Negative Position Feedback Controllers

• Negative position feedback controllers are robust control techniques that exploit negative imaginary (NI) and strictly negative imaginary (SNI) properties for stabilizing resonant flexible structures.
• They achieve high disturbance rejection and robustness to unmodelled dynamics by using colocated force actuators and position sensors with a DC-gain stability criterion.
• Controller synthesis employs methods like integral resonant controllers, LMIs, and Lyapunov-based analysis to ensure stable performance in applications including robotics and UAV control.

Negative position feedback controllers are a class of robust control techniques designed for mechanical systems—especially highly resonant flexible structures—with colocated force actuators and position sensors. These controllers leverage the negative imaginary (NI) and strictly negative imaginary (SNI) properties of the plant and controller transfer functions to achieve high disturbance rejection and inherent robustness to unmodelled dynamics, spillover effects, and parameter uncertainties. They are systematically characterized and designed via internal stability criteria rooted in transfer function matrix inequalities, Laurent expansions, and Lyapunov-based dissipativity concepts (Mabrok et al., 2013, Petersen et al., 2014, Ghallab et al., 2016, Ghallab et al., 2022, Dannatt et al., 2023).

1. Formal Definition and Key Properties

A system is negative imaginary (NI) if its transfer function $G(s)$ (SISO or MIMO, real-rational, proper) satisfies:

• $G(s)$ has no poles in $\operatorname{Re}[s]>0$.
• For all $\omega>0$ for which $G(j\omega)$ is finite, $j\left[G(j\omega)-G(j\omega)^*\right]\geq 0$ (Hermitian positive semidefinite).
• Imaginary axis poles $j\omega_0\neq 0$ are simple with Hermitian, positive semidefinite residue.
• At $s=0$ a double integrator is allowed, but no higher: $\lim_{s\to 0}s^3G(s)=0$, and Laurent coefficients $G_2 = \lim_{s\to 0} s^2 G(s)$, $G_1 = \lim_{s\to 0} s[G(s)-G_2/s^2]$, $G_0 = \lim_{s\to 0} [G(s)-G_2/s^2-G_1/s]$ are appropriately positive semidefinite (Mabrok et al., 2013).

Strictly negative imaginary (SNI) systems strengthen the frequency-domain inequality to $j\left[G(j\omega)-G(j\omega)^*\right]>0$ for all $\omega>0$, and their poles are restricted to $\operatorname{Re}[s]<0$.

The NI property naturally emerges from force-to-position transfer functions of flexible structures with colocated sensors and actuators, as their modal decomposition exhibits a phase lag of $[-\pi,0]$ and Nyquist plot lying below the real axis (Petersen et al., 2014).

2. Feedback Architecture and Robust Stability Criteria

Negative position feedback typically refers to the architecture where the system input (force) is generated by feeding back the measured position through a dynamic compensator, often realized:

$u(s) = -C(s) y(s)$

However, stability analysis is carried out in terms of positive feedback interconnections of the NI plant and SNI controller:

$u_1 = y_2,\quad u_2 = y_1$

for plants $G(s)$ (NI) and controllers $C(s)$ (SNI). Robust internal stability is guaranteed if the DC-gain inequality holds:

$\lambda_{\max}(G(0) C(0)) < 1$

This single scalar or matrix criterion is necessary and sufficient for internal stability of the full closed loop in both SISO and MIMO cases, regardless of order or the presence of free-body (integrator) dynamics (Mabrok et al., 2013, Petersen et al., 2014, Ghallab et al., 2016).

When the plant includes free body modes, stability requires a refined analysis using Laurent expansions and the basis matrix $F$, with conditions on $F^\top \bar{G}(0) F$ and projected matrices $N_f$ for integrator handling (Mabrok et al., 2013).

3. Controller Synthesis Methodologies

Negative position feedback controllers are synthesized as follows:

• Colocated pairing: The force actuator and position sensor ensure the plant is NI or generalized NI.
• SNI controller selection: Popular choices include integral resonant controllers (IRCs):

$$C(s) = (sI + \Gamma\Phi)^{-1}\Gamma - \Delta,\quad \Gamma > 0,\, \Phi > 0,\, \Delta = \Delta^\top$$

which are SNI with easily tunable DC gain and bandwidth (Mabrok et al., 2013, Petersen et al., 2014).

• Tuning: Parameters are tuned to satisfy robust stability, typically through finite-dimensional LMIs involving the controller DC gain and plant Laurent coefficients.
• State-feedback variant: For flexible structures, PID augmentation combined with SNI state-feedback synthesizes closed-loop SNI systems with prescribed stability degree, relying on Riccati-type equations and canonical pole placement (Dannatt et al., 2023).
• Adaptive schemes: Fuzzy Q-learning can be employed to adapt SNI compensator gains online for uncertain, nonlinear plants, preserving robust NI/SNI feedback properties and maintaining stability via the DC gain criterion (Tran et al., 2022).
• General nonlinear systems: The dissipativity-based NNI approach certifies stability for negative position feedback by ensuring supply rate inequalities and a sector-bound condition in the steady-state (Ghallab et al., 2022).

4. Lyapunov and Dissipativity-Based Stability Analysis

Lyapunov theory provides a rigorous stability foundation for NI systems and their feedback interconnections:

• Candidate Lyapunov functions of the form $V(x_1,x_2)=V_1(x_1)+V_2(x_2)-2y_1^\top y_2$ are constructed, where $V_1,V_2$ are storage functions for the plant and controller subsystems (Ghallab et al., 2016).
• The time derivative $\dot{V}$ is shown negative semidefinite, and the DC-gain condition ensures $V$ is positive definite. Stability proofs are constructive and generalize to nonlinear dissipative systems (NNI) via supply rates $\dot{y}^\top u$ (Ghallab et al., 2022).
• Algebraic Riccati equation (ARE) solutions for state- and output-feedback controllers provide sufficient conditions for closed-loop NI-ness and internal stability. Existence of suitable positive semidefinite solutions confirms controller design (Dannatt et al., 2019, Dannatt et al., 2023).

5. Applications and Case Studies

Negative position feedback controllers demonstrate robust modal damping and performance across multiple domains:

Application	Plant Model	Controller Type
Flexible robotic arms	Beam with double integrator (free-body), multimode	IRC, first-order SNI
UAV (quadrotor) control	NI via feedback-linearized dynamics	SNI, fuzzy QL-adaptive
Large piezo-actuated platforms	Multi-modal, collocated force-position	NI positive-feedback, MIMO IRC
Nonlinear mechanical systems	Mass-spring-damper with nonlinear damping	SNI controller, dissipativity framework

In practice, these controllers yield high robustness to parameter uncertainty, strong disturbance rejection, and insensitivity to unmodelled spillover dynamics or higher modes. Extensive simulations and experimental studies confirm faster settling times, smoother control actions, and superior RMSE metrics compared to traditional PID or fixed-gain controllers (Tran et al., 2022, Mabrok et al., 2013, Petersen et al., 2014).

6. Extensions to Nonlinear and High-Order Systems

The negative position feedback framework extends to nonlinear systems via the nonlinear NI (NNI) property, defined by the existence of a storage function $V(x)$ satisfying a supply rate inequality:

$\dot{V}(x(t)) \leq \dot{y}(t)^T u(t)$

Stability in positive feedback interconnection is certified via a sector-bound condition and Lyapunov-type theorem, even for mass–spring–damper systems with nonlinearities and free motion (Ghallab et al., 2022). For plants with arbitrary relative degree or infinite-dimensional modal expansions, controller synthesis and stability checks strictly rely on frequency- and time-domain criteria that do not necessitate reduction or high-order pole placement.

7. Synthesis Recipes and Practical Considerations

For practical synthesis, the methodology reduces to:

• Positive DC-gain tuning: Ensure $\lambda_{\max}(P(0)C(0))<1$ (MIMO) or $P(0)C(0)<1$ (SISO).
• Strictly proper controller: Controller must not introduce RHP poles or non-decaying modes.
• Integrators and free-body motion: Laurent expansion enables handling up to double integrators with refined matrix inequalities for stability (Mabrok et al., 2013).
• Numerical implementation: Recipes involve the computation of transfer matrix Laurent coefficients, finite LMIs, and Riccati equations. Adaptive strategies employ rule-based fuzzy Q-learning to maintain the SNI property under time-varying uncertainties (Tran et al., 2022).

A plausible implication is that the NI/SNI structured approach provides a unified and order-independent robust control paradigm across both linear and nonlinear domains for mechanical systems with collocated actuation and sensing.

---

References (Mabrok et al., 2013, Petersen et al., 2014, Ghallab et al., 2016, Ghallab et al., 2022, Tran et al., 2022, Dannatt et al., 2023, Dannatt et al., 2019)