We are pleased to announce that the paper “A Learning With Errors based Encryption Scheme for Dynamic Controllers that Discloses Residue Signal for Anomaly Detection,” authored by Yeongjun Jang, Joowon Lee, Junsoo Kim, Takashi Tanaka, and Hyungbo Shim, has been accepted for publication in IEEE Transactions on Control of Network Systems. This study was carried out in close collaboration with Professor Takashi Tanaka at Purdue University and Professor Junsoo Kim at Seoul National University of Science and Technology.
Encrypted control systems protect sensitive signals and parameters, but encryption can also hide the residue signals required to detect attacks and faults. This paper introduces a Learning With Errors (LWE)-based homomorphic encryption scheme that automatically discloses only the residue signal to a network-side anomaly detector while keeping the remaining controller signals private. The method exploits the zero-dynamics of a finite-field encrypted system so that the masking term of the encrypted residue remains identically zero. The accompanying security analysis shows that the proposed scheme reveals no information beyond the intended residue signal.
The framework also allows dynamic controllers with non-integer state matrices to operate on encrypted data without re-encryption. By feeding the disclosed residue directly back into the encrypted controller, the design eliminates an extra communication link to the actuator and reduces computational and communication overhead, while supporting operation over an infinite time horizon.
Numerical simulations on a two-mass-spring system, using a CUSUM anomaly detector and encryption parameters designed for 128-bit security, showed that the encrypted controller achieved performance comparable to its unencrypted counterpart and successfully detected an injected sensor attack.
This work takes an important step toward practical control systems that preserve data confidentiality while retaining real-time anomaly detection capabilities.