| 研究者所属（当時） | 資格 | 氏名 | |
|---|---|---|---|
| （代表者） | 理工学術院 国際理工学センター（理工学術院） | 講師 | パン ジェニー |
| （連携研究者） | 情報通信学科 | 教授 | 嶋本薫 |
- 研究成果概要
Future wireless network is expected to provide more reliable communications under various mobile scenarios where classical communication methods suffer severe channel instability. Orthogonal Time Frequency Space (OTFS) is a two-dimensional modulation scheme that transmits data under a channel with significant delay and doppler spread caused by high mobility.
In OTFS system, estimating the channel matrix at the receiver end is essential to perform accurate data detection. In the first phase, we propose a deep learning-based channel estimation employing single superimposing pilot design, which can not only enhance the DD channel fading pattern, but also mitigate the interference caused by data symbols. Simulation results show that the proposed scheme achieves better BER and spectrum efficiency than the threshold methods even at low pilot power.
On the other hand, investigating the potential multipath selection can effectively improve the channel estimation optimization. In the second phase, we propose a BP learning-based self-adjustment scheme which analyzes the historical transmission and feedback the potential number of multipath channels required for estimation when signals with the same DD-domain index are transmitted. Simulation results indicate that the proposal yields better channel estimation performance than various learning-based benchmarks especially in low SNR environment.