Using Linear System Theory and Design for Real Control Problems
The fourth edition of Chi-Tsong Chen's Linear System Theory and Design is one of those textbooks that sits on every controls engineer's shelf. It's not the friendliest book to read cover to cover. It's denser than some alternatives, and the presentation style assumes you're comfortable with matrix algebra and state-space basics before you even get to chapter three. But it covers things most other texts skip or treat too lightly, like singular perturbations and robustness analysis. I've used this book for reference more than I've read it linearly. The chapters on state feedback and observer design are solid. The section on model reduction by balanced truncation is where it actually shines, and it's the part I keep coming back to when someone hands me a tenth-order model that needs to become something simulatable.
Getting the Linear System Theory And Design International Fourth Edition Pdf
The international edition is generally the same content as the US edition with different cover styling and sometimes different pagination on the problem sets. Most people looking for the pdf are graduate students or practicing engineers who need it on a laptop during a project. I won't link to any specific file here, but searching for the ISBN 978-0195109022 along with the edition year will surface what you need across academic repositories and textbook sites. What matters more than the format is knowing which chapters to actually use. Chapter 4 on state-space descriptions and canonical forms is worth reading if you're starting from scratch. Chapter 5 on stability is fine but brief. Chapter 6, controller and observer design, is where the book earns its keep. Chapter 8 on model reduction gets overlooked by students but is critical if you're dealing with high-order systems in practice. I ran into a specific problem last year that made me appreciate this book. I was working on a HVAC control loop for a large building system. The plant model the contractor gave me was a forty-eight state system identified from step response data. My task was to design a compensator that would work on the actual hardware, which had a microcontroller with very limited floating-point capability. The book's treatment of balanced realization and Hankel singular values in chapter 8 directly addressed this. I computed the Hankel singular values, saw that only six states had significant energy, truncated the rest, and designed a sixth-order observer-based controller. The simulation matched the real system within two percent across the full operating range.
The workaround I used wasn't in the book. Chen's presentation assumes continuous-time systems, but the implementation needed to be discrete. I discretized the reduced model using zero-order hold equivalence with a sampling period of 0.05 seconds, then designed the discrete compensator using the pole placement method from chapter 6. The mapping from continuous to discrete poles via z = e^(sT) worked without issue because all the dominant poles were well within the left half plane and far from the Nyquist frequency. One thing the book doesn't emphasize enough: the difference between what works on paper and what works numerically. Condition numbers of controllability and observability Gramians can blow up even for moderate-order systems. When I was reducing that forty-eighth order model, the Gramians had condition numbers above 10^12. Balancing the system first, which is exactly what chapter 8 describes, brought that down to around 10^3 and made the truncation stable. Skipping the balancing step produced a reduced model that was numerically garbage. Another counter-intuitive point that beginners miss: the book presents the separation principle as straightforward, but in practice, placing the observer poles too far left relative to the controller poles creates a situation where measurement noise gets amplified through the observer dynamics. I designed a system once where the observer bandwidth was five times the closed-loop bandwidth, and the output signal was practically unusable due to noise reshaping. The remedy was slowing the observer down and accepting a longer settling time, which the book implies but doesn't stress enough.
Get the Full Details
The exercises in this book are genuinely useful. They're not drill problems. The ones at the end of chapter 6 ask you to design observers with specific performance constraints, which is closer to what you'd actually encounter. The chapter 8 problems on model reduction include numerical cases where you have to verify the reduction quality, which is good training. Limitations of the book are worth stating plainly. It covers linear time-invariant systems almost exclusively. If your application involves time-varying parameters, gain scheduling, or nonlinearities, this book won't help you much past the local linearization step. The robustness coverage is decent but not as thorough as you might want for modern H-infinity type design. For that, you'd need something like Zhou, Doyle, and Glover or Skogestad and Postlethwaite. The discrete-time treatment is also thin. Chapter 3 touches on it but the depth isn't comparable to the continuous-time sections. Another practical issue: the notation shifts slightly between editions and between the main text and the solution manual, which can be annoying when you're cross-referencing. The international edition sometimes drops certain sections that appear in the US version, particularly around computational aspects of the state regulator problem.
If you're using this book self-study, don't try to work every problem. Focus on chapters 4, 6, and 8. Work through the derivations by hand at least once, because the matrix algebra looks simple until you're actually computing inverse of a 12-by-12 system matrix at 2 AM before a deadline. MATLAB's control system toolbox handles the heavy lifting, but understanding what's happening under the command makes debugging so much faster when the results look wrong. The book runs about 600 pages. It's dense. It's not entertaining. It does its job, and it does it well for the problems it covers. If your work stays in the LTI state-space domain, it's one of the better references available. If you're venturing into nonlinear or adaptive territory, keep it as a foundation text but look elsewhere for the advanced material.
