Towards Neuromorphic Processing for Next-Generation MU-MIMO Detection

Abstract

Upcoming physical layer (PHY) processing solutions, leveraging multiple-input multiple-output (MIMO) advances, are expected to support broad transmission bandwidths and the concurrent transmission of multiple information streams. However, the inherent computational complexities of conventional MIMO PHY algorithms pose significant practical challenges, not only in meeting the strict real-time processing latency requirements but also in maintaining practical computational power consumption budgets. Novel computing paradigms, such as neuromorphic computing, promise substantial gains in computational power efficiency. However, it is unknown whether it is feasible or efficient to realize practical PHY algorithms on such platforms. In this work, we evaluate for the first time the potential of neuromorphic computing principles for multi-user (MU)-MIMO detection. In particular, we developed the first spiking-based MU-MIMO simulator that meets practical error-rate targets, suggesting power gains of at least one order of magnitude when realized on actual neuromorphic hardware, compared to conventional processing platforms. Finally, we discuss the challenges and future research directions that could unlock practical neuromorphic-based MU-MIMO systems and revolutionize PHY power efficiency.