Optimizing Quality and Energy Efficiency in Webrtc with ML-Powered Adaptive FEC

Video and audio communication on mobile devices involves dynamic channels with fluctuating error rates along with the added constraints of battery efficiency and resource-limited hardware. Forward error correction (FEC) is a common method for error recovery but introduces computational and bandwidth overhead. To enhance FEC efficiency, machine learning (ML) can adapt error correction based on current channel dynamics. Existing solutions often use complex models, leading to performance issues and inefficiency. Our proposed solution prioritizes energy efficiency and practical deployment by combining Reed-Solomon coding and supervised learning. This approach corrects up to 60% of errors and achieves 2.5 times better energy efficiency than standard WebRTC and 1.7 times better efficiency than non-adaptive Reed-Solomon coding.