Cell-Free MIMO Perceptive Mobile Networks: Cloud vs. Edge Processing

Seongah Jeong, Jinkyu Kang, Osvaldo Simeone, Shlomo ShamaiIEEE Trans. Vehicular Technology 2025 · DOI

TL;DR

This paper explores how to best distribute sensing and communication tasks in future 6G networks using cell-free MIMO. They propose and analyze four different ways to split these tasks between edge devices (access points) and a central cloud, finding that the optimal choice depends heavily on the available network capacity and the desired balance between sensing accuracy and communication speed.

What Problem Does This Paper Address?

Future mobile networks (like 6G) aim to go beyond just communication and integrate sensing capabilities, allowing devices to "perceive" their environment. This is known as "perceptive mobile networks" (PMNs). A key challenge is how to efficiently implement these integrated sensing and communication functions, especially in cell-free MIMO architectures where many distributed access points (APs) cooperate. These APs are connected to a central processing unit (CPU) via fronthaul links, which have limited capacity. The paper addresses the question of where to perform the sensing and communication processing – at the edge (APs) or in the cloud (CPU) – to optimize performance under these constraints. This involves a trade-off: processing at the edge might be faster but requires more sophisticated APs, while cloud processing can leverage more powerful resources but is limited by fronthaul bandwidth.

Methods

The authors propose and analyze four distinct functional splits for PMNs based on cell-free MIMO:

  • CDCS (Cloud-based Decoding and Sensing): Both decoding and sensing happen at the CPU, using compressed signals from APs.
  • CDES (Cloud-based Decoding, Edge-based Sensing): APs handle sensing, then compress channel estimates and signals for the CPU to decode.
  • EDCS (Edge-based Decoding, Cloud-based Sensing): APs estimate channels, decode communication signals, and then forward channel estimates to the CPU for sensing.
  • EDES (Edge-based Decoding and Sensing): Both decoding and sensing are performed individually at each AP.

For each scenario, they formulate a unified design problem to jointly optimize the quantization of signals sent over the fronthaul links. The objective is to maximize the achievable communication rate while meeting sensing requirements and respecting fronthaul capacity limits. They use information-theoretic tools and optimization techniques to analyze these trade-offs.

Datasets

Not discussed in the paper. The paper focuses on theoretical analysis and numerical simulations based on a system model.

Key Insights and Ideas

  • The "Where" Matters: The location of processing (edge vs. cloud) significantly impacts both sensing accuracy and communication rates. There's no one-size-fits-all solution.
  • Fronthaul is King (or Queen): The capacity of the fronthaul links is a critical bottleneck. When fronthaul is limited, edge-based processing becomes more attractive for communication, while cloud-based sensing can be more accurate if the fronthaul can handle the data.
  • Sensing vs. Communication Trade-off: Achieving higher sensing accuracy often comes at the cost of lower communication rates, and vice-versa. The paper quantifies this trade-off for each functional split.
  • Edge Sensing Limitations: Edge-based sensing schemes (CDES, EDES) tend to saturate in performance as the number of APs increases, whereas cloud-based sensing (CDCS, EDCS) can continue to improve.
  • EDCS for High Sensing Accuracy: When high sensing accuracy is paramount, EDCS offers a good balance, achieving high accuracy while still providing decent communication rates.
  • CDES for High Communication Rates: If maximizing communication rate is the priority, CDES is a strong contender, especially when sensing requirements are less stringent.

Key Results

  • ROC Curves: Cloud-based sensing (CDCS and EDCS) generally outperforms edge-based sensing (CDES and EDES) in terms of detection probability for a given false alarm probability, especially when fronthaul capacity is sufficient. EDCS shows an advantage by dedicating fronthaul capacity solely to sensing.
  • Sensing Accuracy vs. Fronthaul Capacity: In low fronthaul capacity regimes, edge-based sensing (EDCS) is preferable. As capacity increases, CDCS becomes competitive, and EDCS excels when capacity is very large.
  • Sensing Accuracy vs. Number of APs: Sensing accuracy improves with more APs for all schemes, but cloud-based sensing shows faster improvement.
  • Ergodic Achievable Rate vs. Fronthaul Capacity: Cloud-based decoding is generally advantageous, particularly when coupled with edge-based sensing. For large fronthaul capacity, cloud-based and edge-based decoding schemes approach similar achievable rates.
  • Ergodic Achievable Rate vs. Number of APs: Achievable rates increase with more APs for both edge and cloud decoding, with cloud-based decoding showing faster growth. EDCS or EDES are preferred for very few APs.
  • Rate-Accuracy Trade-off: EDCS achieves higher sensing accuracy but at the cost of lower communication rates compared to CDCS. CDES offers the largest communication rate by prioritizing communication signals in fronthaul usage.

Limitations and Future Work

  • Non-convexity: The optimization problems for CDCS and CDES are noted as non-convex, requiring heuristic algorithms for solutions.
  • Simplified Model: The paper considers a single stationary target. Future work could explore more complex scenarios with multiple targets, varying locations, delay spreads, and Doppler effects.
  • Multiple Users: The current analysis focuses on a single user. Extending to multiple communication users and multiple sensing targets would be a valuable next step.
  • Heuristic Algorithms: While heuristic algorithms are used, exploring more advanced optimization techniques for the non-convex problems could yield better performance.

Summary updated

BibTeX
@article{jeong2025cell,
  author = {Seongah Jeong and Jinkyu Kang and Osvaldo Simeone and Shlomo Shamai},
  title = {{Cell-Free MIMO Perceptive Mobile Networks: Cloud vs. Edge Processing}},
  journal = {IEEE Trans. Vehicular Technology},
  year = {2025},
  doi = {10.1109/tvt.2025.3534148},
}
Figure