請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/92484
標題: | 多輸入多輸出正交分頻多工低軌道衛星通訊系統考慮非線性功率放大器基於深度學習之下的通道預測器與訊號檢測器 Deep Learning-Based Channel Prediction and Signal Detection for MIMO OFDM LEO Satellite Communication with Nonlinear Power Amplifier |
作者: | 陳柏銉 Po-Yu Chen |
指導教授: | 蘇炫榮 Hsuan-Jung Su |
關鍵字: | 低軌道衛星通信,多輸入多輸出,正交分頻多工,信道預測,信號檢測,非線性功率放大器,深度學習, LEO Satellite Communication,MIMO,OFDM,Channel Prediction,Signal Detection,Nonlinear Power Amplifier Model,Deep Learning, |
出版年 : | 2023 |
學位: | 碩士 |
摘要: | 低軌道衛星通信正在融入第五代行動通訊技術,甚至即將到來的第六代行動通訊技術領域。在低軌道衛星通信中,使用多輸入多輸出的正交分頻多工系統不僅可以提升系統容量,同時還能實現高數據速率傳輸。然而,正交分頻多工系統中固有的高峰均功率比,是個嚴峻的挑戰,可能會造成從發射機發送的訊號,受到功率放大器所引起的非線性失真影響,進而損害接收機的信道估計和信號檢測性能。
在低軌道衛星通信中,由於低軌道衛星的快速移動和長傳輸延遲所引起的「信道老化」,是一個重要問題,使得如何獲取瞬時信道狀態的信息,變得非常具有挑戰性。因此,本文嘗試提出一種基於深度學習下,利用長短時記憶單元的信道預測器,以試圖解決此問題。這個預測器,係透過利用不斷變化的信道之間的相關性來減輕信道老化的影響。本文所提出的具體方法,首先是透過離線方式訓練信道預測器,實現信道特徵的提取,進而運用在未來通道狀態信息的線上預測。當接收端在接收到來自信道預測器的通道狀態信息,以及其接收信號後,會使用本文所提出的透過深度學習後的信號檢測器,該信號檢測器採用長期循環卷積網絡來復原傳送的信號。 此外,本文還考慮衛星上常使用的功率放大器模型,亦即固態功率放大器和行波管放大器。受過訓練的信號檢測器會利用深度學習的能力,減輕這些功率放大器對發送信號所產生的非線性失真。不僅如此,為了在低軌道衛星環境中,實際進行模擬實驗,本文根據第三代合作夥伴計劃和國際電信聯盟無線電通信部門的文檔,建構出了一個實際的低軌道衛星信道模型。依模擬結果顯示,本文所提出的信道預測器經過深度學習後,在歸一化均方誤差方面的性能優於過時的通道狀態信息。此外,本文所提出的基於深度學習的信號檢測器在面對使用不同調變方法與不同功率放大器時,可以相對優於傳統的零強制等化器,實現了更好的符號錯誤率。 Low Earth Orbit (LEO) satellite communication is being integrated into the realms of beyond 5G (B5G) and the forthcoming 6G. The utilization of multiple input multiple output orthogonal frequency division multiplexing (MIMO-OFDM) in LEO satellite communication can not only boost system capacity but also enable high data rate transmissions. Nevertheless, a challenge arises from the high peak-to-average power ratio (PAPR) inherent in OFDM systems, potentially subjecting symbols transmitted from the transmitter to nonlinear distortions from the power amplifier (PA). This, in turn, can deteriorate the performance of channel estimation and signal detection at the receiver. In LEO satellite communication, channel aging presents a significant concern, owing to the rapid movement of satellites and the long transmission delays they introduce. Consequently, obtaining instantaneous channel state information (iCSI) becomes exceedingly challenging. To address this issue, we propose a deep learning (DL)-based channel predictor that leverages long short-term memory (LSTM) units. This predictor mitigates the effects of channel aging by exploiting the correlation of the changing channels. The proposed approach begins with offline training of the channel predictor, enabling channel feature extraction and future CSI prediction online. Upon receiving CSI from the channel predictor, along with the received signals, we propose a DL-based signal detector employing a long-term recurrent convolution network (LRCN) at the receiver to recover the transmitted signal. We also consider power amplifier (PA) models typically used on satellites, namely, solid-state power amplifiers (SSPA) and traveling wave tube amplifiers (TWTA). Utilizing the capabilities of DL, we train networks to mitigate the nonlinear distortion introduced by these PAs on the transmitted signal. Furthermore, to simulate our experiments in a realistic LEO satellite environment, we implement a practical LEO satellite channel model based on documents from 3GPP and ITU-R. Simulation results demonstrate that the proposed DL-based channel predictor exhibits improved performance in terms of normalized mean squared error (NMSE) compared to outdated CSI. Additionally, the proposed DL-based receiver achieves a better symbol error rate (SER) compared to traditional zero-forcing (ZF) equalization when employing various modulation schemes and different types of PA models. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/92484 |
DOI: | 10.6342/NTU202304397 |
全文授權: | 同意授權(限校園內公開) |
顯示於系所單位: | 電信工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-112-1.pdf 目前未授權公開取用 | 4.05 MB | Adobe PDF | 檢視/開啟 |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。