Introduction
As the telecommunications industry pivots toward 6G, the integration of sensing and communication (ISAC) has emerged as a cornerstone technology. Perceptive Mobile Networks (PMNs) aim to transform existing cellular infrastructure into a ubiquitous radar system capable of high-precision environmental monitoring and target tracking. However, the transition from communication-centric to sensing-capable networks faces significant hurdles, primarily due to environmental clutter—unwanted echoes from static objects like buildings and vegetation—and the massive data overhead required for distributed processing.
System Architecture and Protocol Design
The proposed framework utilizes a distributed architecture consisting of a central Base Station (BS) and multiple passive Target Monitoring Terminals (TMTs). This setup allows the network to exploit macro-diversity, observing targets from multiple spatial angles to mitigate blockage effects.
To manage the interference between communication and sensing signals, a two-stage transmission protocol is implemented. The downlink time is partitioned into a Clutter Estimation (CE) period and a Target Sensing (TS) period, as shown in the frame structure below.
- Clutter Estimation (CE) Stage: During this phase, the BS transmits standard communication signals (CS). The TMTs receive these signals along with environmental clutter (CL) to characterize the background noise and estimate the clutter covariance matrix.
- Target Sensing (TS) Stage: The BS transmits an Integrated Sensing and Communication (ISAC) signal. A "null-space" precoder is employed to ensure the sensing signal does not illuminate known clutter patches, thereby minimizing interference. TMTs then capture the sensing echoes (SE) from the target.
AI-Empowered Clutter Estimation: EM-Net
A critical innovation of this research is the development of EM-Net, a deep learning architecture derived from the classical Expectation-Maximization (EM) algorithm. Traditional statistical methods for clutter estimation often require dense data sampling, which is impractical in high-speed mobile environments.
By "unfolding" the iterative EM algorithm into a neural network, the researchers introduced learnable parameters that optimize the estimation process. EM-Net learns to reconstruct the environment's clutter statistics even from sparse data (e.g., 20% sampling). This "physics-inspired" AI approach offers better interpretability and faster convergence than standard black-box deep learning models, typically reaching steady-state performance in just five layers (iterations).
Key Insights and Performance
The study reveals several fundamental properties of networked sensing:
- Macro-Diversity vs. Array Gain: While multiple antennas at a single node provide array gain (improving angular resolution), they do not offer diversity gain in sensing because the signal relies primarily on the Line-of-Sight (LoS) path. True diversity is achieved through the spatial distribution of TMTs.
- Node Selection: The authors derived a "sufficient condition" for node activation, proving that adding more nodes is not always beneficial; poorly positioned nodes can degrade detection probability by introducing excessive noise.
- Interference Suppression: Simulation results showed that the proposed precoding strategy successfully suppressed sensing-induced clutter by over 40 dB, ensuring minimal impact on the primary communication functions.
Conclusion and Future Directions
This framework demonstrates that AI-empowered environment estimation can significantly reduce the data requirements for high-precision sensing in 6G networks. While the simulation results are promising, future research must address practical challenges such as clock synchronization jitter between distributed nodes and extending the model to support high-mobility targets through advanced Doppler processing.
Summary updated