State Estimation with Secrecy against Eavesdroppers

Anastasios Tsiamis, Konstantinos Gatsis, George J. Pappas

TL;DR

This paper introduces a novel "control-theoretic" approach to secrecy in remote state estimation. By strategically and randomly withholding sensor data, a transmitter can exploit the mathematical "phase transition" of Kalman filters to ensure a legitimate user tracks an unstable system with bounded error while an eavesdropper’s estimation error grows to infinity.

What Problem Does This Paper Address?

In the Internet of Things (IoT), sensors often transmit critical data about physical systems (like power grids or industrial plants) over wireless channels. Because wireless signals are broadcast, eavesdroppers can easily intercept this data.

Standard cryptographic solutions (like AES) are often "application-agnostic"—they don't account for the underlying physics of the system being monitored. This paper asks: Can we use the dynamics of the physical system itself to achieve secrecy? Specifically, the authors tackle the challenge of "Perfect Expected Secrecy," where the goal is to keep the legitimate user's estimation error low while forcing the eavesdropper's error to become infinitely large, even if the eavesdropper knows exactly how the system works.

Methods

The authors model the problem using a linear dynamical system and an Intermittent Kalman Filter (IKF) framework. The key components are:

  1. The Secrecy Mechanism: Instead of encrypting the data, the sensor flips a biased coin at each time step to decide whether to send the measurement at all.
  2. Channel Disparity: They assume the legitimate user has a better reception rate ($p_1$) than the eavesdropper's interception rate ($p_2$).
  3. Exploiting the Phase Transition: IKF theory shows there is a "critical probability" ($p_c$) for packet reception. If you receive packets more often than $p_c$, your error stays bounded. If you fall below $p_c$, your error explodes. The authors' mechanism tunes the transmission probability ($p$) such that the user’s effective rate ($p \cdot p_1$) is above $p_c$, but the eavesdropper’s rate ($p \cdot p_2$) falls below it.
  4. Optimization: Since finding the exact error of an IKF is computationally hard, the authors use analytical upper and lower bounds to formulate a bisection-based optimization algorithm. This allows a system designer to minimize user error while guaranteeing a specific "level of pain" (minimum error) for the eavesdropper.

Datasets

This is a theoretical and simulation-based paper; it does not use real-world datasets. The authors evaluate their approach using:

  • Analytical Scalar Examples: To derive closed-form solutions for the optimal transmission probability.
  • Synthetic Second-Order Systems: Numerical simulations of unstable linear systems to demonstrate how estimation errors evolve over time under their secrecy mechanism.

Key Insights and Ideas

  • Instability as a Shield: Usually, unstable system dynamics are a problem to be solved. Here, they are a feature. The faster a system "blows up," the more sensitive the estimation becomes to missing packets, making it easier to "blind" the eavesdropper with fewer withheld measurements.
  • The $p_1 > p_2$ Requirement: The fundamental limit for this mechanism is that the legitimate channel must be strictly better than the eavesdropper's channel. If the eavesdropper has a better "ear" than the user, this specific random-withholding strategy cannot achieve perfect secrecy.
  • Expectation vs. Reality: The authors define secrecy in terms of expected error. They note an important caveat: even if the eavesdropper's average error is infinite, there will be rare moments where they successfully intercept a string of packets and briefly gain an accurate estimate.

Key Results

  • Theorem 1 (Sufficiency and Necessity): The authors prove that perfect expected secrecy is achievable if and only if $p_1 > p_2$.
  • Optimal Tuning: They provide a computationally efficient bisection algorithm (Algorithm 1) to find the optimal transmission rate $p^*$ that balances user utility and secrecy.
  • Numerical Proof of Concept: In a simulated second-order system, the secrecy mechanism resulted in a time-averaged error of 6.2 for the user, while the eavesdropper suffered a massive average error of 2922.7. Without the mechanism, both had comparable errors (1.9 vs 4.6).

Limitations and Future Work

  • Knowledge of the Adversary: The current model assumes the transmitter knows the eavesdropper's interception rate ($p_2$). In practice, this is difficult to estimate.
  • I.I.D. Assumptions: The paper assumes packet drops are independent and identically distributed. Real-world wireless channels often have "bursty" losses (Markovian fading), which would complicate the phase transition analysis.
  • Lack of Feedback: The mechanism currently doesn't use acknowledgments (ACKs) from the user. The authors suggest that incorporating ACKs could allow for even more sophisticated, non-random withholding strategies.
  • Sample Path Secrecy: As noted in the insights, "perfect secrecy in expectation" doesn't mean the eavesdropper is always wrong. Future work could look into stronger definitions of secrecy that bound the error at every single time step.

Summary updated

Abstract

We study the problem of remote state estimation, in the presence of an eavesdropper. An authorized user estimates the state of a linear plant, based on the data received from a sensor, while the data may also be intercepted by the eavesdropper. To maintain confidentiality with respect to state, we introduce a novel control-theoretic definition of perfect secrecy requiring that the user’s expected error remains bounded while the eavesdropper’s expected error grows unbounded. We propose a secrecy mechanism which guarantees perfect secrecy by randomly withholding sensor information, under the condition that the user’s packet reception rate is larger than the eavesdropper’s interception rate. Given this mechanism, we also explore the tradeoff between user’s utility and confidentiality with respect to the eavesdropper, via an optimization problem. Finally, some examples are studied to provide insights about this tradeoff.
BibTeX
@inproceedings{tsiamis2017state,
  author = {Anastasios Tsiamis and Konstantinos Gatsis and George J. Pappas},
  title = {{State Estimation with Secrecy against Eavesdroppers}},
  booktitle = {IFAC-PapersOnLine},
  year = {2017},
  doi = {10.1016/j.ifacol.2017.08.1563},
}

Loading related content...

Figure