Deep learning for free space optics in a data center environment

Laialy Darwesh, Shlomi Arnon

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Scopus citations

Abstract

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.

Original languageEnglish
Title of host publicationLaser Communication and Propagation through the Atmosphere and Oceans VII
EditorsAlexander M.J. Van Eijk, Jeremy P. Bos, Stephen M. Hammel
PublisherSPIE
ISBN (Electronic)9781510621114
DOIs
StatePublished - 1 Jan 2018
EventLaser Communication and Propagation through the Atmosphere and Oceans VII 2018 - San Diego, United States
Duration: 20 Aug 201822 Aug 2018

Publication series

NameProceedings of SPIE - The International Society for Optical Engineering
Volume10770
ISSN (Print)0277-786X
ISSN (Electronic)1996-756X

Conference

ConferenceLaser Communication and Propagation through the Atmosphere and Oceans VII 2018
Country/TerritoryUnited States
CitySan Diego
Period20/08/1822/08/18

Keywords

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

Fingerprint

Dive into the research topics of 'Deep learning for free space optics in a data center environment'. Together they form a unique fingerprint.

Cite this