CR2Net: A Neural Network-Based Classifier for Rician and Rayleigh Channels via Hankelization

Abstract

Channel modeling in a wireless communication system is one of the most fundamental and important issues. Among various channel models, Rician and Rayleigh fading channels are the most orthodoxly used in wireless communications. In this letter, we present a neural network (NN)-based framework that classifies channels having Rician or Rayleigh fading, referred to as CR2Net. The proposed framework is based on the mathematical fact that the Hankelized matrix of a Rician channel has rank-1 property when the K-factor and signal-to-noise ratio (SNR) are infinite. Based on the inference that the corresponding information is still statistically contained in Hankelized matrices of Rician or Rayleigh channels even in a practical environment, we set the input and output of CR2Net to singular values of Hankelized matrices and one-hot encoding vector, respectively. This framework enables NNs to be operated with small sizes of both input vectors and training dataset. Simulation results show that CR2Net has superior classification performance than existing methods such as the conventional NN-based algorithm in various scenarios.