Neural Network-Assisted NLoS Channel Identification in Near-Field Environments: A Linear Fractionalization Approach

초록

This letter introduces a neural network (NN)-assisted non-line-of-sight (NLoS) identification method for near-field environments. Unlike prior methods that depend on raw channel measurements and require high computational complexity, our method introduces a novel channel re-representation that maximizes the learning capability of the NN, referred to as the linear fractional channel, leading to more efficient NLoS identification. The proposed method exploits the fact that the Hankelized matrix of a linear fractional channel becomes rank-1 as both the K-factor and SNR tend to infinity. In a nutshell, we set the input and output of the NN to the singular values of the Hankelized matrix of the linear fractional channel and the one-hot encoding vector, respectively. Numerical results show that the performance of the proposed method has remarkable accuracy compared with existing baseline schemes.

키워드