Skip to content

HInfinityStateFeedback.md

Latest commit

 

History

History
148 lines (108 loc) · 6.88 KB

HInfinityStateFeedback.md

File metadata and controls

148 lines (108 loc) · 6.88 KB
Raw

H∞ State-Feedback Control

Overview & Motivation

Robust control guarantees performance under worst-case conditions rather than average ones. H∞ state-feedback synthesizes a linear gain that minimizes the largest possible ratio of output energy to disturbance energy across all bounded disturbances — the induced $\mathcal{L}_2$ gain. This provable attenuation bound is what safety-critical and certification-driven embedded systems require: not a best-guess response, but a hard upper limit on how badly an unknown disturbance can degrade performance.

Mathematical Theory

Generalized Plant

The design is framed around the generalized plant

$$ x_{k+1} = A x_k + B_1 w_k + B_2 u_k, \quad z_k = C_1 x_k + D_{12} u_k $$

where $w \in \mathbb{R}^{n_w}$ is the exogenous disturbance, $u \in \mathbb{R}^{n_u}$ is the control input, and $z \in \mathbb{R}^{n_z}$ is the performance (error) output.

H∞ Performance Criterion

The controller objective is to find $u = -K x$ such that

$$ \sup_{w \neq 0} \frac{|z|_2}{|w|_2} < \gamma $$

where $\gamma &gt; 0$ is the prescribed attenuation level. Finding the smallest feasible $\gamma$ (the optimal $\gamma^*$) determines the best achievable robustness.

Game-Theoretic Discrete Algebraic Riccati Equation (GARE)

The H∞ state-feedback gain is derived from the discrete Riccati equation

$$ X = A^\top X A - A^\top X B \tilde{R}^{-1} B^\top X A + Q $$

with the augmented input matrix $B = [B_2 \mid B_1]$ and the indefinite weight matrix

$$ \tilde{R} = \begin{pmatrix} I_{n_u} & 0 \ 0 & -\gamma^2 I_{n_w} \end{pmatrix}. $$

The negative block encodes the adversarial role of the disturbance: the disturbance player maximizes while the control player minimizes. A positive-semidefinite stabilizing solution $X \geq 0$ exists if and only if $\gamma$ is above the optimal level $\gamma^*$.

Gain Extraction

From the Riccati solution $X$, the augmented gain is

$$ K_{\text{full}} = (\tilde{R} + B^\top X B)^{-1} B^\top X A. $$

Only the top $n_u$ rows — the control block — form the feedback gain $K$, and the closed-loop map is $A_{\text{cl}} = A - B_2 K$.

Feasibility Conditions

A given $\gamma$ is feasible when:

  1. The GARE has a positive-semidefinite solution $X \geq 0$.
  2. The disturbance block $-\gamma^2 I + B_1^\top X B_1 \prec 0$ is negative definite, confirming that the disturbance remains a genuine maximizer rather than a destabilizing force.

Bisection for Optimal $\gamma$

Neither condition holds for $\gamma < \gamma^$; both hold for $\gamma > \gamma^$. A standard bisection on $[\gamma_{\min}, \gamma_{\max}]$ converges to $\gamma^*$ at a linear rate.

Complexity Analysis

Phase Time Complexity Space Notes
Synthesize $O(\log((\gamma_{\max}-\gamma_{\min})/\epsilon) \cdot n^3)$ $O(n^2)$ Dominated by iterative DARE solves
ComputeControl $O(n_u \cdot n)$ $O(1)$ Single matrix-vector multiply

Step-by-Step Walkthrough

Consider a 2-state discrete plant with one disturbance and one control input.

  1. Bisect: choose $g = (\gamma_{\min} + \gamma_{\max}) / 2$; stack $B = [B_2 \mid B_1]$; build $\tilde{R} = \text{diag}(1, -g^2)$.
  2. Solve GARE: run the iterative DARE solver with indefinite $\tilde{R}$ until convergence.
  3. Check feasibility: verify $X_{ii} \geq 0$ for all $i$ and $-g^2 + (B_1^\top X B_1)_{ii} &lt; 0$.
  4. Update bisection: if feasible, tighten ($\gamma_{\max} \leftarrow g$); otherwise relax ($\gamma_{\min} \leftarrow g$).
  5. Finalize: at convergence, solve GARE at $\gamma_{\max}$, extract the control rows of $K_{\text{full}}$, and verify $A - B_2 K$ is Schur-stable (all eigenvalues inside the unit disk).

Pitfalls & Edge Cases

  • Ill-conditioned GARE near $\gamma^*$: the Riccati solution blows up as $\gamma \to \gamma^*$ from above. The bisection tolerance should not be driven below the float precision of the Riccati solver.
  • Indefinite $\tilde{R}$: the standard DARE assumes positive-definite $R$; using $\tilde{R}$ with a negative block is valid only when the full augmented pair $(A, B)$ is stabilizable and the game saddle-point condition holds. Infeasibility manifests as non-PSD $X$ or violated disturbance-block condition.
  • Disturbance block check: a numerically PSD $X$ does not guarantee feasibility; the disturbance block condition must also be verified explicitly.
  • Float precision: accumulated rounding in many DARE iterations can erode the convergence criterion; using the iterative formulation with a conservative tolerance (relative to 1e-3f) prevents premature acceptance of a diverged iterate.

Variants & Generalizations

  • H∞ output feedback (H∞ LQG): replaces the state $x$ with an observer-based estimate; requires a second (filter) Riccati equation to solve the full information-state problem.
  • Continuous-time H∞: replaces the DARE with the continuous algebraic Riccati equation; directly applicable to analog plants or zero-order-hold designs.
  • Mixed H₂/H∞: constrains the H∞ norm while minimizing the H₂ (LQG) cost — trades average and worst-case performance on a Pareto frontier.

Applications

  • Safety-critical motion control where actuator saturation or load shifts make average-case design insufficient.
  • Vibration suppression under unknown broadband disturbances.
  • Robust attitude control of spacecraft or UAVs subject to unmodeled flexible modes.
  • Robust stabilization of plants with parametric uncertainty encoded as bounded disturbances.

Connections to Other Algorithms

  • LQR ($\gamma \to \infty$): as the adversary weakens, the H∞ gain converges to the LQR gain for the same $(A, B_2, Q, I)$ weights. H∞ is the robust generalization of LQR.
  • DiscreteAlgebraicRiccatiEquation: the inner computational engine; H∞ passes an indefinite weight to it.
  • LQG / Kalman Filter: the stochastic average-case counterpart; H∞ and LQG bound opposite ends of the robustness-optimality trade-off.
  • Sliding Mode Control: a nonlinear alternative to H∞ that achieves robust disturbance rejection without solving a Riccati equation, at the cost of chattering and switching nonlinearity.
  • DurandKerner: used post-synthesis to verify that all eigenvalues of $A - B_2 K$ lie inside the unit disk.

References & Further Reading

  • J. Doyle, K. Glover, P. Khargonekar, B. Francis, "State-Space Solutions to Standard H₂ and H∞ Control Problems," IEEE Trans. Automatic Control, 34(8), pp. 831–847, 1989.
  • B. A. Francis, A Course in H∞ Control Theory, Lecture Notes in Control and Information Sciences, Springer, 1987.
  • K. Zhou, J. C. Doyle, K. Glover, Robust and Optimal Control, Prentice-Hall, 1996.