PSC: A Packet Semantic Communication Framework for Lossy Transmission With Packet Loss Channel

Currently, semantic communication is predominantly confined to wireless communication scenarios, but not the general scenarios that transmit data packets. We propose packet semantic communication (PSC), a new semantic communication framework which achieves reliable transmissions in networks with packet loss. PSC works at the application layer, and it encodes the source data as binary data bits by using deep joint source-channel coding (DeepJSCC) and entropy coding, then transmits data as packets, and the decoder at the receiver performs the decoding process even when the packets are partially lost. We implement the proposed PSC for image transmission with a packet loss channel and compare its performance with the conventional better portable graphics (BPG) codec and learned image compression (LIC) model. The results obtained show that BPG and LIC are highly vulnerable to packet loss, but the proposed PSC can achieve strong robustness against packet loss, even when the packet loss rate (PLR) is as high as 80%.