A Novel Convolutional Neural Network Architecture of Deep Joint Source-Channel Coding for Wireless Image Transmission

Xin Huang,Xiaohui Chen,Li Chen,Huarui Yin,Weidong Wang
DOI: https://doi.org/10.1109/wcsp52459.2021.9613517
2021-01-01
Abstract:We propose a convolutional neural network (CNN) architecture with lower reconstruction distortion and stronger noise adaptability for deep joint source-channel coding for wireless image transmission. Instead of inserting complicated modules to the network, we design the inception modules and change the data flow via skip connection. And we expand the branches of the network using mask and attention mechanism. Experimental results demonstrate that the proposed network achieves lower distortion in additive white Gaussian noise (AWGN) channel. And noise adaptability of the proposed network is better in channel of varying signal-to-noise ratio (SNR).
What problem does this paper attempt to address?