Technique for operating a wireless telecommunications infrastructure

Abstract

A technique for operating a wireless telecommunications infrastructure based on an intent is described. As to a method aspect of the technique, for an action executable in the autonomous operation of the wireless telecommunications infrastructure, a value of a key performance indicator, KPI, of the wireless telecommunications infrastructure and a value of an uncertainty of the KPI value is predicted. An uncertainty of a utility is assessed based on the predicted KPI value and the predicted uncertainty value of the KPI value. The action is selectively executed, wherein the selectivity is dependent on the intent and the assessed uncertainty of the utility.

Description

[0001] Technique for operating a wireless telecommunications infrastructure

[0002] Technical Field

[0003] The present disclosure relates to a technique for operating a wireless telecommunications infrastructure. More specifically, and without limitation, methods and devices are provided for autonomously operating a wireless telecommunications infrastructure according to at least one intent.

[0004] Background

[0005] The Third Generation Partnership Project (3GPP) has defined many different types of services such as Ultra-Reliable Low-Latency Communication (URLLC), enhanced Mobile Broadband (eMBB) and massive Machine-Type Communication (mMTC). Controlling and configuring the delivery of multiple services at the same time, each with certain quality requirements, is beyond the human capacity. Therefore, there is a need for some level of automation, or better yet, a management system with minimal human intervention, such as the use of Cognitive Networks (CN) to manage multiple services.

[0006] Each service may have different requirements. For example, a URLLC service may require its latency to be less than 1 ms and packet loss to be less than 0.1%. In another example, a Web Real-Time Communication (WebRTC) service may require its Quality of Experience (QoE), which is a non-negative integer as an objective qualitative representation to be greater than 4, or mobile Internet of Things (mloT) may require its power consumption to be less than 1 joule per day. The requirements may be referred to as Key Performance Indicators (KPIs), so different services may have different KPIs in terms of number and type. Fulfilment of a service means that all KPIs for that service are successfully met. The aim is to support as many services as possible, given the resources of the telecommunications network. As the number of different requirements increases, it is necessary to take decisions in an autonomous way to fulfil all these KPIs. In an autonomous approach, the requirement comes as a high-level intent, which is a declarative information object that defines the requirements of a service that the telecommunication network is expected to fulfil. A conventional management system of the telecommunication network may dynamically make decisions based on changes of the KPIs predicted for the decisions in order to provide each of the multiple services with its expected quality. However, even if the prediction is just slightly off, the quality may actually be deceased or the conventional management system may jump back and forth between different operating states.

[0007] Summary

[0008] Accordingly, there is a need for a technique that autonomously operates a wireless telecommunications infrastructure more reliable and stable in the absence of human intervention.

[0009] As to a method aspect, a method of autonomously operating a wireless telecommunications infrastructure according to at least one intent is provided. The method comprises or initiates a step of predicting, for an action executable in the autonomous operation of the wireless telecommunications infrastructure, a value of a key performance indicator (KPI) of the wireless telecommunications infrastructure and a value of an uncertainty of the predicted KPI value. The method further comprises or initiates a step of assessing an uncertainty of a utility based on the predicted KPI value and the predicted uncertainty value of the KPI value. The method further comprises or initiates a step of selectively executing the action. The selectivity is dependent on the intent and the assessed uncertainty of the utility.

[0010] By taking the uncertainty of the utility into account in the decision whether the action is executed or not executed, embodiments can refrain from a marginal and / or uncertain improvement of the utility, e.g. if the predicted KPI value is close to (or in) a region where the utility sensitive to small changes of the KPI value, wherein the uncertainty value of the predicted KPI value can provide a measure of closeness. Hence, embodiments can avoid actions that bear a risk of worsening the utility and can lead to a steadier autonomous operation of the wireless telecommunications infrastructure, e.g. with less frequent actions.

[0011] The utility may have an uncertainty according to the assessed uncertainty, e.g. due to a finite accuracy of the predictions and / or due to stochastic dynamics of the wireless telecommunications infrastructure (e.g., because of unpredictable behavior of radio device served by the wireless telecommunications infrastructure). The autonomously operated wireless telecommunications infrastructure may determine (e.g., in the assessing step) and / or make use of the uncertainty of the utility of the KPI value (e.g., in the selectively executing step).

[0012] The claim terminology may be referred to using linguistic variants, for example as follows. A value of the KPI may also be referred to as KPI value. The predicted value of the KPI may also be referred to as the predicted KPI value or predicted KPI or KPI prediction. Furthermore, a value of the uncertainty of a KPI value may also be referred to as uncertainty value of the KPI value or uncertainty value of the KPI or uncertainty of the KPI value or KPI uncertainty value. The predicted value of the uncertainty of the predicted KPI value may also be referred to as the (predicted) uncertainty value of the (predicted) KPI value or the (predicted) uncertainty value of the KPI or the predicted uncertainty of the KPI value or the predicted KPI uncertainty value. Moreover, expressions of the form "at least one of A, B, and C" encompass and disclose the alternatives A or B or C (i.e., any individual feature), as well as any combined alternative (i.e., any subset) such as A and B, or A and C, or B and C, as well as the combination A, B, and C (i.e., a combination of all features).

[0013] The utility may be a utility defined for at least one of the KPI, the action and / or for the wireless telecommunications infrastructure. The utility may be a utility function, e.g. a function of the KPI (i.e., a function of the KPI value). The intent may comprise the utility function.

[0014] For conciseness, the technique is primarily described in terms of one intent, one action, one KPI, and one utility. Any embodiment of the method may be performed (e.g., independently) for multiple intents, and / or by predicting the KPI value and the KPI uncertainty value for each of at least two KPIs, and / or by assessing the uncertainty of the utility as a function of the at least two KPIs, and / or by selectively performing one or more actions. For example, the KPI value and the uncertainty value may be predicted for each KPI and / or for each executable action. Alternatively or in addition, the steps of predicting and assessing may be performed for each of multiple actions (e.g., proposed actions), and the step of selectively executing the action may comprise selecting one action to be executed out of the multiple actions. Optionally, the assessing may comprise cross-checking that a gain in the utility based on the KPI value and the uncertainty value predicted for a first action is not reduced or nullified by a loss in the utility based on the KPI value and the uncertainty value predicted for a second action other than the first action.

[0015] The predicted KPI value and the predicted uncertainty value may specify a probability distribution (e.g., a differential probability distribution or probability density function, or a cumulative distribution function) of the KPI. Alternatively or in addition, the predicted KPI value and / or the predicted uncertainty value may comprise (or may be) one or more parameters of a probability distribution of the KPI. For example, the predicted KPI value may specify a mean value of the KPI (e.g., predicted to be the result of the action). The predicted uncertainty value may specify moments or cumulants of the probability distribution of the KPI, e.g. a variance or standard deviation of the KPI. For example, the probability distribution of the KPI may be a Gaussian distribution specified by the mean value and the variance.

[0016] The uncertainty of the utility may be assessed using the probability distribution of the KPI determined based on the predicted KPI value and the predicted uncertainty value of the KPI.

[0017] The wireless telecommunications infrastructure may comprise at least one of a radio access network (RAN), a backhaul network (BN), and a core network (CN). The wireless telecommunications infrastructure may be an autonomously operated telecommunication network and / or a part of a telecommunication network (e.g., a telecommunication system). The autonomously operated wireless telecommunications infrastructure may be an intent-based network. The terms "network" and "system" hereinafter may be used interchangeably.

[0018] The executing of the action may also be referred to as performing the action. Moreover, the action executable in the autonomous operation may be an optional step (in the sense of a selectively executed step) in the autonomous operation (i.e., the autonomously operating) of the wireless telecommunications infrastructure. The action may be a step of autonomously controlling the wireless telecommunications infrastructure.

[0019] The predicting of the KPI value for the action may predict an effect of the action on the KPI value. For example, the predicted KPI value may be an expectation value of the KPI after executing the action or an estimator of a mean value of the KPI after the action. The KPI may be considered stochastic processes prior to the action and after the action, respectively.

[0020] The assessing of the uncertainty of the utility based on the predicted KPI value and the predicted uncertainty value of the KPI value may relate to the action. The uncertainty may occur when a change of the utility is brought about by executing the action. The uncertainty may be different from the change of the utility. For example, the uncertainty may be an uncertainty of the change (e.g., an uncertainty of the amount of the change of the utility). For example, the change of the utility may be predicted as the change of the KPI value from a KPI value prior to the action to the predicted KPI value. The uncertainty of the utility may be predicted by and / or caused by the predicted uncertainty value of the predicted KPI value.

[0021] The action may change a state of the wireless telecommunications infrastructure. The state of the wireless telecommunications infrastructure may comprise operating parameters (briefly: parameters) of the wireless telecommunications infrastructure. For example, the action may be determined (e.g., proposed) as input to the predicting step) and / or the KPI value and the KPI uncertainty value may be predicted based on the parameters. Examples of the parameters comprise a number of radio device connected to a cell or network node of the wireless telecommunications infrastructure.

[0022] The autonomously operated wireless telecommunications infrastructure may be a self-adapting network. Embodiments of the method can improve efficiency and / or a level of service delivery according to the at least one intent.

[0023] The autonomously operated wireless telecommunications infrastructure may be a cognitive network. The cognitive network due to the autonomous operation to manage multiple services with multiple according to different requirement (e.g., different intents). Herein, the terms "requirement" and "intent" and "criterion" may be used interchangeably.

[0024] The requirements may be strict. Alternatively or in addition, the different requirements may be related to each other (e.g., having correlation and / or causation relationship). The amount of data regarding the requirements and / or the decisions that need to be made may be beyond capacity of a human. Therefore, embodiments of the autonomous operation of the wireless telecommunications infrastructure (e.g., as a cognitive network) can manage multiple services with different requirements without or with minimal human intervention in communication networks.

[0025] From the perspective of a human operator, an "intent" (e.g., a threshold value, KPI_t, for the KPI value) may be understood as the expectation of what the wireless telecommunications infrastructure as an operational system is supposed to deliver and / or how the wireless telecommunications infrastructure behaves. In other words, the at least one intent may be a formal specification of all expectations including requirements, goals and constraints given to the wireless telecommunications infrastructure as a technical system.

[0026] Preferably, the method for the autonomously operating of the wireless telecommunications infrastructure may not execute any action (e.g., control operation) unless the action is related to the fulfillment and / or assurance of the at least one intent. There may be several intents for each service to be provided by the wireless telecommunications infrastructure.

[0027] The performance of the autonomously operated wireless telecommunications infrastructure may be evaluated by quantized indicators such as key performance indicators (KPI). The KPI may be measurable. Multiple KPIs may be associated and / or measured for each intent of the autonomously operated wireless telecommunications infrastructure.

[0028] Alternatively or in addition, each of the at least one intent may comprise or may imply (e.g. in terms of the utility function) a threshold value for the KPI value, which may also be referred to as a target value of a KPI (e.g., KPI_t). The at least one intent may change according to different network configurations. The wireless telecommunications infrastructure may adapt a new network configuration according to a change in network environment (e.g., emergency circumstances, power shut down in some network nodes, a change in the environment such as new building, etc.)

[0029] The method may be performed by at least one device (e.g., at least one node) in the wireless telecommunications infrastructure. Alternatively or in addition, the method may be a computer-implemented method. The predicted KPI value (KPI_p) may be predicted for a given action (a), e.g., according to the at least one intent. Alternatively or in addition, the predicted uncertainty of the utility may be predicted for the predicted KPI value (KPI_p).

[0030] A utility may be a function (e.g., mathematical function) of the KPI value. The utility (e.g., utility function) may be specified by a human administrator derived from another utility. Alternatively or in addition, the utility may be a function of multiple KPI values. The utility value may be understood as a measurable value based on a utility function of the KPI value.

[0031] The at least one intent in the autonomously operated wireless telecommunications infrastructure may be expressed declaratively - , as a utilitylevel target (e.g., goal) that describes the properties of a satisfactory outcome rather than prescribing a specific solution. This may give the autonomously operated wireless telecommunications infrastructure the flexibility to explore various actions (e.g., solution options) and find the optimal one. It may also allow the autonomously operated wireless telecommunications infrastructure to optimize by choosing its own goals that maximize utility value.

[0032] Alternatively or in addition, the intents and / or utility (e.g., utility functions) may be sorted according to a prioritization criterion (e.g., one or more criterions that are input to the method, e.g. set by a human administrator, and / or updated criteria according to the change of one or more conditions and the previously set criteria). For example, in some configurations of the wireless telecommunications infrastructure, the energy consumed by network nodes of the wireless telecommunications infrastructure must be minimal, even at the cost of increased interference, latency, etc. In this case, the uncertainty value of the utility value of the energy consumption can be used for selectively performing the action, even at the cost of reducing the latency.

[0033] The at least one intent may be added to the wireless telecommunications infrastructure offline (e.g., prior to performing the method) and / or during runtime of the wireless telecommunications infrastructure (e.g., while performing the method). Hence, the autonomously operated wireless telecommunications infrastructure may adapt to changed intents and may detect a conflict and execute an action for resolution. The method may further comprise selecting the action from multiple possible (e.g., given) actions according to the assessed (e.g., determined) uncertainty of the utility (e.g. a utility function of the KPI). For example, the action may be selected to minimize the uncertainty of the utility assessed for the respective action. Alternatively or in addition, the method may comprise selecting an action from multiple actions according to the assessed uncertainty of the utility to maximize the utility. Moreover, the selectivity in selectively executing the action may encompass that the autonomously operated wireless telecommunications infrastructure may select no action, e.g. to refrain from executing the action or any of the multiple actions.

[0034] The wireless telecommunications infrastructure may comprise an Intent Management Function (IMF). The IMF may perform the method.

[0035] Herein, a lower bound may be an absolute lower bound (i.e., a hard lower limit) or a statistical lower bound (i.e., the lower bound is exceeded with a certain probability, e.g. 90 % or 99 %).

[0036] The predicted uncertainty value of the predicted KPI value (e.g., according to the method aspect) may comprise at least one of a lower bound for the KPI value, an upper bound for the KPI value, a confidence interval of the predicted KPI value, and a prediction interval of the predicted KPI value.

[0037] The lower bound of the KPI may correspond to a lower bound of the confidence interval or prediction interval of the predicted KPI value. Alternatively or in addition, the lower bound may correspond to a percentile (e.g. 1 % or 10 %) of the predicted KPI value (e.g. determined based on the uncertainty value of the predicted KPI value).

[0038] The uncertainty value of the predicted KPI value may comprise or may be a confidence interval or a prediction interval of the KPI value. For example, the confidence interval may correspond to an interval that comprises a mean value or expectation value of the KPI (resulting from executing the action) with a predefined probability, e.g., specified in units of a standard deviation (o, e.g. a square root of the variance o2) of a probability distribution of the KPI value. The prediction interval may be greater than the confidence interval. For example, the prediction interval may correspond to a combination of the confidence interval and a variability (e.g., scattering) of the actual KPI value around the mean value or expectation value resulting from the action.

[0039] The uncertainty value comprising the confidence or prediction interval may mean that the uncertainty value is indicative of the confidence or prediction interval. For example, the uncertainty value may comprise at least one of a width, a center, a lower bound, and an upper bound of the confidence or prediction interval.

[0040] The linguistic variants of the claim terminology may refer to the lower bound, the confidence interval or the prediction interval of "the predicted KPI value" or for "the KPI value", because these examples of the predicted uncertainty value are indicative of a scattering for "the KPI value" (e.g. as a random variable or a stochastic process) relative to "the predicted KPI value".

[0041] Herein, the predicted KPI value may be denoted by KPI_p. The predicted lower bound of the KPI value may be denoted by KPI_I. The predicted upper bound of the KPI value may be denoted by KPI_u. The prediction interval or the predicted confidence interval may be denoted by Cl (i.e., the symbol Cl is used interchangeably for prediction or confidence interval).

[0042] For the given action, the predicted KPI value and its uncertainty value may be used and / or saved in a memory to be used as a list (e.g., a 5-tuple), for example as <a, KPI_p, KPI_u, KPI_I, Cl> for any further step of the method.

[0043] The Cl may be understood as a confidence level that shows what percentage of the time (e.g., 95 %) the KPI value falls in between KPI_u and KPI_I.

[0044] Confidence intervals may represent a range of values that are likely to contain the true mean value of some response variable based on specific values of one or more predictor variables.

[0045] Prediction intervals may represent a range of values that are likely to contain the true value of some response variable for a single new observation based on specific values of one or more predictor variables.

[0046] The prediction interval may be greater than or equal to confidence interval. Predicting the KPI_p value and / or predicting the uncertainty value (e.g., KPI_u or KPI_I or Cl) may be implemented using different methods. These methods may have different theoretical and / or practical limitations, e.g. when using a neural network (NN) system for the prediction.

[0047] The action (e.g., according to the method aspect) may be executed if the lower bound or the confidence interval or the prediction interval corresponds to an increase of the utility. Alternatively or in addition, the action may not be executed if the lower bound or the confidence interval or the prediction interval corresponds to a decrease of the utility. Alternatively or in addition, the utility may be a function of the KPI value with a non-linear characteristic. The action may be executed if a KPI value of the non-linear characteristic is below the lower bound or the confidence interval or the prediction interval. Alternatively or in addition, the action may not be executed if the KPI value of the non-linear characteristic is included in or above the lower bound or the confidence interval or the prediction interval.

[0048] The non-linear characteristic of the utility function may be a step or an inflection point of the utility function, e.g. a sigmoid function (i.e., a logistic function). The KPI value of the non-linear characteristic may be referred to as a threshold value of the KPI value. The action may be not executed if the predicted KPI value is less than or close (e.g. from above) to the non-linear characteristic of the utility function (e.g., the threshold value), wherein the closeness (e.g., in terms of the KPI value) is measured or specified by the predicted uncertainty value.

[0049] The assessing of the uncertainty of the utility may comprise determining if the lower bound or the confidence interval or the prediction interval corresponds to an increase or decrease of the utility.

[0050] The confidence interval or the prediction interval may correspond to an increase or decrease of the utility, if each KPI value in the confidence interval or the prediction interval, respectively, corresponds to an increase or decrease, respectively, of the utility.

[0051] In any embodiment, the action may be executed based on (i.e., the selectivity may be based on) a criterion for the intent and the assessed uncertainty of the utility. The intent may comprise the criterion. In other words, the action may be executed based on a criterion of the intent for the assessed uncertainty of the utility. By way of example, the utility may be a step function (i.e., the utility may comprise a step at a predefined threshold value for the KPI value). In other words, the assessing of the uncertainty of the utility may comprise assessing whether or not the KPI value is expected to fall below a predefined threshold value and / or assessing the probability for the KPI value falling below the threshold value and / or assessing an expected loss (e.g., a reduction of the utility or a negative utility) if the KPI value falls below the threshold value.

[0052] Herein, the assessing of the uncertainty of the utility may encompass a binary criterion for the utility being either greater or less than a predefined threshold value. As a linguistic variant of the claim terminology, the assessing may also be referred to as assessing a value of an uncertainty of the utility, e.g. to indicate that the assessing is not merely a threshold criterion (for a step function as the utility function) but involves a smooth utility function of the KPI value.

[0053] The assessing of the uncertainty (e.g., according to the method aspect) may comprise determining whether a probability for the action increasing the utility is greater than a probability for the action decreasing the utility and / or whether a predicted increase of the utility greater than a predicted decrease of the utility.

[0054] The assessing (e.g., determining) of the uncertainty may comprise determining whether or not a risk of the action to decrease the utility exceeds the chance of the action to increase the utility. For example, at least two scenarios may be predicted. A first scenario may include a predicted increase of the utility, e.g. based on an upper bound of the KPI value. A second scenario may include a predicted decrease of the utility, e.g. based on a lower bound (e.g., the above or below mentioned lower bound) of the KPI value. The action may be executed if a first probability for the first scenario multiplied by the predicted increase of the utility is greater than a second probability for the second scenario multiplied by the predicted decrease of the utility. Alternatively or in addition, the action may be not executed if the first probability for the first scenario multiplied by the predicted increase of the utility is equal to or less than the second probability for the second scenario multiplied by the predicted decrease of the utility.

[0055] The assessing of the uncertainty (e.g., according to the method aspect) may comprise determining an expectation value or lower bound of the utility based on the predicted KPI value and the predicted uncertainty value of the KPI value. The action may be executed if the expectation value or lower bound of the utility corresponds to an increase of the utility. Alternatively or in addition, the action may not be executed if the expectation value or lower bound of the utility corresponds to a decrease of the utility.

[0056] Herein, the lower bound of the utility may refer to an end of an interval (e.g. prediction or confidence interval) that is less favorable in terms of the KPI or utility.

[0057] In any embodiment, the increase or decrease of the utility may be defined relative to a KPI value prior to the action, e.g. a mean value of the KPI prior to executing the action.

[0058] In a first variant of any embodiment, the expectation value of the utility may be the utility as a function of the KPI value evaluated at the predicted KPI value. Alternatively or in addition, the lower bound of the utility may be the utility function evaluated at the lower bound of the KPI value. Alternatively or in addition, the uncertainty of the utility may correspond to the derivative of the utility function (with respect to the KPI value, i.e. the slop of the utility as a function of the KPI value) at the predicted KPI value multiplied by the predicted uncertainty value of the KPI value.

[0059] In a second variant of any embodiment, the expectation value or lower bound of the utility may be determined by evaluating the utility as a function of the KPI using a probability distribution of the KPI value, e.g. the probability distribution specified by the predicted KPI value and the predicted uncertainty value of the KPI value. Alternatively or in addition, the assessing of the uncertainty of the utility may comprise determining a probability distribution of the utility, e.g. based on the probability distribution of the KPI value. For this end, the utility may be a differentiable and / or strictly monotone function of the KPI value.

[0060] The lower bound of the utility may correspond to a lower bound of a confidence interval or prediction interval of the utility. Alternatively or in addition, the lower bound of the utility may correspond to a percentile (e.g. 1 % or 10 %) of the utility as a random variable (i.e., a stochastic variable) given the predicted KPI value and the predicted uncertainty value or given the probability distribution of the KPI value.

[0061] The action (e.g., according to the method aspect) may be executed if a criterion of the intent is fulfilled for the assessed uncertainty of the utility. A further action may be executed if a further criterion is fulfilled for the assessed uncertainty of the utility or for a further assessed uncertainty of a further utility. The action may not be executed if the further criterion for executing the further action is fulfilled.

[0062] The further criterion may be comprised in the intent or in a further intent of the wireless telecommunications infrastructure.

[0063] The further criterion for executing the further action may relate to the further utility as a function of a further KPI value. For example, the further criterion for executing the further action may be fulfilled, if an expectation value or the lower bound of the utility fulfils a further threshold for executing the further action.

[0064] For example, the action is not executed if the further criterion for executing the further action is fulfilled (e.g., according to the assessment of the utility or the further utility) and the further action counteracts the action. The further action may reverse (i.e., counter, e.g., undo) the action. By assessing whether the action is likely to trigger a reverse action, embodiments of the technique can avoid that the action leads to a reverse action and / or that the action leads to (e.g., temporary) oscillating states of the wireless telecommunications infrastructure.

[0065] Alternatively or in addition, the action is not executed if the further criterion for executing the further action is fulfilled (e.g., according to the assessment of the utility or the further utility) and the further action amplifies the action. The further action may intensify or reinforce the action. By refraining from executing the action that is likely to trigger an amplifying action, embodiments of the method can avoid a chain reaction of actions (e.g., runaway dynamics) in the autonomously operating of the wireless telecommunications infrastructure.

[0066] In addition to any one or each criterion for executing the action, the method may further comprise a counter criterion for not executing the action (i.e., for refraining from executing the respective action) if the counter criterion is fulfilled. In any embodiment, the criterion may be fulfilled if the lower bound or the confidence interval or the prediction interval corresponds to an increase of the utility. Alternatively or in addition, the counter criterion may be fulfilled if the upper bound or the confidence interval or the prediction interval corresponds to a decrease of the utility.

[0067] The intent may comprise the criterion. Alternatively or in addition, the criterion may comprise the utility falling below a predefined threshold value. For example, the intent may comprise the threshold value.

[0068] In the case of multiple intents, each intent may comprise a criterion for the selectivity in the execution of the action. The action may be executed if all criterions are fulfilled. Alternatively or in addition, the action may be not executed if at least one criterion is not fulfilled.

[0069] The intent (e.g., according to the method aspect) may comprise multiple utilities each depending on at least one KPI or the wireless telecommunications infrastructure is autonomously operated according to multiple intents each comprising at least one utility each depending on at least one KPI. The action may be selectively executed and the selectivity is dependent on a combination of the uncertainty assessed for each of the utilities. Alternatively or in addition, the action may not be executed if at least one criterion for the uncertainty assessed for each of the utilities is not fulfilled. Alternatively or in addition, the action may be executed if each criterion for the uncertainty assessed for each of the utilities is fulfilled.

[0070] The combination may comprise a sum of the utilities or a minimum of the utilities. Alternatively or in addition, each utility may be a function of at least one KPI value.

[0071] The predicted uncertainty value of the predicted KPI value (e.g., according to the method aspect) may comprise a probability of the predicted KPI value or a probability for the KPI value being less than the predicted KPI value or a probability for the KPI value falling below a predefined threshold value. Alternatively or in addition, the assessed uncertainty of the utility may comprise a probability for a reduction of the utility or a probability for the utility falling below a predefined threshold value. Each intent may comprise at least one criterion for the selectivity in selectively executing the action in terms of the utility (i.e., the selectivity that is dependent on the intent and the assessed uncertainty of the utility). Each intent and / or each criterion and / or each utility may be defined in terms of at least one KPI.

[0072] The autonomous operation of the wireless telecommunications infrastructure (e.g., according to the method aspect) may comprise at least one closed loop. Alternatively or in addition, each closed loop may comprise at least one intent.

[0073] The method may comprise the closed loop. The closed loop may be a control loop of the wireless telecommunications infrastructure. The closed loop may comprise measuring or receiving a state of the wireless telecommunications infrastructure. The action may alter the state of the wireless telecommunications infrastructure.

[0074] The autonomously operated wireless telecommunications infrastructure may be used for delivering multiple services. Each service may comprise at least one closed loop. Example of the services include ultra-reliable low-latency communication (URLLC) (e.g. robot control), enhanced mobile broadband (eMBB) (e.g. WebRTC), and massive machine-type communication (mMTC) (e.g. massive Internet of Things, mloT). Alternatively or in addition, the services may comprise heterogeneous services (e.g., according to heterogeneous environment of a 5G wireless telecommunications infrastructure). Alternatively or in addition, the services may be competing with each other (e.g., for dynamically balancing radio resources, computational resources, etc.).

[0075] The KPI (e.g., according to the method aspect) may comprise at least one of latency; bit rate; block error rate (BLER); packet loss rate; signal to noise ratio (SNR); and signal to interference and noise ratio (SINR).

[0076] Any of these KPIs may be measurable in the wireless telecommunications infrastructure, e.g. for each cell or each network node (e.g., a base station such as a gNB) of the wireless telecommunications infrastructure or for each radio device (e.g., a user equipment, UE) served by the wireless telecommunications infrastructure. The prediction of the KPI value and / or the uncertainty value of the KPI value (e.g., according to the method aspect) may be performed by an artificial intelligent agent (Al agent).

[0077] The prediction of the KPI value and / or the uncertainty value of the KPI value may be briefly referred to as KPI prediction.

[0078] The Al agent may be part of the autonomously operating of the wireless telecommunications infrastructure. The Al agent may comprise a machine learning system, e.g. a neural network system. Alternatively or in addition, the Al agent may comprise a symbolic logic system (also referred to as symbolic Al system). The neural network system may be trained to perform the KPI prediction, optionally by historical data of the wireless telecommunications infrastructure and / or by simulated data.

[0079] Herein, "agent" may refer to a system or entity that is capable of perceiving its environment (e.g., by measuring or receiving measurements), making decisions (e.g., according to the selectivity in selectively executing the action) or taking actions, and achieving goals or objectives.

[0080] The utility (e.g., according to the method aspect) may be a function of the KPI value.

[0081] The utility may be a (e.g., smooth and / or strictly monotonic) function of the KPI value (e.g., as a random variable) and / or the predicted KPI value (KPI_p). The utility function (e.g., a mapping) may be used to determine how much utility may be obtained by achieving a certain value of the KPI. For example, the utility may be a step-function of the KPI value, which utility value may be zero if the KPI value is below its target (e.g., according to the at least one intent). The utility function may be an input to the autonomous operation, e.g. specified by an external authority (i.e., operator) of the autonomously operated wireless telecommunications infrastructure. As an example of a smooth utility function, the utility may be a nonlinear function of the KPI value. An increment of the utility may be steep at the non-linear characteristic (e.g., at the target or threshold value) compared to a flat increment if the KPI value is above its target or threshold value. That is, the additional utility achieved by increasing the KPI value beyond the target or threshold value may be marginal. The method (e.g., according to the method aspect) may be performed according to more than one intent.

[0082] The method of autonomously operating the wireless telecommunications infrastructure may comprise corresponding steps for each of two or more intents or a combination of the intents. For example, each service to be provided by the wireless telecommunications infrastructure may comprise several intents (e.g., requirements).

[0083] Each intent may be associated with at least one KPI and / or at least one utility. For example, the utility used in assessing may be the minimum of the at least one utility for each KPI or the sum of the at least one utility for each KPI. The executable action may be the same for the at least one KPI and / or the at least one utility. Alternatively or in addition, the executable action may affect each of the at least one KPI.

[0084] The method (e.g., according to the method aspect) may be performed for more than one given action for each of the at least one intent.

[0085] The method of autonomously operating the wireless telecommunications infrastructure may comprise the steps for more than one given action, e.g., two or more actions. The given actions may be a list of actions (e.g., no_action, action_l, action_2, ...).

[0086] The method may comprises selectively executing the action from a list of more than one action (e.g., for each intent). The method may selectively perform an action from a list of given actions (e.g., including no action). No action (e.g., zero action) may be understood as the method refraining from executing the action.

[0087] The method (e.g., according to the method aspect) may further comprise or initiate obtaining information as to a state of the wireless telecommunications infrastructure after executing the action.

[0088] The obtained information may comprise results from measurements, e.g. at one or more network nodes (e.g. base stations such as gNBs) of the wireless telecommunications infrastructure serving one or more radio devices (e.g. user equipments, UEs) or at one or more radio devices served by the wireless telecommunications infrastructure. For example, the obtaining of the information may comprise receiving a measurement report indicative of the information.

[0089] The obtained information (e.g., according to the method aspect) may comprise at least one of a measured KPI value; a KPI lower bound (KPI_I); a KPI upper bound (KPI_u); and a confidence interval (Cl).

[0090] The obtained information may comprise measurement values of the measureable KPI information (e.g., a realization of the predicted KPI value and / or a realization of the predicted uncertainty value of the predicted KPI value).

[0091] The method (e.g., according to the method aspect) may further comprise or initiate comparing the obtained information after executing the action with the predicted KPI value and / or the predicted uncertainty value of the predicted KPI value. Alternatively or in addition, the comparison may comprise training or retraining the Al agent to predict the KPI value and / or the uncertainty value of the predicted KPI value. Alternatively or in addition, the comparison may comprise updating a stored value for the KPI value and / or the uncertainty value of the predicted KPI value for the respective action.

[0092] The stored value may be updated according to a running average or a long-time average, wherein the updated value is a weighted average of the stored value and the obtained value (i.e., according to the obtained information).

[0093] The method (e.g., according to the method aspect) may further comprise or initiate determining a risk factor (Y) for at least one or each pair of the action and the KPI based on the comparison.

[0094] There may be more than one KPI affected by the action or each action. There may be more than one pair comprising one action and one KPI. Accordingly, there may be more than one risk factor y(a, KPI , i = 1, ... , n, for each action a.

[0095] A method for determining the risk factor for the action and KPI pair (e.g., <a, KP I >) may be given as follows: That is, y(a, KPI, t) = 1 - min (^, 1).

[0096] For some intents (e.g., for a maximum latency of the wireless telecommunications infrastructure that is acceptable), the risk factor Y may be determined as:

[0097] The method for determining the risk factor may be based on an implementation of the autonomously operating of the wireless telecommunications infrastructure.

[0098] The risk factor may correspond to the probability that the KPI value falls below the threshold value KPI_t. The risk factor value may be normalized to be between [0, 1]. Alternatively or in addition, the risk factor may be a percentage and / or a number in any predefined range.

[0099] The method (e.g., according to the method aspect) may further comprise or initiate determining an overall risk factor based on a maximum value of all the risk factors determined according to one of the at least one intent.

[0100] For example, the maximum value may be determined over risk factors determined for different KPIs. A symbolic example of the determining of the overall risk factor may comprise:

[0101] The assessing (e.g., according to the method aspect) may comprise determining at least one of an increase of the utility for the action and a decrease of the utility for the action.

[0102] The increase of the utility may be referred to as a utility improvement, U^a). For example, current KPI .

[0103] The decrease of the utility may also be referred to as a utility reduction, URa). For example, the decrease may be assessed for a worst case scenario according to the difference between the KPI value prior to the action (i.e., a current KPI value) and the predicted lower bound of the KPI value:

[0104] UR(a = current KPI — KPI_l(a). An example for the criterion whether or not to execute the action a may be that the expected utility increase,

[0105] (1 — y(a, intent)) ■ U^d), is greater than the expected utility decrease, y(a, intent) ■ UR(d).

[0106] That is, y(n, intent) ■ tR(n) < (1 — y(n, intent)) ■ U^d).

[0107] The selectively executing of the action (e.g., according to the method aspect) may be based on a weighted combination of the determined increase of the utility and the determined decrease of the utility.

[0108] Another example for the criterion whether or not to execute the action a may be:

[0109] The factor (or weight) p may be an input parameter to the method of autonomously operating the wireless telecommunications infrastructure and / or determined by a user (e.g., human operator). The factor p may be set to greater than 1. If the factor p is zero, the action may be executed.

[0110] If the risk factor is zero, the action may be executed. Alternatively or in addition, if the risk factor is greater than zero, the autonomously operating may assess the case (e.g., according to any of the criteria disclosed herein). For example, if the risk factor or Cl is greater than a pre-determined value (e.g., x% that may be determined as an input parameter of the autonomously operating and / or by the user), the wireless telecommunications infrastructure may not select to execute the action. If the risk factor or Cl is less than a pre-determined value (e.g., less than x%) then the autonomously operating telecommunication network may select the action.

[0111] Alternatively or in addition, the autonomously operating may store the action and its related information (e.g., KPIs, Cl ...) for further consideration (e.g., processing) for future time.

[0112] The selectively executed action (e.g., according to the method aspect) may be based on minimizing an absolute value of a difference between the predicted KPI value and the at least one intent, optionally between the predicted KPI value and a threshold value according to the at least one intent.

[0113] For example, the assessing may comprise selecting an action from a list of multiple actions that assumes the minimum min a in list

[0114] The method (e.g., according to the method aspect) may further comprise or initiate updating (after executing the action, e.g. based on the information observed after executing the action) at least one the prediction of the KPI value for the action, the prediction of the uncertainty value of the predicted KPI value for the action, and the risk factor related to the action.

[0115] In order to further evaluate (e.g., assess) the prediction and to help improve the prediction accuracy needed for better performance of the autonomously operating of the wireless telecommunications infrastructure, at least one of the predictions (e.g. the risk factor, the lower bound of the KPI, the upper bound of the KPI, the Cl, etc.) related to the action selected for execution may be updated based on the effect observed after executing the action. For example, by retraining a prediction mechanism or updating stored values for the predicted KPI values and / or the risk factor related to the action (e.g., stored in association with the states resulting from the actions), future predictions of in the autonomously operating of the wireless telecommunications infrastructure can be improved.

[0116] For example, after executing the action a and observing results (e.g., mean value) of the KPI value (i.e. for the pair <a, KP I > of action a and KPI), the lower bound may be updated as follows:

[0117] LB a, KPI, t + 1) = ( LB a, KPI, t - 1) + (1 - / ?) * New LB a, KPI, t)

[0118] Herein, LB a, KPI, t + 1) may be the lower bound that may be used for the next predicting step of the method. The New LB a, KPI, t) may be the lower bound after the selected action is executed and the lower bound is updated with a new measurement. The LB a, KPI, t — 1) is the lower bound before the action is executed.

[0119] The method (e.g., according to the method aspect) may further comprise or initiate time-averaging at least one of the predicted and / or updated KPI value for the respective action; the predicted and / or updated uncertainty value of the KPI value for the respective action; the obtained information after executing the respective action; and the current KPI value.

[0120] The time-averaging may also be referred to as smoothing. The smoothed values may be stored in the memory. The smoothed values may be used by any one of the steps of the method and / or may be updated.

[0121] As to another aspect a computer program product is provided. The computer program product comprising program code portions for performing any one of the steps of the first method aspect when the computer program product is executed on one or more computing devices, optionally stored on a computer-readable recording medium.

[0122] The computer program product may be implemented in a network node (e.g., base station) of a RAN and / or a node of a core network and / or a radio device (e.g., a UE) acting as a relay or gateway to a RAN.

[0123] The computer program product comprises program code portions for performing any one of the steps of the method aspect disclosed herein when the computer program product is executed by one or more computing devices. The computer program product may be stored on a computer-readable recording medium. The computer program product may also be provided for download, e.g., via the radio network, the RAN, the Internet and / or the host computer. Alternatively, or in addition, the method may be encoded in a Field-Programmable Gate Array (FPGA) and / or an Application-Specific Integrated Circuit (ASIC), or the functionality may be provided for download by means of a hardware description language.

[0124] As to a device aspect, a node of a wireless telecommunications infrastructure is provided. The node comprises memory operable to store instructions and processing circuitry operable to execute the instructions, such that the node is operable to predict, for an action executable in the autonomous operation of the wireless telecommunications infrastructure, a value of a key performance indicator (KPI) of the wireless telecommunications infrastructure and a value of an uncertainty of the predicted KPI value. The node is further operable to assess an uncertainty of a utility based on the predicted KPI value and the predicted uncertainty value of the KPI value. The node is further operable to selectively execute the action. The selectivity is dependent on the intent and the assessed uncertainty of the utility.

[0125] The node (e.g., according to the device aspect) may be further operable to perform any one of the steps of the method aspect.

[0126] As to another device aspect, a node supporting a radio access network (RAN) in a wireless telecommunications infrastructure is provided. The node is configured to predict, for an action executable in the autonomous operation of the wireless telecommunications infrastructure, a value of a key performance indicator (KPI) of the wireless telecommunications infrastructure and a value of an uncertainty of the predicted KPI value. The node is further configured to assess an uncertainty of a utility based on the predicted KPI value and the predicted uncertainty value of the KPI value. The node is further configured to selectively execute the action. The selectivity is dependent on the intent and the assessed uncertainty of the utility.

[0127] The node (e.g., according to the other device aspect) may be further configured to perform any one of the steps of the method aspect.

[0128] The node (e.g., according to the device aspect or the other device aspect) may be a core node of a core network (CN) of the wireless telecommunications infrastructure. Alternatively or in addition, the node may be a network node of a radio access network (RAN) of the wireless telecommunications infrastructure, optionally a base station or a central unit (CU) of the network node. As to a system, aspect a communication system is provided. The communication system includes a host computer comprising processing circuitry configured to provide user data; and a communication interface configured to forward user data to a cellular or ad hoc radio network of a wireless telecommunications infrastructure for transmission to a user equipment (UE). The wireless telecommunications infrastructure comprises a radio interface and processing circuitry, the processing circuitry of the wireless telecommunications infrastructure being configured to execute any one of the steps of the method aspect.

[0129] The communication system (e.g., according to the system aspect) may further include a UE. The UE may comprise a radio interface and processing circuitry, the processing circuitry of the UE being configured for radio access to the wireless telecommunications infrastructure.

[0130] E.g. in the communication system according to the system aspect, the wireless telecommunications infrastructure may comprise one or more nodes configured to execute any one of the steps of the method aspect.

[0131] E.g. in the communication system according to the system aspect, the radio network may further comprise a base station, or a radio device functioning as a gateway, which may be configured to communicate with the UE and / or configured to execute any one of the steps of the method aspect.

[0132] E.g. in the communication system according to the system aspect, the processing circuitry of the host computer may be configured to execute a host application, thereby providing the user data; and a processing circuitry of the UE may be configured to execute a client application associated with the host application.

[0133] Any aspect of the technique may be applicable to any type of network, e.g. wireless or cellular or other.

[0134] The technique may be applied in the context of 3GPP New Radio (NR). The technique may be implemented in accordance with a 3GPP specification, e.g., for 3GPP release 17 or 18 or later.

[0135] Furthermore, the technique may be applied in the context of an IEEE standard, for example the IEEE 802.22 standard, which specifies Cognitive Radio (CR) systems for wireless regional area networks. The technique may be embodied based on or by extending the IEEE 802.22 standard, e.g. as to the physical and medium access control layers for Cognitive Radio systems (optionally for operation in TV bands, allowing efficient use of unused spectrum while avoiding interference with licensed users). Alternatively or in addition, the technique may be embodied based on publications of a Cognitive Networks Technical Committee (TCCN) at the IEEE Communications Society. Alternatively or in addition, the technique may be embodied based on or by extending the IEEE P1920.1 standard, which defines air- to-air communications for self-organized ad hoc aerial networks.

[0136] Moreover, the technique may be applied based on or by extending publications of the International Telecommunication Union (ITU), e.g. the ITU-T Recommendation E.712, for operation of cognitive and self-adaptive network systems.

[0137] The radio access network (RAN) may comprise one or more base stations, e.g., performing the method aspect. Alternatively or in addition, the radio network may be a vehicular, ad hoc and / or mesh network comprising two or more radio devices.

[0138] Any of the radio devices may be a 3GPP user equipment (UE) or a Wi-Fi station (STA). The radio device may be a mobile or portable station, a device for machinetype communication (MTC), a device for narrowband Internet of Things (NB-loT) or a combination thereof. Examples for the UE and the mobile station include a mobile phone, a tablet computer and a self-driving vehicle. Examples for the portable station include a laptop computer and a television set. Examples for the MTC device or the NB-loT device include robots, sensors and / or actuators, e.g., in manufacturing, automotive communication and home automation. The MTC device or the NB-loT device may be implemented in a manufacturing plant, household appliances and consumer electronics.

[0139] Whenever referring to the RAN, the RAN may be implemented by one or more base stations.

[0140] The base station may encompass any station that is configured to provide radio access to any of the radio devices. The base stations may also be referred to as cell, transmission and reception point (TRP), radio access node or access point (AP). The base station and / or the relay radio device may provide a data link to a host computer providing the user data to the remote radio device or gathering user data from the remote radio device. Examples for the network node (e.g., base station) may include a 3G base station or Node B (NB), 4G base station or eNodeB (eNB), a 5G base station or gNodeB (gNB), a Wi-Fi AP, and a network controller (e.g., according to Bluetooth, ZigBee or Z-Wave).

[0141] The RAN may be implemented according to the Global System for Mobile Communications (GSM), the Universal Mobile Telecommunications System (UMTS), 3GPP Long Term Evolution (LTE) and / or 3GPP New Radio (NR).

[0142] Any one of the devices, the core node, the base station, the communication system or any node or station for embodying the technique may further include any feature disclosed in the context of the method aspect, and vice versa. Particularly, any one of the units and modules disclosed herein may be configured to perform or initiate one or more of the steps of the method aspect.

[0143] Brief Description of the Drawings

[0144] Further details of embodiments of the technique are described with reference to the enclosed drawings, wherein:

[0145] Fig. 1 shows a schematic block diagram of an embodiment of a device for autonomously operating a wireless telecommunications infrastructure according to at least one intent;

[0146] Fig. 2 shows a flowchart for a method of autonomously operating a wireless telecommunications infrastructure according to at least one intent, which method may be implementable by the device of Fig. 1;

[0147] Fig. 3 shows a schematic diagram of an intent management function according to the prior art;

[0148] Fig. 4 shows a schematic block diagram of an embodiment of the device of Fig. 1;

[0149] Fig. 5 shows the predicted values for a KPI and observed KPI values according to exemplary embodiments of the device of Fig. 1 and the method of Fig. 2. Fig. 6 shows a schematic diagram according the same or another embodiment of the method of Fig. 2.

[0150] Fig. 7 shows an exemplary criterion for executing the action according to an embodiment of the method of Fig. 2.

[0151] Fig. 8 shows exemplary possible problem addressed by an embodiment of the method of Fig. 2 and an exemplary criterion for refraining from executing the action according to an embodiment of the method of Fig. 2.

[0152] Figs. 9A-9D schematically illustrates types of utility functions as functions of the KPI, which may be applicable in any embodiment of the device of Fig. 1 and the method of Fig. 2.

[0153] Fig. 10 shows a schematic sequence diagram according to an embodiment of the method of Fig. 2.

[0154] Fig. 11 shows a schematic block diagram of a network node embodying the device of Fig. 1;

[0155] Fig. 12 shows a schematic block diagram of a core node embodying the device of Fig. 1;

[0156] Fig. 13 schematically illustrates an example telecommunication network connected via an intermediate network to a host computer;

[0157] Fig. 14 shows a generalized block diagram of a host computer communicating via a base station or radio device functioning as a gateway with a user equipment over a partially wireless connection; and

[0158] Figs. 15 and 16 show flowcharts for methods implemented in a communication system including a host computer, a base station or radio device functioning as a gateway and a user equipment. Detailed Description

[0159] In the following description, for purposes of explanation and not limitation, specific details are set forth, such as a specific network environment in order to provide a thorough understanding of the technique disclosed herein. It will be apparent to one skilled in the art that the technique may be practiced in other embodiments that depart from these specific details. Moreover, while the following embodiments are primarily described for a New Radio (NR) or 5G implementation, it is readily apparent that the technique described herein may also be implemented for any other radio communication technique, including a Wireless Local Area Network (WLAN) implementation according to the standard family IEEE 802.11, 3GPP LTE (e.g., LTE-Advanced or a related radio access technique such as MulteFire), for Bluetooth according to the Bluetooth Special Interest Group (SIG), particularly Bluetooth Low Energy, Bluetooth Mesh Networking and Bluetooth broadcasting, for Z-Wave according to the Z-Wave Alliance or for ZigBee based on IEEE 802.15.4.

[0160] Moreover, those skilled in the art will appreciate that the functions, steps, units and modules explained herein may be implemented using software functioning in conjunction with a programmed microprocessor, an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), a Digital Signal Processor (DSP) or a general purpose computer, e.g., including an Advanced RISC Machine (ARM). It will also be appreciated that, while the following embodiments are primarily described in context with methods and devices, the invention may also be embodied in a computer program product as well as in a system comprising at least one computer processor and memory coupled to the at least one processor, wherein the memory is encoded with one or more programs that may perform the functions and steps or implement the units and modules disclosed herein.

[0161] Fig. 1 schematically illustrates a block diagram of an embodiment of a device for autonomously operating a wireless telecommunications infrastructure according to at least one intent. The device is generically referred to by reference sign 100.

[0162] The device 100 comprises a prediction module 102 which, for an action executable in the autonomous operation of the wireless telecommunications infrastructure, predicts a value of a key performance indicator (KPI) of the wireless telecommunications infrastructure and predicts a value of an uncertainty of the predicted KPI value. The prediction module 102 may provide more KPI values predictions for one or multiple possible (e.g. proposed) actions and their corresponding uncertainty values. The prediction module 102 may further provide a measured KPI value and / or a measured uncertainty value after an action has been taken (e.g., an actual KPI value and / or an observed standard deviation of the KPI value) and / or may update its prediction according to the actual KPI value and / or the observed standard deviation of the KPI value.

[0163] The device 100 further comprises an assessing module 104 that assesses (e.g., evaluates) an uncertainty of a utility based on the predicted KPI value and the predicted uncertainty value of the KPI value. The assessing module may further assess multiple uncertainties of multiple utilities based on the corresponding predicted KPI values and the predicted uncertainty values of the KPI values.

[0164] The device 100 further comprises an action module 108 that selectively executes the action. The selectivity may be dependent on the intent and the assessed uncertainty of the utility. For example, the intent may specify the utility function or the predefined threshold value for the KPI or the utility.

[0165] Any of the modules of the device 100 may be implemented by units configured to provide the corresponding functionality.

[0166] The device 100 may be a computer program product comprising program code portion for performing the method of autonomously operating a wireless telecommunications infrastructure according to the at least one intent.

[0167] The device 100 may be implemented in a node of the wireless telecommunications infrastructure. The node and the components of the wireless telecommunications infrastructure controlled according to the action may be in communication (e.g., via a backhaul network), e.g. for executing the action and / or obtaining the information after the action. The node may be a network node 1100 and / or core node 1200 of a wireless telecommunication infrastructure, e.g. as described below.

[0168] Fig. 2 shows an example flowchart of a method 200 for autonomously operating a wireless telecommunications infrastructure according to at least one intent. Optionally, the method 200 may be performed according to more than one intent. The autonomous operation of the wireless telecommunications infrastructure may comprise at least one closed loop, e.g. taking information as to a state of the wireless telecommunications infrastructure resulting from a previous action into account for performing the method 200. Alternatively or in addition, each closed loop may comprise at least one intent.

[0169] In a step 202, for an action (e.g., each proposed action) executable in the autonomous operation of the wireless telecommunications infrastructure, the method 200 may predict 202 a value of a KPI of the wireless telecommunications infrastructure and a value of an uncertainty of the predicted 202 KPI value.

[0170] The intent may comprise multiple utilities, each of which is dependent on at least one KPI value or the wireless telecommunications infrastructure may be operated autonomously according to multiple intents, each of which comprises at least one utility, each of which is dependent on at least one KPI value.

[0171] Examples of the KPI value comprise at least one of jitter, latency, bit rate, block error rate (BLER), packet loss rate (PLR), signal to noise ratio (SNR), and signal to interference and noise ratio (SINR) or any other physical value monitored or measured in the wireless telecommunications infrastructure. Alternatively or in addition, the KPI may comprise non-physical parameters such as service type, network congestion, application requirements, or user expectations.

[0172] The predicted (in step 202) uncertainty value of the predicted KPI value may comprise at least one of a lower bound for the KPI value (KPI / ), an upper bound for the KPI value (KPIu), a confidence interval (Cl) of the predicted KPI value, and a prediction interval of the predicted KPI value. Whenever referring to Cl herein, the technique may also be implemented using the prediction interval instead.

[0173] Alternatively or in addition, the predicted (in step 202) uncertainty value of the predicted KPI value may further comprise a probability of the predicted KPI value or a probability for the KPI value being less than the predicted KPI value or a probability for the KPI value falling below a predefined threshold value.

[0174] Optionally the prediction 202 of the KPI value and / or the uncertainty value of the KPI value may be performed by an artificial intelligent agent (Al agent). In a step 204, the method 200 assesses an uncertainty of a utility based on the predicted (in step 202) KPI value and the predicted (in step 202) uncertainty value of the KPI value.

[0175] The utility may be a function of the KPI value, such as a step function, a linear function and / or any other mathematical function.

[0176] The assessing step 204 may comprise determining 206 at least one of an increase in the utility for the action and a decrease in the utility for the action.

[0177] The assessed 204 uncertainty of the utility may comprise a probability of a decrease of the utility (e.g., utility reduction) or a probability of the utility falling below a predefined threshold value.

[0178] The assessing 204 of the uncertainty may further comprise determining whether a probability for the action increasing the utility is greater than a probability for the action decreasing the utility and / or whether a predicted increase of the utility greater than a predicted decrease of the utility.

[0179] Optionally, the assessing 204 of the uncertainty comprises determining an expectation value or lower bound of the utility based on the predicted 202 KPI value and the predicted 202 uncertainty value of the KPI value.

[0180] In a step 208, the method selectively executes (e.g., perform) the action. The selectivity may depend on the intent and the assessed 204 uncertainty of the utility. Alternatively or in addition, the method 200 may be performed for more than one given action for each of the at least one intent.

[0181] Optionally, the action may be executed 208 if the expectation value or lower bound of the utility corresponds to an increase of the utility. Alternatively or in addition, the action may not be executed 208 if the expectation value or lower bound of the utility corresponds to a decrease of the utility.

[0182] The action may be executed 208 if a criterion of the intent is fulfilled (e.g., met) for the assessed 204 uncertainty of the utility. Optionally, a further action may be executed 208 if a further criterion is fulfilled for the assessed 204 uncertainty of the utility or for a further assessed 204 uncertainty of a further utility. The action may not be executed 208 if the further criterion for executing the further action is fulfilled. This look-ahead mechanism may prevent alternating states or contradicting actions.

[0183] Alternatively or in addition, the action 208 may selectively be executed and the selectivity may be dependent on a combination of the uncertainty assessed in the step 204 for each of the utilities. Alternatively or in addition, the action may not be executed 208 if at least one criterion for the uncertainty assessed 204 for each of the utilities is not fulfilled. Alternatively or in addition, the action 208 may be executed, if each criterion for the uncertainty assessed in the step 204 for each of the utilities is fulfilled.

[0184] The action may be executed in the step 208, if the lower bound or the confidence interval or the prediction interval corresponds to an increase of the utility. Alternatively or in addition, the action may not be executed in the step 208 if the lower bound or the confidence interval or the prediction interval corresponds to a decrease of the utility. Alternatively or in addition, the utility may be a function of the KPI value with a non-linear characteristic. Alternatively or in addition, the action may be executed 208 if a KPI value of the non-linear characteristic is below the lower bound or the confidence interval or the prediction interval. Alternatively or in addition, the action may not be executed 208 if the KPI value of the nonlinear characteristic is within or above the lower bound or the confidence interval or the prediction interval.

[0185] Alternatively or in addition, the selectively executing 208 of the action may be based on a weighted combination of the determined 206 increase of the utility and the determined 206 decrease of the utility.

[0186] The selectively executed 208 action may be based on minimizing an absolute value of a difference between the predicted KPI value and the at least one intent, optionally between the predicted KPI value and a threshold value according to the at least one intent.

[0187] The action selectively executed in the step 208 may be based on minimizing an absolute value of a difference between the predicted KPI value and the at least one intent, optionally between the predicted KPI value and a threshold value according to the at least one intent. Optionally, in a step 210, the method 200 may comprise obtaining information about a state of the wireless telecommunication infrastructure as a result of and / or after executing the action in the step 208. The obtained information in the step 210 may comprise at least one of a measured KPI value (e.g., a time sequence of KPI values); a measured KPI lower bound (KPI_I or KPI / ); a measured KPI upper bound (KPI_u or KPIu); and a measured confidence interval (Cl, e.g. an estimate of the standard deviation), etc.

[0188] Optionally, in a step 212, the method 200 may compare the obtained information of the step 210 (e.g., after executing the action in the step 208) with the KPI_p value predicted in the step 202 and / or the uncertainty value of the predicted KPI_p value predicted in the step 202. Optionally, the comparison 212 may comprise training or re-training the Al agent to predict 202 the KPI value and / or the uncertainty value of the predicted KPI_p value. Optionally, the comparison 212 may comprise updating a stored value for the KPI value and / or the uncertainty value of the predicted KPI_p value for the respective action.

[0189] Optionally, in a step 214, the method may perform determining a risk factor (Y) for at least one or each pair of the action and the KPI value based on the comparison 212.

[0190] Optionally, in a step 216, the method may perform determining an overall risk factor based on a maximum value of all the risk factors determined 214 according to one of the at least one intent.

[0191] Optionally, in a step 218, the method may perform updating, after executing the action in the step 208, the prediction 202 (e.g., a prediction agent). The updating of the prediction 202 may relate to at least one of the KPI value for the action; the uncertainty value of the predicted 202 KPI value for the action; and the risk factor associated with executing the action (if it is selected for execution in the step 208).

[0192] Optionally, in a step 220, the method may perform time-averaging at least one of the predicted 202 and / or updated 218 KPI value for the respective action; the predicted 202 and / or updated 218 uncertainty value of the KPI value for the respective action; the information obtained in the step 210 after executing the respective action in the step 208; and the current KPI value. Embodiments of the method 200 can improve the autonomously operating of the wireless telecommunications infrastructure, e.g. compared to a conventional operated solely based on expectation values in the state of the art, in at least two respects:

[0193] - The method 200 improves the capability of the evaluation agent 104 by considering the uncertainty of the utility, e.g. the prediction confidence and / or the risk of a loss in utility and / or the risk of one or more further actions as subsequent corrective actions responsive to the action. This results in a higher system utility and / or system stability.

[0194] - The method 200 improves the autonomously operating a wireless telecommunications infrastructure and may provide more information to a user (e.g., an operator of the wireless telecommunications infrastructure) regarding what the impact of an action may be by also providing a risk factor. Depending on the risk factor, the operator may decide whether the action shall be applied or not. An action may be a good but also risky, and the operator may judge this using an embodiment of the method 200. For example, the intent may be further indicative of risk aversion (e.g., as quantified by the ratio p).

[0195] To manage the intents, autonomously operating a wireless telecommunications infrastructure (e.g., telecommunications networks) may use closed loops in various domains. A closed loop is a management structure where the system works on a managed entity (e.g., components of the wireless telecommunications infrastructure) with a specific goal (e.g., the intent) and is able to monitor and take action on it via a feedback loop. Each closed loop comprises several agents.

[0196] Fig. 3 shows an example of the architecture of an Intent-based Automation (IBA), which supports the implementation of closed loops according to prior art. Data grounding agents can collect raw data describing the state of the managed environment (i.e., the wireless telecommunications infrastructure). Once issues are identified and stored in the knowledge base, a cognitive core can create goals and select agents to propose actions as solutions. Prediction agents estimate the impact of proposals on the system state before the proposed actions are recommended and executed in the managed environment. When predictions are available, the system estimates their impact (e.g., effect) on all active expectations. The evaluation agent receives the estimated impact of all proposed actions on all active expectations and selects the best action in terms of global network utility or cost. In doing so, this component must identify conflicting actions - actions that improve some expectations (unfulfilled expectations become fulfilled expectations) but degrade other expectations (fulfilled expectations become unfulfilled expectations).

[0197] An example of a possible action might be an action that changes the maximum bit rate (MBR) of UEs registered with the service, or changes the priority level of those services, or moves the service from the cloud to the edge. An action selected to improve a KPI value of one service may degrade the KPI value of another service (or its service). For example, an action may be to increase the priority of the WebRTC service to improve its QoE, but this may negatively impact the packet loss of the URLLC service. This situation can lead to a conflict, which is resolved at the evaluation agent after receiving all the predicted KPI values for each proposed action. Therefore, a good level of prediction is crucial for the IBA to work efficiently.

[0198] In the prior art IBA framework, it is assumed that the impact of a proposed action is predicted without any prediction error and that perfect prediction information is provided to the evaluation agent. This assumption is very strong and cannot hold in practice as there is always some prediction error. The degree of the prediction error can significantly affect the decision of the evaluation agent and therefore can degrade the overall system performance.

[0199] For example, if a good action is proposed, but the prediction is not accurate for some KPI values, then this may lead to a lower network utility, as illustrated by an example below. Suppose there are two intents, Intent A and Intent B. Intent A has a latency expectation (KPI) with a target of 30 ms while Intent B has a QoE KPI value (expectation) with a target of 4. If an intent is not satisfied, it brings 0 utility. If Intent A is satisfied, the utility will be 10. The utility of satisfying intent B is 20.

[0200] Intent A (Utility = 10) Intent B (Utility = 20)

[0201] KPI A -> latency target = 30 ms KPI B -> QoE target = 4 Suppose the proposal agent proposes action a. An example of the effect before and after the action on both KPIs is given below.

[0202] KPI A before action = 40 ms KPI B before action = 4.2

[0203] KPI A predicted for action a = 25 ms KPI B predicted for action a =4.1

[0204] Suppose the following actual KPI values are measured after applying action a:

[0205] Actual KPI A after action applied = 26 ms Actual KPI B after action a = 4.05

[0206] The action a is predicted to improve KPI A and not to degrade KPI B (the value of KPI B will still be above the target). So it is a good action and is applied. The utility before the action a is 20, as KPI A is not met. However, after the action, both KPI values are satisfied, and the total utility becomes 30. Now consider the same scenario, but with a different outcome after action a.

[0207] Propose action a:

[0208] KPI A before action = 40 ms KPI B before action = 4.2

[0209] KPI A predicted for action a = 25 ms KPI B predicted for action a =4.1

[0210] Actual KPI A after action applied = 26 ms Actual KPI B after action a = 3.8

[0211] The only difference is that now, after the action a is applied, the actual KPI B is reduced to 3.8, but it was predicted to be 4.1, i.e. the prediction was not perfect. In this case, the evaluation agent still decides that the action a is good and applies it. However, because of the erroneous prediction, the evaluation agent also makes an erroneous decision, which causes the system utility to decrease to 10. Also, in the next round, another action is proposed to fix KPI B. This can lead to an unstable loop, because now the fix for KPI B can degrade KPI A again and it will go to the initial state to fix KPI A again.

[0212] Therefore, an accurate prediction of the KPI values after an action has been proposed is critical for the evaluation agent to make a good assessment. Given this fact, the prior art IBA architecture lacks this capability. The problem to be addressed may be how to account for and handle the prediction error on KPI values, as well as how to handle the risk caused by a bad prediction, so that the evaluation agent can make a good decision in terms of global system utility. Fig. 4 shows an embodiment of an architecture of Intent-based Automation (IBA) implementing the device 100 for autonomously operating a wireless telecommunications infrastructure according to an embodiment of the method 200. More specifically, the method 200 may be performed by an intent manager for the autonomously operating of the wireless telecommunications infrastructure. An operator (e.g., a user) may predefine the at least one intent via an intent manager framework. The intent manager framework may comprise a reasoner module and a memory for storing raw data and the processed data, network state, etc.

[0213] The intent manager framework may be in communication with at least one data grounding agent. The data grounding agent may receive raw data. The data grounding agent may send the raw data and / or metadata and / or properties to the intent manager framework.

[0214] The intent manager framework may be further in communication with at least one proposal agent. The proposal agent may be further in communication with the data grounding agent, e.g., it may receive raw data from the data grounding agent. The proposal agent may process the received raw data and make a proposal for at least one action according to the intent and the KPI value. The proposal agent may be further in communication with the prediction module 102 (e.g., a prediction agent).

[0215] The prediction module 102 may be in communication with the intent manager framework. The prediction module 102 may be further in communication with the assessing module 104 (e.g., an evaluation agent). The prediction module 102 may also send to the assessing module 104 the predicted uncertainty value for the KPI, e.g. the best and / or the worst predictions (e.g., prediction bound: lower and upper bound of the prediction) of the KPI, and / or a confidence interval (Cl) for these KPI predictions.

[0216] The prediction module 102 may send the predictions of the action on all KPI values and it sends these predicted KPI values to the assessing module 104 with the (at least one) proposed (e.g., recommended) action a. The prediction module 102 may also send its best and / or worst predictions (prediction bound: lower and upper bound of the prediction). More specifically, the prediction module 102 may send a tuple (e.g., a 5-tuple) to the assessing module 104:

[0217] <a, KPI_p, KPI_u, KPI_I, Cl>.

[0218] For example, an action a may be changing a priority (e.g., of certain data packets transmitted and received in the wireless telecommunications infrastructure). A corresponding example of the KPI value may be latency. KPI_p may be the predicted KPI (e.g., 25 ms). KPI_u may be the upper bound of the prediction (e.g., 30 ms) and KPI_I is the lower bound on the prediction (e.g., 22 ms). The Cl may be also the confidence level, which indicates what percentage of the time (e.g., %95) the predicted value falls in between KPI_u and KPI_I . The KPI_u and KPI_I may be calculated based on a prediction interval calculation.

[0219] Alternatively or in addition, there may be multiple prediction modules 102, e.g., for each KPI value with different capabilities.

[0220] For example, some prediction modules 102 may not be able to provide the prediction bounds and confidence intervals. These prediction modules 102 may be labelled as untrustworthy agents, and they may be skipped if there are other prediction modules 102 that are able to provide these information. Alternatively or in addition, obtaining this information from the prediction module 102 may not be difficult, as this information has a well-established scientific and practical background. There may be different methods for obtaining the prediction bounds and these methods may have different theoretical and practical limitations when using neural network (NN) based prediction.

[0221] When an action is proposed and / or executed (e.g., taken), each prediction modules (e.g., latency KPI) may send a new tuple <a', KPI_p', KPI_u', KPI_I', Cl'> to the assessing module(s) 104.

[0222] The assessing module 104 may perform a prediction assessment 204 that uses not only the predicted KPI_p but also the possible bounds of the prediction (e.g., using the 5-tuple).

[0223] The assessing module 104 may assess the proposed action(s) and the predicted KPI_p and its uncertainties (e.g., the 5-tuple). The assessing module(s) 104 may further assess the predicted KPI_p resulting from the action taken. By using this additional information (e.g., 5-tuple, KPI value and its uncertainties), the prediction assessment 204 is more reliable. The intent manager may make a better decision about which action to take based on the more reliable prediction. In other words, the assessing module 104 may assess the choice of action based on the assessed and reliable prediction(s) information.

[0224] The assessing module 104 may measure at least one risk (and / or type of risks) caused by the prediction errors. An explicit risk factor may be determined for the at least one intent. The risk factor may be used in the assessing module 104 for an improved decision making (e.g., increased system utility). The assessing module 104 may pass the result of the assessment to the intent manager framework. The intent manager framework and / or the assessing module 104 may select an action among all the possible actions based on the assessing step 204. The selected (e.g., decided) action may be forwarded to the action module 108 (e.g., actuator agent) for executing the action.

[0225] Alternatively or in addition, the prediction assessment 204 may be performed at the evaluation agent 104 or at a separate module, depending on the implementation.

[0226] After assessing the prediction 204, the assessing module 104 may be more confident about the action it decides to take, and the final action is applied to the real network. The intent manager may then track of the impact of the applied action on the KPI values to assess whether the expected impact is actually achieved and the predictions on the KPI values are as expected. The intent manager may then compare the current KPI values measurement with the history of KPI values measurements to make a further assessment 204. The intent manager may update 218 the information associated with the KPI values, such as the prediction bounds and risk factors, with new information to use in future decision makings.

[0227] Fig. 5 shows an example of the predicted KPI value and the associated uncertainties (e.g., a prediction interval) provided by the prediction module 102, e.g. based on previously measured values of the KPI (black squares). The prediction module 102 may predict (in the step 202) for a target state 1 a KPI value KPI_pl (indicated at reference sign 514) and the associated uncertainties. The associated uncertainties may be understood as the upper prediction bound KPI_ul (indicated at reference sign 512) and lower prediction bound KPI_11 (indicated at reference sign 516). As schematically illustrated in Fig. 5, the prediction may be based on time series data (i.e., the black squares) as the mean value or median for the predicted KPI value 514. The bounds 512 and 516 may correspond to percentiles x% and (100-x)%, e.g. for x=90. Alternatively or in addition, the Cl may indicate what percentage of the time (e.g., 95%) the actual KPI value for the KPI falls in between KPI_ul 512 and KPI_I1 516.

[0228] Similarly, the predicted KPI value and the associated uncertainties (e.g., a prediction interval) provided by the prediction module 102 for a target state 2 may comprise a KPI value KPI_p2 (indicated at reference sign 524) and the associated uncertainties. The associated uncertainties may be the upper prediction bound KPI_u2 (indicated at reference sign 522) and lower prediction bound KPI_I2 (indicated at reference sign 526). As schematically illustrated in Fig. 5, the prediction may be trained based on time series data (i.e., the black squares) as the mean value or median for the predicted KPI value 514. The bounds 512 and 516 may correspond to percentiles x% and (100-x)%, e.g. for x=90. Alternatively or in addition, the Cl may indicate what percentage of the time (e.g., 95%) the actual KPI value for the KPI falls in between KPI_ul 512 and KPI_11 516.

[0229] Herein, prediction 202 may refer to providing both the predicted KPI value and the predicted uncertainty value for the predicted KPI value.

[0230] Furthermore, the prediction 202 may be stateless, i.e., the prediction 202 may depend only on the target state of an action. Alternatively, the prediction 202 may depend on both the source state 1 and the target state 2, in which case the prediction may be trained based on measured values for the KPI as indicated in Fig. 5 for an action a causing the transition from the state 1 to the state 2.

[0231] As a basis for assessing 204 an action a that causes the transition of the wireless telecommunications infrastructure from state 1 to state 2, the prediction module 102 may predict 202 for the target state (i.e., the state resulting after the selected action is executed 208, which is state 2 in the example shown in Fig. 5) a KPI value KPI_p2 shown at reference sign 524 and the associated uncertainties 522 and 526. The assessing module 104 may send its decided (i.e., selected) action to the action module 108 to execute 208 the action.

[0232] The prediction interval and uncertainties provided by the prediction module 102 may be different before an action is taken and after the action is taken. The postaction prediction information (e.g., post action 5-tuple) may be used to update 218 the knowledge base in the intent manager framework for future assessments 204.

[0233] Compared to the conventional architecture in Fig. 3, the assessing step 204 (e.g., prediction assessment) performed by the assessing module 104 may be included as a further step to the IBA in order to assess the consequences of the prediction 202 (e.g., in state 2) and the prediction error (e.g., the uncertainties associated to the KPI_p) to improve the decision of the assessing module 104 (e.g., evaluation agent), e.g. to improve the stability of the states during the autonomous operation.

[0234] The embodiment of the method 200 described with reference to Fig. 4 and 5 may be implemented according to Fig. 6.

[0235] The assessment performed by the assessing module 104 may comprise two steps:

[0236] - Firstly, it may determine a risk factor for each action-KPI value pair and finds a final risk factor for each of the at least one intent.

[0237] - Secondly, it may estimate a potential reduction of the utility based on the prediction 202 including the KPI uncertainty and / or due to the risk factor.

[0238] The prediction assessment 204 may be located at the evaluation agent (shown at reference sign 104 in Fig. 4) or it may be implemented by a separate entity.

[0239] Alternatively or in addition, the prediction assessment 204 and / or the function of the evaluation agent may be implemented by the assessing module 104.

[0240] The prediction assessment 204 (e.g., as a further step in IBA) receives the values <a, KPI_p, KPI_u, KPI_I, Cl> for each action from the prediction module 202, and decides whether it is good to consider this action or not. At least one of the following steps may be performed during a phase of the prediction assessment 204. The selection of the action in the assessing step 204 may be based on the following criterion, e.g. by determining that the following Case 1 and not Case 2 is fulfilled. Examples for the Cases 1 and 2 are illustrated in Figs. 7 and 8, respectively.

[0241] As an example possible case, Case 1, both KPI_u and KPI_I are greater [or lower] than the target KPI value KPI_t (e.g., for a KPI that is greater if better for the utility [or for a KPI that is greater if worse for the utility, respectively]). Optionally, Case 1 may further require that the Cl is greater than a pre-determined value (e.g., x%).

[0242] In Case 1, the evaluation agent 104 may decide (i.e., in the step 204) to consider the action a with the given prediction of the KPI value in the decision. The predetermined value x may be defined by, for example, the network, and / or the operator, and / or it may depend on the intent (e.g., application). The network may require each prediction module 102 to have at least x% Cl.

[0243] As a consequence of the Case 1, the predicted KPI value KPI_p 524 and the actual KPI values 702 are likely (e.g., according to the Cl) both above the target KPI value KPI_t 704 [or both below the target KPI value KPI_t 704]. That means, in Case 1, the assessing module 104 is not confused by the predicted KPI value. For example, the predicted value for the KPI is reliable and stable also in terms of the utility, which may sharply drop below the target value 704 for the KPI. Hence, the action is selected to be executed or selected to be assessed (e.g. in comparison with alternative actions) by the assessing module 104, which is able to make a correct decision based on the predicted value for the KPI.

[0244] In another example case, Case 2, the target KPI value KPI_t 704 is greater than the lower predicted bound KPI_I 526 and less than the upper predicted bound KPI_u 522, and / or the Cl is less than x%. In Case 2, the assessing module 104 may not consider the action at the time of the decision. Optionally, it may still be stored for further consideration (or assessment) at a later time. Not selecting the action can prevent a loss in terms of utility.

[0245] For example, in the Case 2, the predicted KPI value KPI_p 524 may be above the target KPI value 704, but the actual KPI value 702 after executing the action may be below the target [or the predicted KPI value can be below and the actual KPI value can be above the target KPI value 704 for a KPI that is greater if worse for the utility]. In this case, an evaluation agent according to the prior art would determine that the KPI value and / or the corresponding utility will not degrade, so the action will not harm the KPI value, and the action may be considered for executing. However, the actual KPI value 702 is degraded and may cause a higher cost (i.e. a loss in terms of utility).

[0246] According to a variant of any embodiment of the method 200, the one or more actions that pass the phase of the prediction assessment 204 (e.g., taking the uncertainty into account) are evaluated at the evaluation agent 104 (e.g., based on the predicted KPI value). Preferably, untrustworthy predictions and / or problematic action-predictions pairs, such as the Case 2 illustrated in Fig. 8, are not take into consideration by the assessing module 104.

[0247] According to a further variant of any embodiment, there is a risk factor (e.g., for a prediction module 104 with poor prediction performance) for the prediction falling into the Case 2 (e.g., as schematically illustrated in Fig. 8). The method 200 may be implemented as a method for selecting and / or ranking the prediction module 104 or the corresponding prediction in Case 2.

[0248] Each action-KPI value pair may be associated with a risk factor, denoted as Y(a,KPi)- This risk factor measures the risk level of Case 2. For an intent, there may be multiple KPI values that need to be met. Each of these KPI values may have different risk factors, and for the intent, the overall risk factor may be defined as the maximum of Y(a,KPi -

[0249] The risk factor for the intent may be defined as:

[0250] Y(a, intent) = max(y(a,KPI)) -

[0251] The risk factor may be normalized to be between [0, 1]. As an example:

[0252] Intent A Intent B

[0253] Effect of Action a Effect of Action a KPI(B,1) Y(a,KPi(B,i)) “ Case 2

[0254] KPI(A,2) Y(a,KPi(A,2)) - Case 2 KPI(B,2) Y(a,KPi(B,2)) “ Case 2

[0255] KPI(B,3) Y(a,KPI(B,3)) - Case 1, risk 0 As indicated above, the action a leads to Cases 1 for the KPI(A,1) of the intent A and KPI(B,3) of the intent B. But for the other KPIs of these intents, the risk factor is non-zero.

[0256] There are several methods to determine the risk factor for the action-KPI value pair <a, KP I >. One of the methods may be, e.g. considering Case 2 in Fig. 8, when the prediction is above the target, the risk of the actual value falling below the target can be given as: lower bound(t)

[0257] The risk factor in general may depend on the implementation. For example, the above formula may assume an equal distribution of probability in terms of the KPI. For other probability distributions, the probability of the KPI value falling below the target (i.e., the KP l_t) may the risk factor.

[0258] A further variant of any embodiment may decide on an action a with utility reduction based on the associated risk factor.

[0259] Suppose a problem is observed (i.e., a trigger to ensure the at least one intent, e.g., an unmet KPI value), and the proposal agent proposes one or more actions a to fix it. For brevity, it is referred to the one or more proposed actions in singular form herein below. The action may improve the unmet KPI value and the corresponding intent will improve. This improvement will result in some utility improvement.

[0260] If the action is applied it and fixes the observed problem, the utility improvement associated with the taken action may be determined. The utility improvement may be denoted as Ui (a,t).

[0261] In the same example or another example, the proposed action may degrade the other KPI values that are already being met. The degradation in these KPI values may cause some reduction in utility. Using the risk factor, the prediction bounds, and the confidence interval, the expected reduction in utility due to the prediction error may be determined. The utility reduction may be determined depending on the implementation. The reduction may be denoted as UR (a,t). If the improvement (e.g., increase) in utility determined above, is greater than the expected decrease in utility determined above, then the action is accepted to be executed (e.g., applied). In other words, if the following condition holds, the action is accepted:

[0262] Ui(a.t) > pUR(a,t)

[0263] Herein, p is a ratio that may be determined by the network (e.g., system). The ratio p may be set to be greater than 1. If the risk factors are zero, the action is always considered.

[0264] The risk factor may explicitly affect the utility reduction. The utility may be defined in terms of the achieved one or more KPI values. For brevity, only one KPI is considered, i.e. U=U(KPI). In other words, the utility may be a function of the achieved KPI value. The form of this function may vary. Figs 9A to 9D schematically indicate examples of the utility as a function of the KPI value. Depending on whether a greater KPI corresponds to an improvement or deterioration, the utility is either a monotonically increasing function of the KPI or a monotonically decreasing function of the KPI.

[0265] These utility-KPI functions (e.g. mappings) can be used to determine how much utility can be obtained by achieving a certain value of the KPI. For example, for the step function in Fig. 9D, if the KPI is below its target KPI_t, the utility is zero. The form of the utility function may be predefined (e.g., by an operator). In the nonlinear forms of the utility function, e.g. according to Fig. 9B or 9C, if the KPI value is above its target, the additional utility achieved may not be large.

[0266] In Case 2, the maximum KPI value reduction may be calculated as follows.

[0267] The KPIR (a,t) may be defined as the maximum KPI value reduction due to the possibility of Case 2. Since the predicted KPI value KPI_p 524 will fall within the upper prediction bound KPI_u 522 and lower prediction bound KPI_I 526 with a high degree of confidence, the maximum KPI value reduction can be given as follows: In other words, if the Case 2 occurs, the KPI value may be reduced up to its lower bound, and the probability of this case occurring is related to the risk factor. One can determine the predicted KPI value KPI_p by using the maximum KPI value reduction and can map the predicted KPI value to the new utility value by using the functional form of utility and KPI value (e.g., according to one of Figs. 9A-9D). The assessing step 204 may comprise at least one of the following sub-steps:

[0268] 1. Obtaining the risk factor, e.g. the risk factor of a KPI or the risk factor of an intent (e.g. for multiple KPIs);

[0269] 2. Determining the maximum possible KPI value reduction KPIR (a, t), e.g. by using equation above (Maximum KPI reduction). Here, in the worst case, the KPI value resulting from the action will be around its lower bound;

[0270] 3. Determining the predicted KPI value KPI_t, optionally based on and / or after determining how much the KPI value may be reduced using sub-step 2;

[0271] 4. Mapping the predicted KPI value KPI_t to the utility value U=U(KPI_t) by using the given utility-KPI function; and

[0272] 5. Determining a reduction of the utility (i.e., utility loss), UR (a,t), e.g. according to UR=U(current KPI)-U(KPI_t).

[0273] Fig. 10 shows a sequence diagram including additional steps, which may be included in any embodiment of the method 200, e.g. for re-training the prediction module 202 and / or as a basis for selectively executing the decided action a* in the step 208.

[0274] Once it has been decided to apply action a*, the following Table may be created or updated, e.g. filled in, to further assesses 204 the prediction 202 or improve the prediction 202 accuracy.

[0275] Table

[0276] After a decision has been made to execute 208 an action, the autonomously operating a wireless telecommunications infrastructure may track the KPI values to evaluate the actual impact of the taken action on all KPI values as measurements, i.e. information obtained 210, at least one of which may be stored in Table.

[0277] Optionally, the Table comprises a column indicative of the lower bound KPI_I and / or the upper bound KPI_u, e.g. based on the KPI values (e.g., the black squares shown in Fig. 5).

[0278] Based on the information obtained 210, the method 200 can further assess the impact of the action and the prediction error. From the Table, the intent manager framework may check whether the prediction has actually performed as expected in the prediction assessment 204 phase. The information obtained in the step 210 (i.e., the new data) may be used to update at least one of the bounds, the Cl and the risk factor of each <action, KP I > in the step 218, e.g. as follows:

[0279] First, update the Lower Bound (LB) in the equation for each action-KPI value pair with the new data. This step may be done using any one of the known statistical methods. The LB may be smoothed by using a moving average to avoid large fluctuations that could affect the decision of the assessing module 104. For a given action-KPI value pair, the lower bound may be updated as follows:

[0280] LB(a, KPI, t+1) = p-LB(a, KPI, t-1) + (l-p)-New LB(a, KPI, t) Here LB(a, KP I, t+1) may be the LB that will be used for the next decision time if an action a is proposed for the KPI value. New LB(a, KP I, t) may be the lower bound after the action is applied and the bound is updated with a new measurement. LB(a, KPI, t-1) may be the LB before the action is applied. After updating 218 the LB with the new data and performing the smoothing, the risk factor may be recalculated using

[0281] Finally, an updated overall risk and utility score for each action-KPI value pair may be determined in the step 216 of the method 200.

[0282] An example of a corresponding sequence diagram for an IBA closed loop embodying the method 200 is shown in Fig. 10.

[0283] Fig. 11 shows a schematic block diagram for an embodiment of the device 100. The device 100 comprises processing circuitry, e.g., one or more processors 1104 for performing the method 200 and memory 1106 coupled to the processors 1104. For example, the memory 1106 may be encoded with instructions that implement at least one of the modules 102, 104 and 108.

[0284] The one or more processors 1104 may be a combination of one or more of a microprocessor, controller, microcontroller, central processing unit, digital signal processor, application specific integrated circuit, field programmable gate array, or any other suitable computing device, resource, or combination of hardware, microcode and / or encoded logic operable to provide, either alone or in conjunction with other components of the device 100, such as the memory 1106, functionality of a node of the wireless telecommunications infrastructure. For example, the one or more processors 1104 may execute instructions stored in the memory 1106. Such functionality may include providing various features and steps discussed herein, including any of the benefits disclosed herein. The expression "the device being operative to perform an action" may denote the device 100 being configured to perform the action.

[0285] As schematically illustrated in Fig. 11, the device 100 may be embodied by a network node 1100, e.g., functioning as a base station or a gateway or relay UE. The network node 1100 comprises a radio interface 1102 coupled to the device 100 for radio communication with one or more radio devices, e.g., functioning as a UE.

[0286] Fig. 12 shows a schematic block diagram for a further embodiment of the device 100. The device 100 comprises processing circuitry, e.g., one or more processors 1204 for performing the method 200 and memory 1206 coupled to the processors 1204. For example, the memory 1206 may be encoded with instructions that implement at least one of the modules 102, 104 and 108.

[0287] The one or more processors 1204 may be a combination of one or more of a microprocessor, controller, microcontroller, central processing unit, digital signal processor, application specific integrated circuit, field programmable gate array, or any other suitable computing device, resource, or combination of hardware, microcode and / or encoded logic operable to provide, either alone or in conjunction with other components of the device 100, such as the memory 1206, core node functionality. For example, the one or more processors 1204 may execute instructions stored in the memory 1206. Such functionality may include providing various features and steps discussed herein, including any of the benefits disclosed herein. The expression "the device being operative to perform an action" may denote the device 200 being configured to perform the action.

[0288] As schematically illustrated in Fig. 12, the device 100 may be embodied by a core network 1200, e.g., functioning as a Policy Control Function (PCF) of a core network (e.g., a 5GC). The PCF may handle policy enforcement and / or quality of service (QoS) management according to the at least one intent.

[0289] The core network 1200 comprises an interface 1202, e.g. a reference point NG between the access and the core networks. This reference point may be constituted of several interfaces such as N2. The interface 1202 may be coupled to the device 100 for communication with one or more network node of the RAN, e.g., functioning as gNBs.

[0290] With reference to Fig. 13, in accordance with an embodiment, a communication system 1300 includes a telecommunication network 1310, such as a 3GPP-type cellular network, which comprises an access network 1311, such as a radio access network, and a core network 1314. The access network 1311 comprises a plurality of base stations 1312a, 1312b, 1312c, such as NBs, eNBs, gNBs or other types of wireless access points, each defining a corresponding coverage area 1313a, 1313b, 1313c. Each base station 1312a, 1312b, 1312c is connectable to the core network 1314 over a wired or wireless connection 1315. A first user equipment (UE) 1391 located in coverage area 1313c is configured to wirelessly connect to, or be paged by, the corresponding base station 1312c. A second UE 1392 in coverage area 1313a is wirelessly connectable to the corresponding base station 1312a. While a plurality of UEs 1391, 1392 are illustrated in this example, the disclosed embodiments are equally applicable to a situation where a sole UE is in the coverage area or where a sole UE is connecting to the corresponding base station 1312.

[0291] Any of the base stations 1312 and the host computer 1330, and the core network 1314 may embody the device 100.

[0292] The telecommunication network 1310 is itself connected to a host computer 1330, which may be embodied in the hardware and / or software of a standalone server, a cloud-implemented server, a distributed server or as processing resources in a server farm. The host computer 1330 may be under the ownership or control of a service provider, or may be operated by the service provider or on behalf of the service provider. The connections 1321, 1322 between the telecommunication network 1310 and the host computer 1330 may extend directly from the core network 1314 to the host computer 1330 or may go via an optional intermediate network 1320. The intermediate network 1320 may be one of, or a combination of more than one of, a public, private or hosted network; the intermediate network 1320, if any, may be a backbone network or the Internet; in particular, the intermediate network 1320 may comprise two or more sub-networks (not shown).

[0293] The communication system 1300 of Fig. 13 as a whole enables connectivity between one of the connected UEs 1391, 1392 and the host computer 1330. The connectivity may be described as an over-the-top (OTT) connection 1350. The host computer 1330 and the connected UEs 1391, 1392 are configured to communicate data and / or signaling via the OTT connection 1350, using the access network 1311, the core network 1314, any intermediate network 1320 and possible further infrastructure (not shown) as intermediaries. The OTT connection 1350 may be transparent in the sense that the participating communication devices through which the OTT connection 1350 passes are unaware of routing of uplink and downlink communications. For example, a base station 1312 need not be informed about the past routing of an incoming downlink communication with data originating from a host computer 1330 to be forwarded (e.g., handed over) to a connected UE 1391. Similarly, the base station 1312 need not be aware of the future routing of an outgoing uplink communication originating from the UE 1391 towards the host computer 1330.

[0294] By virtue of the method 200 being performed by any one of the base stations 1312 and / or the core node 1314 and / or the host computer 1330, the performance or range of the OTT connection 1350 can be improved, e.g., in terms of increased throughput and / or reduced latency. More specifically, the host computer 1330 may embody the device 100 to perform the method 200 (e.g., in a cloud implementation).

[0295] Example implementations, in accordance with an embodiment of the UE, base station and host computer discussed in the preceding paragraphs, will now be described with reference to Fig. 14. In a communication system 1400, a host computer 1410 comprises hardware 1415 including a communication interface 1416 configured to set up and maintain a wired or wireless connection with an interface of a different communication device of the communication system 1400. The host computer 1410 further comprises processing circuitry 1418, which may have storage and / or processing capabilities. In particular, the processing circuitry 1418 may comprise one or more programmable processors, application-specific integrated circuits, field programmable gate arrays or combinations of these (not shown) adapted to execute instructions. The host computer 1410 further comprises software 1411, which is stored in or accessible by the host computer 1410 and executable by the processing circuitry 1418. The software 1411 includes a host application 1412. The host application 1412 may be operable to provide a service to a remote user, such as a UE 1430 connecting via an OTT connection 1450 terminating at the UE 1430 and the host computer 1410. In providing the service to the remote user, the host application 1412 may provide user data, which is transmitted using the OTT connection 1450. The user data may depend on the location of the UE 1430. The user data may comprise auxiliary information or precision advertisements (also: ads) delivered to the UE 1430. The location may be reported by the UE 1430 to the host computer, e.g., using the OTT connection 1450, and / or by the base station 1420, e.g., using a connection 1460. The communication system 1400 further includes a base station 1420 provided in a telecommunication system and comprising hardware 1425 enabling it to communicate with the host computer 1410 and with the UE 1430. The hardware 1425 may include a communication interface 1426 for setting up and maintaining a wired or wireless connection with an interface of a different communication device of the communication system 1400, as well as a radio interface 1427 for setting up and maintaining at least a wireless connection 1470 with a UE 1430 located in a coverage area (not shown in Fig. 14) served by the base station 1420. The communication interface 1426 may be configured to facilitate a connection 1460 to the host computer 1410. The connection 1460 may be direct, or it may pass through a core network (not shown in Fig. 14) of the telecommunication system and / or through one or more intermediate networks outside the telecommunication system. In the embodiment shown, the hardware 1425 of the base station 1420 further includes processing circuitry 1428, which may comprise one or more programmable processors, application-specific integrated circuits, field programmable gate arrays or combinations of these (not shown) adapted to execute instructions. The base station 1420 further has software 1421 stored internally or accessible via an external connection.

[0296] The communication system 1400 further includes the UE 1430 already referred to. Its hardware 1435 may include a radio interface 1437 configured to set up and maintain a wireless connection 1470 with a base station serving a coverage area in which the UE 1430 is currently located. The hardware 1435 of the UE 1430 further includes processing circuitry 1438, which may comprise one or more programmable processors, application-specific integrated circuits, field programmable gate arrays or combinations of these (not shown) adapted to execute instructions. The UE 1430 further comprises software 1431, which is stored in or accessible by the UE 1430 and executable by the processing circuitry 1438. The software 1431 includes a client application 1432. The client application 1432 may be operable to provide a service to a human or non-human user via the UE 1430, with the support of the host computer 1410. In the host computer 1410, an executing host application 1412 may communicate with the executing client application 1432 via the OTT connection 1450 terminating at the UE 1430 and the host computer 1410. In providing the service to the user, the client application 1432 may receive request data from the host application 1412 and provide user data in response to the request data. The OTT connection 1450 may transfer both the request data and the user data. The client application 1432 may interact with the user to generate the user data that it provides.

[0297] It is noted that the host computer 1410, base station 1420 and UE 1430 illustrated in Fig. 14 may be identical to the host computer 1330, one of the base stations 1312a, 1312b, 1312c and one of the UEs 1391, 1392 of Fig. 13, respectively. This is to say, the inner workings of these entities may be as shown in Fig. 14, and, independently, the surrounding network topology may be that of Fig. 13.

[0298] In Fig. 14, the OTT connection 1450 has been drawn abstractly to illustrate the communication between the host computer 1410 and the UE 1430 via the base station 1420, without explicit reference to any intermediary devices and the precise routing of messages via these devices. Network infrastructure may determine the routing, which it may be configured to hide from the UE 1430 or from the service provider operating the host computer 1410, or both. While the OTT connection 1450 is active, the network infrastructure may further take decisions by which it dynamically changes the routing (e.g., on the basis of load balancing consideration or reconfiguration of the network).

[0299] The wireless connection 1470 between the UE 1430 and the base station 1420 is in accordance with the teachings of the embodiments described throughout this disclosure. One or more of the various embodiments improve the performance of OTT services provided to the UE 1430 using the OTT connection 1450, in which the wireless connection 1470 forms the last segment. More precisely, the teachings of these embodiments may reduce the latency and improve the data rate and thereby provide benefits such as better responsiveness and improved QoS.

[0300] A measurement procedure may be provided for the purpose of monitoring data rate, latency, QoS and other factors on which the one or more embodiments improve. There may further be an optional network functionality for reconfiguring the OTT connection 1450 between the host computer 1410 and UE 1430, in response to variations in the measurement results. The measurement procedure and / or the network functionality for reconfiguring the OTT connection 1450 may be implemented in the software 1411 of the host computer 1410 or in the software 1431 of the UE 1430, or both. In embodiments, sensors (not shown) may be deployed in or in association with communication devices through which the OTT connection 1450 passes; the sensors may participate in the measurement procedure by supplying values of the monitored quantities exemplified above, or supplying values of other physical quantities from which software 1411, 1431 may compute or estimate the monitored quantities. The reconfiguring of the OTT connection 1450 may include message format, retransmission settings, preferred routing etc.; the reconfiguring need not affect the base station 1420, and it may be unknown or imperceptible to the base station 1420. Such procedures and functionalities may be known and practiced in the art. In certain embodiments, measurements may involve proprietary UE signaling facilitating the host computer's 1410 measurements of throughput, propagation times, latency and the like. The measurements may be implemented in that the software 1411, 1431 causes messages to be transmitted, in particular empty or "dummy" messages, using the OTT connection 1450 while it monitors propagation times, errors etc.

[0301] Fig. 15 is a flowchart illustrating a method implemented in a communication system, in accordance with one embodiment. The communication system includes a host computer, a base station and a UE which may be those described with reference to Figs. 13 and 14. For simplicity of the present disclosure, only drawing references to Fig. 15 will be included in this paragraph. In a first step 1510 of the method, the host computer provides user data. In an optional substep 1511 of the first step 1510, the host computer provides the user data by executing a host application. In a second step 1520, the host computer initiates a transmission carrying the user data to the UE. In an optional third step 1530, the base station transmits to the UE the user data which was carried in the transmission that the host computer initiated, in accordance with the teachings of the embodiments described throughout this disclosure. In an optional fourth step 1540, the UE executes a client application associated with the host application executed by the host computer.

[0302] Fig. 16 is a flowchart illustrating a method implemented in a communication system, in accordance with one embodiment. The communication system includes a host computer, a base station and a UE which may be those described with reference to Figs. 13 and 14. For simplicity of the present disclosure, only drawing references to Fig. 16 will be included in this paragraph. In a first step 1610 of the method, the host computer provides user data. In an optional substep (not shown) the host computer provides the user data by executing a host application. In a second step 1620, the host computer initiates a transmission carrying the user data to the UE. The transmission may pass via the base station, in accordance with the teachings of the embodiments described throughout this disclosure. In an optional third step 1630, the UE receives the user data carried in the transmission.

[0303] As has become apparent from above description, at least some embodiments of the technique take the risk of a utility loss due to uncertainty in the predicted KPI values into account. Same or further embodiments allow for an improved IBA framework by evaluating the prediction and making more reliable decisions based on the improved and evaluated prediction. Moreover, embodiments can give more insight and information to an operator of the wireless telecommunications infrastructure.

[0304] Many advantages of the present invention will be fully understood from the foregoing description, and it will be apparent that various changes may be made in the form, construction and arrangement of the units and devices without departing from the scope of the invention and / or without sacrificing all of its advantages. Since the invention can be varied in many ways, it will be recognized that the invention should be limited only by the scope of the following claims.

Claims

Claims1. A method (200) of autonomously operating a wireless telecommunications infrastructure according to at least one intent, the method (200) comprising or initiating: for an action executable in the autonomous operation of the wireless telecommunications infrastructure, predicting (202) a value of a key performance indicator, KPI, of the wireless telecommunications infrastructure and a value of an uncertainty of the predicted (202) KPI value; assessing (204) an uncertainty of a utility based on the predicted (202) KPI value and the predicted (202) uncertainty value of the KPI value; and selectively executing (208) the action, wherein the selectivity is dependent on the intent and the assessed (204) uncertainty of the utility.

2. The method (200) of claim 1, wherein the predicted (202) uncertainty value of the predicted KPI value comprises at least one of a lower bound for the KPI value, an upper bound for the KPI value, a confidence interval of the predicted KPI value, and a prediction interval of the predicted KPI value.

3. The method (200) of claim 2, wherein the action is executed (208) if the lower bound or the confidence interval or the prediction interval corresponds to an increase of the utility and / or wherein the action is not executed (208) if the lower bound or the confidence interval or the prediction interval corresponds to a decrease of the utility; and / or wherein the utility is a function of the KPI value with a non-linear characteristic, and wherein the action is executed (208) if a KPI value of the nonlinear characteristic is below the lower bound or the confidence interval or the prediction interval and / or wherein the action is not executed (208) if the KPI value of the non-linear characteristic is included in or above the lower bound or the confidence interval or the prediction interval.

4. The method (200) of any one of claims 1 to 3, wherein the assessing (204) of the uncertainty comprises determining whether a probability for the action increasing the utility is greater than a probability for the action decreasing the utility and / or whether a predicted increase of the utility greater than a predicted decrease of the utility.

5. The method (200) of any one of claims 1 to 4, wherein the assessing (204) of the uncertainty comprises determining an expectation value or lower bound of the utility based on the predicted (202) KPI value and the predicted (202) uncertainty value of the KPI value, and wherein the action is executed (208) if the expectation value or lower bound of the utility corresponds to an increase of the utility and / or wherein the action is not executed (208) if the expectation value or lower bound of the utility corresponds to a decrease of the utility.

6. The method (200) of any one of claims 1 to 5, wherein the action is executed (208) if a criterion of the intent is fulfilled for the assessed (204) uncertainty of the utility, optionally wherein a further action is executed (208) if a further criterion is fulfilled for the assessed (204) uncertainty of the utility or for a further assessed (204) uncertainty of a further utility, and wherein the action is not executed (208) if the further criterion for executing the further action is fulfilled.

7. The method (200) of any one of claims 1 to 6, wherein the intent comprises multiple utilities each depending on at least one KPI or the wireless telecommunications infrastructure is autonomously operated according to multiple intents each comprising at least one utility each depending on at least one KPI, and wherein the action is selectively executed (208) and the selectivity is dependent on a combination of the uncertainty assessed (204) for each of the utilities, and / or wherein the action is not executed (208) if at least one criterion for the uncertainty assessed (204) for each of the utilities is not fulfilled, and / or wherein the action is executed (208) if each criterion for the uncertainty assessed (204) for each of the utilities is fulfilled.

8. The method (200) of any one of claims 1 to 7, wherein the predicted (202) uncertainty value of the predicted KPI value comprises a probability of the predicted KPI value or a probability for the KPI value being less than the predicted KPI value or a probability for the KPI value falling below a predefined threshold value, and / orthe assessed (204) uncertainty of the utility comprises a probability for a reduction of the utility or a probability for the utility falling below a predefined threshold value.

9. The method (200) of any one of claims 1 to 8, wherein the autonomous operation of the wireless telecommunications infrastructure comprises at least one closed loop, optionally wherein each closed loop comprises at least one intent.

10. The method (200) of any one of claims 1 to 9, wherein the KPI comprises at least one of latency; bit rate; block error rate, BLER; packet loss rate; signal to noise ratio, SNR; and signal to interference and noise ratio, SINR.

11. The method (200) of any one of claims 1 to 10, wherein the prediction (202) of the KPI value and / or the uncertainty value of the KPI value is performed by an artificial intelligent agent, Al agent.

12. The method (200) of any one of claims 1 to 11, wherein the utility is a function of the KPI value.

13. The method (200) of any one of claims 1 to 12, wherein the method (200) is performed according to more than one intent.

14. The method (200) of any one of claims 1 to 13, wherein the method (200) is performed for more than one given action for each of the at least one intent.

15. The method (200) of any one of claims 1 to 14, further comprising or initiating: obtaining (210) information as to a state of the wireless telecommunications infrastructure after executing (208) the action.

16. The method (200) of claim 15, wherein the obtained (210) information comprises at least one of: a measured KPI value; a KPI lower bound, KPI_I; a KPI upper bound, KPI_u; and a confidence interval, Cl.

17. The method (200) of claim 15 or 16, further comprising or initiating: comparing (212) the obtained (210) information after executing (208) the action with the predicted (202) KPI value and / or the predicted (202) uncertainty value of the predicted KPI value, optionally wherein the comparison (212) comprises training or re-training the Al agent to predict (202) the KPI value and / or the uncertainty value of the predicted KPI value; and / or optionally wherein the comparison (212) comprises updating a stored value for the KPI value and / or the uncertainty value of the predicted KPI value for the respective action.

18. The method (200) of any one of claims 1 to 17, further comprising or initiating: determining (214) a risk factor, Y, for at least one or each pair of the action and the KPI based on the comparison (212).

19. The method (200) of claim 18, further comprising or initiating: determining (216) an overall risk factor based on a maximum value of all the risk factors determined (214) according to one of the at least one intent.

20. The method (200) of any one of claims 1 to 19, wherein the assessing (204) comprises determining (206) at least one of: an increase of the utility for the action; and a decrease of the utility for the action.

21. The method (200) of any one of claims 1 to 20, wherein the selectively executing (208) of the action is based on a weighted combination of the determined (206) increase of the utility and the determined (206) decrease of the utility.

22. The method (200) of any one of claims 1 to 21, wherein the selectively executed (208) action is based on minimizing an absolute value of a difference between the predicted KPI value and the at least one intent, optionally between the predicted KPI value and a threshold value according to the at least one intent.

23. The method (200) of any one of claims 1 to 22, further comprising or initiating: updating (218), after executing the action, a prediction of at least one of: the KPI value for the action; the uncertainty value of the predicted (202) KPI value for the action; and the risk factor related to selectively executing (208) the action.

24. The method (200) of any one of claims 1 to 23, further comprising or initiating: time-averaging (220) at least one of: the predicted (202) and / or updated (218) KPI value for the respective action; the predicted (202) and / or updated (218) uncertainty value of the KPI value for the respective action; the obtained (210) information after executing (208) the respective action; and the current KPI value.

25. A computer program product (100) comprising program code portions for performing the steps of any one of the claims 1 to 24 when the computer program product is executed on one or more computing devices (1104; 1204), optionally stored on a computer-readable recording medium (1106; 1206).

26. A node (1100; 1200) of a wireless telecommunications infrastructure, the node (1100; 1200) comprising memory operable to store instructions and processing circuitry operable to execute the instructions, such that the node (1100; 1200) is operable to: predict (202), for an action executable in the autonomous operation of the wireless telecommunications infrastructure, a value of a key performance indicator, KPI, of the wireless telecommunications infrastructure and a value of an uncertainty of the predicted (202) KPI value; assess (204) an uncertainty of a utility based on the predicted (202) KPI value and the predicted (202) uncertainty value of the KPI value; and selectively execute (208) the action, wherein the selectivity is dependent on the intent and the assessed (204) uncertainty of the utility.

27. The node (1100; 1200) of claim 26, further operable to perform any one of the steps of any one of claims 2 to 24.

28. A node (1100; 1200) supporting a radio access network, RAN, in a wireless telecommunications infrastructure, the node (1100; 1200) being configured to: predict (202), for an action executable in the autonomous operation of the wireless telecommunications infrastructure, a value of a key performance indicator, KPI, of the wireless telecommunications infrastructure and a value of an uncertainty of the predicted (202) KPI value; assess (204) an uncertainty of a utility based on the predicted (202) KPI value and the predicted (202) uncertainty value of the KPI value; and selectively execute (208) the action, wherein the selectivity is dependent on the intent and the assessed (204) uncertainty of the utility.

29. The node (1100; 1200) of claim 28, further configured to perform the steps of any one of claim 1 to 24.

30. The node (1100; 1200) of any one of claims 26 to 29, wherein the node is a core node of a core network, CN, of the wireless telecommunications infrastructure, or wherein the node is a network node of a radio access network, RAN, of the wireless telecommunications infrastructure, optionally a base station or a central unit of the base station.

31. A communication system (1300; 1400) including a host computer (1330; 1410) comprising: processing circuitry (1418) configured to provide user data; and a communication interface (1416) configured to forward user data to a cellular or ad hoc radio network (1310) of a wireless telecommunications infrastructure for transmission to a user equipment, UE (1391; 1392; 1430), wherein the wireless telecommunications infrastructure comprises a radio interface (1102; 1437) and processing circuitry (1104; 1438), the processing circuitry (1104; 1438) of the wireless telecommunications infrastructure being configured to execute the steps of any one of claims 1 to 24.

32. The communication system (1300; 1400) of claim 31, further including a UE (1391; 1392; 1430), wherein the UE (1391; 1392; 1430) comprises a radio interface (1102; 1437) and processing circuitry (1104; 1438), the processing circuitry (1104; 1438) of the UE (1391; 1392; 1430) being configured for radio access to the wireless telecommunications infrastructure.

33. The communication system (1300; 1400) of claim 31 or 32, wherein the wireless telecommunications infrastructure comprises one or more nodes (1100; 1200) configured to execute the steps of any one of claims 1 to 24.

34. The communication system (1300; 1400) of any one of claims 31 to 33, wherein the radio network (1310) further comprises a base station (300; 1200; 1312; 1420), or a radio device (1391; 1392; 1430) functioning as a gateway, which is configured to communicate with the UE (1391; 1392; 1430) and / or configured to execute the steps of any one of claims 1 to 24.

35. The communication system (1300; 1400) of any one of claims 31 to 34, wherein: the processing circuitry (1418) of the host computer (1330; 1410) is configured to execute a host application (1412), thereby providing the user data; and a processing circuitry (1104; 1438) of the UE (1391; 1392; 1430) is configured to execute a client application (1432) associated with the host application (1412).

Images