User-Centric Cell-Free Wireless Networks for 6G: Communication Theoretic Models and Research Challenges

TL;DR

This work proposes a scalable, cell-free, user-centric wireless network framework with constant densities of UEs, RUs, and DUs, where each UE is served by a finite-size, dynamically formed cluster of RUs processed at DUs. It introduces ideal partial CSI, two uplink receivers (CLZF and Local LMMSE), and an uplink-downlink duality that enables DL precoding from UL vectors, complemented by subspace projection to mitigate pilot contamination. Numerical results demonstrate that LMMSE-based cluster combining, aided by the duality and dynamic clustering, achieves near-symmetric UL/DL performance and that subspace projection effectively closes much of the gap to ideal CSI. The paper also identifies practical research directions in dynamic clustering, subspace covariance estimation, and UE-RU association mechanisms for real-world deployment.

Abstract

This paper presents a comprehensive communication theoretic model for the physical layer of a cell-free user-centric network, formed by user equipments (UEs), radio units (RUs), and decentralized units (DUs), uniformly spatially distributed over a given coverage area. We consider RUs equipped with multiple antennas, and focus on the regime where the UE, RU, and DU densities are constant and therefore the number of such nodes grows with the coverage area. A system is said scalable if the computing load and information rate at any node in the network converges to a constant as the network size (coverage area) grows to infinity. This imposes that each UE must be processed by a (user-centric) finite-size cluster of RUs, and that such cluster processors are dynamically allocated to the DUs (e.g., as software defined virtual network functions) in order to achieve a balanced computation load. We also assume that the RUs are connected to the DUs through a packet switching network, in order to achieve adaptive routing and load balance. For this model, we define in details the dynamic cluster formation and uplink pilot allocation. As a consequence of the pilot allocation and the scalability constraint, each cluster processor has a partial view of the network channel state information. We define the condition of ``ideal partial CSI'' when the channel vectors that can be estimated are perfectly known (while the ones that cannot be estimated are not know at all). We develop two attractive cluster-based linear receiver schemes for the uplink, and an uplink-downlink duality that allows to reuse such vectors as precoders for the downlink.

Figures (6)

• Figure 1: An example of dynamic clusters and the UE-RU association graph. The graph contains a UE-RU edge $(k,\ell)$ for all $k \in [K]$ and $\ell \in [L]$ such that $k \in {\cal U}_\ell$ and $\ell \in {\cal C}_k$.
• Figure 2: A simple network with $L = 2$ RUs and $K = 6$ users used as an example. The dotted edges correspond to channels vectors that cannot be estimated because of the cluster formation mechanism.
• Figure 3: Sum DL spectral efficiency vs. $\tau_p$ for different numbers of users $K$ for $L=10$ RUs and $M=64$ antennas each. LMMSE with power allocation from duality, LZF with PPA.
• Figure 4: Sum DL spectral efficiency vs. $\tau_p$ for different numbers of users $K$ for $L=20$ RUs and $M=32$ antennas each. LMMSE with power allocation from duality, LZF with PPA.
• Figure 5: Sum DL spectral efficiency vs. $\tau_p$ for different numbers of users $K$ for $L=40$ RUs and $M=16$ antennas each. LMMSE with power allocation from duality, LZF with PPA.

Theorems & Definitions (2)

• Remark 1
• Remark 2