with lower in-
dices implying better controllability.
Observability criterion 2
system dynamics page 13 of 26
 A constant coefficient linear system is completely the transfer function is defined as the ratio of the response am- 11.3.2 System poles and the homogeneous response
observable if and only if the n mn matrix plitude Y (s) to the input amplitude U (s):
Although developed here from the particular response, the
h
| | | | | n1 | |
i transfer function still describes the entire differential equation
Q = C |A C |A2 C | |A C , (129) Y (s) bm sm + bm1 sm1 + + b1 s + b0 of the system, so the homogeneous response is embedded in the
H(s) = = . (134) transfer function. The characteristic equation is the denomi-
U (s) an sn + an1 sn1 + + a1 s + a0 nator of the transfer function set equal to zero, so the system
called the observability matrix, has rank n. The eigenvalues are equal to the system poles (including canceled
number of partitions of Q required to achieve rank The output can now be written as
n is termed the observability index, with lower in- ones, see Sec 11.3.5). This gives the homogeneous solution
dices implying better observability. Note that the
complex conjugates are typically not required. yp (t) = H(s)U (s)est . (135) n
X n
X
yh (t) = Ci ei t = Ci epi t (137)
10.4.5 Stabilizability & detectability i=1 i=1
11.2 Relation to the transfer operator H{}
Weaker versions of controllability and observability are stabiliz-
ability and detectability. The are defined as follows. The only differences for SISO systems between H(s) and H{} where Ci is found from the initial conditions. The locations of
Stabilizability the poles define the n components of the homogeneous response
are in interpretation. H(s) is an algebraic quantity describing as follows:
A linear system is said to be stabilizable if all its the system particular response to an exponential input and can
unstable modes, if any, are controllable. be manipulated using linear algebra, whereas H{} is an opera- 1. pi = : the component Cet is a decaying exponen-
tial,
tor that is independent of the form of the input and implies a 2. pi = 0 : the component C is a constant amplitude,
Detectability causal relationship between system input and output.
3. pi = + : the component Cet is an increasing exponen-
A linear system is said to be detectable if all of its The complex variable s and the differential operator S{} can tial,
unstable modes, if any, are observable. be used interchangeably for LTI systems. 4. pi = j : the component Aet sin(t + ) is a
decaying sinusoid,
11.3 Poles and zeros 5. pi = j : the component A sin(t + ) is a sinusoid,
11 The transfer function[1, p 395] and
The numerator and denominator of Equation (134) can be fac- 6. pi = + j : the component Aet sin(t + ) is an
The transfer function is a representation of input-output dy- tored into the form increasing sinusoid.
namics of linear systems. There are different methods for find-
ing the system transfer function, but the one presented here Note that the larger the distance along the real-axis from the
is called the generalized exponential method (note that other N (s) (s z1 )(s z2 ) . . . (s zm1 )(s zm )
H(s) = =K origin, the greater the exponential decay or increase . Also,
methods give slightly different interpretations of the transfer D(s) (s p1 )(s p2 ) . . . (s pn1 )(s pn ) the larger the distance along the imaginary-axis from the origin,
function). It is developed in terms of the particular solution of (136) the greater the frequency of oscillation .
the system to an exponential input where K = bm /an is the gain constant. For second-order systems,
Confusion can arise from definitions of poles and zeros being
u(t) = U (s)est
p
(130) inconsistent between texts. [1] uses the definition that sets the p1 , p2 = n n 2 1. (138)
numerator and denominator of (136) equal to zero. Another
where the complex variable is s = + j and the amplitude takes the poles and zeros to be the values of s such that the
For underdamped second-order systems,
U (s) is complex in general. It can also be written limit of (136) as s approaches a pole or zero to go to or 0.
These two approaches lead to different poles and zeros in the p
case of pole-zero cancellation, see Sec 11.3.5. 404 p1 , p2 = n jn 2 .Function
1 Transfer
The (139)
u(t) = U (s)e(+j)t = U (s)et (cos t + j sin t), (131)
negative real axis . The poles for an underdamped second-order system therefore lie
which shows that this form of input covers a broad range of in- 11.3.1 The pole-zero plot The following
semicircle figure
with a radius shows
defined by Wn the
and atpole locations
an angle and
defined by the their
value of the relation-
puts of interest, including decaying sinusoidal waveforms. This ships
ratio to
~. n and .
derivation provides a basis for determining the steady-state re- Some system properties can be discovered from the plot of the
sponse characteristics of periodic waveforms. poles and zeros on the complex s-plane as shown in the following #

figure.
Sec. 12.4 System Poles and Zeros 401
g (s)

~ decreasing ~ -t 0
11.1 SISO systems 3(s)
s-plane

Given the classical representation (note this is different from

x- pole . !
,'/f\'
..::~------- +jw,,~

s-plane ,, ',
Equation (34) 0 - zero
~---- j2
I
I
"
, ,'
'
I
W 11

I '

dn y dn1 y dy , I
'
'

,,
I '

an n + an1 n1 + ... + a1 + a0 y
dt dt dt -~=I
,
I
-~wn
10
:'R (s)

dm u dm1 u du o 9\(s)
,,
\

,,
= bm m + bm1 m1 + ... + b1 + b0 u, (132) -2 -I
,
dt dt dt \
\

and the assumed particular solution " \\ *, ~: --------~-jW" ..j ~

\
\

:IE--- - j2 "

yp (t) = Y (s)est (Y (s) C), (133) sdecreasing S -t 0

Figure 12.3: The pole-zero plot for a typical third-order system with one real pole,
a complex conjugate pole pair, and a single real zero .
Figure 12.6: system dynamics page 14 of 26
Definition of the parameters Wn and { for an underdamped second-order
system from the complex conjugate pole location s.

define the components in the homogeneous response . It was shown in Sec. 8.3.1 that the 12.4.3 System Stability
11.3.3 System stability in terms of poles 11.4 Transfer functions of interconnected sys- of this transfer function would not lose any information in this
tems case.
Asymptotic stability An asymptotically stable is a system It is instructive to start with a state space realization of di-
that, with the input identically zero, for all initial conditions, in For systems in series and parallel, the transfer functions can mension n, find the corresponding transfer function H(s), and
the limit as time approaches infinity, the state asymptotically be combined. Two systems in series (cascading series), mean-
approaches zero. Since pole-zero cancellation does not play a find a minimal realization of this of dimension n1 . If n = n1 ,
role in initial-condition response, all the systems eigenvalues are ing the output of the first is the input of the second, if it can the system is completely characterized by H(s), it is both com-
identical to its poles, and vice-versa. For asymptotic stability, be assumed that the second does not load the first (i.e. the pletely controllable and observable, and its poles and eigenval-
all its poles must have negative real parts. A marginally stable first system acts as an ideal source to the second), can combine ues are the same. If n > n1 , then the information about n n1
system requires that at least one pole have zero real part, and transfer functions in the following way modes was lost (and it could have been unstable, who knows).
no poles with positive real parts. An unstable system is one For a scalar transfer function, the order of the irreducible re-
that has at least one pole with a positive real part. alization is the degree of the denominator after all common
H(s) = H1 (s)H2 (s). (140)
Bounded-input-bounded-output stability (BIBO) A pole/zero pairs are canceled.
system is BIBO if, with initial conditions all zero, no bounded For two systems in parallel, meaning that they share the
input causes an unbounded output. Pole-zero cancellation can same input and their outputs are summed, the combined trans-
play a role here, since a transfer function governs the input- fer function can be written as 11.6.1 transmission zeros
output relationship, and an unstable pole could be canceled by
a corresponding zero. While the transfer function may have no Transmission zeros are values of z for which
unstable poles, due to pole-zero cancellation there remains the H(s) = H1 (s) + H2 (s). (141)
" # " #
possibility of an unstable mode being excited by the input. If A zI B A B
the system is not asymptotically stable, it can still be BIBO rank < rank . (145)
stable iff (1) its B matrix is such that the no unstable mode is
11.5 State-space to transfer functions C D C D
excited by any input u(t) or (2) its C matrix is such that the The transfer function matrix can be constructed from the state
unstable mode is not noticed (although this is not practical). equations to be
12 Impedance-based modeling methods
C adj (sI A)B + D det (sI A) [1, p 422]
11.3.4 Interpretation of poles and zeros H(s) = . (142)
det (sI A)
Poles and zeros affect the system output in different ways, de- This chapter is primarily covered in Professor Joseph Garbinis
pending on the s-plane location of the pole or zero, the type of The structure of H(s), the m r matrix, is Appendix A, but I would like to make a few comments here.
input present (sinusoidal, etc.), and the type of output that is
of interest (transient, steady-state, etc.).
H11 (s) ... H1r (s)

Poles Poles, when there is no pole-zero cancellation (see
12.1 Input impedance (impedance of system)
.. .. ..
Sec 11.3.5), are identical to the system eigenvalues, and are al- H(s) =
. . .

(143) When an ideal source is connected to a system, either its across-
ways a subset of them. This means that, along with the initial or through-variable is given, and the other is a function of the
conditions, they always completely determine the homogeneous Hm1 (s) ... Hmr(s) system (e.g. back emf, current draw) and the given source
response of the system, as in (137). Note that this remains true variable. For an across-variable source, Fin (s) = Y (s)Vin (s)
even when pole-zero cancellation occurs (see Sec 11.3.5). where Hij (s) is the transfer function from the jth input to the where Y (s) is the input admittance. For a through-variable
Poles completely determine system stability. Depending on ith output. source, Vin (s) = Z(s)Fin (s), where Z(s) is the input impedance.
the given definition of poles and zeros, canceled poles and ze- For SISO systems,
ros may not be considered poles and zeros, but regardless, even Ignoring causality, Y (s) = 1/Z(s). This is usually just called
these canceled poles govern stability due to initial conditions (so H(s) = C(sI A)1 B + D, (144) the impedance of the system, since this is the impedance that
be sure to include them in an analysis of stability). the source sees (drives).
A periodic input can be described (by a Fourier series) as a which can be calculated using Equation (142).
sum of sinusoidal inputs and, by superposition, we can sum the 12.2 Quick transfer-function generation
responses of the system to each sinusoidal input. This means
that the frequency response function H(j) governs the steady- 11.6 Minimal realizations [2, p 408] This method can be used to either to obtain a general expression
state response. The poles affect such a systems response to for transfer functions between system inputs to outputs, or as a
Definition quick and dirty method to obtain a transfer function between a
periodic inputs by the manner in which they govern the bode
plot construction. Of all possible realizations of H(s), {A, B, C, D} is specific input and output. For the more general method, see [1,
Zeros Zeros are relevant only to response of the system said to be an irreducible (or minimum) realization if p 434]. When reducing the system, be sure to retain those fea-
to the input, since initial condition responses are unaffected by tures required for the transfer function (e.g. dont collapse a
the zeros. For periodic inputs in steady-state, zeros affect the the associated state space has the smallest possible
frequency response function in a manner analogous to the poles. dimension dim(). node if its across-variable is of interest to the transfer function).
Garbini, in the appendix, gives two main tricks to reduce a
A minimal realization is both completely controllable and com- system using impedances and obtain a specific transfer func-
11.3.5 Pole-zero cancellation pletely observable. In the case of a scalar transfer function, the tion: across- and through-variable dividers, and Thevenin and
minimum dimension required is equal to the order of the de- Norton equivalent models.
Pole-zero cancellation occurs occasionally. Recall that this oc- nominator of the transfer function after all common pole-zero
curs for a transfer function, which describes the input-output cancellations are made.
relationship, and tells us nothing of the systems response to A realization obtained from the methods presented earlier, 12.2.1 Across- and through-variable dividers
initial conditions. The canceled pole remains a component of linear graphs, are the most complete descriptions of the internal
the homogeneous response to initial conditions, but is consid- These are presented in Appendix A. The strategy is to reduce a
system. They may not be fully controllable or observable, but system to one or the other model, so the transfer function can
ered not to affect the input-output relationship (think about this is not always important. A minimal realization makes most be found using those models. Sometimes the transfer function
it in terms of Laplace transforms of the input-output differen- sense when the transfer function (scalar or otherwise) is mea- Vi
tial equation - there is no pole-zero cancellation for the initial desired is of the form T (s) = F : the elemental variable desired
sured, which describes the system with input and output and s
condition term). does not describe the internal system. A minimal realization is not matched with the source (across vs through or vice versa).
11.4 Transfer functions of interconnected systems system dynamics page 15 of 26
This can be solved using the relevant elemental relationship (e.g. where H(j) is defined as the frequency response of the system. 13.1.2 Second-order systems of the usual form
1 F From the transfer function
Vi Ci s i
Fs
= Fs
). Second-order systems of the form
H(j) = H(s)|s=j . (151)
12.2.2 Thevenin & Norton equivalent models The magnitude or gain of the frequency response is
2
If the across- or through-variable dividers are unable to allow y + 2n y + n y = K0 u(t) (161)
the desired transfer function to be found, the source equivalent
q
methods will be of great help. |H(j)| = [Re(H(j))]2 + [Im(H(j))]2 ) (152)
The idea for Thevenin models is that any linear system ex-
cited by a single source (across or through) and driving an ex- and the phase angle of the frequency response is where K0 is a constant, has transfer function
ternal load ZL may be modeled as a single across variable source  
Vs (Garbinis Ve ) connected in series with a single impedance Im(H(j))
element Zout (s) (Garbinis Ze ). (j) = tan1 . (153)
Re(H(j)) K0
The idea for Norton models is that any linear system excited H(s) = . (162)
s2 + 2n s + n
2
by a single source (across or through) and driving an external The steady-state solution yss (t) for a system with transfer
load ZL may be modeled as a single through variable source function H(s) (frequency response H(j)) to a sinusoidal input
Fs (Garbinis Fe ) connected in parallel with a single impedance u(t) is
element Zout (s) (Garbinis Ze ). yss (t) = A|H(j)| sin(t + + (j)). (154) The frequency response function is
For construction of these models, see [1, p 441] and follow the
procedure. There arise some situations where the transfer func- This means that for a sinusoidal input, the system responds
tion desired can make use of transforming between Thevenin (in steady-state) with a sinusoidal output at the same fre-
and Norton models. K0 /n2
quency with amplitude scaled by |H(j)| and phase-shifted H(j) = , (163)
(j). |H(j)| is the ratio of the output amplitude to the input 2
(1 2 ) + j(2 n )
12.3 Transducing elements amplitude, sometimes called gain of the system, as a function n
of input frequency . Systems that respond to low frequen-
These can be dealt with using the following method. For trans- cies but attenuate high frequencies are called low-pass filters,
formers, the effective impedance at port 1 of the impedance Z3 while systems that response to high frequencies but attenuate
connected across port 2 is low frequencies are called high-pass filters. and its magnitude and phase are

Z1 (s) = (TF)2 Z3 (s). (146)

13.1 1st - & 2nd -order systems 2
For gyrators, the same quantity is K0 /n
13.1.1 First-order systems |H(j)| = r (164)
2 2
1 First-order systems of the form (1 2 )
n
+ (2 )2
Z1 (s) = (GY)2 . (147) n
Z3 (s)
dy 2/n
+ y = K0 u(t), (155) (j) = tan1 . (165)
dt 1 (/n )2
13 Sinusoidal frequency response of linear
systems [1, p 453] where K0 is a constant, has transfer function

In this section we will investigate the response of a system due K0 The maximum (peak) magnitude occurs at the resonance
H(s) = . (156) frequency
to inputs of the form s + 1
u(t) = A sin(t + ) (148) The frequency response function is
p
where A is amplitude, is the angular frequency ( = 2/T , K0 P = n 1 2 2 2/2 = .707 (166)
where T is the period), and is the phase. H(j) = , (157)
j + 1
Here we deal only with the steady-state response, since the
transient terms of the sinusoidal response (due to ICs) die out and its magnitude and phase are since for .707 no peak occurs and the magnitude is mono-
quickly. tonically decreasing for increasing . Note that this is only valid
We begin with the steady-state (particular) solution to an K0 for systems of the form (161). It was derived from taking the
exponential input u(t) = U (s)est , |H(j)| = p (158) derivative of the magnitude of the frequency response function
( )2 + 1 wrt frequency, setting equal to zero, and solving for . The
yp (t) = Y (s)est = H(s)U (s)est , (149) (j) = tan1 ( ). (159) solution varies if the transfer function is of another form, for
instance, if the transfer function has a zero. The magnitude at
this frequency is
where H(s) is the transfer function and U (s) is the complex in- From Equation (154), the steady-state response to a sinu-
put amplitude. Since we only desire sinusoidal response, s j. soidal input is
This leads to the expression for the complex output amplitude
Y (j) as a function of a gain and the complex input amplitude: K0 2
K0 /n
yss (t) = A p sin(t + + tan1 ( )). (160) |H(j)| = p . (167)
Y (j) = H(j)U (j) (150) ( )2 + 1 2 1 2
12.3 Transducing elements system dynamics page 16 of 26
 484 Sinusoidal Frequency Response of Linear Systems

from the n system poles to a test point s = jw has a magnitude and an angle:

13.2 Bode plots The following procedure can be followed to construct the
Ijw - Pil =
4. graphically add each component (here the phases are, in
ja? + (w Wi)2

Bode plots are magnitude and phase angle plots as functions

Bode magnitude plot: fact, additive, L(s and- onPi) = a linear
tan-I scale).
-ai
(w -
Wi)
1. factor the numerator and denominator of H(s) into con-
of input (driving) frequency . They are typically graphed on
stant, first-order, and second-order terms s.t. each term 13.2.2 Bode plots from pole-zero plots
decibel (dB) and degree ( ) vertical axes and logarithmic hori- has a corresponding entry in the above table, as shown in Fig. 14.20a, with similar expressions for the vectors from the m zeros.
zontal axes. Decibels are found by the formula 2. identify the break frequency associated with each factormagnitudeGivenand a phase
factoredangle transfer
of the complete frequency
function, theresponse
poles may andthen be written
zeros, in
or just
    (from table), the
of the pole-zero
magnitudes andplot,anglesmuch of thesecan be said
component about the magnitude and
vectors:
P A 3. plot the asymptotic of each of the factors on dB-log or phase response, and a magnitude Bode plot can be quickly
Q = 10 log10 dB = 20 log10 dB. (168) sketched. If each factor in a factored frequency response func-
to aI(~w
Pref Aref log-log axes, IH(jw)1 = K [1!=1 - zi)1
(a) draw horizontal line (unless there is a pole or zero at tion, each corresponding
vector from the pole or zero [1i=1
pole
l(jw Pi)1
location
or zero, is considered as a
to the location of j, which
|H(j)| is the ratio of the amplitude of the sinusoidal input to the origin) at the constant-gain value up to the break is the input frequency -m located on the n vertical axis, then the
the amplitude of the sinusoidal output, so in dB, frequency, and magnitude andLH(jw) phase=of the L(jw L
frequency
- Zi) - L
response
L(jw Pi) function are
(b) draw (linear on log) line using the high- slope i=1 i=1
|H(j)|dB = 20 log10 |H(j)| dB. (169) (negative if pole, positive if zero) r1 r2 rm
As defined in Sec. 12.6, |H(j)| if the vector K the pole Pi to the point s = jw has(170)
= from
q q2 qn length
The following table is helpful for quick conversions. 4. graphically add the component plots (multiplication is ad-and an angle (}i from the horizontal, and the1 vector from the zero Zi to the point jw has
dition in log-scale: log(ab) = log a + log b), and length ri and an angle i, as shown=in(
H(j) Fig.1+ 2 +the
14.20b, value
+ of m) the frequency response
dB - power - amplitude relationships 5. round out the corners using known values at break fre-the point jw is (1 + 2 + + n ) (171)
dB Pout /Pin Aout /Ain quencies (3 dB for first-order sections, and dependent on
-40 0.0001 0.01 for quadratic factors). where ri = |(j IH(jw)1 zi=)|,K qrli ...= rm |(j pi )|, K is a constant that
cannot be determined from ql" the qn pole-zero plot, and are the
-30 0.001 0.03162 Phase plots can be constructed by using a similar method. i i
angles pole and LH(jw) = (I + ...
zero vectors + m) - (}I + make
(respectively) ... + (}n)to the horizon-
-20 0.01 0.1 Here is a table for phase-plot properties.
tal.
-10 0.1 0.3162 Asymptotic properties for phase Bode plot
-6 0.25 0.5 jUi jUi
Description jUi] s-plane s-plane
-3 0.5 0.7071 Break frequency Phase @ High-
( = pole, B ( )
B (rad/s) phase ( )
0 1 1 = zero) p]~

3 2 1.414 Const gain 0

6 4 2 at origin 90 o (J (J

10 10 3.162 at origin 90
P2)(01-----1.
20 100 10 Real 1/ 45 90
40 10000 100 Real 1/ 45 90
Conj s n 90 180
Another common measure is the decade, which is synonymous The following (a) observations can be made from (b) these plots:
with the order of magnitude - a factor of 10. Conj s n 90 180
for a 14.20:
Figure real pole: Definition of the vector quantities used in defining the frequency
The procedure for plotting the phase is as follows (first two steps for
response low-,
function the
from the magnitude
pole-zero approaches
plot. In (a) the a finite
vector from a pole (or zero)value,
is
13.2.1 Asymptotic Bode plots from transfer function and
defined, as in(b)increases,
the vectors fromthe magnitude
all poles and zeros in aand
typicalphase
system aredecrease,
shown.
redundant if magnitude plot is already finished):
This method for hand-drawing Bode plots is best-suited when for high-, the magnitude 0, and phase /2;
the transfer function is known (see Section 13.2.2 when the pole- 1. factor the numerator and denominator of H(s) into con- The graphical
for a real method zero: can be very useful in deriving a qualitative picture of a system
zero plot is known). stant, first-order, and second-order terms s.t. each termfrequency response. forFor low-,
example, theconsider
magnitude approaches
the sinusoidal responseaoffinite value,system
a first-order
The general method is to build a higher-order bode plot from has a corresponding entry in the above (magnitude) table,with a pole on theasreal axis increases,
at s = -I/Tthe magnitude
as shown in Fig. and14.21a
phase andincrease,
its Bode plots in
easier-constructed lower-order plots. The magnitude Bode-plot 2. identify the break frequency associated with each factorFig. 14.21 b. Even for
though high-,
the gainthe magnitude
constant K cannot be detennined
, and from phase the/2;
pole-zero plot,
characteristics of some of these are in the following table. (from table), for complex conjugate poles:
3. plot each component phase plot for low-, the magnitude and phase approaches a fi-
(a) if the component is one of the first three in the ta- nite value,
Asymptotic properties for magnitude Bode plot as passes nearest the poles, the magnitude reaches
ble, draw the constant line and skip the rest of the
Description Break sub-steps a maximum (resonance),
High- slope as increases further, the magnitude and phase de-
( = pole, Transfer function frequency (b) plot the break frequency B point (see above table),
(rad/s) (dB/decade) crease,
= zero) (c) if the component is a real pole or zero,
i. plot the horizontal zero-phase up to one decade for high-, the magnitude 0, and phase 2/2 =
Const gain K 0 below B , ;
at origin 1/s -20 ii. plot the high-frequency asymptote (see above ta- for complex conjugate poles:
for low-, the magnitude and phase approaches a fi-
at origin s 20 ble) after one decade above b , nite value,
Real 1/( s + 1) 1/ -20 iii. linearly connect these two lines, drawing a line as passes nearest the zeros, the magnitude reaches
through the Phase @ B , a minimum,
Real s + 1 1/ 20 as increases further, the magnitude and phase in-
2 (d) if the component is a conjugate pair, inspect phase
n crease, and
Conj s n -40 plot shapes in [1, p 476, Figure 14.15] based on ,
s2 +2n s+n
2 for high-, the magnitude , and phase 2/2 = .
s2 +2n s+n
2 but note that the low-, Phase @ B , and High-
Conj s 2
n
n 40 phase will all be satisfied (see above table), The following generalizations can be made.
13.2 Bode plots system dynamics page 17 of 26
 If a system has more poles than zeros, as the frequency 14.1 Fourier analysis of periodic waveforms 14.1.1 Properties of the Fourier series
becomes large, the magnitude tends to zero. (Fourier Series)
If a system has more zeros than poles, as the frequency See [1, p 509] for more, but a few properties are listed here.
becomes large, the magnitude tends to infinity (which can- A periodic function is one that satisfies f (t) = f (t + nT ) n = Linearity If the Fourier series components of two peri-
not happen in a physical system). 1, 2, . . . where T is the period of the function. The funda- odic functions g(t) and h(t) with identical periods T are
If a system has a pair of complex conjugate poles near the mental frequency 0 is defined as 0 2/T . Waveforms are Gn and Hn , then, if a new function f (t) is defined as a
imaginary axis, a peak magnitude is seen as the frequency called harmonics of each other if their fundamental frequencies linear combination of g(t) and h(t),
approaches them. If the poles are on the imaginary axis,
the magnitude is infinite at that frequency. (or periods) are an integer ratio (meaning the greater divided
If a system has a pair of complex conjugate zeros near the by the smaller equals an integer). f (t) = ag(t) + bh(t), (185)
imaginary axis, a minimum (notch) magnitude is seen as The Fourier series is a description of a periodic function as
the frequency approaches them. If the poles are on the a weighted sum of harmonic sinusoidal components. The nth the Fourier components of f (t) are
imaginary axis, the magnitude is zero at that frequency. harmonic component of the series can be written
A pole at the origin (pure integrator) implies an infinite Fn = aGn + bHn . (186)
fn (t) = an cos(n0 t) + bn sin(n0 t) (173)
magnitude at DC-frequency. Even and Odd Functions If the function is even, then
A zero at the origin (pure differentiator) implies a zero = An sin(n0 t + n ) (174)
all bn = 0. If a function is odd, then all an = 0. Note that
gain DC-frequency. where the conversions between each representation are if a function would be odd if it was shifted up or down by
some DC value, assume an = 0, but shift the result by the
an = An sin n and bn = An cos n (175) same shift (see next property).
Constructing magnitude Bode plots
The pole-zero plot provides a good method for constructing a
q
an Interpretation of the zero-frequency (DC) term a0
An = a2n + b2n and n = tan1 . (176) or F0 This term is simply the average value of the func-
magnitude Bode plot. The constant K from Equation (170) bn tion over a period. If a function is shifted up or down,
cannot be determined from the pole-zero plot, so the Bode plot this is the only term in the series that is effected, and it
must be scaled (shifted up or down for log-plot) appropriately. The Fourier series representation of a function f (t) subject
to certain conditions [1, p 509] is is shifted by the same average or DC value.
The key idea is that the vector drawn from the origin to each
pole or zero corresponds to the break frequency (the magnitude 1 X
14.1.2 Line spectra
of the vector) and the phase angle (the angle of the vector). f (t) = a0 + [an cos(n0 t) + bn sin(n0 t)] (177)
487 2 For a waveform decomposed into Fourier components, it is nat-
Break
Sec. 14.7 the pole-zero
Frequency Response andplot into radial
the Pole-Zero Plot regions with boundaries n=1 ural to describe the waveform as a line spectrum. This is most
between the regions being the radial positions of the poles and commonly represented as a magnitude and phase plot versus
zeros. Bode
magnitude These plot radial positions
without knowing correspond
the absolute gain. Theto (equal!)
method described break fre-
here allows 1 X
= a0 + An sin(n0 t + n ). (178) frequency, using (176) or simply the absolute values and phases
thequencies
magnitude plot in to the be sketched by inspection without
Bode magnitude plot drawing
(at this the point,
individualdash-line
component 2
curves. The break
method isfrequencies).
based on the fact that the overallfrom
magnitude n=1 of the complex coefficients |Fn | and Fn . For the complex case,
in the Starting thecurve undergoes
origin, move a change
out- sometimes the magnitude plot is of the coefficient Fn0 in front
in slope at each break frequency. A third, equivalent, complex representation is
ward picking up poles and zeros as you go. The slope of the
The plot
Bode first step is to identify
in each regionthedepends
break frequencies,
on howeither many by factoring
poles and the transfer
zeros of the exponential of Fn = Fn0 en where |Fn | = |Fn0 |.

you have
function carried
or directly topole-zero
from the that region. If you
plot. Consider passpole-zero
a typical a pole,plotthe of aslope
linear X The complex representation gives rise to a two-sided spec-
decreases
system as shownby 20 14.23a.
in Fig. dB; ifTheyou breakpass a zero,
frequencies thefour
for the slope increases
first- and by
second-order f (t) = Fn ejn0 t (179) trum, meaning that negative n gives rise to negative frequency
20 dB.
blocks are allThe formula
at a 'frequency fortothe
equal slopedistance
the radial in eachof theregion
poles oras a function
zeros of
from the origin n= components. The real representation gives a one-sided spec-
the number of poles P and zeros Z between it and
of the s-plane, that is, Wb = J a + w Therefore. all break frequencies may be found
2 2 the origin is trum, for which only positive n and frequencies arise. For con-
by taking a compass and drawing an arc from each pole or zero to the positive imaginary
where version between one- and two-sided spectra, use (180) and (181).
axis. These break frequencies may be transferred directly to the logarithmic frequency axis 1 To convert a two-sided magnitude line spectrum to a one one-
of the Bode plot.
slope = 20(Z P ) dB/decade. (172) Fn = (an jbn ) (180) sided, the plot is merely folded over and doubles its posi-
2 tive half (since it is symmetric about the magnitude axis) or
1 An = 2|Fn |.
2010g lo 1HUw)1
Fn = (an + jbn ). (181)
jw ~;
30r-~--~--~--~--------~--~
2
5 radls
20
So we have a representation of any periodic function as a 14.2 Response of linear systems to periodic
N
, weighted sum of sinusoids. To determine the weights (coeffi- inputs
s-plane " &J 10 .
cients), the following formulas can be used. For the real repre-
" 1.414 rad/s
-0
';'0f----,-----'-----,--~--+-'\>_:__,_-...,....____1
sentations, Given a system with frequency response H(j) and an input
,, ,
x:.--
I
)10
"
.;;;
v_IO ' u(t) of the form of Equations (177) or (178), the output of each
, 2
Z t1 +T
L
, I

I
I I
I o.I radls -20 an = f (t) cos(n0 t)dt, (182) component of the input is
-5 \ -1 -0.1 a _30 L-.L..1....l..Ll.L1.I1_L..L.1..l..1.lliL---1....LLLlJ.llL---L....L.l.llW w T t1
, I N ::::I
yn (t) = |H(jn0 )|An sin[n0 t + n + H(Jn0 )]. (187)
'*--- -)1
0.01 0.03 0.1 0.3
(b)
3 10 30 100
Angular frequency (rad/s)
bn =
2
Z t1 +T
f (t) sin(n0 t)dt, (183)
(a) T t1 By the principle of superposition,
Figure 14.23: Construction of the magnitude Bode plot from the pole-zero
14 Frequency domain methods [1, p 500]
diagram. (a) A typical third-order system and the definition of the break frequencies . and and a0 can be computed separately from Equation (182). For X
. (b) the Bode plot based on changes in slope at the break frequencies. the complex representation, y(t) = yn (t). (188)
Z t1 +T n=0
In the section we develop methods of determining the system re- 1
Because
sponse toaJJmore
low-frequency
generalasymptotes are horizontalWe
input functions. lineswill
with adescribe
gain of 0 dB, a pole
inputs Fn = f (t)ejn0 t dt, (184) Given the same situation but with the input in the form of
or zero does not contribute to the magnitude Bode plot below its break frequency. Each pole T t1
as sums (or integrals) of many sinusoidal inputs and develop, Equation (179),
or zero contributes a change in the slope of the asymptotic plot of 20 dB/decade above
based on the principle of superposition (sum of inputs gives sum
its break frequency. A complex conjugate pole or zero pair defines two coincident breaks
Fn is the complex conjugate of Fn , and F0 can be computed
of of
20individual
dB/decade (oneoutputs),
from each methods forpair),
member of the finding
giving system responses.
a total change in the slope separately (with n = 0) from (184). yn (t) = H(jn0 )Fn ejn0 t (189)
of 40 dB/decade. Therefore, at any frequency w , the slope of the asymptotic magnitude system dynamics page 18 of 26
function depends only on the number of break points at frequencies less than w , or to the
left on the Bode plot. If there are Z break points due to zeros to the left and P break points
and, once again using superposition, 14.4.2 Relationship between H(j) and h(t) transform. If the imaginary axis is in the ROC, the Fourier
transform exists for that function. In operator form
The Fourier transform (spectrum) of the impulse response h(t)
L{f (t)} = F (s) (204)
X
y(t) = yn (t). (190) of a system is the frequency response function H(j):
1
n= L {F (s)} = f (t). (205)
H(j) = F {h(t)} (196)
h(t) = F 1
{H(j)}. (197) See [?, p 545] or Appendix B for common Laplace transforms,
14.3 Fourier analysis of transient waveforms which are typically used in practice.
(Fourier Transforms)
This completely characterizes the system in the frequency or
Some aperiodic (transient) functions, which do not have Fourier time domain, respectively. 15 Approaches to problems
This has important implications for measuring a system fre-
series representations, can be described by a Fourier Transform
quency response function H(j). We can provide an impulse
(also called the spectrum). These functions must be limited in 15.1 Specific mode excitation
time, occur only once, and decay to zero as time becomes large. input (strike it with a hammer at the input) and measure y(t)
A Fourier series of a periodic extension of the function is used in at the output, then take the Fourier transform to obtain H(j) These are typically stated in one of the following ways
the development of the periodic extension, then the interval be- (or transfer function), with which we can predict the output of
tween occurrences of the function taken to infinity, which gives 1. If thereh is no input
i and the initial condition vector is
a function to a variety of inputs.
the Fourier transform. x(0) = 3 1 6 , will the state vector approach zero
The Fourier transform is defined by the following Fourier
transform pair : 14.4.3 Convolution as time gets large?
Z Convolution in the time domain (see Section 8.4), is simply mul- 2. If there is no input, find an initial condition vector such
F (j) = f (t)ejt dt (191) tiplication in the frequency domain: that the state vector approaches zero as time gets large.
h i
1
Z F {f (t) ? g(t)} = F (j)G(j). (198) 3. With initial condition vector x(0) = 3 1 6 (or
f (t) = F (j)ejt d (192)
2 Conversely, convolution in the frequency domain, is a scaled zero), choose a B matrix such that (or with a given B,
multiplication in the time domain: will) the state vector approach zero as time gets large.
and can also be written in terms of the Fourier transform oper-
ator F , F 1 {F (j) ? G(j)} = 2f (t)g(t). (199) 4. There may be some variation of these with outputs instead
of states.
F (j) = F {f (t)} (193)
14.4.4 Frequency response of interconnected systems 5. There may be some variation of these with states or out-
1 put going to zero in a finite time.
f (t) = F {F (j)}. (194)
If two linear systems are connected in cascade (series), and pro-
See [1, pp 528-31] for the properties of the Fourier trans- vided the connection does not effect the output of the first sys- 15.1.1 Approaches
form, including existence, linearity, even and odd functions, tem, the overall frequency response is
There are four main ideas at work here: modal decomposition,
time-shifting, waveform energy, and Fourier transforms of the stability, controllability, and observability. The main approach
derivative of a function. H(j) = H1 (j)H2 (j). (200) here follows.
If two linear systems are connected in parallel (share the 1. Find the eigenvalues of A. Typically at least one is un-
14.4 Fourier transform-based properties of stable for these problems and they are distinct.
input and their outputs sum), the overall frequency response is
linear systems
2. Find the eigenvectors of A.
14.4.1 Response of linear systems to aperiodic inputs H(j) = H1 (j) + H2 (j). (201)
3. Depending on what is required, fully or partially construct
The output spectrum is the input spectrum scaled by the fre- the modal decomposition system with modal state variable
quency response function: 14.5 Laplace Transforms q.

Y (j) = H(j)U (j). (195) Many functions do not have Fourier transforms, including the 4. An initial state in the modal state variable q(0) will not
unit step and ramp functions. The Laplace transform is a gen- excite the unstable mode if it has no component in the
eralized form of the Fourier transform that exists for a much direction of the unstable mode, since the homogeneous
This leads to the following procedure for finding the output of broader range of functions.
a system due to an aperiodic input: The Laplace transform multiplies by a weighting function response can be written as (103),
1. compute the Fourier transform of the input w(t) = et to drive the integral to zero. Now the one-side
xh (t) = Met
transform (assuming f (t) = 0 t < 0) pair is
U (j) = F {u(t)}, = 1 m1 e1 t + 2 m2 e2 t + ... + n mn en t ,
Z
2. form the output spectrum as the product F (s) = f (t)est dt (202) and the initial conditions being applied determine the val-
0 ues of i , and i = 0 will correspond to an eigenvector
Y (j) = H(j)U (j), and 1
Z +j that has no component in the initial condition:
f (t) = F (s)est ds (203)
2j j x(0) = 1 m1 + 2 m2 + ... + n mn .
3. compute the inverse Fourier transform
where s = + j. The region of convergence (ROC) of the in- The best way to think about it is this: we want j cor-
y(t) = F 1 {Y (j)}. tegral in the s-plane is an important quantity for each Laplace responding to the unstable (j-th) eigenvector to be zero.
14.3 Fourier analysis of transient waveforms (Fourier Transforms) system dynamics page 19 of 26
 Lets simply solve for x(0) in terms of , make the j-
component of , the see what x(0) satisfies the equation.
We have
A Impendance-based modeling
Notes on Generalized Impedances by J. L. Garbini
Generalized impedances are an extension of the concept of electrical impedances to systems of other domains.
The table below lists the corresponding driving-point impedance definitions for five different energy modalities.

x(0) = Me0 Mechanical Mechanical Electrical Fluid Thermal

= MI Translational Rotational
Across Variable v, velocity , angular v, voltage p, pressure T, temperature
= M velocity
= 1 m1 + 2 m2 + ... + n mn Through Variable f, force T, torque i, current q, volumetric q, heat flow rate
flow
Impedance Z(s)
V ( s) ( s) V ( s) P ( s) T ( s)
Admittance Z( s) = Z( s) = Z( s) = Z( s) = Z( s) =
So we choose as anything with j = 0 (any and all 1 F ( s) T ( s) I ( s) Q( s) Q( s)
unstable modes), then multiply by M to find x(0). Some- Y ( s) =
Z( s)
thing interesting to note here is that mass, M: inertia, J: capacitor, C fluid capacitor, C thermal capacitor, C
A-Type 1 1 1 1 1

Impedance Z(s)
Ms Js Cs Cs Cs
= M1 x(0) damper, B r. damper, B resistor, R fluid resistor, R thermal resistor, R
D-Type 1 1 R R R
= M1 Mq(0) B B
= q(0) spring, K r. spring, K r inductor, L fluid inductor, L
T-Type s s Ls Ls
K Kr

This shows the connection between the original and de- Series and parallel combinations of impedances and admittances can be combined. In the following V and F
coupled systems. The decoupled system has the identity
represent the across and through variables respectively of any physical domain.
matrix as its eigenvectors (and are therefore orthogonal)
Series Combination Parallel Combination
Elements sharing a Elements sharing a common
(206) common through variable across variable are in parallel.
Z1 Z2
are in series.
The admittance of elements Y1 Y2 Y1+Y2
The impedance of elements connected in parallel is the
Note that the eigenvectors are not necessarily orthogonal connected in series is the Z sum of the individual
to each other (only if A = A| ). sum of the individual admittances.
impedances. Y = Y1 + Y2
Z1 = Z1 + Z2 1 1 Z1Z2
Z= = =
5. Controllability and observability criteria (especially 1) are Y 1
+
1 Z1 + Z2
useful when there is a nonzero input or the output is of Z1 Z2
interest. Specifically, in the modal decomposition state
model we can easily determine if a mode is being excited
by the input and if it is being observed in the output. Simple transfer functions can be determined from impedance/admittance properties.

Across Variable Divider Through Variable Divider

The complex amplitude of The complex amplitude of
References the across variable across a Z1 the through variable through
set of elements in series is a set of elements in parallel is
divided among the divided among the elements
[1] Derek Rowell and David N. Wormley. System Dynamics: Fs Y1 Y2 Y3 F3
elements in proportion Vs Z2 V2 in proportion their
An Introduction. Prentice Hall, 1997.
their impedances. admittances.
Z3 F ( s) Y2
T ( s) = 2 =
[2] William L Brogan. Modern Control Theory. Prentice Hall, Fs ( s) Y1 + Y2 + Y3
third edition, 1991. V ( s) Z2
T ( s) = 2 =
REFERENCES Vs ( s) Z1 + Z2 + Z3 system dynamics page 20 of 26
Thevenin and Norton equivalent networks are useful deriving transfer functions and in modeling systems that Measurement Loading
have a defined load impedance.
Across Variable Measurements

Thevenins Theorem Suppose that we wish to measure an across variable at the output of device
A linear two-terminal network is equivalent to an a device under test with a measurement instrument. The measurement
under Vo Zi
across variable source Ve in series with an equivalent measurement instrument is attached across the terminals of interest. instrument
Linear test
impedance Ze , where Network Of course we desired that the measured variable be undisturbed by
the connection of the instrument. That is, we want Vm to be as
Ze = the impedance of the network with all sources set nearly equal to Vo as possible. We say that the measurement device F
equal to zero, and instrument should not load the device under test. under Vm Zi measurement
Ze instrument
Linear test
Ve = an across variable source equal to the across Ve Ve The output impedance of the device under test is the equivalent
variable that would appear across the open circuit
Network impedance defined by its Thevenin model Zo = Ze for the unloaded
terminals of the network. output terminals. Ze
Ve Zi
Similarly, the input impedance Zi of the measurement instrument is
the Thevenin equivalent impedance defined for its input terminals.
Nortons Theorem
A linear two-terminal network is equivalent to a Connecting the Thevenin model for the device under test to the
through variable source Fe in parallel with an equivalent Linear input impedance of the measurement instrument we have the
impedance Ze , where Network network at the right.

The Thevenin equivalent across variable source is by definition Zo

Ze = the impedance of the network with all sources set
equal to Vo , the value that we wish to measure. Applying the across
equal to zero, and Ve Vm Zi
V ( s) 1
Fe Ze Linear
Fe variable divider rule: m = .
Fe = a through variable source equal to the through Network Vo ( s) 1 + Zo Zi
variable that would flow through the short
circuited terminals of the network. Since we desire that the ratio approach unity, the input impedance
of the measurement instrument must be large in comparison with
the output impedance of the device under test: Zi >> Zo

Source Transformations
Through Variable Measurements
Since any linear two-terminal networks can be Ze
represented by either a Thevenin equivalent or a Fe Ze Alternately, suppose that we wish to measure a through variable in device
Ve measurement
Norton equivalent, the two representations must be a device under test with a measurement instrument. In this case, the under Fo Yi
instrument
equivalent to each other. variable of interest flows through the measurement instrument. We test
desired that the measured variable be undisturbed by the connection
Ve of the instrument. That is, we want Fm to be as nearly equal to Fo as
Fe = device
Ze possible. measurement
under Fm Yi
instrument
test
The output admittance of the device under test is the equivalent
admittance defined by its Nortons model Yo = 1 Ze for the
unloaded output terminals.
Fe Yo Yi
Similarly, the input admittance Yi of the measurement instrument is
the Norton equivalent admittance defined for its input terminals.

Connecting the Norton model for the device under test to the input
admittance of the measurement instrument we have the network at
the right.

The Norton equivalent through variable source is by definition equal

to Fo , the value that we wish to measure. Applying the through
Fe Yo Fm Yi
F ( s) 1
variable divider rule: m = .
Fo ( s) 1 + Yo Yi

Since we desire that the ratio approach unity, the input admittance
of the measurement instrument must be large in comparison with
the output admittance of the device under test: Yi >> Yo

system dynamics page 21 of 26

B Laplace transforms
B.1 Laplace transform properties

system dynamics page 22 of 26

B.2 Laplace transform pairs

B.2 Laplace transform pairs system dynamics page 23 of 26

C Fourier transforms C.2 Fourier transform pairs

C.1 Fourier transform properties

system dynamics page 24 of 26

D Periodic input response

System Response to Periodic Inputs

2
Input 0 = = Fundamental Frequency

Magnitude,

An

Line Spectrum
0 0 20 30 40 50 60 70 80 90
of
Input Input
Phase,

n 0

System
Magnitude,

|H(j)|
0 Frequency
Response
of
System System
Phase, 0

H(j)

Output
Magnitude,

An|H(jn0)|
0 0 20 30 40 50 60 70 80 90 Line Spectrum
of
Output Output
Phase, 0
n + H(jn0)

system dynamics page 25 of 26