Decentralized equalization for massive MU-MIMO on FPGA

Massive multi-user multiple-input multiple-output (MU-MIMO) relies on large antenna arrays that serve tens of user equipments in the same time-frequency resource. The presence of hundreds of antenna elements and radio-frequency (RF) chains at the base station (BS) enables high spectral efficiency via fine-grained beamforming, but poses significant practical implementation challenges. In particular, conventional linear equalization algorithms used in the massive MU-MIMO uplink (users transmit to the BS), such as zero-forcing, typically require centralized architectures, which cause excessively high computational complexity and interconnect bandwidth between the baseband processing unit and the RF chains. In order to mitigate the complexity and bandwidth bottlenecks, we propose a VLSI design of a decentralized feed-forward architecture and a parallel equalization algorithm relying on large-MIMO approximate message passing (LAMA). We use high-level synthesis (HLS) to develop the VLSI architecture and provide corresponding FPGA implementation results. Our results demonstrate that the proposed decentralized LAMA equalizer achieves competitive performance and complexity as existing centralized solutions that have been designed on register-transfer level.