Hardware-Efficient Neural Network-Based Receiver for Intensity-Modulated Direct-Detection Short-Reach Optical Links

Intensity-modulated direct-detection (IMDD) systems dominate short-reach optical communication applications such as data center interconnects. However, the nonlinear IMDD channel as well as the non-ideal response of low-cost devices introduce severe performance penalty. In recent years, various neur...

Full description

Saved in:
Bibliographic Details
Main Authors: Xu, Zhaopeng, Wu, Qi, Chen, Cancan, Ji, Honglin, Yang, Yu, Qiao, Gang, Liang, Junpeng, Wei, Jinlong, Hu, Weisheng, Shieh, William
Format: Conference Proceeding
Language:English
Subjects:
Artificial neural networks
complexity reduction
Integrated optics
intensity-modulated direct-detection
neural network
Optical fiber networks
Performance evaluation
Quantization (signal)
Real-time systems
System performance
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Intensity-modulated direct-detection (IMDD) systems dominate short-reach optical communication applications such as data center interconnects. However, the nonlinear IMDD channel as well as the non-ideal response of low-cost devices introduce severe performance penalty. In recent years, various neural network (NN)-based receivers become popular solutions for IMDD links due to their excellent system performance. Nevertheless, the complexity of NN-based equalizers is an important concern targeting at real-time hardware implementation. In this paper, we provide several complexity reduction approaches towards hardware-efficient NN-based receivers, including multi-symbol equalization, network pruning and quantization.
ISSN:2770-1603
DOI:10.1109/ICAIT59485.2023.10367311