Method and apparatus for providing human interpretable semantic information of network status

Abstract

A method, network node, and computer-readable medium for providing human interpretable semantic information of network states are described herein. The method includes at least receiving a data element and an information set, the data element and the information set identifying a predefined test case pursuant to which a SWNC is executed in a second target communication network; training, based on the received data element and the information set, a network state information model to determine human interpretable semantic information of at least one network state of the identified SWNC during execution in a first target communication network based on the identified network state; applying at least one network model comprising the trained network state information model to the first target communication network to generate at least one human interpretable semantic information of one or more network states of the SWNC determined during execution in the first target communication network; and reporting, by the trained network state information model, at least one human interpretable semantic information associated with the one or more network states.

Description

Technical Field

[0001] This disclosure relates to a method and apparatus for providing human-interpretable semantic information about one or more network states generated during the execution of a network component of a target communication network or a corresponding network digital twin (NDT) of the target communication network. Background Technology

[0002] Any discussion of the background technology in this specification should not be construed as an admission that such technology is well known or constitutes part of common knowledge in the field.

[0003] Network Digital Twins (NDTs) are complex virtual replicas of physical networks, created by integrating real-time data and advanced simulation techniques to reflect the complex behavior, architecture, and performance of real-world communication networks. These digital models provide organizations with powerful tools to gain in-depth, actionable insights into their network operations, thereby facilitating more strategic decision-making and improved operational efficiency. Specifically, by enabling the simulation of various scenarios and network conditions, NDTs allow for the proactive identification and resolution of potential problems, often before they manifest in the physical network. This capability is particularly valuable in complex and dynamic environments such as telecommunications, cloud computing, and the Internet of Things (IoT), where reliability, performance, and scalability are critical requirements. Therefore, NDTs enhance organizations' ability to optimize resources, reduce downtime, accelerate the deployment of new services, and mitigate risk, while also significantly reducing costs.

[0004] Furthermore, to leverage the power of NDT in predefined communication network systems and to further elucidate the positive impacts generated by using NDT, it is worth noting that, as part of the broader ETSI Zero Contact Network and Service Management (ZSM) initiative, specifically work item ZSM015, there has been a long-standing commitment to utilizing the power of NDT, particularly for zero-contact management of networks and services. In this paper, the ZSM framework specifically emphasizes the necessity for automation and elimination of human intervention, enabling intelligent and autonomous network operation. In this context, NDT must support multiple functions, including the ability to provide predictions about network performance and behavior. These predictions are crucial for anticipating network problems, optimizing performance, and ensuring efficient resource allocation. Moreover, the ZSM framework also emphasizes the importance of providing NDT with analytical and diagnostic capabilities, such as Root Cause Analysis (RCA), to help diagnose and resolve problems within a given communication network simulated by NDT.

[0005] In this regard, for NDT to effectively perform these functions, it must model the dynamic behavior of its corresponding communication network and its services at an appropriate level of abstraction. This allows for the prediction of future network states, performance, and behavior, while also enabling the effective correlation of different situations occurring in multi-element communication networks without semantically linking each piece of information identified during simulation.

[0006] However, accurate predictions are not enough; they must also be interpretable. This means that the predicted states and behaviors should ideally be represented in a way that is directly related to the physical network components, rather than existing in abstract modeling or AI-driven latent spaces that are not easily interpretable by network operators or domain experts.

[0007] For example, in the paper "Real-Time Analysis of Multiple Root Causes for Anomalies Assisted by Digital Twin in NFV Environment" by Wang et al. (IEEE Transactions on Network and Service Management, Vol. 19, No. 2, 2022), a root cause analysis method for virtual network functions (VNFs) was proposed to establish fault propagation paths based on the tracking and correlation of different physical and virtual component states.

[0008] However, these outputs often cannot be directly mapped back to the original data input, making labeling network states a complex task. Furthermore, dynamic state models capture state transitions over time, and the generated states are sometimes interrupted, interleaved, or influenced by other states applied to the system or by the environment, adding another layer of complexity and making it difficult to identify the root cause of a given state. Therefore, these models are sensitive to changes in network behavior, and the resulting state trajectories increase the semantic distance between the model output and the original input.

[0009] Therefore, in summary, there is a need to develop an accurate and efficient method to predict the network states occurring during the execution of a given NDT and / or (ideally) in the corresponding communication network simulated by the NDT, and to ensure that these predictions are interpreted in an easily understandable manner. This enables operators and / or users to interpret the predicted network behavior and performance within the context of the actual network, thereby facilitating better decision-making and proactive management of network resources. Summary of the Invention

[0010] The scope of protection sought by the various exemplary embodiments of this disclosure is defined by the independent claims. Exemplary embodiments and features (if any) described in this specification that are not within the scope of the independent claims should be interpreted as examples that aid in understanding the various exemplary embodiments of this disclosure.

[0011] Furthermore, to further clarify the different terms used in this specification, it should be noted that network nodes mentioned in this disclosure can be network elements. Network elements (such as communication elements) can also be terminal equipment, control elements or functions (such as access network elements, such as base stations / BS, gNBs, radio network controllers), core network control elements or functions (such as gateway elements), or other network elements or functions. Furthermore, as described herein, the UE and any other elements, functions, or applications can be implemented by software (e.g., by a computer program product for a computer) and / or by hardware. Moreover, in order to perform their respective processing, the corresponding devices, nodes, functions, or network elements used may include various parts, modules, units, components, etc., required for control, processing, and / or communication / signaling functionality, but are not directly shown herein for ease of understanding.

[0012] In light of this, and to address the aforementioned problems, a computer-readable medium for the method, apparatus, and storage instructions according to the independent claims introduced herein is proposed. This medium is specifically designed to provide a predetermined user or system with human-interpretable semantic information about the network state of a software-defined network component (SWNC) generated in at least one first target communication network. In particular, it is designed to generate information about one or more network states generated during SWNC execution that can be interpreted and correlated by a domain expert.

[0013] Therefore, according to this disclosure, a method can be provided for providing human-interpretable semantic information about the network state of a software-defined networking component (SWNC), the human-interpretable semantic information being generated in a first target communication network, the method comprising: Receive data elements and information sets, which identify the predefined test cases on which SWNC is executed in the second target communication network; Based on the received data elements and information set, a network state information model is trained to determine human-interpretable semantic information of at least one network state of the SWNC identified during execution in the first target communication network based on the identified network state. At least one network model, including a trained network state information model, is applied to a first target communication network to generate at least one human-interpretable semantic information about one or more network states of the SWNC determined during execution in the first target communication network; and A trained network state information model reports at least one human-interpretable semantic information associated with one or more network states to, for example, network components or operators.

[0014] As a result, the subject matter of the invention presented herein specifically relates to a mechanism provided by a network state information model based on artificial intelligence (AI) and machine learning (ML), which is embedded in the initial training process, wherein the corresponding network state information model is initially fed at least with the corresponding test case execution information (i.e., the aforementioned data elements and information set) for a given SWNC. This mechanism is capable of generating human-interpretable semantic information related to one or more network states generated during the (undefined) execution of the corresponding SWNC, thereby achieving the aforementioned effect of ensuring the understandable identification of the generated network states and the conditions existing in the corresponding communication network.

[0015] In some examples herein, the first target communication network can be further defined as a network digital twin (NDT) of either the target or second target communication network. This results in the effect that the provided trained network state information model can be applied both to the same real communication network to which it was trained and to the corresponding NDT simulating that given real communication network. Therefore, a specific effect can be achieved whereby the network state information model provided in this disclosure is trained on real events occurring during the execution of predefined test cases in an actual (i.e., real) communication network, and its application can be further transferred to a simulated network provided by the NDT, thereby allowing the generation of network state information closely related to real-world occurrences.

[0016] In addition, in some examples, the data elements and information set for identifying predefined test cases may further include at least one of the following: the predefined test conditions on which the SWNC is executed, including the expected SWNC behavior and the expected output of the SWNC, which generates state information about the network state when the SWNC is executed under the predefined test conditions, and the observable test output of the SWNC identified after the SWNC is executed using the associated test conditions.

[0017] In this document, predefined test conditions may involve various execution parameters upon which the corresponding SWNC is executed, while the actual content of the test conditions may involve specific parameters related to one or more situations that a given SWNC may face. For example, potential examples of these test conditions may be: different network traffic levels when the SWNC is used for processes related to network load and traffic patterns; network topology, network link quality, security vulnerability parameters, or different QoS parameters under varying network configurations. Furthermore, the expected SWNC behavior may correspond to predefined or ideal performance and operational characteristics of the SWNC, including information related to the corresponding network state generated during the execution of the SWNC under the aforementioned predefined test conditions. For this purpose, as an example, other examples of the expected SWNC behavior may be SWNC characteristics relating to: performance metrics (throughput, latency, packet loss, jitter, error rate, etc.), resource usage (e.g., CPU / memory utilization), state transitions and fault recovery (state changes, recovery behavior), security and anomaly detection (intrusion detection, access control), or operational boundaries (thresholds, limits, stress behavior). Furthermore, "expected output" can refer to the ideal or predefined result or output when SWNC is executed in the corresponding test case, while "observable test output" can refer to the actual result or outcome produced by the network component under predefined test conditions during test case execution. In this regard, the corresponding observable test output information may be much more detailed than the information captured in the expected output, especially since it also takes into account abnormal or unexpected network behavior (such as abnormal bottleneck characteristics).

[0018] Furthermore, in some embodiments, the method for training a network state information model using predefined data elements and information sets may further include: Test cases are executed by performing the SWNC of the second target communication network under associated test conditions and determining the output of the SWNC as observable test outputs to generate data elements and information sets used for training network state information.

[0019] Therefore, in addition to actually training the corresponding network state information model, the mechanism of the invention provided in this paper can also include the execution of actual test cases of the corresponding SWNC in the target communication network. This results in a mechanism that defines a self-sufficient system capable of generating the corresponding observable outputs required for model training by providing a predefined number of conditions used to define test cases. Both systems have the ability to train and apply the corresponding network state information model to a given target communication network and / or NDT. Furthermore, by including the execution of actual test cases in the proposed mechanism, the training process can be further customized according to the actual conditions required to describe the corresponding SWNC behavior, leading to a more accurate and resource-efficient provision of the required human-interpretable semantic information.

[0020] Furthermore, in some examples, when considering the order in which the corresponding network state information model is trained, the corresponding steps for training the network state information model may also include: Based on the received data elements and information set, a network state information model is trained in the state recognition model segment of the network state information model to identify the network state of the SWNC based on the observable output after the SWNC executes in the first target communication network; and Based on the received data elements and information set, as well as the identified network state, the network state information model is trained in the state information generation model segment of the network state information model to generate human-interpretable semantic information associated with the identified network state.

[0021] Therefore, in the first exemplary embodiment of the present invention, the corresponding network state information model may include at least two different functional model segments (state recognition model segment and state information generation model segment), which can be trained and applied independently or at least sequentially, thereby avoiding additional complexity during training and improving resource efficiency.

[0022] In this regard, the first model segment, namely the state recognition model segment, can correspond to a trainable state recognition model function, which is trained and thus configured to recognize the associated network states of the corresponding SWNC when it is executed under unknown conditions (i.e., the actual application of the SWNC in the target communication network or NDT). To this end, the corresponding state recognition model segment can be trained by: receiving the aforementioned data elements and information set to associate a given number of network states with one or more observable outputs generated by the SWNC during process execution; and training the corresponding model segment to associate a given observed output with one or more network states introduced during the training phase (e.g., network state information included in the received data elements and information set).

[0023] Furthermore, the second model segment, namely the network state information model, can correspond to a trainable network state information generation function, which is trained to determine and generate corresponding human-interpretable semantic information based on the corresponding network state identified by the aforementioned state recognition model segment of the network state information model. In this paper, the training of the second model segment can be performed independently of the training of the first model segment, and / or sequentially after the training of the first model segment is completed. Furthermore, the corresponding training can be performed as follows: the second model segment receives one or more network states identified and learned during the training of the first model segment, data elements and information sets for network state recognition, or both, as general training input, and based on this information, trains the corresponding model segment to associate some or all of the identified network states with interpretable information already existing during the execution of the corresponding test cases (e.g., information obtained from data elements and information sets describing a given test case associated with the identified network state), training the second model segment to associate given human-interpretable semantic information with one or more network states identified by the first model segment.

[0024] Therefore, in the first embodiment of the network state information model proposed in this paper, the training of the corresponding network state information model can lead to a bifunctional identification model, wherein, based on two different model segments, human-interpretable semantic information related to one or more network states of the SWNC can be generated through the following steps: in the first step, the current network state of the SWNC is determined based on the given observable SWNC output, and then in the second step, the sought human-interpretable semantic information is generated and associated with the identified network state, both of which are in the same network state information model.

[0025] Based on this, in another example, the step of applying the corresponding network state information model to generate the corresponding human-interpretable semantic information of the SWNC in the first target communication network may further include: Under unknown conditions, perform SWNC in the first target communication network to generate observable output; and The trained network state information model is applied to the first target communication network by: the state recognition model segment of the trained network state information model determining the current network state of the executed SWNC based on the generated observable output; and the state information generation model segment of the network state information model generating at least one human-interpretable semantic information of one or more network states of the SWNC based on the network state determined by the state recognition model segment.

[0026] Therefore, in the first embodiment of the network state information model proposed in this paper, the network state information model includes dual functions and its trained characteristics: identifying the network state of the corresponding SWNC based on the output generated by the SWNC, and subsequently identifying and generating human-interpretable information associated with the identified network state. This can generate a fully automated mechanism that provides understandable information attributable to one or more network states based solely on a single output of the corresponding SWNC. Furthermore, since the corresponding model segment is trained via multiple test cases executed on the same SWNC under different test conditions, high accuracy in network state and information output is enabled. This is because the model segment can "identify" which test case scenarios best match the current operating conditions of the SWNC identified by the network state information model (based on observed SWNC outputs) and provide accurate network state information by invoking the matching test case descriptions (e.g., the set of data elements and information associated with the corresponding test case).

[0027] In contrast, in the second embodiment of the invention, it is also possible that the network state information model disclosed herein may not be configured to simultaneously provide integrated network state identification and network state information generation capabilities, but rather to only identify and generate corresponding human-interpretable information associated with network states previously identified by other (i.e., external) network state identification models. This second embodiment is particularly useful where a given network system already includes a fully functional network state detection mechanism or may have certain specific characteristics that preclude the use of the aforementioned dual-function model (e.g., where the network system may include specific security features that mandate the use of an external state identification model), thereby enabling the network state identification model of this second embodiment to enhance the integrability of existing mature network systems.

[0028] Therefore, and regarding the second embodiment of the invention disclosed herein, another example of the training steps for training the corresponding network state information model may also include the following steps: The received data elements and information set are sent to the network state model, which is configured to identify the network state of the SWNC based on the observable test output of the SWNC, which is being executed in the second target communication network. Based on the transmitted data elements and information set, this network state model is applied to identify the network state of the SWNC being performed in the second target communication network based on the imported data elements and information set; and The network state information model is trained based on the received data elements and information set, as well as the identified network states, to generate human-interpretable semantic information associated with at least one of the identified network states.

[0029] Therefore, regarding the second embodiment of the invention, the same training process can be implemented for recognizing and generating human-interpretable semantic information about the network state of the identified SWNC. In contrast, the difference lies in the dual-functionality of the network state information model in the first embodiment, which requires simultaneous training of a network state recognition model segment for network state recognition and a state information generation model for determining the corresponding information associated with the identified network state. The corresponding training process in the second embodiment may only require training the corresponding network state information determination (previously implemented by the state information generation model), while the determination of the current network state (which is then used to train the network state information model) can be handled separately by the already trained network state model in the corresponding network system. Therefore, in this way, the integrability of the corresponding network state information model can be improved, and resources required for training the corresponding model can also be saved, because the model provided herein only needs to train one specific output (i.e., human-interpretable semantic information), instead of two outputs (i.e., training state recognition and then determining state information based on state recognition).

[0030] Furthermore, the same technical improvement can be achieved by applying the corresponding trained network state information model to the first communication network performing the corresponding SWNC under unknown conditions, especially since in this embodiment, the trained model is only configured to determine and output the corresponding human-interpretable semantic information of the network state previously identified by the trained network state model.

[0031] Therefore, in another embodiment of the present invention, the step of applying at least one network model to the first target communication network in the second embodiment of the present invention may further include the following steps: SWNC is performed in the first target communication network under unknown conditions in order to generate observable output; A network state model is applied by determining the existing network state of the performed SWNC based on the generated observable outputs; and The network state information model is applied by generating at least one human-interpretable semantic information of the network state of SWNC based on the network state determined by the network state model.

[0032] It should also be noted in this paper that the described "network state model" can be a network state model previously used to train the network state information model in the associated training phase, especially to enhance the recognition accuracy of the corresponding trained model. In another example embodiment, it is also possible to use different network state models in the training and application phases to improve the usability of the proposed model. Meanwhile, the already trained network state model can also be trained with the same data / information (i.e., data elements and information sets) used to train the network state information model, but it can also be pre-trained, i.e., trained with previously existing information, thus producing the effect that the "trained network state model" can be regarded as any network state recognition model existing in the system, including conventional or traditional state recognition models. Based on this, the independence of the network state information model thus generated in the second embodiment therefore enables the effective reuse of existing network state models and allows them to be transformed into an extended model mechanism that, in addition to actually recognizing the corresponding state, is also configured to output understandable information about one or more network states among the recognized network states.

[0033] Therefore, based on the two embodiments presented herein, a precise and particularly resource-efficient method can be provided to generate human-interpretable semantic information related to one or more network states generated during the actual execution of the SWNC in the target communication network or the corresponding NDT, thereby achieving the effect of significantly enhancing the prediction and analysis required for normal network operation, such as in the case of fault analysis.

[0034] Furthermore, alternative configurations of the above embodiments can bring further improvements to the proposed mechanism.

[0035] For example, in another example configuration of any of the embodiments described above, the step of reporting at least one human-interpretable semantic information may further include: Annotate one or more network states of the SWNC using at least one associated human-interpretable semantic information; and Report one or more network states to, for example, network components or operators, where associated human-interpretable semantic information is annotated to the network state.

[0036] Therefore, in addition to outputting the given human-interpretable semantic information normally to the corresponding user or system element (e.g., to inform a given user of a specific context / situation discovered during network state detection via state information), the corresponding human-interpretable semantic information can also be used as annotation, wherein the corresponding identified network state is associated with the corresponding network state information. Furthermore, not only the corresponding network state information, but also the associated network states annotated with that information are also output to the user / system element.

[0037] Therefore, based on the above configuration, the network state information model proposed in this paper can also be equipped with an automatic network state information self-annotation function. Through this function, the generated state information can be automatically coupled with and stored to its corresponding network state. In this paper, this may lead to additional improved resource efficiency within the corresponding network system, mainly because, in the case of continuous fault analysis, the corresponding state information can be effectively read from the stored annotated data without repeating the application process of the network state information model, and it is associated with the given network state. Furthermore, by continuously annotating the corresponding network state with information about the SWNC environment / behavior (e.g., fault causes, system parameters, etc.), specific network states or even those representing the root cause of a given network behavior can be effectively identified, as this will aggregate a large amount of similar network state information for additional analysis.

[0038] Furthermore, the actual annotation execution can be performed through different implementations. In the example embodiment, annotations can be stored as metadata along with network status data in the Network Management System (NMS). In other implementations, network devices can also embed annotations into Simple Network Management Protocol (SNMP) or Fraps to provide contextual information for events such as link failures or performance degradation. Additionally, other embodiments can use telemetry data (to which annotations can be appended to real-time telemetry data streams), system logs and event logs, real-time dashboards (displayed as tooltips or labels on a network monitoring dashboard), or via API calls for annotation assignment.

[0039] Furthermore, in another embodiment of the invention, the data elements and information sets that identify predefined test cases correspond to log information generated during the execution of the SWNC continuous integration / continuous development (CI / CD) pipeline. This method also includes: Perform CI / CD pipeline execution on different test cases of SWNC to generate data elements and information sets for training the network state information model. The steps of performing CI / CD pipeline execution include at least: performing test case execution for multiple test cases by repeatedly executing SWNC of the second target communication network under predefined test conditions, and collecting log information generated during test case execution, which includes at least information about: the associated test conditions, the generated network state, and the observable test output of SWNC.

[0040] Therefore, in another configuration of the invention, the execution of test cases for model training can be performed specifically within the CI / CD environment of the corresponding SWNC, and each piece of information required for model training can be obtained through log information and / or protocol inputs and outputs during CI / CD. In this paper, using CI / CD pipelined execution particularly offers the advantage that, for a given SWNC execution, each possible test case condition is tested at least once, resulting in the provision of training data with the widest coverage and thus the most accurate network state information model training. Furthermore, since both CI / CD execution and logging of execution data are typically implemented in the general model implementation workflow of network systems, maximum resource efficiency can be achieved for generating training data (as no additional training data generation process is required), while the corresponding training data is already customized for the execution behavior of the corresponding SWNC, and the trained network state information model is subsequently applied to that SWNC.

[0041] Furthermore, in another example, training the corresponding network state information model may also include at least the following steps: The selection of network states specific to test cases is performed, with each test case associated with data elements and information sets used to train the network state information model; and Train a network state information model to associate the selected network state with data elements and information sets.

[0042] In this regard, an alternative configuration can be implemented during the training process of the proposed network state information model, which may include an initial additional process of selecting only network states specific to the test cases for test case-related model training. Specifically, this is to enable the network state information model to learn the description of the corresponding identified network states based solely on the network states determined for a given SWNC condition and / or behavior.

[0043] In this paper, the reason for this additional process may lie in the fact that during the execution of test cases in general communication networks (especially dynamic networks), many different network states typically arise due to network initialization, general configuration, resource allocation, bootstrapping, etc. The purpose of these operations is, for example, simply to prepare the corresponding communication network (i.e., to bring it to the required conditions) where the key parts of the test cases can be executed. Simultaneously, during each test case, some final steps may also be performed, such as simply releasing resources or disabling configurations. These are unrelated to the conditions / behaviors of the test case itself, resulting in the effect that some network states generated and identified in a given test case often do not originate from the specific test conditions defining the given test case, but rather from the system-related execution requirements. Therefore, by identifying and selecting the corresponding network states specific only to a given test case (and thus the associated test case conditions) before the actual learning phase, and then only importing such specific network states into the network state information model for actual training, unnecessary information in the training phase can be effectively eliminated. Simultaneously, the connection between the identified network states for the corresponding test case and their corresponding conditions / test case descriptions can be narrowed down to the actual network states associated with that test case. Therefore, with the aforementioned additional configurations, a more resource-efficient and accurate training process can be generated.

[0044] Furthermore, the actual process for selecting the initial network state can also be implemented with variant configurations or mechanisms to identify specific network states for corresponding test cases. In this paper, one configuration of such a mechanism can, for example, rely on a direct comparison of the network states identified for the corresponding test case with the network states identified by the remaining test cases executed for training the network state information model.

[0045] Therefore, in another example of the invention proposed herein, the selection of network states specific to the test case can be performed by comparing the network state of the SWNC associated with the test case with the network state of the SWNC that exists during the execution of other test cases in the second target communication network, and then excluding the network states of the SWNC associated with the test case that share a predefined amount of similarity with the network states of other test cases.

[0046] In this regard, it is important to note that the “similarity quantity” defined in the above configuration should not be confused with any specific similarity feature or parameter that may identify the similarity of a given network state to other existing network states, but should be understood as any parameter that can express the degree of probability that a given network state of a test case will also appear in other test case examples.

[0047] For example, possible methods for expressing the similarity of a given network state can be generated by performing cosine similarity on different network states identified in each test case. Specifically, this is done by forming each corresponding network state into a numerical vector and then measuring the similarity of each vector using cosine similarity processing. Other possibilities also include using Euclidean or Manhattan distance calculations, clustering-based methods, feature matching, or heatmap comparisons, each outputting a numerical value representing the similarity of a given network state to network states generated and / or identified in other test cases. Furthermore, to ultimately exclude and / or select given network states for model training, the corresponding numerical similarity values ​​can be additionally compared to a predefined threshold in the final step, effectively specifying the corresponding network states required for subsequent training phases.

[0048] Based on this, the resulting pre-selection of network states for test cases allows for more accurate training of the corresponding network state information model, and thus also allows for more accurate identification of the required human-interpretable semantic information associated with the corresponding network state based on the improved training.

[0049] Furthermore, in another example embodiment, it can further improve the application of the trained model itself, mainly because it can also help specify which types of human-interpretable semantic information should be included in the associated network states determined at the corresponding model application stage.

[0050] Therefore, in another example configuration of the invention proposed herein, it is equally possible, for example, that the human-interpretable semantic information may additionally correspond to at least a combination of information included in the data elements and information sets of different test cases, wherein the SWNC is executed in the second target communication network according to the different test cases; wherein, in order to generate at least one human-interpretable semantic information, the network state information model may additionally identify the network state of the SWNC being executed as a combination of network states associated with one or more test cases in the different test cases, and subsequently generate at least one human-interpretable semantic information based on the information included in the data elements and information sets corresponding to the one or more test cases in the different test cases.

[0051] In this regard, a trained network state information model can be used to determine the correct human-interpretable semantic information associated with one or more given network states through a deterministic model reorganization evaluation step. In this step, when requesting human-interpretable semantic information for network states generated by the SWNC under unknown execution conditions, the corresponding identified network states are initially determined as a combination of network states pre-associated with one or more test cases. Then, the corresponding human-interpretable semantic information is generated based on the data elements and information set associated with the corresponding (combined) test cases. Therefore, since the content to be included in the human-interpretable semantic information depends directly on the specific network state assigned to each previously used test case through this configuration, the aforementioned pre-selection of test case-specific network states further improves the accuracy achieved during the generation of human-interpretable semantic information, mainly because information that might be added from non-specific network states can be effectively omitted.

[0052] Furthermore, the reason for generating the corresponding human-interpretable semantic information as a combination of selected test case descriptions (i.e., the content of data elements and information sets associated with the corresponding test cases) can also be attributed to the fact that, even for well-trained recognition models, semantically describing network states identified under unknown execution conditions (i.e., the actual conditions of SWNC) remains a challenging task. Specifically, when applying the corresponding trained network state information model to actual network deployments, it is important to note that, under normal conditions, the trained model often faces the challenge that the observed network states are caused by multiple effects simultaneously accumulated in the network. As a result, while training the corresponding model based on several independent test cases allows the trained model to associate a predetermined number of network states with a given test case description (these associations can then be used to form the corresponding human-interpretable semantic information), the trained network state information model may still ultimately output fairly diverse and / or erroneous information, especially since it does not take into account the simultaneous accumulation of the aforementioned different network conditions. For this reason, in order to further enhance the accuracy of the description of the network states identified in a given (actual) use case of SWNC, the corresponding network state information model can also be able to determine the network states identified in a given use case as a combination of network states associated with different simultaneous test cases (i.e., different conditions applied to SWNC). This can effectively simulate the aggregation of the aforementioned different network conditions and thus further converge the output human-interpretable semantic information to the actual situation applied in the corresponding use case.

[0053] Furthermore, regarding the actual implementation of the above process, in another example, identifying network states as combinations of network states associated with one or more test cases in different test cases can be specifically performed by: demultiplexing network states associated with different test cases, and identifying combinations of one or more test cases in different test cases, where the combination of the corresponding demultiplexed network states shows a predefined amount of similarity to the network states of the SWNC being executed.

[0054] In this paper, "demultiplexing" specifically refers to the process of separating or decoding multiple network states that are combined to identify the dynamic network state of a given test case into a single transmission or dataset. This produces the effect that each network state within a single network state of the test case can be analyzed accordingly and used to execute the aforementioned test case combination process. Furthermore, various execution configurations known to those skilled in the art can be applied to implement the corresponding demultiplexing, and the present invention is not limited to one such configuration.

[0055] In this regard, demultiplexing can be applied using a label-based resolution system. Specifically, individual network state components can be attached with corresponding labels that uniquely identify the label and / or associated test cases, and resolution is then applied by separating each network state among those labeled with the same test case identifier. Alternatively, other types of identifiers, such as packet header information, message identifiers, or band demultiplexing, can also be applied to form an efficient method for individually separating and identifying each network state among those present in all test cases.

[0056] Furthermore, regarding the "similarity quantity" used to identify and compare a given network state of a test case with the network state of a test case currently analyzed by the trained network state information model, at least the same definition and determination strategy as indicated for identifying test case-specific network states can be implemented, primarily to ensure that individual network states appearing in the system are identified in a consistent manner. Therefore, in the first example, variations of different similarity identification processes can be implemented, such as cosine similarity processing, Euclidean distance, or Manhattan distance calculation, clustering-based methods, or heatmap comparisons, to calculate a corresponding value of similarity between the network state of the test case and the network state of the current test case, where the corresponding value of similarity can then be compared with a predefined threshold to approve or ignore a given network state. Alternatively, in the second example, direct equivalence between two corresponding network states, or equivalence of at least one or more state characteristics (e.g., predefined occurrence time points, network state locations, etc.), can also be enforced for network state similarity identification; therefore, the corresponding identification mechanism should not be limited to a single process.

[0057] Therefore, based on the above mechanism, a resource-efficient and particularly accurate method can be provided to provide human-interpretable semantic information about one or more network states identified in the target communication network or NDT when performing SWNC under unknown conditions.

[0058] In another example, the actual implementation of the corresponding network state information model can typically be further processed in a central control element integrated into the corresponding communication network system. As an example, both model training and application can be implemented in one or more centralized data centers or cloud platforms included in the network system, which typically have sufficient computing resources to perform complex model training. Furthermore, in other configurations or in combination with the aforementioned configurations, the corresponding network model can also be integrated into a network management system (NMS), configured to collect data from various network devices (e.g., routers, switches), thus forming an efficient data source for model training. Additionally, other configurations can choose the network function virtualization (NFV) infrastructure / platform or core network of the corresponding network system as the processing area, primarily because each central system location allows for flexible and resource-efficient model implementation.

[0059] In contrast, in another example configuration of the invention presented herein, it is also possible to locally implement and apply at least one trained network state information model on a given edge device of the corresponding communication network, at least on a user equipment (UE) connected to the communication network via one or more connection types, such as physical or wireless connections, local area networks (LAN / Wi-Fi), routers or gateways, or various virtual network functions (VNFs). Specifically, SWNCs performed in the first and second target communication networks can also be used as such edge devices / UEs.

[0060] Therefore, in another embodiment of the present invention, the SWNC can also correspond to a terminal device of the first target communication network, wherein the terminal device can be at least a user equipment (UE), and wherein The step of applying at least one network model, including a trained network state information model, to the first target communication network can be performed locally at the SWNC.

[0061] The advantages gained through this configuration can be specifically seen in the following facts: the local application of a trained network state information model can lead to enhanced resource and time management within the corresponding network system. Because by locally applying the trained network state information model (including processing information locally available on the corresponding UE) on the SWNC (e.g., the corresponding UE), the UE's computational resources can be used, rather than those provided by the network (core) system, resulting in an enhanced resource allocation mechanism. Furthermore, since the UE's local information can also include device-specific radio interface data (e.g., signal quality), telecommunications service quality (e.g., voice call setup delay, voice call quality, data rate), mobility events (handover and associated success / failure and latency metrics), traffic measurements, etc., which can at least partially define private data that is typically not sent to the corresponding network without the user's consent, additional authorization queries can be omitted, thus leading to a more efficient application process.

[0062] Furthermore, in another example configuration of the invention proposed herein, the output generated by the local application and the trained network state information model, namely the generated human-interpretable semantic information about one or more identified network states, can then be fed back to the SWNC's communication network system, or more specifically, to another given network component found in the corresponding network system, in order to enable subsequent additional actions based on the generated information.

[0063] Therefore, in another embodiment of the invention, the method proposed herein may also include at least the following additional steps: Send at least one network model, including a trained network state information model, to the SWNC; and Based on local data stored in the SWNC, at least one network model including a trained network state information model is applied at the SWNC. The local data includes at least one of the following information: the radio interface of the SWNC, the quality of telecommunications service, and the traffic measurement of the SWNC. The step of reporting human-interpretable semantic information may further include at least: reporting human-interpretable semantic information to the SWNC, or conveying human-interpretable semantic information to another network component of the first target communication network.

[0064] For this reason, another advantage of the above configuration is that it can generate an efficient bidirectional execution pipeline, in which the network state information model can be initially and locally applied to the data stored in the SWNC with the help of the corresponding SWNC, and then the output information thus generated (i.e. human-interpretable semantic information) can be fed back to the network system for further use.

[0065] In this regard, examples of such additional use can be achieved, for instance, by applying subsequent actions based on the generated and / or annotated human-interpretable semantic information, so that the corresponding method of the invention proposed herein can also include the following steps: Perform an action directed to the SWNC or another network component of the first target communication network by at least one human-interpretable semantic information.

[0066] In this paper, additional actions are not necessarily limited to any specific processing implemented within the network system, but can generally refer to any given responsive action that can be taken using the identified and / or annotated human-interpretable semantic information. Examples of such additional actions include: applying subsequent recovery procedures (e.g., in the case of fault identification), modifying network state, alerting and / or notification procedures (capable of notifying a given user / administrator to perform another (selected) action), additional root cause analysis (RCA) steps, traffic rerouting, or subsequent performance monitoring, each depending on the content of the initially generated human-interpretable semantic information. Furthermore, given additional actions can also be executed automatically after the relevant information is transmitted to other corresponding network components, or after authorization by the corresponding administrator, resulting in a flexible input mechanism adaptable to any existing system configuration.

[0067] In another example, it is also possible to implement corresponding retransmissions and subsequent additional actions on the content of the generated human-interpretable semantic information in a feedback loop-based manner in order to converge to a predetermined solution to the problem that may be involved in the generated human-interpretable semantic information (e.g., in the case where the human-interpretable semantic information involves a "storage capacity loss" problem in one or more existing network storage elements, the feedback loop-based additional actions can predictably release additional storage space continuously, in which new storage space is provided step by step until the corresponding problem appears to be solved).

[0068] Therefore, in another embodiment of the present invention, the method proposed herein may further include the following features: at least the following steps are performed in a closed loop until a predefined parameter of the first target communication network changes or a predefined number of iterations of the closed loop is reached: applying at least one network model including a trained network state information model to the first target communication network; having the trained network state information model report at least one human-interpretable semantic information associated with one or more network states; and performing actions indicated by at least one human-interpretable semantic information to the SWNC or other network components of the first target communication network.

[0069] Specifically, this enables efficient automatic recovery processes within a given network system, primarily because even complex problems or complex root cause identification mechanisms can be resolved / applied with high accuracy through appropriate feedback loop processes, while minimizing the required computational resources.

[0070] Therefore, given that the above embodiments and configurations of the invention proposed herein for providing human-interpretable semantic information for network states generated during SWNC execution define an effective way to not only predict network states occurring during SWNC execution, but also ensure that these predictions are interpretable in a way that is understandable and actionable by human domain experts, the invention proposed herein enables operators to interpret predicted network behavior and performance in the context of the actual network, thereby facilitating better decision-making and proactive management of network resources.

[0071] Furthermore, additional variations of the present invention may also correspond to a single use case as follows: a corresponding trained network state information model, a network node configured to process any of the above method steps (including the foregoing use cases) describing a given network element in a given (first) communication network; and at least one computer-readable medium storing instructions that can be executed by at least one processing unit of a machine, thereby causing the machine to perform any of the above method steps.

[0072] Therefore, as another example of the present invention, another method for providing human-interpretable semantic information of the network state of a software-defined networking component (SWNC) according to the present invention, wherein the human-interpretable semantic information is generated in a first target communication network, may include the following steps: At least one network model, including a trained network state information model, is applied to a first target communication network to generate at least one human-interpretable semantic information about one or more network states of a SWNC determined during execution in the first target communication network, wherein the trained network state information model is trained in such a way that, based on the determined network states, the trained network state information model is configured to output human-interpretable semantic information identifying the associated network states in a human-interpretable manner; and A trained network state information model reports at least one human-interpretable semantic information associated with one or more network states to, for example, network components or operators.

[0073] Furthermore, in another embodiment of the present invention, there may also be a network node of a first target communication network, configured to provide human-interpretable semantic information about the network state of a software-defined network component (SWNC), which is generated in the first target communication network, wherein the first network node includes: At least one processor; and At least one memory storing instructions that, when executed by at least one processor, cause the network node to at least: Receive data elements and information sets, which identify the predefined test cases on which SWNC is executed in the second target communication network; Based on the received data elements and information set, a network state information model is trained to determine human-interpretable semantic information of at least one network state of the SWNC identified during execution in the first target communication network, based on the identified network state. At least one network model, including a trained network state information model, is applied to a first target communication network to generate at least one human-interpretable semantic information about one or more network states of the SWNC determined during execution in the first target communication network; and The trained network state information model outputs at least one human-interpretable semantic information associated with one or more network states to network elements or operators.

[0074] Furthermore, regarding the aforementioned network node, the corresponding SWNC can also correspond to a terminal device of the first target communication network, which is at least a user equipment (UE), and wherein the network node is configured to at least apply the trained network state information model and is further configured as follows: Send at least one network model, including a trained network state information model, to SWNC; Based on local data stored in the SWNC, at least one network model, including a trained network state information model, is applied locally at the SWNC. This local data includes at least one of information regarding the SWNC's radio interface, quality of service, and SWNC traffic measurements. The network nodes are configured to report at least human-interpretable semantic information and are also configured to: Output human-interpretable semantic information to the SWNC, or convey human-interpretable semantic information to another network component of the first target communication network, such as a network element or operator.

[0075] Finally, in another example, the invention may also include a computer-readable medium having instructions stored thereon, which, when executed by at least one processing unit of the machine, cause the machine to perform the method according to any of the embodiments and configurations described above.

[0076] Furthermore, while this document will specifically refer to the above applications to describe some exemplary embodiments, it should be understood that this disclosure is not limited to such areas of use and is equally applicable to a broader context.

[0077] It is worth noting that the methods according to this disclosure relate to methods of operating the apparatus according to the above-described exemplary embodiments and variations thereof, and the corresponding statements made regarding the apparatus also apply to the corresponding methods, and vice versa; therefore, for the sake of brevity, similar descriptions may be omitted. Furthermore, even if not explicitly disclosed, the above aspects can be combined in various ways. Those skilled in the art will understand that combinations of these aspects and features / steps are possible unless they create a contradiction that explicitly excludes them.

[0078] Implementations of the disclosed device may also include, but are not limited to, the use of one or more processors, one or more application-specific integrated circuits (ASICs) and / or one or more field-programmable gate arrays (FPGAs). Implementations of the device may also include the use of other conventional and / or custom hardware, such as software-programmable processors, such as graphics processing unit (GPU) processors.

[0079] Furthermore, other embodiments of this disclosure will become apparent during the following discussion and with reference to the accompanying drawings. Attached Figure Description

[0080] Embodiments of this disclosure will now be described by way of example only, with reference to the accompanying drawings, in which:

[0081] Figure 1 An exemplary training process of the network state information model according to a first embodiment of the present invention is illustrated schematically;

[0082] Figure 2 An exemplary application process of the network state information model according to a first embodiment of the present invention is illustrated, which is used to identify and generate human-interpretable semantic information about one or more network states of SWNC that are built as annotations of the corresponding network states.

[0083] Figure 3 An overview of an exemplary network state detection process, illustrated in the present invention, for identifying specific network states of test cases for model training, is provided.

[0084] Figure 4 schematically illustrated Figure 3 The workflow of the network status detection process presented in the document;

[0085] Figure 5 The diagram illustrates an overview of the demultiplexing process used in this invention, which defines the content of human-interpretable semantic information formed as state annotations as a combination of network states discovered during the execution of several test cases.

[0086] Figure 6An exemplary training process of the network state information model according to a second embodiment of the present invention is illustrated schematically;

[0087] Figure 7 An exemplary application process of the network state information model according to a second embodiment of the present invention is illustrated schematically. The network state information model is used to identify and generate human-interpretable semantic information about one or more network states of a network state (SWNC), which is established as an annotation of the corresponding network state.

[0088] Figure 8 The illustration schematically depicts the process of locally applying a corresponding version of a trained network state information model to the edge device of the corresponding communication network or NDT (specifically, to the UE assigned to the communication network or NDT) according to the configuration of the present invention.

[0089] Figure 9 The illustration schematically depicts a system configuration for providing human-interpretable semantic information according to an example configuration based on the present invention. Detailed Implementation

[0090] Below, we will use communication network architectures (such as 5G / NR) based on 3GPP standards for communication networks as examples to describe different exemplary embodiments, but the embodiments are not limited to such architectures. It will be apparent to those skilled in the art that these embodiments can also be applied to other types of communication networks in which mobile communication principles are combined with D2D (device-to-device) or V2X (vehicle-to-everything) configurations (such as SL sidelinks), such as Wi-Fi, Global Microwave Access Interoperability (WiMAX), Bluetooth®, Personal Communication Services (PCS), ZigBee®, Wideband Code Division Multiple Access (WCDMA), systems using Ultra Wideband (UWB) technology, Mobile Ad Hoc Networks (MANETs), wired access, etc. Furthermore, while some examples of the embodiments are described in relation to mobile communication networks without loss of generality, the principles of this disclosure can be extended and applied to any other type of communication network, such as wired communication networks.

[0091] The following examples and embodiments should be understood as illustrative examples only. Although the specification may refer to "a," "an," or "some" examples or embodiments in several places, this does not necessarily mean that each such reference relates to the same(s) examples(s) or embodiments(s), or that the feature applies only to a single example or embodiment. Individual features of different embodiments may also be combined to provide other embodiments. Furthermore, terms such as "comprising" and "including" should be understood not to limit the described embodiments to consisting only of the features mentioned; such examples and embodiments may also include features, structures, units, modules, etc., not explicitly mentioned.

[0092] The basic system architecture of a (telecommunications) communication network, including a mobile communication system (some examples of which are applicable), may include the architecture of one or more communication networks, which include (multiple) radio access network subsystems and (multiple) core networks. This architecture may include one or more communication network control elements or functions, access network elements, radio access network elements, access service network gateways, or base transceivers, such as base stations (BS), access points (AP), NodeBs (NBs), eNBs or gNBs, distributed units (DUs), or centralized / central units (CUs), which control their respective (multiple) coverage areas or cells, and the following devices may communicate with the aforementioned devices via one or more communication beams used to transmit various types of data in multiple access domains through one or more channels: one or more communication stations (such as communication elements or functions, such as user equipment or terminal equipment, such as user equipment (UE), or other devices with similar functions, such as modem chipsets, chips, modules, etc.), which may be part of a site, element, function, or application capable of implementing communication, such as a UE, an element or function usable in a machine-to-machine communication architecture, or attached as a separate element to an element, function, or application capable of implementing communication, etc. In addition, it may include core network elements or network functions, such as gateway network elements / functions, mobility management entities, mobile switching centers, servers, databases, etc.

[0093] The following description provides further details on alternatives, modifications, and differences: gNB includes, for example, nodes that provide NR user plane and control plane protocol termination to the UE and connect to 5GC via the NG interface, for example, according to Section 3.2 of 3GPP TS 38.300 V16.6.0 (2021-06) incorporated by reference.

[0094] The gNB Central Unit (gNB-CU) includes, for example, a logical node that hosts, for example, the gNB's RRC, SDAP, and PDCP protocols, or the en-gNB's RRC and PDCP protocols, and controls the operation of one or more gNB-DUs. The gNB-CU terminates the F1 interface connected to the gNB-DU.

[0095] A gNB Distributed Unit (gNB-DU) may include, for example, a logical node that hosts the RLC, MAC, and PHY layers of, for example, a gNB or en-gNB, and its operation is partially controlled by the gNB-CU. A gNB-DU supports one or more cells. A cell is supported by only one gNB-DU. The gNB-DU terminates the F1 interface connected to the gNB-CU.

[0096] The gNB-CU control plane (gNB-CU-CP) includes, for example, a logical node that hosts the control plane portions of the gNB-CU's RRC and PDCP protocols, for example, for en-gNB or gNB. The gNB-CU-CP terminates the E1 interface connected to the gNB-CU-UP and the F1-C interface connected to the gNB-DU.

[0097] The gNB-CU user plane (gNB-CU-UP) includes, for example, a logical node that hosts, for example, the user plane portion of the PDCP protocol for the gNB-CU for the en-gNB, and the user plane portion of the PDCP protocol for the gNB-CU for the gNB and the SDAP protocol. The gNB-CU-UP terminates the E1 interface connected to the gNB-CU-CP and the F1-U interface connected to the gNB-DU, for example, according to Section 3.1 of 3GPP TS 38.401 V16.6.0 (2021-07) incorporated by reference.

[0098] Different functional divisions between central and distributed units are possible, for example, referred to as options: Option 1 (similar to the split of 1A): - The functional breakdown in this option is similar to the 1A architecture in a DC. The RRC is located in the central unit. PDCP, RLC, MAC, physical layer, and RF are located in the distributed units. Option 2 (similar to the splitting of 3C products): - The functional breakdown in this option is similar to the 3C architecture in a DC. RRC and PDCP are located in the central unit. RLC, MAC, physical layer, and RF are located in the distributed units. Option 3 (Split within RLC): - Low RLC (part of the RLC functionality), MAC, physical layer, and RF are located in the distributed unit. PDCP and high RLC (another part of the RLC functionality) are located in the central unit. Option 4 (RLC-MAC split): - MAC, physical layer, and RF are located in the distributed unit. PDCP and RLC are located in the central unit. Alternatively, for example, according to Section 11 of 3GPP TR 38.801 V14.0.0 (2017-03) incorporated by reference.

[0099] gNB supports different protocol layers, such as Layer 1 (L1) – the physical layer.

[0100] NR's Layer 2 (L2) is split into the following sublayers: Media Access Control (MAC), Radio Link Control (RLC), Packet Data Convergence Protocol (PDCP), and Service Data Adaptation Protocol (SDAP), where: - The physical layer provides a transmission channel to the MAC sublayer; - The MAC sublayer provides logical channels to the RLC sublayer; - The RLC sublayer provides RLC channels to the PDCP sublayer; - The PDCP sublayer provides radio bearers to the SDAP sublayer; - The SDAP sublayer provides QoS flows to 5GC; - Comp. refers to header compression, while Segm. refers to segmentation; - The control channels include (BCCH, PCCH).

[0101] Layer 3 (L3) includes, for example, Radio Resource Control (RRC), as per Section 6 of 3GPP TS38.300 V16.6.0 (2021-06) incorporated by reference.

[0102] RAN (Radio Access Network) nodes or network nodes, such as gNBs, base stations, gNB CUs or gNB DUs or portions thereof, may be implemented using, for example, means having at least one processor and / or at least one memory (including computer-readable instructions (computer programs)) configured to support and / or supply and / or process functions and / or features associated with the CUs and / or DUs, and / or at least one protocol (sub) layer of the RAN, such as layer 2 and / or layer 3.

[0103] The gNB CU and gNB DU portions can, for example, be co-located or physically separated. The gNB DU can even be further split, for example, into two parts, one including processing equipment and the other including antennas. The Central Unit (CU) can also be referred to as BBU / REC / RCC / C-RAN / V-RAN, O-RAN, or a portion thereof. The Distributed Unit (DU) can also be referred to as RRH / RRU / RE / RU, or a portion thereof. Thereafter, in various exemplary embodiments of this disclosure, the CU-CP (or more generally referred to as CU) can also be referred to as a (first) network node supporting at least one of the Central Unit Control Plane Functionality or Layer 3 Protocol of the Radio Access Network; similarly, the DU can be referred to as a (second) network node supporting Layer 2 Protocol or Distributed Unit Functionality of the Radio Access Network.

[0104] gNB-DU supports one or more cells and can therefore be used as a serving cell, for example, for user equipment (UE).

[0105] User equipment (UE) may include wireless or mobile devices, devices with a radio interface for interacting with the RAN, smartphones, vehicle-mounted devices, IoT devices, M2M devices, etc. Such UEs or devices may include: at least one processor; and at least one memory including computer program code; wherein the at least one memory and the computer program code are configured, together with the at least one processor, to cause the device to perform at least certain operations, such as an RRC connection to the RAN. For example, the UE is configured to generate a message (e.g., including a cell ID) to be transmitted to the RAN via radio (e.g., to reach and communicate with the serving cell). The UE may generate, transmit, and receive RRC messages containing one or more RRC PDUs (Packet Data Units).

[0106] User equipment (UE) can have different states (e.g., according to sections 4.2.1 and 4.4 of 3GPP TS 38.331V16.5.0 (2021-06) which are incorporated by reference).

[0107] When an RRC connection has been established, the UE is in, for example, the RRC_CONNECTED state or the RRC_INACTIVE state.

[0108] In the RRC_CONNECTED state, the UE can: - Store the AS context; - Sending / receiving unicast data to / from the UE; - Monitor the control channel associated with the shared data channel to determine whether data has been scheduled to that data channel; - Provides channel quality and feedback information; - Perform neighboring cell measurements and measurement reports.

[0109] The RRC protocol includes, for example, the following main functions: - RRC connection control; - Measurement configuration and reporting; - Establishing / modifying / releasing measurement configurations (e.g., intra-frequency, inter-frequency, and inter-RAT measurements). - Setting and releasing the measurement gap; - Measurement report.

[0110] The general functions and interconnections of the described elements and functions (which also depend on the actual network type) are known to those skilled in the art and are described in the corresponding specifications; therefore, detailed descriptions may be omitted herein for the sake of brevity. However, it should be noted that, in addition to those described in detail below, several additional network elements and signaling links may be employed for communication to or from elements, functions, or applications, such as communication endpoints, communication network control elements (such as servers, gateways, radio network controllers), and other elements in the same or other communication networks.

[0111] The communication network architecture considered in the examples of the embodiments may also be able to communicate with other networks, such as the public switched telephone network or the Internet. The communication network may also be able to support the use of cloud services for virtual network elements or their functions. It should be noted that the virtual network portion of a telecommunications network may also be provided by non-cloud resources, such as internal networks, etc. It should be understood that network elements and / or corresponding functionalities of access systems, core networks, etc., can be implemented using any nodes, hosts, servers, access nodes, or entities suitable for such purposes. Typically, network functions can be implemented as network elements on dedicated hardware; as software instances running on dedicated hardware; or as virtualized functions instantiated on a suitable platform (e.g., cloud infrastructure).

[0112] Furthermore, network elements (such as communication elements, like UEs, terminal devices, control elements or functions, such as access network elements, like BSs, gNBs, radio network controllers, core network control elements or functions, like gateway elements, or other network elements or functions as described herein) and any other elements, functions, or applications can be implemented via software (e.g., via computer program products for computers) and / or via hardware. To perform their respective processing, the corresponding devices, nodes, functions, or network elements used may include various parts, modules, units, components, etc. (not shown) required for control, processing, and / or communication / signaling functionality. Such components, modules, units, and parts may include, for example, one or more processors or processor units (including one or more processing sections for executing instructions and / or programs and / or processing data), storage units or memory units or components for storing instructions, programs, and / or data (e.g., ROM, RAM, EEPROM, etc.), working areas for serving as processors or processing sections, etc.; input or interface components for inputting data and instructions via software (e.g., floppy disks, CD-ROMs, EEPROMs, etc.); user interfaces for providing users with the possibility of monitoring and manipulation (e.g., screens, keyboards, etc.); and other interfaces or components for establishing links and / or connections under the control of processor units or processing sections (e.g., wired and radio interface components, radio interface components including, for example, antenna units, components for forming radio communication sections, etc.), etc., wherein the corresponding components forming the interfaces (such as radio communication sections) may also be located at remote sites (e.g., radio heads or radio stations, etc.). It should be noted that, in this specification, a processing section should not only be considered as a physical part representing one or more processors, but also as a logical division of the processing tasks performed by one or more processors. It should be understood that, according to some examples, a so-called "liquid" or flexible network concept can be adopted, in which the operation and functionality of network elements, network functions, or other entities in the network can be flexibly performed in different entities or functions (such as nodes, hosts, or servers). In other words, the "division of labor" among the network elements, functions, or entities involved may vary depending on the specific circumstances.

[0113] Now, before going into detail about the exemplary embodiments of this disclosure, it is still worthwhile to briefly review some exemplary general aspects of machine learning that may be considered helpful in understanding this disclosure, particularly the techniques / processes related to AI (artificial intelligence) and / or ML (machine learning).

[0114] ML frameworks and technologies are being deployed increasingly extensively in 5G networks (including RAN, core network, and management systems / functionalities), and their application scale is expected to expand further with the development of technologies such as advanced 5G and 6G networks. Furthermore, the telecommunications industry is widely focusing on the research and standardization of ML frameworks and functionalities in various standards forums, and is committed to leveraging ML capabilities to improve use cases within individual standards organizations and their working groups. For example, 3GPP is actively involved in researching ML functions and their different applicability in various working groups. For instance: - SA (Service and Systems Aspect) 5 handles the WI (Work Item) aspects of PM (Performance Management) and KPI (Key Performance Indicators) enhancements for 5G and advanced systems, typically focusing on specifying the PMs and KPIs required to measure the performance of different network entities. Research on SA5 typically focuses on the management aspects of training, retraining, testing, and reasoning. - 3GPP SA2 typically deals with some advanced topics about ML, such as enhancing the sharing of trained ML models and supporting FL (Federated Learning) in the 5C core.

[0115] Furthermore, the RAN3 research project on enhancing data collection for NR and EN-DC describes a functional framework for RAN intelligence. For brevity, a detailed description of this framework is omitted as it is considered easily understood by those skilled in the art. In summary, the framework typically proposes two options for ML model training and deployment. The first option is to train the ML model at the RAN and deploy the same model within the RAN itself for inference. The second option is to train the ML model at the OAM and deploy the model at the RAN for inference. In either of these options, inference occurs at the RAN. SA5 subsequently conducted research to understand the impact of the aforementioned RAN3 research on SA5.

[0116] Network entities in the RAN and core are also adopting AI / ML frameworks in the standard. Generally, each network entity can train, retrain, and / or deploy multiple ML models for inference. It is also being considered to understand and improve the efficiency of AI / ML usage across different AI / ML-enabled network entities by providing appropriate auxiliary information to analyze the feasibility of the framework.

[0117] Looking ahead to advanced 5G and 6G, it is foreseeable that many RAN, core, and management use cases can be driven by AI / ML-based analytics. Therefore, a large number of AI / ML models can be trained and / or deployed for inference directly within network entities such as UEs, base stations (e.g., gNodeBs), core network functions, and / or management functions.

[0118] As explained above, this disclosure generally aims to provide an accurate and efficient method for predicting not only network states occurring during the execution of a given NDT, but also, in the best case, network states occurring in the corresponding communication network simulated by the NDT, and ensuring that these predictions can be interpreted in a manner understandable and operable by human domain experts. To this end, this disclosure specifies the generation of human-interpretable semantic information related to one or more network states identified by a given SWNC or corresponding NDT for a communication network, based on a trained network state information model, as previously mentioned.

[0119] In this regard, Figure 1 The first exemplary process flow of the training process of the corresponding network state information model according to the first exemplary embodiment of the present invention is shown.

[0120] In this document, as a preliminary introduction, "human-interpretable semantic information" should be understood as human-interpretable, that is, readable information related to one or more network states identified during SWNC execution, and may include at least, particularly, information about the causes of the relevant network states (e.g., the corresponding network conditions leading to the corresponding network state), different network state characteristics (e.g., network components affected by the corresponding network state, the time of occurrence, duration, etc.), or solutions for predefined network conditions (e.g., appropriate solutions adapted to a given identified network state). This information can help the relevant network operators and network components understand the current network conditions present during SWNC execution, thereby enabling appropriate and targeted countermeasures to be taken in the event of potential system failures (e.g., in the event of unexpected system behavior / output).

[0121] In addition, in such Figure 1 In the embodiment shown, the corresponding human-interpretable semantic information is established and generated as annotations of the associated network states. That is, information annotated to one or more network states pointed to by the human-interpretable semantic information, thereby making the above network state information model an automatic self-annotating network state information model.

[0122] In addition, see now Figure 1 The content shown, Figure 1 The indicated model training typically involves two independent execution segments, ultimately resulting in a fully trained network state information model.

[0123] In this regard, the first of the two execution segments defines the test case execution segment, which is configured to provide the necessary training data to the network state information model, specifically to train the corresponding (untrained) network state information model to associate one or more network states (generated by the execution of a given SWNC) among the corresponding identified network states with a given human-interpretable network state description.

[0124] Therefore, in such Figure 1 In the embodiment shown, the CI / CD pipeline execution process is performed, wherein test cases are implemented (step SA01). Each test case has a test case description TCD1…TCDN, which includes test conditions (e.g., the scenario and context in which the SWNC is tested), expected SWNC behavior (e.g., the actions that the SWNC will produce), and expected output (e.g., the selected numerical / log outputs that the SWNC should produce during the execution of the test case). Finally, by executing the corresponding SWNC under the known and selected execution conditions, corresponding observable SWNC outputs SWOT1…SWOTN are generated for each test case, thereby defining the actual output of the corresponding SWNC for the corresponding test case conditions.

[0125] Subsequently, during the subsequent training process (step SA02), the untrained network state information model can then receive test case descriptions TCD1…TCDN (i.e., a set of information about each test case, including at least the test case conditions, expected SWNC behavior, and expected output) and corresponding observable SWNC outputs SWOT1…SWOTN as training samples. This allows it to associate the corresponding network states known from the test cases with the information obtained from the test case descriptions. Simultaneously, it also learns to identify / determine the occurrence of one or more network states based on the observed SWNC outputs as input to the initial network state information model. As a result, the first state recognition model segment (…) of the network state information model… Figure 1 The first model function (represented by the top bar shown in the dashed box) learns to represent the (observable) output of a given SWNC as a combination of several abstract network states associated with that output (e.g., "learned states LS1…LSN" learned for the corresponding observable SWNC outputs SWOT1…SWOTN used for each test case). Furthermore, as a model segment generated from the second state information ( Figure 1The second model function (shown in the bottom bar with the dashed box in the middle) is the network state information model. The trained network state information model also learns to associate different combinations of human-interpretable semantic information (learned annotations AN1…ANN for each test case) with the corresponding abstract network states. This is mainly achieved by associating each network state known during the execution of each test case with the corresponding test case conditions and / or descriptions.

[0126] Figure 2 An exemplary illustration of the continuous application of a trained network state information model is further shown, which is used to identify / generate appropriate human-interpretable semantic information aspects related to the network state of a SWNC performed under primarily unknown conditions.

[0127] Specifically, such as Figure 2 As can be seen from the configuration shown, the application process of the trained network state information model can also be divided into two different processing segments, each of which can be processed individually and / or sequentially to generate the sought human-interpretable semantic information.

[0128] In this paper, the first processing segment corresponds to the actual input process and is used to introduce the observed output of the SWNC generated during unknown execution conditions, which in Figure 2 This is summarized under the heading "Operating Environment".

[0129] In this section, SWNC is executed in a real-world network environment under unknown conditions (step SBO1). Depending on the number of times, type, or conditions under which SWNC is executed, one or more observable SWNC (SW) outputs SWO1…SWON are generated. These outputs can be used to generate human-interpretable semantic information.

[0130] In this regard, it is important to note that the "real-world network environment" in which SWNC is executed can correspond to the same communication network used to train the corresponding model. In another embodiment, it is equally possible to apply the trained model to other communication networks, particularly to NDTs simulating the aforementioned communication networks used for training. Therefore, the proposed model application process can be specifically designed for network state identification and interpretation of SWNCs used in NDTs, while the corresponding identification model is pre-trained with real network data (test cases executed in actual communication networks). This results in a more accurate identification / interpretation of the current network state using the trained network state information model, primarily because the model's output is based on real-world conditions rather than those artificially simulated by the NDT.

[0131] Furthermore, as a next step, each of the generated observable outputs SWO1…SWON is fed into the trained network state information model in order to identify the sought human-interpretable semantic information (step SBO2).

[0132] In this paper, compared to CI / CD-based learning scenarios, there is no direct information about the expected SWNC behavior and execution conditions in this configuration. However, since the same SWNC is typically used for test case execution, even in actual execution scenarios, the SWNC should produce the same type of observable output as it does within a controlled test environment. Therefore, based on the fact that the corresponding network state information model has been trained through previous test case executions to associate a given number of network states with their corresponding observable outputs, and subsequently associates given human-interpretable semantic information with several identified network states, the trained network state information model can thus “identify” which previous test case scenario best matches the current operating conditions (based on the corresponding observed SWNC outputs SWO1…SWON), and generate human-interpretable semantic information (formed as network state annotations) by invoking the matching test case description (i.e., the content of the data elements and information set used to define the execution of the corresponding test case) and possibly modifying / expanding / adjusting it to fit the currently observed conditions.

[0133] Therefore, based on the aforementioned characteristics of the trained network state information model, the trained network state information model takes into account the corresponding observable outputs SWO1…SWON, and then identifies the corresponding (abstract) network state PS1…PSN and associated human-interpretable semantic information (i.e., the predicted annotations PA1…PAN) for each corresponding SWNC execution process.

[0134] Here, in Figure 2 The second processing segment illustrated in the figure shows the corresponding generated network states PS1…PSN and the generated human-interpretable semantic information, which is summarized under the heading "Model Prediction". Furthermore, it should be noted that, as mentioned earlier, each corresponding generated / identified network state and human-interpretable semantic information can be generated by different processing segments of the network state information model, namely the first state identification model segment of the network state information model (…). Figure 2 The top bar shown in the dashed box indicates that this model segment can identify / predict the corresponding abstract network states PS1…PSN for a given execution process based on the corresponding observable outputs SWO1…SWON; and the second state information generation model segment ( Figure 2The bottom bar (shown in the dashed box) is a model segment that associates different human-interpretable semantic information (predicted annotations PA1…PAN) with the identified network states generated by the state recognition model segment. Therefore, the general process generated by the trained network state information model may initially aim to use the state recognition model segment to generate the corresponding abstract network states PS1…PSN from the input observable SWNC outputs SWO1…SWON; simultaneously, using the state information generation model segment, based on the generated network states PS1…PSN as input, it determines the corresponding human-interpretable semantic information, which is then formed as the network state annotations PA1…PAN.

[0135] Figure 3 Further exemplary illustrations of network states discovered in multiple different test cases TC1, TC2, and TCN are shown, which are used to illustrate the additional network state selection process provided in this invention.

[0136] Specifically, in the actual training process that associates specific human-interpretable semantic information with the network state of the corresponding test case, the network state information model can also select the network state of a given test case before the actual training phase. These network states are specific to the corresponding test case only, and the training is then performed using only the selected network states.

[0137] In this regard, the reason for this additional process can be seen from the fact that during test case execution, many different network states typically arise due to initialization, general configuration, resource allocation, bootstrapping, etc. The purpose of these operations is merely to prepare the network (i.e., to bring it to the conditions) where the key parts of the test cases can be executed. Furthermore, the final network state may also be generated simply by resource release, configuration deactivation, etc., which are also completely unrelated to the actual characteristics of the test cases. Therefore, since only a portion of all network states identified for a test case are actually identified as specific conditions defined by the corresponding test case, selecting only those network states specific to a given test case for subsequent model training may lead to improved model output. This is mainly because a model generated and trained in this way can correctly associate the corresponding human-interpretable semantic information with the correctly specified network states.

[0138] As described in this article, when dealing with many test cases (such as...) Figure 1 As shown in the exemplary illustration, the process of selecting the corresponding network state specific only to a given test case can be implemented by the network state information model during its actual training or preferably before its actual training, and the network state sequences observed during each test case can be directly compared with each other.

[0139] In this regard, Figure 3 The illustration shows an exemplary composition of network states for different test cases executed before model training. Network states labeled with the letter 'ad' are defined as common (i.e., non-specific) network states for all test cases and belong to the initial (i.e., configuration) process step. Similarly, network states with the letters 'xz' are also common to all test cases and belong to a given final step of the test case. Network states labeled with numbers can be defined as network states specific to the corresponding test case.

[0140] Figure 4 Further illustrations show examples of the network selection process performed during and / or before actual model training.

[0141] Here, in the first step (step SCO1), in addition to the observable SWNC output SWOT1…SWOTN obtained after the execution of the corresponding test case, data elements and information sets related to the given test case are also received (that is, Figure 1 The test case description shown in the figure is TCD1…TCDN).

[0142] In the next step (Step SCO2), based on the received information and output, corresponding network states are created for each test case, thereby generating a different sequence of network states for each test case, such as... Figure 3 exemplified in .

[0143] Subsequently, for each test case's network state sequence (step SCO3), the corresponding network state sequence is compared with the network state sequences of other test cases to identify network states that appear more than once (i.e., those that may form non-specific network states for the test cases). Typical state identification mechanisms known in the art can be used for this purpose. Furthermore, in other embodiments, it is also possible to implement various similarity identification processes, such as Euclidean distance or Manhattan distance calculation, clustering-based methods, feature matching, or heatmap comparison, each outputting a numerical value representing the similarity of a given network state to network states generated and / or identified in other test cases. This method may be particularly effective when the current network states include similar but not identical characteristics (e.g., due to subtle inconsistencies within the corresponding test case configurations), enabling efficient identification of non-specific network states even in these cases. Furthermore, to define a specific similarity level at which network states are considered the same, a predefined threshold can be introduced to evaluate the similarity values ​​generated by a single similarity identification process.

[0144] In step SCO4, non-test case-specific network states can be removed from the sequence of network states representing a given test case. These states may specifically correspond to the network states at the beginning or end of the sequence (due to the configuration features described above).

[0145] Finally, in step SCO5, a given network state information model can be trained solely based on the network state specific to the selected test cases and the test case description provided as additional input for each test case (step SCO6). In this regard, the final training step described in step SCO5 can be equivalent to... Figure 1 The difference in the training process mentioned above is that the network state information model only receives the selected network states (rather than all network states) as training input parameters.

[0146] Figure 5 Further illustration shows another example configuration to be added to the application process of this invention, wherein the corresponding human-interpretable semantic information is generated by: identifying a given sequence of network states executed by (unknown) SWNC as a combination of network states found in multiple test cases, and subsequently generating human-interpretable semantic information based on the test case descriptions of these combined test cases (i.e., defining the set of data elements and the content of information for the corresponding test cases used for model training). Figure 5 The prediction annotation PA is shown in the figure.

[0147] In this paper, the reason for this additional configuration can be seen from the fact that when a trained network state information model is applied to a real-world network deployment (such as...), Figure 2 As shown in the diagram, the challenge this model faces is that the observed network state is caused by multiple events occurring simultaneously in the network. While executing test cases results in observing network states isolated from the execution of other test cases, in real-world networks, multiple events are occurring concurrently. In such cases, the effect is as if multiple test cases are executed in parallel, thus providing a mixed network state. This complicates the network states observed in real-world networks because they are not specific to any particular test case.

[0148] To address this complexity, as another step to improve the efficiency of the corresponding network state information model in real-world networks, the model can also perform demultiplexing on the corresponding network states found for a given real-world use case. The main purpose is to separate the corresponding network states and represent them as a combination of network states specific to one or more existing test cases.

[0149] In this paper, the actual concept of this demultiplexing process is as follows: Figure 5 As shown in the diagram.

[0150] First, receive the observations from the real network formed by performing SWNC under unknown conditions (that is, the observable SWNC output SWO), which is equivalent to... Figure 2 The steps corresponding to SWO1…SWON are shown in the middle.

[0151] Subsequently, based on the observed output SWO, the obtained network state sequence is determined primarily using a trained network state information model, which is equivalent to... Figure 2 This corresponds to the steps in SBO2. In this paper, the resulting network state sequence is a mixture of network states caused by simultaneous or overlapping events in the network, which can be regarded as multiple test cases being executed in parallel / overlapping.

[0152] Therefore, since high-coverage test cases are typically used in this invention, it is foreseeable that any network state can be represented by a combination of network states present in one or more test cases executed against the corresponding SWNC.

[0153] Based on this, the trained network state information model can be configured in the next step to initially detect which network states are specific to a given test case, rather than common states (see about...). Figure 3 and Figure 4 (The explanation), and then, with the help of the demultiplexing process, identify combinations of specific network states included in one or more test cases, which can form specific network states defined by the desired test case.

[0154] In this regard, Figure 5 An exemplary illustration shows the sequence of network states determined by performing SWNC under unknown conditions. Figure 5 (The top network state sequence is shown in the figure). In this paper, seven non-specific network states defined by states ad and xz are found at the beginning and end of the sequence, resulting in the effect that the remaining network states 1, 3, 5, 4, 7, and 2 define the specific network states for the corresponding execution test cases. Furthermore, in order to identify the corresponding specific network states as combinations of network states for one or more test cases, the trained network state information model can further perform the above-mentioned specific network state identification process for each test case previously executed for model training, and then demultiplex each network state of each test case for combinatorial analysis.

[0155] In this article, Figure 5 In the example illustrated, for instance, it is found through a trained network state information model that a specific network state sequence 1, 3, 5, 4, 7, 2 for a test case can be formed by combining the network state sequences of test case 1TC1 and test case 2, because the two form the same network state sequence found for the corresponding test case in a combined manner.

[0156] Therefore, in the subsequent steps after identifying the above combination of test cases, the trained network state information model can then generate corresponding human-interpretable semantic information (i.e., based on the combined description of test cases 1 and 2) Figure 5 The prediction shown in the annotation PA is mainly because the corresponding use case can be properly described as the execution of two test cases under the condition of simultaneous occurrence.

[0157] Furthermore, as the actual implementation method for performing the corresponding network state comparison and identification process, any identification process known to those skilled in the art can also be performed. Additionally, methods targeting… Figure 3 and Figure 4 The same similarity identification process is indicated by the mechanism to ensure consistent network state identification across the entire network system in use.

[0158] Furthermore, generating human-interpretable semantic information from descriptions / information from multiple test case descriptions can be performed by an additional large language model (LLM) by receiving cues from the original test cases, explanations about the expected downstream tasks (i.e., merging multiple descriptions into a single description), and possibly further explanations about the relationships between the identified test cases (e.g., the order in which the identified test cases occur—let the LLM interpret the temporal order or potential causal dependencies; if a test case is dominant because it is represented by multiple repetitions of the network state associated with that test case—then let the LLM focus on the description of that dominant test case, etc.).

[0159] Figure 6 and Figure 7 The second embodiment of the invention is further illustrated, wherein the corresponding functions are split into two independent models compared to the dual-function characteristics of the network state information model (function 1: identifying network state from observable output; function 2: identifying human-interpretable semantic information from the identified network state).

[0160] Specifically, such as Figure 6 and Figure 7 The content shows that the network state information model (i.e. Figure 6 and Figure 7 The “annotation model” shown under SDO3 and SEO3 in this case can be configured only to recognize human-interpretable semantic information formed as network state annotations based on the identified network states, while the corresponding network state recognition is handled by another (trained) network state model (i.e. Figure 6 and Figure 7 The implementation of the network state model shown in SDO2 and SEO2.

[0161] Therefore, the general process of this corresponding second embodiment for training and / or applying the corresponding network state information model can be compared with... Figure 1 and Figure 2 The process is the same as that of the first embodiment illustrated herein, except that the process initially performed by the state recognition model segment of the network state information model is now performed by an additional (external) trained network state model. In this document, the advantages gained by the configuration shown in the second embodiment are evident from the fact that the trained and existing network state model can be effectively combined with the additionally provided network state information model, thereby enabling the latter to be effectively integrated into a network system that already contains a functional network state model. Furthermore, since a trained network state model already exists for recognizing the corresponding network state based on the observable SWNC output, additional training of the corresponding network state model can be effectively omitted, thus enabling a more resource-efficient information generation process.

[0162] Therefore, it is important to note that in this case, when the training of the network state information model begins ( Figure 6 Step SDO3 in the network state model ( Figure 6 The SDO2 in the model must have been trained. It is also important to note that the network state model does not need to be trained using the same test cases as the network state information model described in this paper. Therefore, this independence between the network state model and the network state information model makes it particularly possible to reuse existing network state models and transform them into mechanisms for providing human-interpretable semantic information about one or more identified network states.

[0163] Figure 8 Another preferred application example of the proposed network state information model is further illustrated as a flowchart for the local application of the corresponding model on a network edge device (specifically referred to as a user equipment (UE) in this document).

[0164] Building upon this, the inventions disclosed herein can also be applied to the context of UE-network interoperability to improve the services provided by the network. As a general concept for this application example, leveraging the knowledge present in the test case description, the network state information model is initially developed by the network / operator in conjunction with... Figure 1 or Figure 6 The same approach illustrated is used for centralized training, where the required data (i.e., the data elements and information set needed for training) is collected centrally from the network, which is the result of serving a large number of edge devices (e.g., UEs) over a long period of time (i.e., a given SWNC used for training can be defined as one of the UEs present in the network system).

[0165] Furthermore, after successfully training the network state information model, the corresponding application example further specifies that the trained model (or a lightweight version thereof) is delivered to the specific UE for which it was trained, whereby the model then processes information locally available on the UE by specifically using the UE's computing resources.

[0166] In this paper, local data may include device-specific radio interface data (e.g., signal quality), telecommunications service quality (e.g., voice call setup delay, voice call quality, data rate), mobility events (handover and associated success / failure and latency metrics), traffic measurements, etc. Therefore, by processing this information, a trained network state information model can provide human-interpretable semantic information, while also taking into account UE-specific conditions. This results in the generated human-interpretable semantic information helping to identify potential problems or anomalies facing the corresponding device, or confirming that the device and network (such as voice calls or data connections) are operating as expected. This information can then be presented locally to the device's user / owner (i.e., the user of a given network system) or communicated to the network for processing by other network engineers, so that this additional information can be used for, for example, to troubleshoot network communication service failures or improve the network's relationship with users (e.g., proactively contacting users to inform them of potential service problems and offer compensation, while network engineers are already working to resolve the issue).

[0167] Therefore, the benefits of transmitting the model to the UE are: it can reduce the computational burden on the network to collect and process massive amounts of data, and it can access potentially private data that cannot be transmitted to the network without the explicit consent of the device owner.

[0168] In addition, the execution of the corresponding UE-side model can be triggered by an initial indication from the network side, which indicates that the network is not performing well in a specific area or for a specific device; or the device owner may have contacted the communication network operator to report a problem with a certain network service provided by the operator (e.g., voice calls cannot be made or data rates are too low), thereby attracting the operator's attention and prompting the operator to conduct further investigation using the UE-side model.

[0169] Furthermore, the explicit methodological steps used for the corresponding UE-side model execution are in Figure 8 The flowchart shown in the figure is shown separately.

[0170] In this paper, in the first step (step SFO1), a given edge device (e.g., UE) can receive a trained network state information model from the network (i.e., where the network state information model further forms the generated human-interpretable semantic information into network state annotations). Figure 8 The “trained self-annotated network state model”.

[0171] Subsequently, in the second step (step SFO2), the trained network state information model is applied locally to the UE, mainly by inputting UE-specific data into the model.

[0172] Following this, in the third step (step SFO3), based on the aforementioned model application, the corresponding trained network state information model outputs human-interpretable semantic information related to the network state generated by the UE (i.e., annotations of the network state in this example), which describes the operation of the corresponding UE device in view of predefined network characteristics (such as network performance or quality of service).

[0173] Subsequently, in the fourth step (step SFO4), the generated information / annotations will be transmitted to the network operator or other network components within the network system, primarily to perform additional actions based on the generated information.

[0174] Additionally, optionally in the fifth step (step SFO5), the generated information / annotations can also be presented to the owner of the corresponding device, for example, to inform the owner of potential errors and / or fault characteristics found in the system / device.

[0175] at last, Figure 9 An exemplary connection configuration of a device, such as a network node, is illustrated, in which the network state information model described above can be integrated and / or implemented.

[0176] In this paper, the corresponding apparatus covers the training phase ( Figure 9 The first half) and the application phase of the trained model ( Figure 9 (The lower half) Both.

[0177] During training, subsequently generated human-interpretable semantic information (i.e., [missing information]) is included as a description of test cases. Figure 9 The “annotation” shown in the figure solicits “truth” information. The corresponding network state information model (i.e., the “self-annotated network state model” SAM1) interfaces with the test case (TC) and, through the connected network NW, the model can receive data generated during the execution of the test case (TC).

[0178] Furthermore, in applying trained network state information models (i.e. Figure 9 During the “Trained Self-Annotated Network State Model (SAM2)” phase, the model continuously interfaces with the network (NW) to receive data, but no longer connects to the test cases (TC). Furthermore, the trained model can also communicate with the network to provide control information and influence the network in a closed-loop manner by responding to specific types of network states or specific types of human-interpretable semantic information detected by the trained network state information model.

[0179] In this regard, predefined network states and / or human-interpretable semantic information that can trigger network-side actions may include anomalous conditions or indications of potential anomalies, with known remedies (e.g., network states or information indicating a memory leak in network function software can be resolved by restarting the network function instance; another network function or information indicating a radio link failure at the target cell leading to handover failure may trigger a change in the handover threshold to delay the handover of the target cell until its radio conditions are sufficient to maintain the connection of the network device to be handed over). Based on this, automatically providing actions controlled by state or human-interpretable semantic information generated by the network state information model in the network reduces the time spent in suboptimal or erroneous conditions and improves the overall user experience and the quality of service provided by the network system.

[0180] Furthermore, the network state information model can provide human-interpretable semantic information / annotations to network operators and / or relevant equipment owners (NOs), thereby allowing for transparent and reliable output to each relevant network user.

[0181] It should be noted that the examples of embodiments of this disclosure are applicable to a variety of different network configurations. In other words, the examples shown in the above figures (which serve as the basis for the examples discussed above) are merely illustrative and do not limit this disclosure in any way. That is, in the corresponding operating environment, examples of embodiments of this disclosure based on defined principles can be used in combination with existing and proposed new functionalities.

[0182] It should also be noted that the disclosed example embodiments can be implemented using hardware and / or software configurations in various ways. For example, the disclosed embodiments can be implemented using dedicated hardware and / or executable software associated thereon. The components and / or elements in the accompanying drawings are merely examples and do not limit the scope or functionality of any hardware, software combined with hardware, firmware, embedded logic components, or combinations of two or more such components in implementing particular embodiments of this disclosure.

[0183] It should also be noted that the specification and drawings are merely illustrative of the principles of this disclosure. Those skilled in the art will be able to implement various solutions, although these solutions are not explicitly described or shown herein, but they embody the principles of this disclosure and are included within the spirit and scope of this disclosure. Furthermore, all examples and embodiments outlined in this disclosure are primarily intended for illustrative purposes only to aid the reader in understanding the principles of the proposed methods. Additionally, all statements regarding the principles, aspects, and embodiments of this disclosure, as well as specific examples, provided herein are intended to cover their equivalents.

[0184] List of abbreviations: Abbreviation Description DU Distributed Unit CU Centralized / Central Unit NB NodeB AP access point BS base station SWNC Software-Defined Networking Component NDT Network Digital Twin OAM Operation, Management and Maintenance QoS (Quality of Service) SDO Standards Organization UE User Equipment

Claims

1. A method for providing human-interpretable semantic information about the network state of a software-defined networking component (SWNC), wherein the human-interpretable semantic information is generated in a target communication network or a network digital twin (NDT) of the target communication network, comprising: Receive data elements and information sets, the data elements and information sets identifying predefined test cases on which the SWNC is executed in the target communication network; Based on the received data elements and information set, a network state information model is trained to determine human-interpretable semantic information of at least one network state of the SWNC identified during execution in the target communication network or the NDT, based on the identified network state. At least one network model, including the trained network state information model, is applied to the target communication network or the NDT to generate at least one human-interpretable semantic information of one or more network states of the SWNC determined during execution in the target communication network or the NDT. as well as The trained network state information model reports at least one human-interpretable semantic information associated with the one or more network states.

2. The method according to claim 1, wherein: The data elements and information set for identifying predefined test cases include at least one of the following: the predefined test conditions on which the SWNC is executed, including the expected SWNC behavior of the state information about the network state generated by the SWNC when it is executed under the predefined test conditions, the expected output of the SWNC, and the observable test output of the SWNC identified after the SWNC is executed using the associated test conditions.

3. The method according to claim 2, further comprising: Test case execution is performed by executing the SWNC of the target communication network under the associated test conditions and determining the output of the SWNC as the observable test output to generate the data elements and information set used for training network state information.

4. The method according to claim 1 or 2, wherein: The training of the network state information model further includes: Based on the received data elements and information set, the network state information model is trained in the state recognition model segment of the network state information model to identify the network state of the SWNC based on the observable output after the SWNC executes in the target communication network or the NDT; and Based on the received data elements and information set, and the identified network state, the network state information model is trained in the state information generation model segment of the network state information model to generate human-interpretable semantic information associated with the identified network state.

5. The method of claim 4, wherein applying at least one network model to the target communication network or the NDT further comprises: Under unknown conditions, the SWNC is executed in the target communication network or the NDT of the target communication network in order to generate the observable output; The trained network state information model is applied to the target communication network or the NDT by: the state recognition model segment of the trained network state information model determining the current network state of the executed SWNC based on the generated observable output; And the state information generation model segment of the network state information model generates at least one human-interpretable semantic information of the one or more network states of the SWNC based on the network state determined by the state recognition model segment.

6. The method according to claims 1 to 2, wherein the training of the network state information model further comprises: The received data elements and information set are sent to a network state model, which is configured to identify the network state of the SWNC based on the observable test output of the SWNC, which is being executed in the target communication network. Based on the transmitted data elements and information set, the network state model is applied to identify the network state of the SWNC being executed in the target communication network based on the imported data elements and information set. as well as The network state information model is trained based on the received data elements and information set and based on the identified network states to generate human-interpretable semantic information associated with at least one of the identified network states.

7. The method of claim 6, wherein applying at least one network model to the target communication network or the NDT further comprises: The SWNC is performed under unknown conditions in the target communication network or the NDT of the target communication network in order to generate an observable output; The network state model is applied by determining the existing network state of the performed SWNC based on the generated observable output; as well as The network state information model is applied by generating at least one human-interpretable semantic information of the network state of the SWNC based on the network state determined by the network state model.

8. The method of claim 1 or 2, wherein the report of the at least one human-interpretable semantic information further comprises: The one or more network states of the SWNC are annotated using at least one associated human-interpretable semantic information; as well as The report describes one or more network states, wherein the associated human-interpretable semantic information is annotated into the network states.

9. The method according to claim 1, wherein: The data elements and information sets that identify predefined test cases correspond to log information generated during the execution of the SWNC's continuous integration / continuous development (CI / CD) pipeline; as well as The method further includes: Perform CI / CD pipeline execution on different test cases of the SWNC to generate the data elements and information set for training the network state information model. The steps of performing CI / CD pipeline execution include at least: performing test case execution for multiple test cases by repeatedly executing the SWNC of the target communication network under predefined test conditions, and collecting log information generated during test case execution, the log information including at least information about: the associated test conditions, the generated network state, and the observable test output of the SWNC.

10. The method of claim 1, wherein the training of the network state information model further comprises: Make selection of network states specific to the test cases, the test cases being associated with the data elements and information sets used to train the network state information model; as well as The network state information model is trained to associate the selected network state with the data elements and information set.

11. The method of claim 10, wherein The selection of the network state specific to the test case is performed by comparing the network state of the SWNC associated with the test case with the network state of the SWNC that existed during the execution of other test cases in the target communication network, and excluding the network state of the SWNC associated with the test case that shares a predefined amount of similarity with the network state of the other test cases.

12. The method of claim 1, wherein... The human-interpretable semantic information corresponds at least to a combination of information included in the data elements and information sets in different test cases, and the SWNC is executed in the target communication network according to the different test cases; wherein In order to generate the at least one human-interpretable semantic information, the network state information model identifies the network state of the SWNC being executed as a combination of network states associated with one or more test cases in the different test cases, and generates the at least one human-interpretable semantic information based on the information set of the data elements and information sets included in the identification of the corresponding one or more test cases in the different test cases.

13. The method of claim 12, wherein Identifying the network state as a combination of network states associated with one or more test cases among the different test cases is performed by: demultiplexing the network states associated with the different test cases, and identifying combinations of one or more test cases among the different test cases, wherein the combination of the corresponding demultiplexed network states shows a predefined amount of similarity to the network state of the SWNC being executed.

14. The method of claim 1, wherein... The SWNC corresponds to the terminal device of the target communication network or the NDT, wherein the terminal device is at least a user equipment (UE). The step of applying at least one network model, including the trained network state information model, to the target communication network or the NDT is performed locally at the SWNC.

15. The method of claim 14, further comprising: The at least one network model, including the trained network state information model, is transmitted to the SWNC; Based on local data stored in the SWNC, at least one network model including the trained network state information model is applied at the SWNC, the local data including at least one of the following: the radio interface of the SWNC, the quality of telecommunications service, and the traffic measurement of the SWNC; as well as The step of reporting the human-interpretable semantic information further includes: reporting the human-interpretable semantic information to the SWNC, or conveying the human-interpretable semantic information to the target communication network or another network component of the NDT.

16. The method according to claim 1, further comprising: Perform an action indicated by the at least one human-interpretable semantic information to the SWNC, the target communication network, or another network component of the NDT.

17. The method of claim 16, wherein: At least the following steps are performed in a closed loop until a predefined parameter of the target communication network or the NDT changes or a predefined number of iterations of the closed loop is reached: applying at least one network model, including the trained network state information model, to the target communication network or the NDT; having the trained network state information model report the at least one human-interpretable semantic information associated with the one or more network states; and performing the actions indicated by the at least one human-interpretable semantic information on the SWNC or other network components of the target communication network or the NDT.

18. A method for providing human-interpretable semantic information about the network state of a software-defined networking component (SWNC), said human-interpretable semantic information being generated in a target communication network or a network digital twin (NDT), comprising: At least one network model, including a trained network state information model, is applied to the target communication network or the NDT to generate at least one human-interpretable semantic information of one or more network states of the SWNC determined during execution in the target communication network or the NDT, wherein the trained network state information model is trained in such a way that, based on the determined network states, the trained network state information model is configured to output the human-interpretable semantic information that identifies the associated network states in a human-interpretable manner. as well as The trained network state information model reports at least one human-interpretable semantic information associated with the one or more network states.

19. A network node of a target communication network, configured to provide human-interpretable semantic information about the network state of a software-defined networking component (SWNC), the human-interpretable semantic information being generated in the target communication network of the network node or in a network digital twin (NDT) of the target communication network, wherein the first network node comprises: At least one processor; as well as At least one memory storing instructions, which, when executed by the at least one processor, cause the first network node to at least: Receive data elements and information sets, the data elements and information sets identifying predefined test cases on which the SWNC is executed in the target communication network; Based on the received data elements and information set, a network state information model is trained to determine human-interpretable semantic information of at least one network state of the SWNC identified during execution in the target communication network or the NDT, based on the identified network state. At least one network model, including the trained network state information model, is applied to the target communication network or the NDT to generate at least one human-interpretable semantic information of one or more network states of the SWNC determined during execution in the target communication network or the NDT. as well as The trained network state information model outputs at least one human-interpretable semantic information associated with the one or more network states.

20. The network node according to claim 19, wherein: The SWNC corresponds to a terminal device of the target communication network or the NDT, the terminal device being at least a user equipment (UE), wherein the network node is at least used to apply the trained network state information model, the network state information model being further configured to: The at least one network model, including the trained network state information model, is transmitted to the SWNC; Based on local data stored in the SWNC, at least one network model, including the trained network state information model, is locally applied at the SWNC. The local data includes at least one of the following: the SWNC's radio interface, quality of service, and traffic measurements of the SWNC; and The network nodes are configured to at least report the human-interpretable semantic information, and the network nodes are further configured to: The human-interpretable semantic information is reported to the SWNC or conveyed to the target communication network or another network component of the NDT.

21. A computer-readable medium having instructions stored thereon, which, when executed by at least one processing unit of a machine, cause the machine to perform the method according to any one of claims 1 to 18.

22. A computer-readable medium having instructions stored thereon, which, when executed by at least one processing unit of a machine, cause the machine to perform the method according to claim 18.

Images