Link Availability Prediction Framework for Sub-Terahertz Wireless LANs Based on Multi-Band Propagation Information

Authors:
Open Access
Publisher website
Google Scholar
Abstract
This study proposes a link availability prediction framework for sub-terahertz wireless local area networks (LANs). In the sub-terahertz band, to combat severe path loss between an access point (AP) and a station (STA), it is necessary to form a narrow beam. Exhaustive beam search due to a large number of candidate beams causes significant processing overhead in establishing wireless channels between an AP and a STA. To address this issue, we propose a novel framework to predict link availability that utilizes the propagation information of lower frequency bands. To incorporate realistic channel characteristics, we measured the propagation characteristics of signals in the millimeter wave and microwave bands using off-the-shelf wireless devices under various AP/STA positions and orientations. We analyzed the measured characteristics of each band by quantifying the fluctuation and separability of the features in both linear and nonlinear spaces. The analysis revealed complementary characteristics: channel state information (CSI) in the microwave band showed lower fluctuation and higher separability in linear space, while the received signal strength indicator (RSSI) in the millimeter wave band showed greater robustness in linear space and better separability in nonlinear space. Motivated by this complementarity, we designed a probabilistic neural network (PNN)-based architecture that integrates neural encoders to predict the link availability of the sub-terahertz band. Numerical experiments using the measured data demonstrate high performance, and show robustness to fluctuations and adaptability to variations in AP/STA positions and orientations. This study verifies the effectiveness of the proposed link prediction framework for sub-terahertz wireless LANs.
Funding Information
  • Research and Development for Expansion of Radio Wave Resources (JPJ000254)

This publication has 34 references indexed in Scilit: