WO2025105990A1

WO2025105990A1 - Intent management in 5g core

Info

Publication number: WO2025105990A1
Application number: PCT/SE2023/051154
Authority: WO
Inventors: Danesh DAROUI; Swarup Kumar Mohalik
Original assignee: Telefonaktiebolaget LM Ericsson AB
Current assignee: Telefonaktiebolaget LM Ericsson AB
Priority date: 2023-11-14
Filing date: 2023-11-14
Publication date: 2025-05-22
Anticipated expiration: 2026-05-14

Abstract

A node in a communications network core may receive, from a service consumer network function (NF), a request for an analytics report regarding a network state. The node may receive an intent from the service consumer NF representing a desired network state. The node may determine, from the received intent, at least one key performance indicator (KPI) associated with the network state. Responsive to said determining, the node may monitor a value of the at least one KPI to determine whether at least one action to fulfill the received intent has been triggered. By monitoring the KPI value, the node may measure accuracy of the analytics report for further improvement.

Description

Intent Management in 5G Core

TECHNICAL FIELD

[0001] The present disclosure relates to intent management in 5G Core (5GC).

BACKGROUND

[0002] In 5GC, a network data analytics function (NWDAF) generates analytics reports for service consumer network functions (NFs) upon request. The service consumer NFs take suitable actions after receiving the analytics reports. Knowing actions that have been taken or potentially taken by the service consumer NFs would be helpful for the NWDAF to verify the accuracy of the generated analytics reports. Such knowledge would also be helpful for the NWDAF to recommend a particular action to the service consumer NF when generating the analytics report. However, for reasons discussed below, it is infeasible for the NWDAF to gain the knowledge of all possible actions triggered by the service consumer NFs.

[0003] First, there are many possible actions that can be taken by the service consumer NFs in the network. Such actions are so diverse that makes it infeasible to store information about all actions in one place, i.e., the NWDAF.

[0004] Second, revealing information about actions can result in privacy violations and/or intellectual property violations. For example, actions can relate to user equipment (UE), policy, vendor or operator data, and authentication, among others, which shall not be exposed unless certain criteria are fulfilled.

[0005] Third, even if the NWDAF knows all possible actions, there might exist conflicts among different actions. Conflict resolution among different actions can be a sophisticated task. [0006] In view of the above, there is a need to enable the NWDAF to efficiently measure the accuracy of the generated analytic reports, without exchanging information about actions between the NWDAF and the service consumer NFs. Moreover, without such exchange of information about actions, there is also a need to enable the NWDAF to provide recommendations to the service consumer NFs when the service consumer NFs request analytic reports, so as to assist the service consumer NFs to trigger efficient actions to reach a network state desired by the service consumer NFs.

SUMMARY

[0007] The present disclosure enables communications between service consumer NFs and NWDAF in the form of intents to express requirements and/or expectations with respect to the network state, without revealing any information about actions taken by the service consumer NFs. The service consumer NFs may be enhanced to include an intent management function (IMF) to process intents and trigger proper actions to fulfill the intents. The service consumer NFs may provide their intents to the NWDAF as feedback in response to receiving analytic reports from the NWDAF. The NWDAF may be enhanced to process intents and extract a list of key performance indicators (KPIs). The KPIs may be affected by the actions that are triggered to fulfill the intents. By monitoring the KPIs, the NWDAF may know whether any action has been taken by a service consumer NF without the need to know about details of the triggered actions, such as the type of the triggered actions.

[0008] Further, the NWDAF may be enhanced to implement a machine learning (ML) model to predict a network state desired by a service consumer NF, and then use the prediction for recommendation purposes. For example, the NWDAF may send a set of recommendations in the form of intents to the service consumer NF without proposing any action to be taken by the service consumer NF. [0009] According to one aspect of the disclosure, a method may be performed by one or more nodes in a communications network core. The node(s) may include an NWDAF. The node(s) may receive, from a service consumer NF, a request for an analytics report regarding a network state. The node(s) may receive an intent from the service consumer NF representing a desired network state. The node(s) may determine, from the received intent, at least one KPI associated with the network state. Responsive to said determining, the node(s) may monitor a value of the at least one KPI to determine whether at least one action to fulfill the received intent has been triggered.

[0010] In some embodiments, the request for the analytics report may include an analytics identifier.

[0011] In some embodiments, the node(s) may output the analytics report to the service consumer NF based on the request. The intent received from the service consumer NF may be based on the analytics report.

[0012] In some embodiments, the node(s) may train an ML model to predict a network state desired by another service consumer NF. To train the ML model, the node(s) may use one or more of the following: the monitored value of the at least one KPI before the at least one action has been triggered, the monitored value of the at least one KPI after the at least one action has been triggered until the intent is removed, a type of the service consumer NF, and the desired network state represented by the received intent.

[0013] In some embodiments, the node(s) may receive, from the service consumer NF, at least one criterion for removing the intent.

[0014] In some embodiments, in the determining step, a trimmed IMF may extract the at least one KPI from the intent submitted by the service consumer NF.

[0015] In some embodiments, the node(s) may include processing circuitry configured to perform the above method. [0016] According to another aspect of the disclosure, a method performed by one or more nodes in a communications network core. The node(s) may include an NWDAF. The node(s) may receive, from a service consumer NF, a request for an analytics report regarding a network state. The node(s) may output the analytics report to the service consumer NF based on the request. The node(s) may receive an intent from the service consumer NF as feedback to the analytics report. The node(s) may detect whether an action relevant to the generated analytics report has been triggered.

[0017] In some embodiments, the node(s) may determine, from the received intent, at least one KPI that can be affected by the action. The node(s) may monitor a value of the at least one KPI to detect whether the action has been triggered.

[0018] In some embodiments, the node(s) may measure accuracy of the analytics report based on the detection.

[0019] In some embodiments, the node(s) may include processing circuitry configured to perform the above method.

[0020] According to yet another aspect of the disclosure, a method may be performed by one or more nodes in a communications network core. The node(s) may include an NWDAF. The node(s) may receive, from a service consumer NF, a request for an analytics report regarding a network state. An ML model may predict a value of at least one KPI that can be affected by at least one action that is triggered after the service consumer NF receives the analytics report. The predicted value may describe a network state desired by the service consumer NF. The node(s) may generate a recommendation intent describing the predicted network state desired by the service consumer NF. The node(s) may send the recommendation intent to the service consumer NF.

[0021] In some embodiments, the node(s) may output the analytics report to the service consumer NF based on the request. [0022] In some embodiments, the node(s) may include processing circuitry configured to perform the above method.

[0023] According to one aspect of the disclosure, a method may be performed by a service consumer NF. The service consumer NF may send, to an NWDAF, a request for an analytics report regarding a network state. The service consumer NF may receive the analytics report from the NWDAF. The service consumer NF may determine, based on the analytics report, an intent representing a desired network state. The service consumer NF may send the intent to at least one of an IMF and the NWDAF.

[0024] In some embodiments, the service consumer NF may receive a notification from the IMF that at least one action to fulfill the intent has been triggered.

[0025] In some embodiments, the service consumer NF may determine whether the desired network state is reached. The service consumer NF may determine and submit a new intent to the IMF and the NWDAF, when the desired network state is not reached.

[0026] In some embodiments, the service consumer NF may send an intent removal request to the IMF and the NWDAF.

[0027] In some embodiments, the service consumer NF may send, to at least one of the IMF and the NWDAF, at least one criterion for removing the intent.

[0028] In some embodiments, the at least one criterion may include one or more of the following: timeout, network situation, and at least one KPI value.

[0029] In some embodiments, the service consumer NF may receive at least one recommendation intent from the NWDAF.

[0030] In some embodiments, determining the intent may include determining the intent based on the received at least one recommendation intent.

[0031] In some embodiments, determining the intent may include discarding the received at least one recommendation intent and creating a new intent for sending to the IMF and the NWDAF.

[0032] In some embodiments, the service consumer NF may include processing circuitry configured to perform the above method.

[0033] Of course, the present invention is not limited to the above features and advantages. Indeed, those skilled in the art will recognize additional features and advantages upon reading the following detailed description, and upon viewing the accompanying drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The above and other objects, features and advantages will be more apparent from the following description of embodiments with reference to the accompanied drawings.

[0035] FIG. 1 illustrates example components of a communications network core relevant to one aspect of the technology.

[0036] FIG. 2 illustrates a sequence diagram related to providing feedback by the service consumer NF to the NWDAF according to one aspect of the technology.

[0037] FIG. 3 illustrates a flow chart corresponding to the sequence illustrated in FIG. 2 according to one aspect of the technology.

[0038] FIG. 4 illustrates a sequence diagram related to providing recommendations by the NWDAF to the service consumer NF according to one aspect of the technology.

[0039] FIG. 5 illustrates a flow chart corresponding to the sequence illustrated in FIG. 4 according to one aspect of the technology.

[0040] FIG. 6 illustrates a flow chart of a method performed by an NWDAF according to one aspect of the technology.

[0041] FIG. 7 illustrates a flow chart of another method performed by the NWDAF according to one aspect of the technology. [0042] FIG. 8 illustrates a flow chart of yet another method performed by the NWDAF according to one aspect of the technology.

[0043] FIG 9 illustrates a flow chart of a method performed by a service consumer NF according to one aspect of the technology.

[0044] FIG. lOAis a block diagram showing internal components of the service consumer NF according to one aspect of the technology.

[0045] FIG. 10B is a block diagram showing internal components of an IMF according to one aspect of the technology.

[0046] FIG. 10C is a block diagram showing internal components of the NWDAF according to one aspect of the technology.

DETAILED DESCRIPTION

[0047] Notably, modifications and other embodiments of the disclosed invention(s) will come to mind to one skilled in the art having the benefit of the teachings presented in the foregoing descriptions and the associated drawings. Therefore, it is to be understood that the invention(s) is/are not to be limited to the specific embodiments disclosed and that modifications and other embodiments are intended to be included within the scope of this disclosure. Although specific terms may be employed herein, they are used in a generic and descriptive sense only and not for purposes of limitation.

1. Overview

[0048] FIG. 1 illustrates a communications network core 100 relevant to the present disclosure. The core 100 may include one or more service consumer NFs 102, one or more IMFs 104 and an NWDAF 106, each of which is discussed in detail below. The core 100 may comprise many other functions not illustrated. 1.1 Service Consumer Network Function

[0049] Service consumer NFs 102 may include policy control function (PCF), user plane function (UPF), application functions (AFs), and Operations, Administration and Maintenance (0AM), among other possibilities.

[0050] The service consumer NF 102 may send, to the NWDAF 106, a request for an analytics report regarding a network state. The request may include an analytics identifier (ID), such as UE mobility, user data congestion, and network performance, among other possibilities. The analytics report may include one or more of the following: current data of the network state, historical data of the network state, and a prediction of the network state. In one example, the service consumer NF 102 may send a request to the NWDADF inquiring UE mobility for a specific UE or a group of UEs within a certain period of time.

[0051] The service consumer NF 102 may receive the analytics report from the NWDAF 106. Based on the analytics report, the service consumer NF 102 or its associated IMF 104 may trigger an action. By way of example, the service consumer NF 102 may be a PCF, which may send a request to the NWDAF 106 for an analytics report on user data congestion. If the analytics report predicts that user data congestion will occur, the PCF or its associated IMF 104 may take an action to prevent congestion in the network. If the analytics report predicts no user data congestion, the PCF may release resources. In another example, the service consumer NF 102 may be a UPF, which may ask for an analytics report regarding mobility prediction for a UE. The analytics report may provide a predicted trajectory of the UE. Based on the predicted trajectory, the UPF or its associated IMF 104 may take suitable actions to allocate resources in advance before the UE reaches a certain network node, e.g, gNodeB (gNB).

[0052] In some embodiments, based on the analytics report, the service consumer NF 102 may evaluate the network state, and determine an intent representing its desired network state. The intent may express requirements/ expectations that reflect the desired network state. The service consumer NF 102 may send the intent to the NWDAF 106. By communicating with the NWDAF 106 about its intent, the service consumer NF 102 may avoid revealing any information about its actions to the NWDAF 106.

[0053] The service consumer NF 102 may send the same intent to the IMF 104 to trigger one or more proper actions. Once the IMF 104 triggers the action(s), the service consumer NF 102 may receive a notification from the IMF 104 that the action(s) to fulfill the intent has been triggered.

[0054] Thereafter, the service consumer NF 102 may evaluate the network state to determine whether the desired network state is reached. When the desired network state is not reached or when the network state is not as expected, the service consumer NF 102 may compose a new intent, and submit the new intent to the IMF 104 and the NWDAF 106. The service consumer NF 102 may continuously evaluate the network state, generate a new intent and submit the new intent to the IMF 104 and the NWDAF 106 in a closed loop until the desired network state is achieved, e.g, within a specified deadline.

[0055] The service consumer NF 102 may receive at least one recommendation intent from the NWDAF 106. The service consumer NF 102 may take the recommendation intent(s) into account, and choose the best intent for submission to the IMF 104 and the NWDAF 106. Alternatively, the service consumer NF 102 may completely discard the recommendation intent(s) and create a new intent for sending to the IMF 104 and the NWDAF 106.

[0056] The service consumer NF 102 may send an intent removal request to the IMF 104 and the NWDAF 106. The service consumer NF 102 may specify one or more criteria to remove the intent if the intent is not fulfilled. The criteria may include one or more of the following: timeout, network situation, and at least one KPI value associated with the network state, among other possibilities. The service consumer NF 102 may send the criteria to the IMF 104 and/or the NWDAF 106 for them to remove the intent. 1.2 Intent Management Function (IMF)

[0057] Each service consumer NF 102 may be associated with an IMF 104. The IMF 104 may be part of the service consumer NF 102 or may be an independent entity separate from the service consumer NF 102. The IMF 104 may process the intent(s) provided by the service consumer NF 102, and trigger one or more suitable actions to fulfill any requirement or expectation expressed in the intent(s).

[0058] As shown in FIG. 1, the IMF 104 may include an Intent Management Framework (IMFr) 120 and a plurality of agents such as a data grounding agent 122, a proposal agent 124, a prediction agent 126, an evaluation agent 128 and an actuator agent 130. The IMFr may include a reasoner 132 and a knowledge base 134.

[0059] After the IMF 104 receives an intent from the service consumer NF 102, the reasoner 132 may extract knowledge objects from the intent and store the knowledge objects in the knowledge base 133. Depending upon the type of the intent, e.g., fulfillment or assurance, the reasoner 132 may define a goal to be reached. The IMFr 120 may collect solution proposals from the proposal agent 124, evaluate one or more non-conflicting proposals that may achieve the goal by using the evaluation agent 128, and actuate the proposal(s) to trigger one or more actions through the actuator agent 130.

[0060] In some embodiments, once the intent is fulfilled, the IMF 104 may automatically remove the intent.

[0061] In some embodiments, the IMF 104 may not transfer to the service consumer NF 102 any information about the type of action(s) triggered by the IMF 104. Instead, the IMF 104 may only inform the service consumer NF 102 that a relevant action or actions to fulfill the intent have been triggered.

[0062] The IMF may handle conflict resolution. If the IMF receives multiple intents of identical goals from different network entities, the IMF may prevent redundant actions and apply only the necessary action(s) to reach the common goal. The IMF may detect the redundant actions when evaluating actions proposed by the proposal agent 124. Example redundant actions are provided below.

[0063] In one example, the service consumer NF 102 may receive an analytics report from the NWDAF 106 predicting that the service consumer NF will be overloaded. The service consumer NF may need more computational resources, e.g., CPU and memory (vertical scaling). At the same time, an NF orchestrator may realize that the load on the service consumer NF is increasing, so that the NF orchestrator may start to instantiate new instances (horizontal scaling). The service consumer NF 102 and the NF orchestrator may submit their respective intents to the IMF 104. The IMF 104 may trigger actions to satisfy only one intent, while avoiding triggering similar actions that would result in a waste of resources, e.g, vertical and horizontal scaling of computational resources, and/or using or creating new PDU sessions, among other possibilities.

[0064] In another example, the service consumer NF 102 may be a PCF, which may receive a prediction from the NWDAF 106 about imminent degradation in quality of service (QoS), e.g., degradation may occur in 5 minutes. As a result, the PCF may start allocating more bandwidth by selecting and executing a policy. At the same time, an application function (AF) may receive the same prediction from the NWDAF 106 when requesting an analytics report, and may start to use a code with lower bandwidth demands. The PCF and the AF may submit their respective intents to the IMF 104. The IMF 104 may prevent triggering redundant actions with similar effects. For example, the IMF 104 may prevent triggering redundant actions that would affect the same KPIs.

1.3 Network Data Analytics Function (NWDAF)

[0065] The NWDAF 106 may receive, from the service consumer NF 102, a request for an analytics report regarding a network state. The request may include an analytics identifier. Example analytics identifiers may include UE mobility, user data congestion, and network performance, among other possibilities. The NWDAF 106 may output the analytics report to the service consumer NF 102 based on the request. The analytics report may include statistics from the past and/or predictions for the future. For instance, a service consumer NF 102 may ask for UE mobility for a specific UE or a group of UEs within a certain period of time. The NWDAF 106 may respond with a predicted UE trajectory within that period of time.

[0066] The NWDAF 106 may receive an intent from the service consumer NF 102. The intent may represent a network state desired by the service consumer NF. The intent received from the service consumer NF 102 may be based on the analytics report. The NWDAF 106 may receive the intent from the service consumer NF as feedback to the analytics report.

[0067] The NWDAF may determine one or more KPIs from the intent received from the service consumer NF 102. The KPI(s) may be associated with the network state. The KPI(s) may be associated with the analytics report generated by the NWDAF 106. The KPI(s) may be affected by the action(s) that serve to fulfill the intent of the service consumer NF 102. After determining the KPI(s), the NWDAF may monitor the KPI(s) value to determine whether one or more actions relevant to the analytics report have been triggered. By monitoring the KPI(s) and collecting information about the KPI(s), the NWDAF 106 may realize whether any relevant action(s) has been triggered by the service consumer NF 102 or its associated IMF 104 to fulfill the intent of the service consumer NF 102.

[0068] For example, the NWDAF 106 may determine whether and how the triggered action(s) may have any impact on the KPI(s) that are associated with the analytics report generated by NWDAF.

[0069] The NWDAF 106 may remove the intent when not used any more for monitoring purposes. The NWDAF 106 may receive, from the service consumer NF, 102 at least one criterion for removing the intent. [0070] Referring to FIG. 1, the NWDAF 106 may include one or more of the following: a trimmed version of Intent Management Function (tIMF) 140, model training logical function (MTLF) 142, and analytics logical function (AnLF) 144, among others.

1.3.1 tIMF

[0071] The tIMF 140 may serve to handle all intents received by the NWDAF 106. The tIMF 140 may include an IMFr 146, and one or more agents such as a proposal agent 148, a prediction agent 150 and an evaluation agent 152, among other possibilities.

[0072] The tIMF 140 may process the intent received from the service consumer NF 102. The tIMF 140 may extract KPI(s) that may be affected when an action(s) to fulfill the intent is triggered. The KPI(s) may correlate to requirements/expectations in the intent. For example, the tIMF 140 may extract a state vector which contains information about the KPI(s) that may be affected when the intent is fulfilled.

[0073] The tIMF 140 may not propose or trigger actions.

1.3.2 MTLF and ANLF

[0074] The MTLF 142 may train and provide a first ML model for generating an analytics report. When a request for an analytics reports is received, the AnLF 144 may perform inference on the first ML model to generate the analytics report.

1.3.3 Other Functions

1.3.3.1 Measure Accuracy of Analytics Reports

[0075] The NWDAF 106 may monitor the KPI(s) extracted by the tIMF 140 to determine whether any action(s) relevant to the generated analytics report has been triggered. By monitoring the KPI(s), the NWDAF may determine whether any action has been triggered by the service consumer NF 102 or its associated IMF 104 without the need to know any detail about the action (e.g., the type of the action). Knowing whether any action has been triggered may help the NWDAF 106 measure the accuracy of the generated analytics report. For example, the NWDAF 106 may measure the accuracy of the analytics report based on determining whether any action relevant to the generated analytics report has been triggered.

[0076] To measure the accuracy of the analytics report, the NWDAF may measure any deviation between predicted values of the analytics report and the ground truth, and assess whether any triggered action has an impact on the prediction made by the analytics report. For instance, the service consumer NF 102 may be a PCF which asks for an analytics report regarding user data congestion. If the NWDAF 106 predicts that congestion will happen, the PCF may trigger one or more actions to prevent the congestion. Later, the NWDAF 106 may collect data to measure if the prediction is accurate. The ground truth may show no congestion. The feedback, e.g., intent, from the PCF may let the NWDAF 106 know that one or more actions have been taken. Although there is a difference between the prediction made by the analytics report and the ground truth, the NWDAF may conclude, based on the feedback from the PCF, that the NWDAF ’s prediction about the congestion is correct, and the congestion has been prevented because of the action(s) taken by the PCF.

1.3.3.2 Train and Use a Second ML Model

[0077] The NWDAF 106 may train a second ML model to predict a network state desired by a service consumer NF. The second ML model may also be referred to as a recommendation intents ML (RIM) model. When receiving a request for an analytics report from the service consumer NF 102, the NWDAF 106 may rely on the RIM model to predict the network state desired by the service consumer NF 102, and generate one or more recommended intents to the service consumer NF 102. The NWDAF 106 may provide the recommendation to the service consumer NF 102 proactively, without receiving any intent from the service consumer NF 102. [0078] In one example, the NWDAF 106 may train the RIM model for analytics IDs and/or service consumer NF types. [0079] In one example, the NWDAF 106 may train the RIM model to predict a desired, expected, required and/or achievable network state for a service consumer NF of a specific type, e.g., access and mobility management function (AMF), PCF, and session management function (SMF), among others. When a service consumer NF of a specific type submits an intent, the NWDAF 106 may train the RIM model, based on the submitted intent, to predict the expected network state for the specific NF type. In particular, the NWDAF 106 may train the RIM model by using collected monitoring data when the intent is submitted, and/or when actions are triggered. The training data may include one or more of the following: (1) KPI values in the intent submitted by the service consumer NF before actions are applied, (2) KPI values in the intent submitted by the NF after actions are applied until the intent is removed, (3) NF type, e.g., PCF, AMF, SMF, among others, and (4) extracted requirements and expectations in the submitted intent. Regarding (2), the KPI values may be used regardless whether the intent is fulfilled or not.

[0080] When receiving a request for an analytics report from a service consumer NF of a specific type, the NWDAF 106 may use the trained RIM model to predict a future state of the network that is expected by the service consumer NF that has requested the analytics report. The predicted future network state may be a network state desired by the service consumer NF. [0081] To perform the prediction, the trained second ML model may be fed with the current KPI values, timestamp, and NF type of the service consumer NF. The trained RIM model may predict a value of at least one KPI that can be affected by at least one action that is triggered after the service consumer NF receives the analytics report. The predicted KPI value may describe the network state desired by the service consumer NF.

[0082] In one embodiment, the KPI value(s) predicted by the RIM model may be different from the prediction made in the analytics report that is generated by the first ML model, even if the values for the same KPI(s) are predicted. The KPI value(s) in the analytics report may be predicted based on the previous network state, while the KPI value(s) predicted by the RIM model may be predicted based on the previous behavior of the service consumer NFs of the same type.

[0083] Based on the prediction from the RIM model, the NWDAF 106 may create intents which describe the predicted network state in the form of expectation/requirements. For example, the NWDAF 106 may generate one or more recommendation intents that describe the predicted network state desired by the service consumer NF. The NWDAF 106 may utilize its own internal logic to compose the recommendation intent(s) based on the predicted KPI value(s). The NWDAF may send the recommendation intents to the service consumer NF. The service consumer NF may optionally choose the best intent from the recommendation intents. Alternatively, the service consumer NF may discard all recommendation intents and create a new intent and submit it to its associated IMF.

[0084] By providing recommendation intents to the service consumer NF, the NWDAF avoids proposing any action to the service consumer NF for attaining the desired network state.

[0085] In one embodiment, training and inference on the RIM model may be performed by the MTLF 142 and the AnLF 144, respectively, or by one or more dedicated logical functions. For example, the MTLF 142 may train and/or generate the RIM model, while the AnLF 144 may use or perform inference on the RIM model. In another example, one or more dedicated logical functions other than the MTLF 142 and the AnLF 144 may train and use the RIM model.

2. Example Methods of Operations

[0086] FIG. 2 illustrates a sequence diagram related to providing feedback by the service consumer NF 102 to the NWDAF 106. Dashed lines may represent optional steps.

[0087] At step 202, the service consumer NF 102 may send a request to the NWDAF 106 with an analytics ID. [0088] At step 204, the NWDAF 106 may generate an analytics report and deliver it to the service consumer NF 102.

[0089] At step 206, the service consumer NF 102 may optionally generate an intent which describes expectations and/or requirements for the network state that shall be fulfilled, and send the intent to the NWDAF 106.

[0090] At step 208, the service consumer NF 102 may optionally send the intent to the IMF 104 to set a goal. The IMF 104 may in turn trigger one or more actions to fulfill the intent.

[0091] At steps 206 and/or 208, the service consumer NF may specify criteria to remove the intent if it is not fulfilled. Such criteria may include one or more of the following: timeout, network situation, and KPI value(s), among other possibilities. If the intent is fulfilled, the intent may be automatically removed by the IMF 104. The tIMF 140 within the NWDAF 106 may, based on its internal logic, remove the intent when not used anymore for monitoring purposes.

[0092] At step 210, the IMF 104 may optionally inform the service consumer NF 102 that one or more actions have been triggered. The IMF 104 may only inform the service consumer NF that a relevant action or actions to fulfil the intent are triggered. No details of the triggered action(s) may be transferred from the IMF 104 to the service consumer NF 102.

[0093] At 212, the NWDAF 106 may optionally process the intent and extract a state vector. The state vector may contain one or more KPIs to be collected. The KPIs may be affected when an action or actions are triggered to fulfill the intent. Steps 206 and 208 may be repeated until the service consumer NF 102 is satisfied with the network state. The service consumer NF 102 may re-submit intents to the NWDAF 106 and the IMF 104 in a loop.

[0094] At 214, the NWDAF 106 may optionally monitor one or more KPIs extracted from the intent submitted by the service consumer NF 102. By monitoring the KPI(s), the NWDAF 106 may determine if any relevant action or actions have been triggered that can affect the prediction made in the generated analytics report. By monitoring the KPI(s), the NWDAF may realize if any relevant action(s) has had a positive or negative impact with respect to the prediction made in the analytics report. The results of monitoring may be collected by the NWDAF 106 for further training the first ML model.

[0095] At step 216, the service consumer NF 102 may optionally inform the NWDAF 106 that an action or actions have been triggered. The action(s) may be relevant to the prediction made in the generated analytics report. Similar to step 210, no information about the type of action(s) may be transferred from the service consumer NF 102 to the NWDAF 106, as it is not feasible for the NWDAF to keep information about all possible actions in the network. The service consumer NF 102 may only inform the NWDAF 106 that a relevant action or actions to fulfil the intent are triggered. The NWDAF 106 may not know the type of action(s).

[0096] For instance, the service consumer NF 102 may be a UPF which asks the NWDAF 106 for prediction about UE communication pattern for a group of UEs. The NWDAF 106 may send the prediction in the form of bandwidth usage, applications, quality, and data rate, among others. Then, the UPF may send feedback to the NWDAF 106, and inform the NWDAF 106 that some actions that have been triggered. The NWDAF may not know details about the actions, e.g., whether the UPF has asked the PCF to update policies, has asked the SMF to create a new PDU session, or has modified resource allocation in edge servers. All these actions may be triggered and may have different impacts on KPIs which would affect the way that the NWDAF 106 measures the accuracy of the prediction made in the generated analytics report. Based on the present technology, the UPF may submit an intent to the NWDAF, and then the NWDAF may know that the UPF is interested in updating a set of KPI values. The NWDAF may not know what type of actions are triggered. To monitor accuracy of predictions made in the analytics report, it would be enough for the NWDAF to monitor KPI values extracted from the intent and then determine how relevant actions have changed KPI values. [0097] At 218, if the intent is not fulfilled and the defined criteria for removing the intent are met, then the service consumer NF 102 may optionally ask the NWDAF 106 to remove the intent. In turn, the tIMF 140 within the NWDAF 106 may remove the intent.

[0098] At 220, if the service consumer NF 102 has sent an intent removal request to the NWDAF 106, then the service consumer NF 102 may optionally send the same request to the IMF 104.

[0099] At 222, the NWDAF 106 may optionally use the monitoring data collected from step 214 to train the RIM model. The RIM model may be used to recommend intents to a service consumer NF, which may be the same service consumer NF that submits the request at step 202, or may be a different service consumer NF.

[0100] FIG. 3 illustrates a flow diagram corresponding to the sequence diagram of FIG. 2. At 302, the service consumer NF 102 may send a request to the NWDAF 106 asking for an analytics report. At step 304, the NWDAF 106 may send the analytics report with one or more predictions to the service consumer NF 102. The service consumer NF 102 may generate an intent which describes expectations and/or requirements for the network state that shall be fulfilled. At 306, the service consumer NF 102 may submit the intent to the IMF 104 and the NWDAF 106. The service consumer NF 102 may specify criteria for removing the intent if it is not fulfilled.

[0101] At step 308, the NWDAF 106 may process the intent and extract one or more relevant KPIs using the tIMF 140.

[0102] At 310, the NWDAF 106 may monitor values of the extracted KPI(s).

[0103] At 312, the service consumer NF 102 may determine if any action has been triggered. For example, the IMF 104 may inform the service consumer NF 102 that an action or actions have been triggered. [0104] At 314, if at least one action relevant to the previously requested analytics report has been triggered, the service consumer NF 102 may inform the NWDAF 106 about the triggered action. The NWDAF 106 may monitor the KPI value(s) after the action(s) has been triggered.

[0105] At 316, if no action has been triggered, the service consumer NF 102 may assess whether to remove the intent. If the service consumer NF 102 decides not to remove the intent, then the NWDAF 106 may continue monitoring the KPI values.

[0106] At 318, the NWDAF 106 may use the monitoring data collected at step 310 to train the RIM model offline.

[0107] FIG. 4 illustrates a sequence diagram with respect to providing recommendation intents by the NWDAF 106 to the service consumer NF 102. Dashed lines may indicate optional steps. At 402, the service consumer NF 102 may send a request to the NWDAF 106 with an analytics ID.

[0108] At 404, the NWDAF 106 may generate an analytics report and deliver it to the service consumer NF 102.

[0109] At 406, the NWDAF 106 may optionally use the trained RIM model (step 222 of FIG 2 or step 318 of FIG. 3) to predict one or more KPI value(s) that can be suitable for the service consumer NF 102.

[0110] At 408, based on the predicted KPI value(s), the NWDAF 106 may optionally create one or more intents to be sent to the service consumer NF as recommendation intents. The recommendation intents may reflect the prediction made in step 406. The recommendation intents may be composed based on the internal logic of the NWDAF 106.

[oni] At 410, the NWDAF 106 may optionally send the recommendation intents to the service consumer NF 102. The service consumer NF 102 may either choose the best intent from the recommendation intents or discard all of the recommendation intents. [0112] At 412, the service consumer NF 102 may optionally choose the best intent in the recommendation intents based on its own internal logic.

[0113] At 414, the service consumer NF 102 may send the chosen intent to the NWDAF 106. [0114] At 416, the service consumer NF may send the intent to the IMF 104 to set a goal. In turn, the IMF 104 may trigger one or more actions to fulfill the intent. At both steps 414 and 416, the service consumer NF 102 may specify criteria to remove the intent if it is not fulfilled. Such criteria may include one or more of the following: timeout, network situation, and KPI values, among others. If the intent is fulfilled, the intent may be automatically removed by the IMF 104. The NWDAF 106, e.g., the tIMF 140 within the NWDAF 106, may exercise its internal logic to remove the intent when not used any more for monitoring purposes.

[0115] At 418, the IMF 104 may optionally inform the service consumer NF 102 that one or more actions have been triggered. No information about the type of actions may be transferred from the IMF 104 to the service consumer NF 102. The IMF 104 may only inform the service consumer NF that relevant actions to fulfil the intent are triggered.

[0116] At 420, the service consumer NF 102 may inform the NWDAF 106 that one or more actions have been triggered. No information about the type of actions may be transferred from the service consumer NF 102 to the NWDAF 106. The service consumer NF 102 may only inform the NWDAF 106 that a relevant action or actions to fulfil the intent are triggered.

[0117] At 422, the NWDAF 106 may optionally process the intent and extract a state vector. The state vector may contain one or more KPIs that may be affected by one or more actions that fulfill the intent.

[0118] At 424, the NWDAF 106 may optionally monitor one or more KPIs extracted from the intent submitted by the service consumer NF 102. By monitoring the KPI(s), the NWDAF 106 may determine if one or more relevant actions have been triggered that can affect the prediction made in the analytics report provided by the NWDAF 106. The results of monitoring may be collected by the NWDAF 106 for further training at step 430.

[0119] At 426, if the intent is not fulfilled and the defined criteria for removing the intent are met, then the service consumer NF 102 may send an intent removal request to the NWDAF 106, asking the NWDAF 106, e. g. , the tIMF 140 within the NWDAF 106, to remove the intent. [0120] At 428, if the service consumer NF 102 has sent the intent removal request to the NWDAF 106 at step 426, the service consumer NF 102 may optionally send the same request to the IMF 104.

[0121] At 430, the NWDAF 106 may use the monitoring data collected at step 424 to train the RIM model. The RIM model may be used to recommend intents to the service consumer NF 102.

[0122] FIG. 5 illustrates a flow diagram corresponding to the sequence diagram of FIG. 4. At 502, the service consumer NF 102 may send a request to the NWDAF 106 asking for an analytics report.

[0123] At 504, the NWDAF 106 may generate the analytics report and deliver it to the service consumer NF 102.

[0124] At 506, the NWDAF 106 may use the trained RIM model to predict one or more KPI values that are desired by the service consumer NF 102.

[0125] At 508, based on the predicted KPI value(s), the NWDAF 106 may generate one or more recommendation intents, and send the recommendation intents to the service consumer NF 102.

[0126] At 510, the service consumer NF 102 may exercise its own internal logic to choose the best intent from the recommendation intents. [0127] At 512, the service consumer NF 102 may submit the chosen intent to the IMF 104 and the NWDAF 106. The service consumer NF 102 may specify criteria to remove the intent if the intent is not fulfilled.

[0128] At 514, the NWDAF 106 may extract one or more relevant KPIs from the intent by using the tIMF 140. The KPIs may be affected by one or more actions that fulfill the intent.

[0129] At 516, the NWDAF 106 may monitor values of the KPIs extracted from the intent. [0130] At 518, the service consumer NF 102 may determine if any action has been triggered. For example, the IMF 104 may inform the service consumer NF 102 that an action or actions have been triggered.

[0131] At 520, if at least one action relevant to the previously requested analytics report has been triggered, the service consumer NF 102 may inform the NWDAF about the triggered action. The NWDAF 106 may monitor the KPI value(s) after the action(s) has been triggered.

[0132] At 522, if no action has been triggered, the service consumer NF 102 may assess whether to remove the intent. If the service consumer NF 102 decides not to remove the intent, then the NWDAF 106 may continue monitoring the KPI values.

[0133] At 524, the NWDAF 106 may use the monitoring data collected at step 516 to train the RIM model offline.

[0134] FIG. 6 is a flow chart illustrating an example method performed by one or more nodes in a communications network core. The node(s) may include an NWDAF 106.

[0135] At 602, the node(s) may receive, from a service consumer NF 102, a request for an analytics report regarding a network state.

[0136] At 604, the node(s) may receive an intent from the service consumer NF 102 representing a desired network state.

[0137] At 606, the node(s) may determine, from the received intent, at least one key performance indicator (KPI) associated with the network state. [0138] At 608, responsive to said determining, the node(s) may monitor a value of the at least one KPI to determine whether at least one action to fulfill the received intent has been triggered. [0139] FIG. 7 is a flow chart illustrating an example method performed by one or more nodes in a communications network core. The node(s) may include an NWDAF 106.

[0140] At 702, the node(s) may receive, from a service consumer NF, a request for an analytics report regarding a network state.

[0141] At 704, the node(s) may output the analytics report to the service consumer NF 102 based on the request.

[0142] At 706, the node(s) may receive an intent from the service consumer NF 102 as feedback to the analytics report.

[0143] At 708, the node(s) may detect whether an action relevant to the generated analytics report has been triggered.

[0144] FIG. 8 is a flow chart illustrating an example method performed by one or more nodes in a communications network core. The node(s) may include an NWDAF 106.

[0145] At 802, the node(s) may receive, from a service consumer NF 102, a request for an analytics report regarding a network state.

[0146] At 804, an ML model, e.g., the RIM model, may predict a value of at least one key performance indicator (KPI) that can be affected by at least one action that is triggered after the service consumer NF 102 receives the analytics report. The predicted value may describe a network state desired by the service consumer NF 102.

[0147] At 806, the node(s) may generate a recommendation intent describing the predicted network state desired by the service consumer NF 102.

[0148] At 808, the node(s) may send the recommendation intent to the service consumer NF

102. [0149] FIG. 9 is a flow chart illustrating an example method performed by a service consumer NF 102. At 902, the service consumer NF 102 may send, to an NWDAF 106, a request for an analytics report regarding a network state.

[0150] At 904, the service consumer NF 102 may receive the analytics report from the NWDAF 106.

[0151] At 906, the service consumer NF 102 may determine, based on the analytics report, an intent representing a desired network state.

[0152] At 908, the service consumer NF 102 may send the intent to at least one of an IMF 104 and the NWDAF 106.

3. Use Case Examples

3.1 First Use Case Example

[0153] In a first use case example, the service consumer NF 102 may be a PCF. The PCF may send a request to the NWDAF 106 for an analytics report regarding user data congestion. The analytics report may predict whether the user data congestion would occur.

[0154] Table 1 shows different situations of predictions by the NWDAF and the ground truth. The first column shows an index for each situation. The second column shows predictions by the NWDAF 106 with respect to whether user data congestion would occur. The third column shows the ground truth, namely, whether use data congestion happened. The fourth column shows whether the prediction made by the NWDAF 106 is accurate in view of the ground truth.

Table 1 - NWDAF prediction and ground truth [0155] In a first scenario, as indicated in the first row of Table 1, the NWDAF 106 (e.g., the first ML model of the NWDAF) predicted that there would be congestion. The ground truth indicates that congestion happened. Since the NWDAF predicted correctly, there is no need to update its first ML model.

[0156] In a second scenario, as indicated in the second row of Table 1, the NWDAF 106 predicted that there would be congestion. The ground truth indicates that congestion did not happen. Without any feedback from the PCF on whether any remedial action was executed by the PCF to prevent the congestion, the NWDAF 106 would not know whether in this situation, the prediction by the first ML model was accurate. Without any feedback from the PCF, the NWDAF would not know whether to update the first ML model. The present technology resolves this dilemma by enabling feedback from the PCF, wherein the feedback indicates whether the PCF took any remedial action to prevent congestion. Further discussion is provided in Table 2 below.

[0157] In a third scenario, as indicated in the third row of Table 1, the NWDAF predicted that there would be no congestion. The ground truth indicates that congestion happened. In this case, the PCF did not take any remedial action because of the prediction made by the NWDAF. Clearly, the prediction made by the ML model is wrong. As a result, the NWDAF needs to update its first ML model.

[0158] In a fourth scenario, as indicated in the fourth row of Table 1, the NWDAF 106 predicted that there would be no congestion. The ground truth indicates that congestion did not happen. The PDF did not take any action as a result of the NWDAF’s prediction. Here, the prediction is correct. As a result, the NWDAF 106 does not need to update its first ML model. [0159] The present technology enables the PCF to provide feedback to the NWDAF 106 indicating whether any action has been taken. Table 2 distinguishes between two situations by using feedback from the PCF. In Table 2, the first column shows predictions by the NWDAF 106 with respect to whether user data congestion would occur. The second column shows whether the PCF took any action. The third column shows the ground truth, namely, whether use data congestion happened. The fourth column shows whether the prediction made by the NWDAF was correct.

Table 2 - Refining the second scenario in Table 1 using feedback from the PCF

[0160] In a first situation as shown in the first row of Table 2, the NWDAF predicted congestion, and the PCF took a remedial action to prevent the congestion. As a result, the prediction made by the NWDAF was correct. There is no need to update the first ML model in the NWDAF 106.

[0161] In a second situation as shown in the second row of Table 2, the NWDAF 106predicted congestion, and the PCF did not take any remedial action. The congestion did not happen, implying that the prediction by the first ML model of the NWDAF was wrong. As a result, the NWDAF should update the first ML model.

[0162] Thus, by using feedback mechanism of the present technology, the NWDAF 106 knows if the PCF has taken relevant actions. As part of the feedback mechanism, the PCF provides its intent to the NWDAF. By monitoring one KPI(s) related to the intent, the NWDAF may determine if any change in the KPIs has been happened. This information may be used internally by the NWDAF 106 to assess the accuracy of the generated analytics reports. If the values predicted in the analytics report are different from the collected ground truth, the NWDAF 106 or the MTLF 142 of the NWDAF 106 may retrain the first ML model, for example, by using different data sources to collect different training data. 3.2 Second Use Case Example

[0163] In a second use case example, a service consumer NF 102 may ask an NWDAF 106 to predict if the service consumer NF will be overloaded or if there will be any degradation in the network performance. If the NWDAF 106 predicts overload in the service consumer NF and/ or predicts degradation in the network performance, then the RIM model of the NWDAF 106 may create a list of intents. The NWDAF 106 may send the list of intents to the service consumer NF 102 as recommendations.

[0164] The service consumer NF 102 may then choose the best intent from the recommendation list, and submit the chosen intent to the NWDAF 106 and the IMF 104. Alternatively, the service consumer NF 102 may discard the recommendation list, and create a new intent and submit the new intent to the NWDAF 106 and the IMF 104. Once the IMF 104 receives the intent from the service consumer NF 102, the IMF 104 may trigger one or more actions to allocate more resources to prevent network performance degradation. The IMF 104 may trigger an action or actions to satisfy only one intent, while avoiding triggering similar actions that would result in a waste of resources, e.g., vertical, and horizontal scaling of computational resources, and/or using or creating new PDU sessions, among other possibilities. The IMF 104 may also be configured to resolve possible conflicts between actions (not necessarily redundant actions) to satisfy different intents.

4. Example Components of Entities

[0165] FIGS. 10A-10C illustrate example components of the service consumer NF 102, the IMF 104, and the NWDAF 106, respectively. Each may include processing circuitry 1002a, 1002b and 1002c communicatively coupled with memory 1004a, 1004b and 1004c, respectively. The memory 1004a, 1004b and 1004c may store instructions executable by the processing circuitry 1002a, 1002b and 1002c to perform methods described in this disclosure. [0166] The processing circuitry 1002a, 1002b and 1002c may be configured to process instructions and data and may be configured to implement any sequential state machine operative to execute instructions stored as machine-readable computer programs in the memory 1004a, 1004b and 1004c. The processing circuitry may be implemented as one or more hardware-implemented state machines (e.g., in discrete logic, field-programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), etc.); programmable logic together with appropriate firmware; one or more stored computer programs, general -purpose processors, such as a microprocessor or digital signal processor (DSP), together with appropriate software; or any combination of the above. For example, the processing circuitry 1002a, 1002b and 1002c may include multiple central processing units (CPUs).

[0167] The memory 1004a, 1004b and 1004c may be or be configured to include memory such as random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic disks, optical disks, hard disks, removable cartridges, flash drives, and so forth. In one example, the memory 1004a, 1004b and 1004c includes one or more application programs, such as an operating system, web browser application, a widget, gadget engine, or other application, and corresponding data. The memory 1004a, 1004b and 1004c may store any of a variety of various operating systems or combinations of operating systems.

[0168] In general, the various exemplary embodiments may be implemented in hardware or special purpose chips, circuits, software, logic or any combination thereof. For example, some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software which may be executed by a controller, microprocessor or other computing device, although the disclosure is not limited thereto. While various aspects of the exemplary embodiments of this disclosure may be illustrated and described as block diagrams, flow charts, or using some other pictorial representation, it is well understood that these blocks, apparatus, systems, techniques or methods described herein may be implemented in, as non-limiting examples, hardware, software, firmware, special purpose circuits or logic, general purpose hardware or controller or other computing devices, or some combination thereof.

[0169] Various modifications and adaptations to the foregoing exemplary embodiments of this disclosure may become apparent to those skilled in the relevant arts in view of the foregoing description, when read in conjunction with the accompanying drawings. However, any and all modifications will still fall within the scope of the non-limiting and exemplary embodiments of this disclosure.

Claims

1. A method for intent management performed by one or more nodes ( 106) in a communicati on s network core (100), the method comprising: receiving, from a service consumer network function (NF) (102), a request for an analytics report regarding a network state; receiving an intent from the service consumer NF (102) representing a desired network state; determining, from the received intent, at least one key performance indicator (KPI) associated with the network state; and responsive to said determining, monitoring a value of the at least one KPI to determine whether at least one action to fulfill the received intent has been triggered.

2. The method of claim 1, wherein the request includes an analytics identifier.

3. The method of any one of the preceding claims, further comprising: outputting the analytics report to the service consumer NF based on the request, wherein the intent received from the service consumer NF is based on the analytics report.

4. The method of any one of the preceding claims, further comprising: training a machine learning (ML) model to predict a network state desired by another service consumer NF, by using one or more of the following: the monitored value of the at least one KPI before the at least one action has been triggered; the monitored value of the at least one KPI after the at least one action has been triggered until the intent is removed; a type of the service consumer NF; and the desired network state represented by the received intent.

5. The method of any one of the preceding claims, further comprising: receiving, from the service consumer NF, at least one criterion for removing the intent.

6. The method of any one of the preceding claims, wherein the determining includes extracting the at least one KPI by a trimmed Intent Management Function (tIMF).

7. The method of any one of the preceding claims, wherein the one or more nodes include a network data analytics function (NWDAF).

8. A method for intent management performed by one or more nodes (106) in a communications network core (100), the method comprising: receiving, from a service consumer network function (NF) (102), a request for an analytics report regarding a network state; outputting the analytics report to the service consumer NF (102) based on the request; receiving an intent from the service consumer NF (102) as feedback to the analytics report; and detecting whether an action relevant to the generated analytics report has been triggered.

9. The method of claim 8, further comprising: determining, from the received intent, at least one key performance indicator (KPI) that can be affected by the action; and monitoring a value of the at least one KPI to detect whether the action has been triggered.

10. A method for intent management performed by one or more nodes (106) in a communications network core (100), the method comprising: receiving, from a service consumer network function (NF) (102), a request for an analytics report regarding a network state; predicting, by a machine learning (ML) model, a value of at least one key performance indicator (KPI) that can be affected by at least one action that is triggered after the service consumer NF (102) receives the analytics report, the predicted value describing a network state desired by the service consumer NF (102); generating a recommendation intent describing the predicted network state desired by the service consumer NF (102); and sending the recommendation intent to the service consumer NF (102).

11. A method for intent management performed by a service consumer network function (NF) (102), the method comprising: sending, to a network data analytics function (NWDAF) (106), a request for an analytics report regarding a network state; receiving the analytics report from the NWDAF (106); determining, based on the analytics report, an intent representing a desired network state; and sending the intent to at least one of an intent management function (IMF) (104) and the NWDAF (106).

12. The method of claim 11, further comprising: receiving a notification from the IMF that at least one action to fulfill the intent has been triggered.

13. The method of any one of claims 11 to 12, further comprising: determining whether the desired network state is reached; and determining and submitting a new intent to the IMF and the NWDAF, when the desired network state is not reached.

14. A node in a communications network core comprising processing circuitry configured to perform the method according to any one of claims 1 to 7.

15. A node in a communications network core comprising processing circuitry configured to perform the method according to any one of claim 8 to 9.

16. A node in a communications network core comprising processing circuitry configured to perform the method according to claim 10.

17. A service consumer network function comprising processing circuitry configured to perform the method according to any one of claims 11 to 13.