Coding Theory for Scalable Media Delivery
summary
Claims useable fountain coding: a RaptorQ systematic fountain code is broadcast as repair symbols only while devices fetch small HTTP prefixes to complete decoding, cutting edge-cache load and storage writes. A 91 MB file is split into 10 source blocks (~9 MB), sub-blocks of 223 KB, symbol size 36 bytes, about 6,300 symbols per decoding, and survives induced 20–30% loss with on-the-fly decoding and no large postprocessing. At zero overhead decode fails ~1% and each extra symbol cuts failure by ~100×.
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:31Application layer erasure coding
- 2:22Packet transmission
- 2:58Application Layer erasure codes
- 4:51AL-FEC and PHY-FEC working together
- 8:20What is a fountain code?
- 8:47LT encoding
- 9:31Belief propagation decoding
- 12:36Choosing from Soliton distribution
- 13:10Raptor codes in standards
- 14:40Major technical features (first appearance)
- 18:23Raptor codes computational complexity
- 18:35Raptor codes overhead
- 19:383GPP LTE Broadcast (MBMS) Service Layer
- 23:22File download completion using HTTP 1.1 byte range requests
- 24:50LTE broadcast offload service for HTTP
- 28:37Just-in-Time recovery demo
related talks
17mTooth: Toward Optimal Balance of Video QoE and Redundancy Cost by Fine-Grained FEC in...
10mEnhancing CDN Performance and Cutting Egress Costs in Large Video Libraries
16mSwift: Adaptive Video Streaming with Layered Neural Codecs
11mRTC @Scale, Resilience and Encryption Session | Live Q&A
13mlo-fi spy: video, steganography, and you
15mUsing The Power of Edge Compute To Enhance Streaming Delivery
22mProviding better video experiences for the next billion users.
15mWhat Codec should I use?