Unsupervised Support Vector Machines for Nonlinear Blind Equalization in CO-OFDM

Elias Giacoumidis, Athanasios Tsokanos, M. Ghanbarisabagh, S. Mhatli, L. P. Barry

Research output: Contribution to journalArticlepeer-review

6 Citations (Scopus)
24 Downloads (Pure)

Abstract

A novel blind nonlinear equalization (BNLE) technique based on the iterative re-weighted least square is experimentally demonstrated for single- and multi-channel coherent optical orthogonal frequency-division multiplexing. The adopted BNLE combines, for the first time, a support vector machine-learning cost function with the classical Sato or Godard error functions and maximum likelihood recursive least-squares. At optimum launched optical power, BNLE reduces the fiber nonlinearity penalty by ~1 (16-QAM single-channel at 2000 km) and ~1.7 dB (QPSK multi-channel at 3200 km) compared to a Volterra-based NLE. The proposed BNLE is more effective for multi-channel configuration: 1) it outperforms the “gold-standard” digital-back propagation and 2) for a high number of subcarriers the performance is better due to its capability of tackling inter-subcarrier four-wave mixing.
Original languageEnglish
Pages (from-to)1091 - 1094
Number of pages4
JournalIEEE Photonics Technology Letters
Volume30
Issue number12
Early online date4 May 2018
DOIs
Publication statusPublished - 15 Jun 2018

Keywords

  • Optical OFDM
  • fiber nonlinearity compensation
  • machine learning
  • optical fiber communication

Fingerprint

Dive into the research topics of 'Unsupervised Support Vector Machines for Nonlinear Blind Equalization in CO-OFDM'. Together they form a unique fingerprint.

Cite this