Accelerating Maximum-Likelihood Detection in Massive MIMO: A New Paradigm With Memristor Crossbar Based In-Memory Computing Circuit

Massive multi-input multi-output (MIMO) signal processing algorithms heavily rely on high-dimension matrix operations, which impose excessively high computational complexity. Moreover, in the post-Moore era, the performance of the classical von Neumann computing architecture is facing severe limitations. The memristor crossbar based in-memory computing holds the potential to break the memory wall and enhance the circuit's energy efficiency by orders-of-magnitude. In this paper we present a memristor crossbar based circuit design for performing the optimal maximum likelihood (ML) MIMO detection with high computation parallelism at the base station equipped with a large-scale antenna array. Simulation results show that, despite considering the circuit programming error and the memristor bit-precision, the bit error rate (BER) of the proposed circuit is almost the same as that achieved by the digital computer. Moreover, the proposed circuit achieves roughly the same level of computing performance as the NVIDIA QUADRO GV100, but with about 40 times higher energy efficiency.