An H2-norm approach to performance analysis of networked control systems under multiplicative routing transformations

Ruslan Seifullaev, André M. H. Teixeira

TL;DR

This paper introduces a mathematical framework to analyze a sneaky class of "multiplicative" cyber-attacks on networked control systems (NCS). By using $H_2$-norms, the authors quantify how an adversary can reroute or scale sensor data to degrade system performance while remaining invisible to standard energy-based anomaly detectors.

What Problem Does This Paper Address?

Most research on secure control systems focuses on additive attacks, where an adversary injects a malicious signal (like noise or a false data offset) into the system. Because these attacks add external energy, they are often caught by anomaly detectors that monitor "residuals"—the difference between expected and actual system behavior.

However, this paper tackles multiplicative (or routing) attacks. In this scenario, the attacker doesn't inject new signals; instead, they manipulate the structure of the network. They might alter routing tables, change sensor fusion weights, or permute measurement channels. These attacks are "energy-neutral" because they don't add external signals, making them exceptionally difficult to detect. The core challenge is understanding how these structural changes impact system stability and performance, and how an attacker might find the "sweet spot" that maximizes damage while minimizing the footprint left for detectors.

Methods

The authors propose an $H_2$-norm framework to evaluate the trade-off between performance degradation and detectability.

  1. Modeling the Attack: They model the attack as a transformation matrix $R$ that modifies the measurement vector $y$. The resulting closed-loop system remains linear but has dynamics that depend explicitly on this unknown matrix $R$.
  2. The Performance Ratio: They formulate an optimization problem that seeks to maximize the ratio between the steady-state energy of the performance output (the error we want to keep small) and the residual output (the signal used for detection).
  3. LMI Formulations: To make this computationally feasible, they derive equivalent Linear Matrix Inequality (LMI) conditions. This allows engineers to efficiently calculate the worst-case performance ratio for a given system.
  4. Analytical Bounds: They establish theoretical upper bounds on how much a system can be degraded under both small and general perturbations of the routing matrix, providing a "safety envelope" for system designers.

Datasets

Not discussed in the paper. As a theoretical control theory paper, it does not use external real-world datasets. Instead, it validates the framework using a numerical simulation of a 3-state linear system with specific observer and controller gains.

Key Insights and Ideas

  • Structural Alignment: The most effective (and stealthy) attacks occur when the routing transformation "aligns" with the system's internal modes. Specifically, if an attack excites a slow or marginally stable mode that is visible to the performance output but hidden from the residual output, the attacker achieves maximum impact.
  • Energy-Neutral Sabotage: Multiplicative attacks can be viewed as a "preparatory stage." By reshaping the feedback pathways without adding energy, an attacker can make the system inherently more fragile before launching a more aggressive, active attack.
  • The $H_2$ Advantage: While $H_\infty$ metrics are common for additive attacks, the authors argue that $H_2$-norms are physically more meaningful for multiplicative faults because they capture how steady-state energy propagates under the stochastic excitation (noise) naturally present in the system.

Key Results

  • Exact Representation: The paper proves that the ratio of performance energy to residual energy can be exactly represented using $H_2$-norms.
  • Numerical Validation: In a simulated example, the authors showed that an attack-free system had a performance-to-residual ratio of 2.46. By choosing a specific diagonal routing matrix (reducing signal power by about 30%), an attacker could increase that ratio to 5.35.
  • Stealthy Optimization: When constrained to stay below a specific detection threshold (e.g., residual energy < 2), the adversary was still able to achieve a performance ratio of 4.65, demonstrating that significant degradation is possible even under strict stealth requirements.

Limitations and Future Work

  • Linearity: The current framework is restricted to linear time-invariant (LTI) systems. Real-world systems with non-linearities or time-varying dynamics would require more complex analysis.
  • Stability Assumption: The analysis assumes the attacker wants to keep the system stable to avoid immediate detection. It doesn't fully explore the transition point where a multiplicative attack intentionally pushes a system into instability.
  • Mitigation Strategies: While the paper focuses on analysis, the authors suggest that future work should look at synthesis—designing controllers and observers specifically to be robust against these routing manipulations, perhaps by using "private excitation signals" to make the residuals harder to cancel out.

Summary updated

BibTeX
@inproceedings{seifullaev2026h,
  author = {Ruslan Seifullaev and André M. H. Teixeira},
  title = {{An H2-norm approach to performance analysis of networked control systems under multiplicative routing transformations}},
  booktitle = {ECC},
  year = {2026},
}
Figure