16.20.2 Principles

Support of AI/ML for NG-RAN requires inputs from neighbour NG-RAN nodes (e.g., predicted information, feedback information, measurements) and/or UEs (e.g., measurement results).

Signalling procedures used for the exchange of information to support AI/ML for NG-RAN, are use case and data type agnostic, which means that the intended usage (e.g., input, output, feedback) of the data exchanged via these procedures is not indicated.

AI/ML algorithms and models are out of 3GPP scope. Model-specific performance information, e.g. model performance indicators specified in clause 6 of TS 28.105 [64], is not exchanged over NG-RAN interfaces in TS 38.401 [4].

Support of AI/ML for NG-RAN does not apply to ng-eNB.

For the deployment of AI/ML for NG-RAN the following scenarios may be supported:

• AI/ML Model Training is located in the OAM and AI/ML Model Inference is located in the NG-RAN node;
• AI/ML Model Training and AI/ML Model Inference are both located in the NG-RAN node.

AI/ML Model Training follows the definition of the "ML model training" as specified in clause 3.1 of TS 28.105 [64]. An AI/ML Model needs to be trained, validated and tested before deployment for AI/ML Model Inference.

AI/ML Model Inference follows the definition of the "AI/ML inference" as defined in clause 3.1 of TS 28.105 [64].