Intelligent Coding And Transmission For Distributed Multi-view

We study the problem of deep joint source-channel coding in distributed multiview scenarios.In the distributed multi-viewpoint scenario,each viewpoint is a source,and each source sends image information to the common receiver through an independent noisy channel.In particular,we consider a pair of images captured by two cameras that may have overlapping fields of view transmitted over a wireless channel and reconstructed in a central node.This challenging problem involves designing a practical code that utilizes source and channel correlation to improve transmission efficiency without additional transmission overhead.To solve this problem,we need to consider the common information between the images transmitted by the two sources and the differences between the two transmission channels.In this case,we propose a deep neural network solution that includes lightweight edge encoders and a powerful center decoder.In addition,in the decoder,we propose a novel channel state information aware cross attention module to highlight the overlapping information of the images sent by two sources and take advantage of the correlation between the two noise feature maps.Our results show an impressive improvement in reconstruction quality at both links by utilizing the noise representation of the other link.In addition,compared with the separated scheme using capacity-achieving channel code,this scheme has certain competitiveness.