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.

翻译：本文提出了一个全面的通信理论模型，用于描述由均匀分布在给定覆盖区域内的用户设备（UE）、无线单元（RU）和分布式单元（DU）构成的无蜂窝用户中心网络的物理层。我们考虑配备多天线的RU，并聚焦于UE、RU和DU密度恒定、因此节点数量随覆盖区域增长的情况。若网络中任意节点的计算负载和信息速率随网络规模（覆盖区域）无限增大而收敛至常数，则称该系统具有可扩展性。这要求每个UE必须由用户中心的有限尺寸RU簇处理，且此类簇处理器需动态分配至DU（例如作为软件定义的虚拟网络功能），以实现均衡的计算负载。我们进一步假设RU通过分组交换网络连接至DU，以实现自适应路由和负载均衡。针对该模型，我们详细定义了动态簇形成机制与上行导频分配方案。由于导频分配与可扩展性约束，每个簇处理器仅能获取网络信道状态信息的部分视图。我们定义了“理想部分CSI”的条件：可估计的信道向量被精确已知（而不可估计的向量完全未知）。针对上行链路，我们提出了两种基于簇的线性接收机方案，并推导了上下行对偶性，使得这些向量可复用为下行预编码器。