Optimal Privacy-Aware Co-Design of Quantizer and Controller in Networked Control Systems

Researchers Chuanghong Weng and Ehsan Nekouei have introduced a novel framework for securing networked control systems (NCS) against privacy attacks. In their paper 'Optimal Privacy-Aware Co-Design of Quantizer and Controller in Networked Control Systems', they address a critical vulnerability: when a system (like a smart building's HVAC) sends sensor measurements to a remote controller, an adversary can infer private operational inputs from the transmitted data. The core innovation is treating privacy as a co-design problem, where the data-compressing 'quantizer' and the system 'controller' are optimized together, not separately.

Their method formulates this as a stochastic control problem with a mutual information regularizer, which mathematically quantifies privacy leakage. Using dynamic programming, they derived coupled Bellman equations, revealing that the optimal controller is deterministic, while the optimal quantizer actively manipulates the adversary's beliefs in a closed loop. To make this tractable for complex systems, they jointly parameterized the quantizer and controller and trained them using policy gradient methods—a reinforcement learning technique. A binary classification model was used to approximate the hard-to-calculate mutual information loss.

The framework was successfully validated through numerical experiments on a building control system, a common and sensitive NCS application. This demonstrates a practical path to deploying AI-driven control in environments like smart grids, industrial IoT, or autonomous vehicles without exposing proprietary or sensitive state information. The work, available on arXiv under identifier 2604.08860, represents a significant step toward trustworthy autonomous systems where performance and privacy are guaranteed by design, not added as an afterthought.