Enabling Cross-Technology Semantic Communication Via Knowledge-Driven Deep Joint Source-Channel Coding

In the Internet of things (IoTs), heterogeneous devices employing incompatible technologies, like WiFi, ZigBee, and Bluetooth, often coexist. Simplifying connections among them is crucial for ubiquitous connectivity in sixth-generation (6G) networks. While cross-technology communication (CTC) enables direct transmission, it faces reliability and efficiency issues. This paper introduces cross-technology semantic communication (CTSC), a novel paradigm to overcome these limitations. We address two key challenges. First, to overcome the prohibitive training cost of adapting to diverse physical-layer incompatibilities for each new device pair, we propose a knowledge-driven deep joint source-channel coding scheme. It incorporates existing CTC algorithms as foundational knowledge and uses a loosely coupled structure for transfer learning, significantly reducing training overhead. Second, to tackle the challenge of reconstructing images from imperfect semantic information—impaired by both random distortions and complete vector losses—we introduce a novel self-attention-based error correction method that repairs semantic information at the receiver without requiring retransmissions. Extensive simulations on image transmission scenarios demonstrate that our proposed scheme significantly reduces transmission overhead and improves structural similarity compared to CTC.