CCSS: Secure Dual-Function Radar Communication Systems Assisted by Intelligent Reflecting Surfaces

As next-generation wireless systems strive to incorporate sensing capabilities alongside communication functionalities, the efficient allocation of spectrum between radar and communication functions has become a prominent focus. One promising solution is the utilization of dual-function radar-communication (DFRC) systems. These systems leverage a single device and a single waveform to enable simultaneous sensing and communication, resulting in improved spectrum utilization, hardware efficiency, and power savings.

To further enhance the performance of DFRC systems in challenging channel conditions, an emerging technology called Intelligent Reflecting Surfaces (IRS) can be employed. IRS involves a planar array of numerous low-cost, real-time configurable elements that intelligently manipulate the transmitted waveform, creating a smart propagation environment. While DFRC systems excel at achieving high performance in both sensing and communication, they are vulnerable to potential eavesdroppers due to the embedded communication information within the probing waveform.

This project aims to develop novel IRS-aided DFRC system designs that deliver reliable, high-rate information to the intended communication receiver while minimizing the information accessible to eavesdroppers. Successful implementation of secure DFRC systems will have wide-ranging benefits across applications such as autonomous driving vehicles, unmanned aerial vehicles, surveillance, search and rescue operations, and advanced manufacturing processes involving networked robots.

This project addresses optimal IRS-aided DFRC system design and encompasses two thrusts. Thrust 1 considers design in dynamic environments, such as urban communication environments. A natural framework for dynamically optimizing the system parameters in that context is Deep Reinforcement Learning (RL), since it is adaptive, data-driven and does not require pre-annotated data.

In Thrust 1, a novel and principled deep RL framework will be developed that incorporates domain knowledge and ensures convergence under gradient descent dynamics. An off-policy actor-critic approach will be considered, and the role of the Neural Tangent Kernel (NTK) of the critic network to the training stability of the deep RL algorithm and its performance to unseen or rarely experienced events will be investigated.

Intelligent eavesdroppers will be addressed by formulating the performance optimization problem as a multi-agent reinforcement learning problem that considers both cooperation and competition. Thrust 2 considers DFRC systems transmitting Orthogonal Frequency Division Multiplexing (OFDM) waveforms and introduces the novel concept of Directional Modulation via Time-Modulated IRS (TM-IRS), which allows for more flexible secure system design.

With TM-IRS, each element of the IRS periodically turns ON/OFF across OFDM symbols. By carefully designing the periodic activation pattern as well as the IRS parameters and the transmit antenna weights, the signal can be delivered intact in a desired direction while appearing scrambled in all other directions. TM-IRS offers a large number of degrees of freedom for system design, thus enhancing the secure operation of DFRC systems.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.