6. State-Space Modeling and Modern ControlIntroduction and ContextClassical control analysis is usually built around transfer functions, block diagrams, and frequency-response methods. These tools remain valuable, particularly for single-input single-output systems. However, many contemporary engineering systems are multivariable, internally coupled, subject to nonzero initial conditions, and implemented digitally. For such systems, a description based only on input–output behavior is often incomplete. State-space modeling addresses this limitation by representing the internal dynamic condition of a system through a set of state variables. The state-space viewpoint provides a unified foundation for modeling, simulation, stability analysis, controllability, observability, state feedback, pole placement, integral control, and observer design. It also forms the conceptual bridge between classical control and modern methods such as optimal control, robust control, model predictive control, and estimation. A continuous-time linear time-invariant system is represented as \[ \dot{x}(t)=Ax(t)+Bu(t),\qquad y(t)=Cx(t)+Du(t), \] where \(x\in\mathbb{R}^n\) is the state vector, \(u\in\mathbb{R}^m\) is the input, and \(y\in\mathbb{R}^p\) is the output. The matrices \(A,B,C,D\) define the state-space realization. Learning ObjectivesAfter studying this chapter, the student should be able to: 1. Formulate continuous-time state models from differential equations, physical laws, and transfer functions. 2. Interpret states, system matrices, eigenvalues, eigenvectors, and modes. 3. Derive and use the state-transition matrix for zero-input and zero-state responses. 4. Test controllability and observability and explain their engineering significance. 5. Construct controllable, observable, and modal canonical realizations. 6. Design state-feedback controllers using pole-placement methods. 7. Add integral action to eliminate steady-state tracking error. 8. Design full-order observers and observer-based compensators. 9. Recognize modeling assumptions, numerical issues, implementation risks, and practical limitations. 6.1 From Differential Equations to State ModelsA state is a variable that, together with the future input, completely determines the future behavior of a system. For an energy-storage system, states are often capacitor voltages, inductor currents, positions, velocities, temperatures, or chemical concentrations. A state vector need not be unique; many different coordinate choices can describe the same physical dynamics. Consider the second-order equation \[ \ddot{y}+a_1\dot{y}+a_0y=b_0u. \] Choosing \(x_1=y\) and \(x_2=\dot y\) gives \[ \dot{x}_1=x_2, \] \[ \dot{x}_2=-a_0x_1-a_1x_2+b_0u, \] \[ y=x_1. \] Thus, \[ A=\begin{bmatrix}0&1\\-a_0&-a_1\end{bmatrix},\quad B=\begin{bmatrix}0\\b_0\end{bmatrix},\quad C=\begin{bmatrix}1&0\end{bmatrix},\quad D=0. \] For a mechanical mass–spring–damper system, \[ m\ddot q+c\dot q+kq=F, \] selecting position and velocity as states gives \[ \dot{x}=\begin{bmatrix}0&1\\-k/m&-c/m\end{bmatrix}x+\begin{bmatrix}0\\1/m\end{bmatrix}F. \] The state variables should be physically meaningful where possible, but mathematical convenience is also important. A model with physically measurable states may not be the most convenient form for pole placement or numerical computation. Transfer Function to State SpaceFor a strictly proper transfer function \[ G(s)=\frac{b_{n-1}s^{n-1}+\cdots+b_1s+b_0}{s^n+a_{n-1}s^{n-1}+\cdots+a_1s+a_0}, \] one realization is the controllable companion form: \[ A=\begin{bmatrix} 0&1&0&\cdots&0\\ 0&0&1&\cdots&0\\ \vdots&&&\ddots&\vdots\\ 0&0&0&\cdots&1\\ -a_0&-a_1&-a_2&\cdots&-a_{n-1} \end{bmatrix}, \quad B=\begin{bmatrix}0\\0\\\vdots\\0\\1\end{bmatrix}, \] with \(C=[b_0\ b_1\ \cdots\ b_{n-1}]\) and \(D=0\), subject to the chosen polynomial convention. A nonzero direct term is represented by \(D\), and care is required when the transfer function is improper or has pole-zero cancellations. A transfer function specifies only the input–output map. It does not uniquely specify internal states. Two realizations related by a nonsingular similarity transformation are dynamically equivalent: \[ z=T^{-1}x, \quad A_z=T^{-1}AT, \quad B_z=T^{-1}B, \quad C_z=CT. \] The transfer function remains unchanged, although the numerical values and interpretation of the states differ. 6.2 Eigenstructure, Modes, and StabilityThe eigenvalues of \(A\) are the roots of \[ \det(sI-A)=0. \] For a continuous-time LTI system, asymptotic stability requires every eigenvalue of \(A\) to have a strictly negative real part. Eigenvalues with positive real parts produce growing modes; eigenvalues on the imaginary axis require further qualification and generally prevent asymptotic stability. If \(A\) is diagonalizable, then \[ A=V\Lambda V^{-1}, \] where \(\Lambda\) contains the eigenvalues. The state response is a combination of modal responses. Real eigenvalues generate exponential modes, while complex conjugate eigenvalues generate damped oscillations. The real part determines decay rate and the imaginary part determines oscillation frequency. A repeated eigenvalue may have fewer independent eigenvectors than its algebraic multiplicity. In that case, a Jordan form describes generalized modes containing terms such as \(te^{\lambda t}\). This distinction matters in stability and transient-response analysis. The characteristic polynomial determines eigenvalues, but eigenvectors determine how states participate in each mode. In multivariable systems, two systems may have identical eigenvalues but substantially different transient behavior because their eigenvectors and input/output directions differ. Non-normal matrices can also produce large transient amplification even when all eigenvalues indicate asymptotic stability. 6.3 State Transition and System ResponseThe state-transition matrix is defined by \[ \Phi(t)=e^{At}=I+At+\frac{A^2t^2}{2!}+\cdots. \] For the state equation \(\dot x=Ax+Bu\), the solution is \[ x(t)=e^{A(t-t_0)}x(t_0)+\int_{t_0}^{t}e^{A(t-\tau)}Bu(\tau)d\tau. \] The first term is the zero-input response caused by the initial state. The integral is the forced response caused by the input. The output is \[ y(t)=Cx(t)+Du(t). \] Useful properties include \(\Phi(0)=I\), \(\Phi(t_1+t_2)=\Phi(t_1)\Phi(t_2)\), and \(d\Phi/dt=A\Phi=\Phi A\). For a constant input \(u(t)=u_0\), the forced response can be evaluated through integration or, when \(A\) is nonsingular, by using \(A^{-1}(e^{At}-I)Bu_0\). The transfer function follows from the state model by taking Laplace transforms with zero initial conditions: \[ G(s)=C(sI-A)^{-1}B+D. \] This expression demonstrates the relationship between internal dynamics and the external input–output description. It also shows why pole-zero cancellations can conceal internal modes in a transfer function. 6.4 Controllability and ObservabilityA state is controllable if an admissible input can move the system from that state to a desired state in finite time. For an \(n\)-state single-input or multi-input system, the controllability matrix is \[ \mathcal C=[B\ AB\ A^2B\ \cdots\ A^{n-1}B]. \] The pair \((A,B)\) is completely controllable if \[ \operatorname{rank}(\mathcal C)=n. \] If the rank is deficient, some state directions cannot be influenced by the input. The Popov–Belevitch–Hautus test provides an alternative: \[ \operatorname{rank}[\lambda I-A\ \ B]=n \] for every eigenvalue \(\lambda\) of \(A\). Observability is the dual concept. A system is observable if its initial state can be uniquely reconstructed from input and output measurements over a finite interval. The observability matrix is \[ \mathcal O=\begin{bmatrix}C\\CA\\CA^2\\\vdots\\CA^{n-1}\end{bmatrix}. \] The pair \((A,C)\) is completely observable if \(\operatorname{rank}(\mathcal O)=n\). The PBH test is \[ \operatorname{rank}\begin{bmatrix}\lambda I-A\\C\end{bmatrix}=n \] for every eigenvalue \(\lambda\). Controllability and observability are structural properties of a model, but their numerical conditioning determines practical usefulness. A system may be theoretically controllable yet require unreasonably large actuator effort in some direction. Similarly, it may be observable in theory but highly sensitive to measurement noise. Gramian-based measures help quantify these issues. For a stable system, the controllability Gramian satisfies \[ AW_c+W_cA^T+BB^T=0, \] and the observability Gramian satisfies \[ A^TW_o+W_oA+C^TC=0. \] Small Gramian eigenvalues indicate weakly controllable or weakly observable directions. 6.5 Realizations and Canonical FormsA realization is a particular quadruple \((A,B,C,D)\) that produces a transfer function. A realization is minimal when it is both controllable and observable. Minimality is important because uncontrollable or unobservable states do not contribute fully to the external behavior and may conceal unstable internal dynamics. The controllable canonical form is useful for pole placement because the characteristic polynomial coefficients appear directly in the last row of \(A\). The observable canonical form is the dual representation obtained by transposition. Modal form is useful when eigenvalues are distinct and the transformation matrix is well conditioned; it separates independent modes. Jordan form is theoretically informative but often numerically unsuitable because it is highly sensitive to perturbations. Canonical forms should be viewed as analytical tools rather than mandatory implementation forms. Physical-coordinate models are usually preferable for interpreting sensors, actuators, constraints, and energy flows. Balanced realizations, which balance controllability and observability measures, are often more appropriate for model reduction and numerical computation. 6.6 State-Feedback DesignSuppose all states are available for feedback and the control law is \[ u=-Kx+r, \] where \(r\) is a reference-related signal. The closed-loop dynamics are \[ \dot{x}=(A-BK)x+Br. \] The feedback gain \(K\) changes the closed-loop eigenvalues. Arbitrary pole placement is possible if and only if \((A,B)\) is controllable, subject to practical limitations. For single-input systems, Ackermann’s formula is \[ K=e_n^T\mathcal C^{-1}\phi(A), \] where \(e_n^T=[0\ \cdots\ 0\ 1]\) and \(\phi(s)\) is the desired characteristic polynomial. For multi-input systems, numerical algorithms based on orthogonal transformations are generally preferred. Direct use of symbolic formulas can be ill-conditioned, especially for high-order systems. Pole placement is not simply a matter of choosing poles as far left as possible. Fast poles can increase actuator demand, amplify sensor noise, magnify unmodeled dynamics, and create sampling challenges. State feedback alone does not guarantee zero steady-state error to a constant reference. If the plant has suitable type or if a reference precompensator is accurately calculated, tracking may be acceptable, but parameter uncertainty can cause residual error. Integral action provides a more robust solution. Integral ActionLet the tracking error be \(e=r-y\), and introduce an integrator state \(z\): \[ \dot z=r-y. \] For \(y=Cx+Du\), the augmented model is \[ \begin{bmatrix}\dot x\\\dot z\end{bmatrix} = \begin{bmatrix}A&0\\-C&0\end{bmatrix} \begin{bmatrix}x\\z\end{bmatrix} + \begin{bmatrix}B\\-D\end{bmatrix}u+ \begin{bmatrix}0\\I\end{bmatrix}r. \] A control law such as \(u=-K_xx-K_iz\) can place the augmented poles. Integral action must be accompanied by anti-windup protection when actuators saturate. Otherwise, the integrator continues accumulating error while the actuator is constrained, producing overshoot and long recovery times. 6.7 Observers and Output-Feedback ControlIn many systems, not every state can be measured economically or reliably. A full-order Luenberger observer estimates the state using \[ \dot{\hat x}=A\hat x+Bu+L(y-C\hat x). \] Defining the estimation error \(e_o=x-\hat x\) gives \[ \dot e_o=(A-LC)e_o. \] Therefore, the observer gain \(L\) is selected so that the eigenvalues of \(A-LC\) have desirable locations. Observer pole placement is possible when \((A,C)\) is observable. The duality between controller and observer design follows from applying state-feedback methods to \((A^T,C^T)\). A common guideline is to make observer poles faster than controller poles, but excessive speed increases sensitivity to measurement noise and model errors. The appropriate bandwidth depends on sensor quality, sampling rate, actuator dynamics, and unmodeled high-frequency behavior. Combining the control law \(u=-K\hat x\) with the observer produces an observer-based compensator. For a linear time-invariant system, the separation principle states that the controller poles and observer poles can be designed independently, and the combined closed-loop poles are the union of the two sets. This principle does not remove practical constraints: saturation, nonlinearities, quantization, delays, and unstable unmodeled dynamics can invalidate ideal conclusions. 6.8 Discrete Implementation and Computational PracticeA continuous model implemented digitally must be discretized. With zero-order-hold sampling period \(T_s\), \[ x[k+1]=A_dx[k]+B_du[k], \] where \[ A_d=e^{AT_s},\qquad B_d=\int_0^{T_s}e^{A\tau}B\,d\tau. \] Continuous-time pole locations map to discrete-time poles through \(z=e^{sT_s}\). Sampling too slowly can distort the intended dynamics and reduce achievable control bandwidth. Numerical scaling, sensor filtering, computational delay, and finite word length should be considered during implementation. Modern software packages can compute matrix exponentials, rank, Gramians, pole-placement gains, and observer gains. Such tools improve productivity but do not replace modeling judgment. Rank tests based on poorly scaled matrices may give misleading results; balancing, scaling, singular-value analysis, and physical validation are important. Simulation should include disturbances, parameter variation, sensor noise, saturation, initialization errors, and sampling effects rather than relying only on nominal step responses. 6.9 Illustrative Case: Speed Regulation of a DC MotorBackgroundConsider a simplified armature-controlled DC motor whose state variables are armature current \(i_a\) and angular speed \(\omega\). The voltage input is \(v_a\), and the measured output is speed. The model is \[ \dot i_a=-\frac{R}{L}i_a-\frac{K_e}{L}\omega+\frac{1}{L}v_a, \] \[ \dot\omega=\frac{K_t}{J}i_a-\frac{b}{J}\omega-\frac{1}{J}T_L, \] where \(T_L\) is load torque. Situation and ProblemA speed command must be tracked despite load-torque changes. Measuring current and speed is possible, but current measurement may be noisy and the load torque is not directly measured. A proportional speed controller may leave a steady-state error when the load changes. AnalysisWith \(x=[i_a\ \omega]^T\), the motor becomes a second-order state model. Controllability depends on the voltage input influencing current and current influencing speed. Observability depends on the speed measurement and the coupling between current and speed. In realistic parameter ranges, both properties generally hold, but numerical conditioning should still be checked. ApproachA state-feedback law \(v_a=-Kx+k_r r\) can assign the motor’s closed-loop poles. To reject constant load torque and remove speed offset, an integrator state \(z=\int(r-\omega)dt\) is added, and the voltage command becomes \[ v_a=-K_xx-K_iz. \] If current is not measured, a Luenberger observer estimates it from voltage and speed. The observer is designed with poles faster than the desired motor-control poles, but not so fast that current-sensor noise is excessively amplified. Voltage saturation is included in simulation, together with anti-windup logic. OutcomeThe resulting architecture combines state feedback, integral action, and state estimation. It can provide fast speed regulation, load-disturbance rejection, and reduced sensor requirements. Performance must be verified under parameter changes in resistance, inertia, friction, and motor constants. Learning PointsThe example illustrates that state-space design is not merely pole assignment. Successful engineering requires a model, controllability and observability assessment, disturbance analysis, actuator constraints, observer tuning, and validation under uncertainty. 6.10 Limitations, Risks, and Professional PracticeState-space models are commonly linearized around an operating point. Large excursions may invalidate the linear model, especially in magnetic actuators, power converters, aircraft dynamics, and chemical processes. Parameter uncertainty can shift poles and reduce robustness. Unmodeled delays and actuator dynamics can undermine a design that appears satisfactory in simulation. Common mistakes include confusing state variables with outputs, omitting direct feedthrough, using inconsistent sign conventions, placing observer poles excessively fast, ignoring actuator saturation, and interpreting numerical rank without considering conditioning. Another error is treating a transfer-function cancellation as proof that an internal unstable mode has disappeared; hidden unstable modes may remain uncontrollable or unobservable. Professional design should document assumptions, units, operating conditions, sensor and actuator limits, software versions, numerical tolerances, and validation tests. In safety-critical applications, independent review, traceability, fault handling, cybersecurity controls, and compliance with applicable industry standards are essential. A controller should fail safely when sensors, communications, or computation become unavailable. 6.11 Emerging DirectionsState-space methods now support digital twins, online parameter estimation, fault diagnosis, autonomous systems, and model predictive control. Kalman filtering extends observer concepts to stochastic systems by combining models with noisy measurements. Data-driven and learning-based controllers can complement physics-based models, but they introduce questions of extrapolation, explainability, verification, and safe exploration. A useful engineering direction is hybrid modeling: physical state equations provide structure, while data refine uncertain parameters or unmodeled effects. Best Practices for State-Space Design- Define states from energy storage, dynamic order, and measurement requirements.
- Check dimensions, units, signs, and equilibrium conditions before design.
- Verify the model against differential equations and limiting cases.
- Test controllability and observability using rank and conditioning measures.
- Prefer numerically stable algorithms over symbolic formulas for large systems.
- Place controller and observer poles with actuator, sensor, sampling, and robustness limits in mind.
- Include integral anti-windup, saturation, noise, delays, and parameter uncertainty in validation.
- Preserve physical meaning and document every coordinate transformation.
- Treat software output as a design aid, not as a substitute for engineering judgment.
Chapter SummaryState-space modeling represents a dynamic system through its internal state, input, and output variables. The matrices \(A,B,C,D\) describe dynamics, actuation, measurement, and direct feedthrough. Similarity transformations generate equivalent realizations, while minimal realizations are both controllable and observable. Eigenvalues determine modal stability and transient characteristics, and the matrix exponential provides the state-transition solution. Controllability determines whether inputs can influence all states; observability determines whether states can be reconstructed from measurements. State feedback permits closed-loop pole placement when controllability is present. Integral augmentation removes steady-state tracking error, while observers estimate unmeasured states when observability is available. Practical implementation requires attention to numerical conditioning, saturation, noise, delays, uncertainty, sampling, safety, and validation. Key TermsState: A variable that, with the input, determines future system behavior. State vector: An organized vector containing the system states. State-space model: A set of first-order differential equations describing state and output dynamics. Realization: A particular \((A,B,C,D)\) representation of an input–output system. Similarity transformation: A nonsingular coordinate transformation preserving the transfer function. State-transition matrix: The matrix \(e^{At}\) describing autonomous state evolution. Eigenvalue: A root of \(\det(sI-A)\) that characterizes a system mode. Eigenvector: A direction in state space associated with an eigenvalue. Controllability: The ability to move the state to a desired condition using the input. Observability: The ability to reconstruct the state from measured inputs and outputs. Controllability matrix: \([B\ AB\ \cdots\ A^{n-1}B]\). Observability matrix: The matrix formed by stacking \(C,CA,\ldots,CA^{n-1}\). Minimal realization: A realization that is both controllable and observable. Canonical form: A structured realization useful for analysis or design. State feedback: A control law formed from measured or estimated states. Pole placement: Selection of feedback gains to assign closed-loop eigenvalues. Integral action: Feedback involving the time integral of tracking error. Observer: A dynamic estimator of unmeasured system states. Separation principle: The result that controller and observer poles can be designed independently for an ideal linear system. Gramian: A matrix quantifying finite-energy controllability or observability. Review QuestionsShort Answer Questions1. What is meant by a state, and why is a transfer function alone sometimes insufficient? 2. State the continuous-time state-space equations and identify the role of each matrix. 3. What condition on the eigenvalues of \(A\) ensures asymptotic stability? 4. Define controllability and observability in physical terms. 5. Why is integral action added to a state-feedback controller? Long Answer Questions1. Derive a state-space model for a second-order mass–spring–damper system and explain the meaning of each state. 2. Explain the state-transition matrix and derive the complete solution of \(\dot x=Ax+Bu\). 3. Compare controllable, observable, minimal, modal, and canonical realizations. 4. Develop the design procedure for state feedback with integral action. 5. Explain full-order observer design, estimation-error dynamics, and the separation principle. Analytical / Application Questions1. For \(A=\begin{bmatrix}0&1\\-4&-3\end{bmatrix}\), \(B=\begin{bmatrix}0\\1\end{bmatrix}\), and \(C=[1\ 0]\), test controllability and observability, and determine whether the open-loop system is asymptotically stable. 2. A plant has an observable but poorly conditioned output model. Explain why a theoretically valid fast observer may perform badly and propose practical remedies. 3. Design an evaluation plan for a digitally implemented state-feedback controller with an observer and integral action. Include sampling, saturation, noise, uncertainty, and safety tests. 22. Multiple Choice Questions with AnswersQuestion 1For the model ẋ = Ax + Bu, which condition guarantees complete controllability of an n-state system? A. rank(C) = n B. rank([B AB ... A^(n−1)B]) = n C. All eigenvalues of C are negative D. det(A) = 0 Correct Answer: B. rank([B AB ... A^(n−1)B]) = n Question 2The continuous-time state-transition matrix for an LTI system is A. A^t B. tA C. e^(At) D. (sI−A)B Correct Answer: C. e^(At) Question 3If all eigenvalues of A have strictly negative real parts, the autonomous system ẋ = Ax is A. Marginally stable only B. Asymptotically stable C. Uncontrollable D. Unobservable Correct Answer: B. Asymptotically stable Question 4Which statement best describes a minimal realization? A. It has the largest possible number of states B. It is diagonal in every coordinate system C. It is both controllable and observable D. It has no input matrix Correct Answer: C. It is both controllable and observable Question 5For state feedback u = −Kx, the closed-loop state matrix is A. A + BK B. A − BK C. K − AB D. A − KC Correct Answer: B. A − BK Question 6The main purpose of adding integral action to a tracking controller is to A. Increase sensor bandwidth B. Eliminate steady-state error for suitable constant references and disturbances C. Make every plant controllable D. Remove the need for an actuator Correct Answer: B. Eliminate steady-state error for suitable constant references and disturbances Question 7For the observer x̂̇ = Ax̂ + Bu + L(y − Cx̂), the estimation-error dynamics are governed by A. A + LC B. A − LC C. A − BK D. C − LA Correct Answer: B. A − LC Question 8Arbitrary closed-loop pole placement by state feedback is possible, in the ideal linear model, when A. The plant is observable B. The plant has no direct feedthrough C. The pair (A,B) is controllable D. The matrix A is symmetric Correct Answer: C. The pair (A,B) is controllable Question 9Why can excessively fast observer poles be undesirable? A. They always make the plant uncontrollable B. They can amplify measurement noise and model uncertainty C. They eliminate all closed-loop dynamics D. They prevent digital implementation in every case Correct Answer: B. They can amplify measurement noise and model uncertainty Question 10Under zero-order-hold discretization with sampling period T, the discrete state matrix is A. A_d = A/T B. A_d = e^(AT) C. A_d = T^A D. A_d = I − AT for all systems Correct Answer: B. A_d = e^(AT)
|