Fully-Decentralized MIMO Equalization Designs: Bidirectional-Chain Architecture and Efficient Algorithms

The existing massive multiple-input multiple-output (MIMO) systems predominantly adopt the centralized architecture. As the number of antennas and the transmission bandwidth become large, the amount of baseband data sampled from antennas increases dramatically. The huge amount of bus data traffic imposes a heavy burden on the baseband processing unit (BPU). To tackle this issue, in this paper, we propose a fully-decentralized bidirectional-chain (DBC) architecture without a central node for information exchange. The DBC architecture partitions all antennas into multiple clusters, and each cluster is allocated a processing module, which forms a local processing unit (LPU). By distributing the bus data traffic to multiple LPUs, the DBC architecture can effectively lower the required bandwidth between antennas and processing units. Since the number of LPUs can be dynamically adjusted, it also offers high flexibility. To fully exploit the architecture, we tailor an efficient equalization approach for the DBC architecture, which can significantly reduce the demand for bus bandwidth. To further reduce the computational complexity and enhance the performance (e.g., the bit error rate - BER), we design an efficient algorithmic deep network, referred to as the bidirectional-chain equalization deep-unfolding network (BCEDUN), by unfolding the iteration operations of the original optimization algorithm. Numerical results validate the effectiveness and superiority of our DBC architecture, demonstrating notable improvements in terms of computational complexity and baseband data traffic.