FastNet for Symbol Detection in Massive MIMO Systems

In massive multiple-input-multiple-output (MIMO) systems, a major limiting factor for symbol detection is the amount of computational complexity required. Symbol detection in unquantized massive MIMO systems have been studied in the context of both traditional and machine learning methods. In this paper, we propose a hybrid framework that replaces some neural network layers with simple gradient descent layers to reduce complexity. Simulations showed that a judicious choice of the number of such layers can lead to significant reduction in the range of 35-50%, in computational complexity, with marginal change in performance.