Key Concepts in Feedback Control Systems

Why This Matters

Feedback control systems are the backbone of modern engineering, from cruise control in your car to temperature regulation in industrial processes. When you're tested on Control Theory, you're not just being asked to recall definitions. You're being evaluated on whether you understand how systems behave dynamically, why stability matters, and what tools engineers use to predict and shape system performance. The concepts here connect mathematical representations, stability criteria, and controller design into a unified framework.

Don't just memorize transfer functions or PID tuning rules in isolation. Know what problem each concept solves: transfer functions let you analyze dynamics algebraically, stability methods tell you if a system will behave safely, and controllers give you the tools to shape that behavior. When you see an exam question, ask yourself: Is this about representing the system, analyzing its stability, or designing a controller? That mental framework will guide you to the right approach every time.

---

Mathematical Representations

Before you can analyze or control a system, you need a mathematical model that captures its dynamics. These representations translate physical behavior into equations you can manipulate and solve.

Transfer Functions

• Relates input to output in the Laplace domain, transforming differential equations into algebraic expressions using $G(s) = \frac{Y(s)}{U(s)}$
• Poles and zeros determine system behavior. Poles are roots of the denominator and directly indicate stability and response characteristics. Zeros are roots of the numerator and affect the shape of the transient response (for example, a zero near a dominant pole can increase overshoot).
• Foundation for frequency response analysis. Substitute $s = j\omega$ to get $G(j\omega)$, which tells you how the system responds to sinusoidal inputs at each frequency $\omega$.

State-Space Representation

• Describes systems using first-order differential equations in the form $\dot{x} = Ax + Bu$, $y = Cx + Du$. This naturally handles multiple inputs and outputs (MIMO systems).
• State variables capture all internal information needed to predict future behavior, not just the input-output relationship. Think of them as the minimum set of variables that fully describe the system's current condition.
• Enables modern control techniques like optimal control (LQR), state feedback, and observer design that transfer functions can't easily support.

Block Diagrams

• Visual representation of signal flow showing how inputs, outputs, feedback loops, and subsystems connect in a control architecture.
• Simplifies complex systems by breaking them into cascaded, parallel, or feedback configurations with clear reduction rules. For a standard negative feedback loop, the closed-loop transfer function is $\frac{G(s)}{1 + G(s)H(s)}$.
• Essential for identifying feedback loops. Understanding where signals are measured and where they're injected is critical for correctly writing the system equations.

---

Stability Analysis Methods

A control system is useless if it's unstable. Outputs that grow without bound or oscillate uncontrollably are dangerous. These methods determine whether a closed-loop system will behave predictably.

Stability Analysis Fundamentals

• A system is stable when all closed-loop poles have negative real parts. This means the system returns to equilibrium after disturbances rather than diverging.
• Pole locations in the complex $s$-plane tell the whole story. Poles in the left half-plane (LHP) indicate stability. Poles in the right half-plane (RHP) mean instability. Poles exactly on the imaginary axis indicate marginal stability, where the system sustains oscillations without growing or decaying.
• BIBO stability (bounded-input, bounded-output) ensures that any bounded input produces a bounded output. For linear time-invariant systems, BIBO stability is equivalent to all poles being in the LHP.

Root Locus Method

The root locus traces the paths of closed-loop poles as a parameter (usually gain $K$) varies from 0 to $\infty$. Poles start at the open-loop pole locations (when $K = 0$) and migrate toward the open-loop zeros (as $K \to \infty$).

• Design tool for selecting gain values. You can visually identify what gain ranges keep all poles in the stable left half-plane.
• Reveals transient response characteristics. The position of the dominant closed-loop poles indicates the damping ratio $\zeta$, natural frequency $\omega_n$, and whether the response will oscillate. Poles closer to the real axis are more heavily damped; poles farther from the origin respond faster.
• Key construction rules include: loci exist on the real axis to the left of an odd number of real poles and zeros, and asymptotes guide the loci toward infinity when there are more poles than zeros.

Nyquist Stability Criterion

• Determines closed-loop stability from the open-loop frequency response. You plot $G(j\omega)H(j\omega)$ as $\omega$ varies from $-\infty$ to $+\infty$, creating a closed contour in the complex plane.
• Counts encirclements of the critical point $-1 + j0$. The relationship $Z = N + P$ connects the number of clockwise encirclements ($N$) and the number of open-loop RHP poles ($P$) to the number of unstable closed-loop poles ($Z$). For stability, you need $Z = 0$.
• Handles systems with time delays and non-minimum phase behavior that other methods struggle with. This makes it especially useful for real-world systems where you may only have experimental frequency response data.

---

Frequency-Domain Tools

Analyzing how systems respond to sinusoidal inputs at different frequencies reveals stability margins and performance limits. These tools are essential for practical controller tuning.

Bode Plots

• Two plots: magnitude (in dB) and phase (in degrees) versus frequency on a logarithmic scale. Magnitude is computed as $20\log_{10}|G(j\omega)|$.
• Asymptotic approximations make hand-sketching possible. Each pole contributes a slope change of $-20$ dB/decade at its break frequency, and each zero contributes $+20$ dB/decade. You build the full plot by summing individual contributions.
• Directly shows bandwidth and frequency response characteristics. The bandwidth (roughly where magnitude drops 3 dB below its low-frequency value) correlates with speed of response: higher bandwidth means faster response.

Gain and Phase Margins

These are your primary measures of how robustly stable a system is, not just whether it's stable.

• Gain margin measures how much you can increase $K$ before instability. Find the frequency where phase equals $-180°$ (the phase crossover frequency), then read how far the magnitude is below 0 dB at that frequency. That gap is your gain margin.
• Phase margin measures how much additional phase lag the system can tolerate. Find the frequency where magnitude equals 0 dB (the gain crossover frequency), then read how far the phase is above $-180°$. That gap is your phase margin.
• Typical design targets are gain margin $> 6$ dB and phase margin between $30°$ and $60°$. These provide a reasonable buffer against modeling errors and parameter variations.

Frequency-Domain Analysis

• Evaluates system behavior across a spectrum of input frequencies. This is critical for understanding how noise (typically high frequency), disturbances, and reference signals affect the output.
• Complements time-domain analysis. Frequency response characterizes steady-state sinusoidal behavior, while time-domain analysis shows transient response to steps and impulses. Together they give a complete picture.
• Foundation for loop shaping, where you design controllers by manipulating the open-loop frequency response to achieve desired gain and phase margins while meeting bandwidth requirements.

---

Time-Domain Performance

Ultimately, you care about how the system output evolves over time in response to commands and disturbances. These concepts quantify what "good performance" actually means.

Time-Domain Analysis

• Examines response to standard test inputs. Step inputs reveal speed and overshoot, impulse inputs reveal the system's natural modes, and ramp inputs reveal tracking ability.
• Key metrics include rise time $t_r$, settling time $t_s$, and percent overshoot $M_p$. These specifications drive controller design requirements. For a second-order system, $M_p$ depends only on $\zeta$, and $t_s$ depends on $\zeta \omega_n$.
• Directly connects to pole locations. The real part of a pole determines how fast the exponential envelope decays (affecting $t_s$), while the imaginary part determines oscillation frequency. A larger damping ratio $\zeta$ means less overshoot.

System Response (Transient and Steady-State)

• Transient response describes behavior immediately after a change. It's dominated by system poles and initial conditions, typically lasting a few time constants (roughly $4/(\zeta\omega_n)$ for 2% settling in a second-order system).
• Steady-state response is the long-term behavior after transients decay. It determines tracking accuracy and whether the output actually reaches the desired value.
• Second-order system approximations are powerful. Many higher-order systems behave approximately like $\frac{\omega_n^2}{s^2 + 2\zeta\omega_n s + \omega_n^2}$ when one pair of poles dominates. Here $\zeta$ is the damping ratio and $\omega_n$ is the natural frequency.

Error Analysis

Steady-state error measures the persistent gap between the desired output and the actual output after transients die out. The system type (the number of free integrators in the open-loop transfer function) determines how well the system tracks different input types.

• Type 0 systems have finite steady-state error to a step input and infinite error to ramps.
• Type 1 systems track steps with zero error but have finite error to ramp inputs.
• Type 2 systems track both steps and ramps perfectly but have finite error to parabolic (acceleration) inputs.

The error constants $K_p$, $K_v$, and $K_a$ quantify this behavior:

• $K_p = \lim_{s \to 0} G(s)$ (position error constant)
• $K_v = \lim_{s \to 0} sG(s)$ (velocity error constant)
• $K_a = \lim_{s \to 0} s^2 G(s)$ (acceleration error constant)

For example, steady-state error to a unit step is $e_{ss} = \frac{1}{1 + K_p}$, and to a unit ramp is $e_{ss} = \frac{1}{K_v}$.

---

Controller Design

Once you understand system behavior, you need tools to shape it. Controllers modify the system dynamics to meet performance and stability requirements.

PID Controllers

The PID controller combines three control actions, each addressing a different aspect of the error signal $e(t)$:

1. Proportional ($K_p e$): Responds to the current error. Higher $K_p$ reduces steady-state error and speeds up response, but too much gain causes excessive overshoot or instability.
2. Integral ($K_i \int e \, dt$): Accumulates past error. This adds a free integrator to the loop, which eliminates steady-state error (effectively increasing the system type by one). The tradeoff is that integral action can slow the response and cause integral windup if the actuator saturates.
3. Derivative ($K_d \frac{de}{dt}$): Responds to the rate of change of error, providing a predictive or damping effect. It improves transient response but amplifies high-frequency noise, so it's often used with a low-pass filter in practice.

PID controllers are used in over 90% of industrial control loops due to their simplicity and effectiveness. Tuning methods like Ziegler-Nichols provide starting points based on the system's ultimate gain and ultimate period, but achieving the right balance between speed, stability, and noise sensitivity typically requires iteration.

Feedback and Feedforward Control

• Feedback control measures the output and adjusts the input to reduce error. It's a reactive approach that handles unknown disturbances but inherently introduces some delay in correction.
• Feedforward control measures disturbances directly and compensates before they affect the output. It's faster but requires an accurate model of how the disturbance enters the system.
• Combining both yields the best performance. Feedforward handles predictable, measurable disturbances quickly, while feedback corrects for model errors and unmeasured effects.

Controllability and Observability

These are fundamental properties of a state-space model that determine whether certain control strategies are even possible.

• Controllability asks: can you drive the system from any initial state to any desired state using the available inputs? The controllability matrix $\mathcal{C} = [B \; AB \; A^2B \; \cdots \; A^{n-1}B]$ must have full row rank (rank $n$, where $n$ is the number of states). If a state is uncontrollable, no input can influence it.
• Observability asks: can you determine all internal states from the output measurements alone? The observability matrix $\mathcal{O} = [C; \; CA; \; CA^2; \; \cdots; \; CA^{n-1}]^T$ must have full column rank. If a state is unobservable, you can't reconstruct it from measurements, which means you can't build an observer (state estimator) for it.
• Both are prerequisites for advanced control. You can't apply state feedback to states you can't reach (controllability), and you can't estimate states you can't see (observability).

---

Quick Reference Table

---

Self-Check Questions

1. What do transfer functions and state-space representations have in common, and when would you choose one over the other?

2. A system has all poles in the left half-plane but a phase margin of only 5°. Is it stable? Would you trust it in practice? Explain your reasoning.

3. Compare the root locus method and Bode plot analysis: what type of information does each provide, and how do they complement each other in controller design?

4. If a system has zero steady-state error to a step input but finite error to a ramp input, what can you conclude about the system type? How would adding integral action change this?

5. Explain why a system might be stable but not controllable, or controllable but not observable. Give a physical interpretation of what each situation means for controller design.