When using an untrusted AF as data source, NWDAF may consider the data source rating results. The rating of untrusted AF is based on the quality of data collected.
Such rating may be triggered when the accuracy check based on the calculation between the predicted and ground-truth data indicates low performance, while the untrusted data source rating may be performed based on NWDAF internal logic. In the selection of the appropriate data sources, the NWDAF may use the rate of untrusted AF data sources as a criterion to calculate the expected confidence degree.
The process of rating untrusted AF data sources is depicted in Figure 6.2.13.2. For realizing potential issues, the NWDAF containing AnLF subscribes to the NWDAF containing MTLF, which performs an accuracy calculation based on the predicted and ground-truth data or alternatively the NWDAF containing AnLF can calculate the accuracy locally by comparing the predicted and ground-truth data.
NWDAF containing AnLF subscribes to NWDAF containing MTLF for obtaining an ML model using the Nnwdaf_ModelProvision_Subscribe service operation. The NWDAF containing AnLF may include a threshold (as described in clause 6.2E.2) to indicate when the NWDAF containing MTLF needs to execute the accuracy monitoring operations.
Option 1: Accuracy report from NWDAF containing MTLF
An accuracy report is sent to the NWDAF containing AnLF, e.g. when the reporting threshold is met by invoking Nnwdaf_MLModelProvision_Notify service operation.
NWDAF containing AnLF is aware that the ML Model used has a low accuracy either by receiving the accuracy report in step 2b or monitoring the accuracy by itself in step 2c. NWDAF containing AnLF determines that it needs to check further the data sources and compute data source rating. The decision conditions upon which it needs to initiate data source rating for a data source is based on NWDAF containing AnLF implementation.
NWDAF containing AnLF initiates rating of a data source by requesting and receiving supplementary data, i.e. via Nnwdaf_DataManagement_Fetch / Ndccf_DataManagement_Notify, from different data sources (if available) to verify the data source quality or correctness. Such data can be for example performance data from the OAM which are supplementary to the data from untrusted AFs, or data from UPF supplementary to the data from untrusted AFs.
NWDAF containing AnLF may send the untrusted AF data source rating to UDSF, if available. NWDAF containing AnLF uses the Nudsf_ UnstructuredDataManagement_Create service operation.
The NWDAF containing AnLF retrieves the untrusted AF rating of the data sources from the UDSF using the to use Nudsf_ UnstructuredDataManagement_Query service operation.
The NWDAF containing AnLF may use the rate of untrusted AF data sources as a criterion to calculate the confidence level of the respective analytics output.
The NWDAF containing AnLF provides the analytics output to the analytics consumer, using the Nnwdaf_AnalyticsSubscription_Notify service operation.
In the case of ML model (re)training, if the NWDAF containing MTLF is the same NWDAF containing AnLF in step 5b, it may also use the rate of untrusted AF data sources by performing steps 7b and 8-9 and then, (re)trains the ML model.
The MDA functional overview and service framework in Figure 5.1-1 as defined in TS 28.104 is used by NWDAF to trigger the MDA MnS to request analytics from the MDA Management Function.
Before NWDAF requests analytics from the MDA Management Function, the NWDAF firstly discovers the MDA Management Function via the MnS discovery service producer as defined in clause 5 of TS 28.537.
Initially MDAF(s) or MDA MnS producers register their capabilities, i.e. MnS information or MnS profile as described in clause 5 of TS 28.537 to a MnS discovery service producer. The MnS discovery service producer may contain all or partial information related to the capabilities of MDA MnS producer.
NWDAF discovers the MDA Management Function from the MnS discovery service producer by sending a MnS producer discovery service operation. The service operation may include the following parameters:
requestedMDAType: indicates a specific MDA capability such as Slice coverage analysis, Mobility performance analysis as defined in clause 7.2 of TS 28.104;
Area of Interest;
Network Slice information (i.e. NetworkSliceInfo including a DN (Distinguished Name) of the NetworkSlice managed object relating to the network slice instance associated to the S-NSSAI and NSI ID if available as defined in TS 28.541).
If more than one MDAF information is received, NWDAF selects the highest suitable MDAF and gets the address of the MDAF from Mns information. Then NWDAF sends analytics request to the MDA Management Function by triggering a MDARequest service operation as defined in clause 9.3.2 of TS 28.104. The service operation may include the following parameters:
requestedMDAOutputs: contains one or multiple MDAOutputPerMDAType, each of which consists of mDAType and optional MDAOutputIEFilters. The mDAType indicates type of analytics such as Slice coverage analysis, Mobility performance analysis, fault prediction analysis as defined in TS 28.104.
reportingMethod: method for analytics output (it can only be notification-based reporting since file-based, and streaming are not supported in NWDAF).
reportingTarget: indicates the reporting target of the MDA outputs (in case the MDA MnS consumer is different from the one that issued the request).
analyticsScope: Area of Interest, Network Slice information, NF type information, etc.
startTime: indicates the start time of the analytics requested by the MnS consumer.
stopTime: indicates the stop time of the analytics requested by the MnS consumer.
NWDAF provides the analytics response back to the analytics consumer after processing the MDA Function analytics together with other data receives from NF sources according to the Analytics ID defined in clause 6.