Method for updating parameter value of at least one parameter related to quality of experience (QOE), device, system, and corresponding computer program

Abstract

To provide a method for updating a parameter value of at least one parameter associated with Quality of Experience (QoE) of an application in a wireless communications system for improving the QoE by adjusting relevant QoS parameters based on measured QoE and identified system features.SOLUTION: A method implemented in a network controller device includes: a step (10) of determining at least one feature related to a wireless communication system, the feature being responsible for a deviation in QoE measurements for an application; a step (20) of obtaining at least one Quality of Service (QoS) parameter linked to the deviation from the identified feature; and a step (30) of updating a parameter value of a QoS parameter linked to the deviation.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to managing applications deployed in communication networks. More specifically, the present invention relates to determining the resources required to ensure the Quality of Experience (QoE) of an application within a communication network. The present invention finds application, for example, in implementing applications within non-public communication networks. [Background technology]

[0002] Controlling the variation of Quality of Experience (QoE) of deployed applications in wireless communication networks (more specifically Standalone Non-Public Networks (SNPNs)) is one of the most important aspects of optimizing application deployment in industrial networks. This control of QoE is - Reduced QoE variations relative to the nominal performance regime, Outliers, i.e., significant degradation of QoE relative to the reference performance regime, are detected and a warning is sent to the network controller regarding the parameters responsible for the QoE degradation. The present invention comprises adjusting parameters of a wireless communication network so as to

[0003] Model predictive control (MPC) is one possible approach to ensure this QoE optimization and control. MPC operates within a dynamic environment and utilizes predictive models to forecast future states. This ability to consider the future is believed to enable proactive measures to address potential issues before they have a significant impact on QoE. MPC's strength lies in its ability to handle multiple variables simultaneously, ensuring a comprehensive approach to QoE management.

[0004] Indeed, one of the advantages of MPC is its adaptability to changing conditions. Network dynamics, influenced by factors such as varying traffic loads (due to application flows and transaction loads), require real-time adjustments. In this regard, MPC performs well, providing a closed-loop control system that continuously receives feedback on the network's performance, allowing it to adapt and improve the QoE over time. Typically, MPC aims to stabilize the QoE achieved by a wireless network deployment for an application around a reference nominal QoE performance provided by a model. This stabilization is achieved by applying controls to several parameters that determine the behavior of the deviation of the QoE obtained from the deployment relative to the model. To achieve this result, the first step is to find a set of parameters (features) that, when applied to the wireless network, reduce the QoE fluctuations relative to the reference QoE. The optimization metric can be a cost function of the deviation between the reference QoE and the measured QoE over a decreasing time window of length T, i.e., the time interval [t, t + T - 1]. A typical cost function of the deviation is the squared error of the difference between the measured QoE and the reference QoE, and the control parameters are determined as shown below:

number

[0005] Another important aspect to consider when designing QoE control via MPC is reducing the amplitude (energy) of the controls (parameters) to ensure stability of the MPC and provide warning about possible abnormal QoE behavior of the system.

[0006] Both problems, i.e., reducing the complexity of state-of-the-art MPC and improving its stability, are related to the specific choice of parameters that must be made in the optimization logic. To solve these problems, adaptations of conventional MPC are required. Summary of the Invention [Problem to be solved by the invention]

[0007] The present disclosure aims to improve this situation. Therefore, a method and system for reducing the complexity of MPC for QoE control issues and alert propagation is disclosed. More specifically, a method is proposed for updating parameter values ​​of at least one parameter related to the quality of experience of the execution of at least one application, the at least one application comprising an application server exchanging data with application clients via a wireless communication system, the method being implemented in a network controller device of the wireless communication system. [Means for solving the problem]

[0008] According to the present disclosure, the method comprises the following steps: - determining from QoE measurements of at least one application at least one characteristic related to the wireless communication system, the characteristic contributing to a deviation between the QoE measurements of the at least one application and an application QoE model associated with the at least one application, the application QoE model being configured to predict variations in the QoE of the application; - obtaining from the at least one feature at least one QoS parameter associated with the deviation; - updating a parameter value of at least one QoS parameter associated with the deviation; Contains at least one repeat of

[0009] This method allows for continuous adjustment of QoS parameters that contribute to the QoE measured by an application, while substantially reducing the necessary resources used to determine which parameters are most responsible for QoE deviation, and thus allows for rapid response to required needs for QoS parameter adjustment.

[0010] According to one feature, the QoE measurements are obtained from the application client and / or from quality of experience QoE monitoring at the application server.

[0011] Therefore, it is possible to collect information from various sources, which helps to respond quickly to changes in QoE.

[0012] According to one feature, the step of determining at least one feature responsible for the deviation between the QoE measurement and the QoE obtained from an application QoE model associated with at least one application comprises: - obtaining QoE measurements for at least one application and parameter identifier generating a QoE metric associated with a wireless communication system; - calculating deviations of QoE measurements from a QoE predicted from an application QoE model, the model being learned during nominal application operation in the wireless communication system and / or the model being obtained from an application management layer; - analyzing the deviation of QoE and determining a nominal deviation area and an outlier QoE deviation area; - determining at least one feature included in the nominal deviation region and the outlier QoE deviation region; Includes.

[0013] This makes it possible to consider only those parameters that have a severely adverse effect on the deviation.

[0014] According to one feature, obtaining QoE measurements for the at least one application includes measuring the satisfaction of at least one application client with a particular configuration of the wireless communication system.

[0015] The satisfaction level of an application client may be provided by several satisfaction labels provided by the application client on demand or continuously during its execution.

[0016] According to one feature, a particular configuration of a wireless communication system includes at least one set of QoS parameters associated with at least one communication session opened for at least one application.

[0017] According to one feature, obtaining QoE measurements for the at least one application includes measuring satisfaction of at least one application client obtained from a digital twin of a wireless communication system in which execution of the at least one application is simulated.

[0018] According to one feature, the step of obtaining at least one QoS parameter associated with the deviation comprises: - calculating a first contribution of each of the at least one feature to the deviation of the QoE in a nominal deviation region and in an outlier QoE deviation region; - determining a QoS parameter linked to the determined at least one characteristic of the nominal and outlier deviations of the QoE from the first contribution of each of the at least one characteristic to the deviation of the QoE; - determining a second contribution of each of the at least one feature in the application QoE model prediction and obtaining a QoS parameter associated with the determined at least one feature that takes priority over the application QoE model prediction; Includes.

[0019] According to one feature, the step of updating the parameter value of at least one QoS parameter associated with the deviation comprises: - determining parameter values ​​for updating at least a portion of at least one QoS parameter; applying the updated parameter values ​​to the determined QoS parameters for updating; Includes.

[0020] According to one feature, the method further comprises updating the application QoE model in response to at least some of the at least one feature.

[0021] According to one feature, the application QoE model relates at least one satisfaction measure of at least one application to at least one QoS parameter and at least one configuration parameter of the wireless communication system.

[0022] According to one feature, the wireless communication system comprises a 5G standalone non-public network.

[0023] According to another aspect, the present disclosure relates to a device for updating a parameter value of at least one parameter related to a quality of experience of an execution of at least one application, the at least one application comprising an application server that exchanges data with an application client via a wireless communication system, the device being part of a network controller device of the wireless communication network. For example, the device may repeatedly: - determining from the QoE measurements of the at least one application at least one characteristic related to the wireless communication system, the characteristic contributing to a deviation between the QoE measurements of the at least one application and an application QoE model associated with the at least one application, the application QoE model being configured to predict a variation in the QoE of the application; - obtaining at least one QoS parameter from the at least one feature, the QoS parameter being associated with the deviation; updating the parameter value of at least one QoS parameter that is associated with the deviation; At least one module is provided for this purpose.

[0024] According to another general aspect of at least one embodiment, a non-transitory computer-readable medium is provided that includes data content generated according to any of the described methods for updating parameter values ​​or variants.

[0025] According to another general aspect of at least one embodiment, there is provided a signal including data representing at least one tensor of at least one layer or sublayer of at least one deep neural network, generated according to any of the described methods for updating parameter values ​​or variations.

[0026] According to another general aspect of at least one embodiment, a computer program product is provided that includes instructions that, when executed by a computer, cause the computer to perform any of the described methods for updating a parameter value or variant.

[0027] According to another general aspect of at least one embodiment, there is provided a computer-readable non-transitory program storage device tangibly embodying a program of instructions executable by a computer to perform at least one of the methods of the present disclosure in any of its embodiments.

[0028] According to another general aspect of at least one embodiment, there is provided a computer-readable storage medium comprising instructions that, when executed by a computer, cause the computer to perform at least one of the disclosed methods in any of the embodiments.

[0029] Although not explicitly described, the devices of the present disclosure may be adapted to carry out the methods of the present disclosure in any of their embodiments.

[0030] Although not explicitly described, the present embodiments relating to methods or corresponding signals, devices, and computer-readable storage media may be used in any combination or subcombination. [Brief explanation of the drawings]

[0031] The present disclosure will be better understood on reading the following description, given by way of in no way limiting example, and made with reference to the figures in which:

[0032] [Figure 1] 1 illustrates steps of a method according to an exemplary embodiment. [Figure 2] 1 illustrates a system in which the present disclosure is intended to be practiced; [Figure 3] 1 illustrates a system in which the present disclosure is implemented; [Figure 4] FIG. 1 illustrates an example of a device designed to update parameters according to the method of the present disclosure.

[0033] In these figures, the same reference numbers from one figure to another indicate the same or similar elements. For clarity, the elements shown are not drawn to scale unless otherwise noted. Also, unless required by the language of the description, the order of steps depicted in these figures is provided for illustrative purposes only and is not meant to limit the disclosure as may be applied to the same steps performed in different orders. DETAILED DESCRIPTION OF THE INVENTION

[0034] The problem addressed by this disclosure is reducing the complexity and improving the stability of QoE control loops through model predictive control (MPC) strategies.

[0035] As previously disclosed, the present disclosure proposes reducing the dimensionality of the MPC by performing a root cause analysis to identify a reduced set of parameters (i.e., a reduced set of QoS parameters) that most strongly influence the measured QoE deviation from the QoE predicted by the model for a given time period. Additionally, if it is determined that the control parameters (those that cause QoE deviations) also have a strong impact on the application model performance (i.e., the fit between the application model and the measured QoE), the application QoE model is finally updated.

[0036] In other words, complexity reduction is achieved by reducing the number of parameters that affect QoE deviation from the reference QoE obtained from the application model. Improvement of MPC stability is achieved by determining a separate list of parameters that are adversely affecting the abnormal behavior of QoE deviation and implementing corrective measures that send alerts to relevant entities. Determining a reduced set of relevant control parameters for both large (outlier) and small deviations of QoE is achieved in one embodiment by feature extraction techniques (e.g., AI) and feature importance assessment of deviation labels for prediction.

[0037] 1 , the present disclosure relates to a method for updating parameter values ​​of one or more parameters related to a Quality of Experience (QoE) of an execution of an application (App), the application (App) comprising an application server (AS) exchanging data with an application client (AC) via a wireless communication system (SNPN), the method being implemented in a network controller device (NCD) of the wireless communication system (SNPN). - a step 10 of determining, from QoE measurements of at least one application App, at least one feature Fea related to the wireless communication system, the feature being responsible for the deviation between the QoE measurements of the at least one application App and an application QoE model associated with the at least one application App, the application QoE model being configured to predict the QoE variations of the application App; a step 20 of obtaining, from at least one feature Fea, at least one QoS parameter associated with the deviation; a step 30 of updating a parameter value of at least one QoS parameter associated with the deviation; This includes repetition of

[0038] More specifically, the step 10 of determining the features Fea responsible for the deviation between the QoE measurements and the application QoE model associated with at least one application App comprises: a step 101 of obtaining QoE measurements for at least one application App label and parameter identifier generating QoE metric characteristics related to a wireless communication system SNPN; - a step 102 of calculating the deviation of the QoE measurement from the QoE predicted from an application QoE model, the model being learned during nominal application operation in the wireless communication system and / or the model being obtained from an application App management layer; a step 103 of analyzing the deviations in QoE and determining nominal deviation regions and outlier QoE deviation regions; a step 104 of determining at least one feature Fea included in the nominal deviation region and the outlier QoE deviation region; Includes.

[0039] It should be noted that the application QoE model is used to predict nominal behavior in terms of QoE. One possible model is the Mean Opinion Score (MoS) (MoS = 5 (good); MoS = 1 (bad)), a mean quality MoS value. For example, the model provides a distribution of MoS depending on the QoS parameters of the network. This kind of model can be trained periodically, during the application's optimal operation period, or provided as input parameters by the application. Other application QoE models are of course possible, for example, providing specific labels depending on a subset of parameters, e.g., opinion scores on data rate reception, opinion scores on data decoding speed, etc., and these scores are aggregated as a function of an aggregation function (or discrete function) that forms the formula of the application QoE model.

[0040] Therefore, this disclosure proposes reducing the dimensionality of the MPC by performing a root cause analysis to find a reduced set of parameters that most strongly influence the measured QoE deviation from the QoE predicted by the application QoE model. If it is determined that all or some of the reduced set of parameters also strongly influence the model performance itself, the application QoE model may also be updated.

[0041] Depending on the implementation, the application QoE model associated with the application(s) (App) includes one or more pre-trained artificial intelligence models. More specifically, according to the present disclosure, predicted QoE can be achieved using an artificial neural network (ANN) model trained to link quality of service (QoS) parameters (and network configurations) to predicted QoE values. The application QoE model is designed to process at least one network key performance indicator (KPI) as input and generate predicted QoE values ​​as output. To train a neural network for QoE prediction in a communication system, the following method is adopted, which includes: A data collection phase in which relevant network KPIs associated with QoS and QoE labels from applications are obtained, which may include parameters such as channel quality indicators, user throughput, data volume, modulation order, and coding, which are essential for assessing the network's ability to meet QoS requirements and affect QoE; a feature selection phase, which utilizes statistical methods such as ANOVA F-tests to select the most relevant features for training the application QoE model, making it possible to enforce a first restriction of the QoS features and network parameters initially selected in the model with respect to the generated QoE labels; a model training phase for developing an application QoE model having multiple layers for processing the selected features and training the application QoE model using the collected data, where the model is designed to map input KPIs to corresponding QoE labels; Evaluation and validation: The trained application QoE model is evaluated using appropriate validation techniques, such as cross-validation, to ensure its robustness and generalizability. Includes.

[0042] More specifically, according to one embodiment, the application QoE model is acquired (trained and tuned) offline using a digital twin of the SNPN. The parameters of this digital twin are modified to generate as many conditions as possible for the SNPN to capture various possibilities for network conditions while the application is functioning (i.e., while the application server is exchanging data with the application client). The digital twin of the SNPN is a computer simulation model that replicates the assets, information, and processes of an active SNPN, along with its operating environment and application traffic. It uses historical and real-time data to simulate the SNPN, providing benefits such as network visibility, training, and improved network operation. According to the present disclosure, the digital twin is used to study the behavior of applications that exchange data over a physical network under various conditions. The digital twin works by creating a virtual representation of the network and enabling analysis, diagnosis, emulation, and control of the physical network based on the data. In the context of the present disclosure, the digital twin is interfaced with a software application running in real time to collect application satisfaction data across various operating conditions.

[0043] In particular, we have shown that the network parameters that can be used to evaluate the QoE of an SNPN are the following: RAN parameters: network-centric Radio Access Network (RAN) parameters, including aspects such as signal strength, interference, and handover performance; IP transport parameters: Parameters related to IP transport, including packet loss, delay, and jitter, that evaluate the performance of IP-based communications in a network; -Context-Aware QoS / QoE: Context-aware QoS / QoE considerations, including understanding the impact of network context, such as location, device type, etc., on an application's QoE; -Throughput: the average throughput of a network, measuring the amount of data transferred per unit time; -End-to-end delay: the delay experienced by a data packet from source to destination, which has a direct negative impact on real-time applications. Packet loss ratio: the ratio of lost packets to the total number of packets sent, which evaluates the reliability of the network; Jitter: Variation in the delay of received packets, which affects real-time communications and the smoothness of some applications. Fairness index: a measure of the fairness of resource allocation among users, which is important to ensure fair QoE for all users in the network; - Channel Spectral Efficiency: The efficiency of spectrum usage in a channel, which directly impacts the capacity and performance of the network; -Throughput efficiency: The efficiency of data transfer in the network, which is essential to maximize the utilization of network resources. Signal-to-Interference and Noise Ratio (SINR): The ratio of the received signal power to the combined interference and noise power, which is an important indicator of signal quality in a network. It was decided to include at least one of the following:

[0044] In addition, we believe that the QoS parameters to be used to train the application QoE model should be the following parameters: - Maximum flow bit rate (downlink / uplink), which specifies the maximum bit rate that can be provided by the network for downlink and uplink data transmission; Allocation and Retention Priority (ARP), which is used to prioritize the allocation and retention of network resources for particular services or users, ensuring that higher priority services receive preferential treatment; -Packet Delay Budget (PDB), which defines the maximum tolerable end-to-end delay of a packet, which is important for real-time and delay-sensitive applications; - priority levels, which determine the priority level of a particular service or user, affecting resource allocation and QoS treatment within the network; Reflected QoS Attributes (RQAs) that reflect the QoS attributes associated with QoS flows and provide essential information for QoS management and control; Notification controls used to control notifications of QoS-related events and changes in the network, ensuring effective QoS management and maintenance; - Flow bit rate, which specifies the bit rate associated with individual QoS flows and allows for fine-grained control and management of data transmission within the network; - Aggregate bit rate, which defines the combined bit rate of multiple QoS flows and provides an overview of the network capacity and performance; -Maximum packet loss rate, which indicates the maximum tolerable rate of lost packets for a QoS flow and ensures reliable and robust data transmission It was also decided to include at least one of the following:

[0045] As previously revealed herein, all of these parameters are subject to some important variation, and therefore the range of possible values ​​dramatically expands the complexity of identifying the parameters responsible for the discrepancy between the expected QoE and the actual expressed (or measured) QoE. However, thanks to the use of the proposed method, and more specifically thanks to the dimensionality reduction made to the representation in question, one obtains a more accurate assessment of the parameters that actually matter while assessing or determining the root causes of QoE variation.

[0046] In an exemplary implementation of the method of the present disclosure, a modified "control network" node (device) is implemented, as will be elucidated in relation to Figures 2 and 3. This exemplary embodiment will be described with respect to the exemplary context of a control network node, as illustrated in Figure 2, which shows a general setup for dynamic QoE-aware optimization of an SNPN deployment.

[0047] On the left side of Figure 2, an application flow (AppLyrFlw) is set up between an application server (AppSrv) and an application client (AppClt) traversing a standalone non-public network (SNPN). The application layer flow (AppLyrFlw) is converted into an internal SNPN session flow (SNPNSessFlw) used to transport application layer packets in the SNPN domain, and this session is controlled by a network configuration node (NtwkConf). The network configuration node (NtwkConf) outputs QoS features (QoSF) that describe the session QoS parameters of the SNPN.

[0048] In this embodiment, the closed-loop network control relies on a model of application QoE (AppMdl) that predicts the application's quality of experience from the QoS features (QoSF) obtained from the deployment and provides a prediction of future values ​​of the QoS features (QoSFsP). Depending on the application deployment, the application QoE model may be implemented in the form of an artificial intelligence process (e.g., by using a convolutional neural network or a U-Net++ neural network), or in the form of a linear decision tree, a decision forest, or a mixture of several decision-making models that can predict both future values ​​of the predicted QoE (LblsPred). In other words, the model is, for example, a regression-like function that provides future QoE values ​​as a function of the QoS values. More specifically, the model may be a deep learning engine using multiple layers of a neural network to predict the value of the predicted QoE (LblsPred), or it may be based on classification machine learning techniques that predict the class of typical predicted QoE (LblsPred) from the measured QoS features. Additionally, the application QoE model (AppMdl) may be embedded in an electronic device that is connected to or part of the standalone non-public system (SNPN), for example as a probe, and that collects data from various parts of the SNPN's network. The application QoE model (AppMdl) may be stored in a local database that is locally available at the edge cloud system of the standalone non-public system (SNPN).

[0049] The measured (CurLbl) and predicted QoE (LblsPred) values ​​and predicted QoS features (QoSFsP) are input to a network control node (NtwKCtrl), which determines from the input parameters the actions (UpdParVal) to be applied to the network to adapt internal SNPN parameters so that application QoE is optimized. The network configuration node (NtwkConf) determines the parameters to be applied to the SNPN to set up an SNPN session (SNPNSessFlow) so that the quality of experience is optimized for the application deployment. Thus, the principle of model predictive control of the SNPN is to optimize the variation of the client's measured quality of experience against the application QoE model (AppMdl) by applying adjustments to the SNPN's radio and session parameters.

[0050] In other words, MPC attempts to predict key variables involved in the quality of experience dynamics of SNPNs over a moving look-ahead horizon and solves an accurate optimization problem based on the predictions obtained from the model.Multiple parameters can be optimized for MPC, such as the time horizon for optimization, or multiple metrics that can be included to process measured QoE, predicted QoE, and QoS features.

[0051] According to the present disclosure, a modified "control network" device is introduced to reduce the complexity (dimensionality) of the problem to be solved, as illustrated in FIG. 3 , and it comprises two modules. The general architecture remains the same overall. The first module (DRM) manages the dimensionality reduction, which results in obtaining only a reduced set of QoS features that contribute to the label prediction variation (the predicted QoE (LblsPred) generated by the application QoE model, i.e., which label is generated by the application QoE model according to the current measurement parameters; the label can be, for example, an MoS, i.e., an aggregate function or a multi-MoS of continuous or discrete functions). The second module, MFIM, performs an evaluation of the importance of QoS features (some of which may have been discovered by the first module) for the application QoE model itself and updates the application QoE model itself to improve its stability over time.

[0052] More specifically, - The Dimensionality Reduction (DRM) node computes from the input variables (measured QoE, predicted QoE, measured feature QoS, predicted feature QoS) a reduced set of features that necessarily have a strong influence on the difference between measured and predicted QoE. - The Model Feature Importance Node (MFIM) calculates the importance of the reduced set of QoS features relative to the actual application QoE model and updates the application QoE model (AppMdl) if necessary.

[0053] Achieving these results, especially the reduction in dimensionality, assumes that the application QoE model (AppMdl) is maintained, as explained above. More specifically, a prerequisite for the implementation of this method is based on an application model for QoE, which links application satisfaction indicators (i.e., represented QoE in the form of labels) to the QoS characteristics of the SNPN and the configuration parameters of the SNPN. The application QoE model can take several forms, namely: The application model is obtained, for example, by training a neural network using a selected training dataset (QoE labels, QoS features, network parameters), The application model may be obtained by training a stacked set of neural networks with some selected training datasets (QoE labels, QoS features, network parameters), The application model may be obtained by training a convolutional neural network using a selected training dataset (QoE labels, QoS features, network parameters); The application model may be obtained by training a recurrent neural network with a selected training dataset (QoE labels, QoS features, network parameters).

[0054] Using the application QoE model (AppMdl), these two nodes (modules) are used to find a smaller set of variables for MPC with reduced network dimensionality, and feature importance is used to find the network configuration to be applied for control, and finally update the application QoE model. These two modules (or nodes) implement one embodiment of the method disclosed above. More specifically, in such an implementation, the following iterative method is processed: - obtaining QoE measurements of deployed applications and / or sets of applications (QoE labels), network conditions described using QoS features, and parameters generating the QoE metrics (features) and QoS features, a. Measurements are obtained by measuring the satisfaction of a client application with a particular configuration of the SNPN, e.g., the application client provides labels representing the current QoE, which can be linked to some QoS parameters associated with these QoE labels; b. In a complementary or alternative method, measurements are obtained by running a simulation of a digital twin of the SNPN deployment in which some QoS parameters are changed or downgraded; - Calculating the deviation of the measured QoE from the QoE predicted from the application QoE model. In one embodiment, the deviation calculation is done by calculating the squared difference between the measured and predicted QoE, where the label is in the form of a number or by calculating a function of the squared difference, in another embodiment the deviation is estimated by a neural network that has previously trained a model of deviation over the features (QoS and network parameters). - analyzing the calculated QoE deviations and determining nominal and outlier feature deviation regions. According to one embodiment, the nominal feature deviation region is determined as the region of the mean QoE deviation. According to one embodiment, the nominal feature deviation region is determined as the mean QoE deviation in a specific region fixed by the SNPN management layer. According to one embodiment, the outlier feature deviation region is a high QoE deviation fixed by the SNPN management layer. According to one embodiment, the outlier feature deviation region is a high QoE deviation level learned by the Isolation Forest. According to one embodiment, the nominal feature deviation region is determined as the mean QoE deviation. The mean QoE deviation is obtained from a dataset obtained from QoE measurements and QoE values ​​predicted by a QoE model, from which the outlier feature deviation region is extracted. According to one embodiment, the nominal and outlier feature deviation regions are based on clustering performed on a dataset obtained from QoE measurements and QoE values ​​predicted by the QoE model. According to one variant, the clustering is performed using deep clustering techniques. According to one variant, the clustering is performed using the K-means algorithm. - evaluating the importance of the features of the analyzed deviation regions. The feature importance determines the QoS parameters that are most responsible for the nominal and high (outlier) deviations of the QoE. According to one embodiment, the feature importance is calculated by the Shapley index of the QoS and parameter features relative to the model of the deviation of the QoE. According to another embodiment, namely b, the feature importance is calculated by evaluating the correlation between the feature and the QoE deviation. The importance of these features relates to the level of correlation between the feature and the QoE deviation. - evaluating the importance of the features of the application QoE model, which determines the QoS parameters that have the strongest influence on the application QoE model. According to one embodiment, the importance of the features is calculated by the Shapley index of the QoS and parameter features for the QoE model. According to another embodiment, the importance of the features is calculated by evaluating the correlation between the features and the QoE deviation. The importance of these features relates to the level of correlation between the features and the QoE data. - determining an action to be performed on the determined parameters based on the feature importance of the deviations and the application QoE model, where the action consists of updating (increasing or decreasing) parameters that exhibit a high feature importance index for deviations of the QoE dataset without impacting the application QoE model. The following possibilities are available for action determination: The action could be to update the parameters that show the most feature importance to QoE outliers if these parameters do not have a strong impact on the application QoE model. In this case, an alert is delivered to the SNPN management layer to notify the deployment of future abnormal events. - If features with high importance also have a strong impact on the application QoE model, the application QoE model is updated based on the importance of the features. - Apply the operation to the determined parameters of the SNPN deployment and return to the first step.

[0055] The proposed method is fully automatic, allowing for the management of thousands of parameters and the tuning of QoS to specific QoE requirements or objectives in real time. Of course, this result cannot be achieved by manually managing these parameters, as this would result in parameterization that is not adaptable to rapidly changing environments as a real-time process. Implementation of the disclosed method reduces both the complexity of parameter tuning and deviation from a stable point. It should be noted that the present disclosure is not limited to the above exemplary embodiment. Variations of the above exemplary embodiment are also within the scope of the present disclosure.

[0056] As shown in FIG. 4, a device DV for updating parameter values ​​of at least one parameter related to the quality of experience (QoE) of the execution of at least one application (App) may include an antenna system AS coupled to a processing circuit including a processor PROC and a memory MEM via an interface IN, as presented above. Alternatively, as a complement to, or as a variant of the antenna system AS, necessary data (e.g., measured QoE, network parameters, Qos parameters) may also be received via a communication link CL, enabling the device to determine which parameters to update in accordance with the present disclosure. The memory stores at least instructions for a computer program according to the present disclosure. By performing this type of processing, and in particular by implementing dimensionality reduction and key feature identification according to the method of the present disclosure, it has been proven that the device's processor can identify both key features in the loss of QoE and key features of the model in a short time while stabilizing the application QoE model. This effect is obviously not possible by attempting to calculate or obtain these operation sequences manually or by thought, as this would be too time-consuming and completely pointless given the complexity of the problem to be solved.

[0057] In a preferred embodiment, the computer system or computer device comprises one or more processors (which may belong to the same computer or different computers) and one or more memories (magnetic hard disks, optical disks, electronic memories, or any computer-readable storage medium) in which a computer program product is stored in the form of a set of program code instructions to be executed to perform all or part of the steps of the determination method. Alternatively, or in combination thereof, the computer system may comprise one or more programmable logic circuits (FPGAs, PLDs, etc.) and / or one or more special purpose integrated circuits (ASICs, etc.) adapted to perform all or part of said steps of the determination method. In other words, the computer system comprises a set of means constituted by software (specific computer program products) and / or hardware (processors, FPGAs, PLDs, ASICs, etc.) for performing the steps of the determination method.

Claims

1. 1. A method for updating a parameter value of at least one parameter related to a Quality of Experience (QoE) of the execution of at least one application (App), said at least one application (App) comprising an Application Server (AS) exchanging data with an Application Client (AC) via a wireless communication system (SNPN), said method being implemented in a Network Controller Device (NCD) of said wireless communication system (SNPN), said method comprising the following steps: - determining (10) from the QoE measurements of the at least one application (App) at least one feature (Fea) related to the wireless communication system that contributes to a deviation between the QoE measurements of the at least one application (App) and an application QoE model associated with the at least one application (App), the application QoE model being configured to predict a variation in the QoE of the application (App); - obtaining (20) from said at least one feature (Fea) at least one QoS parameter associated with said deviation; updating (30) a parameter value of said at least one QoS parameter associated with said deviation; The method of claim 1, further comprising at least one repetition of

2. The method of claim 1 , wherein the QoE measurements are obtained from the Application Client (AC) and / or from monitoring the Quality of Experience (QoE) at the Application Server (AS).

3. The step (10) of determining the at least one feature (Fea) responsible for the deviation between the QoE measurement and the QoE obtained from the application QoE model associated with the at least one application (App) comprises: - obtaining (101) QoE measurements of said at least one application (App) (label) and parameter identifiers generating QoE metrics (characteristics) related to said wireless communication system (SNPN); Calculating (102) the deviation of the QoE measurement from the QoE predicted from the application QoE model, the model being learned during nominal application operation in the wireless communication system and / or the model being obtained from an application (App) management layer; - analyzing the deviation of the QoE and determining a nominal deviation region and an outlier QoE deviation region (103); determining (104) the at least one feature (Fea) included in the nominal deviation region and the outlier QoE deviation region; The method of claim 1 , comprising:

4. 3. The method of claim 2, wherein the step of obtaining (101) QoE measurements for the at least one application (App) comprises measuring the satisfaction of the at least one application client (AC) with a particular configuration of the wireless communication system (SNPN).

5. 5. The method of claim 4, wherein the specific configuration of the wireless communication system includes at least one set of QoS parameters associated with at least one communication session opened for the at least one application (App).

6. 3. The method of claim 2, wherein the step of obtaining (101) QoE measurements for the at least one application (App) comprises measuring satisfaction of the at least one application client (AC) obtained from a digital twin of the wireless communication system (SNPN) in which execution of the at least one application (App) is simulated.

7. The step (20) of obtaining the at least one QoS parameter associated with the deviation comprises: Calculating (201) a first contribution of each of said at least one feature (Fea) to said deviation of said QoE in a nominal deviation region and an outlier QoE deviation region; determining (202) the QoS parameter associated with the determined at least one feature of the nominal and outlier deviations of the QoE from the first contribution of each of the at least one feature (Fea) to the deviation of the QoE; determining a second contribution of each of the at least one feature (Fea) in the prediction of the application QoE model, and obtaining the QoS parameter associated with the determined at least one feature (Fea), which takes priority over the prediction of the application QoE model (203); The method of claim 1 , comprising:

8. The step (30) of updating a parameter value of said at least one QoS parameter associated with said deviation comprises: determining parameter values ​​for updating at least a portion of the at least one QoS parameter; applying the updated parameter values ​​to the updated determined QoS parameters; The method of claim 1 , comprising:

9. The method according to claim 1 , further comprising updating the application QoE model in response to at least some of the at least one feature (Fea).

10. 10. The method according to claim 1, wherein the application QoE model associates at least one satisfaction indicator of the at least one application with at least one QoS parameter and at least one configuration parameter of the wireless communication system (SNPN).

11. 10. The method of claim 1, wherein the wireless communication system (SNPN) comprises a 5G standalone non-public network.

12. 1. A device for updating a parameter value of at least one parameter related to a Quality of Experience (QoE) of an execution of at least one application (App), said at least one application (App) comprising an Application Server (AS) exchanging data with an Application Client (AC) via a wireless communication system (SNPN), said device being part of a Network Controller Device (NCD) of said wireless communication system (SNPN), said device comprising: determining, from QoE measurements of at least one application (App), at least one feature (Fea) related to a wireless communication system, the feature being responsible for a deviation between the QoE measurements of the at least one application (App) and an application QoE model associated with the at least one application (App), the application QoE model being configured to predict a variation in the QoE of the application (App); obtaining at least one QoS parameter associated with said deviation from said at least one feature (Fea); updating a parameter value of the at least one QoS parameter associated with the deviation; 20. A device comprising: at least one module for:

13. A computer program product comprising instructions that, when executed by at least one processor, configure the at least one processor to perform the method of any one of claims 1 to 11.

Images