Using analytics reports from 5gc as input to imf

Abstract

Methods, in or associated with a communications network, for recommending analytics data to a node or function in or operatively connected to the communications network. An example method comprises the steps of collecting (1410) network information comprising current and historical data for the state of the communications network and applying (1420) the network information and intent information defining or relating a goal for the communications network to a Machine Learning, ML, model, to identify an agent, node, or function relevant to the intent information and one or more corresponding analytics reports available from the communications network, for recommendation to the identified agent, node, or function. The example method further comprises sending (1430) the one or more corresponding analytics reports to the identified agent, node, or function.

Description

[0001] USING ANALYTICS REPORTS FROM 5GC AS INPUT TO IMF

[0002] TECHNICAL FIELD

[0003] This disclosure is generally related to communications networks and is more particularly related to the selection, distribution, and use of analytics data in such networks.

[0004] BACKGROUND

[0005] The 5G core network (5GC) currently under development and being deployed by members of the 3rd- Generation Partnership Project (3GPP) comprises several components called Network Functions (NF). Each NF oversees certain specific tasks in the network. One of these NFs is the Network Data Analytics Function (NWDAF), which is an NF that provides analytics reports that contain information about the operation of the 5GC to any other NF, upon request. These analytics reports can be used to take various actions in the network to improve user experience. An analytics report can be either predictions about the future state of the network or statistics from the previous network state and parameters.

[0006] The NWDAF supports a set of analytics IDs, where each analytics ID represent a certain type of analytics report, e.g., a User Data Congestion report, a UE Mobility report, an NF load report, etc. A detailed discussion of network data analytics and corresponding improvements to the 5G system to support network data analytics can be found in the 3GPP document 3GPP TS 23.288 v18.2.0, “Architecture enhancements for 5G System (5GS) to support network data analytics service,” 2023- 06.

[0007] One key application for the analytics data found in the analytics reports is as input to automated network functions. Increasingly, communications networks will include autonomous systems, which are systems capable of sensing their environment and operation and adapting their behavior accordingly, with little or no human input.

[0008] In the development and deployment of autonomous network functions, “intent”-based management is a key concept. As defined in TM Forum [IG1253], “Intent is the formal specification of all expectations including requirements, goals, and constraints given to a technical system.” An Intent Management Function (IMF) is thus an entity that operates in an autonomous system on intents. It can assume two roles (often simultaneously): intent owner, to send intents to other IMFs, and intent handler, which receives intents from other IMFs.

[0009] The system reads in requirements, constraints, and definition of network parameters / KPIs that describe the desired behavior of the network in the form of intents. However, an intent doesn’t have any information about how to reach the network state or behavior, but only expresses the requirements that should be fulfilled.

[0010] An example IMF might consist of two main components: 1 . Reasoner: Responsible for setting the goals to check whether a goal can be satisfied or not, by performing inference and executing the rules on a knowledge base.

[0011] 2. Knowledge Base: Stores knowledge objects (also known as facts) that are extracted from the intents and will be used by the reasoner. It also stores the data grounded from the underlying system and state of the environment.

[0012] Agents, together with the reasoner and knowledge base, form this example IMF. Agents in the IMF communicate with the reasoner by submitting rules. Rules are executed by the reasoner on the knowledge base, and then results are sent back to the agents. The agents handle different operations e.g., collection of data, action proposal to reach the goals, analysis of the proposed action to predict the impact on the network, etc.

[0013] SUMMARY

[0014] In the absence of comprehensive insight regarding the network’s behavior and state, the actions triggered by IMF to fulfill the requirements in an intent can have undesired and / or unexpected consequences. An IMF can benefit from the information that can be offered by 5GC, with the lack of such information leading to less efficient or, in the worst-case, inefficient actions proposed by the IMF’s agents. Consequently, this can cause inaccurate predictions of the impact of the proposed actions and as a result degradation in the network’s performance.

[0015] In an IMF, agents need information from different aspects of the network to be able to make correct decisions. For instance, action proposal agents would need information about different NFs operating in the network to know how an action can possibly affect other key performance indicators (KPIs) of network operation. A narrow view of the network that is limited to simple action-to-KPI mapping might not be enough to efficiently propose actions and predict the impact. Moreover, without having knowledge about upcoming events in the network, e.g., congestion, network performance, etc., agents might have incomplete knowledge for the tasks they are assigned to.

[0016] Similar problems may arise in the context of trace-related operations, i.e., the use of session, subscriber, and / or equipment traces to diagnose and debug network problems and / or to improve network performance.

[0017] The techniques, apparatuses, and systems described herein address these problems by providing, in some embodiments, a dedicated agent referred to herein as a 5GC Interface Agent (5GCIA), which might be part of an IMF itself, in some embodiments, or as part of a separate node and / or function, in others, acts as an interface between the various agents in an IMF and the 5GC, and in particular might act as an interface between the IMF and the NWDAF, or between a trace collection entity (TCE) or other entity dealing with trace-related information and the NWDAF, for example. More generally, this agent can operate as a selection or a recommendation entity that interacts with the agents in an intent-based function and the network. An example method, in or associated with a communications network, for recommending analytics data to a node or function in or operatively connected to the communications network comprises the steps of collecting network information comprising current and historical data for the state of the communications network and applying the network information and intent information defining or relating a goal for the communications network to a Machine Learning (ML) model, to identify a processing entity, e.g., an agent, node, or function, relevant to the intent information and one or more corresponding analytics reports available from the communications network, for recommendation to the identified processing entity. The example method further comprises sending the one or more corresponding analytics reports to the identified processing entity.

[0018] Another example method, in or associated with a communications network, for recommending analytics data to a node or function in or operatively connected to the communications network, comprises the steps of receiving an indication of one or more key performance indicators (KPIs) for the communications network and identifying, from a data structure mapping KPI(s) to analytics reports, based on the indication of one or more KPIs, one or more analytics reports available from the communication network. This example method further comprises sending, to the node or function, the identified analytics reports.

[0019] The methods summarized above, as well as network node apparatuses corresponding to these example methods, are described in detail below, along with several variations thereof.

[0020] The advantages provided by some embodiments of the techniques and apparatuses described herein include that:

[0021] 1 . Information about relevant network KPIs from the past (statistics) is delivered to the agents in an IMF, allowing these agents to better measure the effect of the proposed / evaluated actions.

[0022] 2. Information about relevant network KPIs in the future (predictions) is delivered to the agents in an IMF, to assist the agents in accurately proposing, evaluating, and predicting the impact of the proposed actions, accurately.

[0023] 3. An interactive scheme, involving the sending of network data and the receiving of scores indicating the relevance of the data, may be used to improve the 5GCIA’s ability to provide more relevant data to the agents.

[0024] 4. Machine-Learning (ML) models and KPI Analytics Reports (KAR) graphs or other data structures mapping KPIs to relevant analytics data may be used in this invention to predict / choose the most suitable analytics reports for an agent.

[0025] BRIEF DESCRIPTION OF THE FIGURES

[0026] Figure 1 illustrates an example Intent Management Function (IMF) with a 5GC, according to some embodiments.

[0027] Figure 2 shows an example anatomy of a 5GC Interface Agent (5GCIA) and interaction between IMF and 5GC. Figure 3 illustrates sources used by a Machine-Language Model Trainer (MMT) for training a Machine-Language (ML) model.

[0028] Figure 4 shows a ML model used by a 5GCIA to predict analytics reports for use for IMF agents.

[0029] Figure 5 is a sequence diagram for an example Analytics Reports Recommendation Unit (ARRU), according to some embodiments.

[0030] Figure 6 shows an example KPI Analytics Reports (KAR) graph.

[0031] Figure 7 is a sequence diagram for an example Analytics Reports Selection Unit (ARSU), according to some embodiments.

[0032] Figure 8 illustrates an ARSU use case, showing signaling according to some embodiments.

[0033] Figure 9 shows an example KAR used in the ARSU use case, omitting weights.

[0034] Figure 10 illustrates an example scenario in a network.

[0035] Figure 11 illustrates intent-driven MnS as in 3GPP TS 28.312.

[0036] Figure 12 shows an example deployment of a 5GCIA in a 3GPP network.

[0037] Figure 13 illustrates an example of a 5GCIA in collaboration with a Trace Collection Entity (TCE).

[0038] Figure 14 is a process flow diagram illustrating an example method according to some embodiments.

[0039] Figure 15 is a process flow diagram illustrating an example method according to some other embodiments.

[0040] Figure 16 is a block diagram illustrating an example network node, according to some embodiments.

[0041] DETAILED DESCRIPTION

[0042] The techniques, apparatuses, and systems described herein may be used to improve the accuracy and efficiency of proposal, prediction, and evaluation agents, by providing advanced insight about network state and performance to the agents through one or more of several mechanisms whereby agents are provided access to the most relevant analytics reports offered by an NWDAF in the core network.

[0043] According to some embodiments of the solutions detailed below, a dedicated agent referred to herein as a 5GC Interface Agent (5GCIA) is introduced. This 5GCIA, which may, of course, be labeled with a different name, operates between an IMF and the core network to provide analytics reports to any of proposal, prediction, or evaluation agents. As detailed below, the 5GCIA may benefit from an internal logical function to train an ML model. The trained ML model can then be used to predict which analytics reports can be useful for an agent to propose more efficient action, predict the impact of the actions more accurately, or perform more accurate evaluation over the proposed actions.

[0044] In some embodiments, the 5GCIA described herein can also accept requests from the agents, in the form of KPIs that an agent is interested in. The 5GCIA can go through a probability graph to translate the given KPIs to the information that can be offered by the most relevant analytics reports. Then, a proposal may be sent to the agent, which can optionally choose between the proposed analytics report or discard the proposal from 5GCIA.

[0045] The 5GCIA might be part of an IMF itself, in some embodiments, or may instead form part of a separate node and / or function, in others. In either case, the 5GCIA may act as an interface between the various agents in an IMF and the 5GC. For instance, the 5GCIA might act as an interface between the IMF and the NWDAF, or as an interface between a trace collection entity (TCE) or other entity dealing with trace-related information and the NWAF, in various examples. More generally, this agent can operate as a selection or a recommendation entity that interacts with the agents in an intentbased function and the network.

[0046] The architecture of one embodiment of the proposed solution is shown in Figure 1 , where a 5GCIA 120 in this example forms part of an IMF, communicating with other agents of the IMF as well as with the 5GC 110.

[0047] In this solution, a Network Data Analytics Function (NWDAF), which is an NF in the 5GC 110, may be utilized to retrieve information that is supposed to be used by the agents in the IMF 100 to (1) make more accurate predictions, (2) perform more precise evaluations, and (3) propose more effective actions. In the proposed scheme, the 5GCIA 120 is the service consumer to the NWDAF, receiving analytics reports from the NWDAF and then delivering them to the agents. Generally, any NF can be considered as a source of data to 5GCIA. For the purpose of describing details of the techniques herein, the NWDAF is assumed to be used to get information about network’s behavior.

[0048] A feedback mechanism may also be introduced to this scheme, whereby agents provide feedback to 5GCIA about the received analytics reports, e.g., a “score” or other indication of the relevance of a given report or group of reports to the agents’ activities.

[0049] Agents may communicate with the 5GCIA 120 in the form of Key Performance Indicators (KPIs) or other metrics related to network operation in which they are interested. The 5GCIA must then decide which analytics reports can be beneficial to an agent with respect to the KPIs or other parameters that are specified as important. As briefly noted above, feedback may be provided by the agents to indicate the relevance of analytics data / reports that are provided to the agents according to this scheme. This can be implemented, for example, as a scoring system whereby agents provide a “score” for each of the received analytics reports, where the score indicates a given agent’s determination of the relevance between its KPIs of interest and the received analytics reports. The 5GCIA may use these scores to improve the process of selecting / recommending analytics reports and delivering them to the agents.

[0050] Figure 2 illustrates an example representation of a 5GCIA 120, communicating with an IMF 100 and a 5GC 110. In this illustration, the 5GCIA 120 is external to the IMF 100, but it may be part of the IMF 100 in some embodiments.

[0051] The instance of 5GCIA 120 shown in Figure 2 comprises the following components:

[0052] 1. ML Model Trainer (MMT) 210

[0053] 2. Analytics Reports Recommendation Unit (ARRU) 220

[0054] 3. Analytics Reports Selection Unit (ARSU) 2230

[0055] The 5GCIA 120 trains a ML model internally. This model is then used by the ARRU 220. The data collection for training and ML model provisioning is done by the MMT 210. The ARRU 220 uses the ML model to perform inference and to generate recommendations to be sent to the agents in an IMF. After training is completed and the ML model is in use, the ARRU is responsible to detect any possible drift in an ML model and then inform MMT to re-train or prepare another model.

[0056] Information to be delivered to an agent is retrieved from the network e.g., from the NWDAF in 5GC 110. A discovery mechanism, e.g., as defined by 3GPP specifications, is used to query and find an instance of NWDAF to send requests to. The Network Repository Function (NRF) is utilized to collect information about an instance of NWDAF since several instances of each NF can co-exist in the network.

[0057] The ARSU 230 receives KPIs (and / or other network parameters, metrics, etc.) that the agents are interested in. The ARSU 230 internally decides which analytics reports might give the most benefits to the agents who have submitted the request. As a whole, the 5GCIA 120 tries to find the most relevant analytics reports that can help the agents regarding the submitted KPIs.

[0058] In the following subsections, each of the 5GCIA components shown in Figure 2 is described in more detail.

[0059] First is the MMT 210. The MMT 210 collects data from various data sources in the network to train an ML model that is then used to predict the recommendation list of analytics reports. ML model training can be realized, for example, by using free open-source libraries like for example PyTorch.

[0060] A local storage is used to store collected data and / or trained ML models. Training data can be collected from various data sources, including:

[0061] 1 . Network parameters from the core network, e.g., network parameters, KPIs, historical data about network state such as congestion, network performance, slice level information, etc.

[0062] 2. The scores that have been given to previous predictions to improve the accuracy of an ML model.

[0063] 3. Mapping between an agent type and a list of relevant analytics reports. 4. Goals that are set by goal setting agents. The goals are retrieved from knowledge base.

[0064] After a ML model is trained, it is provisioned to ARRU 220 for inference. Models can be stored in a local storage in the IMF’s domain, for example, for downloading by the ARRU 220. Other storage units may be used instead, e.g., an Analytics Data Repository Function (ADRF) inside 5GC to store ML models, training data, etc., but since the use of external storage will increase the traffic between IMF and the network, it may be preferred to have the storage as close as possible to the agents.

[0065] Figure 3 illustrates examples of sources that may be used by the MMT 210 to collect data for the purpose of ML model training. Data about the goals that are set, e.g., by agents in an IMF, are one input to the MMT 210. Network KPIs and other parameters that describe the network behavior and state, are another input. The MMT 210 may also use a mapping between an agent and a set of relevant analytics reports that might be useful for training the ML model, and then to be used by ARRU for inference. MMT 210 may also uses the scores, or other feedback provided by the IMF, where this feedback reflects the accuracy assessment of the previous predictions. Such scores can be used to decide, for example, whether an ML model is accurate enough, or to decide whether further actions such as training with larger data set or re-training needs to be triggered. The MMT 210 can also utilize internal logic to measure the accuracy of the ML model it is using.

[0066] To train the ML model to be used by ARRU towards finding the most relevant analytic reports, one of the promising approaches is to implement it through supervised learning methodologies. This generated model will be used to map the relevant input data, i.e., network parameters and goals, to predict the outcome, in the form of a list of relevant analytics reports for agents. To generate a model capable of learning complex relations and features, Deep Neural Networks (DNN) may be one feasible solution, where the output layer consists of multiple classes representing the analytic reports. Then the model can pick the analytics reports with the higher probability values of being the related reports for the agents.

[0067] Next is the ARRU 220. The ARRU 220 receives an ML model that has been trained by MMT 210, and then collects information from both the knowledge base and the network to perform inference. The input to an ML model is one or more goal(s) that have been set by one or more agents, as well as data characterizing current and historical network states. Based on the goal(s) and current state of the network, the ARRU 220 will use the ML model to try to find the most relevant analytics reports for a given agent or agents.

[0068] Figure 4 illustrates, at a high-level, the ML model used by ARRU 220. The output of the model may include a prediction of an agent type, e.g., a proposal agent, prediction agent, or evaluation agent, and a list of recommended analytics reports, which may be sorted based on their importance / relevance for the agent. This list may be provided to an agent or agents, e.g., in the IMF - when an agent receives the list of recommended analytics reports, it can discard the recommendation or choose those analytics reports that it finds relevant and respond back to ARRU 220 for further preparation. When the ARRU 220 receives the response about the chosen analytics reports, it will retrieve those reports and deliver them to the respective agent. Alternatively, one or more analytics reports may be delivered to a given agent without first sending the list and receiving a request, in some embodiments or instances.

[0069] Figure 5 is an example sequence diagram for a 5GCIA, including MMT 210 and ARRU 220. The illustrated sequence includes the following steps:

[0070] 1 . MMT 210 collects data about network behavior and state i.e., network parameters, KPIs, and historical data.

[0071] 2. Having a goal set by IMF and the current state of the network and historical data about the network behavior, a mapping between agent type and relevant analytics reports that might be useful for that type of agent is created by IMF and sent to the local storage. This will be used by MMT as a training data set.

[0072] 3. Mapping dates are stored in the local storage.

[0073] 4. MMT 210 collects the mapping data from the local storage.

[0074] 5. MMT 220 starts the training process to prepare an ML model for prediction.

[0075] 6. When an ML model is ready, it will be provisioned to ARRU 220 to be used for prediction of analytics reports that will be proposed to an agent.

[0076] 7. ARRU 220 receives a goal which is set by IMF. The goal is set by IMF when the requirements in an intent cannot be met. In that case, IMF created an issue with a goal that should be reached to fulfil intent’s expectations.

[0077] 8. ARRU 220 collects information about the network including current state and historical data.

[0078] 9. ARRU 220 uses the received goal and network information as input to the ML model trained by MMT 210 for inference and predicts what analytics reports might be useful for an agent with a specific type.

[0079] 10. ARRU 220 sends the recommended analytics reports to the agent (See Figure 4).

[0080] 11 . [OPTIONAL] The agent picks the most relevant analytics reports and sends their identifiers back to ARRU 220, for collection of the reports.

[0081] 12. [OPTIONAL] ARRU 220 sends the requests to receive the interesting analytics reports to NWDAF in the core network.

[0082] 13. [OPTIONAL] NWDAF answers back to ARRU 220 with the requested analytic reports.

[0083] 14. [OPTIONAL] ARRU 220 sends the received analytics reports to the agent.

[0084] 15. [OPTIONAL] The agent sends a score it was given to the recommendation made by ARRU 220.

[0085] 16. [OPTIONAL] ARRU 220 sends the received score from the agent to MMT 210 to adjust the ML model and improve the accuracy of the model.

[0086] As was shown in Figure 2, the 5GCIA 120 may also include an ARSU 230. An ARSU 230 can act as a selection entity, meaning that it receives requests from the agents and delivers back the analytics reports. Since the IMF doesn’t have any detailed insight about 5GC, it is assumed that detailed information about the available analytics reports is not exposed to IMF. Instead, an agent would send one or more KPIs that it is interested in, such as latency, along with a duration for the request and a type of analytics report, which can be either prediction or statistics. ARSU 230 then uses a stored mapping of KPIs to analytics reports, e.g., a KPI Analytics Reports (KAR) graph which is a Markov Chain, to find the analytics reports with the highest probabilities to be relevant to the submitted KPIs. The number of analytics reports which highest ranks that shall be sent to an agent can be configured within the 5GCIA 120.

[0087] In Figure 6, an example of a KAR is illustrated. The Markov Chain in this example KAR is created initially by a 5GCIA 120, and the probability values on each edge are assigned using the extra information about the network, i.e., how KPIs are correlated to network parameters, NFs, and finally analytics reports. In this example of KAR, the layer for the network parameters is optional, meaning that the KPIs in some embodiments or instances can be directly correlated to NFs and then to the analytics reports that are relevant for specific NF(s). In some embodiments or instances, after receiving feedback from an agent, e.g., a score indicating a relevance of one or more previously supplied analytics reports for an agent, the ARSU 230 will go through the path it has recently traversed and ty to adjust the weights in the KAR. Therefore, the probability values in the edges of a KAR can be dynamically adjusted by ARSU 230.

[0088] The scores that are given by agents to the selections / recommendations from the 5GCIA 120 can be as simple as a true / false value, where “true” means that the selected / recommended analytics reports were relevant to the submitted KPI and “false” means they were not. The score can also be a scalar value e.g., [0, 1], to represent the score in a more precise form.

[0089] It should be noted that more layers can be added to the KAR shown in Figure 6, to increase the granularity when the graph is being traversed. Other layers can be a domain, a group of analytics reports e.g., UE related, performance related, etc.

[0090] Requests that have been received from the agents and responses from ARSU 230 to the agents may be stored in a local storage in IMF’s domain. This historical information may then be used by MMT 210 to find the pattern between the KPIs (shown in Figure 3), and analytics reports and the scores that are received from the agents, when an ML model is being trained. It should be noted that the trained ML model will only be used by ARRU 220, despite that some information from ARSU 230 may be used in the training process.

[0091] Figure 7 shows a sequence diagram relevant to the ARSU 230. This example sequence includes the following steps:

[0092] 1 . IMF (an agent) sends the KPIs in which it is interested to the ARSU 230.

[0093] 2. ARSU 230 traverses the KAR (See Figure 6) to find the most relevant analytics reports for the respective KPIs.

[0094] 3. When ARSU 230 finds the relevant analytics reports, it sends requests to the NWDAF in the core network.

[0095] 4. NWDAF answers back to ARSU 230 with the requested analytics reports.

[0096] 5. ARSU 230 sends the received analytics reports from NWDAF to the agent. 6. The agent sends a score it determines for the received analytics reports to ARSU 230.

[0097] 7. ARSU 230 uses the received score to adjust the weights in the KAR.

[0098] The operation of the 5GCIA can be illustrated with some use cases. As an use case, assume that a proposal agent intends to reduce the latency for a group of UEs e.g., UEs with URLLC requirements. Two scenarios are discussed here, which are:

[0099] 1 . The proposal agent responsible for generating action proposals for latency KPI, communicates with 5GCIA and asks for the relevant analytics report that can assist it to propose efficient actions. This may be regarded as an ARSU use case.

[0100] 2. 5GCIA takes the initiative and predicts what analytics reports can be useful for the proposal agent with respect to the goal which is set to reduce latency and network state and KPIs. This may be regarded as an ARRU.

[0101] In the first of these scenarios, the proposal agent submits a latency KPI, duration, and an analytics type, set to prediction, to the 5GCIA. The 5GCIA receives this information and then the ARSU goes through KAR and ends up with several relevant analytics reports to be sent back to the proposal agent.

[0102] The diagram illustrated in Figure 8 illustrates the communication between the proposal agent and 5GCIA. In this diagram the proposal agent sends a request for latency KPI, as shown at (1). Then, the 5GCIA goes through the KAR shown in Figure 9 - this example KAR consists of several layers i.e., from the top: KPIs, network parameters, NFs, and analytics reports. The traversal of this KAR by the 5GCIA is shown in Figure 8 as (2). The 5GCIA sends the most suitable analytics reports obtained by traversing the KAR back to the proposal agent, as shown at (3). The proposal agent will then choose the analytics reports of interest and send them back to 5GCIA, as shown at (4). 5GCIA will fetch the received analytics reports from the core network, as shown at (5), and then deliver the results to the proposal agent, as shown at (6). Finally, the proposal agent sends a score to 5GCIA regarding the received analytics reports, as shown at (7). The score indicates how relevant the set of analytics reports sent by the ARSU was. A high score means that the analytics reports were relevant to the KPI, and a low score means that 5GCIA will need to adjust the weights in the KAR for more precise selection.

[0103] Considering the ARSU use case, ARSU starts with the KAR illustrated in Figure 9 and continues traversing until it reaches the leaf nodes in the graph. At the end, the analytics reports for Congestion, QoS, UE Communication, and Network Performance will be selected. It should be noted that specific weights are not shown in the presented KAR, for simplicity. Having all weights in the graph, ARSU will be able to sort the selected analytics reports based on the accumulated weights. Upon receiving the selected analytics reports, the agent will send back a score which is used by 5GCIA to update the weights in the KAR.

[0104] In the second scenario, the 5GCIA initially takes a defined goal from the knowledge base where the goal in this example is to reduce the latency. Moreover, 5GCIA monitors and collects the current and historical states of the network for inference. Then, the ARRU uses the ML model that has been trained by MMT. The output of the ML model will be the agent type, e.g., proposal agent, and a list of analytics IDs that have the most relevance to the goal that is set. Considering that the goal was to reduce the latency, then it may be assumed for example that the data about the historical state of the network shows a congestion in the past while the current state of the network indicates high traffic from a group of UEs in the same tracking area that the goal is targeting. Taking all assumptions into account, the 5GCIA might recommend the proposal agent to ask for UE Mobility, Network Performance, and User Data Congestion predictions.

[0105] Figure 10 demonstrates the scenario explained in this use case. The base station with high traffic is serving a group of UEs with high traffic demands. Since it determines that a goal is to reduce the latency, the ARRU recommends the proposal agent to ask for UE mobility analytics data, which can reveal information about the UEs that will be served by the loaded base station in the future. Other recommended analytics reports, e.g., Network Performance and User Data Congestion, can also give information about whether the network will be degraded in the future and useful information about the congestion in the network.

[0106] All this information will help the proposal agent to propose actions with respect to the state of the network in the future and how the network will be changed as time goes by. The rationale behind proposing UE mobility prediction by ARRU, for example, would be the fact that the ML model has already learnt that UE mobility prediction can help the proposal agent deciding to use a closer application server to serve the UE. Having the knowledge about how UEs will be in the future will affect the action if the latency is intended to be reduced.

[0107] The 3GPP standards document 3GPP TS 28.312 v17.3.1 , “Management and orchestration; Intent driven management services for mobile networks,” 2023-04, introduces intent-driven API via management interfaces in SA5. As seen in Figure 11 , which is reproduced from that document, an MnS consumer can request an intent that the MnS producer (i.e., IMF) is expected to fulfill (e.g., via closed-loop operations). The MnS producer may consume other management services to satisfy the intent, including non-intent-driven management service interfaces.

[0108] The 3GPP standard document 3GPP TS 28.533 v17.3.0, “Management and orchestration;

[0109] Architecture framework,” 2023-03, introduces the Management Data Analytics Service (MDAS), providing a way to gather analytics from different domains (e.g., RAN, Core Network) by using well- defined standard interfaces. This means that the IMF, which is fulfilling the role of MnS producer in the intent view, can take the role of an MDAS consumer of analytics in the MDAS architecture. The 5GCIA can then be considered as a 3GPP MDAS producer, as shown in Figure 12, which leverages the exposure of analytics from the core network for performing the techniques herein. The realization of exposure of data analytics depends on the deployment option for 5GCIA. In a trusted domain, analytics subscription can be based on direct communication between 5GCIA and NWDAF. When the NWDAF is deployed in a non-trusted domain (e.g., AF) instead, exposure of data analytics is mediated by NEF (Network Exposure Function). The 5GCIA functionality described above may be used for purposes other than recommending and providing analytics data to agents in an IMF. Another application for this functionality is the enhancement of trace-related collection and analysis.

[0110] In communication networks, traces can be used, in addition to performance measurements, for specific analysis purposes, such as root cause of a malfunctioning UE, optimization of resource usage, etc. The Trace Collection Entity (TCE), which is responsible for logging traces, can be considered as an MnS consumer. In this direction, the 5GCIA may be used to improve the efficiency of TCE for delivering relevant analytic reports in addition to trace.

[0111] The TCE and concepts associated with traces are described in the 3GPP document 3GPP TS 32.421 V17.4.0, “Telecommunication management; Subscriber and equipment trace; Trace concepts and requirements,” 2022-09. As shown in Figure 13, a TCE can be enhanced with the 5GCIA functionality. This entity contains records that are information elements or signaling messages from control signaling and / or the characteristics of the user data. Benefiting from 5GCIA functionality that provides analytics reports based on predictions of network state in the future or statistics from the states in the past can be useful for TCE to collect data more accurately.

[0112] Two alternative approaches may be considered. In the first one, the TCE’s functionality is extended to receive the analytic reports, since there is no traceable interface defined between TCE and a data analytics function (e.g., NWDAF). In a second alternative, an intermediate translation layer receives the analytics reports from 5GCIA. Then it provides trace reporting to TCE after translating the analytics reports into trace reports. In addition to the representation in Figure 13, 5GCIA can be considered as a part of the management system.

[0113] The study in the 3GPP document 3GPP TR 37.816 v16.0.0, “Study on RAN -centric data collection and utilization for LTE and NR,” 2019-07, explains use cases of Self-Organizing Network (SON) / Minimization of Drive Testing (MDT) and other use cases related to data collection and utilization and identified potential solutions for them. Coverage optimization, indoor MDT improvements, QoS verification via MDT can be listed as exemplary use cases for MDT. Following these use cases, multiple measurements should be supported for MDT performance such as packet delay, packet loss, etc. and management of MD. These can be collected, recommended, and / or provided by the 5GCIA functionality described herein.

[0114] One approach is an optional service that extends the interface between the management system and legacy entities by adding a new parameter. The parameter is called Trace Context (TC), with the type of string that contains the context of the tracing information that is intended to be collected by TCE as well as meta information about the entity that is sending tracing request such as centralized CCO-5, subscriber A12345, etc. Using this parameter, the 5GCIA would be able to provide most relevant analytics reports either statistics or predictions to TCE. Like the solution described above for an IMF, the 5GCIA will then translate a trace context (TC) to one or a set of analytics reports using a trained ML model. Such information i.e., TC and predicted list of relevant analytics reports, are used as training dataset to train an ML model by MMT and used by ARRU. It should be noted that while the TC as an optional parameter can be standardized, the value of TC cannot be standardized, due to the diversity of contexts and because it is assumed to be an agreement between the vendor and the operator. IA non-standardized optional parameter called Use Case Context (UCC) is introduced has been introduced in 3GPP standards - this can be seen as a relevant example that already exists in the 5GC domain. Table 1 illustrates some examples are demonstrated to show how a TC can be translated into information that will be provided by a 5GCI. The TCs are chosen from the use cases that are listed in 3GPP TS 32.421 V17.4.0, “Telecommunication management; Subscriber and equipment trace; Trace concepts and requirements,” 2022-09. This table shows how several TCs can be mapped to analytics reports from 5GC, which in turn will be delivered to TCE.

[0115] Table 1

[0116] To implement this extended 5GCIA functionality, it is proposed that a separate optional signal, which might be called Tracing Property Signal (TPS), for example, to be sent to 5GCIA whenever a Trace Session Activation is initiated by management system. This TPS signal is optionally sent to 5GCIA whenever one of the signals described in clauses 4.1.1 .2-4.1 .1 .9 in the 3GPP document 3GPP TS 32.422 V17.10.0, “Telecommunication management; Subscriber and equipment trace; Trace control and configuration management,” 2023-03, is sent by a management system. An example specification of TPS that could be proposed to 3GPP for standardization is as follows:

[0117] - begin example 3GPP proposal text -

[0118] 4. 1. 1.x 5GCIA tracing properties mechanism [Optional]

[0119] When the management system activates a trace session, it may optionally send a Tracing Property Signal (TPS) to 5GCIA. The following trace control and configuration parameter is received by 5GCIA from the management function:

[0120] - Trace Context: The context of the trace that is used by TCE. (NOTE)

[0121] NOTE: The values of this parameter are not standardized.

[0122] Upon receiving TPS, 5GCIA delivers a list of recommended analytics reports to TCE by using the internal trained ML model.

[0123] - end example 3GPP proposal text -

[0124] A process of activating a trace session by management system, in conjunction with 5GCIA is described here, with respect to the ARRU functionality described above:

[0125] 1 . MMT inside 5GCIA trains an ML model with a training dataset which associates TCs to different lists of relevant analytics reports.

[0126] 2. The management system initiates a trace session activation based on the legacy procedures.

[0127] 3. The management system optionally sends TC to 5GCIA.

[0128] 4. [If TC is received by 5GCIA]: 5GCIA uses the trained ML model to predict what analytics reports might be useful to TCE. TCE will then decide whetherto fetch any of the proposed analytics reports by sending request to 5GCIA or discard the whole.

[0129] 5. TCE optionally gives score to the received list of analytics reports.

[0130] 6. [If the score is received by 5GCIA]: 5GCIA will use the score to improve the accuracy of the used ML model by triggering relevant action e.g., retraining the ML model, use another ML model, etc.

[0131] In view of the several techniques, examples, and use cases described above, it will be appreciated that Figure 14 illustrates an example method, for use in or in association with a communications network, for recommending analytics data to a node or function in or operatively connected to the communications network. This method may be carried out by hardware implementing a 5GCIA function like that described above (although other names for this function may be used). This function may form part of an IMF, in some embodiments or entities, or may be implemented as a standalone function or part of some other function, in others. The method shown in Figure 14 and described here is intended to be a generalization of many of the techniques and examples detailed above. More particularly, the illustrated method is intended to encompass those techniques described above in connection with a so-called ARRU. Thus, where the terminology or phrasing used in the figure and below differs slightly from that used above, the terminology and phrasing used here should be understood as at least encompassing similar terms phrasing used above, where reasonably possible. The method as illustrated does, however, assume that an ML model has already been trained. The illustrated method may thus be extended with preparatory steps for collecting data and training the model.

[0132] As shown at block 1410, the illustrated example method comprises the step of collecting network information comprising current and historical data for the state of the communications network, as shown at block 1420. The method further comprises applying the network information and intent information defining or relating a goal for the communications network to an ML model, to identify a processing entity relevant to the intent information and one or more corresponding analytics reports available from the communications network, for recommendation to the identified agent, node, or function. “Processing entity” as used herein means any agent, node, or functionality. This is shown at block 1420. The method further comprises sending the one or more corresponding analytics reports to the identified processing entity, as shown at block 1430.

[0133] As noted above, the method shown in Figure 14 may be carried out by an IMF in or operatively connected to the communications network, in some embodiments or instances. In others, the method may be carried out in a network node and / or network function separate from an in IMF in or operatively connected to the communications network. In these latter embodiments or instances, the method may further comprise receiving the intent information from the separate IMF.

[0134] In some embodiments, the method may comprise the intermediate steps shown at blocks 1422, 1424, and 1426. These steps are sending, to the IMF, an indication of an identified agent and an indication of the one or more corresponding analytics reports, and receiving, in response to said indication, a request for one or more of the corresponding analytics reports, as shown at blocks 1422 and 1424. As shown at block 1426 requested analytics reports are then retrieved, e.g., from an NWDAF, before being forwarded to the IMF.

[0135] In some embodiments, the network information applied to the ML model comprises Trace Context (TC) information. In some of these embodiments or instances, the method may comprise sending the one or more corresponding analytics reports to a Trace Collection Entity (TCE) or to a translation function for use in trace reporting to the TCE.

[0136] In some of the embodiments or instances described above, the method may comprise retrieving the one or more corresponding analytics reports from a Network Data Analytics Function (NWDAF) in the communications network. As was described above, the ML model may be tuned, using feedback related to the analytics reports recommended and provided by this method. Thus, the method may further comprise, in some embodiments, the step of receiving a score indicating a relevance of an analytics report previously sent to a processing entity, and tuning the ML model, in response to the score. These steps are shown at blocks 1440 and 1450. In some embodiments or instances, the tuning of the ML model may comprise retraining the ML model with training data collected from one or more additional data sources and / or collected at a different sampling rate, compared to training data previously used to train the model. Likewise, in some embodiments or instances, the tuning of the ML model may comprise retraining the ML model using a different feature selection, compared to a feature selection previously used to train the model.

[0137] Figure 15 illustrates another example method, for use in or in association with a communications network, for recommending analytics data to a node or function in or operatively connected to the communications network. Again, this method may be carried out by hardware implementing a 5GCIA function like that described above (although other names for this function may be used). This function may form part of an IMF, in some embodiments or entities, or may be implemented as a standalone function or part of some other function, in others.

[0138] Once again, the method shown in Figure 15 and described here is intended to be a generalization of many of the techniques and examples detailed above. More particularly, the illustrated method is intended to encompass those techniques described above in connection with a so-called ARSLL Thus, where the terminology or phrasing used in the figure and below differs slightly from that used above, the terminology and phrasing used here should be understood as at least encompassing similar terms phrasing used above, where reasonably possible.

[0139] As shown at block 1510, the method includes the step of receiving an indication of one or more key performance indicators (KPIs) for the communications network. This may be received, for example, from an agent in an IMF.

[0140] The method further comprises the step of identifying, from a data structure mapping KPI(s) to analytics reports, based on the indication of one or more KPIs, one or more analytics reports available from the communication network. This is shown at block 1520. As shown at block 1530, the method further comprises sending the identified analytics reports to the node or function.

[0141] In some embodiments or instances, as discussed above, this data structure may be a KPI-Analytics Reports graph. This graph may be in the form of a Markov chain, for example.

[0142] In some embodiments or instances, the method may comprise retrieving the analytics reports from a NWDAF in the communications network.

[0143] In some embodiments, the data structure used to identify analytics reports may be “tuned,” based on feedback received in response to analytics reports supplied according to the method. Thus, the method may comprise, in some embodiments or instances, receiving a score indicating a relevance of an analytics report previously sent to a node or function, and adjusting one or more parameters of the data structure, in response to the score. These steps are shown at blocks 1540 and 1550.

[0144] Embodiments herein also include corresponding equipment for performing either or both of the methods shown in Figure 14 and 15, or variants thereof. This equipment may be referred to as a “network node,” may include processing circuitry and power supply circuitry. The processing circuitry is configured to perform any of the steps of any of the embodiments described above for the equipment. The power supply circuitry is configured to supply power to the equipment.

[0145] Embodiments further include equipment comprising processing circuitry and memory. The memory contains instructions executable by the processing circuitry whereby the equipment is configured to perform any of the steps of any of the embodiments described above for the equipment.

[0146] More particularly, the equipment described above may perform the methods herein and any other processing by implementing any functional means, modules, units, or circuitry. In one embodiment, for example, the equipment comprise respective circuits or circuitry configured to perform the steps shown in Figure 14 and / or Figure 15. The circuits or circuitry in this regard may comprise circuits dedicated to performing certain functional processing and / or one or more microprocessors in conjunction with memory. For instance, the circuitry may include one or more microprocessor or microcontrollers, as well as other digital hardware, which may include digital signal processors (DSPs), special-purpose digital logic, and the like. The processing circuitry may be configured to execute program code stored in memory, which may include one or several types of memory such as read-only memory (ROM), random-access memory, cache memory, flash memory devices, optical storage devices, etc. Program code stored in memory may include program instructions for executing one or more telecommunications and / or data communications protocols as well as instructions for carrying out one or more of the techniques described herein, in several embodiments. In embodiments that employ memory, the memory stores program code that, when executed by the one or more processors, carries out the techniques described herein.

[0147] Figure 16 illustrates an example network node 1600 as implemented in accordance with one or more embodiments. As shown, the equipment 1600 includes processing circuitry 1610 and communication circuitry 1620. The communication circuitry 1620 is configured to transmit and / or receive information to and / or from one or more other nodes, e.g., via any communication technology. The processing circuitry 1610 is configured to perform processing described above, e.g., in Figure 14 or Figure 15, such as by executing instructions stored in memory 1630. The processing circuitry 1610 in this regard may implement certain functional means, units, or modules.

[0148] Those skilled in the art will also appreciate that embodiments herein further include corresponding computer programs. A computer program comprises instructions which, when executed on at least one processor of equipment, cause the equipment to carry out any of the respective processing described above. A computer program in this regard may comprise one or more code modules corresponding to the means or units described above.

[0149] Embodiments further include a carrier containing such a computer program. This carrier may comprise one of an electronic signal, optical signal, radio signal, or computer readable storage medium.

[0150] In this regard, embodiments herein also include a computer program product stored on a non- transitory computer readable (storage or recording) medium and comprising instructions that, when executed by a processor of equipment, cause the equipment to perform as described above.

[0151] Embodiments further include a computer program product comprising program code portions for performing the steps of any of the embodiments herein when the computer program product is executed by equipment. This computer program product may be stored on a computer readable recording medium.

[0152] 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.

[0153] ABBREVIATIONS

[0154] Abbreviation Explanation

[0155] 5GC 5G Core

[0156] 5GCIA 5G Core Interface Agent

[0157] NWDAF Network Data Analytics Function

[0158] NF Network Function

[0159] IMF Intent Management Function

[0160] KPI Key Performance Indicator

[0161] UE User Equipment

[0162] ADRF Analytics Data Repository Function

[0163] QoS Quality of Service

[0164] ML Machine Learning

[0165] NRF Network Repository Function

[0166] MMT ML Model Trainer

[0167] ARRU Analytics Reports Recommendation Unit

[0168] ARSU Analytics Reports Selection Unit DNN Deep Neural Network

[0169] KAR KPI Analytics Reports

[0170] SON Self-Organizing Network

[0171] MDT Minimization of Drive Testing MnS Management Service

[0172] MDAS Management Data Analytics Service

[0173] MDT Minimization of Drive Testing

[0174] SON Self-Organizing Networks TC Trace Context TCE Trace Collection Entity

Claims

CLAIMSWhat is claimed is:1 . A method, in or associated with a communications network, for recommending analytics data to a node or function in or operatively connected to the communications network, the method comprising: collecting (1410) network information comprising current and historical data for the state of the communications network; applying (1420) the network information and intent information defining or relating a goal for the communications network to a Machine Learning, ML, model, to identify a processing entity relevant to the intent information and one or more corresponding analytics reports available from the communications network, for recommendation to the identified processing entity; and sending (1430) the one or more corresponding analytics reports to the identified processing entity.

2. The method of claim 1 , wherein the method is carried out by an Intent Management Function, IMF, (100) in or operatively connected to the communications network.

3. The method of claim 1 , wherein said method is carried out in a network node and / or network function separate from an Intent Management Function, IMF, (100) in or operatively connected to the communications network, and wherein the method further comprises receiving the intent information from the IMF.

4. The method of claim 3, wherein the method further comprises: sending (1422), to the IMF (100), an indication of an identified agent and an indication of the one or more corresponding analytics reports; receiving (1424), in response to said indication, a request for one or more of the corresponding analytics reports; retrieving (1426) the requested analytics reports; and forwarding the retrieved analytics reports to the IMF (100).

5. The method of any one of claims 1-3, wherein the network information comprises Trace Context, TC, information and wherein the method comprises sending the one or more corresponding analytics reports to a Trace Collection Entity, TCE, or to a translation function for use in trace reporting to the TCE.

6. The method of any one of claims 1-5, wherein the method comprises retrieving the one or more corresponding analytics reports from a Network Data Analytics Function, NWDAF, in the communications network.

7. The method of any one of claims 1-6, wherein the method further comprises:receiving (1440) a score indicating a relevance of an analytics report previously sent to a processing entity; and tuning(1450) the ML model, in response to the score.

8. The method of claim 7, wherein tuning the ML model comprises retraining the ML model with training data collected from one or more additional data sources and / or collected at a different sampling rate, compared to training data previously used to train the model.

9. The method of claim 7 or 8, wherein tuning the ML comprises retraining the ML model using a different feature selection, compared to a feature selection previously used to train the model.

10. A network node for use in or with a communications network and adapted to carry out a method according to any one of claims 1-9.11 . A computer program product comprising program instructions for execution by processing circuitry in a network node, the program instructions being configured to cause the network node to carry out a method according to any one of claims 1-9.

12. A network node (1600) for use in or with a communications network, the network node (1600) comprising: communications interface circuitry (1620) configured to communicate with one or more other network nodes; processing circuitry (1610) operatively coupled to the communications interface circuitry; and memory (1630) in or operatively coupled to the processing circuitry (1610) and storing program instructions for execution by the processing circuitry (1610), whereby the network node is configured to: collect network information comprising current and historical data for the state of the communications network; apply the network information and intent information defining or relating a goal for the communications network to a Machine Learning, ML, model, to identify a processing entity relevant to the intent information and one or more corresponding analytics reports available from the communications network, for recommendation to the identified processing entity; and send the one or more corresponding analytics reports to the identified processing entity.

13. The network node (1600) of claim 12, wherein the network node (1600) is configured to carry out a method according to any of claims 2-9.

14. A method, in or associated with a communications network, for recommending analytics data to a node or function in or operatively connected to the communications network, the method comprising:receiving (1510) an indication of one or more key performance indicators, KPIs, for the communications network; identifying (1520), from a data structure mapping KPI(s) to analytics reports, based on the indication of one or more KPIs, one or more analytics reports available from the communication network; and sending (1530), to the node or function, the identified analytics reports.

15. The method of claim 14, wherein the data structure is a KPI-Analytics Reports graph.

16. The method of claim 15, wherein the KPI-Analytics Reports graph is in the form of a Markov chain.

17. The method of any one of claims 14-16, wherein the method comprises retrieving the analytics reports from a Network Data Analytics Function, NWDAF, in the communications network.

18. The method of any one of claims 14-17, further comprising: receiving (1540) a score indicating a relevance of an analytics report previously sent to a node or function; and adjusting (1550) one or more parameters of the data structure, in response to the score.

19. A network node (1600) for use in or with a communications network and adapted to carry out a method according to any one of claims 14-18.

20. A computer program product comprising program instructions for execution by processing circuitry in a network node, the program instructions being configured to cause the network node to carry out a method according to any one of claims 14-18.

21. A network node (1600) for use in or with a communications network, the network node comprising: communications interface circuitry (1620) configured to communicate with one or more other network nodes; processing circuitry (1610) operatively coupled to the communications interface circuitry; and memory (1630) in or operatively coupled to the processing circuitry (1610) and storing program instructions for execution by the processing circuitry (1610), whereby the network node is configured to: receive an indication of one or more key performance indicators, KPIs, forthe communications network; identify, from a data structure mapping KPI(s) to analytics reports, based on the indication of one or more KPIs, one or more analytics reports available from the communication network; and sending, to the node or function, the identified analytics reports.

22. The network node (1600) of claim 21 , wherein the network node (1600) is configured to carry out a method according to any of claims 15-18.

Images