Page |1
Ordinary Differential Equations (ODEs)
1. Notation
The systems we wish to control are modelled by ordinary differential equations
(ODEs) with constant coefficients. In this unit, we review some key concepts from
MATH 375 about ODEs and introduce some new notation. This basic knowledge is
essential for the analysis and design of feedback control systems. In the next unit,
we will build on these concepts to create an intuitive yet rigorous mathematical
framework for this course.
An ODE with constant coefficients has the general form
ππππ π¦π¦ (ππ) (π‘π‘) + ππππ−1 π¦π¦ (ππ−1) (π‘π‘) + β― + ππ1 π¦π¦ ′ (π‘π‘) + ππ0 π¦π¦(π‘π‘)
= ππππ π’π’(ππ) (π‘π‘) + ππππ−1 π’π’(ππ−1) (π‘π‘) + β― + ππ1 π’π’′ (π‘π‘) + ππ0 π’π’(π‘π‘),
(1)
for all π‘π‘ ∈ (−∞, ∞), where π¦π¦(π‘π‘) is the system output and π’π’(π‘π‘) is the system input.
These functions of time are called signals and represent changing physical
variables, such as force, position, speed, and voltage. Although these signals are
real valued (i.e. π’π’(π‘π‘), π¦π¦(π‘π‘) ∈ β), we sometimes consider complex-valued signals
π’π’(π‘π‘), π¦π¦(π‘π‘) ∈ β, as their real parts give real-valued solutions of (1). The derivative of
π¦π¦(π‘π‘) is denoted by π¦π¦′(π‘π‘), and the nth derivative is denoted π¦π¦ (ππ) (π‘π‘). The coefficients
ππππ , ππππ are assumed real. The order of the system is given by ππ, and usually, ππ > ππ.
In this course, ODEs are used to represent 3 types of system: a) plants, which are
physical systems we wish to control; b) controls, which output a dynamic control
signal in response to a measured error (the control ODE is usually implemented in
software); and c) a closed-loop system (CLS) obtained by connecting the control to
the plant. The input to the CLS is usually a reference trajectory that the plant output
is desired to follow, and its output is the tracking error (which is kept small by the
control action).
In this unit, we consider the inputs and outputs to be smooth (i.e., infinitely
differentiable) signals, denoted π’π’, π¦π¦ ∈ ππ ∞ . This notation is read as, “π’π’ and π¦π¦ are
both elements of the set ππ ∞ (the set of all smooth signals)”. The set ππ ∞ is called a
linear space (or vector space), and its elements (smooth signals) are (infinite
dimensional) vectors.
Page |2
Note that whereas π¦π¦ ∈ ππ ∞ is a signal, π¦π¦(π‘π‘) ∈ β is a number. The signal π¦π¦′ ∈ ππ ∞ is
the derivative of π¦π¦, and π¦π¦′(π‘π‘) ∈ β is the value of the derivative at time π‘π‘. Thus,
differentiation applies to a signal π¦π¦ ∈ ππ ∞ , not a number π¦π¦(π‘π‘) ∈ β.
Some signals are defined by a formula that maps each π‘π‘ ∈ β into π¦π¦(π‘π‘) ∈ β. We can
name these signals by their formulas using the notation π¦π¦ = {π¦π¦(π‘π‘)}. The curly
braces create a signal π¦π¦ from its formula π¦π¦(π‘π‘). For example, π¦π¦ = {5π‘π‘ 2 } is the signal
defined by π¦π¦(π‘π‘) = 5π‘π‘ 2 for all π‘π‘ ∈ (−∞, ∞). This notation avoids having to assign
variable names (like π¦π¦) to every signal we wish to reference. For example, the
statement “for all π‘π‘ ∈ β, if π¦π¦(π‘π‘) = 5π‘π‘ 2 , then π¦π¦′(π‘π‘) = 10π‘π‘” can be expressed more
concisely and anonymously as {5π‘π‘ 2 }′ = {10π‘π‘}.
ππ
The derivative operator is denoted by the symbol π·π·, i.e. π·π· = ππππ
. Hence, for π¦π¦ ∈ ππ ∞ ,
π·π·π·π· = π¦π¦ ′ = π¦π¦Μ , where the symbols π·π·, prime, and dot all mean differentiation. For
example, π·π·{3} = {3}′ = {0}, which says that the derivative (i.e. slope) of the
constant signal {3} is the zero signal. Signals should not be confused with numbers.
For example, the number 3 represents a scalar operator on signals (i.e. scaling by
3) whereas {3} is a signal that has a value of 3 at every time π‘π‘ . Similarly, the signal
π·π·{3} = {0} is different from the operator π·π·3 = 3π·π·, which multiplies a signal by 3
and then differentiates it (or vice versa, since it does not matter in which order
these 2 operators are applied). An exception to this distinction is that we use the
symbol 0 to mean both the number zero and the signal {0}, as this does not cause
any problems.
Solving (1) is greatly facilitated by operator notation. Since π¦π¦ ′ = π·π·π·π·, π¦π¦ ′′ = π·π·2 π¦π¦,
and π¦π¦ (ππ) = π·π·ππ π¦π¦, we may write (1) concisely as
ππ(π·π·)π¦π¦ = π§π§(π·π·)π’π’,
(2)
where ππ(π·π·) = ππππ π·π·ππ + ππππ−1 π·π·(ππ−1) + β― + ππ1 π·π· + ππ0 and π§π§(π·π·) = ππππ π·π·ππ +
ππππ−1 π·π·(ππ−1) + β― + ππ1 π·π· + ππ0 are polynomial differential operators. We may
divide (2) through by ππππ to make the leading coefficient of ππ(π·π·) equal to 1.
Page |3
2. Natural Response of a First Order System
If π’π’ = 0 in (2), we obtain the homogeneous ODE
ππ(π·π·)π¦π¦ = 0.
(3)
The general solution of (3) is called the natural response (or free response or
unforced response) of the system (2) since the input π’π’ = 0. The natural response
is important because it defines whether the system is stable and how stable it is,
which is a measure of performance. Fortunately, finding the natural response (i.e.
solving (3)) is easy for any order ππ of ππ.
If the ODE (3) is first order, it reduces to π¦π¦ ′ (π‘π‘) + ππ0 π¦π¦(π‘π‘) = 0, or if we drop the
subscript on the coefficient ππ0 ,
π¦π¦ ′ (π‘π‘) + ππππ(π‘π‘) = 0,
(4)
which corresponds to ππ(π·π·) = π·π· + ππ in (3). The solution of (4) is π¦π¦(π‘π‘) = ππππ −ππππ ,
where ππ ∈ β is an arbitrary constant. This may be verified by substituting π¦π¦(π‘π‘) =
ππππ −ππππ into (4) and checking that the left-hand side reduces to zero.
In (3), we say that the polynomial differential operator ππ(π·π·) annihilates π¦π¦ because
it maps π¦π¦ to zero. Therefore, solving (3) amounts to finding all signals π¦π¦ that are
annihilated by ππ(π·π·). For the first order case (4) , the proof of this annihilation is
ππ(π·π·)π¦π¦ = (π·π· + ππ){ππππ −ππππ }
= {−ππππππ −ππππ + ππππππ −ππππ }
= {0}.
(5)
The general solution π¦π¦(π‘π‘) = ππππ −ππππ represents an infinite number of solutions since
ππ is arbitrary. To determine the value of ππ, additional information about the output
is needed, such as the initial value π¦π¦(0) of the output. Setting π‘π‘ = 0 in the
solution π¦π¦(π‘π‘) = ππππ −ππππ gives π¦π¦(0) = ππππ −ππ0 = ππ β 1 = ππ, and therefore the solution
may be written in terms of the initial condition as π¦π¦(π‘π‘) = π¦π¦(0)ππ −ππππ .
However, in practice, we usually don’t know (or care about) any initial conditions
or the values of arbitrary constants in the natural response. The only important
constant is ππ, which depends not on the input or initial conditions, but only on the
system (4) itself.
Page |4
If −ππ < 0, then the solution π¦π¦(π‘π‘) = ππππ −ππππ decays to zero and is called stable. If
ππ = 0, then π¦π¦(π‘π‘) does not change from its initial value and is called marginally
unstable. If −ππ > 0, then π¦π¦(π‘π‘) blows up exponentially and is called unstable. A key
requirement of feedback control is to give a stable closed-loop system.
3. Natural Response of an n-th Order ODE
Let us now generalize the results of the previous section to higher order systems.
The roots (or zeros) of a polynomial function ππ(π π ) are the values of π π ∈ β for which
ππ(π π ) = 0.The Fundamental Theorem of Algebra (FTA) states that a polynomial ππ of
order ππ has ππ complex roots (including multiplicities). If the coefficients of the
polynomial are real, then its roots are either real or occur in complex conjugate
pairs.
A consequence of the FTA is that a polynomial differential operator ππ(π·π·) factors
into a product of real polynomials that are either first order (π·π· − ππ), which has a
real root ππ, or second order (π·π·2 + ππππ + ππ), which has 2 complex conjugate roots.
If a root of ππ(π·π·) is repeated, then the corresponding factor will also be repeated,
such as (π·π· − ππ)2 or (π·π·2 + ππππ + ππ)3 .
Suppose ππ(π·π·) factors as ππ(π·π·) = ππ1 (π·π·)ππ2 (π·π·). If ππ1 (π·π·)π¦π¦1 = 0, then ππ(π·π·)π¦π¦1 =
ππ2 (π·π·)ππ1 (π·π·)π¦π¦1 = ππ2 (π·π·)0 = 0. Similarly, if ππ2 (π·π·)π¦π¦2 = 0, then ππ(π·π·)π¦π¦2 = 0. Also,
since ππ(π·π·) is a linear operator, we conclude that ππ(π·π·)(ππ1 π¦π¦1 + ππ2 π¦π¦2 ) =
ππ1 ππ(π·π·)π¦π¦1 + ππ2 ππ(π·π·)π¦π¦2 = 0 for any constants ππ1 , ππ2 . Thus, π¦π¦ = ππ1 π¦π¦1 + ππ2 π¦π¦2 is a
solution of the homogeneous ODE ππ(π·π·)π¦π¦ = 0.
Applying the above reasoning to all irreducible factors of ππ(π·π·) gives us the general
solution of ππ(π·π·)π¦π¦ = 0 as the linear combination of all signals that are annihilated
by a factor of ππ(π·π·). Since these irreducible factors are only 1st or 2nd order
polynomials, possibly repeated, we only need to find out what signals are
annihilated by 1st or 2nd order polynomial differential operators to obtain the
natural response of any system.
Page |5
4. Natural Response of an n-th Order ODE with Real Roots
The simplest case is when all roots of ππ are all real and distinct. If ππ1 is a real root
of ππ, then π·π· − ππ1 is a factor of ππ(π·π·). Since {ππ1 ππ ππ1π‘π‘ } is annihilated by π·π· − ππ1 , it
solves ππ(π·π·)π¦π¦ = 0. If ππ2 is another real root of ππ, then π·π· − ππ2 is a factor of ππ(π·π·),
and hence {ππ2 ππ ππ2π‘π‘ } is another solution. Since ππ(π·π·) is a linear operator, it also
annihilates the linear combination {ππ1 ππ ππ1π‘π‘ + ππ2 ππ ππ2π‘π‘ }. If ππ is a second order, then
ππ1 and ππ2 are the only roots of ππ(π·π·), and {ππ1 ππ ππ1π‘π‘ + ππ2 ππ ππ2π‘π‘ } is the general solution.
Example 1: Find the general solution π¦π¦ of
(π·π·3 + 5π·π·2 + 6π·π·)π¦π¦ = 0.
(6)
π¦π¦ = {ππ1 + ππ2 ππ −2π‘π‘ + ππ3 ππ −3π‘π‘ }.
(7)
Solution: Since π·π·3 + 5π·π·2 + 6π·π· = π·π·(π·π· + 2)(π·π· + 3), and since π·π·{ππ 0π‘π‘ } = π·π·{1} =
0 and (π·π· + 2){ππ −2π‘π‘ } = 0 and (π·π· + 3){ππ −3π‘π‘ } = 0, the general solution is
Let us now consider the case where ππ(π·π·) has a repeated real root. The simplest
case is when the repeated root is ππ = 0. If this root is repeated ππ times, then ππ(π·π·)
includes the factor (π·π· − ππ)ππ = (π·π· − 0)ππ = π·π·ππ , which represents ππ-times
differentiation. Recall from calculus that the derivative operator π·π· annihilates
constant signals, π·π·2 annihilates linear signals, and more generally, π·π·ππ annihilates
any polynomial signal of order ππ − 1. Hence, π·π·ππ οΏ½ππππ−1 π‘π‘ (ππ−1) + β― + ππ1 π‘π‘ + ππ0 οΏ½ = 0,
and the solution of ππ(π·π·)π¦π¦ = 0 includes a general polynomial of order ππ − 1.
Before tackling the case of a nonzero real root ππ repeated ππ times, let us first derive
the exponential shift formula. Consider the derivative of {ππ(π‘π‘)ππ ππππ } where ππ ∈ ππ ∞
(i.e. ππ belongs to the space of infinitely differentiable signals). We apply the product
rule to obtain
π·π·{ππ(π‘π‘)ππ ππππ } = {ππ ′ (π‘π‘)ππ ππππ + ππ(π‘π‘)ππππ ππππ }
Applying (8) repeatedly gives
= {ππ ππππ ((π·π· + ππ)ππ)(π‘π‘)}.
π·π·ππ {ππ(π‘π‘)ππ ππππ } = {ππ ππππ ((π·π· + ππ)ππ ππ)(π‘π‘)},
(8)
(9)
which tells us that we can move the exponential function outside of the
differentiation operator if we shift π·π·ππ to (π·π· + ππ)ππ before applying it to the second
Page |6
signal ππ in the product. Applying (9) to each term of a polynomial differential
operator ππ(π·π·) gives
ππ(π·π·){ππ(π‘π‘)ππ ππππ } = {ππ ππππ (ππ(π·π· + ππ)ππ)(π‘π‘)}.
(10)
If ππ is a polynomial of order ππ − 1 and ππ(π·π·) = (π·π· − ππ)ππ , we obtain
(π·π· − ππ)ππ {ππ(π‘π‘)ππ ππππ } = {ππ ππππ ((π·π· − ππ + ππ)ππ ππ)(π‘π‘)}
= {ππ ππππ (π·π·ππ ππ)(π‘π‘)}
= {ππ ππππ (0)}
= 0,
(11)
since π·π·ππ annihilates any polynomial ππ of order ππ − 1. Hence, the general solution
of (π·π· − ππ)ππ π¦π¦ = 0 is
π¦π¦ = {ππ(π‘π‘)ππ ππππ },
(12)
where ππ is any real polynomial signal of order ππ − 1.
Example 2: Find the general solution π¦π¦ of
π·π·2 (π·π· + 2)3 (π·π· + 3)π¦π¦ = 0.
(13)
π¦π¦ = {ππ0 + ππ1 π‘π‘ + (ππ0 + ππ1 π‘π‘ + ππ2 π‘π‘ 2 )ππ −2π‘π‘ + ππππ −3π‘π‘ }.
(14)
Solution: Applying (12) to each factor in this equation gives
5. Natural Response of an ODE with Complex Roots
First consider a second order polynomial with imaginary roots: ππ(π·π·) = (π·π·2 + ππ2 ),
where ππ ∈ β is a real constant. The two roots of ππ are ±ππππ, where ππ 2 = −1. Since
π·π·2 {sin ππππ} = {−ππ2 sin ππππ}, we obtain (π·π·2 + ππ2 ){ππ1 sin ππππ} = 0. Similarly,
π·π·2 {cos ππππ} = {−ππ2 cos ππππ} gives (π·π·2 + ππ2 ){ππ2 cos ππππ} = 0. Together, these give
the general solution of ππ(π·π·)π¦π¦ = 0 for ππ(π·π·) = (π·π·2 + ππ2 ), namely
π¦π¦ = {ππ1 cos ππππ + ππ2 sin ππππ}.
(15)
In other words, since (π·π·2 + ππ2 ) annihilates both {sin ππππ} and {cos ππππ}, it
annihilates any linear combination of these signals.
Page |7
Now let us consider the more general case of a second order polynomial ππ(π π ) with
complex roots (not just imaginary ones). Such a polynomial may be written in the
form ππ(π π ) = (π π + ππ)2 + ππ2 , which has 2 complex conjugate roots at π π = −ππ ± ππππ
(check this by substitution).
Let us now use exponential shift to solve ((π·π· + ππ)2 + ππ2 )π¦π¦ = 0. Since (π·π· + ππ)2 +
ππ2 is obtained by shifting π·π· to π·π· + ππ in the expression π·π·2 + ππ2 , we apply
exponential shift (10) to the known solution (15) of (π·π·2 + ππ2 )π¦π¦ = 0 to obtain the
solution π¦π¦ of ((π·π· + ππ)2 + ππ2 )π¦π¦ = 0, namely
π¦π¦ = {ππ −ππππ (ππ1 cos ππππ + ππ2 sin ππππ)}.
(16)
Verify this by substituting this solution into ((π·π· + ππ)2 + ππ2 )π¦π¦ = 0 and applying
exponential shift.
Example 3: Find the general solution π¦π¦ of
(π·π·2 + 4π·π· + 13)π¦π¦ = 0.
(17)
π¦π¦ = {ππ −2π‘π‘ (ππ1 cos 3π‘π‘ + ππ2 sin 3π‘π‘)}.
(18)
Solution: We first express ππ(π·π·) = π·π·2 + 4π·π· + 13 in the form (π·π· + ππ)2 + ππ2 .
Setting these equal and comparing the π·π· terms on both sides gives ππ = 4/2 = 2.
Plugging ππ = 2 into the second expression and comparing the constant terms gives
ππ2 = 9, and so ππ = 3. Substituting these values into our general solution (16) gives
Finally, let us consider the case when the roots of ππ(π π ) are complex and repeated
ππ times. These roots occur as a complex conjugate pair ππ, ππΜ
, where ππ = −ππ + ππππ.
Hence, ππ(π·π·) = (π·π· − ππ)ππ (π·π· − ππΜ
)ππ = ((π·π· + ππ)2 + ππ2 )ππ . Since the exponential
shift rules (10) and (11) still hold when ππ is complex, we will follow the same
approach as in (11) to show that ππ(π·π·) annihilates ReοΏ½ππ(π‘π‘)ππ ππππ οΏ½ = {ππ(π‘π‘) ππ ππππ cos ππππ},
for any real polynomial ππ(π‘π‘) of order ππ − 1. Since ππ(π·π·) = ((π·π· + ππ)2 + ππ2 )ππ is a
real operator, it commutes with the Re operator (which extracts the real part of a
number or signal). This gives
ππ(π·π·){ππ(π‘π‘) ππ ππππ cos ππππ} = ππ(π·π·)ReοΏ½ππ(π‘π‘)ππ ππππ οΏ½
= Re(ππ(π·π·)οΏ½ππ(π‘π‘)ππ ππππ οΏ½)
Page |8
= Re((π·π· − ππΜ
)ππ (π·π· − ππ)ππ οΏ½ππ (π‘π‘)ππ ππππ οΏ½)
ππ
= Re(οΏ½π·π· − ππΜ
οΏ½ οΏ½ππ ππππ ((π·π· − ππ + ππ)ππ ππ)(π‘π‘)οΏ½)
ππ
= Re(οΏ½π·π· − ππΜ
οΏ½ οΏ½ππ ππππ (π·π·ππ ππ)(π‘π‘)οΏ½)
ππ
= Re(οΏ½π·π· − ππΜ
οΏ½ οΏ½ππ ππππ (0)οΏ½)
= 0.
(19)
Similarly, {ππ(π‘π‘) ππ ππππ sin ππππ} = ImοΏ½ππ(π‘π‘)ππ ππππ οΏ½ , and since ππ(π·π·) is real,
ππ(π·π·)ImοΏ½ππ(π‘π‘)ππ ππππ οΏ½ = Im(ππ(π·π·)οΏ½ππ(π‘π‘)ππ ππππ οΏ½). Applying the process of
(19) but taking imaginary parts instead of real parts gives ππ(π·π·){ππ(π‘π‘) ππ ππππ sin ππππ} =
0. Hence, ππ(π·π·) = ((π·π· + ππ)2 + ππ2 )ππ annihilates
π¦π¦ = {ππ −ππππ (ππ1 (π‘π‘) cos ππππ + ππ2 (π‘π‘) sin ππππ)},
(20)
where ππ1 (π‘π‘) andππ2 (π‘π‘) are arbitrary real polynomials of order ππ − 1. If ππ = 0, this
solution reduces to that obtained in (12) for a real repeated root −ππ, and if the root
is not repeated, then ππ1 (π‘π‘) and ππ2 (π‘π‘) reduce to real constants.
Example 4: Find the general solution π¦π¦ of
(π·π·2 + 4π·π· + 13)3 π¦π¦ = 0.
(21)
Solution: Since (18) gave the solution π¦π¦1 (π‘π‘) = {ππ −2π‘π‘ (ππ1 cos 3π‘π‘ + ππ2 sin 3π‘π‘)} for the
case of non-repeated roots, we simply replace each coefficient with a general
second order polynomial to obtain the general solution
π¦π¦ = {ππ −2π‘π‘ ((ππ10 + ππ11 π‘π‘ + ππ12 π‘π‘ 2 ) cos 3π‘π‘ + (ππ20 + ππ21 π‘π‘ + ππ22 π‘π‘ 2 ) sin 3π‘π‘)}. (22)
Note that this solution has 6 undetermined coefficients because it is the general
solution to a sixth order ODE. The values of these can be found if 6 initial conditions
are provided.
Page |9
6. Natural Signals
A signal is called natural if it is the natural response of some linear, constant
coefficient ODE (2). Equivalently, every natural signal is annihilated by some
polynomial differential operator ππ(π·π·). Let ππ denote the space of all natural signals.
We have shown that a natural signal is a sum of terms that each have the form (20),
which we call a natural term. Each natural term is annihilated by an operator ππ(π·π·)
that has a root ππ = −ππ + ππππ of multiplicity ππ; i.e., ππ(ππ) = 0. Both constants ππ and
ππ appear in the natural term (20).
The roots of ππ in (2) are called the poles of the ODE. The roots of π§π§ are called the
zeros of the ODE. Hence, the natural response is determined by the poles. Once
these poles are found (e.g. by factoring ππ manually or using Matlab), we obtain its
natural response by writing a natural term (20) for each pole and adding up all the
terms.
The simplest case of (20) is when ππ = 0, which gives sin ππππ = 0 and cos ππππ = 1.
In addition, if the multiplicity ππ = 1, then then ππ1 (π‘π‘) and ππ2 (π‘π‘) reduce to real
constants, and hence the natural response reduces to the first order response
π¦π¦(π‘π‘) = ππππ −ππππ considered in our first example.
Try sketching the first order response π¦π¦(π‘π‘) = ππππ −ππππ . Since this solution to the ODE
is valid for all π‘π‘ ∈ (−∞, ∞), the time axis should include negative times. If −ππ > 0,
then π¦π¦(π‘π‘) = ππππ −ππππ is small for large negative times and grows exponentially to the
right, called an unstable response. The plot interpolates the point π¦π¦(0) = ππ.
Although we call ππ the initial condition, it makes more sense to call it the middle
condition since, from the viewpoint of negative times, it occurs in the future.
If −ππ < 0, then π¦π¦(π‘π‘) = ππππ −ππππ grows large to the left and decays to the right, called
a stable response. This response is the mirror image about the π¦π¦ axis of the previous
case. If the sign of the constant ππ is opposite to what you assumed in your first plot,
then the plot is flipped about the time axis.
If ππ > 1, then the pole −ππ is repeated ππ times, and ππ1 (π‘π‘) and ππ2 (π‘π‘) become
polynomials of order ππ − 1. Still assuming that ππ = 0, these polynomials have little
effect on the growth rate at large positive or negative times (unless ππ = 0) since
exponential growth or decay is much faster than polynomial growth. However,
these polynomials do affect the response near the origin, such as the value of π¦π¦(0).
P a g e | 10
Now try plotting the case of ππ = 1, ππ = 0, and ππ ≠ 0. Then, the response (20)
reduces to π¦π¦ = {ππ1 cos ππππ + ππ2 sin ππππ}, which can also be written as π¦π¦ =
{π΄π΄ cos(ππππ + ππ)}, where π΄π΄ = οΏ½ππ1 2 + ππ2 2 is the resultant amplitude of the cosine
wave and ππ = atan2(ππ2 , ππ1 ) its phase shift. Your plot should therefore be that of a
cosine that has a peak value of π΄π΄, but shifted leftward by a time of ππ/ππ.
Finally, suppose both ππ ≠ 0, and ππ ≠ 0. Then, the response will be the cosine wave
of the last case but scaled by an exponential ππ −ππππ that decays to the right if −ππ <
0 and grows to the right if −ππ > 0. Try sketching this. If ππ > 1, then the response
will be similar, except near the origin where the polynomial factors have an effect.
A simple way of visualizing all these responses is as the real part of a complex valued
response οΏ½ππ(π‘π‘)ππ ππππ οΏ½, where ππ ∈ β and ππ(π‘π‘) ∈ β is a polynomial with complex
coefficients. Even if ππ(π‘π‘) is not constant, the behaviour of this response is primarily
due to the factor of ππ ππππ = ππ −ππππ ππ πππππ‘π‘ , which traces a spiral in the complex plane. The
ππ πππππ‘π‘ factor gives circular motion at an angular speed of ππ. The ππ −ππππ factor is the
radius, which spirals into the origin if −ππ < 0 and away from it if −ππ > 0. The real
part of this complex motion is the horizontal component, which is a decaying or
growing cosine wave.
The most important feature of all these natural responses is their decay (stability)
or growth (instability) as time increases. This depends on the number −ππ, which is
the real part of the pole. If there are multiple poles, then the real part of the
rightmost pole determines stability since if that pole is stable, then every pole to
the left of it is also stable. If the rightmost pole is unstable, then even if the other
poles are stable, the total response is still unstable since adding signals that decay
to zero to an unstable signal still gives an unstable signal.
7. Solving ODEs for any Natural Input
Now consider the forced ODE (2) with a natural input π’π’ ∈ ππ. We will show that the
output π¦π¦ is also natural and its poles are the poles of the ODE combined with those
of π§π§(π·π·)π’π’. If π¦π¦ = π¦π¦ππ is a particular solution of (2), then so is π¦π¦ = π¦π¦ππ + π¦π¦β , where π¦π¦β
is any solution to the homogeneous ODE (2). This may be verified by substituting
π¦π¦ = π¦π¦ππ + π¦π¦β into (2) and setting ππ(π·π·)π¦π¦β = 0. Hence, (2) has infinitely many
solutions.
P a g e | 11
The input term π§π§(π·π·)π’π’ of (2) with π’π’ ∈ ππ can be shown to be natural as follows. The
derivative π·π·π·π· of a polynomial signal ππ is a polynomial, and hence so is π·π·2 ππ, etc.
and hence so is π§π§(π·π·)ππ. Similarly, π·π·{ππ ππππ } = ππ{ππ ππππ }, π·π·2 {ππ ππππ } = ππ2 {ππ ππππ }, etc., and
hence π§π§(π·π·){ππ ππππ } = π§π§(ππ){ππ ππππ }, and so π§π§(π·π·) maps exponentials into exponentials.
Finally, consider a complex natural term οΏ½ππ (π‘π‘)ππ ππππ οΏ½, where ππ ∈ β and ππ(π‘π‘) ∈ β is a
polynomial with complex coefficients. The exponential shift property (10) gives
π§π§(π·π·)οΏ½ππ(π‘π‘)ππ ππππ οΏ½ = οΏ½ππ ππππ (π§π§(π·π· + ππ)ππ)(π‘π‘)οΏ½ = οΏ½ππ(π‘π‘)ππ ππππ οΏ½, where ππ = π§π§(π·π· + ππ)ππ is a
polynomial with complex coefficients. Hence, π§π§(π·π·) maps complex natural signals
into complex natural signals.
Since every real natural signal π’π’ is the real part of a complex natural signal π’π’ππ , and
since π§π§(π·π·) is a real operator, we have π§π§(π·π·)π’π’ = π§π§(π·π·)Re(π’π’ππ ) = Re(π§π§(π·π·)π’π’ππ ), which
is a real natural signal. Hence, π§π§(π·π·) maps real natural signals into real natural
signals. Therefore if π’π’ in (2) is a real natural signal, then so is π§π§(π·π·)π’π’ in (2).
To solve (2), we choose π¦π¦ to have the same form as the input π§π§(π·π·)π’π’ but with
undetermined coefficients. If ππ(π·π·) has a root in common with the poles of π§π§(π·π·)π’π’,
then the order of the corresponding polynomial in π¦π¦ must be ππ + ππ, where ππ is the
order of the corresponding polynomial in π§π§(π·π·)π’π’, and ππ is the multiplicity of the
common root. This accounts for any reductions in the order of any polynomial
factors in π¦π¦ caused by ππ(π·π·). Examples will demonstrate.
Example 5: Find the general solution π¦π¦ of
if π’π’ = {π‘π‘ππ π‘π‘ + ππ 2π‘π‘ }.
(π·π· + 2)(π·π· + 1)π¦π¦ = (π·π· + 3)π’π’,
(23)
Solution: Applying (π·π· + 3) to π’π’ gives (π·π· + 3){π‘π‘ππ π‘π‘ + ππ 2π‘π‘ } = {(4π‘π‘ + 1)ππ π‘π‘ + 5ππ 2π‘π‘ }.
This modified input to (23) produces an output of the form π¦π¦ = π¦π¦ππ =
{(ππ0 + ππ1 π‘π‘)ππ π‘π‘ + ππππ 2π‘π‘ }. Here, the polynomials in the output are chosen to have the
same order as those in the input so that the coefficients can be matched. Applying
(π·π· + 2)(π·π· + 1) to π¦π¦ππ and equating the result to the input {(4π‘π‘ + 1)ππ π‘π‘ + 5ππ 2π‘π‘ }
gives
(6ππ0 + 5ππ1 + 6ππ1 π‘π‘)ππ π‘π‘ + 12ππππ 2π‘π‘ = (1 + 4π‘π‘)ππ π‘π‘ + 5ππ 2π‘π‘ .
(24)
Solving for the 3 unknown coefficients gives ππ0 = − 7⁄18, ππ1 = 2/3, ππ = 5/12.
7
5 2π‘π‘
ππ . The general unforced
This gives the particular solution π¦π¦ππ = οΏ½23π‘π‘ − 18
οΏ½ππ π‘π‘ + 12
P a g e | 12
solution π¦π¦β is obtained by setting π’π’ = 0 in (23), which gives π¦π¦β = ππ1 ππ −π‘π‘ + ππ2 ππ −2π‘π‘ .
Adding this solution to π¦π¦ππ gives the general solution to the forced ODE as
2
7
(25)
5
π¦π¦(π‘π‘) = οΏ½3π‘π‘ − 18οΏ½ ππ π‘π‘ + 12ππ 2π‘π‘ + ππ1 ππ −π‘π‘ + ππ2 ππ −2π‘π‘ .
π·π·+3
2+3
5 2π‘π‘
The term 12
ππ could also be found directly as (π·π·+2)(π·π·+1) {ππ 2π‘π‘ } = (2+2)(2+1) {ππ 2π‘π‘ } =
{ππ 2π‘π‘ } since π·π·{ππ 2π‘π‘ } = 2{ππ 2π‘π‘ }. If initial conditions π¦π¦(0) and π¦π¦′(0) are given, then
ππ1 and ππ2 can be determined.
5
12
Example 6: Find the general solution π¦π¦ of
if π’π’ = {π‘π‘ππ −π‘π‘ + ππ −3π‘π‘ }.
(π·π· + 2)(π·π· + 1)π¦π¦ = (π·π· + 3)π’π’,
(26)
Solution: Applying (π·π· + 3) to π’π’ gives (π·π· + 3){π‘π‘ππ −π‘π‘ + ππ −3π‘π‘ } = {(2π‘π‘ + 1)ππ −π‘π‘ }. This
modified input to (26) has pole ππ = −1 in common with a root of ππ(π·π·) =
(π·π· + 2)(π·π· + 1) of multiplicity ππ = 1. The corresponding polynomial (2π‘π‘ + 1) in
(π·π· + 3)π’π’ has order ππ = 1. Hence, the order of the corresponding polynomial in π¦π¦
must be ππ + ππ = 2. Therefore, we write the solution π¦π¦ = π¦π¦ππ = {(ππ1 π‘π‘ + ππ2 π‘π‘ 2 )ππ −π‘π‘ }
with coefficients to be determined. We omitted a constant term ππ0 because this
will be annihilated by (π·π· + 1) and will be added to the general solution anyway.
Substituting this solution and (π·π· + 3)π’π’ = {(2π‘π‘ + 1)ππ −π‘π‘ } into (26) gives
(π·π· + 2)(π·π· + 1){(ππ1 π‘π‘ + ππ2 π‘π‘ 2 )ππ −π‘π‘ } = {(2π‘π‘ + 1)ππ −π‘π‘ }.
(27)
Using exponential shift to evaluate the l.h.s. and comparing its terms with the r.h.s.
gives ππ2 = 1 and ππ1 = −1, so π¦π¦ = π¦π¦ππ = {(π‘π‘ 2 − π‘π‘)ππ −π‘π‘ } solves (26). The general
solution of (26) is this solution plus the general solution of the homogeneous ODE
obtained by setting π’π’ = 0 in (26). This gives π¦π¦ = {(π‘π‘ 2 − π‘π‘ + ππ0 )ππ −π‘π‘ + ππππ −2π‘π‘ }. This
general solution has 2 undetermined coefficients ππ0 and ππ since (26) is a secondorder ODE. These coefficients can be found from two initial conditions, such as
π¦π¦(0) = π¦π¦0 and π¦π¦′(0) = π¦π¦1 . Substituting π¦π¦(π‘π‘) = (π‘π‘ 2 − π‘π‘ + ππ0 )ππ −π‘π‘ + ππππ −2π‘π‘ into
these initial conditions gives π¦π¦0 = ππ0 + ππ and π¦π¦1 = −(1 + ππ0 + 2ππ), which may
then be solved for ππ0 and ππ.
This example illustrates two tricky features that can occur when solving ODEs. The
first is that the term ππ −3π‘π‘ in the input π’π’ did not appear in the output π¦π¦. This is
P a g e | 13
because it was annihilated by π§π§(π·π·) = π·π· + 3. The modified input of π§π§(π·π·)π’π’ =
{(2π‘π‘ + 1)ππ −π‘π‘ } determines the form of π¦π¦. The other tricky feature is that the output
term (π‘π‘ 2 − π‘π‘ + ππ0 )ππ −π‘π‘ contained a higher-order polynomial than that of the input
term (2π‘π‘ + 1)ππ −π‘π‘ that produced it. This is because one of the poles of the ODE,
namely ππ = −1 , happened to match a pole of the input π’π’. This is called pure
resonance.
The fact that an ODE has an infinite number of solutions π¦π¦ for every given input π’π’
is undesirable. We can avoid this by requiring that present values of the output do
not depend on future values of the input. This causality constraint turns the ODE
into an operator, which maps each input into a unique output.
In the next unit, we will develop an algebra for such operators (and more general
operators) where operators behave just like real or complex numbers, providing a
simple mathematical framework for feedback control analysis and design.
8. Frequency Response
To conclude this unit, let us consider the response of an ODE to 2 special natural
inputs that are closely related: exponentials and sinusoids.
Consider the ODE (2) with a complex exponential input π’π’ = {ππππ π π π π }, where ππ, π π ∈ β.
Let us assume the constraint of the previous section: that the real part of π π is
greater than the real part of every pole of the ODE. This implies that ππ(π π ) ≠ 0. The
unique solution is then π¦π¦ = πΊπΊ(π π )π’π’, where the complex constant πΊπΊ(π π ) =
π§π§(π π )/ππ(π π ). Verify this solution by substitution. If we use ππ to denote the operator
from input signals to output signals, then the function πΊπΊ(π π ) is called the
(operational) Laplace transform of ππ. The number πΊπΊ(π π ) is the eigenvalue of ππ
corresponding to the eigensignal π’π’ = {ππππ π π π π }.
A special case of this result is when π π = ππππ. Writing ππ in polar form as ππ = |ππ|ππ ππ∠ππ ,
where ∠ππ denotes the angle that ππ makes with the real axis, we find the real part
of π’π’ = {ππππ πππππ‘π‘ } to be
π’π’ππ = ReοΏ½ππππ πππππ‘π‘ οΏ½
= ReοΏ½|ππ|ππ ππ∠ππ ππ πππππ‘π‘ οΏ½
P a g e | 14
= |ππ|ReοΏ½ππ ππ(πππ‘π‘+∠ππ) οΏ½
= |ππ|{cos(πππ‘π‘ + ∠ππ)},
(28)
which we call a general sinusoid. For example, ππ = −ππ gives π’π’ππ = sin(πππ‘π‘). We will
now show that a solution of (2) for a general sinusoidal input is also a general
sinusoid at the same frequency but different amplitude and phase.
If π¦π¦ solves (2) when π’π’ = {ππππ πππππ‘π‘ }, then since ππ and π§π§ are real polynomials,
ππ(π·π·)π¦π¦ = π§π§(π·π·)π’π’
Re(ππ(π·π·)π¦π¦) = Re(π§π§(π·π·)π’π’)
ππ(π·π·)Re(π¦π¦) = π§π§(π·π·)Re(π’π’)
ππ(π·π·)π¦π¦ππ = π§π§(π·π·)π’π’ππ ,
and π¦π¦ππ = Re(π¦π¦) is obtained from the exponential solution π¦π¦ = πΊπΊ(ππππ)π’π’ as
(29)
π¦π¦ππ = ReοΏ½πΊπΊ(ππππ)ππππ πππππ‘π‘ οΏ½
= ReοΏ½|πΊπΊ(ππππ)|ππ ππ∠πΊπΊ(ππππ) |ππ|ππ ππ∠ππ ππ πππππ‘π‘ οΏ½
= |ππ||πΊπΊ(ππππ)|ReοΏ½ππ ππ(πππ‘π‘+∠ππ+∠πΊπΊ(ππππ)) οΏ½
= |ππ||πΊπΊ(ππππ)|{cos(πππ‘π‘ + ∠ππ + ∠πΊπΊ(ππππ))}.
(30)
Hence, the amplitude of the output π¦π¦ππ equals the amplitude of the input multiplied
by the magnitude of the complex number πΊπΊ(ππππ) = π§π§(ππππ)/ππ(ππππ), and the phase of
the output equals the phase of the input plus the angle of πΊπΊ(ππππ).
Example 7: Find the sinusoidal solution π¦π¦ of
if π’π’ππ = {sin 5π‘π‘}.
(π·π· + 2)(π·π· + 1)π¦π¦ππ = (π·π· + 3)π’π’ππ ,
Solution: Since πΊπΊ(ππππ) =
π§π§(ππππ)
ππ(ππππ)
π¦π¦ππ = 0.212{sin(5π‘π‘ − 1.533)}.
(31)
3+5ππ
= (2+5ππ)(1+5ππ) = 0.212ππ −1.533 , the solution is
The complex-valued function ππ(ππ) = πΊπΊ(ππππ) of the real variable ππ is called the
frequency response of the system (2). Its magnitude is called the magnitude
P a g e | 15
response of the system, and its angle is called the phase response. The magnitude
versus frequency is typically plotted on log-log axes, and the phase versus
frequency is plotted on linear-log axes. These plots are called Bode plots and give
a complete picture of how the system responds to sinusoids. The system bandwidth
is defined as the range of input frequencies ππ for which |πΊπΊ(ππππ)| ≥ |πΊπΊ(0)|/√2,
where the output is considered to respond well to the input.
0
You can add this document to your study collection(s)
Sign in Available only to authorized usersYou can add this document to your saved list
Sign in Available only to authorized users(For complaints, use another form )