Control-System Foundations and Architectures

Introduction

A control system is an arrangement of physical components, computation, and communication links that causes a system—the plant—to behave in a desired manner. Examples include motor-speed regulation, aircraft attitude control, optical-positioning systems, power-converter control, and temperature regulation in semiconductor processing. At postgraduate level, control design is not merely the selection of a controller. It involves modelling uncertainty, selecting an architecture, defining measurable performance, analysing stability, and understanding trade-offs among speed, accuracy, robustness, noise sensitivity, and implementation cost.

The central engineering problem is to make the controlled output follow a reference despite plant dynamics, disturbances, parameter variations, sensor imperfections, and actuator limitations.

Learning Objectives

After studying this chapter, the reader should be able to:

1. distinguish open-loop, closed-loop, feedback, and feedforward architectures;

2. formulate error, regulation, tracking, and disturbance-rejection objectives;

3. derive sensitivity relationships and interpret their practical meaning;

4. reduce single- and multiloop block diagrams and construct signal-flow graphs;

5. relate time-domain specifications to frequency-domain measures;

6. classify system type and predict steady-state error;

7. evaluate architecture choices using a realistic engineering case.

1. Fundamental Terminology and Mathematical Model

A typical single-input single-output system contains:

• Reference, \(r(t)\): desired command;
• Controller, \(C(s)\): computes the control action;
• Actuator: converts the command into physical input;
• Plant, \(P(s)\): process being controlled;
• Sensor, \(H(s)\): measures the output;
• Output, \(y(t)\): controlled variable;
• Disturbance, \(d(t)\): unwanted input or environmental influence;
• Measurement noise, \(n(t)\): corruption introduced in sensing or transmission.

For linear time-invariant systems, Laplace-transform models are commonly written as transfer functions. A transfer function is defined under zero initial conditions as

\[

G(s)=\frac{Y(s)}{U(s)}.

\]

It is a model of input-output behaviour, not necessarily a complete physical description. State-space models are preferable when there are multiple inputs and outputs, nonzero initial conditions, nonlinearities, or internal-state questions.

The plant is the process; the controller is the designed decision element; the forward path is the product of elements between the summing junction and output; and the feedback path carries measured output information back for comparison.

2. Open-Loop and Closed-Loop Control

Open-loop control

An open-loop system does not use measured output to correct its action. If the forward transfer function is \(G(s)\), then

\[

Y(s)=G(s)R(s).

\]

Examples include a timed washing-machine cycle and a stepper motor operated without position feedback. Open-loop systems are inexpensive and often reliable when the plant is predictable, but they cannot automatically compensate for load changes, model error, or disturbances.

Closed-loop control

In a negative-feedback system with controller \(C(s)\), plant \(P(s)\), and sensor \(H(s)\), define the error

\[

E(s)=R(s)-H(s)Y(s).

\]

The control signal is \(U(s)=C(s)E(s)\), and \(Y(s)=P(s)U(s)\). Hence

\[

Y=C P(R-HY),

\]

so

\[

\boxed{\frac{Y(s)}{R(s)}=\frac{C(s)P(s)}{1+C(s)P(s)H(s)}}.

\]

The loop transfer function is

\[

L(s)=C(s)P(s)H(s),

\]

and the characteristic equation is \(1+L(s)=0\). Negative feedback normally improves accuracy and disturbance rejection, but excessive loop gain or phase lag can produce oscillation or instability.

3. Feedback, Sensitivity, and Robustness

For a standard negative-feedback loop, define the sensitivity and complementary-sensitivity functions:

\[

S(s)=\frac{1}{1+L(s)}, \qquad T(s)=\frac{L(s)}{1+L(s)}.

\]

They satisfy \(S(s)+T(s)=1\). The function \(S\) governs error and many disturbance effects; \(T\) governs reference transmission and measurement-noise transmission in common configurations.

Suppose the forward product is \(G(s)\), while the feedback path is fixed. A small plant perturbation \(\Delta G\) produces, to first order,

\[

\boxed{\frac{\Delta Y}{Y}\approx S(s)\frac{\Delta G}{G}}.

\]

Equivalently,

\[

\frac{(\Delta Y/Y)}{(\Delta G/G)}\approx S(s).

\]

This relation is valid for small perturbations and at a specified complex frequency or operating condition. It is not an exact nonlinear identity. When \(|L|\gg1\), \(|S|\ll1\), so the closed-loop response is relatively insensitive to plant-gain uncertainty. This robustness is purchased with bandwidth, actuator effort, noise transmission, and possible stability-margin reduction.

For a parameter \(p\), the logarithmic sensitivity of a quantity \(F\) is

\[

S_F^p=\frac{p}{F}\frac{\partial F}{\partial p}.

\]

For a closed-loop transfer function \(T=G/(1+GH)\), with respect to \(G\), this sensitivity is \(1/(1+GH)=S\). Sensitivity is therefore a quantitative robustness measure, not merely a qualitative statement that feedback is beneficial.

4. Regulation, Tracking, and Servomechanisms

Tracking means following a time-varying reference. Regulation means maintaining a specified output, often constant, despite disturbances. A servomechanism is a feedback system whose principal function is command tracking of a mechanical or electromechanical variable such as position, velocity, or angle. A regulator may have a constant reference but must reject disturbances.

For a plant disturbance entering additively at the plant input,

\[

Y=PC(R-HY)+PD,

\]

therefore

\[

\frac{Y}{R}=\frac{PC}{1+PCH}=T_r,

\qquad

\frac{Y}{D}=\frac{P}{1+PCH}=PS.

\]

Thus high loop gain reduces the output effect of an input disturbance. If disturbance enters at the output, \(Y=PC(R-HY)+D_o\), giving \(Y/D_o=S\). These formulas depend on the stated injection points; moving a disturbance changes its transfer function.

5. Feedforward and Combined Architectures

Feedforward measures a disturbance or command and acts before the plant output is affected. If a disturbance \(D\) enters through \(P_d\), while the control path is \(P_u\), an ideal disturbance-cancellation element is

\[

F_d(s)=-\frac{P_d(s)}{P_u(s)},

\]

provided the disturbance is measurable, the models are sufficiently accurate, and the inverse is causal and implementable. Exact cancellation is often impossible because of delays, nonminimum-phase zeros, noise, or uncertain dynamics.

A practical architecture combines feedforward with feedback. Feedforward handles predictable, measurable disturbances rapidly; feedback corrects model mismatch and unmeasured disturbances. Feedforward alone has no inherent correction mechanism, while feedback alone may respond only after an error develops.

6. System Type and Steady-State Error

For unity feedback, the error is

\[

E(s)=\frac{R(s)}{1+G(s)},

\]

where \(G(s)=C(s)P(s)\) is the forward transfer function. If the closed loop is stable, the final-value theorem gives

\[

e_{ss}=\lim_{t\to\infty}e(t)=\lim_{s\to0}sE(s).

\]

The theorem requires that all poles of \(sE(s)\) lie in the open left half-plane, except possibly the simple pole at the origin required by the theorem's standard conditions.

The system type is the number of uncancelled integrators (poles at the origin) in the forward path. For a unity-feedback system:

• Type 0 has finite step error, generally infinite ramp error;
• Type 1 has zero step error and finite ramp error;
• Type 2 has zero step and ramp error, and finite parabolic error.

For example, if \(G(s)=K/s\) and \(R(s)=1/s\) (unit step),

\[

E(s)=\frac{1/s}{1+K/s}=\frac{1}{s+K},

\quad e_{ss}=0.

\]

For a unit ramp \(R=1/s^2\), \(E=1/[s(s+K)]\), and \(e_{ss}=1/K\). Stability must be established before applying these conclusions.

7. Block Diagrams: Rules and Worked Multiloop Reduction

A block diagram represents functional relationships. Blocks in cascade multiply; parallel paths add algebraically; a summing point forms a signed sum; and a takeoff point duplicates a signal. Moving a summing point or takeoff point is permissible only when the intervening gain is adjusted according to block-diagram equivalence rules.

Worked reduction

Consider a system with an inner negative-feedback loop around \(G_2\), an outer negative-feedback loop, and forward blocks \(G_1\) and \(G_3\). Let the inner feedback be \(H_2\), and the outer feedback be \(H_1\). The inner loop reduces to

\[

G_{\mathrm{inner}}=\frac{G_2}{1+G_2H_2}.

\]

The equivalent forward path for the outer loop is

\[

G_f=G_1G_{\mathrm{inner}}G_3

=\frac{G_1G_2G_3}{1+G_2H_2}.

\]

Consequently,

\[

\boxed{\frac{Y}{R}=\frac{G_1G_2G_3}{1+G_2H_2+G_1G_2G_3H_1}}.

\]

This reduction assumes the inner-loop signals and feedback connections permit the inner loop to be isolated without altering pickoff or summing-point relationships. In a general multiloop diagram, reduction by inspection can be unsafe; Mason's gain formula or state-space formulation is preferable.

8. Signal-Flow Graphs and Mason's Formula

A signal-flow graph (SFG) consists of nodes representing variables and directed branches labelled by transmittances. A branch from node \(x_j\) to node \(x_i\) means that the contribution to \(x_i\) is the branch gain times \(x_j\).

Mason's overall gain formula is

\[

\frac{Y}{R}=\frac{\sum_k P_k\Delta_k}{\Delta},

\]

where \(P_k\) is the gain of the \(k\)-th forward path, \(\Delta=1-\)(sum of individual loop gains) + (sum of products of two non-touching loop gains) - (sum of products of three non-touching loop gains) +..., and \(\Delta_k\) is formed by excluding loops touching the \(k\)-th forward path.

For the simple negative-feedback graph, the forward path is \(P_1=G\), and the loop gain is \(L=-GH\). Thus \(\Delta=1-(-GH)=1+GH\), and

\[

Y/R=G/(1+GH).

\]

For the multiloop example above, the graph has forward path \(P_1=G_1G_2G_3\), individual loops \(L_1=-G_2H_2\) and \(L_2=-G_1G_2G_3H_1\). These loops touch the forward path and each other, so there are no non-touching loop products. Therefore

\[

\Delta=1+G_2H_2+G_1G_2G_3H_1,

\]

which reproduces the block-diagram result. This construction demonstrates why SFGs are particularly useful when several loops, feedforward paths, or noncanonical signal arrangements coexist.

9. Design Specifications in Time and Frequency Domains

Time-domain specifications

For a stable underdamped second-order dominant model,

\[

T(s)=\frac{\omega_n^2}{s^2+2\zeta\omega_n s+\omega_n^2},

\]

where \(\omega_n\) is natural frequency and \(\zeta\) is damping ratio. For \(0<\zeta<1\),

\[

\omega_d=\omega_n\sqrt{1-\zeta^2},

\]

\[

M_p=100\exp\left(-\frac{\pi\zeta}{\sqrt{1-\zeta^2}}\right)\%,

\]

\[

t_s\text{ (2% criterion)}\approx\frac{4}{\zeta\omega_n}.

\]

The settling approximation is based on the exponential envelope and is most useful when higher-order poles are sufficiently faster than the dominant pair. Rise time depends on the precise definition and should not be treated as a universal formula.

Frequency-domain specifications

The frequency response is obtained by evaluating \(T(j\omega)\). Important specifications include:

• Bandwidth: often the frequency at which \(|T(j\omega)|\) falls to \(1/\sqrt{2}\) of its low-frequency value (the \(-3\) dB point);
• Resonant peak \(M_r\): maximum magnitude of \(T(j\omega)\), when a peak exists;
• resonant frequency \(\omega_r\): frequency at which that peak occurs;
• gain crossover frequency \(\omega_{gc}\): frequency where \(|L(j\omega)|=1\);
• phase crossover frequency \(\omega_{pc}\): frequency where \(\arg L(j\omega)=-180^\circ\);
• phase margin: \(PM=180^\circ+\arg L(j\omega_{gc})\);
• gain margin: reciprocal of \(|L(j\omega_{pc})|\) in absolute units, or \(-20\log_{10}|L(j\omega_{pc})|\) dB.

Margins are local frequency-domain indicators of robustness for the assumed loop and should not be confused with guaranteed stability under arbitrary uncertainty.

For the second-order closed-loop model, a useful relation is

\[

M_r=\frac{1}{2\zeta\sqrt{1-\zeta^2}},

\]

when \(\zeta<1/\sqrt{2}\); otherwise no resonant peak above the zero-frequency value exists. Thus low damping produces overshoot and a frequency-response peak. Increasing \(\omega_n\) generally makes the response faster, while changing \(\zeta\) primarily affects overshoot and transient shape.

Frequency-domain and time-domain specifications are related but not identical. Bandwidth usually correlates with rise time, approximately \(t_r\propto1/\omega_{BW}\) for a specified response family. Phase margin correlates with damping and overshoot, but the relationship is model-dependent and loses accuracy for high-order, nonminimum-phase, or strongly delayed systems. A credible postgraduate design therefore checks both Bode/Nyquist measures and simulated or measured time responses.

Worked specification interpretation

Suppose a design is approximated by \(\zeta=0.5\) and \(\omega_n=10\) rad/s. Then

\[

M_p=100e^{-\pi(0.5)/\sqrt{0.75}}\approx16.3\%,

\]

and

\[

t_s\approx4/(0.5\times10)=0.8\text{ s}.

\]

These are approximate because the real system may contain additional poles and zeros. A frequency-response study must still verify bandwidth, peak magnitude, and stability margins.

10. Practical Architecture Selection

Open loop is suitable where repeatability is high and measurement is difficult or expensive. Feedback is preferred when disturbances and uncertainty are significant. Feedforward is valuable when major disturbances are measurable and a sufficiently accurate model is available. Cascaded control places a fast inner loop around a subsystem and a slower outer loop around the overall process; this can improve disturbance rejection and isolate actuator or sensor dynamics, but loop interactions must be analysed.

Implementation introduces sampling, quantisation, computation delay, saturation, dead zones, and sensor noise. Integral action improves low-frequency tracking and regulation but can cause windup when an actuator saturates. Anti-windup logic, rate limits, bumpless transfer, fault detection, and safe fallback modes are therefore part of professional control design.

Applied Case: Precision DC-Motor Positioning

Background

Consider an illustrative precision positioning stage driven by a DC motor. The command is shaft position, while load torque varies during operation and encoder measurements contain quantisation noise.

Problem

An open-loop voltage command cannot maintain position when load torque changes. A pure proportional controller reduces error but leaves a steady offset under constant load torque. High gain improves stiffness but may amplify encoder noise and reduce stability margin because of motor and amplifier phase lag.

Analysis and approach

A cascaded architecture is selected: an inner current loop regulates motor torque, an intermediate velocity loop damps mechanical motion, and an outer position loop tracks the command. Feedforward velocity and acceleration terms provide anticipatory action. Feedback remains responsible for correcting modelling errors and unmeasured torque disturbances.

The position loop is specified by allowable overshoot, settling time, bandwidth, and minimum phase margin. The designer first verifies the inner loop is substantially faster than the outer loop, then tunes the outer controller using the measured plant. Saturation limits are included in simulation, and integral action is protected by anti-windup. Encoder filtering is designed so that noise attenuation does not introduce unacceptable phase lag near crossover.

Outcome and learning points

The combined architecture can provide low steady-state error, rapid tracking, and improved rejection of load changes compared with open-loop control. The case also shows that no single metric is sufficient: time response, frequency response, actuator limits, sensor noise, and robustness must be evaluated together. Safety requires limiting commanded torque and defining a controlled response to encoder failure or runaway detection.

Challenges, Governance, and Contemporary Practice

Control systems increasingly interact with networked computers, cloud services, and machine-learning components. AI may assist plant identification, fault diagnosis, or gain scheduling, but a learned model does not remove the need for stability analysis, validation, explainability, and constraint enforcement. Safety-critical systems require traceability, independent verification, cybersecurity protection, deterministic fallback behaviour, and compliance with applicable sector standards. Ethical design includes preventing unsafe operation, protecting operational data, and ensuring that automated decisions remain within human-approved limits.

Best practice is to define requirements quantitatively, state modelling assumptions, identify uncertainty, select architecture deliberately, analyse nominal and worst-case behaviour, test saturation and faults, and validate the implementation on hardware before deployment.

Chapter Summary

Control systems manipulate plant inputs to achieve tracking, regulation, and disturbance rejection. Open-loop systems are simple but cannot correct output errors; closed-loop systems use feedback and the characteristic equation \(1+L=0\) to regulate behaviour. Sensitivity quantifies robustness: for a small plant perturbation, \(\Delta Y/Y\approx S(\Delta G/G)\), with \(S=1/(1+L)\). Feedforward can anticipate measurable disturbances, while feedback corrects residual errors and uncertainty. System type predicts steady-state error only when stability and final-value-theorem conditions hold. Block diagrams and signal-flow graphs provide complementary reduction tools, including Mason's formula for complex interconnections. Time-domain specifications such as overshoot and settling time must be interpreted alongside bandwidth, resonant peak, crossover frequencies, and stability margins. Professional design also accounts for noise, saturation, delays, implementation constraints, safety, and cybersecurity.

Key Terms

Control system: Arrangement that regulates a plant variable.

Plant: Physical process being controlled.

Controller: Element that generates control action.

Reference: Desired value or trajectory.

Error: Difference between reference and measured output.

Feedback: Return of measured output information for comparison.

Feedforward: Control action based on a measured command or disturbance.

Disturbance: Unwanted influence on system behaviour.

Tracking: Following a time-varying reference.

Regulation: Maintaining a desired condition despite disturbances.

Servomechanism: Feedback system primarily intended for command tracking.

Loop transfer function: Product of transfer functions around a loop.

Sensitivity: Relative response of a quantity to a relative parameter change.

Complementary sensitivity: \(T=L/(1+L)\), governing common reference/noise transmission.

System type: Number of uncancelled integrators in the forward path.

Bandwidth: Frequency range over which response follows input adequately.

Phase margin: Distance from phase crossover to the \(-180^\circ\) condition at gain crossover.

Signal-flow graph: Directed graph representing variable relationships.

Mason's formula: Algebraic formula for gain of a signal-flow graph.

Anti-windup: Method limiting integrator accumulation during actuator saturation.

Review Questions

Short Answer Questions

1. What distinguishes open-loop from closed-loop control?

2. Define sensitivity and complementary sensitivity.

3. What is meant by system type?

4. Why is feedforward normally combined with feedback?

5. Define gain crossover frequency and phase margin.

Long Answer Questions

1. Derive the closed-loop transfer function of a negative-feedback system and discuss the characteristic equation.

2. Explain regulation, tracking, and disturbance rejection using transfer functions for different disturbance injection points.

3. Describe block-diagram reduction and derive the multiloop result presented in this chapter.

4. Explain Mason's gain formula and apply it to a single negative-feedback graph.

5. Relate damping ratio, overshoot, settling time, bandwidth, and phase margin, including limitations of these relationships.

Analytical / Application Questions

1. For a unity-feedback system with forward path \(G(s)=K/[s(s+2)]\), determine the steady-state error for a unit step and a unit ramp, assuming closed-loop stability. Interpret the result.

2. A plant has a 10% small relative gain error at a frequency where \(|S|=0.2\). Estimate the magnitude of the resulting relative output error and state the approximation used.

3. Design an architecture for a motor drive subject to measurable load disturbances and encoder noise. Justify the use of feedback, feedforward, cascade control, filtering, and anti-windup.

Multiple Choice Questions with Answers

Question 1

For a standard negative-feedback loop with loop transfer function L(s), which expression is the sensitivity function?

A. S(s)=1+L(s)

B. S(s)=1/(1+L(s))

C. S(s)=L(s)/(1-L(s))

D. S(s)=1-L(s)

Correct Answer: B. S(s)=1/(1+L(s))

Question 2

A small 10% relative plant variation occurs where the sensitivity magnitude is 0.2. The approximate relative output variation is

A. 0.02%

B. 0.2%

C. 2%

D. 50%

Correct Answer: C. 2%

Question 3

For unity feedback with forward path G(s)=K/s and a unit-step input, assuming closed-loop stability, the steady-state error is

A. 1/K

B. K

C. Zero

D. Infinite

Correct Answer: C. Zero

Question 4

Which architecture acts on a measurable disturbance before its effect propagates through the plant?

A. Feedforward

B. Open-loop calibration only

C. Measurement noise shaping

D. Output-only regulation

Correct Answer: A. Feedforward

Question 5

The characteristic equation of a standard negative-feedback system is

A. 1-L(s)=0

B. G(s)-H(s)=0

C. 1+L(s)=0

D. L(s)=0

Correct Answer: C. 1+L(s)=0

Question 6

For a second-order dominant system with zeta=0.5 and omega_n=10 rad/s, the approximate 2% settling time is

A. 0.08 s

B. 0.4 s

C. 0.8 s

D. 5 s

Correct Answer: C. 0.8 s

Question 7

In Mason's formula, Delta is formed from

A. Only forward-path gains

B. Individual loops and products of non-touching loops with alternating signs

C. Only sensor gains

D. The sum of all node values

Correct Answer: B. Individual loops and products of non-touching loops with alternating signs

Question 8

The -3 dB bandwidth convention corresponds to a magnitude ratio of approximately

A. 0.1

B. 0.5

C. 0.707

D. 2

Correct Answer: C. 0.707

Question 9

A Type 1 unity-feedback system generally has what steady-state error for a unit-step input, if stable?

A. Zero

B. Finite nonzero error

C. Infinite error

D. An error equal to the damping ratio

Correct Answer: A. Zero

Question 10

Phase margin is evaluated at the frequency where

A. The closed-loop phase is zero

B. The loop magnitude equals one

C. The plant gain is maximum

D. The sensor output is zero

Correct Answer: B. The loop magnitude equals one