5G • 5G-Advanced (Rel-18)
Two-Sided AI Architecture
Joint UE-gNodeB Collaborative AI/ML Architecture
Collaborative AI framework where paired neural network models run simultaneously at both the UE and the gNodeB to solve joint physical layer problems.
Technical Explanation
In 3GPP Release 18, Category 2 AI workflows denote 'Two-Sided' models where the network and the terminal execute complementary halves of a shared neural network architecture. For example, in CSI compression, the encoder resides on the UE modem and the decoder resides on the gNodeB baseband, necessitating coordinated training, interface synchronization, and joint model versioning.
Key Functions
- Synchronizes paired neural network weights between UE modems and base stations
- Enables joint autoencoder training for asymmetric uplink/downlink physical channels
- Defines standard interface feedback for end-to-end model parameter validation
- Maintains version compatibility across heterogeneous device chipsets
Specifications
3GPP TR 38.843, TS 38.214 (Rel-18)
Interfaces
Uu (PUCCH/PUSCH)F1-C
Want to memorize 5G concepts like this one?
Study it with SuperMemo SM-2 spaced repetition flashcards.