Feedback Systems, 2nd Edition (Åström & Murray, 2020)
A current, freely published 573-page systems-and-control textbook that integrates nonlinear dynamics, state-space methods, transfer functions, frequency-domain analysis and design, PID, robust performance, fundamental limits, and control architecture. It is the principal modern reference used to reconcile the supplied TAU courses.
Karl J. Åström and Richard M. Murray's second edition is the modern backbone for the TAU material. The authors and Princeton University Press make the complete book available through the official Caltech/FBS site. Unlike a narrow classical-control text, it begins with feedback as a systems principle, develops nonlinear dynamics before specializing to linear representations, and reconnects local design rules to robustness, uncertainty, saturation, and architecture.
Structure and role #
| Chapters | Main contribution to GTK |
|---|---|
| 1–4 | Feedback and feedforward principles, modeling methodology, and physical examples across multiple domains |
| 5–6 | Qualitative nonlinear dynamics, equilibria, phase portraits, Lyapunov stability, linear systems, forced response, and linearization |
| 7–8 | State feedback, integral action, observability, estimation, output feedback, and controller structure |
| 9–10 | Transfer functions, block diagrams, Bode plots, Nyquist, margins, minimum-phase structure, and generalized gain/phase concepts |
| 11–12 | PID implementation, sensitivity functions, performance specifications, loop shaping, root locus, and frequency-domain design |
| 13–14 | Uncertainty, robust performance, nonlinear effects, Bode's integral, gain-crossover limits, and robust pole placement |
| 15 | System architecture, interaction, adaptation, learning, and application-domain design patterns |
The book does not replace specialized digital-control treatment, but it establishes the continuous and systems-level semantics against which a discrete implementation must be checked. It is especially valuable where the older supplied notes present a construction without its limitations: pole-zero cancellation, bandwidth increase, integral action, high loop gain, describing functions, or local linearization.
Principles carried into the entries #
- Feedback is a map of causal paths. Tracking, disturbance rejection, noise transmission, robustness, and control effort have different transfer functions even when they share one denominator.
- A representation is a view, not a different system. State space, transfer functions, time response, Bode plots, Nyquist curves, and root loci expose different properties and have different hidden assumptions.
- Linearization is local. A stable linearized equilibrium supplies local information; it does not certify a distant operating point, actuator constraint, or global region of attraction.
- Performance and robustness are coupled. Large low-frequency gain can reduce error, but crossover, delay, unmodeled dynamics, noise, and actuator authority constrain where gain may be placed.
- Fundamental limits cannot be tuned away. Right-half-plane zeros, unstable poles, delays, sensitivity integrals, and causality impose tradeoffs that compensation can relocate but not erase.
- Implementation closes the proof. Saturation, windup, mode transfer, scheduling, sensing, and computation belong to the closed-loop system.
Visual audit #
The archived 573-page PDF was text-extracted and every page was rendered. All 23 contact sheets were inspected, including front matter, references, and index. No page is text-empty. The book contains a dense, high-quality visual language: multi-domain feedback examples, nonlinear nullclines and phase portraits, time responses, state-feedback structures, block reductions, Bode and Nyquist constructions, sensitivity plots, root loci, robust-performance geometry, fundamental-limit illustrations, and top-down/bottom-up architectures.
GTK does not reproduce those figures. The new topic set uses independent SVGs tailored to article width and to a single inference per figure. This avoids shrinking textbook plates, removes example-specific clutter, and permits dark/light themes. The source audit still determines what must not disappear: sign conventions, operating points, pole and zero locations, encirclement direction, crossover definitions, invariant-set boundaries, continuous/discrete signal boundaries, and the distinction between a causal block diagram and a merely algebraic manipulation.
Use boundary #
This source is authoritative for the book's stated mathematics and design framework, not for every current product or software API. Current MathWorks, MIT, ARM, and other primary/official documentation is used where the GTK entry makes claims about contemporary tools, numerical algorithms, or embedded realization behavior.