An NN-Aided Near-and-Far-Field Classifier via Channel Hankelization in XL-MIMO Systems

Abstract

Compared to classical communication systems, sixth-generation (6G) communication requires higher data rates, lower latency, improved energy efficiency, and more diverse users. To satisfy these many requirements, the extremely large-scale massive multiple-input multiple-output (XL-MIMO) system is attracting attention as a promising technology in 6G communication. Depending on the distance between a transmitter and a receiver, the electromagnetic radiation channels in XL-MIMO systems are divided into two models: near-field and far-field channels. The main difference between far-field and near-field is the phase-linearity, resulting in a need for a differentiated system design such as beam management. As a consequence, it is essential to classify near-field and far-field. This paper presents a new neural network (NN)-aided framework for classifying near-field and far-field using the partially captured channel in downlink scenarios in XL-MIMO systems. It is based on the mathematical reasoning that an effective latent space can be constructed with a small amount of data by using the singular values of the channel Hankelization. Briefly, it is to determine the one-hot encoding vector corresponding to each field and learn the singular values of the Hankelized channel matrix. It is noteworthy that this framework operates using the short length of input vectors and the small size of the training dataset. Simulation results show that the proposed method shows the detection rate of about 90% in almost all scenarios. Interestingly, the proposed method shows almost 100% of detection ratio in high SNR environments. It is believed that the proposed method shows superior performance than naive approaches in various environments, discovering the suitable domain to classify near-field and far-field channels.