China Telecom tests AI-driven satellite video transmission

China Telecom Research Institute announced the successful completion of a trial for AI-driven, cross-modal semantic communication over high-orbit satellites. The experiment, conducted in collaboration with Beijing University of Posts and Telecommunications and Pengcheng Lab, utilized China Telecom's cloud-network convergence test platform and the AsiaSat 9 satellite.

The trial introduced a novel integration of a semantic knowledge base with Joint Source-Channel Coding (JSCC) technology. This approach significantly enhanced satellite-to-ground link efficiency, reporting an improvement of over 3.5 times compared to conventional H.265 video compression. The method selectively transmits key video frames using JSCC and represents non-key frames as text vectors mapped to a knowledge base, effectively lowering data dimensionality while maintaining visual fidelity.

This resulted in measured transmission efficiencies 5.5 times higher than H.264 and 3.5 times higher than H.265, while preserving comparable video quality (Frechet Inception Distance < 50). Furthermore, the experiment addressed challenges in multi-modal semantic communication, improving transmission quality by over 70%. This was achieved through generative semantic recovery based on the knowledge base and cross-modal alignment of video and audio data.

China Telecom stated that this advancement in semantic communication technology represents a key step towards 6G developments. The company plans to deepen its collaboration with academic and research partners to drive innovation, standardization, and industrial progress in these critical areas.