FHECAP: An Encrypted Control System With Piecewise Continuous Actuation

Song Bian, Yunhao Fu, Dongxu Zhao, Haowen Pan, Yuexiang Jin, Jiayue Sun, Hui Qiao, Zhenyu Guan

TL;DR

This paper introduces FHECAP, a novel framework for encrypted control systems that can handle non-linear actuator behaviors, a common challenge in real-world applications. By employing fully homomorphic encryption (FHE) and a clever piecewise evaluation strategy, FHECAP achieves significant speedups while maintaining system accuracy and security, making secure cloud-based control more practical.

What Problem Does This Paper Address?

Modern control systems, especially those operating in critical environments like spacecraft, often rely on cloud-based controllers for computational power. However, transmitting sensitive control data to the cloud raises significant security and privacy concerns, leading to the exploration of homomorphic encryption (HE). While existing HE-based control systems can handle linear operations, they struggle with non-linearities, particularly those introduced by actuators (e.g., saturation, dead zones). These non-linearities are crucial for system stability and protecting hardware, but their homomorphic evaluation is computationally expensive and often imprecise with current HE schemes. This gap limits the applicability of secure encrypted control to a broader range of real-world systems.

Methods

FHECAP tackles this challenge by:

  • Piecewise Function Evaluation: The core innovation is breaking down complex non-linear actuator functions into smaller, manageable piecewise segments. Each segment is then evaluated using specialized HE primitives. This approach avoids the need for high-degree polynomial approximations (which are computationally intensive and prone to errors in HE) or imprecise look-up tables.
  • Cross-Scheme HE Infrastructure: FHECAP leverages multiple HE schemes, specifically LWE (for TFHE) and RLWE (for CKKS), to optimize different parts of the computation. LWE is used for non-linear functions, while RLWE is employed for linear transformations.
  • Homomorphic Operators: The framework utilizes a suite of homomorphic operators, including:
  • Homomorphic Linear Transformation: Efficiently computes matrix-vector multiplication over encrypted data, using coefficient-domain encoding for speed.
  • Homomorphic Ciphertext Conversion: Bridges the gap between RLWE and LWE ciphertexts to enable subsequent non-linear function evaluations.
  • Homomorphic Non-Linear Actuation: Implements the piecewise evaluation using operators like HomComp (homomorphic comparison), CB (ciphertext-to-ciphertext conversion), and CMUX (homomorphic selector).
  • Automated Noise Analysis: The paper provides a rigorous theoretical analysis of noise growth throughout the FHECAP protocol, crucial for ensuring system stability and determining appropriate HE parameters.

Datasets

Not discussed in the paper. The evaluation uses simulated control system models and micro-benchmarks.

Key Insights and Ideas

  • Piecewise is Powerful: Decomposing non-linear functions into smaller, simpler pieces is a highly effective strategy for achieving accurate and efficient homomorphic evaluation, overcoming limitations of traditional polynomial approximation or LUT-based methods in HE.
  • Cross-Scheme Synergy: Combining different HE schemes (LWE and RLWE) for distinct computational tasks (non-linear vs. linear) offers a significant advantage in terms of both performance and expressiveness.
  • Actuator Non-Linearities are Key: The paper highlights that actuator non-linearities are not just an afterthought but a critical component of many control systems, and addressing them securely is essential for practical encrypted control.
  • Noise Management is Crucial: The detailed noise analysis demonstrates that while HE introduces noise, careful parameter selection and protocol design can keep it within acceptable bounds for stable control.

Key Results

  • Speedup: FHECAP achieves a 4x to 1000x reduction in computational latency compared to existing state-of-the-art encrypted controllers for linear transformations.
  • Non-Linear Function Evaluation: The framework can efficiently evaluate a wide range of complex non-linear functions (saturation, dead-zone, relay) over encrypted data.
  • Spacecraft Rendezvous: In a simulated spacecraft rendezvous scenario, FHECAP demonstrated real-time control with negligible accuracy loss, showcasing its practical applicability. The errors introduced by encrypted control were found to be extremely small and did not accumulate over time.
  • Performance Comparison: For non-linear function evaluation, FHECAP is at least 1.4x faster than PA-based methods and 1.8x faster than LUT-based methods.

Limitations and Future Work

  • Parameter Selection Complexity: While the paper provides a framework for parameter selection, finding the optimal balance between security, performance, and accuracy can still be complex.
  • Generalization of Non-Linear Functions: The paper focuses on specific types of non-linearities. Extending the framework to handle even more complex or arbitrary non-linear functions might require further research.
  • Real-World Hardware Implementation: The current evaluation is based on simulations. Implementing and testing FHECAP on actual hardware would be a valuable next step.
  • Interaction Interval: The study on spacecraft rendezvous indicated that the interaction interval plays a critical role in system self-correction. Further investigation into optimizing this interval for encrypted control systems could be beneficial.

Summary updated

BibTeX
@article{bian2025fhecap,
  author = {Song Bian and Yunhao Fu and Dongxu Zhao and Haowen Pan and Yuexiang Jin and Jiayue Sun and Hui Qiao and Zhenyu Guan},
  title = {{FHECAP: An Encrypted Control System With Piecewise Continuous Actuation}},
  journal = {IACR Cryptol. ePrint Arch.},
  year = {2025},
}

Loading related content...

Figure