Page 7 - 3GPP_Highlights_I4

Introducing AI Feedback
and ML while Training
maintaining data Model Training
the current 5G
architecture was
a challenge that Model Model
RAN3 addressed Deployment / Performance
Update
Feedback
very effectively. In Data
order to facilitate Collection
such a task, RAN3 Inference
defined a functional data Output
framework aiming Model Inference Actor
at providing
guidance on how
different AI/ML
functions interwork, Functional Framework for RAN Intelligence
as shown: TR 37.817, fig. 4.2-1 TM

Deployment options For each use case a set of feedback information was identified. This
information indicates how the system performance is affected by the
On the basis of the principles and framework above, AI/ML based operations and it can be used, for example, to trigger
RAN3 developed different deployment options for model retraining. Feedback information is case-based and measured
the AI/ML functions: after AI/ML based decisions are taken. It may include, but it is not
• AI/ML Model Training is located in the OAM limited to, indications on QoS and UE performance, energy efficiency
and AI/ML Model Inference is located in the measurements, RAN resource status metrics.
gNB (gNB-CU for split RAN). See full details of the study in TR 37.817 – ‘Study on enhancement for
• AI/ML Model Training and AI/ML Model data collection for NR and ENDC’.
Inference are both located in the gNB (gNB-CU
for split RAN).
What’s next?
For every use case, RAN3 identified sets of AI/ML
inputs, outputs and feedback information that were The study on AI/ML support in 5G RAN concluded that, during normative
considered most relevant to carry out and evaluate phase:
AI/ML based processes.
• The working principles agreed during the study shall remain valid
A variety of input information was described. Inputs
may be generated by different entities such as • The functional framework should be used as a guideline
UEs, neighbouring RAN nodes, and RAN nodes •
hosting the AI/ML inference process. Examples of The identified use cases should be the focus of the work and the
input information are UE location information, RAN solutions developed during the study should be taken as baseline
energy efficiency metrics and RAN resource status A Work Item based on the above conclusions was approved in March
measurements. 2022 during RAN#95-e and is available in RP-220635.
The identified AI/ML outputs consist of a variety of Look out for progress on the new Work Item on Artificial Intelligence
predicted metrics and actions. Examples could be (AI)/Machine Learning (ML) for NG-RAN and consider joining us in our
predicted energy efficiency levels, predicted RAN journey into the “uncharted” territory of AI/ML in NG-RAN.
resource status metrics and mobility decisions for the
purpose of energy efficiency improvements or load https://www.3gpp.org/specifications-groups (select RAN3)
balancing optimisation.

|
Issue 04 - Ma y 2 022 07