Deep Joint Source-Channel Coding Over Cooperative Relay Networks

TL;DR

Simulation results show that the proposed DeepJSCC-AF, DF, and PF schemes are superior to the digital baselines with BPG compression with polar codes and provide a graceful performance degradation with deteriorating channel quality.

Abstract

This paper presents a novel deep joint source-channel coding (DeepJSCC) scheme for image transmission over a half-duplex cooperative relay channel. Specifically, we apply DeepJSCC to two basic modes of cooperative communications, namely amplify-and-forward (AF) and decode-and-forward (DF). In DeepJSCC-AF, the relay simply amplifies and forwards its received signal. In DeepJSCC-DF, on the other hand, the relay first reconstructs the transmitted image and then re-encodes it before forwarding. Considering the excessive computation overhead of DeepJSCC-DF for recovering the image at the relay, we propose an alternative scheme, called DeepJSCC-PF, in which the relay processes and forwards its received signal without necessarily recovering the image. Simulation results show that the proposed DeepJSCC-AF, DF, and PF schemes are superior to the digital baselines with BPG compression with polar codes and provides a graceful performance degradation with deteriorating channel quality. Further investigation shows that the PSNR gain of DeepJSCC-DF/PF over DeepJSCC-AF improves as the channel condition between the source and relay improves. Moreover, DeepJSCC-PF scheme achieves a similar performance to DeepJSCC-DF with lower computational complexity.

Figures (8)

• Figure 1: Illustration of the half-duplex relay channel.
• Figure 2: The architectures of the DNNs used to parameterize $f_s, g$ and $f_{PF}, f_{DF}$ functions are shown in (a) and (b), respectively.
• Figure 3: The processing of DeepJSCC-AF, DF and PF at the relay $(\mathrm{R})$ and destination $(\mathrm{D})$, where $\beta$ is the scaling factor while $w_{rd}$ and $w_{sd}$ are the MRC coefficients in \ref{['equ:MRC']}. $\bm{SNR}$ denotes the collection of channel qualities, consisting of $SNR_{sr}, SNR_{rd}$ and $SNR_{sd}$.
• Figure 4: DeepJSCC-PF model with CA modules trained at varying SNR $(\gamma)$ values is compared to the models trained at a fixed $\gamma$ when tested at different $\gamma_{test}$. We set $SNR_{sr} = 12$ dB.
• Figure 5: Comparison between DeepJSCC-AF and DeepJSCC-DF with $SNR_{sr} \in \{0, 12, 24, \infty\}$ dB. We also include the non-cooperative scheme deepjscc as a benchmark.