Deep learning for free space optics in a data center environment

被引：2

作者：

Darwesh, Laialy ^{[1
]}

Arnon, Shlomi ^{[1
]}

机构：

[1] Ben Gurion Univ Negev, Elect & Comp Engn Dept, POB 653, IL-841405 Beer Sheva, Israel

来源：

LASER COMMUNICATION AND PROPAGATION THROUGH THE ATMOSPHERE AND OCEANS VII | 2018年 / 10770卷

关键词：

Free space optics; data center; orthogonal frequency division multiplexing; multiple input multiple output; machine learning; deep learning; neural network; fully connected; convolutional neural networks; fully convolutional network; recurrent neural network; deep neural network; on-off keying; NEURAL-NETWORKS; MASSIVE MIMO;

D O I：

10.1117/12.2321025

中图分类号：

TM [电工技术]; TN [电子技术、通信技术];

学科分类号：

0808 ; 0809 ;

摘要：

Over the last few years, there has been an exponential increase in the amount of communication network traffic, where the data center (DC) is a major building block of this network. However current DCs face various problems in the light of current demands, such as high power consumption, low scalability and low flexibility. It is necessary to build a new high speed data center which could support this exponential growth. One of the technologies that could scale up the performance of the data center is free space optical (FSO) communication. FSO communication could provide an adaptive, flexible and dynamic network that could meet the performance requirements of future DCs. However, no one has characterized the optical communication channel in DC. In DC there is an HVAC system that causes non-homogeneous changes in temperature and air velocity that can affect the performance of the optical signal. In this work, we demonstrate that by using deep learning algorithms for channel estimation and signal detection, without knowledge of the channel model, we can improve the signal detection and increase the performance of the optical communication in DC environment.

引用

页数：9