Coding Schemes for Securing Cyber-Physical Systems Against Stealthy Data Injection Attacks

Fei Miao, Quanyan Zhu, Miroslav Pajic, George J. Pappas

TL;DR

This paper introduces a lightweight, low-cost coding scheme to protect Cyber-Physical Systems (CPS) against "stealthy" false data injection attacks. By multiplying sensor outputs with a secret, periodically changing matrix, the system forces sophisticated attacks to reveal themselves to standard fault detectors without requiring the high computational overhead of traditional cryptographic encryption.

What Problem Does This Paper Address?

Cyber-physical systems—such as power grids, autonomous vehicles, and medical devices—rely on sensor data to estimate their physical state and make control decisions. An intelligent attacker who knows the mathematical model of the system can craft "stealthy" data injection attacks on the communication channels.

These attacks are dangerous because they are mathematically designed to slowly drive the system's state estimation error to infinity while keeping the "residue" (the difference between expected and actual sensor readings) incredibly small. As a result, standard statistical fault detectors (like $\chi^2$ detectors used alongside Kalman filters) never trigger an alarm. While encrypting all sensor data could prevent this, traditional bit-level encryption is often too computationally expensive and introduces too much communication overhead for resource-constrained embedded sensors.

Methods

The authors propose a linear algebraic defense mechanism that acts as a lightweight alternative to encryption:

  • Sensor Output Coding: Before transmitting data, sensors multiply their measurements by a secret, invertible "coding matrix" ($\Sigma$). The central estimator/controller knows this matrix and decodes the data upon receipt.
  • Breaking the Stealth: Stealthy attacks rely on precise alignment with the system's physical dynamics (specifically, its unstable eigenvectors). The coding matrix effectively rotates the sensor outputs. If an attacker injects a stealthy signal designed for the uncoded system, the secret rotation pushes their signal out of the "stealthy subspace." This causes the fault detector's residue to blow up, triggering the alarm.
  • Matrix Generation (Algorithm 1): The authors provide an algorithm to generate feasible coding matrices using Givens rotations, ensuring the attacker's signal is always misaligned.
  • Time-Varying Defense (Algorithm 3): An advanced attacker might try to eavesdrop on the coded data to reverse-engineer $\Sigma$. To counter this, the authors model the attacker's learning process (Algorithm 2) and propose a heuristic for the defender to periodically update $\Sigma$ before the attacker can gather enough data to solve the underlying noisy bilinear equations.

Datasets

Not discussed in the paper. The authors evaluate their theoretical claims using mathematical proofs and numerical simulations of a 2-dimensional Linear Time-Invariant (LTI) system, rather than empirical datasets.

Key Insights and Ideas

  • Active monitors aren't enough: Simply adding random noise or authentication signals to the control input (actuators) does not help detect pure sensor data injection attacks. A different approach is required.
  • The geometry of stealth: Stealthy attacks work by hiding in specific mathematical blind spots of the system. A simple rotation matrix ($\Sigma$) is enough to shift the attacker's injected vector out of these blind spots.
  • Security through moving targets: Even if an attacker knows the system's physical dynamics, keeping the coding matrix secret and changing it periodically creates a moving target. Learning the matrix requires solving noisy bilinear equations, which is computationally difficult and time-consuming.
  • Drastically lower overhead: The proposed matrix multiplication requires only $O(n^3 + p^3)$ operations (where $n$ is the number of states and $p$ is the number of sensors). This is vastly cheaper than computing highly complex non-linear cryptographic primitives, making it ideal for embedded architectures.

Key Results

  • Theoretical Guarantee: The authors mathematically prove that if the coding matrix $\Sigma$ misaligns the attacker's signal from the system's unstable eigenvectors, the estimation residue will approach infinity, guaranteeing detection.
  • Simulation Performance: In a simulated 2D LTI system, an attack that remained completely undetected on an uncoded system triggered the alarm in just 12 seconds once the coding matrix was applied.
  • Resilience to Reverse-Engineering: Even when an attacker was allowed to eavesdrop for 20 time steps to estimate the secret coding matrix, their newly crafted attack still triggered the system's alarm at 50 seconds. This proves the system has ample time to rotate the matrix (e.g., every 5 seconds) to stay permanently ahead of the attacker.

Limitations and Future Work

  • Limitations: The scheme assumes that the sensor measurements are not corrupted before the coding is applied (i.e., the attack must happen in the communication channel, not via physical tampering of the sensor itself). It also assumes perfectly synchronized clocks between sensors and controllers to update the matrices simultaneously.
  • Future Work: The authors plan to explore coding schemes for systems with structural constraints. For example, in systems where sensors are sparse, the coding matrix might be forced to have zeros in specific places. They also suggest looking into distributed coding schemes where multiple different matrices are applied across various parts of a large-scale system.

Summary updated

Abstract

This paper considers a method of coding the sensor outputs in order to detect stealthy false data injection attacks. An intelligent attacker can design a sequence of data injection to sensors and actuators that pass the state estimator and statistical fault detector, based on knowledge of the system parameters. To stay undetected, the injected data should increase the state estimation errors while keep the estimation residues small. We employ a coding matrix to change the original sensor outputs to increase the estimation residues under intelligent data injection attacks. This is a low-cost method compared with encryption schemes over all sensor measurements in communication networks. We show the conditions of a feasible coding matrix under the assumption that the attacker does not have knowledge of the exact coding matrix. An algorithm is developed to compute a feasible coding matrix, and, we show that in general, multiple feasible coding matrices exist. To defend against attackers who estimates the coding matrix via sensor and actuator measurements, time-varying coding matrices are designed according to the detection requirements. A heuristic algorithm to decide the time length of updating a coding matrix is then proposed.
BibTeX
@article{miao2017coding,
  author = {Fei Miao and Quanyan Zhu and Miroslav Pajic and George J. Pappas},
  title = {{Coding Schemes for Securing Cyber-Physical Systems Against Stealthy Data Injection Attacks}},
  journal = {IEEE Trans. Control. Netw. Syst.},
  year = {2017},
  url = {https://openreview.net/forum?id=_Zs2ktWXuGS},
}

Loading related content...

Figure