Fundamentals of Linear Control:
A Concise Approach

Maurício C. de Oliveira

Chapter 3: Transfer-Function Models

linearcontrol.info/fundamentals

3.1 The Laplace Transform

Definition

\[\begin{align*} \label{eq:laplace} F(s) &= \mathcal{L} \{ f(t) \} = \int_{0^-}^{\infty} f(t) \, e^{-s t} \, dt \end{align*}\]

Example

\[\begin{align*} f(t) &= 1 & F(s) &= \int_{0^-}^{\infty} e^{-s t} \, dt = \left . \frac{- e^{-s t}}{s} \right |_{0^-}^\infty = 0 - \frac{-1}{s} = \frac{1}{s}, \qquad \operatorname{Re}(s) > 0 \end{align*}\]

Region of convergence

\[\begin{align*} \operatorname{Re}(s) > \alpha_c = 0 \end{align*}\]

Inverse Laplace Transform

\[\begin{align*} \mathcal{L}^{-1} \{ F(s) \} &= \frac{1}{2 \pi j} \lim_{\omega \rightarrow \infty} \int_{\alpha - j \omega}^{\alpha + j \omega} F(s) \, e^{s t} \, ds, & \alpha > \alpha_c \end{align*}\]

See book for a lot more!

Table of Transform Pairs

Table of Properties

3.2 Linearity, Causality, and Time-Invariance

Dynamic model

\[\begin{align*} y(t) = G(u(t),t) \end{align*}\]

Linearity

\[\begin{align*} y_1(t) &= G(u_1(t),t), & y_2(t) &= G(u_2(t),t), & y(t) &= G(\alpha_1 \, u_1(t) + \alpha_2 \, u_2(t), t) = \alpha_1 \, y_1(t) + \alpha_2 \, y_2(t) \end{align*}\]

Time-invariance

\[\begin{align*} y(t) &= G(u(t),t), & y(t, \tau) &= G(u(t - \tau),t) = y(t - \tau) \end{align*}\]

Impulse response

Sifting property of the impulse

\[\begin{align*} u(t) &= \int_{-\infty}^\infty \! u(\tau) \, \delta(t - \tau) \, d\tau = \int_{0^-}^t \! u(\tau) \, \delta(t - \tau) \, d\tau, & u(t) &= 0, \quad t < 0 \end{align*}\]

Linearity

\[\begin{align*} y(t) &= \int_{0^-}^\infty u(\tau) \, g(t, \tau) \, d\tau = \int_{0^-}^\infty g(t, \tau) \, u(\tau) \, d\tau, & g(t, \tau) &= G(\delta(t - \tau), t) \end{align*}\]

Causality

\[\begin{align*} g(t, \tau) &= 0, \quad t < \tau & & \implies & y(t) &= \int_{0^-}^t g(t, \tau) \, u(\tau) \, d\tau \end{align*}\]

Impulse response

Causality + Time-invariance

\[\begin{align*} g(t, \tau) &= g(t - \tau) = 0, \quad t < \tau & & \implies & y(t) &= \int_{0^-}^t g(t - \tau) \, u(\tau) \, d\tau \end{align*}\]

Laplace transform

\[\begin{align*} Y(s) &= G(s) \, U(s), & U(s) &= \mathcal{L} \{ u(t) \}, & G(s) &= \mathcal{L} \{ g(t) \}, & Y(s) &= \mathcal{L} \{ y(t) \} \end{align*}\]

\(G\) is the transfer-function

3.3 Differential Equations and Transfer-Functions

Car model

\[\begin{align*} \dot{y}(t) + \frac{b}{m} \, y(t) - \frac{p}{m} u(t) &= \frac{p}{m} u(t) \end{align*}\]

Laplace transform

\[\begin{align*} \mathcal{L} \left \{ \dot{y}(t) + \frac{b}{m} \, y(t) - \frac{p}{m} u(t) \right \} &= s Y(s) - y(0^-) + \frac{b}{m} Y(s) - \frac{p}{m} U(s) = 0 \end{align*}\]

Transfer-function

\[\begin{align*} Y(s) &= G(s) \, U(s) + F(s) \, y(0^-), & G(s) &= \frac{\frac{p}{m}}{s + \frac{b}{m}}, & F(s) &= \frac{1}{s + \frac{b}{m}} \end{align*}\]

Impulse response

\[\begin{align*} g(t) &= \mathcal{L}^{-1} \left \{ \frac{\frac{p}{m}}{s + \frac{b}{m}} \right \} = \frac{p}{m} e^{-(\frac{b}{m})t}, \quad t \geq 0 \end{align*}\]

3.4 Integration and Residues

Theorem 3.1. Cauchy’s residue theorem

If \(f\) is analytic inside and on the the positively oriented simple closed countour \(C\) except at singular points \(p_k\), \(k = 1, \ldots, n\), inside \(C\) then \[\begin{align*} \int_{C} f(s) \, ds &= 2 \pi j \sum_{k = 1}^{n} \operatorname{Res}_{s = p_k} f(s) \end{align*}\]

Notes

  • \(\operatorname{Res}_{s = p_k} f(s)\) is the residue of \(f\) at \(s = p_k\)
  • \(\operatorname{Res}_{s = p_k} f(s)\) is simpler to calculate than the original integral
  • \(C\) is positively oriented simple contour
  • Multiply by ‘\(-1\)’ if negatively oriented
  • singularities are isolated, inside and not on the contour
  • singularities outside \(C\) do not matter

Example

Contours

Function

\(\displaystyle f(s) = \frac{s - 1}{s (s + 1)}\)

Integrals

\[\begin{align*} \frac{1}{2 \pi j} \int_{C_1} f(s) \, ds &= \operatorname{Res}_{s = 0} f(s) + \operatorname{Res}_{s = -1} f(s) \\ \int_{C_2} f(s) \, ds &= - \int_{C_1} f(s) \, ds \\ \int_{C_3} f(s) \, ds &= 0 \end{align*}\]

Application: Inverse Laplace transform

\[\begin{align*} t &> 0: & \lim_{\rho \rightarrow \infty} \int_{C^\rho_-} F(s) \, e^{s t} \, ds &= 0 & & \implies & \mathcal{L}^{-1}\{F(s)\} &= \frac{1}{2 \pi j} \lim_{\rho \rightarrow \infty} \int_{\Gamma_-^\alpha} F(s) \, e^{s t} \, ds \\ & & & & & & &= \sum_{k = 1}^{n} \operatorname{Res}_{s = p_k} \left (F(s) \, e^{s t} \right ) \\ t &< 0: & \lim_{\rho \rightarrow \infty} \int_{C^\rho_+} F(s) \, e^{s t} \, ds &= 0 & & \implies & \mathcal{L}^{-1}\{F(s)\} &= \frac{1}{2 \pi j} \lim_{\rho \rightarrow \infty} \int_{\Gamma_+^\alpha} F(s) \, e^{s t} \, ds = 0 \end{align*}\]

Residue

\[\begin{align*} \operatorname{Res}_{s = s_0} f(s) &= \int_{C \in D} f(s) \, ds, & D &= \{ s : 0 < |s - s_0| < R \} \end{align*}\]

Laurent series expansion

\[\begin{align*} f(s) &= \frac{ g(s)}{(s - s_0)^m}, \quad g(s_0) \neq 0, \quad g \text{ is analytic at } s_0 \end{align*}\]

Calculus of residues

\[\begin{align*} \operatorname{Res}_{s = s_0} f(s) &= \begin{cases} g(s_0), & m = 1 \\[1ex] \displaystyle \frac{g^{(m -1)}(s_0)}{(m-1)!} , & m \geq 2 \end{cases} \end{align*}\]

3.5 Rational Functions

Canonical form

\[\begin{align*} F(s) &= \frac{N(s)}{D(s)}, & & N, D \text{ are polynomials} \end{align*}\]

Notes

  • Zeros: \(N(s) = \beta \, (s - z_1) \cdots (s - z_m) = 0\)
  • Poles: \(D(s) = (s - p_1) \cdots (s - p_n) = 0\), \(D\) is monic
  • Order: \(n\)
  • Proper: \(m \leq n\)
  • Strictly Proper: \(m < n\)

Inverse Laplace transform

\[\begin{align*} & F \text{ is strictly proper} & & \implies & \mathcal{L}^{-1}\{F(s)\} &= \sum_{k = 1}^{n} \operatorname{Res}_{s = p_k} \left (F(s) \, e^{s t} \right ), \quad t \geq 0 \end{align*}\]

Simple poles

\[\begin{align*} F(s) \, e^{s t} &= \frac{ G(s)}{(s - p_k)}, & G(s) &= (s - p_k) \, F(s) \, e^{s t}, & G(p_k) &\neq 0 \end{align*}\]

Residues

\[\begin{align} \operatorname{Res}_{s=p_k} \left ( F(s) \, e^{s t} \right ) &= G(p_k) = k_k \, e^{p_k t}, & k_k &= \lim_{s \rightarrow p_k} (s - p_k) F(s) \end{align}\]

Inverse Laplace transform

\[\begin{align*} \mathcal{L}^{-1}\{F(s)\} &= \sum_{k = 1}^{n} k_k e^{p_k t}, \quad t \geq 0 \end{align*}\]

Expansion in partial fractions

\[\begin{align*} F(s) &= \mathcal{L} \left \{ \sum_{k = 1}^{n} k_k e^{p_k t} \right \} = \sum_{k = 1}^{n} k_k \mathcal{L} \left \{ e^{p_k t} \right \} = \sum_{k = 1}^{n} \frac{k_k}{s - p_k} \end{align*}\]

Example: car model step response

Step input

\[\begin{align*} u(t) &= \tilde{u} \, 1(t), & U(s) &= \frac{\tilde{u}}{s} \end{align*}\]

Laplace transform of the output

\[\begin{align*} Y(s) &= \frac{\tilde{u}}{s} \times \frac{\frac{p}{m}}{s + \frac{b}{m}} + \frac{1}{s + \frac{b}{m}} \, y_0 \end{align*}\]

Inverse Laplace transform

\[\begin{align*} k_1 &= \lim_{s \rightarrow 0} s \, Y(s) = \frac{p}{b} \tilde{u}, & k_2 &= \!\!\!\lim_{s \rightarrow -\frac{b}{m}} \left ( s + \frac{b}{m} \right ) Y(s) = y_0 - \frac{p}{b} \tilde{u} \end{align*}\] \[\begin{align*} y(t) &= \sum_{k = 1}^2 \operatorname{Res}_{s = p_k} \left (Y(s) e^{s t} \right ) = k_1 e^{p_1 t} + k_2 e^{p_2 t} = \frac{p}{b} \tilde{u} + \left ( y_0 - \frac{p}{b} \tilde{u} \right ) e^{-\frac{b}{m} t}, \quad t > 0 \end{align*}\]

Proper functions

Split \(F\) as \(F_1 + F_2\) where \(F_1\) is a constant and \(F_2\) is strictly proper, e.g. \[\begin{align*} F(s) &= \frac{s}{s + 1} = F_1(s) + F_2(s), & F_1(s) &= 1, & F_2(s) &= - \frac{1}{s + 1} \end{align*}\]

Repeated pole

\[\begin{align*} F(s) \, e^{s t} &= \frac{G(s)}{(s - s_k)^m}, & G(s) &= (s - s_k)^m F(s) \, e^{s t}, & G(s_k) &\neq 0, & m &\geq 2 \end{align*}\]

Residue

\[\begin{align*} \operatorname{Res}_{s = p_k} \left (F(s) e^{s t} \right ) = \frac{G^{(m-1)}(p_k)}{(m-1)!} = & \left ( k_{k,1} t^{m-1} + k_{k,2} t^{m-2} + \cdots + k_{k,m} \right ) e^{p_k t} \\ k_{k,j} &= \frac{1}{(m-j)!(j-1)!} \lim_{s \rightarrow p_k} \frac{d^{j-1}}{d s^{j-1}} (s - p_k)^{m} F(s) \end{align*}\]

Example: closed-loop car model ramp response

Ramp input

\[\begin{align*} \bar{y}(t) &= \mu \, t \, 1(t), & \bar{Y}(s) &= \frac{\mu}{s^2} \end{align*}\]

Laplace transform of the output

\[\begin{align*} Y(s) &= H(s) \bar{Y}(s) = \frac{b_1}{s + a_1} \times \frac{\mu}{s^2}, & a_1 &= \frac{b}{m} + \frac{p}{m} K, & b_1 &= \frac{p}{m} K \end{align*}\]

Inverse Laplace transform

\[\begin{align*} k_1 &= \lim_{s \rightarrow -a_1} ( s + a_1 ) Y(s) = \frac{\mu \, b_1}{a_1^2}, & k_{2,1} &= \lim_{s \rightarrow 0} s^2 Y(s) = \frac{\mu \, b_1}{a_1}, & k_{2,2} &= \lim_{s \rightarrow 0} \frac{d}{d s} s^2 Y(s) = - \frac{\mu \, b_1}{a_1^2} \end{align*}\] \[\begin{align*} y(t) &= \sum_{k = 1}^2 \operatorname{Res}_{s = p_k} \left (Y(s) e^{s t} \right ) = \frac{\mu \, b_1}{a_1^2} \left [ e^{-a_1 t} + a_1 \, t - 1 \right ], \quad t \geq 0 \end{align*}\]

3.6 Stability

Bounded-Input-Bounded-Output (BIBO) stability

\[\begin{align*} | u(t) | &\leq M_u < \infty & & \implies & | y(t) | &\leq M_y < \infty \end{align*}\]

Necessary and sufficient condition

\[\begin{align} \|g\|_1 := \int_{0^-}^\infty |g(\tau)| \, d \tau < M_g < \infty \end{align}\]

Proof of sufficiency

\[\begin{align*} |y(t)| &= \int_{0^-}^t |g(\tau)| |u(t - \tau)| \, d \tau \leq M_u \int_{0^-}^t |g(\tau)| \, d \tau \leq M_y, \quad M_y = M_u M_g < \infty \end{align*}\]

Bounded-Input-Bounded-Output (BIBO) stability

\[\begin{align*} | u(t) | &\leq M_u < \infty & & \implies & | y(t) | &\leq M_y < \infty \end{align*}\]

Necessary and sufficient condition

\[\begin{align} \|g\|_1 := \int_{0^-}^\infty |g(\tau)| \, d \tau < M_g < \infty \end{align}\]

Proof of necessity

\[\begin{align*} u_T(t) &= \begin{cases} \operatorname{sign}(g(T - t)), & 0 \leq t \leq T \\ 0, & t < 0 \text{ or } t > T \end{cases} \end{align*}\] is such that \(|u_T(t)| \leq 1\) and \[\begin{align*} y_T(T) &= \int_{0^-}^T g(T - \tau) \, u_T(\tau) \, d \tau = \int_{0^-}^T |g(\tau)| \, d \tau \rightarrow \|g\|_1 \text{ for } T \text{ large } \end{align*}\]

Frequency-domain

Asymptotic stability

\(G(s)\) has no poles on the imaginary axis or on the right-hand side of the complex plane

Same as BIBO stability

One side is easy: because \(|e^{-s t}| \leq 1\) for all \(\operatorname{Re}(s) \geq 0\) \[\begin{align*} |G(s)| = | \mathcal{L} \{ g(t) \} | = \left | \int_{0^-}^\infty g(t) \, e^{-s t} \, dt \right | \leq \int_{0^-}^\infty |g(t)| |e^{-s t}| \, dt \leq \int_{0^-}^\infty |g(t)| \, dt = \|g\|_1 \end{align*}\] hence \(G\) has no poles such that \(\operatorname{Re}(s) \geq 0\) because \(G\) is bounded in \(\operatorname{Re}(s) \geq 0\)

Frequency-domain

Asymptotic stability

\(G(s)\) has no poles on the imaginary axis or on the right-hand side of the complex plane

Same as BIBO stability

The other side is more complicated, unless \(G\) is rational. In this case, if \(G\) is strictly proper and with simple poles \[\begin{align*} |g(t)| &\leq \sum_{k = 1}^{n} |k_k| |e^{p_k t}| = \sum_{k = 1}^{n} |k_k| e^{\operatorname{Re}(p_k) t}, & t &> 0 \end{align*}\] and because \(\operatorname{Re}(p_k) < 0\) \[\begin{align*} \| g \|_1 \leq \int_{0^-}^{\infty} |g(t)| \, dt \leq \sum_{k = 1}^{n} |k_k| \int_{0^-}^{\infty} e^{\operatorname{Re}(p_k) t} \, dt = \sum_{k = 1}^{n} \frac{|k_k|}{-\operatorname{Re}(p_k)} < \infty \end{align*}\] Conclusion is the same if \(G\) is proper and/or has repeated poles

3.7 Transient and Steady-State Response

Frequency-domain splitting

\[\begin{align*} Y(s) &= Y_{-}(s) + Y_{0}(s) + Y_{+}(s) \end{align*}\]

  • \(Y_-\) has all poles on \(\operatorname{Re}(s) < 0\)
  • \(Y_0\) has all poles on \(\operatorname{Re}(s) = 0\)
  • \(Y_+\) has all poles on \(\operatorname{Re}(s) > 0\)

Time-domain splitting

\[\begin{align*} y(s) &= y_{-}(s) + y_{0}(s) + y_{+}(s) \end{align*}\]

  • \(y_{\mathrm{tr}}(t) = y_-(t) = \mathcal{L}^{-1} \{ Y_-(s) \}\) is the transient response
  • \(y_{\mathrm{ss}}(t) = y_0(t) = \mathcal{L}^{-1} \{ Y_0(s) \}\) is the steady-state response, bounded or with at most polynomial growth
  • \(y_+(t) = \mathcal{L}^{-1} \{ Y_+(s) \}\) is unbounded with exponential growth

3.8 Frequency Response

Time-invariant linear system

Transfer-function

\[\begin{align*} G(s) \end{align*}\]

Input signal

\[\begin{align} u(t) &= A \cos(\omega t + \phi) \end{align}\]

Steady-state output response

\[\begin{align} y_{\mathrm{ss}}(t) &= A |G(j \omega)| \cos(\omega t + \phi + \angle G(j \omega)) \end{align}\]

Not usually true for time-varying linear systems!

System behavior can be “plotted”: Bode plots

Bode plots

Examples (from Section 7.8)

\[\begin{align*} L_0 &= \frac{67}{s (s^2 + 49)}, & L_\pi &= \frac{67}{s (s^2 - 49)} \end{align*}\]

Magnitude

\(20 \log_{10} |G(j \omega)|\) versus \(\log_{10} \omega\)

Phase:

\(\angle G(j \omega)\) versus \(\log_{10} \omega\)

See Chapter 7 for much more!

3.9 Norms of Signals and Systems

Some norms of signals

\[\begin{align*} \| y \|_1 &= \int_{0^-}^{\infty} | y(t) | \, dt, & \| y \|_2 &= \left ( \int_{0^-}^{\infty} y(t)^2 \, dt \right )^{1/2}\!\!\!\!, & \| y \|_\infty &= \sup_{t \geq 0} | y(t) | \end{align*}\]

BIBO stability

\[\begin{align*} \| y \|_\infty \leq \|g \|_1 \, \| u \|_\infty. \end{align*}\]

Signal norms are equivalent

\[\begin{align*} \| y \|_p &\leq \|g \|_1 \, \| u \|_p, & p = 1, 2, \ldots, \infty. \end{align*}\]

Some norms of systems

Plancherel and Parseval’s theorem

\[\begin{align*} \| y \|_2^2 &= \int_{0^-}^{\infty} y(t)^2 \, dt = \int_{-\infty}^{\infty} y^*(t) \, y(t) \, dt = \frac{1}{2 \pi} \int_{-\infty}^{\infty} Y(j \omega) Y(-j \omega) \, d\omega = \frac{1}{2 \pi} \int_{-\infty}^{\infty} |Y(j \omega)|^2 \, d\omega = \| Y \|^2_2 \end{align*}\]

\(H_2\) norm

\[\begin{align*} \| y \|_\infty &\leq \| G \|_2 \| u \|_2, & \| g \|_2 &= \| G \|_2 \end{align*}\]

Computation can be done using residues!

\(H_\infty\) norm

\[\begin{align*} \| y \|_2 &\leq \| G \|_\infty \|u \|_2, & \| G \|_\infty &= \sup_{\omega \in \mathbb{R}} | G(j \omega) | \end{align*}\]

Value can be obtained from a Bode plot!

Learn more