Networked Sensing With AI-Empowered Interference Management: Exploiting Macro-Diversity and Array Gain in Perceptive Mobile Networks

Lei Xie, Shenghui Song, Khaled B. Letaief

TL;DR

This paper introduces a novel two-stage protocol for "perceptive mobile networks" (PMNs) that integrates sensing capabilities into existing cellular infrastructure. By intelligently managing interference between communication and sensing, and leveraging distributed sensing nodes (SNs) with AI-powered techniques, the system achieves enhanced target detection while minimizing communication overhead.

What Problem Does This Paper Address?

Future wireless networks are envisioned to go beyond communication and provide sensing services for applications like autonomous driving and environmental monitoring. Perceptive Mobile Networks (PMNs) aim to achieve this by integrating sensing into cellular networks. However, this integration faces significant challenges:

  • Interference Management: Communication signals can interfere with sensing operations, and vice-versa. This requires sophisticated methods to estimate and mitigate interference, particularly "clutter" (echoes from communication signals).
  • Distributed Sensing: Utilizing multiple sensing nodes (SNs) for collaborative sensing offers benefits like macro-diversity but introduces communication and computation burdens, especially with the need for synchronization and data sharing.
  • Efficiency: Traditional sensing methods can be computationally intensive and require substantial data, which is a challenge for power-limited devices like the proposed sensing nodes (TMTs).

This paper addresses these challenges by proposing a framework for networked sensing in PMNs that effectively manages interference and leverages distributed SNs, while also introducing an efficient AI-driven approach for clutter estimation.

Methods

The paper proposes a multi-faceted approach:

  • Two-Stage Protocol: A novel protocol divides sensing into two consecutive periods:
  • Clutter Estimation (CE) Period: Communication signals are used to estimate the "clutter" (interference for sensing).
  • Target Sensing (TS) Period: The estimated clutter information is used to manage interference during actual target detection. A specific sensing signal design ensures the clutter statistics remain consistent between the two periods.
  • Networked Detector: A Generalized Likelihood Ratio Test (GLRT) based detector is derived to fuse information from multiple sensing nodes (TMTs), exploiting macro-diversity and array gain for improved target detection.
  • SN Selection Algorithm: A sufficient condition is derived to determine when adding a new sensing node improves performance, leading to a TMT selection algorithm that optimizes resource utilization.
  • AI-Empowered Clutter Estimation (CE): To reduce communication and computation workload, a distributed, model-driven deep learning algorithm (EM-Net) is proposed. This algorithm utilizes partially sampled data and unfolds an Expectation-Maximization (EM) algorithm, making it sample-efficient.

Datasets

Not discussed in the paper. The paper relies on simulated data generated based on a defined system model and parameters.

Key Insights and Ideas

  • Separating CE and TS is Crucial: The two-stage protocol effectively decouples the estimation of interference (clutter) from the actual target sensing, allowing for better management of both.
  • Macro-Diversity vs. Array Gain in Sensing: While multiple antennas provide array gain and angular resolution for sensing, they do not offer diversity gain as they do in communication. This is because sensing primarily relies on the Line-of-Sight (LoS) component, and non-LoS components are treated as clutter.
  • AI for Efficient Clutter Estimation: The EM-Net algorithm demonstrates that deep learning can significantly improve clutter estimation efficiency, especially when dealing with limited data samples, which is critical for power-constrained sensing nodes.
  • Optimizing SN Deployment: The TMT selection algorithm highlights that simply adding more sensing nodes doesn't always guarantee better performance; a strategic selection based on contribution is necessary.

Key Results

  • Improved Detection Performance: Simulations show that the proposed networked sensing detector, leveraging macro-diversity and array gain, significantly enhances target detection probability compared to single-node sensing.
  • Efficient Clutter Estimation: The EM-Net algorithm outperforms traditional EM and Sample Covariance Matrix (SCM) methods in terms of detection probability, especially with limited data samples. It achieves comparable performance with fewer samples.
  • TMT Selection Benefits: The TMT selection algorithm effectively identifies the optimal number of sensing nodes, balancing performance gains with system complexity.
  • Validation of Theoretical Models: The simulation results closely match the theoretical predictions for detection probability, validating the analytical framework.

Limitations and Future Work

  • Swerling I Model Assumption: The target detection model assumes a Swerling I model (point-like, static or slow-moving target), neglecting Doppler effects. Future work could extend this to more complex target models.
  • Full-Duplex Maturity: While the paper's approach avoids direct full-duplex operation at the sensing nodes (TMTs), the introduction mentions that full-duplex technology for communication is still maturing.
  • Computational Complexity of EM-Net: While EM-Net is sample-efficient, its overall computational complexity can be higher than SCM for a large number of samples. However, its advantage lies in performance with limited samples.
  • Generalizability of EM-Net: The paper focuses on a specific type of clutter (low-rank in mmWave). Further research could explore the applicability of EM-Net to different clutter models and environments.
  • Real-world Deployment Challenges: The paper focuses on theoretical and simulation aspects. Real-world deployment would involve addressing practical challenges like synchronization accuracy, channel estimation errors, and hardware imperfections.

Summary updated

Abstract

Sensing will become an important service of future wireless networks to assist innovative applications such as autonomous driving and environment monitoring. Perceptive mobile networks (PMNs) were proposed to incorporate sensing capability into current cellular networks. However, the interference management between sensing and communication, as well as the collaborative sensing by multiple sensing nodes (SNs), faces significant challenges. In this paper, we first propose a two-stage protocol to tackle the interference between two sub-systems, where the echoes created by communication signals, i.e., interference for sensing, are estimated in the clutter estimation (CE) stage and then utilized for interference management in the target sensing (TS) stage. Then, a networked sensing detector is derived to exploit the perspectives provided by multiple SNs for sensing the same target. The macro-diversity from multiple SNs, the array gain, and the higher angular resolution from multiple receive antennas of each SN are then investigated to reveal the benefit of networked sensing. Furthermore, we derive the sufficient condition for one SN’s contribution to be positive, based on which a SN selection algorithm is proposed. To reduce the communication workload, we propose a distributed model-driven deep-learning algorithm that utilizes partially-sampled data for CE. Simulation results demonstrate the benefits of networked sensing and validate the higher efficiency of the proposed CE algorithm than existing methods.
BibTeX
@article{xie2023networked,
  author = {Lei Xie and Shenghui Song and Khaled B. Letaief},
  title = {{Networked Sensing With AI-Empowered Interference Management: Exploiting Macro-Diversity and Array Gain in Perceptive Mobile Networks}},
  journal = {IEEE J. Sel. Areas Commun.},
  year = {2023},
  url = {https://openreview.net/forum?id=oJD5tgg8AM},
}
Figure