We present an AI-based framework for semantic transmission of multimedia data over band-limited, time-varying channels. The method targets scenarios where large content is split into multiple packets, with an unknown number potentially dropped due to channel impairments. Using joint source-channel coding (JSCC), our approach achieves reliable semantic reconstruction with graceful quality degradation as channel conditions worsen, eliminating the need for retransmissions that cause unacceptable delays in latency-sensitive applications such as video conferencing and robotic control. The framework is compatible with existing network protocols and further enables intelligent congestion control and unequal error protection. A tunable design parameter allows balancing robustness at low channel quality against fidelity at high channel quality. Experiments demonstrate significant robustness improvement over state-of-the-art baselines in both image and video domains.

翻译：我们提出了一种基于人工智能的框架，用于在带宽受限、时变信道上的多媒体数据语义传输。该方法针对大容量内容被分割为多个数据包、且因信道损伤可能丢失未知数量数据包的场景。通过联合信源信道编码，我们的方法实现了可靠的语义重建，并在信道条件恶化时实现平缓的质量降级，从而消除了在视频会议和机器人控制等延迟敏感应用中导致不可接受延迟的重传需求。该框架与现有网络协议兼容，并进一步支持智能拥塞控制和不等差错保护。一个可调设计参数允许在低信道质量下的鲁棒性与高信道质量下的保真度之间进行权衡。实验证明，在图像和视频领域，该方法相较于最先进的基线模型均实现了显著的鲁棒性提升。