| 研究課題/領域番号 |
23K16870
|
|---|
| 研究種目 |
若手研究
|
| 配分区分 | 基金 |
|---|
| 審査区分 |
小区分60060:情報ネットワーク関連
|
|---|
| 研究機関 | 奈良先端科学技術大学院大学 |
|---|
研究代表者 |
Chen Na 奈良先端科学技術大学院大学, 先端科学技術研究科, 助教 (80838342)
|
|---|
| 研究期間 (年度) |
2023-04-01 – 2026-03-31
|
| 研究課題ステータス |
交付 (2023年度)
|
| 配分額 *注記 |
4,810千円 (直接経費: 3,700千円、間接経費: 1,110千円)
2025年度: 1,170千円 (直接経費: 900千円、間接経費: 270千円)
2024年度: 1,170千円 (直接経費: 900千円、間接経費: 270千円)
2023年度: 2,470千円 (直接経費: 1,900千円、間接経費: 570千円)
|
|---|
| キーワード | Beam Management / Deep Learning / Massive MIMO / IRS / Heterogeneous Network / Cell-Free Communication |
|---|
| 研究開始時の研究の概要 |
The B5G/6G systems are confronted with high transmission requirements and a harsh wireless communication environment. This research considers a radio over fiber (RoF) supported multi-layer cell-free heterogeneous network (HetNet) architecture to achieve an efficient transmission in complex scenarios. We first propose the HetNet model and deep learning (DL) method for cooperative beam management considering the nonlinear optic fiber channel and the cell-free wireless channel, providing a possible solution for future wireless communication networks with high throughput and robustness.
|
| 研究実績の概要 |
This research focus on deep learning based heterogeneous network (HetNet) beam management. We aim at developing a deep learning empowered cooperative beam management scheme for the RIS and radio-over-fiber (RoF) and intelligent reflecting surface (IRS) assisted cell-free (CF) HetNet system. We also aim at carrying out a hardware demo with campus local 5G experiment network to confirm the model performance. Comparing with existing study, this is a comprehensive study on deep learning model design for complex communication system beam cooperation with demo implementation, which is of high importance for 5G/6G systems. We mainly did the following study in the past year:
1. We developed deep learning schemes for massive multiple-input multiple-output (mMIMO) beam training. Specifically, we introduced contrastive learning mechanism with Transformer model for reducing the training overhead and improve model efficiency.
2. We developed deep learning algorithms for IRS-assisted mMIMO communication. Specifically, we considered a practical semi-passive IRS design to collect the channel information. Then we designed hybrid convolutional neural network (CNN) encoder-based Transformer deep learning model for IRS beam selection. The model achieved high prediction performance and spectral efficiency.
3. We implemented some hardware demo for RIS. We designed LC-based IRS structure and mushroom style patch design for IRS. Besides, we measured the local 5G channel character and radio-over-fiber channel character.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
Based on the theoretic analysis of the IRS-assisted mMIMO system and algorithm simulation, we designed multiple effective deep learning models achieving high beam management accuracy and spectrum efficiency improvement for massive MIMO and IRS-assisted systems. More general distributed deep learning model design considering CF access is under process.
For hardware demo, we purchased RoF devices and improved demo system scale. We simulated the IRS hardware design with simulation. We also made some basic IRS hardware models and measured the reflection performance.
Overall, the research is progressing smoothly.
|
| 今後の研究の推進方策 |
In the next step, we will continue with the following 3 aspects:
1. We will continue with the RoF nonlinear character analysis. Specifically, cooperational effect of the multiple RoF links.
2. Based on the study on modeling the CF-IRS network, we will publish journal and conference papers. Specifically, we will discuss how to reduce the training overhead and de-centralized cooperation among network devices.
3. We will accomplish international cooperation and domestic cooperation with the hardware implementation of IRS and design our apporach with deep learning algorithm loaded. We plan to apply for patents about the new IRS model with deep learning algorithm design.
|
