自編碼神經網路設計混合預編碼與合併器應用於毫米波多輸入多輸出系統;Hybrid Precoder and Combiner Design by Using Autoencoder Neural Network in Millimeter Wave MIMO Systems

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Communication Engineering > Electronic Thesis & Dissertation > Item 987654321/83858

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/83858

Title:	自編碼神經網路設計混合預編碼與合併器應用於毫米波多輸入多輸出系統;Hybrid Precoder and Combiner Design by Using Autoencoder Neural Network in Millimeter Wave MIMO Systems
Authors:	賴福緯;Lai, Fu-Wei
Contributors:	通訊工程學系
Keywords:	第五代通訊;毫米波;多輸入多輸出;深度學習;自編碼;5G;millimeter wave (mmWave);multiple-input multiple-output (MIMO);deep learning (DL);Autoencoder (AE)
Date:	2020-08-19
Issue Date:	2020-09-02 17:16:23 (UTC+8)
Publisher:	國立中央大學
Abstract:	本論文是將深度學習(DL)架構結合毫米波之多輸入多輸出的混合預編碼與合併器設計。在傳統的數位預編碼架構中，每根天線都會對應一個專用射頻鏈，但這在多輸入多輸出的系統中，天線數量將會非常龐大，所以混合預編碼正是改善傳統數位預編碼的問題，藉由巨量的訓練資料訓練，預測出類比預編碼的射頻鏈應有的形式，並與數位預編碼結合，降低所需的射頻鏈數量，達成降低硬體成本及高功率消耗的問題。論文中，是採用半監督是學習的方式，預測出類比的射頻鏈。使用自編碼(Autoencoder)的神經網路設計，因自編碼的內神經有對稱性質，藉此找出相關的類比預編碼與結合器的射頻鏈，而我們可利用最小平方解獲得數位的預編碼與結合器。在不同的數據流情況下，從內神經網路結果產生的頻譜效率證明設計的推測是可執行的。 ;We apply deep learning (DL) structure to hybrid precoding and combining design in millimeter wave (mmWave) communication and multiple-input multiple-output (MIMO) system. In the traditional digital precoding structure, each antenna corresponds to a dedicated RF chain, the amount of antenna will be very large. Hybrid precoding and combining will improve traditional full digital precoding problem. By huge amount of training data, predict the analog precoding and combining RF chain, this can reduce the cost and power consumption. We try to use semi-supervised learning, predict the analog RF chain. Using Autoencoder neural network design because Autoencoder neurons have symmetric properties, to find the related analog precoding and combining RF chain, digital precoder and combiner can be calculated by least square solution. In the different cases of data streams, we calculate the spectrual efficiency by neural network, it can be proved that the design is feasible.
Appears in Collections:	[Graduate Institute of Communication Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	272	View/Open

社群 sharing

Loading...