| 電子情報通信学会 研究会発表申込システム 講演論文 詳細 |
技報閲覧サービス [ログイン] 技報アーカイブ |
| トップに戻る | 前のページに戻る | [Japanese] / [English] |
| 講演抄録／キーワード | ||
| 講演名 | 2025-11-07 17:20 Physics-Informed Autoencoder for Robust Channel Reconstruction in Rician Fading LEO Satellite Systems Rizky Pratama Hudhajanto（NAIST）・Duong Quang Thang（OIT）・Na Chen・○Minoru Okada（NAIST）  SAT2025-61 |
|
| 抄録 | （和） | Low Earth Orbit (LEO) satellite systems face significant challenges in channel estimation due to dynamic orbital motion, Doppler shifts, and Rician fading, which complicate reliable communication with ground users. This paper proposes a physics-informed autoencoder (PINN AE) framework that integrates physical priors—such as satellite positions, azimuth angles, Doppler shifts, and distances—into the autoencoder architecture, augmented by a Line-of-Sight (LoS) regularization term in the loss function. The model is trained on simulated beamformed channel data generated from LEO satellite communication scenario serving several users in major Japanese cities, incorporating path loss, shadowing, and Rician fading with varying K-factors. Comparative evaluations against a vanilla autoencoder demonstrate that the PINN AE achieves up to 10% lower test mean squared error (MSE) in LoS-dominant regimes with balanced physics weighting, while exhibiting superior denoising performance (14–20% reduced MSE to LoS channels) across configurations. Furthermore, in data-limited scenarios, the PINN AE provides 15–20% relative MSE improvements, leveraging physical constraints for enhanced generalization. Simulation results validate the framework's efficacy for robust channel reconstruction, with implications for efficient non-terrestrial networks. Future extensions could incorporate adaptive physics terms for broader fading environments. |
| （英） | Low Earth Orbit (LEO) satellite systems face significant challenges in channel estimation due to dynamic orbital motion, Doppler shifts, and Rician fading, which complicate reliable communication with ground users. This paper proposes a physics-informed autoencoder (PINN AE) framework that integrates physical priors—such as satellite positions, azimuth angles, Doppler shifts, and distances—into the autoencoder architecture, augmented by a Line-of-Sight (LoS) regularization term in the loss function. The model is trained on simulated beamformed channel data generated from LEO satellite communication scenario serving several users in major Japanese cities, incorporating path loss, shadowing, and Rician fading with varying K-factors. Comparative evaluations against a vanilla autoencoder demonstrate that the PINN AE achieves up to 10% lower test mean squared error (MSE) in LoS-dominant regimes with balanced physics weighting, while exhibiting superior denoising performance (14–20% reduced MSE to LoS channels) across configurations. Furthermore, in data-limited scenarios, the PINN AE provides 15–20% relative MSE improvements, leveraging physical constraints for enhanced generalization. Simulation results validate the framework's efficacy for robust channel reconstruction, with implications for efficient non-terrestrial networks. Future extensions could incorporate adaptive physics terms for broader fading environments. | |
| キーワード | （和） | Low Earth Orbit satellites / channel reconstruction / physics-informed neural networks / Doppler compensation / autoencoders / Rician fading / non-terrestrial networks / deep learning |
| （英） | Low Earth Orbit satellites / channel reconstruction / physics-informed neural networks / Doppler compensation / autoencoders / Rician fading / non-terrestrial networks / deep learning | |
| 文献情報 | 信学技報, vol. 125, no. 227, SAT2025-61, pp. 54-63, 2025年11月. | |
| 資料番号 | SAT2025-61 | |
| 発行日 | 2025-10-30 (SAT) | |
| ISSN | Online edition: ISSN 2432-6380 | |
| 著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します．(許諾番号：10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) | |
| PDFダウンロード | SAT2025-61 |
|
| 研究会情報 | |
| 研究会 | SAT KOSST |
| 開催期間 | 2025-11-06 - 2025-11-07 |
| 開催地（和） | Hotel Sunshine (韓国・大田) |
| 開催地（英） | Hotel Sunshine (Korea, Daejeon) |
| テーマ（和） | 衛星通信技術及び一般 (JC-SAT 2025: 韓国KOSST連催) |
| テーマ（英） | Satellite communication technology, etc. (JC-SAT 2025 co-hosted by IEICE and KOSST) |
| 講演論文情報の詳細 | |
| 申込み研究会 | SAT |
| 会議コード | 2025-11-SAT-KOSST |
| 本文の言語 | 英語 |
| タイトル（和） | |
| サブタイトル（和） | |
| タイトル（英） | Physics-Informed Autoencoder for Robust Channel Reconstruction in Rician Fading LEO Satellite Systems |
| サブタイトル（英） | |
| キーワード(1)（和/英） | Low Earth Orbit satellites / Low Earth Orbit satellites |
| キーワード(2)（和/英） | channel reconstruction / channel reconstruction |
| キーワード(3)（和/英） | physics-informed neural networks / physics-informed neural networks |
| キーワード(4)（和/英） | Doppler compensation / Doppler compensation |
| キーワード(5)（和/英） | autoencoders / autoencoders |
| キーワード(6)（和/英） | Rician fading / Rician fading |
| キーワード(7)（和/英） | non-terrestrial networks / non-terrestrial networks |
| キーワード(8)（和/英） | deep learning / deep learning |
| 第1著者 氏名（和/英/ヨミ） | Rizky Pratama Hudhajanto / Rizky Pratama Hudhajanto / |
| 第1著者 所属（和/英） | Nara Institute of Science and Technology (略称： NAIST) Nara Institute of Science and Technology (略称： NAIST) |
| 第2著者 氏名（和/英/ヨミ） | Duong Quang Thang / Duong Quang Thang / |
| 第2著者 所属（和/英） | Osaka Institute of Technology (略称： OIT) Osaka Institute of Technology (略称： OIT) |
| 第3著者 氏名（和/英/ヨミ） | Na Chen / Na Chen / |
| 第3著者 所属（和/英） | Nara Institute of Science and Technology (略称： NAIST) Nara Institute of Science and Technology (略称： NAIST) |
| 第4著者 氏名（和/英/ヨミ） | Minoru Okada / Minoru Okada / |
| 第4著者 所属（和/英） | Nara Institute of Science and Technology (略称： NAIST) Nara Institute of Science and Technology (略称： NAIST) |
| 講演者 | 第4著者 |
| 発表日時 | 2025-11-07 17:20:00 |
| 発表時間 | 25分 |
| 申込先研究会 | SAT |
| 資料番号 | SAT2025-61 |
| 巻番号（vol） | vol.125 |
| 号番号（no） | no.227 |
| ページ範囲 | pp.54-63 |
| ページ数 | 10 |
| 発行日 | 2025-10-30 (SAT) |
[研究会発表申込システムのトップページに戻る]