Limited CSI Feedback System for Massive MISO: A Neural Network-Aided SVD Approach

Abstract

This article introduces a novel approach for a limited but important form of channel state information (CSI) feedback in massive multiple-input single-output (MISO) systems. Leveraging feedforward neural network (FFNN) enhancement to estimate the singular values of the channel, the proposed method optimizes the estimation of channel characteristics between access points (APs) and user equipment (UE). Through the exchange of dominant singular values, the joint neural network system enables an accurate estimation of channel singular values despite noisy channel constraints. This innovative technique significantly improves the efficiency and scalability of channel characteristic estimation in complex MISO environments, representing a substantial advancement in wireless communication systems. Simulations for a complex Gaussian channel and a 2-D indoor scenario at the L band with a channel rank of 16 confirm the joint FFNN’s ability to estimate all channel singular values by measuring only the received signal using four orthogonal beams, and a measurement campaign at the millimeter frequency band validates the estimation of 32 singular values using eight orthogonal DFT beams. Moreover, the neural network exhibits superior robustness to noisy channels compared with conventional deterministic sample mean estimation methods. Finally, it is shown that the performance of FFNN improved using the attention mechanism.