Digital twin entity for replicating a physical wireless network

Abstract

The present disclosure relates to a digital twin entity (110) for replicating a physical wireless network (120) that is representable as a plurality of network nodes (121a, 121b) and interactions (122) between the network nodes (121a, 121b) in a heterogeneous structure (123) having nodes and edges of different type. The digital twin entity comprises: an interface (130) with the physical wireless network for receiving information about the network nodes and interactions between the network nodes of the physical wireless network. The information can be represented as a plurality of heterogeneous structures derived from measurements of the network nodes, each heterogeneous structure representing a structure of the physical wireless network obtained by a respective network node. The digital twin entity comprises: a service mapping model, SMM, subsystem (140) configured to provide a replica of the physical wireless network based on an Artificial Intelligence and / or Machine Learning (AI / ML) model (141). The AI / ML model is trained with the heterogeneous structures derived from the measurements of the network nodes. The AI / ML model is configured to generate a digital representation (702) for each network node of the physical wireless network.

Description

[0001] DIGITAL TWIN ENTITY FOR REPLICATING A PHYSICAL WIRELESS NETWORK

[0002] TECHNICAL FIELD

[0003] The present disclosure relates to the field of Artificial Intelligence (Al) and Machine Learning (ML) applied to mobile networks such as 3GPP systems. In particular, the disclosure relates to a digital twin entity for replicating a physical wireless network and to methods to enable a digital twin model of radio access network based on heterogeneous structural information.

[0004] BACKGROUND

[0005] Mobile networks, such as 3GPP systems, are inherently dynamic due to the high mobility of users. To enable mentioned network applications, use of a digital twin model for network management has been proposed to reduce OAM costs. Such a digital twin model of wireless network needs to be accurate and efficient to synchronize model to what is occurring in the wireless network in real-time. To rapidly replicate the behavior of wireless networks and thus successfully build a wireless network digital twin, the digital twin model is required to be accurate, fast and adaptable. In addition, wireless nodes may suddenly appear. In such scenario, digital twin model should account these different network sizes with decreased amount of model updates as well as coping with increasing size of wireless network, e.g., when mobile network is highly loaded during peak hours. The digital twin instance should be scalable with respect to the synchronization to the varying complexity of physical wireless network. Digital twin models of the radio access network which are simultaneously accurate and computationally efficient are lacking. Digital twin models need to be computationally efficient to rapidly assess multiple what-if scenarios and configurations in a safe manner even in a realtime closed-loop.

[0006] SUMMARY

[0007] This disclosure provides a solution for a fast, accurate and scalable digital representation of the physical wireless network.

[0008] The foregoing and other objects are achieved by the features of the independent claims. Further implementation forms are apparent from the dependent claims, the description and the figures. The disclosure presents a digital twin entity for replicating a physical wireless network that is representable as a plurality of network nodes and interactions between the network nodes in a heterogeneous structure having nodes and edges of different type.

[0009] The digital twin entity is fast (i.e. , computationally efficient) in replicating network behavior and generating digital representations of the wireless network. These enables the evaluation of many what-if wireless network scenarios / configurations in a short amount of time. By being computationally efficient, multiple what-if wireless network scenarios / configurations may be quickly evaluated. Real-time closed-loop network optimization using a digital twin model is enabled, and more optimized solutions are transferred to the physical wireless network.

[0010] The digital twin entity is accurate to reflect the wireless system behavior at the network- and network node-level, in the sense that it has the ability to precisely estimate key performance indicators (KPIs) with respect to each wireless node (e.g., downlink data rate per user, cellthroughput per BS, etc.). In addition, the model can accurately replicate network behaviour under different dynamics and characteristics of wireless network (e.g., UE status changes, different functionalities, different wireless phenomena). Some examples of dynamics of wireless network include but are not limited to procedures such as handovers (HO), changes in scheduled frequency resources for UEs, changes in wireless channels, changes in inter-cell interference and cell-load fluctuations due to mobility, etc...

[0011] The digital twin entity is scalable to account different wireless network topologies / scenarios (e.g., unseen wireless network deployments / loads during training) with reduced model update and reduced performance degradation. It is also able to cope with increasing size of wireless network (e.g., dense small cell deployments). In terms of implementation in a real wireless system (e.g., 5G system), the digital twin instance is scalable to multiple requests from multiple network applications in RAN, CORE (e.g., 5G SBA) and / or end-users, in order to be integrated seamlessly in a service-based architecture.

[0012] The digital twin entity presented in this disclosure can be deployed as digital twin instance(s) in wireless networks in existing or new network functions (NFs) in RAN / Core. The digital twin entity or model can be based on a 5G service-based architecture (SBA) by exposing a common digital twin service to other NFs. In this way, the digital twin instance can be self-contained and loosely coupled with other NFs, thus facilitating its deployment without impacting existing NFs in RAN and CORE. The digital twin instance is scalable with respect to multiple network applications (via the northbound interface) requesting digital twin services. The digital twin entity presented in this disclosure can integrate AI / ML for in-network optimization in RAN. The digital twin can provide services inter alia to two general network applications i) what-if performance analysis, and ii) generation of synthetic data for AI / ML model training to support the envisioned use cases. Other network applications that don’t exist yet are expected to be supported by the digital twin of wireless network.

[0013] The solution presented in this disclosure is based on the following ideas: Wireless network nodes and the underlying wireless phenomena between them are represented as nodes and interactions of different type in a heterogeneous structure (e.g., a heterogeneous graph). Heterogeneous structures are fed as samples into the HGNN (heterogeneous graph neural network) model so that the model learns to simulate the underlying wireless phenomena. The topology of the input heterogeneous structures is constructed based on wireless measurements of network nodes. The input heterogeneous structure defines the computation graph of HGNN-based model. By monitoring structural changes in physical wireless network, the computation graph of the digital twin model is synchronized with the structure of wireless phenomena in physical wireless network. A structural change means a change in interaction between network nodes (e.g., user executes HO (handover), changes in inter-cell interference, etc.). Different wireless interaction types in the input heterogeneous structure signal computation with different independent trainable parameters per wireless interaction type in the HGNN model. These trainable parameters are shared across nodes interacting with the same wireless phenomena (e.g., wireless communication and inter-cell interference).

[0014] These characteristics of the HGNN-based digital twin model enables significantly more efficient and more accurate computation to infer network behavior than un-structured models or homogeneous structured models as well as improved generalization to different network topologies. After training the HGNN-based digital twin model with specific learning procedures (e.g., based on self-supervised / contrastive graph representation learning), the model generates general wireless vector representations (GWVR) per network node for all node types. GWVRs are inspired from word embeddings in Natural Language Processing. GWVRs are given as input for any network application (e.g., performance in what-if scenario, network prediction, QoS prediction, network planning, AI / ML model training).

[0015] To enable the concept, a mechanism and functions are proposed to enable the creation and adaptation of models of wireless networks based on heterogeneous graph neural networks that generate general wireless vector representations. The features that enable the concept are the following: - A logical Data Instruction Entity (DIE) function as described below. These configurable instructions, referred to as Local Neighborhood Construction Rules (LNCRs) therein, are given to all wireless nodes in physical wireless network to monitor relevant changes of the local wireless situation. The instructions may be configured by the network or e.g., by CAM. A Heterogeneous Local Wireless Structure (HLWS) is constructed from the local wireless situation per Wireless Interaction Type. The HLWS consists of a set of neighbors for a given wireless node and the wireless interactions between the node and its neighbors. Wireless nodes register with DIE and said entity manages registration information type as defined below.

[0016] - A set of Local Neighborhood Construction Rules (LNCRs) per node type as described below to monitor relevant changes in physical wireless network and notify these changes to create / adapt digital twin model residing in digital twin instance. As different type of local information is available at each type of wireless node, LCNRs are defined per node type and one rule entry is defined per wireless interaction type. For each rule entry in a LCNR, after an event is triggered, the action is to notify the digital twin instance of the wireless network as described in Sec 2.2. Wireless nodes notify relevant wireless interaction changes in wireless network to digital twin instance.

[0017] - A logical Structure Assemble Entity (SAE) function as described below to assemble local structural changes per wireless interaction type reported by all wireless nodes across the physical wireless network to build a heterogeneous global wireless structure (HGWS) as described below that represents the structure of wireless network at a given instance. The assembled HGWS is given as an input sample to functional model(s) (e.g., a HGNN model) within the Service Mapping Model(s) (described below). With a HGNN-based model as functional model, the structure of the input HGWS assembled by the SAE defines the computation graph of the HGNN-based model.

[0018] - Messages to adapt digital twin model to changes in wireless network and provide digital representations to network applications. To enable above mentioned functions, messages need to be exchanged via the southbound interface (SI) (digital twin instance and physical wireless network) and the northbound interface (Nl) (digital twin instance and network applications). Wireless nodes register with DIE providing local information including but not limited available local measurement(s) per node, node IDs and node type. Thereafter, DIE communicates to wireless nodes a LNCR depending on their node type and available local measurements. Nodes start measurements and once an event from associated LCNR is triggered, wireless nodes communicate a HLWS specifying local relevant changes on who interacts wirelessly with whom (with source and target node ID, e.g., source BS ID and target UE ID in downlink communication) and the form of wireless interaction (with wireless interaction type ID as described below) to SAE. SAE assembles all HLWSs communicated by the wireless nodes across the network to create a HGWS. The HGWS (to which the feature information of the nodes and edges is attached) is given as an input sample to functional model(s) (e.g., a HGNN model) within the Service Mapping Model(s) (described below). Based on the HGWS given as input sample, the SMM generates general wireless vector representations (GWVR) per node in the HGWS (e.g., two different GWVRs type UE for UE a and UE b and one GWVR type BS for BS x). These GWVRs are provided to network applications upon request via the Nl. The GWVRs can be used for any network application. Network applications include but are not limited to the following: performance in what-if network scenario, network prediction, network optimization, network planning, AI / ML model training, etc...

[0019] In order to describe the disclosure in detail, the following terms, abbreviations and notations will be used:

[0020] 5G Fifth-generation mobile network

[0021] LTE Long Term Evolution

[0022] NR New Radio

[0023] UE User Equipment

[0024] BS Base Station

[0025] RAN Radio Access Network

[0026] NF Network Function

[0027] AF Application Function

[0028] HGNN Heterogeneous Graph Neural Network

[0029] GNN Graph Neural Network

[0030] HG Heterogeneous Graph

[0031] DIE Data Instruction Entity

[0032] LNCR Local Neighborhood Construction Rules per Node Type HLWS Heterogeneous Local Wireless Structure

[0033] HGWS Heterogeneous Global Wireless Structure

[0034] SAE Structure Assemble Entity

[0035] SMM Service Mapping Model(s)

[0036] DTEM Digital Twin Entity Management

[0037] DR Data Repository

[0038] Nl Northbound Interface SI Southbound Interface

[0039] GWVR General Wireless Vector Representation

[0040] GAM Operation, Administration and Management

[0041] DRL Deep Reinforcement Learning

[0042] Heterogeneous Graph Neural Networks (HGNNs): as defined in “T. Kipf: Deep Learning with Graph Structured Representations, PhD Thesis in University of Amsterdam, 2020”. Relational (in the reference) refers to heterogeneous.

[0043] Heterogeneous Graphs (HGs): a data structure comprised by nodes of different type and the edges of between said nodes which can also be of different type, together with the feature information attached to them. In a HG, different type of information (i.e. , different data type and dimensionality) is attached to the nodes and edges. In this invention, user equipment (UE) and base stations (BS) are represented with nodes of different type in a HG which have different type of local information (e.g., position information for the former and resource block (RB) utilization for the latter). In addition, these wireless nodes interact between each other in different ways, e.g., BSs communicate in downlink with UEs while also interfering other UEs connected to neighboring BSs. This is reflected by edges of different type, a communication wireless interaction type and an interference wireless interaction type, respectively. In HGs, edges are directed, i.e., nodes between an edge can either be source node or target node depending on the direction of the edge. These edges signal computation with different independent trainable parameters per wireless interaction type in the learning model (detailed below).

[0044] Data Instruction Entity (DIE): DIE is a logical entity with the function of defining a set of configurable rules per node type (e.g., UE and BSs in wireless network), referred to Local Neighborhood Construction Rules (defined below) to be sent to nodes. One rule entry is defined per wireless interaction type (detailed in Embodiment Section). The goal of these rules is to track the relevant structural changes in the physical wireless network (e.g., handovers (HO), changes in inter-cell interference, etc.) and trigger a notification of the changes to be sent to the digital twin instance of the wireless network. DIE manages registration information from wireless nodes (e.g., ID, node type, available local node measurements (s) per node etc.) that form part in the digital twinning process of wireless network. Wireless nodes register with DIE, and upon registration, DIE communicates the set of configurable instructions per node type, referred to as local neighborhood construction therein. Wireless Node Type: refers to the classification of wireless nodes based on their functionality (e.g., UEs allow users to access radio access network services, BSs are a central connection point for a wireless device to communicate in the radio access, RIS is a programmable structure that controls the wireless propagation of electromagnetic waves, network relay is a repeater that extracts data from received signal, applies the noise correction techniques and retransmits signal, etc.). Classification may also include more granular differentiation between wireless nodes. Wireless nodes of different type have different available local node measurements (e.g., UEs measure downlink channel quality such as received signal reference power (RSRP), downlink signal-to-interference ratio (SINR) from nearby BSs and can obtain position information from end-user application, UEs measure sidelink channel quality such as Received Signal Strength Indicator (RSSI), BSs measure resource block (RB) utilization, cellload as number of UEs attached, uplink channel quality, etc...).

[0045] Local Neighborhood Construction Rules (LNCR) per Node Type: LNCRs are a set of input network parameters rules (e.g., as mathematical expressions, event-based triggers, conditional statements, etc...) that enable the detection of relevant interaction changes between the wireless nodes based on their local measurements or local network events (detailed in Embodiment Section). As different type of local information is available at each type of wireless node, LCNRs are defined per node type and one rule entry is defined per wireless interaction type. For each rule entry in a LCNR, after an event is triggered, the action is to notify the digital twin instance of the wireless network with relevant changes on which wireless node interacts with which wireless node (i.e., with source and target node ID) and form of wireless interaction (i.e., with wireless interaction type ID (defined below)).

[0046] Wireless Interaction type: refers to the classification of wireless interactions between wireless nodes based on the characteristics of said wireless interactions (e.g., downlink communication is the transmission of data from a BS to a UE with specific configurations at said transmitter BS and said receiver UE, downlink inter-cell interference is the unintended disruption by BSs to the received wireless signal of nearby UEs attached to neighboring BSs, uplink communication is the transmission of data from a UE to a BS with specific configurations at said transmitter UE and said receiver BS, sidelink communication is the transmission of data from a transmitter UE and a receiver UE, etc...). Classification may also include more granular differentiation between wireless interactions.

[0047] Heterogeneous Local Wireless Structure (HLWS) change: A notification reported by wireless nodes in wireless network about local relevant structural changes. By structural changes, we mean changes in wireless interactions between wireless nodes. Wireless interaction changes are specified with source and target wireless node ID together with the wireless interaction type of said wireless interaction change. Examples of wireless interaction changes include but are not limited to: changes in received interference, handover from source BS to target BS, change of frequency band or radio access technology, addition of secondary wireless link to secondary serving BS, etc. These notifications from the wireless nodes across the wireless network are reported to the Structure Assemble Entity (defined below) residing within digital twin instance.

[0048] Heterogeneous Global Wireless Structure (HGWS): A data structure that represents the structure of the wireless network at a given time instance. HGWS is comprised by multiple HLWSs from same time instance assembled together. HGWS captures wireless interactions between wireless nodes beyond their one-hop local neighborhood. HGWS represent the structure of the physical wireless network at a given time instance. HWGS includes feature information attached to the wireless nodes and wireless interactions.

[0049] Structure Assemble Entity (SAE): SAE is a logical entity with the function of assembling HLWSs reported by wireless nodes across the wireless network to build a HGWS (e.g., a heterogeneous graph).

[0050] General Wireless Vector Representations (GWVR): GWVRs are low-dimensional, learned continuous vector representations (i.e. , real-valued numbers) of the wireless network, where a GWVR is generated per wireless node by the digital twin functional model (i.e., a HGNN). GWVRs capture local structural information per wireless interaction type together with local and neighborhood feature information. GWVRs are defined per wireless node, per edge as a function of the GWVRs of the source node and target node and per graph as a function of all GWVRs of wireless nodes within HGWS. Different types of GWVRs (e.g., with different dimensionality) are generated per type of node and type of edge (e.g., a UE GWVR and a BS GVR). GWVRs are useful for any network application (defined in Sec. 3.5). For example, GWVRs can be given as input to other functions to compute wireless network KPIs (e.g., downlink rate for a UE wireless node or cell-throughput for a BS wireless node, etc...).

[0051] Digital Twinning Process: refers to the process in which wireless nodes periodically send local and structural information to digital twin instance (defined in Sec. 3.5.) and said instance internally replicates the behavior of wireless nodes. Network applications (defined in Sec. 3.5.) can request digital representations in the form of GWVRs of different parts of physical wireless network to the digital twin instance. Said instance efficiently provides GWVRs in real-time, including alternative what-if scenarios. If closed-loop control, digital twin instance may send control information to wireless nodes to perform control.

[0052] Representation to Phenomena Functional Mapping: refers to a functional mapping with GWVRs as input and a real-world phenomenon parameter as output (e.g., downlink rate, delay jitter, uplink packet error rate, uplink spectral efficiency, etc...). The phenomena are measurable in physical wireless network. Thus, said functional mapping allows network applications evaluate the outcome of different what-if network configurations with GWVRs.

[0053] According to a first aspect, the disclosure relates to a digital twin entity for replicating a physical wireless network that is representable as a plurality of network nodes and interactions between the network nodes in a heterogeneous structure having nodes and edges of different type, the digital twin entity comprising: an interface with the physical wireless network for receiving information about the network nodes and interactions between the network nodes of the physical wireless network, the information being representable as a plurality of heterogeneous structures derived from measurements of the network nodes, each heterogeneous structure representing a structure of the physical wireless network obtained by a respective network node; a service mapping model, SMM, subsystem configured to provide a replica of the physical wireless network based on an Artificial Intelligence and / or Machine Learning, AI / ML, model, the AI / ML model being trained with the heterogeneous structures derived from the measurements of the network nodes; wherein the AI / ML model is configured to generate a digital representation for each network node of the physical wireless network.

[0054] Such a digital twin entity allows for a fast, accurate and scalable digital representation of the physical wireless network.

[0055] In an exemplary implementation of the digital twin entity, the AI / ML model comprises a heterogeneous graph Neural Network, HGNN.

[0056] By using such a HGNN, a model using nodes and edges of different type can be applied which can accurately replicate the structure of the physical wireless network with its different kinds of base stations (BS) and user equipment (UE) and connections in between.

[0057] In an exemplary implementation of the digital twin entity, the AI / ML model is configured to generate a general wireless vector representation for each network node of the physical wireless network. Such GWVRs are easy to compute because they are low-dimensional, learned continuous vector representations (i.e., real-valued numbers) of the wireless network, where a GWVR is generated per wireless node by the digital twin functional model (i.e., a HGNN). By such GWVRs local structural information per wireless interaction type together with local and neighborhood feature information can be efficiently captured.

[0058] In an exemplary implementation of the digital twin entity, the digital twin entity comprises: a digital twin entity management subsystem configured to monitor performance and resource consumption of the AI / ML model.

[0059] By such digital twin entity management subsystem performance and resource consumption of the AI / ML model can be efficiently monitored.

[0060] In an exemplary implementation of the digital twin entity, the digital twin entity comprises: a data repository configured to collect and store the information about the network nodes and the interactions between the network nodes of the physical wireless network by collecting and updating real-time data of the network nodes via the interface with the physical wireless network.

[0061] In this data repository all relevant information for replication of the physical wireless network can be stored and is available for further processing.

[0062] In an exemplary implementation of the digital twin entity, the digital twin entity comprises: an interface with a network applications system for receiving digital twin service requests from a plurality of network applications and providing services of the digital twin entity to the plurality of network applications.

[0063] By such interface, network applications can easily request services of the digital twin entity.

[0064] In an exemplary implementation of the digital twin entity, the digital twin service requests comprise at least one of: request for an identification of at least one network node of the physical wireless network for which a digital representation is requested by a network application; request of a time stamp for the digital representation; and optional information comprising at least one of a phenomena ID and one or more What-if network configuration IDs; wherein the services of the digital twin entity comprise at least one of: digital representation of the at least one network node of the physical wireless network specified in the digital twin service request; time stamp for the digital representation; and optional information comprising at least one of a representation to phenomena functional mapping as per phenomena ID and an optimal network configuration.

[0065] The phenomena ID is optional information in the Digital Twin Service Request. The phenomena ID is associated with the Representation to Phenomena Functional Mapping as defined above (see ..Representation to Phenomena Functional Mapping") and as described below with respect to Figures 1 and 2.

[0066] Another optional information in the Digital Twin Service Request are What-if Network Configuration ID(s) together with the desired what-if network configurations / states to be evaluated including expected future states / configurations (e.g., alternative UE position(s), alternative BS deployment and / or resource scheduling policy, alternative cell-load, etc... ,) per What-if Network Configuration ID as described below with respect to Figures 1 and 2.

[0067] Optional information in the Digital Twin Service Message includes: representation to Phenomena Functional Mapping as per Phenomena ID in service request, e.g., as described below with respect to Figures 1 and 2. Further optional information in the Digital Twin Service Message includes the optimal network configuration for the optimization action specified in the service request, e.g., as described below with respect to Figures 1 and 2.

[0068] By such information, services of the digital twin entity can be easily and efficiently identified.

[0069] In an exemplary implementation of the digital twin entity, the digital twin entity comprises: a data instruction entity, DIE, configured to define a set of configurable rules per node type of the network nodes; wherein the DIE is configured to register the network nodes of the physical wireless network and upon registration communicate the set of configurable rules to the network nodes.

[0070] Such DIE entity or function enables the definition of a set of configurable instructions to collect local structural information as HLWS from wireless nodes.

[0071] In an exemplary implementation of the digital twin entity, the set of configurable rules are local neighborhood construction rules, LNCR, that enable detection of interaction changes between the network nodes based on local measurements or local network events of the network nodes.

[0072] By such LNRCs, relevant structural changes in physical wireless network can be efficiently monitored per wireless interaction type and these HLWS changes can be notified to create / adapt digital twin model residing in digital twin instance. Adapting model enables significantly more efficient and more accurate computation to infer network behavior than unstructured models.

[0073] In an exemplary implementation of the digital twin entity, the DIE is configured to receive local information from the network nodes comprising at least one of: an identifier of a respective network node; a node type of the respective network node; available measurements of the respective network node; a communication functionality of the respective network node.

[0074] This enables efficient aggregation of HLWS changes reported by wireless nodes across the physical wireless network.

[0075] In an exemplary implementation of the digital twin entity, the digital twin entity comprises: a structure assemble entity, SAE, configured to build a Heterogeneous Global Wireless data structure, HGWS, that represents a structure of the physical wireless network at a given time instance based on notifications of the network nodes reporting local structural changes of the physical wireless network; wherein the SAE is configured to provide the HGWS to the SMM subsystem for adapting the AI / ML model to the local structural changes of the physical wireless network.

[0076] This enables the aggregation of HLWS changes reported by wireless nodes across the physical wireless network to build a HGWS.

[0077] Local structural changes of the physical wireless network are also referred to as Heterogeneous Local Wireless Structure (HLWS) changes.

[0078] In an exemplary implementation of the digital twin entity, the notifications of the network nodes reporting the local structural changes of the physical wireless network comprise at least one of the following: a wireless interaction type ID, a source wireless node, a target wireless node, an optional trigger signal that triggered the local structural change report of the network nodes.

[0079] The trigger signal represents a triggering measurement (e.g., downlink RSRP received at target UE from new source serving BS during HO event, inter-cell interference measurement received at target UE from source interfering BS, sidelink RSSI received at target UE from source UE, etc...). The local structural changes of the physical wireless network, i.e., HLWS changes, are communicated from network nodes to SAE (which resides in digital twin entity).

[0080] In an exemplary implementation of the digital twin entity, the digital twin entity is configured for implementation in a 3GPP radio access network, RAN, entity of the physical wireless network.

[0081] The digital twin entity can be flexible implemented in different entities of the 3GPP radio access network.

[0082] In an exemplary implementation of the digital twin entity, the SMM subsystem and the SAE entity are configured for implementation in a data analytics function component of a 3GPP radio access network, RAN, entity of the physical wireless network; and wherein the DIE entity is configured for implementation in an operation and maintenance, GAM, entity of a 3GPP core network.

[0083] This allows flexible implementation. Parts of the digital twin entity can be implemented in the RAN and other parts can be implemented in the core network.

[0084] In an exemplary implementation of the digital twin entity, the SMM subsystem and the SAE entity are configured for implementation in a data analytics function component of a 3GPP core network; and the DIE entity is configured for implementation in an operation and maintenance, 0AM, entity of the 3GPP core network.

[0085] This allows flexible implementation. The digital twin entity can be implemented in the core network, in particular distributed over different entities of the core network.

[0086] In an exemplary implementation of the digital twin entity, the digital twin entity is configured for implementation in a mobile edge computing, MEC, server connected to the physical wireless network.

[0087] This allows flexible implementation. The digital twin entity can be implemented in a single MEC server that is connected to the physical wireless network.

[0088] According to a second aspect, the disclosure relates to a method for replicating a physical wireless network by a digital twin entity, wherein the physical wireless network is representable as a plurality of network nodes and interactions between the network nodes in a heterogeneous structure having nodes and edges of different type, the method comprising: receiving information about the network nodes and interactions between the network nodes of the physical wireless network, the information being representable as a plurality of heterogeneous structures derived from measurements of the network nodes, each heterogeneous structure representing a structure of the physical wireless network obtained by a respective network node; providing a replica of the physical wireless network based on an Artificial Intelligence and / or Machine Learning, AI / ML, model, the AI / ML model being trained with the heterogeneous structures derived from the measurements of the network nodes; and generating a digital representation for each network node of the physical wireless network by the AI / ML model.

[0089] Such a method for replicating a physical wireless network by a digital twin entity allows for a fast, accurate and scalable digital representation of the physical wireless network.

[0090] BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Further embodiments of the disclosure will be described with respect to the following figures, in which:

[0092] Figure 1 shows a schematic diagram illustrating the system architecture of a wireless network system 100 with digital twin entity 110 according to the disclosure;

[0093] Figure 2 shows a schematic diagram illustrating the system architecture of the wireless network system 100 with a more detailed illustration of the digital twin entity 110;

[0094] Figure 3 shows an exemplary message chart 300 illustrating message exchange between involved entities of the wireless network system 100;

[0095] Figures 4a, 4b and 4c show exemplary network diagrams illustrating the creation of new wireless nodes in HGWS assembled in SAE;

[0096] Figure 5 shows an exemplary message chart 500 illustrating message exchange with digital twin instance in gNB according to a first embodiment;

[0097] Figure 6 shows an exemplary 3GPP wireless network system 600 according to the first embodiment, where the complete digital twin instance is implemented in a RAN entity; Figure 7 shows an implementation example of a physical wireless network 120 represented as a heterogeneous structure 123;

[0098] Figure 8 shows an implementation example of a heterogeneous structure 123 and its associated heterogeneous computation graph 123b of digital twin model;

[0099] Figure 9 shows an exemplary message chart 900 illustrating message exchange with SMM / SAE at RANDAF in RAN and DIE at OAM in core network according to a subembodiment of the first embodiment;

[0100] Figure 10 shows an exemplary 3GPP wireless network system 1000 according to the subembodiment of the first embodiment, where parts of the digital twin instance are implemented in a RAN entity and the DIE is implemented at OAM function in the core network;

[0101] Figure 11 shows an exemplary message chart 1100 illustrating message exchange with SMM / SAE at NWDAF in core network and DIE at OAM in core network according to a second embodiment;

[0102] Figure 12 shows an exemplary 3GPP wireless network system 1200 according to the second embodiment, where parts of the digital twin instance are implemented in NWDAF and the DIE is implemented at OAM function in the core network;

[0103] Figure 13 shows an exemplary message chart 1300 illustrating message exchange with SMM / SAE / DIE at MEC (mobile edge computing) according to a third embodiment; and

[0104] Figure 14 shows an exemplary 3GPP wireless network system 1400 according to the third embodiment, where the digital twin instance is implemented at MEC.

[0105] DETAILED DESCRIPTION OF EMBODIMENTS

[0106] In the following detailed description, reference is made to the accompanying drawings, which form a part thereof, and in which is shown by way of illustration specific aspects in which the disclosure may be practiced. It is understood that other aspects may be utilized and structural or logical changes may be made without departing from the scope of the present disclosure. The following detailed description, therefore, is not to be taken in a limiting sense, and the scope of the present disclosure is defined by the appended claims. It is understood that comments made in connection with a described method may also hold true for a corresponding device or system configured to perform the method and vice versa. For example, if a specific method step is described, a corresponding device may include a unit to perform the described method step, even if such unit is not explicitly described or illustrated in the figures. Further, it is understood that the features of the various exemplary aspects described herein may be combined with each other, unless specifically noted otherwise.

[0107] Figure 1 shows a schematic diagram illustrating the system architecture of a wireless network system 100 with digital twin entity 110 according to the disclosure.

[0108] The wireless network system 100 is composed of three main system components: the physical wireless network 120, the wireless network digital twin instance 110 (also referred herein as digital twin entity 110 or digital twin instance) and network applications 180. These system components are detailed in Figure 1 (and also in Figure 2) and described in the following.

[0109] Physical Wireless Network:

[0110] A wireless system operating with multiple radio access technologies (RATs) is composed of multiple types of radio access nodes (network nodes 121a in Figure 1). Such radio access nodes include but are not limited to macro-cell and micro-cell BSs belonging to different RATs, non-terrestrial networks (NTN) such as satellites, remote radio heads (RRHs), intelligent reflecting surfaces (IRSs), etc.... These radio access nodes serve UEs (network nodes 121b in Figure 1) with different functionalities / capabilities (e.g., conventional smartphones, automotive communication systems, Internet of Things (loT) devices, robot communication system, actuator communication system, etc...), with different communication functionalities (e.g., multi-element antenna topology, beamforming, support of different frequency bands, etc...) and running different types of services (e.g., enhanced mobile broadband, extended reality (XR), ultra-reliable and low latency communication, etc...).

[0111] The radio access nodes (network nodes 121a in Figure 1) and user equipment (network nodes 121b in Figure 1) may communicate in downlink, uplink and sidelink. Wireless communication may occur with different frequency bands. Radio access nodes may induce interference to neighbouring UEs (or neighbouring BSs) when using same resources.

[0112] For the digital twinning process, the physical wireless network 120 communicates via the Southbound Interface (SI) 130 to the wireless network digital twin instance 110. The SI 130 is responsible for information exchange between the wireless network digital twin instance 110 and the physical wireless network 120. Wireless Network Digital Twin Instance:

[0113] The wireless network digital twin instance 110 includes three key system sub-components: Service Mapping Models (SMMs) system sub-component 140, Digital Twin Entity Management (DTEM) system sub-component 150 and Data Repository (DR) system sub-components 160.

[0114] Service Mapping Model(s) (SMM) 140: SMM 140 is a subsystem within the digital twin instance 110 that provides data modeling and data model instances for various network applications. The HGNN-based model (also referred to as AI / ML model 141) resides within the SMM subcomponent 140 and it generates a general vector representation (detailed below) for each network node 121a, 121b in wireless network 120. Within the SMM sub-component 140 resides functional models and basic models. Functional models refer to models of physical wireless network 120 for a specific function (e.g., state monitoring, traffic analysis, fault diagnosis, performance assurance, etc...). Basic models refer to network topology, environment information, and operational state of network nodes 121a, 121 b of wireless network 120.

[0115] Digital Twin Entity Management (DTEM) 150: DTEM 150 is a subsystem which monitors performance and resource consumption of individual models residing in SMM 140, visualizes and can be enabled to control various elements of network digital twin 110.

[0116] Data Repository (DR) 160: DR 160 is a subsystem that collects and stores network data to build digital twin models that reside in SMM sub-component 140 by collecting and updating real-time data of network nodes 121a, 121 b in physical wireless network 120.

[0117] Northbound Interface (Nl) 170: Nl 170 is a network application-facing interface between the wireless network digital twin 110 and the network applications 180. The Nl 170 is responsible for information exchange between network wireless network digital twin 110 and network applications 180.

[0118] Southbound Interface (SI) 130: SI 130 is a physical twin-facing interface between the wireless network digital twin 110 and the physical wireless network 120. The SI 130 is responsible for information exchange between the wireless network digital twin 110 and the physical wireless network 120.

[0119] Network Applications

[0120] The network applications system component 180 refers to various applications including but not limited to network optimization, network prediction, performance in what-if wireless scenario, network planning, network management / orchestration and AI / ML model training. These network applications may be requested by different entities within the network including but not limited to Operation, Administration and Maintenance (OAM) functions, application functions (AFs), NFs, BSs, AMF, SMF, UPF, etc...

[0121] Names, Functions, Structures Related to the disclosure

[0122] New Functions and structure related to the disclosure are defined above.

[0123] The procedure to create and adapt digital twin model(s) of wireless network 120 and send representation(s) to network applications 180 may comprise the following seven steps:

[0124] Creation Procedure (Figure 4a shows an example of the creation of new wireless node in HGWS in SAE)

[0125] Step 1 :

[0126] A new wireless node 121 b appears in wireless network 120. Wireless nodes within network are taking part in a digital twinning process as described above. New wireless nodes 121a, 121 b in physical wireless network 120 register to DIE to form part in the digital twinning process via the SI 130. Wireless nodes 121a, 121b provide local information to DIE including but not limited to their:

[0127] - Wireless node ID,

[0128] - Wireless node type as defined above,

[0129] - available local node measurements as defined above,

[0130] - and node communication functionality (OPTIONAL: If UE, provide configured QoS requirement(s) associated to their PDU session. Node-related information available in other NFs may also be provided to DIE).

[0131] Step 2:

[0132] Upon initial registration of wireless nodes during the initial creation of digital twin functional model(s), DIE 210 (see Figure 2) specifies LNCR per node type with user input (e.g., OAM). Additionally, LCNR may be provided dynamically by DTEM 150. DIE 210, based on available local node measurements and distinct wireless node types, specifies the number of LNCRs (i.e., one per wireless node type) and wireless interaction types (i.e., one rule entry of the LCNR per wireless interaction type). Using the provided available local node measurements, DIE 210 specifies configurable events that track the topological structure of the local wireless situation of nodes per wireless interaction type (i.e., a rule entry of a particular LNCR). DIE 210 communicates the LNCRs per node type to wireless nodes 121a, 121b via the SI 130. LCNRs defined in DIE 210 can be defined dynamically or with user input with OAM function.

[0133] Adaptation Procedure (Figure 4b shows the adaptation of the HGWS in SAE 220 (see Figure 2) when said wireless node changes state.)

[0134] Step 3:

[0135] Wireless nodes 121a, 121b configure internal process, start measurements and track events according to the provided LNCRs.

[0136] Step 4:

[0137] Network events, based on the rule entries / conditional statements specified in LNCR, trigger HLWS changes on wireless nodes. Wireless nodes 121a, 121 b send HLWS change(s) to the SAE 220 (see Figure 2) as but not limited to:

[0138] - Wireless interaction type ID as defined above,

[0139] - source wireless node,

[0140] - target wireless node,

[0141] - OPTIONAL: triggering measurement (e.g., downlink RSRP (reference signal received power) received at target UE (user equipment) from new source serving BS (base station) during HO (handover) event, inter-cell interference measurement received at target UE from source interfering BS, sidelink RSSI (received signal strength indicator) received at target UE from source UE, etc...).

[0142] If multiple events occur simultaneously (e.g., simultaneous increase of downlink inter-cell interference induced by another neighboring BS and also a HO event occurs), events can be multiplexed and reported in the same HLWS change. HLWS changes are reported to SAE 220.

[0143] Step 5:

[0144] The SAE 220 assembles HLWSs received from wireless nodes 121a, 121b across wireless network 120 and builds HGWS of physical wireless network. SAE 220 communicates the HGWS to the SMM function. Upon reception of the HGWS, the functional model residing in SMM 140 as defined above updates its computation graph based on the HGWS. With the HGWS given as input sample, functional model generates a GWVR per node in the input HGWS. The digital twinning process for new wireless node is active. Step 6:

[0145] Digital twin instance 110 is replicating physical wireless network 120 and awaits digital twin service requests 171 from network applications 180 via the Nl 170. Network applications 180 include but are not limited to network optimization, network prediction, performance in what-if wireless scenario, network planning, network management and orchestration, and AI / ML model training ran by other NFs / AFs such as Network Data Analytics Function (NWDAF), gNB, UPF, etc... Network applications 180 request 171 digital twin service to the digital twin instance 110. The digital twin service request 171 includes but is not limited to:

[0146] - wireless node ID(s) for which a digital representation (GWVRs) are requested by network applications,

[0147] - time-stamps for GWVRs,

[0148] - OPTIONAL: Phenomena ID associated with the Representation to Phenomena Functional Mapping as defined above,

[0149] - OPTIONAL: What-if Network Configuration ID(s) together with the desired what-if network configurations / states to be evaluated including expected future states / configurations (e.g., alternative UE position(s), alternative BS deployment and / or resource scheduling policy, alternative cell-load, etc... ,) per What-if Network Configuration ID. Digital twin instance 110 may have the capability to generate what-if network configuration / states as well.

[0150] - OPTIONAL: network applications 180 such as trial-and-error network optimization (e.g., resource allocation based on Reinforcement Learning, Bayesian Optimization, Brute-Force, etc...), can optionally request directly optimal network configurations (i.e., optimization variables) to avoid communication overhead (multiple message requests / responses or subscription to exchange representations and optimization actions). Digital twin entity management (DTEM) 150 residing in digital twin instance 110 can perform optimization, coined as virtual control, using SMM functional model(s). DTEM 150 provides optimal network configuration to network application 180. Network application 180 can thereafter control the physical wireless network 120 with found network configuration using digital twin models.

[0151] Step 7:

[0152] SMM 140 residing in digital twin instance 110 provides GWVRs as per the digital twin service requests 171 to network applications 180. The digital twin service response 172 includes but is not limited to:

[0153] - GWVRs per wireless node ID(s) specified in service request 171 , OPTIONAL: GWVRs per wireless node ID per Network Configuration ID specified in service request 171 may be provided. GWVRs of what-if scenarios can be generated based on history information of digital twin instance replicating physical wireless network.

[0154] - Time-stamps for GWVRS, - OPTIONAL: Representation to Phenomena Functional Mapping as per Phenomena ID in service request 171.

[0155] - OPTIONAL: DTEM 150 residing in digital twin instance 110 can provide optimal network configuration for the optimization action specified in the service request 171.

[0156] Figure 2 shows a schematic diagram illustrating the system architecture of the wireless network system 100 with a more detailed illustration of the digital twin entity 110.

[0157] In particular, the wireless network system 100 in Figure 2 corresponds to the wireless network system 100 in Figure 1 with additional Data Instruction Entity (DIE) 210 and Structure Assemble Entity (SAE) 220 residing in the wireless network digital twin instance 110 (also referred herein as digital twin entity 110).

[0158] As described above with respect to Figure 1 , the wireless network system 100 is composed of three main system components: the physical wireless network 120, the wireless network digital twin instance 110 (also referred herein as digital twin entity 110 or digital twin instance) and network applications 180.

[0159] The digital twin entity 110 can be used for replicating a physical wireless network 120 that is representable as a plurality of network nodes 121a, 121 b and interactions 122 between the network nodes 121a, 121b in a heterogeneous structure 123 having nodes and edges of different type.

[0160] The digital twin entity 110 comprises an interface 130 (also denoted as SI) with the physical wireless network 120 for receiving information about the network nodes 121a, 121b and interactions 122 between the network nodes 121a, 121b of the physical wireless network 120. This information can be represented as a plurality of heterogeneous structures 123 derived from measurements of the network nodes 121a, 121 b, each heterogeneous structure 123 representing a structure of the physical wireless network 120 obtained by a respective network node 121a, 121 b.

[0161] The digital twin entity 110 comprises a service mapping model (SMM) subsystem 140 configured to provide a replica of the physical wireless network 120 based on an Artificial Intelligence and / or Machine Learning (AI / ML) model 141. This AI / ML model 141 is trained with the heterogeneous structures 123 derived from the measurements of the network nodes 121a, The AI / ML model 141 is configured to generate a digital representation 702 (see Figure 7 for more details) for each network node 121a, 121b of the physical wireless network 120.

[0162] The AI / ML model 141 may comprise a heterogeneous graph Neural Network (HGNN) as described above.

[0163] The AI / ML model 141 may be configured to generate a general wireless vector representation (GWVR) 702 as defined above for each network node 121a, 121b of the physical wireless network 120.

[0164] The digital twin entity 110 may comprise a digital twin entity management subsystem 150 configured to monitor performance and resource consumption of the AI / ML model 141.

[0165] The digital twin entity 110 may comprise a data repository 160 configured to collect and store the information about the network nodes 121a, 121b and the interactions 122 between the network nodes 121a, 121 b of the physical wireless network 120 by collecting and updating real-time data of the network nodes 121a, 121b via the interface 130 with the physical wireless network 120.

[0166] The digital twin entity 110 may comprise an interface 170 (also denoted as Nl) with a network applications system 180 for receiving digital twin service requests 171 from a plurality of network applications 181 , 182 and providing services 172 of the digital twin entity 110 to the plurality of network applications 181 , 182.

[0167] The digital twin service requests 171 may comprise at least one of: request for an identification of at least one network node 121a, 121b of the physical wireless network 120 for which a digital representation 702 is requested by a network application; request of a time stamp for the digital representation 702.

[0168] The services 172 of the digital twin entity 110 may comprise at least one of: digital representation 702 of the at least one network node 121a, 121 b of the physical wireless network 120 specified in the digital twin service request 171 ; time stamp for the digital representation 702.

[0169] The digital twin entity 110 may comprise a data instruction entity (DIE) 210, configured to define a set of configurable rules 211 per node type of the network nodes 121a, 121b. The DIE 210 may be configured to register the network nodes 121a, 121 b of the physical wireless network 120 and upon registration communicate the set of configurable rules 211 to the network nodes 121a, 121b.

[0170] The set of configurable rules 211 can be local neighborhood construction rules (LNCR) that enable detection of interaction changes between the network nodes 121a, 121b based on local measurements or local network events of the network nodes 121a, 121 b.

[0171] The DIE 210 may be configured to receive local information from the network nodes 121a, 121 b comprising at least one of: an identifier of a respective network node; a node type of the respective network node; available measurements of the respective network node; a communication functionality of the respective network node.

[0172] The digital twin entity 110 may comprise a structure assemble entity (SAE) 220, configured to build a Heterogeneous Global Wireless data structure (HGWS) that represents a structure of the physical wireless network 120 at a given time instance based on notifications of the network nodes 121a, 121 b reporting local structural changes of the physical wireless network 120.

[0173] The SAE 220 may be configured to provide the HGWS to the SMM subsystem 140 for adapting the AI / ML model 141 to the local structural changes of the physical wireless network 120.

[0174] Local structural changes of the physical wireless network are also referred to as Heterogeneous Local Wireless Structure (HLWS) changes.

[0175] The notifications of the network nodes 121a, 121 b reporting the local structural changes of the physical wireless network 120 may comprise at least one of the following: a wireless interaction type ID, a source wireless node, a target wireless node, an optional trigger signal that triggered the local structural change report of the network nodes 121a, 121b.

[0176] The trigger signal represents a triggering measurement (e.g., downlink RSRP received at target UE from new source serving BS during HO event, inter-cell interference measurement received at target UE from source interfering BS, sidelink RSSI received at target UE from source UE, etc...).

[0177] The local structural changes of the physical wireless network, i.e., HLWS changes, are communicated from network nodes to SAE (which resides in digital twin entity). The digital twin entity 110 may be configured for implementation in a 3GPP radio access network (RAN) entity 610 as shown in Figure 6 of the physical wireless network 120.

[0178] The SMM subsystem 140 and the SAE 220 entity may be configured for implementation in a data analytics function component 1010 as shown in Figure 10 of a 3GPP radio access network (RAN) entity of the physical wireless network 120. The DIE entity 210 may be configured for implementation in an operation and maintenance (GAM) entity 1020 as shown in Figure 10 of a 3GPP core network 1001.

[0179] The SMM subsystem 140 and the SAE 220 entity may be configured for implementation in a data analytics function component 1030 as shown in Figure 12 of a 3GPP core network 1001. The DIE entity 210 may be configured for implementation in an operation and maintenance (GAM) entity 1020 as shown in Figure 12 of the 3GPP core network 1001.

[0180] The digital twin entity 110 may be configured for implementation in a mobile edge computing (MEC) server 1410 as shown in Figure 14 connected to the physical wireless network 120.

[0181] The disclosure also introduces a method for replicating a physical wireless network 120 by a digital twin entity 110 as described above, wherein the physical wireless network is representable as a plurality of network nodes 121a, 121 b and interactions 122 between the network nodes in a heterogeneous structure 123 having nodes and edges of different type.

[0182] Such a method comprises: receiving information about the network nodes and interactions between the network nodes of the physical wireless network, the information being representable as a plurality of heterogeneous structures derived from measurements of the network nodes, each heterogeneous structure representing a structure of the physical wireless network obtained by a respective network node as described above.

[0183] Such a method comprises: providing a replica of the physical wireless network based on an Artificial Intelligence and / or Machine Learning, AI / ML, model, the AI / ML model being trained with the heterogeneous structures derived from the measurements of the network nodes as described above.

[0184] Such a method comprises: generating a digital representation for each network node of the physical wireless network by the AI / ML model as described above. Figure 3 shows an exemplary message chart 300 illustrating message exchange between involved entities of the wireless network system 100.

[0185] The message chart 300 illustrates communication between the digital twin entity 110 including SMM 140, SAE 220 and DIE 210 as shown in Figure 2 and the network nodes 121a, 121b of the physical wireless network 120 including UE1 to UEn, BS1 , BS2 and BS3 and the NF / AF.

[0186] After initial procedure registration 301 between DIE 210 and the physical wireless network 120 including UE1 to UEn, BS1 , BS2 and BS3 and the NF / AF, a message 302 is sent from DIE 210 to UE1 to communicate local neighborhood construction rules per node type. UE1 configures process 302a and starts measurements 303 with digital twin entity 110. Then, UE1 sends message 304 to SAE 220 to trigger update of topology of local graph per interaction type. SAE 220 sends message 305 to SMM 140 to notify updated global graph. Then SMM 140 updates computation graph 305a.

[0187] After the status has changed 305b in UE1 , e.g., by handover execution, UE1 sends message 305c to SAE 220 to trigger local neighborhood update per interaction type. SAE 220 sends message 305d to SMM 140 to notify updated global graph. Then, SMM 140 updates computation graph 305e.

[0188] NF / AF sends message 306 to SMM 140 for digital twin service request, e.g., what-if network scenario performance. SMM 140 replies with message 307 to communicate general vector representation and optionally output function. Then, NF / AF can assess performance what-if scenario 307a.

[0189] Figures 4a, 4b and 4c show exemplary network diagrams illustrating the creation of new wireless nodes in HGWS assembled in SAE.

[0190] In Figure 4a, a new wireless node 121b (UE1) appears in wireless network 120. Wireless nodes within network are taking part in a digital twinning process as described above. The new wireless node 121b for UE1 registers to DIE 210 to form part in the digital twinning process via the SI 130. Wireless nodes 121a, 121 b provide local information to DIE 210 as described above.

[0191] In Figure 4b, adaptation of the HGWS in SAE 220 is shown when said wireless node changes state. Here, a status change event occurred, e.g., when UE1 performs handover. The status change event can also apply to other node types, e.g. base stations 121a. The SAE 220 assembles HLWSs received from wireless nodes 121a, 121 b across wireless network 120 and builds HGWS of physical wireless network. SAE 220 communicates the HGWS to the SMM function. Upon reception of the HGWS, the functional model residing in SMM 140 as defined above updates its computation graph based on the HGWS. With the HGWS given as input sample, functional model generates a GWVR per node in the input HGWS. The digital twinning process for new wireless node is active.

[0192] Digital twin instance 110 is replicating physical wireless network 120 according to Figure 4c and awaits digital twin service requests 171 from network applications 180 via the Nl 170.

[0193] Figure 5 shows an exemplary message chart 500 illustrating message exchange with digital twin instance in gNB according to a first embodiment.

[0194] This first embodiment is based on the implementation of the whole digital twin instance 110, i.e., including SMM / SAE / DIE functions 140, 220, 210, in a RAN entity such as an AN (access network) (e.g., gNB 1) or a RAN Data Analytics Function (e.g., RANDAF) of 3GPP RAN. Here, another AN in RAN (e.g., gNB 2) requests digital twin services to the digital twin instance 110 to consume digital twin services for a specific network application (e.g., what-if performance analysis or Al-based network optimization). Other network applications include AI / ML model training (e.g., QoS prediction). The message exchange for this first embodiment is shown by Figure 5.

[0195] The message chart 500 illustrates communication between the digital twin entity 110 including SMM 140, SAE 220, DIE 210 and AN1 and the network nodes 121a, 121b of the physical wireless network 120 including UE1 to UEn, AN2 and AN3.

[0196] After initial procedure registration 501 between DIE 210, AN1 and the physical wireless network 120 including UE1 to UEn, AN2 and AN3, a message 502 is sent from DIE 210 to UE1 to communicate local neighborhood construction rules per node type. UE1 configures process 502a and starts measurements 503 with digital twin entity 110. Then, UE1 sends message 504 to SAE 220 to trigger update of topology of local graph per edge type. SAE 220 sends message 505 to SMM 140 to notify updated global graph. Then SMM 140 updates computation graph 505a.

[0197] After the status has changed 505b in UE1 , e.g., by handover execution, UE1 sends message 505c to SAE 220 to trigger local neighborhood update per edge type. SAE 220 sends message 505d to SMM 140 to notify updated global graph. Then, SMM 140 updates computation graph 505e.

[0198] AN2 sends message 506 to SMM 140 for digital twin service request, e.g., what-if network scenario performance. SMM 140 replies with message 507 to communicate general vector representation and optionally output function. Then, AN2 can assess performance what-if scenario 507a.

[0199] Steps 1-2) are contained in general procedure as described above with respect to Figure 1. LNCR is communicated via the SI implemented through the llu and Xn interface to wireless nodes involved in the digital twinning process.

[0200] Step 3) is contained in general procedure as described above with respect to Figure 1. An implementation example of a LNCR for node type UE and LNCR for node type BS is described below. Here, the wireless interaction type is implemented as an edge type.

[0201] Step 4) is contained in general procedure as described above with respect to Figure 1. Wireless nodes communicate HLWS change(s) to SAE implemented inside gNB via Uu and Xn interface.

[0202] Step 5) is contained in general procedure as described above with respect to Figure 1. SAE notifies updated HGWS to the SMM also implemented inside gNB 1. An implementation example of the HGWS as a heterogeneous graph is shown in Figure 7 (for brevity, node and edge feature information attached is omitted). The SMM function is implemented but not limited to the model class of HGNN models. An implementation example of the heterogeneous computation graph of a HGNN model in the SMM function is shown in Figure 8.

[0203] Step 6) is contained in general procedure as described above with respect to Figure 1. gNB 2 requests a digital twin service including a Phenomena ID from digital twin instance in gNB 1. In this embodiment, Phenomena ID is an Analytics ID. gNB 2 runs an Al-based trial-and-error optimization (e.g., RL-based or brute force resource allocation) as a network application.

[0204] Step 7) is contained in general procedure as described above with respect to Figure 1. Response includes a Representation to Phenomena Functional Mapping according to Phenomena ID in request. Network application may request further digital twin representations in what-if specific scenarios (e.g., updated optimization network actions) in a closed loop until convergence, as shown in Figure 5. After an optimal configuration is found by gNB 2 using digital twin instance 110, gNB 2 applies the resource allocation configuration found using digital twin for more optimized performance.

[0205] The SI interface can be realized for: a) gNB-gNB pairs, b) gNB-UE pairs, c) RANDAF-gNB and d) RANDAF-UE pairs via the Xn interface for the gNB-gNB pairs, via the llu interface for the gNB-UE pairs, new interfaces defined to connect new RAN entities such as a RANFAF with gNB for RANDAF-gNB pairs, and through Uu and new interfaces for RANDAF-gNB using gNB as relay for UE-RANDAF pairs. The Nl interface can be realized via the Xn, N1 and / or N2 interfaces. Figure 6 shows the message flow in 3GPP network.

[0206] Figure 6 shows an exemplary 3GPP wireless network system 600 according to the first embodiment, where the complete digital twin instance is implemented in a RAN entity.

[0207] The 3GPP wireless network system 600 includes a control plane with the entities: AF, AMF, GAM and SMF and a data plane with the entities: AN1 , UE1 , AN2, UE2, AN3, UE3, UE4, UPF and DN as defined in 3GPP.

[0208] The digital twin entity 110 with SMM 140, SAE 220 and DIE 210 is implemented in a 3GPP RAN entity 610, here the AN1 entity.

[0209] Local neighborhood construction rules (LNCR) per node type as described above with respect to Figures 1 and 2 are delivered via a first path 601 from AN1 to UE1 and AN2, from AN2 to UE2 and AN3 and from AN3 to UE3 and UE4.

[0210] Heterogeneous local wireless structure (HWLS) updates as described above with respect to Figures 1 and 2 are delivered via a second path 602 from UE4 and UE3 to AN3, from AN3 and UE2 to AN2 and from AN2 and UE1 to AN1.

[0211] General wireless vector representation (GWVR) as described above with respect to Figures 1 and 2 is delivered via a third path 603 from AN1 to AN2.

[0212] Table 1 is an implementation example of a LNCR for node type UE. Wireless interaction types are implemented as edge type. Rule Entry 1 corresponds to Edge Type 1 with description associated to “Uu DL communication” whose measurement triggering event corresponds to HO execution event and Action corresponds to step 4) as described above with respect to Figure 1. Rule Entry 2 corresponds to Edge Type 2 with description associated to “Uu DL Intercell Interference”. Event for Rule 2 is triggered if RSRP is greater than a specific threshold.

[0213] Table 1 : implementation example of a LNCR for node type UE Table 2 is an implementation example of a LNCR for node type BS. Wireless interaction types are implemented as edge type. Rule Entry 1 corresponds to Edge Type 1 with description associated to “llu llu communication” whose measurement triggering event corresponds to HO execution event and Action corresponds to step 4) as described above with respect to Figure 1. Rule Entry 2 corresponds to Edge Type 2 with description associated to “llu UL Inter- cell Interference”. Event for Rule 2 is triggered if SRS is greater than a specific threshold.

[0214] Table 2: implementation example of a LNCR for node type BS

[0215] Figure 7 shows an implementation example of a physical wireless network 120 represented as a heterogeneous structure 123.

[0216] The physical wireless network 120 depicted on the left side of Figure 7 includes in this example three base stations BS1 , BS2, BS3 as network nodes 121a and four user equipment UE1 , UE2, UE3, UE4 as network nodes 121b. Interactions 122 between the base stations 121a and the UEs 121b are denoted by the connections between the different network nodes 121a, 121 b. The physical wireless network 120 can be represented as a heterogeneous structure 123 (see right side of Figure 7) with nodes and edges of different type, i.e. with first type network nodes 121a (BS node type) representing BS1 , BS2, BS3 and second type network nodes 121b (UE node type) representing UE1 , UE2, UE3, UE4 and interactions 122 in between.

[0217] For each UE or UE node type a general wireless vector representation (GWVR) can be determined and associated to the respective UE.

[0218] Figure 7 thus shows an implementation Example of Physical Wireless Network 120 represented as a heterogeneous graph or structure 123 (i.e., HGWS). SMM can generate a GWVR per wireless node (for brevity, GWVRs are shown for UE wireless nodes).

[0219] Figure 8 shows an implementation example of a heterogeneous structure 123 and its associated heterogeneous computation graph 123b of digital twin model.

[0220] On the left side of Figure 8, a wireless network and wireless phenomena are shown as heterogeneous graphs, i.e., graphs with multiple types of nodes and edges.

[0221] In this example, three BS node type nodes are communication with one UE node type node. The heterogeneous structure 123 comprises a communication edge type path 801 from BS1 to UE1 and two interference edge type paths 802 from BS2 to UE1 and BS3 to UE1.

[0222] The SMM function is implemented but not limited to the model class of HGNN models. Figure 8 shows an implementation example of the heterogeneous computation graph of a HGNN model in the SMM function.

[0223] On the right-hand side of Figure 8, an associated heterogeneous computation graph 123b of digital twin model is shown as one example.

[0224] Each input of a respective base station BS1 , BS2, BS3 is weighted, compared to thresholds, added and normalized to compute a general wireless vector representation 702 for UE1.

[0225] Figure 9 shows an exemplary message chart 900 illustrating message exchange with SMM / SAE at RANDAF in RAN and DIE at OAM in core network according to a subembodiment of the first embodiment. This sub-embodiment of the first embodiment 1 is based on the partial implementation of the digital twin instance 110 in RAN, where SMM / SAE functions 140, 220 are co-located at a RAN Data Analytics Function (RANDAF) and the DIE function 210 is located at an OAM function in CORE. In this sub-embodiment, a network application (e.g., AI / ML model training) is run by NWDAF to train QoS prediction model and deploy model(s) for inference. The objective of the network application is to predict QoS, such as downlink throughput, that a UE will experience typically several seconds ahead. The QoS prediction model is trained on synthetic data generated by the digital twin instance 110, i.e., GWVRs per wireless node generated by SMM 140. Digital twin instance 110 not only allows to generate multiple synthetic data for corner scenarios but also its generated GWVRs capture relevant wireless structure and thus enabling decreased sample complexity for downstream network applications such as QoS prediction. Another benefit is the privacy of RAN information with respect to CORE, and 3GPP compliance in the information exchange between RAN and CORE.

[0226] As the DIE function 210 is located at an OAM function in CORE, DIE 210 communicates the LCNRs via the SBA message bus of the 5G Core (5GC), N1 / N2, llu and Xn interfaces to the respective wireless nodes. HLWS changes are communicated via the llu and Xn interfaces, as well as new interfaces define for RANDAF-UE and RANDAF-gNB pairs. A digital twin service request 906 from NWDAF is sent via a common bus connecting all NFs in CORE, e.g. the SBA message bus of the 5G Core (5GC), and the N2 interface to the RAN entities. Digital twin representations (i.e., GWVRs per node) are sent 907 via the N2 interface and SBA message bus of 5GC to the NWDAF. NWDAF can train QoS prediction models with GWVRs as input. Figure 10 shows the message flow in 3GPP network.

[0227] The message chart 900 illustrates communication between the digital twin entity 110 including SMM 140 and SAE 220 and Die 210 and the network nodes 121a, 121b of the physical wireless network 120 including UE1 to UEn, AN2, AN3 and AN1.

[0228] After initial procedure registration 901 between DIE 210, AN1 and the physical wireless network 120 including UE1 to UEn, AN2 and AN3, a message 902 is sent from DIE 210 to UE1 to communicate local neighborhood construction rules per node type. UE1 configures process 902a and starts measurements 903 with digital twin entity 110. Then, UE1 sends message 904 to SAE 220 to trigger update of topology of local graph per edge type. SAE 220 sends message 905 to SMM 140 to notify updated global graph. Then SMM 140 updates computation graph 905a. After the status has changed 905b in UE1 , e.g., by handover execution, UE1 sends message 905c to SAE 220 to trigger local neighborhood update per edge type. SAE 220 sends message 905d to SMM 140 to notify updated global graph. Then, SMM 140 updates computation graph 905e.

[0229] NF / AF, e.g., NWDAF sends message 906 to SMM 140 for digital twin service request, e.g., what-if network scenario performance. SMM 140 replies with message 907 to communicate general vector representation and optionally output function. Then, NF / AF, e.g. NWDAF can perform AL / ML model training (e.g., QoS prediction) 907a.

[0230] Figure 10 shows an exemplary 3GPP wireless network system 1000 according to the subembodiment of the first embodiment, where parts of the digital twin instance are implemented in a RAN entity and the DIE is implemented at OAM function in the core network.

[0231] The 3GPP wireless network system 1000 includes a control plane 1001 with the entities: AF, AMF, OAM and SMF and a data plane with the entities: AN 1 , UE1 , AN2, UE2, AN3, UE3, UE4, UPF and DN as defined in 3GPP.

[0232] Part of the digital twin entity 110 with SMM 140 and SAE 220 are implemented in RANDAF 1010 of a 3GPP RAN entity 610, here the AN1 entity. The other part of digital twin entity 110 with DIE 210 is implemented in OAM 1020 of control plane 1001 or core network.

[0233] Local neighborhood construction rules (LNCR) per node type as described above with respect to Figures 1 and 2 are delivered via a first path 601 from OAM 1020 to AMF, from AMF to AN1 , AN2 and AN3 and from AN1 to UE1 , AN2 to UE2 and AN3 to UE3 and UE4.

[0234] Heterogeneous local wireless structure (HWLS) updates as described above with respect to Figures 1 and 2 are delivered via a second path 602 from UE4 and UE3 to AN3, f4rom AN3 and UE2 to AN2, from AN2 and UE1 to AN1 and from AN1 to RANDAF 1010.

[0235] General wireless vector representation (GWVR) as described above with respect to Figures 1 and 2 is delivered via a third path 603 from RANDAF 1010 to AN 1 , from AN1 to AMF and from ANF to AF.

[0236] Figure 11 shows an exemplary message chart 1100 illustrating message exchange with SMM / SAE at NWDAF in core network and DIE at OAM in core network according to a second embodiment. This second embodiment is based on a partial implementation of the digital twin instance 110, i.e., the SMM / SAE functions 140, 220 at NWDAF and DIE 210 at OAM function of the 3GPP core network. In this implementation, another entity called Repository, e.g., a Network Repository Function (NRF), collects registration information at 5GC of wireless nodes involved in the digital twinning process as well as to discover other NFs present in 3GPP core network. Involved NF / AFs execute standard procedures for capability and service exposure as per TS. 23.502.

[0237] In this embodiment, a network application (e.g., AI / ML model training) is run by another NF / AF, e.g. NWDAF 2, to train QoS prediction model(s). The QoS prediction model is trained on synthetic data generated by the digital twin instance 110, i.e., GWVRs per wireless node generated by SMM 140.

[0238] As now all the functions associated to the digital twin instance are located in 3GPP Core Network, these functions discover the services and information of other NF / AFs via a Repository, i.e., the NRF.

[0239] Step 1) Wireless nodes involved in digital twinning process register to the NRF.

[0240] Step 1.1) DIE solicits the Repository for wireless node discovery.

[0241] Step 1.2) The Repository responds with a wireless node discovery response including the information relating to the wireless node IDs involved in the digital twinning process, node types, available local measurements per nodes and communication functionalities of wireless nodes.

[0242] Further steps are contained within general procedure as described above with respect to Figures 1 and 2. As the DIE function 210 is located at an OAM function in CORE, DIE 210 communicates the LCNRs via the SBA message bus of the 5G Core (5GC), N1 / N2, llu and Xn interfaces to the respective wireless nodes. HLWS changes are communicated via the llu and Xn interfaces, N1 , SBA message bus of the 5G Core (5GC) to NWDAF 1.

[0243] A digital twin service request from NWDAF 2 is sent via a common bus connecting all NFs in CORE, e.g., the SBA message bus of the 5G Core (5GC) to NWDAF 2. Digital twin representations (i.e., GWVRs per node) are sent via the SBA message bus of 5GC to the other NFs / AF, i.e., NWDAF 2. NWDAF 2 can train QoS prediction models with GWVRs as input. Figure 12 shows the message flow in 3GPP network. The message chart 1100 illustrates communication between the digital twin entity 110, e.g., implemented in NWDAF including SMM 140 and SAE 220 and DIE 210 and Repository (e.g. NRF) and the network nodes 121a, 121 b of the physical wireless network 120 including UE1 to UEn, AN2, AN3 and AN1 and NF / AF (e.g., NWDAF2).

[0244] After initial procedure registration 1101 between Repository, AN1 and the physical wireless network 120 including UE1 to UEn, AN2 and AN3, a message 1101a is sent from DIE to Repository for wireless node discovery. Repository answers with message Wireless Node Response 1101 b. then a message 1102 is sent from DIE 210 to UE1 to communicate local neighborhood construction rules per node type. UE1 configures process 1102a and starts measurements 1103. Then, UE1 sends message 1104 to SAE 220 to trigger update of topology of local graph per edge type. SAE 220 sends message 1105 to SMM 140 to notify updated global graph. Then SMM 140 updates computation graph 1105a.

[0245] After the status has changed 1105b in UE1 , e.g., by handover execution, UE1 sends message 1105c to SAE 220 to trigger local neighborhood update per edge type. SAE 220 sends message 1105d to SMM 140 to notify updated global graph. Then, SMM 140 updates computation graph 1105e.

[0246] NF / AF, e.g., NWDAF2 sends message 1106 to SMM 140 for digital twin service request, e.g., what-if network scenario performance. SMM 140 replies with message 1107 to communicate general vector representation and optionally output function. Then, NF / AF, e.g., NWDAF2 can perform AL / ML model training (e.g., QoS prediction) 1107a.

[0247] Figure 12 shows an exemplary 3GPP wireless network system 1200 according to the second embodiment, where parts of the digital twin instance are implemented in NWDAF and the DIE is implemented at OAM function in the core network.

[0248] The 3GPP wireless network system 1200 includes a control plane 1001 with the entities: AF, AMF, OAM, SMF, NRF and NWDAF and a data plane with the entities: AN1 , UE1 , AN2, UE2, AN3, UE3, UE4, UPF and DN as defined in 3GPP.

[0249] Part of the digital twin entity 110 with SMM 140 and SAE 220 are implemented in NWDAF 1030 in core network. The other part of digital twin entity 110 with DIE 210 is implemented in OAM 1020 of control plane 1001 in core network. Local neighborhood construction rules (LNCR) per node type as described above with respect to Figures 1 and 2 are delivered via a first path 601 from OAM 1020 to AMF, from AMF to AN1 , AN2 and AN3 and from AN1 to UE1 , AN2 to UE2 and AN3 to UE3 and UE4.

[0250] Heterogeneous local wireless structure (HWLS) updates as described above with respect to Figures 1 and 2 are delivered via a second path 602 from UE4 and UE3 to AN3, from UE2 to AN2, from UE1 to AN1 , from AN1 , AN2 and AN3 to AMF and from AMF to NWDAF 1030.

[0251] General wireless vector representation (GWVR) as described above with respect to Figures 1 and 2 is delivered via a third path 603 from NWDAF 1030 to AF.

[0252] Figure 13 shows an exemplary message chart 1300 illustrating message exchange with SMM / SAE / DIE at MEC (mobile edge computing) according to a third embodiment.

[0253] This embodiment is based on a full implementation of the digital twin instance 110, i.e., the SMM / SAE / DIE functions 140, 220, 210, at a MEC server connected to 3GPP system. In this implementation, another entity called Repository, e.g., a Network Repository Function (NRF), collects registration information at 5GC of wireless nodes involved in the digital twinning process as well as to discover other NFs present in 3GPP network. Involved NF / AFs execute standard procedures for capability and service exposure as per TS. 23.502.

[0254] MEC applications, may or may not use the Network Exposure Function (NEF) to access the service of untrusted entities external to the domain.

[0255] In this embodiment, a network application (e.g., AI / ML model training) is run by another NF / AF, e.g. NWDAF 1 , to train QoS prediction model(s). The QoS prediction model is trained on synthetic data generated by the digital twin instance 110, i.e., GWVRs per wireless node generated by SMM. As for current implementation, logically MEC server resides in Data Network (DN).

[0256] As the SMM / SAE / DIE functions are implemented in a MEC server, DIE 210 communicates the LCNRs via the N6, N3, Xn and llu interfaces to the respective wireless nodes. HLWS changes are communicated via the llu and Xn interfaces, N3, N6 Interfaces to MEC server residing in DN. A digital twin service request from NWDAF 1 is sent via a common bus connecting all NFs in CORE, e.g. the SBA message bus of the 5G Core (5GC), N4 and N6 interfaces to MEC server. Digital twin representations (i.e., GWVRs per node) are sent via N4 / N6 interfaces and the SBA message bus of 5GC to the other NFs / AF, i.e. , NWDAF 1. NWDAF 1 can train AI / ML models (e.g., QoS prediction models) with GWVRs as input.

[0257] The message chart 1300 illustrates communication between the digital twin entity 110, e.g., implemented in MEC including SMM 140, SAE 220 and DIE 210 and Repository (e.g. NRF) and the network nodes 121a, 121 b of the physical wireless network 120 including UE1 to UEn, AN2, AN3 and AN1 and NF / AF (e.g., NWDAF1).

[0258] After initial procedure registration 1301 between Repository, AN1 and the physical wireless network 120 including UE1 to UEn, AN2 and AN3, a message 1301a is sent from DIE to Repository for wireless node discovery. Repository answers with message Wireless Node Response 1301 b. then a message 1302 is sent from DIE 210 to UE1 to communicate local neighborhood construction rules per node type. UE1 configures process 1302a and starts measurements 1303. Then, UE1 sends message 1304 to SAE 220 to trigger update of topology of local graph per edge type. SAE 220 sends message 1305 to SMM 140 to notify updated global graph. Then SMM 140 updates computation graph 1305a.

[0259] After the status has changed 1305b in UE1 , e.g., by handover execution, UE1 sends message 1305c to SAE 220 to trigger local neighborhood update per edge type. SAE 220 sends message 1305d to SMM 140 to notify updated global graph. Then, SMM 140 updates computation graph 1305e.

[0260] NF / AF, e.g., NWDAF1 sends message 1306 to SMM 140 for digital twin service request, e.g., what-if network scenario performance. SMM 140 replies with message 1307 to communicate general vector representation and optionally output function. Then, NF / AF, e.g., NWDAF1 can perform AL / ML model training (e.g., QoS prediction) 1307a.

[0261] Figure 14 shows an exemplary 3GPP wireless network system 1400 according to the third embodiment, where the digital twin instance is implemented at MEC.

[0262] The 3GPP wireless network system 1400 includes a control plane 1001 with the entities: AF, AMF, QAM, SMF, NEF, NRF and NWDAF and a data plane with the entities: AN1 , UE1 , AN2, UE2, AN3, UE3, UE4, UPF and DN and MEC 1410 as defined in 3GPP.

[0263] The digital twin entity 110 with SMM 140, SAE 220 and DIE 210 is implemented in MEC server 1410. Local neighborhood construction rules (LNCR) per node type as described above with respect to Figures 1 and 2 are delivered via a first path 601 from MEC server 1410 to UPF, from UPF to AN3, AN2 and AN 1 , from AN3 to UE3 and UE4, from AN2 to UE2 and from AN1 to UE1.

[0264] Heterogeneous local wireless structure (HWLS) updates as described above with respect to Figures 1 and 2 are delivered via a second path 602 from UE4 and UE3 to AN3, from UE2 to AN2, from UE1 to AN1 , from AN1 , AN2 and AN3 to UPF and from UPF to MEC server 1410.

[0265] General wireless vector representation (GWVR) as described above with respect to Figures 1 and 2 is delivered via a third path 603 from MEC server 1410 via UPF and SMF to AF.

[0266] In this disclosure, a logical DIE function or entity 210 is introduced. This enables the definition of set of configurable instructions to collect local structural information as HLWS from wireless nodes.

[0267] A new set of LNCRs per node type as introduced in this disclosure allows to monitor relevant structural changes in physical wireless network per wireless interaction type and notify these HLWS changes to create / adapt digital twin model residing in digital twin instance. Adapting model enables significantly more efficient and more accurate computation to infer network behavior than un-structured models.

[0268] Introduction of a logical SAE function or entity 220, and local and global structures to adapt HGNN model residing in SMM function or entity 140 enables the aggregation of HLWS changes reported by wireless nodes across the physical wireless network to build a HGWS.

[0269] Introduction of messages to adapt digital twin model to changes in wireless network and provide digital representations, i.e. GWVRs, enables the Digital twin instance generate GWVRs per node defined in HGWS given as input sample and communicate it to any network application.

[0270] While a particular feature or aspect of the disclosure may have been disclosed with respect to only one of several implementations, such feature or aspect may be combined with one or more other features or aspects of the other implementations as may be desired and advantageous for any given or particular application. Furthermore, to the extent that the terms "include", "have", "with", or other variants thereof are used in either the detailed description or the claims, such terms are intended to be inclusive in a manner similar to the term "comprise". Also, the terms "exemplary", "for example" and "e.g." are merely meant as an example, rather than the best or optimal. The terms “coupled” and “connected”, along with derivatives may have been used. It should be understood that these terms may have been used to indicate that two elements cooperate or interact with each other, regardless whether they are in direct physical or electrical contact, or they are not in direct contact with each other.

[0271] Although specific aspects have been illustrated and described herein, it will be appreciated by those of ordinary skill in the art that a variety of alternate and / or equivalent implementations may be substituted for the specific aspects shown and described without departing from the scope of the present disclosure. This application is intended to cover any adaptations or variations of the specific aspects discussed herein.

[0272] Although the elements in the following claims are recited in a particular sequence with corresponding labeling, unless the claim recitations otherwise imply a particular sequence for implementing some or all of those elements, those elements are not necessarily intended to be limited to being implemented in that particular sequence.

[0273] Many alternatives, modifications, and variations will be apparent to those skilled in the art in light of the above teachings. Of course, those skilled in the art readily recognize that there are numerous applications of the disclosure beyond those described herein. While the present disclosure has been described with reference to one or more particular embodiments, those skilled in the art recognize that many changes may be made thereto without departing from the scope of the present disclosure. It is therefore to be understood that within the scope of the appended claims and their equivalents, the disclosure may be practiced otherwise than as specifically described herein.

Claims

CLAIMS1. A digital twin entity (110) for replicating a physical wireless network (120) that is representable as a plurality of network nodes (121a, 121 b) and interactions (122) between the network nodes (121a, 121 b) in a heterogeneous structure (123) having nodes and edges of different type, the digital twin entity (110) comprising: an interface (130) with the physical wireless network (120) for receiving information about the network nodes (121a, 121b) and interactions (122) between the network nodes (121a, 121b) of the physical wireless network (120), the information being representable as a plurality of heterogeneous structures (123) derived from measurements of the network nodes (121a, 121 b), each heterogeneous structure (123) representing a structure of the physical wireless network (120) obtained by a respective network node (121a, 121 b); a service mapping model, SMM, subsystem (140) configured to provide a replica of the physical wireless network (120) based on an Artificial Intelligence and / or Machine Learning, AI / ML, model (141), the AI / ML model (141) being trained with the heterogeneous structures (123) derived from the measurements of the network nodes (121a, 121 b); wherein the AI / ML model (141) is configured to generate a digital representation (702) for each network node (121a, 121 b) of the physical wireless network (120).

2. The digital twin entity (110) of claim 1 , wherein the AI / ML model (141) comprises a heterogeneous graph Neural Network, HGNN.

3. The digital twin entity (110) of claim 1 or 2, wherein the AI / ML model (141) is configured to generate a general wireless vector representation (702) for each network node (121a, 121 b) of the physical wireless network (120).

4. The digital twin entity (110) of any of the preceding claims, comprising: a digital twin entity management subsystem (150) configured to monitor performance and resource consumption of the AI / ML model (141).

5. The digital twin entity (110) of any of the preceding claims, comprising: a data repository (160) configured to collect and store the information about the network nodes (121a, 121b) and the interactions (122) between the network nodes (121a, 121b) of the physical wireless network (120) by collecting and updating real-time data of the network nodes (121a, 121b) via the interface (130) with the physical wireless network (120).

6. The digital twin entity (110) of any of the preceding claims, comprising: an interface (170) with a network applications system (180) for receiving digital twin service requests (171) from a plurality of network applications (181, 182) and providing services (172) of the digital twin entity (110) to the plurality of network applications (181, 182).

7. The digital twin entity (110) of claim 6, wherein the digital twin service requests (171) comprise at least one of: request for an identification of at least one network node (121a, 121b) of the physical wireless network (120) for which a digital representation (702) is requested by a network application; request of a time stamp for the digital representation (702); and optional information comprising at least one of a phenomena ID and one or more What-if network configuration IDs; wherein the services (172) of the digital twin entity (110) comprise at least one of: digital representation (702) of the at least one network node (121a, 121b) of the physical wireless network (120) specified in the digital twin service request (171); time stamp for the digital representation (702); and optional information comprising at least one of a representation to phenomena functional mapping as per phenomena ID and an optimal network configuration.

8. The digital twin entity (110) of any of the preceding claims, comprising: a data instruction entity, DIE (210), configured to define a set of configurable rules (211) per node type of the network nodes (121a, 121b);wherein the DIE (210) is configured to register the network nodes (121a, 121b) of the physical wireless network (120) and upon registration communicate the set of configurable rules (211) to the network nodes (121a, 121 b).

9. The digital twin entity (110) of claim 8, wherein the set of configurable rules (211) are local neighborhood construction rules, LNCR, that enable detection of interaction changes between the network nodes (121a, 121b) based on local measurements or local network events of the network nodes (121a, 121 b).

10. The digital twin entity (110) of claim 8 or 9, wherein the DIE (210) is configured to receive local information from the network nodes (121a, 121b) comprising at least one of: an identifier of a respective network node; a node type of the respective network node; available measurements of the respective network node; a communication functionality of the respective network node.

11. The digital twin entity (110) of any of claims 8 to 10, comprising: a structure assemble entity, SAE, (220), configured to build a Heterogeneous Global Wireless data structure, HGWS, that represents a structure of the physical wireless network (120) at a given time instance based on notifications of the network nodes (121a, 121 b) reporting local structural changes of the physical wireless network (120); wherein the SAE (220) is configured to provide the HGWS to the SMM subsystem (140) for adapting the AI / ML model (141) to the local structural changes of the physical wireless network (120).

12. The digital twin entity (110) of claim 11 , wherein the notifications of the network nodes (121a, 121b) reporting the local structural changes of the physical wireless network (120) comprise at least one of the following: a wireless interaction type ID,a source wireless node, a target wireless node, an optional trigger signal that triggered the local structural change report of the network nodes (121a, 121b).

13. The digital twin entity (110) of any of the preceding claims, wherein the digital twin entity (110) is configured for implementation in a 3GPP radio access network, RAN, entity (610) of the physical wireless network (120).

14. The digital twin entity (110) of any of claims 8 to 12, wherein the SMM subsystem (140) and the SAE (220) entity are configured for implementation in a data analytics function component (1010) of a 3GPP radio access network, RAN, entity of the physical wireless network (120); and wherein the DIE entity (210) is configured for implementation in an operation and maintenance, GAM, entity (1020) of a 3GPP core network (1001).

15. The digital twin entity (110) of any of claims 8 to 12, wherein the SMM subsystem (140) and the SAE (220) entity are configured for implementation in a data analytics function component (1030) of a 3GPP core network (1001); and wherein the DIE entity (210) is configured for implementation in an operation and maintenance, GAM, entity (1020) of the 3GPP core network (1001).

16. The digital twin entity (110) of any of claims 8 to 12, wherein the digital twin entity (110) is configured for implementation in a mobile edge computing, MEC, server (1410) connected to the physical wireless network (120).

Images