In cell-free massive multiple-input-multiple-output system with wireless backhaul, the distributed access points (APs) and the center processing unit (CPU) are connected via wireless links. Hence, the limited backhaul bandwidth becomes a critical challenge to uplink transmission. To save the bandwidth while maintaining high transmission accuracy, we propose to deploy non-orthogonal transmissions in backhaul link and jointly optimize the detection computation mappings at the APs and the CPU under the non-orthogonal backhaul. First, we formulate the joint design problem subject to backhaul bandwidth constraint aiming at a better end-to-end transmission accuracy. Then, the non-trivial problem is parameterized and solved with a novel model-driven deep neural network, where wireless backhaul is integrated as a neural computing layer by exploiting the reciprocity between non-orthogonal transmission and additive operation. Evaluations show that, the proposed integration method outperforms the conventional approaches by a margin in both backhaul bandwidth cost and the symbol error rate.

中文翻译：

---

无蜂窝大规模MIMO的非正交无线回程设计：一种集成的计算和通信方法

在具有无线回程的无单元大规模多输入多输出系统中，分布式访问点（AP）和中央处理单元（CPU）通过无线链路连接。因此，有限的回程带宽成为上行链路传输的关键挑战。为了节省带宽并保持较高的传输精度，我们建议在回程链路中部署非正交传输，并在非正交回程下共同优化AP和CPU的检测计算映射。首先，我们制定了受回程带宽约束的联合设计问题，旨在获得更好的端到端传输精度。然后，使用新型模型驱动的深度神经网络对非平凡问题进行参数化和求解，利用非正交传输和加性运算之间的互易性，将无线回程集成为神经计算层。评估表明，所提出的集成方法在回程带宽成本和符号错误率方面均优于常规方法。