Low Complexity Rate-Adaptive Deep Joint Source Channel Coding for Wireless Image Transmission Using Tensor-Train Decomposition

Man Xu, Yuanhui Liang, Chan Tong Lam, Benjamin Ng, Sio Kei Im

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

3 Citations (Scopus)

Abstract

We propose a low complexity deep joint source channel coding (DJSCC) with adaptive rate control model for wireless image transmission using Tensor-Train (TT) decomposition. TT is used to decompose the weights of the convolutional layers in the model in order to reduce the number of parameters and the amount of computation. According to the characteristics of the convolution layer and that of TT decomposition, the optimal decomposition ranks and the corresponding weight distribution of the network for the DJSCC with adaptive rate control model are obtained. We evaluated the performance of the proposal low complexity DJSCC with adaptive rate control model using CIFAR-10 and found that compared with the model without TT decomposition, the proposed Rate-Adaptive DJSCC model using TT decomposition can reduce the number of parameters to 52.91%, and multiplication calculation to 63.61%, with a PSNR performance degradation of about 1 dB.

Original languageEnglish
Title of host publication2022 7th International Conference on Signal and Image Processing, ICSIP 2022
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages707-712
Number of pages6
ISBN (Electronic)9781665495639
DOIs
Publication statusPublished - 2022
Event7th International Conference on Signal and Image Processing, ICSIP 2022 - Suzhou, China
Duration: 20 Jul 202222 Jul 2022

Publication series

Name2022 7th International Conference on Signal and Image Processing, ICSIP 2022

Conference

Conference7th International Conference on Signal and Image Processing, ICSIP 2022
Country/TerritoryChina
CitySuzhou
Period20/07/2222/07/22

Keywords

  • adaptive rate control image transmission
  • deep joint source channel coding
  • tensor train decomposition

Fingerprint

Dive into the research topics of 'Low Complexity Rate-Adaptive Deep Joint Source Channel Coding for Wireless Image Transmission Using Tensor-Train Decomposition'. Together they form a unique fingerprint.

Cite this