A Direct learning approach for neural network based pre-distortion for coherent nonlinear optical transmitter

Gil Paryanti, Hananel Faig, Lior Rokach, Dan Sadot

Research output: Contribution to journalArticlepeer-review

Abstract

High throughput coherent optical transmitters are key components in future optical communication infrastructure. However, these transmitters are often distorted with the nonlinearity of their components. A potential approach for compensating nonlinearity is by applying digital pre-distortion methods based on the Volterra series or one of its derivatives. However, the Volterra series-based solution is complex to implement, difficult to scale, and its simplified versions may not yield the desired performance. Recently digital pre distortion solutions based on neural networks were proposed, which may benefit from the generality of neural networks and can be more easily scaled. These solutions are often based on non-standard neural network architectures which require complex neurons-based architectures or being based on indirect training approach which suffer from noise enhancement. In this article, a novel method for neural network-based pre-distortion with direct learning is proposed. The direct learning with neural network does not assume a specific transmitter model and does not suffer from noise enhancement. The method assumes standard neural network inference architecture and is applied to a coherent nonlinear optical transmitted with long-short-Term memory neural network. The overall performance and complexity of the direct learning method is compared with the indirect approach and with the Volterra series-based solution, showing significant advantage in performance, especially in cases of severe nonlinearity and noise conditions.

Original languageAmerican English
Article number9050812
Pages (from-to)3883-3896
Number of pages14
JournalJournal of Lightwave Technology
Volume38
Issue number15
DOIs
StatePublished - 1 Aug 2020

Keywords

  • Artificial neural networks
  • digital pre-distortion techniques
  • optical fiber communication
  • optical modulation
  • optical transmitters
  • quadrature amplitude modulation

All Science Journal Classification (ASJC) codes

  • Atomic and Molecular Physics, and Optics

Fingerprint

Dive into the research topics of 'A Direct learning approach for neural network based pre-distortion for coherent nonlinear optical transmitter'. Together they form a unique fingerprint.

Cite this