Low-Delay Analog Distributed Joint Source-Channel Coding using SIRENs

We consider the design of joint source-channel coding (JSCC) schemes for two multiterminal source-channel coding problems - namely, the transmission of a Gaussian source with side information at the receiver and the transmission of bivariate Gaussian sources over two independent Gaussian channels. O...

Full description

Saved in:
Bibliographic Details
Published in:2021 29th European Signal Processing Conference (EUSIPCO) pp. 1601 - 1605
Main Authors: Xuan, Ziwei, Narayanan, Krishna
Format: Conference Proceeding
Language:English
Published: EURASIP 23.08.2021
Subjects:
Computational complexity
Computational modeling
deep learning
Distributed source-channel coding
Encoding
Europe
joint sourcechannel coding
Optimization
Receivers
Signal processing
SIRENs
ISSN:2076-1465
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:We consider the design of joint source-channel coding (JSCC) schemes for two multiterminal source-channel coding problems - namely, the transmission of a Gaussian source with side information at the receiver and the transmission of bivariate Gaussian sources over two independent Gaussian channels. Our focus is on low-delay transmission. We formulate the design problem as optimization of an autoencoder (AE) model, and show that sinusoidal representation networks (SIRENs) are a good choice due to their inherent periodicity and the ability of stretched sinusiods to cover the source space. We show that SIRENs outperform parametric ReLU based networks. The complexity of the proposed method scales better with source dimension than the best traditional schemes known in the literature while their performance is comparable to or better than that of the best traditional schemes. We demonstrate that the spontaneously learned encoder mappings share resemblance to the classical Wyner-Ziv mappings for JSCC with side information, and exhibits structured patterns in the case of distributed coding that are interpretable.
ISSN:2076-1465
DOI:10.23919/EUSIPCO54536.2021.9616186