It is generally believed that downlink cell-free networks perform best under centralized implementations where the local channel state information (CSI) acquired by the access-points (AP) is forwarded to one or more central processing units (CPU) for the computation of the joint precoders based on global CSI. However, mostly due to limited fronthaul capabilities, this procedure incurs some delay that may lead to partially outdated precoding decisions and hence performance degradation. In some scenarios, this may even lead to worse performance than distributed implementations where the precoders are locally computed by the APs based on partial yet timely local CSI. To address this issue, this study considers the problem of robust precoding design merging the benefits of timely local CSI and delayed global CSI. As main result, we provide a novel distributed precoding design based on the recently proposed team minimum mean-square error method. As a byproduct, we also obtain novel insights related to the AP-CPU functional split problem. Our main conclusion, corroborated by simulations, is that the opportunity of performing some local precoding computations at the APs should not be neglected, even in centralized implementations.

翻译：普遍认为，下行无蜂窝网络在集中式实现中表现最佳，此时接入点(AP)获取的本地信道状态信息(CSI)被转发至一个或多个中央处理单元(CPU)，以基于全局CSI计算联合预编码器。然而，主要受限于前传能力不足，这一过程会引入一定延迟，可能导致预编码决策部分过时，进而造成性能下降。在某些场景下，这甚至可能比分布式实现的性能更差——在分布式方案中，预编码器由AP基于局部但及时的本地CSI计算得到。为解决此问题，本研究探讨融合及时本地CSI与延迟全局CSI优势的鲁棒预编码设计问题。作为主要成果，我们基于近期提出的团队最小均方误差方法，提出了一种新型分布式预编码设计。作为副产品，我们还获得了关于AP-CPU功能划分问题的新见解。经仿真验证的核心结论是：即使对于集中式实现，也不应忽视在AP端执行部分本地预编码计算的潜力。