A Variational Auto-Encoder Model for Underwater Acoustic Channels

Document Type

Conference Proceeding

Publication Date

11-22-2021

Department

Department of Electrical and Computer Engineering

Abstract

An underwater acoustic (UWA) channel model with high validity and re-usability is widely demanded. In this paper, we propose a variational auto-encoder (VAE)-based deep generative model which learns an abstract representation of the UWA channel impulse responses (CIRs) and can generate CIR samples with similar features. A customized training process is proposed to avoid the model collapse and being trapped in a gradient pit. The proposed deep generative model is validated using field experimental data sets.

Publication Title

WUWNet 2021 - 15th ACM International Conference on Underwater Networks and Systems

ISBN

9781450395625

Share

COinS