MetaSensing: Intelligent Metasurface Assisted RF 3D Sensing by Deep Reinforcement Learning

Jingzhi Hu, Hongliang Zhang, Kaigui Bian, Marco Di Renzo, Zhu Han, Lingyang Song

Abstract

Using RF signals for wireless sensing has gained increasing attention. However, due to the unwanted multi-path fading in uncontrollable radio environments, the accuracy of RF sensing is limited. Instead of passively adapting to the environment, in this paper, we consider the scenario where an intelligent metasurface is deployed for sensing the existence and locations of 3D objects. By programming its beamformer patterns, the metasurface can provide desirable propagation properties. However, achieving a high sensing accuracy is challenging, since it requires the joint optimization of the beamformer patterns and mapping of the received signals to the sensed outcome. To tackle this challenge, we formulate an optimization problem for minimizing the cross-entropy loss of the sensing outcome, and propose a deep reinforcement learning algorithm to jointly compute the optimal beamformer patterns and the mapping of the received signals. Simulation results verify the effectiveness of the proposed algorithm and show how the sizes of the metasurface and the target space influence the sensing accuracy.

Introduction

Traditional radio frequency (RF) sensing systems are inherently passive. They rely on signals transmitted from a source, which then interact with the environment through reflection, diffraction, and scattering before reaching a receiver. In complex indoor environments, multipath fading and unpredictable interference often obscure the target object, leading to degraded sensing performance.

To overcome these limitations, researchers from Peking University have proposed MetaSensing, a framework that transforms the radio environment from a passive medium into an active, programmable component of the sensing system. By deploying an Intelligent Metasurface, the system can actively shape the propagation environment to create favorable paths for 3D object detection.

System model of MetaSensing featuring a metasurface, transceiver, and target space.

Methodology

The MetaSensing framework involves the joint optimization of metasurface beamforming patterns and the signal processing pipeline. This is modeled as a Markov Decision Process (MDP) and solved using a custom Deep Reinforcement Learning (DRL) approach.

The PRPG Algorithm

The core innovation is the Progressing Reward Policy Gradient (PRPG) algorithm, which iteratively trains two interconnected neural networks:

  • Policy Network: Determines the sequence of metasurface configurations. It utilizes a Symmetric MLP architecture to exploit the physical fact that the order of beamforming patterns does not change the sensing outcome, reducing parameter complexity.
  • Sensing Network: Maps the received signal differences to the probability of an object existing in specific 3D grids. It includes a model-aided decoder that uses physical domain knowledge to provide a "rough draft" of the environment, which the MLP then refines.

Operational Phases

The sensing process is divided into four distinct phases: synchronization, calibration, data collection, and data processing. During the data collection phase, the metasurface cycles through $K$ configurations ($c_1$ to $c_K$) while the receiver captures the resulting signal variations.

Timing diagram of the MetaSensing operational phases including data collection and processing.

Key Results and Insights

The proposed system was evaluated using high-fidelity electromagnetic simulations, assuming a 3.198 GHz signal and a discretized 3D target space.

  • Superior Accuracy: The PRPG algorithm achieved significantly lower cross-entropy loss compared to benchmarks like random metasurface patterns or systems without the model-aided decoder. The system converges rapidly, with results closely matching ground-truth object shapes after roughly $10^4$ training epochs.
  • Beating MIMO: A MetaSensing system with a single-antenna transceiver outperformed traditional MIMO (Multiple-Input Multiple-Output) systems (up to $5 \times 5$ arrays) that used static reflection surfaces, proving the efficiency of active environment control.
  • Scaling Laws: Increasing the size of the metasurface improves accuracy by providing more "design freedom," though benefits plateau once the surface is large enough to resolve the target grids. The study also noted that 3D sensing is inherently more challenging than 2D due to the increased grid density and potential for occlusion.

Conclusion

MetaSensing represents a paradigm shift in RF sensing by moving from passive adaptation to active environment programming. By leveraging intelligent metasurfaces and DRL, the framework achieves high-accuracy 3D sensing even in complex multipath environments. Future work will focus on hardware realization and adapting the system to highly dynamic, real-world scenarios.

Summary updated

Abstract

Using RF signals for wireless sensing has gained increasing attention. However, due to the unwanted multi-path fading in uncontrollable radio environments, the accuracy of RF sensing is limited. Instead of passively adapting to the environment, in this paper, we consider the scenario where an intelligent metasurface is deployed for sensing the existence and locations of 3D objects. By programming its beamformer patterns, the metasurface can provide desirable propagation properties. However, achieving a high sensing accuracy is challenging, since it requires the joint optimization of the beamformer patterns and mapping of the received signals to the sensed outcome. To tackle this challenge, we formulate an optimization problem for minimizing the cross-entropy loss of the sensing outcome, and propose a deep reinforcement learning algorithm to jointly compute the optimal beamformer patterns and the mapping of the received signals. Simulation results verify the effectiveness of the proposed algorithm and show how the sizes of the metasurface and the target space influence the sensing accuracy.
BibTeX
@article{hu2021metasensing,
  author = {Jingzhi Hu and Hongliang Zhang and Kaigui Bian and Marco Di Renzo and Zhu Han and Lingyang Song},
  title = {{MetaSensing: Intelligent Metasurface Assisted RF 3D Sensing by Deep Reinforcement Learning}},
  journal = {IEEE J. Sel. Areas Commun.},
  year = {2021},
  url = {https://openreview.net/forum?id=NM6kHrH7fa},
}
Figure