Online Regularization of Complex-Valued Neural Networks for Structure Optimization in Wireless-Communication Channel Prediction

Abstract

This paper proposes online-learning complex-valued neural networks (CVNNs) to predict future channel states in fast-fading multipath mobile communications. CVNN is suitable for dealing with a fading communication channel as a single complex-valued entity. This framework makes it possible to realize accurate channel prediction by utilizing its high generalization ability in the complex domain. However, actual communication environments are marked by rapid and irregular changes, thus causing fluctuation of communication channel states. Hence, an empirically selected stationary network gives only limited prediction accuracy. In this paper, we introduce regularization in updates of the CVNN weights to develop online dynamics that can self-optimize its effective network size in response to such channelstate changes. It realizes online adaptive, highly accurate and robust channel prediction with dynamical adjustment of the network size. We characterize its online adaptability in a series of simulations and our practical wireless-propagation experiments demonstrate that the proposed channel prediction scheme provide 2.5 dB and 5.5 dB improvement of bit error rate (BER) at 10 −3 and 5 × 10 −4 , and achieve 10 −5 BER with E b /N 0 = 23 − 24 dB. INDEX TERMS Adaptive communications, channel prediction, channel state information (CSI), complexvalued neural network (CVNN), fading, 5G wireless communications (5G-NR)

Neural Network Fusion Processing and Inverse Mapping to Combine Multisensor Satellite Data and Analyze the Prominent Features

“…Inverse mapping is a dynamics that traces which input is prominent for which output by paying attention to the signal flow in the network. It is based on teacher-signal backpropagation dynamics where the teacher signal propagates backward from the output layer towards the input layer [37], [38]. Though the dynamics shows learning results similar to those of usual error backpropagation in an ordinary (real-valued) neural network, they provide us with a direct suggestion on the inverse mapping as follows.…”

Enhanced AI-Based CSI Prediction Solutions for Massive MIMO in 5G and 6G Systems

“…For instance, Ref. [31] proposes a CvNN network for per-path CSI prediction and introduces regularization when updating of the CvNN weights. In [25], the authors propose a temporal frame structure for updating the network.…”

Polarization-Aware Prediction of Mobile Radio Wave Propagation Based on Complex-Valued and Quaternion Neural Networks

“…QNNs have found many applications [5] such as color image processing by associating RGB channels with one another [6]- [8], robot-kinematic model construction [9], [10], controller design [11], [12], polarimetric synthetic aperture radar (PolSAR) land classification [13], [14], and so forth. On the other hand, complex-valued neural networks (CVNNs) are excellent in learning the wireless channel characteristics consisting of amplitude and phase for channel prediction [15], [16].…”