5G • 5G-Advanced (Rel-18)
Joint Source-Channel Coding (JSCC)
Deep Learning Joint Source and Channel Coding for Air Interface
End-to-end neural network pipeline that maps raw multimedia data directly into channel transmission symbols, bypassing separate compression and channel coding.
Technical Explanation
Shannon's separation theorem assumes infinite block lengths, which breaks down in ultra-low latency real-time communication. Rel-18 explores deep JSCC where an autoencoder maps video/audio pixels directly into constellation points adapted to current SNR, avoiding the cliff effect and providing graceful video degradation when channel conditions plummet.
Key Functions
- Eliminates the catastrophic cliff effect where video streams abruptly freeze
- Provides graceful visual degradation proportional to instantaneous channel SNR
- Bypasses separate video codecs (H.265/AV1) and forward error correction (LDPC)
- Dramatically slashes end-to-end latency for real-time robotic tele-operation
Specifications
3GPP TR 38.843 (Rel-18)
Interfaces
Uu (User Plane)
Want to memorize 5G concepts like this one?
Study it with SuperMemo SM-2 spaced repetition flashcards.