Collaborative Learning based Symbol Detection in Massive MIMO

Abstract

Massive multiple-input multiple-output (MIMO) system is a core technology to realize high-speed data for 5G and beyond systems. Though machine learning-based MIMO detection techniques outperform conventional symbol detection techniques, in large user massive MIMO, they suffer from maintaining an optimal bias-variance trade-off to yield optimal performance from an individual model. Hence, in this article, collaborative learning based low complexity detection technique is proposed for uplink symbol detection in large user massive MIMO systems. The proposed detection technique strategically ensembles multiple fully connected neural network models utilizing iterative meta-predictor and reduces the final estimation error by smoothing the variance associated with individual estimation errors. Simulations are carried out to validate the performance of the proposed detection technique under both perfect and imperfect channel state information scenarios. Simulation results reveal that the proposed detection technique achieves a lower bit error rate while maintaining a low computational complexity as compared to several existing uplink massive MIMO detection techniques.