Free-Ride Transmission of Semantic Features in Wireless Video Surveillance Systems

This paper is concerned with the wireless video surveillance systems, which were widely deployed and now augmented by edge computing. Armed with the edge computing, on-device local intelligence algorithms like machine learning (ML) can be utilized to extract specific semantics such as events classification in terms of risk levels which are vital for downstream tasks, say video retrieval. Different from the emergent semantic communications, not only these extracted semantic features (for further use) but also the raw video (by legal requirement) need to be sent to the surveillance center. This application scenario is also different from those for the conventional edge computing. The main objective of this paper is to propose a cost-effective scheme for such extra semantic data transmission in a free-ride way that has mild impact on the existing communication link and requires neither bandwidth expansion nor extra transmission power. The basic idea is to superimpose in the binary field the extra bits on the coded payload data. Numerical results show that, with the 5G low-density parity-check (LDPC) codes, simultaneously transmitting semantic features along with the payload data delivers more reliable semantic features and has a negligible effect on the quality of the payload data.