An Adaptive Forward Error Correction Method based on Deep Learning for Real-Time Video Transmission

With the continuous evolution of network technology, real-time video communication has witnessed a surge in popularity. However, real-time video transmission encounters challenges, such as latency, packet loss and network fluctuations, all of which can significantly impact video quality. Conventional Forward Error Correction (FEC) techniques, characterized by fixed redundancy, face difficulties in adapting to dynamic network conditions. This paper proposed a novel adaptive FEC algorithm based on Long Short-Term Memory (LSTM) neural networks, designed to elevate real-time video transmission performance. The algorithm employs frame-level Reed-Solomon coding to effectively tackle network packet loss. Furthermore, this paper introduces the concept of uneven protection, acknowledging the varied impact of different frames on video quality. Experimental results attest to the effectiveness of the proposed adaptive algorithm in dynamically selecting appropriate redundancy levels, thereby achieving Video Multi- method Assessment Fusion (VMAF) scores exceeding 80 in real- time video transmission.