Bandwidth-Agile Image Transmission With Deep Joint Source-Channel Coding

We propose deep learning based communication methods for adaptive-bandwidth transmission of images over wireless channels. consider the scenario in which are transmitted progressively layers time or frequency, and such can be aggregated by receivers order to increase quality their reconstructions. investigate two scenarios, one sent sequentially, incrementally contribute refinement a reconstruction, another independent retrieved any order. Those scenarios correspond well known problems successive refinement xmlns:xlink="http://www.w3.org/1999/xlink">multiple descriptions , respectively, context joint source-channel coding (JSCC). DeepJSCC- $l$ an innovative solution that uses convolutional autoencoders, present three architectures with different complexity trade-offs. To best our knowledge, this is first practical multiple-description JSCC scheme developed tested information sources Numerical results show learn transmit source negligible losses end-to-end performance compared single transmission. Moreover, has comparable state art digital progressive schemes challenging low signal-to-noise ratio (SNR) small bandwidth regimes, additional advantage graceful degradation channel SNR.