
Control Systems
Drag poles around the s-plane, watch a root locus grow, read gain and phase margins off a live Bode plot and tune a PID loop until it stops overshooting
Eighteen chapters, 20 interactives, few words. Every chapter is mapped to the Anna University, VTU, SPPU, Mumbai University and KTU units it covers: modelling mechanical, electrical and electromechanical systems, block diagrams and signal flow graphs, first- and second-order time response and steady-state error, Routh-Hurwitz stability, root locus, Bode and Nyquist plots with gain and phase margins, lead and lag compensators, PID control with Ziegler-Nichols tuning, state-space models, pole placement and digital control.
- AUEE3503Control Systems
- AUEC3351Control Systems
- VTUBEE602Control Systems
- VTUBEC403Control Systems
- SPPU303150Control System Engineering
- SPPU204192Control Systems
- MUEEC503Control Systems
- MUECC305Electronic Instrumentation and Control Systems
- KTUPCECT503Control Systems
- KTUPCEET601Control Systems
By Skill-Lync · Updated 17 Nov 2026 · 18 chapters · 20 interactives · free, no sign-up
Introduction: open loop, closed loop and what feedback buys you
Which syllabus covers thisAU EE3503 Unit I; EC3351 Unit IVTU BEE602 Module 1; BEC403 Module 1SPPU 303150 Unit 1; 204192 Unit IMU EEC503 Module 1; ECC305 Module 3KTU PCECT503 Module 1
What is the difference between an open-loop and a closed-loop system?
An open-loop system acts on a setting and never checks the result: a toaster timer, a washing machine cycle, a fixed-voltage fan. A closed-loop system measures the output, compares it with the reference and drives the error towards zero: a cruise control, an air-conditioner thermostat, a motor speed loop. Closed loop costs a sensor, but it corrects for disturbances and drift that open loop cannot see.
What are the blocks of a feedback control system?
Reference input (the set point), error detector (a summing junction that subtracts the measurement), controller (decides the correction), actuator (motor, valve, heater that applies it), plant (the thing being controlled), and the feedback sensor that measures the output. A disturbance enters at the plant. Every block diagram in this book is some arrangement of these six.
What does negative feedback change?
It divides the sensitivity to plant changes by , so a 10% drift in barely moves the output. It divides the effect of disturbances by the same factor. It widens the bandwidth (the loop responds faster) and can linearise a nonlinear plant. The prices: overall gain falls by , you need a sensor, and too much loop gain can make a stable plant oscillate.
How are control systems classified?
Linear or nonlinear; time-invariant or time-varying; continuous or discrete (sampled by a computer); single-input single-output or multivariable. By job: a regulator holds a constant set point against disturbances (voltage regulator, thermostat), a servo or tracking system follows a changing reference (antenna, robot joint). Feedforward adds a correction computed from a measured disturbance before it affects the output.
A plant has gain G = 100. You close the loop with H = 0.1. If G drifts up by 10%, the closed-loop gain changes by about:
The closed-loop gain goes from 100/11 = 9.091 to 110/12 = 9.167, a change of 0.83%. The small-change estimate 10%/(1 + GH) = 0.91% is close; the exact figure is a little lower because the change is not infinitesimal.
Worked examples
Sensitivity to a plant change · Forward gain G = 100, feedback H = 0.1; G rises to 110
- = 100/11 = 9.091
- Sensitivity = 0.0909
- New = 110/12 = 9.167, a change of 0.83% for a 10% change in G
Answer: 0.83% instead of 10%
Disturbance rejection on a speed loop · Motor speed = 100 rpm/V × input − 30 rpm per N·m of load; load 4 N·m; closed loop adds a controller of 0.1 V per rpm of error
- Open loop: the speed simply drops 30 × 4 = 120 rpm
- Loop gain = 100 × 0.1 = 10
- Closed loop: drop = 120/(1 + 10) = 10.9 rpm
Answer: 10.9 rpm instead of 120 rpm
Feedback speeds up a first-order plant · G = 9/(s + 1), unity feedback
- Closed loop:
- Time constant 1 s → 0.1 s, bandwidth 1 → 10 rad/s
- DC gain falls from 9 to 0.9
Answer: 10× faster, at one tenth of the gain
Mathematical models of mechanical, electrical and electromechanical systems
Which syllabus covers thisAU EE3503 Unit I; EC3351 Unit IVTU BEE602 Module 1; BEC403 Module 1SPPU 303150 Unit 1; 204192 Unit IMU EEC503 Module 2; ECC305 3.2KTU PCECT503 Module 1
How do you model a mechanical translational system?
Three elements: mass resists acceleration, damper resists velocity, spring resists displacement. Draw a free-body diagram for each mass and add the forces (D'Alembert). One mass on a spring and damper pushed by a force gives a second-order equation, and with zero initial conditions its Laplace transform is the transfer function.
And a rotational system?
Same three elements in rotation: inertia , viscous friction , torsional stiffness , acted on by a torque. Gear trains move inertia and friction between shafts: seen from the motor, a load behind an reduction looks smaller by , which is why a small motor can drive a heavy load through a gearbox.
How are electrical networks modelled?
With Kirchhoff's laws on resistors, inductors and capacitors: , , . A series RLC driven by a voltage gives the same second-order equation as the mass-spring-damper. Take Laplace transforms with zero initial conditions, use impedances , and , and the transfer function falls out of a voltage divider.
What are force-voltage and force-current analogies?
Two systems with the same differential equation behave identically. Force-voltage: force ↔ voltage, mass ↔ inductance, damper ↔ resistance, spring ↔ 1/capacitance, velocity ↔ current (series circuit). Force-current: force ↔ current, mass ↔ capacitance, damper ↔ conductance, spring ↔ 1/inductance (parallel circuit). Before computers, engineers built the circuit to test the machine.
How is a DC motor modelled?
Two coupled equations. Electrical: , where the back EMF is proportional to speed. Mechanical: , torque proportional to current. In SI units . Eliminate the current and you get speed over voltage, a second-order transfer function that is nearly first order because the electrical time constant is tiny.
A mass-spring-damper (m = 2 kg, b = 8 N·s/m, k = 50 N/m) and a series RLC (L = 0.5 H, R = 4 Ω, C = 0.02 F). Which overshoots more?
Both have damping ratio ζ = 0.4: the mass from b/(2√(mk)) = 8/20, the circuit from (R/2)√(C/L) = 2 × 0.2. Overshoot depends only on ζ, so both overshoot 25.4%. The circuit is twice as fast (ωn 10 vs 5 rad/s).
Worked examples
Mass-spring-damper · m = 2 kg, b = 8 N·s/m, k = 50 N/m, step force 10 N
- = 5.0 rad/s, = 8/20 = 0.4
- Final deflection = F/k = 10/50 = 0.2 m
Answer: ωn = 5 rad/s, ζ = 0.4, settles at 0.2 m
Its electrical analogues · The same mass-spring-damper
- Force-voltage (series RLC): L = m = 2 H, R = b = 8 Ω, C = 1/k = 0.02 F
- Force-current (parallel RLC): C = m = 2 F, 1/R = b → R = 0.125 Ω, L = 1/k = 0.02 H
Answer: L 2 H, R 8 Ω, C 0.02 F (series)
Series RLC · R = 4 Ω, L = 0.5 H, C = 0.02 F, output across C
- = 10 rad/s
- = 2 × 0.2 = 0.4
Answer: ωn = 10 rad/s, ζ = 0.4
Inertia through a gearbox · Motor Jm = 0.01 kg·m², load JL = 0.5 kg·m², BL = 1 N·m·s/rad, load torque 20 N·m, 10:1 reduction
- = 0.01 + 0.5/10² = 0.015000000000000001 kg·m²
- = 1/10² = 0.010000000000000002 N·m·s/rad
- Load torque at the motor = 20/10 = 2 N·m
Answer: 0.015000000000000001 kg·m² seen by the motor
Transfer functions, poles and zeros, servomotors and control components
Which syllabus covers thisAU EE3503 Unit I; EC3351 Unit IVTU BEE602 Module 1SPPU 303150 Units 1 and 6MU EEC503 Module 2KTU PCECT503 Module 1
What exactly is a transfer function?
The Laplace transform of the output divided by the Laplace transform of the input, with all initial conditions zero. It describes a linear time-invariant system completely, and it is the same whatever the input. Its denominator set to zero is the characteristic equation; the highest power of there is the order of the system.
What do poles and zeros tell you?
Poles are the roots of the denominator: each gives a mode in the response, so their real parts set how fast things decay and their imaginary parts set how fast they oscillate. Any pole in the right half-plane means instability. Zeros are roots of the numerator: they change how much of each mode appears, not which modes exist. Poles at the origin set the type of the system.
How does an armature-controlled DC servomotor behave?
Field fixed, armature voltage varied. Because the armature inductance is small, the electrical pole sits far out on the negative real axis and the motor acts almost like a first-order lag from voltage to speed, with mechanical time constant . Add an integrator for position. A field-controlled motor has a large field time constant, so it responds more slowly and suits small drives.
What is a two-phase AC servomotor?
An induction motor with two windings at 90°: a fixed reference phase and a control phase fed with a variable voltage. Its rotor has high resistance (often a thin drag cup), so the torque-speed curve is nearly a straight line with negative slope; that line gives a stable, almost linear first-order model. Used in low-power instrument servos.
Which components close the loop in a position servo?
A potentiometer pair acts as the error detector: output voltage proportional to the angle difference, in V/rad. A tachogenerator gives a voltage proportional to speed for velocity feedback. A synchro transmitter-control transformer pair does the same as the potentiometers for AC systems without sliding contacts. A gear train sets the speed and torque ratio.
A DC motor's armature inductance doubles (2 mH → 4 mH). Its step response from voltage to speed:
The electrical pole moves from about −487 to about −237 rad/s, still 17 times faster than the mechanical pole near −13.7. The mechanical time constant of 76.9 ms still dominates. The interactive shows both responses.
Worked examples
Poles, zeros, order and type ·
- Zero: −2. Poles: 0, −5, −1 ± j2
- Order 4 (degree of denominator), type 1 (one pole at the origin)
- Velocity constant = 10 × 2/(5 × 5) = 0.8
Answer: Order 4, type 1, Kv = 0.8
DC servomotor transfer function · Ra = 1 Ω, La ≈ 0, K = 0.05 V·s/rad = N·m/A, J = 2 × 10⁻⁴ kg·m², B = 1 × 10⁻⁴ N·m·s/rad
- = 0.0001 + 0.0025 = 0.0026000000000000003
- = 76.9 ms
- = 19.23 rad/s per V = 184 rpm/V
Answer: ω/V = 19.2/(0.0769s + 1)
Same motor with La = 2 mH · As above, La = 2 mH
- Characteristic equation:
- Poles: -13.34 and -487.2 rad/s
- The electrical pole is 35 times faster: the first-order model is fine
Answer: Dominant pole -13.34 rad/s
Error detector and tacho constants · 10-turn potentiometers across 10 V; tachogenerator 2.5 V per 1,000 rpm
- = 10 V/(10 × 2π rad) = 0.1592 V/rad
- = 2.5/(1,000 × 2π/60) = 0.0239 V per rad/s
Answer: Kp = 0.159 V/rad, Kt = 0.0239 V·s/rad
Block diagrams and block diagram reduction
Which syllabus covers thisAU EE3503 Unit I; EC3351 Unit IVTU BEE602 Module 2; BEC403 Module 2SPPU 303150 Unit 1; 204192 Unit IMU EEC503 Module 2; ECC305 3.3KTU PCECT503 Module 1
What is a block diagram made of?
Blocks (each a transfer function, output = block × input), summing points (add or subtract signals, signs marked), take-off points (a signal branches unchanged to several places) and arrows for signal flow. It shows how signals move between components, but it is not unique: the same system can be drawn in many ways.
What are the reduction rules?
Blocks in series multiply. Blocks in parallel add. A loop with forward path and feedback becomes (plus sign for negative feedback). To untangle crossing loops, move a summing point or take-off point past a block, compensating with that block or its inverse. Reduce inner loops first, then work outwards.
How do you move a take-off or summing point?
Moving a take-off point ahead of block (towards the input): put in the branch. Moving it after : put in the branch. Moving a summing point ahead of : put in the added branch. Moving it after : put in the branch. The check is always the same: every signal must arrive with the same value as before.
How do you handle a disturbance input?
Linear systems obey superposition: set the reference to zero, find the output due to the disturbance alone, then set the disturbance to zero and find the output due to the reference, and add. For a disturbance entering just before the plant , the closed loop shrinks its effect by .
Inner loop: G2 = 1/(s + 1) with feedback H1 = 2. Outer loop: G1 = 10, unity feedback. The closed-loop pole is at:
The inner loop gives 1/(s + 1 + 2) = 1/(s + 3). The outer loop then gives 10/(s + 3 + 10) = 10/(s + 13).
Worked examples
Series blocks in a unity loop · , in series, unity negative feedback
- Series:
- Loop:
Answer: 5/(s² + 6s + 10)
Nested loops · , with inner feedback , outer unity feedback
- Inner:
- Outer:
Answer: 10/(s + 13)
Disturbance at the plant input · Controller G1 = 20, plant G2 = 1/(s + 2), H = 1; steady disturbance D
- Open loop: C/D at s = 0 is G2(0) = 0.5
- Closed loop: G2/(1 + G1G2) at s = 0 = 0.5/(1 + 10) = 0.0455
Answer: 11 times smaller with the loop closed
Signal flow graphs and Mason's gain formula
Which syllabus covers thisAU EE3503 Unit I; EC3351 Unit IVTU BEE602 Module 2; BEC403 Module 2SPPU 303150 Unit 1; 204192 Unit IMU EEC503 Module 2; ECC305 3.3KTU PCECT503 Module 1
What are the terms of a signal flow graph?
Nodes are signals; branches are directed gains between them. An input node has only outgoing branches, an output node only incoming. A forward path runs from input to output touching no node twice. A loop starts and ends at the same node. Path gain and loop gain are products of branch gains. Loops are non-touching if they share no node.
How does Mason's gain formula work?
Find every forward path and every loop. = 1 − (sum of all loop gains) + (sum of products of every pair of non-touching loops) − (triples) + ... Then is with every loop touching path removed. The formula gives the overall gain in one step, no redrawing.
How do you draw a signal flow graph from a block diagram?
Every summing point and every take-off point becomes a node; every block becomes a branch with its transfer function as the gain; a subtraction becomes a branch with a minus sign. Then apply Mason. For an electrical network, write the node or loop equations and draw one node per variable.
How do you check a Mason result?
Set every transfer function to a number and solve the node equations directly as simultaneous linear equations; the output must match. The interactive does exactly this for the graph on screen. A quick sanity check: if all loops touch every forward path, every is 1.
A single forward path of gain 10 passes through two loops, L1 = −2 and L2 = −3, that do not touch each other. The overall gain is:
Δ = 1 − (L1 + L2) + L1L2 = 1 + 5 + 6 = 12. The path touches both loops, so Δ1 = 1 and T = 10/12. Forgetting the non-touching product gives the wrong 10/6.
Worked examples
Three touching loops · Forward path G1G2G3; loops −G2H1, −G2G3H2, −G1G2G3. Numbers: G1 = 2, G2 = 3, G3 = 4, H1 = 0.5, H2 = 0.25
- All loops touch the path, so Δ1 = 1
- = 24/(1 + 1.5 + 3 + 24) = 0.8136 (node equations give the same)
Answer: 0.8136
Two non-touching loops · Path gain 10 through loops L1 = −2 and L2 = −3, which share no node
- Δ = 1 − (−2 − 3) + (−2)(−3) = 12
- Δ1 = 1 (the path touches both)
Answer: T = 10/12 = 0.833
A feedforward path · P1 = G1G2, P2 = G3 (bypasses G2); one loop −G2H around G2. G1 = 5, G2 = 2, G3 = 1, H = 0.5
- Δ = 1 + G2H = 2
- Δ1 = 1 (P1 touches the loop), Δ2 = Δ = 2 (P2 does not)
- = (10 + 2)/2
Answer: T = 6
Time response: test signals and first-order systems
Which syllabus covers thisAU EE3503 Unit II; EC3351 Unit IIVTU BEE602 Module 3; BEC403 Module 3SPPU 303150 Unit 2; 204192 Unit IIMU EEC503 Module 3; ECC305 4.1KTU PCECT503 Module 2
What are the transient and steady-state parts of a response?
The transient is the part that dies away: it comes from the system's poles and tells you how fast and how smoothly the output gets there. The steady state is what is left as time goes to infinity: it tells you how accurately the output follows. Design usually trades one against the other.
Why use step, ramp, parabolic and impulse inputs?
They are simple, easy to generate in a lab and easy to transform: step (a sudden change of set point), ramp (constant velocity, like tracking a moving target), parabola (constant acceleration), impulse 1 (a sharp kick). The impulse response is the inverse Laplace transform of the transfer function itself.
How does a first-order system respond to a step?
Exponentially, with no overshoot. After one time constant it reaches 63.2% of the final value; after , 98%, the usual 2% settling time. To a ramp it follows with a constant lag of . A thermometer, an RC filter and a DC motor speed loop are all close to first order.
What is the difference between order and type?
Order is the highest power of in the denominator of the closed-loop (or open-loop) transfer function: it counts the energy-storing elements. Type is the number of open-loop poles at the origin, the number of integrators in the loop: it decides which inputs the loop can follow with zero steady-state error (chapter 9).
A thermometer with τ = 10 s is plunged from 20 °C into 80 °C water. After 15 s it reads about:
20 + 60(1 − e^−1.5) = 20 + 60 × 0.777 = 66.6 °C. It needs 4τ = 40 s to come within 2% of 80 °C.
Worked examples
Thermometer reading · τ = 10 s, step from 20 °C to 80 °C
- At t = 15 s: 66.6 °C
- Within 2% of 80 °C after 4τ = 40 s
Answer: 66.6 °C at 15 s
Ramp-following lag · First-order sensor, τ = 0.5 s, input ramps at 2 units/s
- In steady state the output follows the ramp, delayed by τ
- Lag error = 2 × 0.5 = 1 unit
Answer: 1 unit behind
Closed-loop integrator · G = 4/s with unity feedback
- Closed loop: 4/(s + 4) = 1/(0.25s + 1)
- Settling (2%): 4τ = 1 s
- Rise time 10–90%: τ ln 9 = 0.549 s
Answer: τ = 0.25 s
Second-order systems and time-domain specifications
Which syllabus covers thisAU EE3503 Unit II; EC3351 Unit IIVTU BEE602 Module 3; BEC403 Module 3SPPU 303150 Unit 2; 204192 Unit IIMU EEC503 Module 3; ECC305 4.1KTU PCECT503 Module 2
What is the standard second-order system?
, with natural frequency and damping ratio . = 0: undamped, oscillates for ever. 0 < < 1: underdamped, overshoots and rings. = 1: critically damped, fastest with no overshoot. > 1: overdamped, sluggish. Most control loops are tuned underdamped, around = 0.4–0.8.
How do the poles' positions map to the response?
The distance of the poles from the origin is . The angle from the negative real axis satisfies . The real part sets the decay (settling time). The imaginary part sets the ringing frequency (peak time). Poles on a radial line share the same overshoot.
What are the time-domain specifications?
Delay time, rise time (0 to 100% for underdamped), peak time , peak overshoot , settling time (within 2% or 5%) and steady-state error. Overshoot depends on alone: 25% at 0.4, 9.5% at 0.6, 4.3% at 0.7.
What is the gain-damping trade-off?
For a plant with unity feedback, and . Raising the gain makes the loop faster and its steady-state error smaller, but lowers , so it overshoots more. Getting both needs a compensator (chapter 15) or derivative action (chapter 16).
You double the gain K of K/(s(s + 6)) from 25 to 50. The overshoot:
ζ = 6/(2√K) falls from 0.6 to 0.424, and overshoot depends on ζ alone: e^(−ζπ/√(1−ζ²)) goes from 9.5% to 22.9%. The loop is faster (ωn 5 → 7.07 rad/s) but rings more.
Worked examples
All the specifications · , unity feedback, unit step
- Closed loop: : ωn = 5 rad/s, ζ = 0.6
- ωd = 4 rad/s, θ = cos⁻¹0.6 = 53.13°
- = (π − 0.927)/4 = 0.554 s, = π/4 = 0.785 s
- = 9.48%, ≈ 4/3 = 1.33 s
Answer: Mp 9.5%, ts 1.33 s
Damping for an overshoot limit · Overshoot must not exceed 10%
- = 2.303/√(9.870 + 5.302) = 0.591
Answer: ζ ≥ 0.59
Gain for ζ = 0.5 · , unity feedback
- Characteristic equation : 2ζωn = 4 → ωn = 4
- K = ωn² = 16
- Overshoot = 16.3%
Answer: K = 16
Reading ζ and ωn from a test · A step test shows 25% overshoot, peak at 0.5 s
- ζ = 1.386/√(9.870 + 1.922) = 0.404
- ωd = π/0.5 = 6.283 rad/s
- ωn = ωd/√(1 − ζ²) = 6.87 rad/s
Answer: ζ = 0.40, ωn = 6.87 rad/s
Effect of extra poles and zeros, dominant poles and higher-order systems
Which syllabus covers thisAU EE3503 Unit II; EC3351 Unit IIVTU BEE602 Module 3SPPU 204192 Unit IIIMU EEC503 Module 3KTU not in syllabus
What are dominant poles?
The poles closest to the imaginary axis. Their modes decay slowest, so after the first instant they are all you see. If the other poles are at least five times further left, a higher-order system can be approximated by its dominant pair, keeping the original DC gain. That is how most designs are done with second-order formulas.
What does an extra pole do?
An extra real pole adds lag: the response rises later and more slowly, with less overshoot. As the pole moves towards the origin it starts to dominate and the system looks first order. As it moves far left its effect vanishes. In root-locus terms it bends the branches towards the right-half plane, lowering the stable gain range.
And an extra zero?
A left-half-plane zero adds a term proportional to the derivative of the response, so the output rises faster and overshoots more. A zero close to the origin can produce large overshoot even with good damping. A right-half-plane zero (non-minimum phase) makes the output first move the wrong way: a boiler drum level, or an aircraft pitching before it climbs.
Can a zero cancel a pole?
In the transfer function, yes: a zero sitting on a pole removes that mode from the input-output response, and an almost-coincident pair (a dipole) leaves only a small, slow tail. Never cancel a right-half-plane pole this way: the unstable mode is still inside the system and any mismatch lets it grow.
Adding a zero at s = −1 to a system with ζ = 0.5, ωn = 1 (16.3% overshoot). The overshoot becomes:
The zero adds the derivative of the response to itself, so the rise is sharper and the peak higher: simulation gives 29.8%. Damping alone no longer predicts the overshoot.
Worked examples
Dominant-pole approximation ·
- Poles: −10 and −1 ± j3; the far pole is 10× further left
- Keep the DC gain of 1: approximate as
- ζ = 1/√10 = 0.316 → predicted overshoot 35.1%
- Exact simulation: 33.2% (the far pole still trims the peak a little)
Answer: 35.1% approximate vs 33.2% exact
A zero raises overshoot · against
- Same poles: ζ = 0.5, ωn = 1
- Without the zero: 16.3% overshoot
- With the zero at −1: 29.8% (simulated)
Answer: 16.3% → 29.8%
Non-minimum phase undershoot · , unit step
- Partial fractions:
- = −1: the output starts downwards
- Minimum at t = 0.5 s: c = 1 − 2e^−0.5 = -0.213
Answer: Dips to -0.213 before rising to 1
Steady-state error, system type and error constants
Which syllabus covers thisAU EC3351 Unit IIVTU BEE602 Module 3; BEC403 Module 3SPPU 303150 Unit 2; 204192 Unit IIMU EEC503 Module 3; ECC305 4.1KTU PCECT503 Module 2
How is steady-state error found?
For unity feedback the error is . If the closed loop is stable, the final value theorem gives the steady-state error as the limit of at . Always check stability first: the final value theorem gives nonsense for an unstable loop.
What are Kp, Kv and Ka?
Position constant , velocity constant , acceleration constant , all as . Step error is , ramp error , parabolic error for unit inputs. They come straight from the low-frequency end of the open-loop transfer function.
How does system type decide the error?
Type 0 (no integrator): finite step error, cannot follow a ramp. Type 1 (one integrator): zero step error, finite ramp error, cannot follow a parabola. Type 2: zero step and ramp error, finite parabolic error. Each integrator in the loop removes the error for one more class of input, at a cost in stability margin.
What about non-unity feedback and disturbances?
For non-unity feedback, convert to an equivalent unity loop , or compute the actual error directly. For a disturbance, find with the reference at zero and apply the final value theorem: an integrator in the controller, before the point where the disturbance enters, removes a constant disturbance completely.
G = K/(s(s + 4)) must follow a 2 rad/s ramp with at most 0.05 rad error. With that gain, the step overshoot is about:
Kv = K/4 ≥ 2/0.05 = 40, so K ≥ 160. Then ζ = 4/(2√160) = 0.158 and the overshoot is 60%: accuracy and damping fight, which is why compensators exist.
Worked examples
Type 1 loop · , unity feedback
- Kp = ∞, Kv = 10/2 = 5, Ka = 0
- Unit step: ess = 0. Unit ramp: ess = 1/5 = 0.2
- Parabola: ess = ∞
Answer: 0, 0.2, ∞
Type 0 loop ·
- Kp = 50/10 = 5
- Step error = 1/(1 + 5) = 0.1667
Answer: 16.7% of the step
Gain for a ramp spec · , ramp 2 rad/s, error ≤ 0.05 rad
- Need Kv = K/4 ≥ 2/0.05 = 40 → K ≥ 160
- Then ζ = 4/(2√160) = 0.158
- Overshoot 60.5%: too oscillatory without compensation
Answer: K = 160, ζ = 0.16
Type 2 loop · ; input
- Closed loop is stable (5 × 20 > 20)
- Ka = 20 × 1/5 = 4
- ess = 1/Ka = 0.25
Answer: 0.25
Stability and the Routh-Hurwitz criterion
Which syllabus covers thisAU EE3503 Unit II; EC3351 Unit IVVTU BEE602 Module 3; BEC403 Module 4SPPU 303150 Unit 3; 204192 Unit IIIMU EEC503 Module 3; ECC305 5.1KTU PCECT503 Module 3
What does stability mean for a linear system?
Bounded-input bounded-output (BIBO) stable: every bounded input gives a bounded output. That holds exactly when every closed-loop pole has a negative real part. Simple poles on the imaginary axis give sustained oscillation (marginal stability); repeated ones, or any pole in the right half-plane, give a response that grows without limit.
Is there a quick first check?
Yes: for all roots to be in the left half-plane, every coefficient of the characteristic polynomial must be present and of the same sign. A missing term or a sign change means unstable, no further work. For second-order polynomials that condition is also sufficient; from third order up you need the Routh array.
How do you build the Routh array?
Write the coefficients in two rows, alternating. Each new element is a cross-multiplication of the two rows above divided by the first element of the row just above. The number of sign changes in the first column equals the number of right-half-plane roots. For a gain , require every first-column entry to be positive and you get the stable range.
What are the special cases?
A zero in the first column (rest of the row non-zero): replace it by a small and continue. A whole row of zeros: roots symmetric about the origin (often a pair on the imaginary axis). Form the auxiliary polynomial from the row above, replace the zero row by its derivative and continue. The auxiliary polynomial's roots are those symmetric roots.
How do you test relative stability?
Substitute into the characteristic equation and apply Routh to the new polynomial in . No sign changes means every root lies left of , so every mode decays at least as fast as , a settling-time guarantee.
K/(s(s + 2)(s + 4)) with unity feedback. What happens at K = 48?
The s¹ row is (48 − K)/6, zero at K = 48. The auxiliary polynomial 6s² + 48 = 0 gives s = ±j2.83: two poles on the imaginary axis, a sustained oscillation. Above 48, two poles cross into the right half-plane.
Worked examples
A stable cubic ·
- Rows: s³: 1, 11; s²: 6, 6
- s¹: (6 × 11 − 1 × 6)/6 = 10; s⁰: 6
- First column 1, 6, 10, 6: no sign change
Answer: Stable (roots −1, −2, −3)
Range of K · , unity feedback
- s¹ row: (48 − K)/6 > 0, s⁰ row: K > 0
- At K = 48: 6s² + 48 = 0 → ω = √8 = 2.828 rad/s
Answer: 0 < K < 48
Row of zeros ·
- s⁴: 1, 5, 6; s³: 2, 4; s²: 3, 6; s¹: 0 (whole row)
- Auxiliary , → s¹ row: 6; s⁰: 6
- No sign change, but A(s) = 0 gives s = ±j1.414
Answer: Marginally stable, oscillates at 1.414 rad/s
Zero in the first column ·
- s⁴: 1, 2, 3; s³: 1, 2; s²: 0 → ε, 3
- s¹: (2ε − 3)/ε → large negative; s⁰: 3
- Column signs +, +, +, −, +: two changes
Answer: Two roots in the right half-plane
Root locus: construction and design
Which syllabus covers thisAU EE3503 Unit II; EC3351 Unit IVVTU BEE602 Module 4; BEC403 Module 4SPPU 303150 Unit 3; 204192 Unit IIIMU EEC503 Module 5; ECC305 5.2KTU PCECT503 Module 3
What is the root locus?
The paths the closed-loop poles trace on the s-plane as the gain goes from 0 to infinity. A point is on the locus if the open-loop transfer function there has an angle of an odd multiple of 180° (angle condition); the gain that puts a pole there comes from the magnitude condition.
What are the construction rules?
Branches start at open-loop poles (K = 0) and end at zeros or at infinity. Real-axis segments with an odd number of poles plus zeros to their right are on the locus. branches go to infinity along asymptotes at angles from a centroid. Breakaway points satisfy . Imaginary-axis crossings come from Routh.
How do you find departure and arrival angles?
At a complex pole, the departure angle is 180° minus the sum of angles from the other poles plus the sum of angles from the zeros. At a complex zero, the arrival angle is 180° plus the angles from the poles minus the angles from the other zeros. They show which way the branch leaves a complex pole.
How is the locus used for design?
Draw the line for the damping you want ( from the negative real axis), find where it crosses the locus, read from the magnitude condition. Adding a zero pulls the locus left (more stable, the idea behind lead and PD); adding a pole pushes it right (the cost of integral action).
On K/(s(s + 1)(s + 3)), the branches from 0 and −1 meet and leave the real axis at:
dK/ds = 0 gives 3s² + 8s + 3 = 0, so s = −0.451 or −2.215. Only −0.451 lies on the segment between 0 and −1. It is not halfway because the pole at −3 pushes the breakaway towards the origin.
Worked examples
Full sketch data ·
- Centroid (0 − 1 − 3)/3 = -1.333; asymptotes at 60°, 180°, 300°
- Breakaway: → s = -0.451 (K = 0.631)
- Routh on : K = 12 at the jω axis, ω = √3 = 1.732 rad/s
Answer: Stable for 0 < K < 12
Gain for ζ = 0.5 ·
- The locus leaves the axis at −2 and runs vertically
- ζ = 0.5 → 60° line: meets it at s = −2 + j3.46
- K = |s| × |s + 4| = 4 × 4
Answer: K = 16 (matches chapter 7)
Departure angle · , poles −1 ± j1.414
- Angle from the zero at −2 to −1 + j1.414: 54.74°
- Angle from the other pole: 90°
- φd = 180° + 54.74° − 90°
Answer: φd = 144.74°
Frequency response and frequency-domain specifications
Which syllabus covers thisAU EE3503 Unit III; EC3351 Unit IIIVTU BEE602 Module 4; BEC403 Module 4SPPU 303150 Unit 4; 204192 Unit IVMU EEC503 Module 6; ECC305 6.1KTU PCECT503 Module 3
What is frequency response?
Feed a stable linear system a sine wave and, once the transient dies, the output is a sine of the same frequency, scaled by and shifted by . Sweep and you have the frequency response. It can be measured on a real plant with a signal generator, with no model at all.
What are the frequency-domain specifications?
Resonant peak (the maximum of the closed-loop magnitude), resonant frequency , bandwidth (where the magnitude falls to 0.707, −3 dB), cut-off rate, and on the open loop the gain margin and phase margin. A large means a lightly damped loop; a wide bandwidth means a fast one.
How do time and frequency specifications correlate?
For the standard second-order loop both are set by and . Resonant peak and overshoot both grow as falls; there is no resonant peak at all above = 0.707. Bandwidth is roughly proportional to , so rise time falls as bandwidth rises. Phase margin in degrees is close to 100 for below about 0.6.
Why bother with frequency methods?
They work with measured data, they handle time delays exactly (a delay is a pure phase lag), they show noise and resonance problems directly, and they make compensator design a matter of reshaping a curve. Bode, polar and Nyquist plots in the next two chapters are three ways of drawing the same .
A loop has ζ = 0.8. Its closed-loop resonant peak Mr is:
A resonant peak exists only for ζ < 0.707. Above that the closed-loop magnitude falls steadily from 1 as frequency rises.
Worked examples
Resonance and bandwidth · Closed loop ζ = 0.4, ωn = 5 rad/s
- = 1/(2 × 0.4 × 0.917) = 1.364 (2.70 dB)
- = 5√(1 − 0.32) = 4.123 rad/s
- = 6.873 rad/s; step overshoot 25.4%
Answer: Mr 1.36, ωb 6.87 rad/s
From Mr back to overshoot · Measured Mr = 1.25
- = 0.8 → ζ² = (1 − √0.36)/2 = 0.2
- ζ = 0.4472
- = e^(−π/2) = 20.8%
Answer: ζ = 0.447, about 21% overshoot
Sinusoidal steady state · , input 2 sin 5t
- |G(j5)| = 10/√50 = 1.414
- ∠G(j5) = −tan⁻¹(5/5) = −45°
Answer: Output 2.828 sin(5t − 45°)
Bode plots, gain margin and phase margin
Which syllabus covers thisAU EE3503 Unit III; EC3351 Unit IIIVTU BEE602 Module 4; BEC403 Module 4SPPU 303150 Unit 5; 204192 Unit IVMU EEC503 Module 6; ECC305 6.2KTU PCECT503 Module 3
Why are Bode plots drawn in decibels against log frequency?
Because a transfer function is a product of factors, and logarithms turn products into sums. Each factor's magnitude in dB and its phase can be drawn separately and added. On log axes every first-order factor becomes two straight lines meeting at its corner frequency, so a sketch takes minutes.
What does each factor contribute?
A constant : a flat line at . A pole at the origin: −20 dB/decade and −90°. A real pole : flat, then −20 dB/decade after , with the true curve 3 dB below the corner and phase −45° there. A zero is the mirror image. A complex pole pair falls at −40 dB/decade with a resonant hump set by .
What are gain margin and phase margin?
At the phase crossover frequency (phase = −180°), gain margin is how many dB the gain could rise before the loop goes unstable. At the gain crossover frequency (magnitude 0 dB), phase margin is how much more lag the loop could take. Both positive means a stable closed loop for minimum-phase systems; 6 dB and 30–60° are usual targets.
How do you get a transfer function from a measured Bode plot?
Fit straight-line asymptotes to the measured magnitude. The low-frequency slope gives the type (0, −20, −40 dB/decade for type 0, 1, 2); each change of slope by ±20 dB/decade marks a corner, a zero or pole at that frequency; ±40 with a hump marks a complex pair, whose peak gives . The level of the low-frequency line gives . Check against the phase plot.
G = 10/(s(s + 1)) with unity feedback. Its phase margin is about:
The gain crosses 0 dB at ω = 3.08 rad/s, where the phase is −90° − tan⁻¹3.08 = −162°. PM = 18°: stable but very oscillatory (about 60% step overshoot). The phase never reaches −180°, so the gain margin is infinite.
Worked examples
Phase margin ·
- |G(jω)| = 1: ω⁴ + ω² − 100 = 0
- ω² = (−1 + √401)/2 → ωgc = 3.084 rad/s
- PM = 180° − 90° − tan⁻¹(3.084) = 17.96°
Answer: PM ≈ 18.0°, GM infinite
Gain margin ·
- Phase −180°: tan⁻¹(ω/2) + tan⁻¹(ω/10) = 90° → ω² = 20
- |G(j4.47)| = 20/(4.47 × 4.90 × 10.95) = 1/12
- GM = 20 log 12 = 21.58 dB
Answer: 21.6 dB: the gain can rise 12× (K = 240, as Routh says)
Asymptote sketch ·
- Time-constant form:
- At ω = 1 the low-frequency line is 20 log 0.4 = -7.96 dB
- Slopes: −20 to 2 rad/s, 0 from 2 to 10, −20 from 10 to 50, −40 beyond
Answer: Corners at 2, 10 and 50 rad/s
Polar plots and the Nyquist stability criterion
Which syllabus covers thisAU EE3503 Unit III; EC3351 Units III and IVVTU BEC403 Module 5; BEE602 objective 5SPPU 303150 Unit 4; 204192 Unit IVMU EEC503 Module 6; ECC305 6.3KTU PCECT503 Module 3
What is a polar plot?
drawn as a single curve in the complex plane as goes from 0 to infinity: distance from the origin is the magnitude, angle is the phase. Type 0 starts on the positive real axis; type 1 starts at infinity along −90°; each extra pole adds −90° to where the curve ends at the origin. Where it crosses the negative real axis gives the gain margin.
What is the Nyquist criterion?
Map a contour enclosing the whole right half-plane (the imaginary axis plus a huge semicircle) through the open-loop . Count , the clockwise encirclements of the point −1. Then the number of closed-loop poles in the right half-plane is , where is the number of open-loop poles there. Stable means .
What about poles on the imaginary axis?
The contour must not pass through them, so it detours round each one on a tiny semicircle into the right half-plane. A pole at the origin then maps into a huge semicircle sweeping clockwise through 180° per pole, closing the plot at infinity. Draw it, or the encirclement count will be wrong.
Why use Nyquist when Bode is easier?
Bode's margin rules assume a minimum-phase open loop. Nyquist works for anything: an open loop that is itself unstable (, then you need counter-clockwise encirclements to be stable), time delays, and plots with several crossings. It also shows how close the curve comes to −1, the most honest robustness measure.
G = K/(s − 1), open-loop unstable (P = 1). For K = 2 the Nyquist plot circles −1 once anticlockwise. The closed loop is:
Anticlockwise means N = −1, so Z = N + P = 0. Check: 1 + K/(s − 1) = 0 gives s = 1 − K = −1, a stable pole. Feedback stabilises this plant for any K > 1.
Worked examples
Type 0, third order ·
- Denominator at jω: (6 − 6ω²) + j(11ω − ω³)
- Imaginary part zero at ω = √11 = 3.317 rad/s, where the real part is −60
- So the plot crosses the negative real axis at −K/60
Answer: Stable for K < 60; at K = 6, GM = 20 dB
Type 1, third order ·
- Denominator at jω: −3ω² + jω(2 − ω²)
- Crosses the real axis at ω = √2 with value −K/6
- K = 1: GM = 20 log 6 = 15.56 dB. K = 10: −1 is circled twice clockwise, N = 2, P = 0
Answer: Stable for K < 6; at K = 10, Z = 2 RHP poles
Open-loop unstable plant · , P = 1
- Plot: a circle from −K (ω = 0) to 0, centre −K/2
- K > 1: −1 is inside, circled once anticlockwise, N = −1
- Z = N + P = 0
Answer: Stable for K > 1 (closed-loop pole at 1 − K)
Lead, lag and lag-lead compensators
Which syllabus covers thisAU EE3503 Units III and V; EC3351 Unit IIIVTU BEE602 Module 5; BEC403 Module 5SPPU 303150 Unit 6MU ECC305 4.2KTU PCECT503 Module 3
Why add a compensator?
Gain alone cannot satisfy both a steady-state spec and a transient spec: raising for accuracy kills the damping (chapters 7 and 9). A compensator is a small network or algorithm in series with the plant that reshapes the loop: more phase near crossover for damping (lead), or more low-frequency gain for accuracy (lag), or both.
How does a lead compensator work?
with : a zero before a pole. It adds positive phase between and , up to at their geometric mean, and lifts the gain there by dB. Placed at the new crossover, it raises the phase margin and widens the bandwidth: faster and better damped. Cost: more high-frequency noise.
And a lag compensator?
A pole before a zero, both placed well below the gain crossover (about a decade). It leaves the phase near crossover almost untouched but lets you raise the low-frequency gain by the pole-zero ratio , cutting the steady-state error by up to times. Cost: a slow tail in the step response and less bandwidth.
When do you use lag-lead?
When the plant needs both: more damping and lower steady-state error. The lead part fixes the phase margin and speed, the lag part, at much lower frequency, fixes the error constant. An electrical lag-lead network uses two resistors and two capacitors; in software it is just two first-order filters in series.
How is a lead network built from R and C?
A resistor with a capacitor across it, in series, followed by a resistor to ground. The capacitor passes fast changes around , which is the phase lead. , , and the network attenuates DC by , which an amplifier makes up.
A lead compensator with a = 10 adds at most how much phase?
sin φmax = (10 − 1)/(10 + 1) = 0.818, φmax = 54.9°. One lead stage rarely adds more than 60°; beyond that, use two in series.
Worked examples
Maximum phase lead · Lead compensator with a = 4
- sin φmax = 3/5 → φmax = 36.87°
- Gain lift at ωm: 10 log 4 = 6.02 dB
Answer: 36.9°
RC lead network · R1 = 1 MΩ, C1 = 1 µF, R2 = 250 kΩ
- aT = R1C1 = 1 s; T = R1R2C1/(R1 + R2) = 0.19999999999999998 s → a = 5
- φmax = sin⁻¹(4/6) = 41.81°
- ωm = 1/(T√a) = 2.236 rad/s; DC gain 1/5
Answer: 41.8° at 2.24 rad/s
Lead design for 45° phase margin · (Kv = 10 fixed), PM is 18°; target 45°
- Phase to add: 45 − 18.0 + 5 = 32.0° → a = (1 + sin φ)/(1 − sin φ) = 3.26
- New crossover where |G| = −10 log a = -5.13 dB: ω = 4.19 rad/s
- T = 1/(ω√a) = 0.1322 s, aT = 0.4308 s
- Check: PM = 45.5°; step overshoot 60% → 27%
Answer: Gc = (1 + 0.431s)/(1 + 0.132s)
PID controllers and Ziegler-Nichols tuning
Which syllabus covers thisAU EE3503 Unit V; EC3351 Unit IIVTU BEE602 Module 5; BEC403 Module 3SPPU 303150 Unit 6; 204192 Unit VIMU not in syllabusKTU PCECT503 Module 3
What is on-off control, and what is a dead zone?
The simplest controller: full output when the error is positive, zero when it is negative, like a fridge thermostat. To stop it chattering the contacts, a dead zone (differential gap) is added: it switches on below one threshold and off above another. The output then cycles continuously inside the band, which is fine for temperature but not for position.
What do P, I and D each do?
Proportional: output proportional to error. Fast, but a type 0 plant keeps an offset of . Integral: output grows while any error remains, so the offset goes to zero, at the cost of overshoot and windup when the actuator saturates. Derivative: reacts to the rate of change, adding damping, but amplifies noise, so it is filtered and usually taken on the measurement.
How does the Ziegler-Nichols ultimate-gain method work?
With integral and derivative off, raise until the loop oscillates steadily: that gain is and the period is . Then set P: ; PI: , ; PID: , , . It aims for quarter-decay response, quite aggressive, so it is a starting point to fine-tune from.
And the reaction-curve method?
Open loop: step the actuator and record the S-shaped response. Draw the tangent at the inflection point: where it crosses the starting level gives the apparent dead time , its rise to the final value gives the time constant . For process gain : P ; PI , ; PID , , . No risky oscillation needed.
Plant 2/(s + 1) under P control with Kp = 4. The steady-state error to a unit step is:
Kp(0) × G(0) = 8, so the error is 1/(1 + 8) = 11.1%. Raising Kp shrinks it but never removes it; integral action does.
Worked examples
Offset under P control · Plant 2/(s + 1), Kp = 4, unit step
- Loop gain at DC = 4 × 2 = 8
- ess = 1/(1 + 8) = 0.1111
Answer: 11.1% offset; PI removes it
Ultimate-gain tuning · Plant
- Characteristic equation
- Routh: 3 × 3 = 1 + K → Ku = 8; ω = √3 → Pu = 3.628 s
- PID: Kp = 4.8, Ti = 1.814 s, Td = 0.453 s
- So Ki = Kp/Ti = 2.646, Kd = Kp·Td = 2.177
Answer: Kp 4.8, Ti 1.81 s, Td 0.45 s
Reaction-curve tuning · Step test: process gain K = 2, dead time L = 0.8 s, time constant T = 4 s
- PID: Kp = 1.2 × 4/(2 × 0.8) = 3.0
- Ti = 2L = 1.6 s, Td = 0.5L = 0.4 s
Answer: Kp 3.0, Ti 1.6 s, Td 0.4 s
State-space models, canonical forms and the state transition matrix
Which syllabus covers thisAU EE3503 Unit IV; EC3351 Unit VVTU BEE602 Module 5; BEC403 Module 5SPPU 204192 Unit VMU EEC503 Module 4KTU PCECT503 Module 4
What are state variables?
The smallest set of variables that, with the input, fixes the system's whole future: usually the energy stores, such as capacitor voltages, inductor currents, positions and velocities. An n-th order system has n states. The model is two matrix equations; it handles many inputs and outputs, non-zero initial conditions and time-varying systems that transfer functions cannot.
Is the state model unique?
No. Physical-variable form uses measurable quantities. Phase-variable (controllable canonical) form uses the output and its derivatives and reads straight off the transfer function's coefficients. Observable canonical form is its transpose. Diagonal form, from a similarity transform with the eigenvectors, decouples the modes. All share the same eigenvalues and the same transfer function.
How do you get the transfer function from a state model?
Take Laplace transforms with zero initial state: . The denominator is , so the eigenvalues of are the poles. A pole can vanish from if it is cancelled, which signals an uncontrollable or unobservable mode (chapter 18).
What is the state transition matrix?
. It carries the initial state forward: with no input, . Properties: , , . With an input, add the convolution of with .
What do the eigenvalues show in a phase portrait?
For two states, plot against . Real negative eigenvalues: a stable node, trajectories slide in along the eigenvectors. Complex with negative real part: a stable focus, spirals in. Purely imaginary: a centre, closed orbits. Opposite signs: a saddle, everything escapes except along one line. Trace and determinant of decide which.
A = [[0, 1], [−2, −3]]. Starting from x = (1, 0) with no input, after 1 s x1 is about:
Φ11(t) = 2e^−t − e^−2t, so x1(1) = 2 × 0.368 − 0.135 = 0.600. Both modes (−1 and −2) are decaying; the slower one dominates.
Worked examples
Controllable canonical form ·
- Denominator coefficients go in the last row: A = [[0, 1], [−2, −3]]
- B = [0; 1]
- Numerator coefficients form C = [3 1], D = 0
Answer: A = [[0, 1], [−2, −3]], B = [0; 1], C = [3 1]
Transfer function from the model · A = [[0, 1], [−2, −3]], B = [0; 1], C = [1 0]
- = [[s + 3, 1], [−2, s]]/(s² + 3s + 2)
- picks the (1, 2) entry
Answer: G = 1/(s² + 3s + 2); poles −1, −2 = eigenvalues of A
State transition matrix · Same A, x(0) = [1; 0], no input
- → partial fractions per entry
- Φ(t) = [[2e^−t − e^−2t, e^−t − e^−2t], [−2e^−t + 2e^−2t, −e^−t + 2e^−2t]]
- x1(1) = 2e^−1 − e^−2 = 0.6004
Answer: x1(1 s) = 0.600
Series RLC in state form · R = 4 Ω, L = 0.5 H, C = 0.02 F; states i and vC
- L di/dt = v − Ri − vC; C dvC/dt = i
- A = [[−8, −2], [50, 0]], B = [2; 0]
- det(sI − A) = s² + 8s + 100 → eigenvalues −4 ± j9.165
Answer: ωn = 10, ζ = 0.4, as in chapter 2
Controllability, observability, state feedback and digital control
Which syllabus covers thisAU EE3503 Units IV and V; EC3351 Unit VVTU not in syllabusSPPU 204192 Unit VI (digital controllers)MU not in syllabusKTU PCECT503 Module 4
What is controllability?
A system is controllable if some input can drive the state from anywhere to anywhere in finite time. Kalman's test: the matrix must have rank . If it does not, some mode cannot be influenced by the input at all, and no controller can move its pole.
And observability?
Observable if the initial state can be worked out from the output over a finite time. Test: must have rank . An unobservable mode is invisible at the output, which shows up as a pole-zero cancellation in the transfer function. You need observability to build an observer that estimates unmeasured states.
How does state feedback place the poles?
Feed back every state: . The closed loop is , and if the system is controllable can put the poles anywhere you like. Pick them from a and spec, match the characteristic polynomial (or use Ackermann's formula) to get , then choose for zero steady-state error.
What changes when a computer runs the loop?
The controller samples the output every seconds and holds its output constant between samples (zero-order hold). The plant is then described by a pulse transfer function in , and stability means every closed-loop pole inside the unit circle. Sampling adds lag of about , so a gain that is fine in continuous time can go unstable when is too long.
How fast should you sample?
Shannon says at least twice the highest frequency of interest; practice says 10 to 30 times the closed-loop bandwidth, so the hold's lag stays small and the response looks smooth. A motor current loop with 1 kHz bandwidth runs at 10–20 kHz; a temperature loop with a time constant of minutes is happy with a sample a second.
K/(s(s + 1)) is stable for every K in continuous time. Sampled with a zero-order hold every 1 s, it goes unstable above K ≈:
The constant term of the closed-loop polynomial, 0.368 + 0.264K, must stay below 1, so K < 2.39. The hold's delay costs phase that the continuous loop never had to give.
Worked examples
Controllable but not observable · A = [[0, 1], [−2, −3]], B = [0; 1], C = [1 1]
- [B AB] = [[0, 1], [1, −3]]: determinant −1 → rank 2, controllable
- [C; CA] = [[1, 1], [−2, −2]]: determinant 0 → rank 1, not observable
- Why: G = (s + 1)/[(s + 1)(s + 2)], the mode at −1 is cancelled
Answer: Controllable, not observable
Pole placement · Same A, B; want ζ = 0.6, ωn = 5 rad/s; output x1
- Desired: s² + 6s + 25
- A − BK = [[0, 1], [−2 − k1, −3 − k2]] → s² + (3 + k2)s + (2 + k1)
- k1 = 23, k2 = 3; DC gain of x1 is then 1/25, so N = 25
Answer: K = [23 3], N = 25
Critical gain of a sampled loop · G = K/(s(s + 1)) with ZOH, T = 1 s
- G(z) = K(0.368z + 0.264)/[(z − 1)(z − 0.368)]
- Closed loop: z² + (0.368K − 1.368)z + (0.368 + 0.264K) = 0
- Jury: 0.368 + 0.264K < 1 → K < 2.392; the other two conditions give K > 0 and K < 26.4
Answer: Stable for 0 < K < 2.39
Keep going
- GATE EE Control Systems: solved time-response and stability questions
- GATE EE Bode, Nyquist and state-space questions, solved
- Control Theory for PMSM course
- MATLAB and Simulink Basics course
- System Modelling and Simulation using Simulink and Simscape
- Free lessons: control theory for PMSM motors
- Mechatronics interactive book
- Robotics interactive book
- Electrical Machines interactive book
- All Skill-Lync interactive books
Recommended reading
- Control Systems Engineering, I. J. Nagrath and M. Gopal
- Automatic Control Systems, B. C. Kuo and F. Golnaraghi
- Modern Control Engineering, K. Ogata
- Control Systems Engineering, N. S. Nise
- Modern Control Systems, R. C. Dorf and R. H. Bishop
- Control Systems: Principles and Design, M. Gopal
- Control Systems, A. Anand Kumar
- Feedback and Control Systems (Schaum's Outline), J. J. DiStefano, A. R. Stubberud and I. J. Williams
How this book was made
Chapters follow the official syllabi listed above. Every interactive's formulas and every worked-example number were checked by script. Explanations, animations and examples are original to Skill-Lync. Topics informed by MIT OpenCourseWare 6.302 (Feedback Systems), CC BY-NC-SA. All explanations, animations and examples are original to Skill-Lync; MIT does not endorse this page.