2024-03-29T08:18:24Zhttps://ipsj.ixsq.nii.ac.jp/ej/?action=repository_oaipmhoai:ipsj.ixsq.nii.ac.jp:002038082023-04-27T10:00:04Z01164:04061:10116:10175
Preliminary Investigation of Machine Learning-based Subcarrier Selection for AoA Estimation Using Wi-Fi CSIPreliminary Investigation of Machine Learning-based Subcarrier Selection for AoA Estimation Using Wi-Fi CSIenghttp://id.nii.ac.jp/1001/00203713/Technical Reporthttps://ipsj.ixsq.nii.ac.jp/ej/?action=repository_action_common_download&item_id=203808&item_no=1&attribute_id=1&file_no=1Copyright (c) 2020 by the Information Processing Society of JapanGraduate School of Information Science and Technologies, Osaka UniversityGraduate School of Information Science and Technologies, Osaka UniversityGraduate School of Information Science and Technologies, Osaka UniversityCommunication Science Laboratories, NTTAccess Network Service Systems Laboratories, NTTAccess Network Service Systems Laboratories, NTTZesheng, CaiTakuya, MaekawaTakahiro, HaraKazuya, OharaTomoki, MurakamiHirantha, AbeysekeraWith the rapid development of wireless sensing technologies, context awareness based on Wi-Fi channel state information (CSI) has been actively studied. Specifically, methods for estimating the angle of arrival (AoA) extracted from CSI attract researchers' attention for its ability to reveal the signal paths from a transmitter to receiver. However, different subcarriers in CSI have different sensitivities and thus it is essential to exploit suitable subcarriers for AoA estimation. Traditional approaches are limited to the number of CSI packets and specialized capture devices, decreasing the opportunity for deployment. In this work, we propose to select the proper subcarriers even with one single packet for AoA estimation by utilizing machine learning without specialized devices. Our system exploits the fine-grained compressed channel state information with common IEEE 802.11ac devices and thus has the potential to be widely deployed. Instead of selecting antenna pairs from CSI measurements, which has limited combinations due to the number of receiver antennas, the developed algorithm studies the properties of subcarriers within all antenna pairs and tries to find the subcarriers with small AoA errors by using classification and regression method, e.g., Random Forest and Support Vector Regression (SVR) in our system. Our extensive experiments demonstrate that our system can accurately select the proper subcarriers with high environmental robustness for AoA estimation using Wi-Fi CSI.With the rapid development of wireless sensing technologies, context awareness based on Wi-Fi channel state information (CSI) has been actively studied. Specifically, methods for estimating the angle of arrival (AoA) extracted from CSI attract researchers' attention for its ability to reveal the signal paths from a transmitter to receiver. However, different subcarriers in CSI have different sensitivities and thus it is essential to exploit suitable subcarriers for AoA estimation. Traditional approaches are limited to the number of CSI packets and specialized capture devices, decreasing the opportunity for deployment. In this work, we propose to select the proper subcarriers even with one single packet for AoA estimation by utilizing machine learning without specialized devices. Our system exploits the fine-grained compressed channel state information with common IEEE 802.11ac devices and thus has the potential to be widely deployed. Instead of selecting antenna pairs from CSI measurements, which has limited combinations due to the number of receiver antennas, the developed algorithm studies the properties of subcarriers within all antenna pairs and tries to find the subcarriers with small AoA errors by using classification and regression method, e.g., Random Forest and Support Vector Regression (SVR) in our system. Our extensive experiments demonstrate that our system can accurately select the proper subcarriers with high environmental robustness for AoA estimation using Wi-Fi CSI.AA11838947研究報告ユビキタスコンピューティングシステム(UBI)2020-UBI-6555182020-02-242188-86982020-02-28