The paper proposes a retraining-free blockage pre- diction scheme that exploits angular power profiles (APPs) en- abling us to proactively control system parameters of millimeter- wave (mmWave) communications systems. Specifically, we pro- pose a feature extraction method using some minor components obtained by the principal component analysis (PCA) of the sample covariance matrix of APPs. Moreover, to cope with environmental changes without model retraining, we employ a vector database indexed by the feature vectors of the minor components as the machine learning model, and perform prediction of the obstacle positions and received signal-to-noise ratio (SNR) via a nearest-neighbor search of the database. Numerical experiments using measured APPs from an indoor mmWave communication environment demonstrate that the proposed method achieves prediction accuracy comparable to that of a supervised model using LightGBM, while the proposed approach does not require any retraining. Moreover, we confirm that prediction accuracy is preserved even when the database contains data from multiple environments with different obstacle velocities.
| 氏名 | コース | 研究室 | 役職/学年 |
|---|---|---|---|
| TONG XIAOQING | データ科学コース | 信号情報処理 （データ科学イノベ ーション教育協力講座） | 博士2回生 |
| 林 和則 | その他の専攻・大学 | 信号情報処理 （データ科学イノベ ーション教育協力講座） | 教授 |
| Koji Yamamoto | その他の専攻・大学 | Department of Information and Human Science, Kyoto Institute of Technology, Japan | 教授 |
| Kohei Mitani | --- 未設定 | 信号情報処理 （データ科学イノベ ーション教育協力講座） | --- 未設定 |
| Takuto Arai | --- 未設定 | NTT Access Network Service Systems Laboratories, Japan | --- 未設定 |
| Shuki Wai | --- 未設定 | NTT Access Network Service Systems Laboratories, Japan | --- 未設定 |
| Tatsuhiko Iwakuni | --- 未設定 | NTT Access Network Service Systems Laboratories, Japan | --- 未設定 |
| Daisei Uchida | --- 未設定 | NTT Access Network Service Systems Laboratories, Japan | --- 未設定 |