Acquiring updated values of parameters of a digital twin

Abstract

A method, performed by an update function node of a digital twin, wherein the digital twin is associated with a network, is disclosed. The method comprises receiving, from a first network node, a request indicating that one or more parameters of the digital twin are to be updated; based on one or more parameters that represent a state of the network and / or one or more constraints, determining, for each of the one or more parameters: whether to obtain an updated value of the parameter from a second network node, or whether to generate an updated value of the parameter; for one or more of the one or more parameters: in response to determining to obtain the updated value of the parameter from the second network node, obtaining the updated value of the parameter from the second network node, or in response to determining to generate the updated value of the parameter, generating the updated value of the parameter; and sending, to the first network node, a response to the request, the response comprising the obtained updated values and / or the generated updated values.

Description

[0001] Acquiring updated values of parameters of a digital twin

[0002] Technical Field

[0003] The present disclosure relates to methods performed by an update function node of a digital twin, to a method performed by a first network node, and to a method performed by a second network node. The present disclosure also relates to an update function node, a first network node and a second network node.

[0004] Digital twins (DTs), in the context of mobile networks, are virtual representations of deployed network functionality. This functionality can be contained within one or multiple core network nodes (for example, a network function (NF) node), or can be part of the Radio Access Network (RAN) (for example, radio resource management, beam management, etc.). DTs may also span both a core network and a RAN, and in these cases are known as "end-to-end” DTs.

[0005] DTs are important tools for mobile networks, serving primarily two roles. Firstly, DTs may provide a safe environment in which a network configuration can be tested. Such a "network configuration” may vary from use case to use case. In certain cases, artificial intelligence (Al) algorithm predictions may be involved. For example, in a power saving use case, a network configuration may be a plan to turn on or off one or more radio frequency (RF) ports of a radio unit (RU), based on a traffic demand that has been predicted by an Al algorithm, in order to save power and to serve users with a certain Quality of Service (QoS). In another example, considering a radio resource management (RRM) use case, the network configuration may be the assignment of downlink physical resource blocks (PRBs) to users based on the data traffic that has been predicted to be transmitted to these users. Such a prediction can also be made by an Al algorithm.

[0006] This practice can also be extended to a learning process itself, as well as for inferencing. For example, in reinforcement learning (RL), algorithms may for at least some of the learning duration, engage in so-called "exploration”, in which an "action space” is explored, vis-a-vis network configuration, oftentimes randomly selecting values within that action space during the exploration. A DT offers a safe environment to evaluate the output of these algorithms as they learn, without normal network operation being affected.

[0007] Secondly, DTs may provide an environment for the verification and validation of Al algorithms prior to their deployment. In such an environment, several aspects of a respective algorithm's performance can be evaluated (for example, the accuracy of the predictions, the use of resources, the impact of inferencing time on network operation, etc.). For example, a digital twin may be utilised to provide a safe environment for learning as part of a network slice portioning process. Figure 1 shows an example of a network slice partitioning process.

[0008] With reference to Figure 1 , network slice partitioning is performed on the air interface (that is, the interface between a UE and a radio base station). Every K amount of Transmission Time Intervals (TTIs), which in 5G is 2ms by default, a decision is made by a partitioner on the RU on how to allocate Frequency / Time resources (OFDM spectrum) to different network slices. The decision is based on information relating to the historical use of slices, which is input into a time-series predictor (e.g. a recurrent neural network or a regressor). The time-series predictor then outputs a predicted future use. Because the partitions are measured in Physical Resource Blocks (PRBs), which in turn are bounded by TTIs and frequency, the predictor should provide an output every X amount of TTIs, where can X range from a few TTIs (e.g., 2-3) to several TTIs (e.g., 10-11).

[0009] In the example shown in Figure 1 , the partitioning algorithm works in Decision Time Intervals (DTIs) that span 4 TTIs. In the example shown in Figure 1 , there are three types of slices, two slices of ultra reliable low latency communication (URLLC) of the types URLLC (Streaming) and URLLC (loT), and one best-effort of the type enhanced mobile broadband. In Figure 1 , every DTI (that is, every 4 TTIs), a predictor identifies the needs of the slices, and a partitioning algorithm decides the partition size (e.g., using a fairness-based allocation algorithm). Within the partition, other algorithms may allocate individual PRBs to different UEs.

[0010] The predictor model is typically deployed in a cell in order to provide the partitioning algorithm with predictions of network slice usage. The model may have some baseline weights and would have to be incrementally trained at the point of the deployment, in order to understand the different traffic patterns exhibited by the UE attached to this particular cell, and for the different network slices of the mobile network. Training this model in a real network may cause issues and may negatively impact UE traffic, as the algorithm would need to make predictions on the usage of network slices initially randomly, as it explores a list of viable alternatives until it learns which predictions are optimal. A DT may be used in this use case to provide a safe environment for this training.

[0011] A digital twin may also be utilised to provide a safe environment for learning as part of a beam sweeping operation. In a beam sweeping operation, a beamforming antenna sweeps a sector in order to determine the best beam for a UE to attach to. In this operation, a predictor model can be used in order to minimize the geographical search space, and therefore shorten the time required and reduce the energy resources required for the beam sweep. Figure 2a shows a beam sweeping operation in which a beamforming antenna performs beam sweeping for a whole sector in order to allocate the best beam to a UE. Figure 2b shows a beam sweeping operation in which a beamforming antenna performs beam sweeping for a part of a sector (as recommended by a predictor model) in order to allocate the best beam to a UE. In the example shown in Figure 2b, it will be appreciated that fewer beam candidates (in this example, 3, as opposed to 6) need to be considered during the beam sweeping process, shortening the time required and reducing the energy resources required for the beam sweep compared to the beam sweeping operation shown in Figure 2a.

[0012] As is the case with network slice partitioning, the predictor model in the beam sweeping use case would also need to be trained with local data, as mobility patterns of UEs and geographical features are different in every cell. A DT again may be used in the beam sweeping use case to provide a safe environment for this training.

[0013] DTs may also be useful in use cases where multiple different solutions need to be considered. This could be the case, for example, in autonomous networks and intent-driven architectures. For example, a network management solution is considered, where an intent provided by an enterprise needs to maintain a certain quality of service (QoS) for its own UE (e.g. a certain latency ceiling and a certain minimum acceptable throughput). In this example, several different agents may offer different proposals. For example, one agent may propose to change the policies at the Policy Control Function (PCF) node. Another agent may then propose to offload cells from the UE that do not belong to the enterprise and transfer them to local Wi-Fi small cells, to avoid congestions on the air interface. A third agent may then propose to create new virtual routes for the data packets transmitted toward / sent from the UE on the core network and backhaul, using a combination of routing rules for physical routers and re-configuration of software defined network (SDN) controllers. However, at the same time, the network needs to manage other intents as well, for example, maintaining a certain level of power consumption, but also the QoS for other enterprises that may be sharing the network. Testing the effectiveness of each proposal towards the QoS intent of the enterprise, but also testing whether each proposal violates the other aforementioned intents can be achieved using a digital twin. The digital twin could then test all three proposals, in order to safely pick the most effective proposal which also has the least adverse effect on the other intents.

[0014] It will be appreciated that, for a DT to be utilized effectively, the DT should remain synchronized with the mobile network. In order to achieve this, the DT must continuously observe data from the real network and use this data to update its own internal state. Maintaining synchronization between the digital twin and the real network can be costly, due to the overhead of observing and transporting the network state data. It will be appreciated that an update on the values of parameters of the digital twin will depend on the hardware and the current load of the cell. For example, in certain deployments, updated DT parameters may be aggregated and reported every 15 minutes. In addition, the current load of the network may result in delays in updated DT parameter reporting. If for example, the baseband board, where all the updated DT parameters are aggregated, is overloaded, reporting of these parameters may be delayed. Furthermore, some use cases, such as the network slice partitioning use case described with reference to Fig 1., require more fine-grained updates in much shorter time intervals (e.g. in milliseconds). Other use cases, such as the beam sweeping use case as described with reference to Fig 2. , may not require frequent updates, but may instead require timely updates. For example, for the beam sweeping use case, this could mean that whenever a UE attaches, the DT must synchronize in order to investigate beam selection strategies. This should happen very quickly (in the order of ms), in order for the process to have any benefit over the full beam sweep. However, given the large parameter update intervals, it would be highly unlikely that this synchronization would happen within the required time interval for the process to be beneficial.

[0015] There may also be use cases where a large amount of data is required for a DT update. Considering the proposal agents use case described above, the number of parameters needed would be in the order of thousands, including up to date synchronization of core network nodes such as the policy control function (PCF), the user plane function (UPF), the SDN controller configuration, the physical router configuration, power consumption performance monitor (PM) counters, etc. This could result in a DT update being several megabytes long.

[0016] Therefore, there currently exist challenges relating to the provision of updated parameters for a digital twin, in order to enable the digital twin to remain synchronized with the network.

[0017] It is an aim of the present disclosure to provide methods, an update function node, a first network node, a second network node and a computer readable medium which at least partially address one or more of the challenges discussed above.

[0018] According to a first aspect of the present disclosure, there is provided a method performed by an update function node of a digital twin, wherein the digital twin is associated with a network. The method comprises receiving, from a first network node, a request indicating that one or more parameters of the digital twin are to be updated; based on one or more parameters that represent a state of the network and / or one or more constraints , determining, for each of the one or more parameters: whether to obtain an updated value of the parameter from a second network node, or whether to generate an updated value of the parameter using a generative model; for one or more of the one or more parameters: in response to determining to obtain the updated value of the parameter from the second network node, obtaining the updated value of the parameter from the second network node, or in response to determining to generate the updated value of the parameter using a generative model, generating the updated value of the parameter using a generative model; and sending, to the first network node, a response to the request, the response comprising the obtained updated values and / or the generated updated values.

[0019] According to another aspect of the present disclosure, there is provided a method, performed by an update function node of a digital twin, of training a first machine learning model to generate updated values of parameters of a digital twin, wherein the digital twin is associated with a network. The method comprises sending, to a second network node, a request for one or more parameters that represent a state of the network and one or more updated values of parameters of the digital twin; receiving, from the second network node, one or more parameters that represent a state of the network and one or more updated values of parameters of the digital twin; determining whether the received parameters and the received updated values were received within a particular time period; in response to determining that the received parameters and the received updated values were received within the particular time period, forming a training data set comprising the received parameters and the received updated values; and using the training dataset to train the first machine learning model to generate updated values of parameters of the digital twin based on one or more parameters that represent a state of the network.

[0020] According to another aspect of the present disclosure, there is provided a method performed by a first network node. The method comprises sending, to an update function node of a digital twin, a request indicating that one or more parameters of the digital twin are to be updated, wherein the digital twin is associated with a network; and receiving, from the update function node, a response to the request, the response comprising one or more updated values of parameters of the digital twin that have been obtained from a second network node, and / or one or more updated values of parameters of the digital twin that have been generated using a generative modelby the update function node.

[0021] According to another aspect of the present disclosure, there is provided a method performed by a second network node. The method comprises providing, to an update function node of a digital twin, one or more updated values of parameters of the digital twin, wherein the digital twin is associated with a network.

[0022] According to another aspect of the present disclosure, there is provided a computer program product comprising a computer readable medium, the computer readable medium having computer readable code embodied therein, the computer readable code being configured such that, on execution by a suitable computer or processor, the computer or processor is caused to perform a method according to any one of the aspects or examples of the present disclosure.

[0023] According to another aspect of the present disclosure, there is provided an update function node of a digital twin associated with a network. The update function node is configured to: receive, from a first network node, a request indicating that one or more parameters of the digital twin are to be updated; based on one or more parameters that represent a state of the network and / or one or more constraints, determine, for each of the one or more parameters: whether to obtain an updated value of the parameter from a second network node, or whether to generate an updated value of the parameter using a generative model; for one or more of the one or more parameters: in response to determining to obtain the updated value of the parameter from the second network node, obtain the updated value of the parameter from the second network node, or in response to determining to generate the updated value of the parameter using a generative model, generate the updated value of the parameter using a generative model; and send, to the first network node, a response to the request, the response comprising the obtained updated values and / or the generated updated values.

[0024] According to another aspect of the present disclosure, there is provided an update function node of a digital twin associated with a network. The update function node comprises a processor and a memory, said memory containing instructions executable by said processor, whereby said update function node is operative to: receive, from a first network node, a request indicating that one or more parameters of the digital twin are to be updated; based on one or more parameters that represent a state of the network and / or one or more constraints, determine, for each of the one or more parameters: whether to obtain an updated value of the parameter from a second network node, or whether to generate an updated value of the parameter using a generative model; for one or more of the one or more parameters: in response to determining to obtain the updated value of the parameter from the second network node, obtain the updated value of the parameter from the second network node, or in response to determining to generate the updated value of the parameter using a generative model, generate the updated value of the parameter using a generative model; and send, to the first network node, a response to the request, the response comprising the obtained updated values and / or the generated updated values.

[0025] According to another aspect of the present disclosure, there is provided an update function node of a digital twin associated with a network. The update function node is configured to: send, to a second network node, a request for one or more parameters that represent a state of the network and one or more updated values of parameters of the digital twin; receive, from the second network node, one or more parameters that represent a state of the network and one or more updated values of parameters of the digital twin; determine whether the received parameters and the received updated values were received within a particular time period; in response to determining that the received parameters and the received updated values were received within the particular time period, form a training data set comprising the received parameters and the received updated values; and use the training dataset to train the first machine learning model to generate updated values of parameters of the digital twin based on one or more parameters that represent a state of the network.

[0026] According to another aspect of the present disclosure, there is provided an update function node of a digital twin associated with a network. The update function node comprises a processor and a memory, said memory containing instructions executable by said processor, whereby said update function node is operative to: send, to a second network node, a request for one or more parameters that represent a state of the network and one or more updated values of parameters of the digital twin; receive, from the second network node, one or more parameters that represent a state of the network and one or more updated values of parameters of the digital twin; determine whether the received parameters and the received updated values were received within a particular time period; in response to determining that the received parameters and the received updated values were received within the particular time period, form a training data set comprising the received parameters and the received updated values; and use the training dataset to train the first machine learning model to generate updated values of parameters of the digital twin based on one or more parameters that represent a state of the network.

[0027] According to another aspect of the present disclosure, there is provided a first network node. The first network node is configured to: send, to an update function node of a digital twin, a request indicating that one or more parameters of the digital twin are to be updated, wherein the digital twin is associated with a network; and receive, from the update function node, a response to the request, the response comprising one or more updated values of parameters of the digital twin that have been obtained from a second network node, and / or one or more updated values of parameters of the digital twin that have been generated by the update function node.

[0028] According to another aspect of the present disclosure, there is provided a first network node. The first network node comprises a processor and a memory, said memory containing instructions executable by said processor, whereby said second network node is operative to: send, to an update function node of a digital twin, a request indicating that one or more parameters of the digital twin are to be updated, wherein the digital twin is associated with a network; and receive, from the update function node, a response to the request, the response comprising one or more updated values of parameters of the digital twin that have been obtained from a second network node, and / or one or more updated values of parameters of the digital twin that have been generated by the update function node. According to another aspect of the present disclosure, there is provided a second network node. The second network node is configured to: provide, to an update function node of a digital twin, one or more updated values of parameters of the digital twin, wherein the digital twin is associated with a network.

[0029] According to another aspect of the present disclosure, there is provided a second network node. The second network node comprises a processor and a memory, said memory containing instructions executable by said processor, whereby said second network node is operative to: provide, to an update function node of a digital twin, one or more updated values of parameters of the digital twin, wherein the digital twin is associated with a network.

[0030] Brief Description of the Figures

[0031] For a better understanding of the present disclosure, and to show more clearly how it may be carried into effect, reference will now be made, by way of example, to the following drawings in which:

[0032] Figure 1 shows an example of a network slice partitioning process;

[0033] Figure 2a shows a beam sweeping operation;

[0034] Figure 2b shows a beam sweeping operation;

[0035] Figure 3 illustrates an example architecture;

[0036] Figure 4 is a flow chart illustrating process steps in a method performed by an update function node of a digital twin;

[0037] Figure 5 is a flow chart illustrating process steps in a method performed by an update function node of a digital twin, of training a first machine learning model to generate updated values of parameters of a digital twin;

[0038] Figure 6 is a flow chart illustrating process steps in a method performed by a first network node;

[0039] Figure 7 is a flow chart illustrating process steps in a method performed by a second network node;

[0040] Figure 8 is a block diagram illustrating functional modules in an example of an update function node;

[0041] Figure 9 is a block diagram illustrating functional modules in an example of a first network node; Figure 10 is a block diagram illustrating functional modules in an example of a second network node;

[0042] Figure 11 shows an example signalling exchange;

[0043] Figure 12 shows another example signalling exchange; and

[0044] Figure 13 shows yet another example signalling exchange.

[0045] Detailed Description

[0046] Aspects and examples of the present disclosure thus propose an update function node that determines whether to, and how to, "split” between obtaining updated values of parameters of a digital twin from a network and generating synthetic updated values of parameters of a digital twin, given factors such as the state of the network and constraints relating to the acquisition the updated values. The update function node may also then generate synthetic updated values of parameters of a digital twin using machine-learning techniques.

[0047] In certain embodiments of the present disclosure, given an update request for a digital twin that may be constrained by a response time, an acceptable report accuracy and / or other parameters, and based on the state of the real network in terms of observability capability, the update function node may choose to either produce some or all of the updated values of parameters of the digital twin using machine-learning techniques, while retrieving the rest of the updated values, if any, from the real network.

[0048] Figure 3 illustrates an example architecture 300 for implementing methods according to the present disclosure. The example architecture 300 comprises a first network node (or a third party node) 302, an update function node (or a digital update function node) 304 for a digital twin, and a second network node (or a network data collection function node) 306.

[0049] The first network node 302 is an entity that initiates an update request for the digital twin. The first network node 302 may comprise the digital twin itself, or alternatively may comprise a management function of the digital twin.

[0050] The update function node 304 is an entity that determines which updated parameter values of the digital twin should be collected from the real network, and which updated parameter values of the digital twin should be generated using a first machine learning model. This determination may be performed by a decision function 308 of the update function node 304. The decision function 308 may be a logical component of the update function node 304.

[0051] The update function node 304 is also responsible for generating synthetic updated parameter values of the digital twin using the first machine learning model. The first machine learning model may generate these synthetic updated parameter values based on a current or predicted network state. The first machine learning model may have been trained to learn probability distribution functions of update parameter values of the digital twin.

[0052] The update function node 304 is also responsible training the first machine learning model, and is also responsible for training a second machine learning model to predict a future network state, based on a current network state. The update function node 304 is also responsible for using the second machine learning model to predict a future network state. This training may be performed by a learning function 310 of the update function node 304. The learning function 310 may be a logical component of the update function node 304.

[0053] In some embodiments, particularly for use cases which involve a RAN, the update function node 304 may be an xAPP or an rAPP, residing in the non-Real Time Radio Intelligence Controller (non-RT RIC) of Open RAN (O-RAN) architecture. rAPPs have low duration control loops (between 10ms and 1 sec) and may be utilised in the use case described with reference to Figure 1. xApps have higher duration control loops (greater than a second) and may be utilised in the use case described with reference to Figure 2. When more complex digital twins involving multiple network nodes beyond a RAN are involved, such as the proposal agent use case described above, the update function node 304 may be a core network function node such as a network data and analytics function (NWDAF) node. In this case, the digital twin update may be a service provided by the NWDAF node, which can be either provided to external entities via a Network Exposure Function (NEF) node, or internally to other network function (NF) nodes.

[0054] The second network node 306 may be any node in a network that is used for collecting data such as updated digital twin parameters or parameters representing a network state. It will be appreciated that the second network node 306 may be a logical entity, which may reside in different physical locations depending on the type of use case the digital twin is deployed for. For example, for RAN type of use cases such as the ones described with reference to Figs. 1 and 2, the second network node 306 can be located in the baseband board. In other examples, such as the autonomous network use case described above, the second network node 306 may be a logical component of an operations support system (OSS) node. A first signalling exchange performed in an operational phase of the update function node 304, between the first network node 302, the update function node 304 and the second network node 306, is illustrated in Figure 3. A second signalling exchange performed in a learning phase of the update function node 304, between the update function node 304 and the second network node 306, is also illustrated in Figure 3. It will be appreciated that elements of the learning phase may also be executed during the operational phase, and the signalling exchange performed during in the learning phase of the update function node 304 may also be executed as a background task during the operational phase of the update function node 304.

[0055] The first signalling exchange is as follows:

[0056] At step 312, the first network node 302, sends, to the update function node 304, a request indicating that one or more parameters of the digital twin are to be updated. The request may specify the one or more parameters, or alternatively, may comprise an identifier of the digital twin. The identifier of the digital twin may then be resolved by the update function node 304, or the second network node 306, in order to determine the one or more parameters. The request may also comprise one or more constraints relating to the acquisition of the updated parameter values for the digital twin. These constraints may comprise, for example, a time period in which the update function node 304 is to provide a response to the request, and / or a constraint on the quality of the updated parameter values that are to be provided in a response to the request.

[0057] At step 314, the update function node 304 determines, a first set of the one or more parameters for which updated values are to be collected from the network and a second set of the one or more parameters for which updated values are to be generated using the first machine learning model. This determination (of which updated parameter values of the digital twin should be collected from the real network, and which updated parameter values of the digital twin should be generated using a first machine learning model) may be based on a current network state and / or a future network state. The future network state may be predicted using a second machine learning model at the update function node 304. Additionally or alternatively, this determination may be based on the one or more constraints comprised in the request received from the first network node 302.

[0058] This determination, when based on a state of the network, enables updated parameter values for the digital twin to be acquired in a manner that ensures normal network operation, and enables the digital twin to remain synchronised with the network regardless of whether the updated values can be obtained from the real network or not, as will be described in greater detail below. At step 316, the update function node 304 obtains updated values of the first set of the one or more parameters from the second network node 306, and generates updated values of the second set of the one or more parameters using the first machine learning model.

[0059] At step 318, the update function node 304 sends, to the first network node 302, a response to the request comprising the obtained updated values and the generated updated values. The first network node 302 may then use these updated values to update the state of the digital twin, thus enabling the digital twin to remain synchronised with the network.

[0060] The second signalling exchange is as follows:

[0061] At step 320, the update function node 304 obtains parameters that represent a network state, and updated values of parameters of the digital twin, from the second network node 306. This data may be collected from the second network node 306 as the data becomes available, or alternatively, is collected when said collection does not compromise network functionality. The parameters that represent the network state may indicate the overall status of the network in terms of load, health, and energy reserves.

[0062] The update function node 304 then trains a second machine-learning model for predicting a future state of the network, based on a current state of the network, using the obtained parameters that represent the network state. The second machine-learning model may use a time-series algorithm such as a recurrent neural network or a regressor, for example. The update function node 304 then also trains a first machine learning model to generate updated values of parameters of the digital twin, based on a network state. The first machine learning model may be a transformer, which generates updated values of parameters of the digital twin in response to receiving a network state or a cell state as a prompt.

[0063] An overview of methods which may be performed according to different examples of the present disclosure are now described with reference to Figures 4-7. The methods described with reference to Figures 4 and 5 may be performed by the update function node 304. The method described with reference to Figure 6 may be performed by the first network node 302. The method described with reference to Figure 7 may be performed by the second network node 306.

[0064] Figure 4 is a flow chart illustrating process steps in a method 400 performed by an update function node of a digital twin. The digital twin is associated with a network. The update function node may be a non-Real Time Radio Intelligence Controller, or may be a Network Data and Analytics Function node. In step 402, the method 400 comprises receiving, from a first network node, a request indicating that one or more parameters of the digital twin are to be updated. The first network node may comprise the digital twin, or may comprise a management function of the digital twin.

[0065] The request may comprise an identifier of the digital twin. The request may further identify the one or more parameters. In some embodiments, the method 400 may further comprise determining, based on the identifier of the digital twin, the one or more parameters. The request may additionally or alternatively identify the one or more constraints described with reference to step 404.

[0066] The one or more constraints may comprise at least one of: a threshold value to be exceeded by a confidence level associated with the first machine learning model, a time period in which the update function node is to provide, to the first network node, the response to the request, a proportion of obtained updated values and a proportion of generated updated values to be provided in the response to the request, a validity period that is to apply to the updated values provided in the response to the request, and a maximum throughput that is to apply to providing the response to the request.

[0067] In step 404, based on one or more parameters that represent a state of the network and / or one or more constraints, the method 400 comprises determining, for each of the one or more parameters: cwhether to obtain an updated value of the parameter from a second network node, or whether to generate an updated value of the parameter.

[0068] The second network node may comprise a network data collector function.

[0069] In some embodiments, the step of determining comprises, for each of the one or more parameters: determining, based on the one or more parameters that represent the state of the network, whether the second network node is able to provide an updated value of the parameter; in response to determining that the second network node is able to provide the updated value of the parameter, determining to obtain the updated value of the parameter from the second network node; and in response to determining the second network node is not able to provide the updated value of the parameter, determining to generate the updated value of the parameter.

[0070] That is, in some embodiments, the update function node may determine to obtain as many updated parameter values from the network as the network is able to provide. The update function node may then determine to generate the remaining updated parameter values. The step of generating the updated value of the parameter may comprise generating the updated value of the parameter using a first machine learning model. The first machine learning model may be a generative model (for example, a generative adversarial network (GAN), a transformer, or a variational Autoencoder (VAE)), or may be a discriminative model (for example, a regression-based model or a recurrent neural network).

[0071] In some embodiments, in which the one or more constraints comprise the threshold value, the step of determining may comprise: in response to the confidence level associated with the first machine learning model exceeding the threshold value, determining, for each of the one or more parameters, to generate an updated value of the parameter using the first machine learning model; and in response to a confidence level associated with the first machine learning model failing to exceed the threshold value, determining, for each of the one or more parameters, to obtain an updated value of the parameter from the second network node.

[0072] That is, in some embodiments, the update function node may determine to generate all of the updated parameter values when the first machine learning model is capable of generating updated parameter values of an acceptable quality.

[0073] In some embodiments in which the one or more constraints comprise the time period, the step of determining may comprise, for each of the one or more parameters: determining whether an updated value of the parameter can be obtained from the second network node within the time period; determining, based on the one or more parameters that represent the state of the network, whether the second network node is able to provide an updated value of the parameter; in response to determining that the updated value of the parameter can be obtained from the second network node within the time period and in response to determining that the second network node is able to provide the updated value of the parameter, determining to obtain the updated value of the parameter from the second network node; and in response to determining that the updated value of the parameter cannot be obtained from the second network node within the time period, or in response to determining the second network node is not able to provide the updated value of the parameter, determining to generate the updated value of the parameter using the first machine learning model.

[0074] That is, in some embodiments, the update function node may determine to obtain as many updated parameter values from the network as the network is able to provide, and that can be obtained within the time period in which the update function node is to provide as response to the request. The update function node may then determine to generate the remaining updated parameter values. The first machine learning model may have been trained to generate updated values of parameters of the digital twin based on one or more parameters that represent the state of the network.

[0075] The one or more parameters that represent the state of the network may comprise at least one of: one or more parameters that represent a predicted future state of the network, and one or more parameters that represent a current state of the network.

[0076] The method 400 may further comprise using a second machine learning model to generate the one or more parameters that represent the predicted future state of the network. The second machine learning model may have been trained to generate, based on the one or more parameters that represent the current state of the network, the one or more parameters that represent the predicted future state of the network.

[0077] The one or more parameters that represent the state of the network may describe at least one of: a load of the network, a configuration of the network, a radio environment of the network, and a physical environment of the network.

[0078] In step 406, for one or more of the one or more parameters, the method 400 comprises: in response to determining to obtain the updated value of the parameter from the second network node, obtaining the updated value of the parameter from the second network node, or in response to determining to generate the updated value of the parameter using a first machine learning model, generating the updated value of the parameter using a first machine learning model.

[0079] In step 408, the method 400 comprises sending, to the first network node, a response to the request, the response comprising the obtained updated values and / or the generated updated values. The response to the request may further comprise an indication of the proportion of obtained updated values and the proportion of generated updated values provided in the response.

[0080] Figure 5 is a flow chart illustrating process steps in a method 500 performed by an update function node of a digital twin, of training a first machine learning model to generate updated values of parameters of a digital twin. The digital twin is associated with a network. The update function node may be a non-Real Time Radio Intelligence Controller, or may be a Network Data and Analytics Function node. The first machine learning model may be a generative model (for example, a generative adversarial network (GAN), a transformer, a variational Autoencoder (VAE)), or may be a discriminative model (for example, a regression-based model or a recurrent neural network). In step 502, the method 500 comprises sending, to a second network node, a request for one or more parameters that represent a state of the network and one or more updated values of parameters of the digital twin. The second network node may comprise a network data collector function. The one or more parameters that represent a state of the network may describe at least one of: a load of the network, a configuration of the network, a radio environment of the network, or a physical environment of the network.

[0081] In step 504, the method 500 comprises receiving, from the second network node, one or more parameters that represent a state of the network and one or more updated values of parameters of the digital twin.

[0082] In step 506, the method 500 comprises determining whether the received parameters and the received updated values were received within a particular time period. This determines whether the data has been obtained in a timely manner.

[0083] In step 508, in response to determining that the received parameters and the received updated values were received within the particular time period, the method 500 comprises forming a training data set comprising the received parameters and the received updated values. The training dataset may comprise a set of input training data comprising the received parameters, and a set of output training data comprising the received updated values.

[0084] In step 510, the method 500 comprises using the training dataset to train the first machine learning model to generate updated values of parameters of the digital twin based on one or more parameters that represent a state of the network.

[0085] In some embodiments, the method 500 may further comprise generating an updated value of a parameter of the digital twin using the first machine learning model.

[0086] Figure 6 is a flow chart illustrating process steps in a method 600 performed by a first network node. The first network node may comprise the digital twin, or may comprise a management function of the digital twin.

[0087] In step 602, the method 600 comprises sending, to an update function node of a digital twin, a request indicating that one or more parameters of the digital twin are to be updated. The digital twin is associated with a network. The update function node may be a non-Real Time Radio Intelligence Controller, or may be a Network Data and Analytics Function node. The request may comprise an identifier of the digital twin. Additionally or alternatively, the request may identify the one or more parameters. Additionally or alternatively, the request may identify one or more constraints. The one or more constraints may comprise at least one of: a threshold value to be exceeded by a confidence level associated with the first machine learning model, a time period in which the update function node is to provide, to the first network node, the response to the request, a proportion of obtained updated values and a proportion of generated updated values to be provided in the response to the request, a validity period that is to apply to the updated values provided in the response to the request, and a maximum throughput that is to apply to providing the response to the request.

[0088] In step 604, the method 600 comprises, receiving, from the update function node, a response to the request, the response comprising one or more updated values of parameters of the digital twin that have been obtained from a second network node, and / or one or more updated values of parameters of the digital twin that have been generated by the update function node. The response to the request may further comprise an indication of the proportion of obtained updated values and the proportion of generated updated values provided in the response. The second network node may comprise a network data collector function.

[0089] The one or more updated values of parameters of the digital twin that have been generated by the update function node may have been generated using a first machine learning model. The first machine learning model may be a generative model (for example, a generative adversarial network (GAN), a transformer, a variational Autoencoder (VAE)), or may be a discriminative model (for example, a regression-based model or a recurrent neural network).

[0090] Figure 7 is a flow chart illustrating process steps in a method 700 performed by a second network node. The second network node may comprise a network data collector function.

[0091] In step 702, the method 700 comprises providing, to an update function node of a digital twin, one or more updated values of parameters of the digital twin. The digital twin is associated with a network. The update function node may be a non-Real Time Radio Intelligence Controller, or may be a Network Data and Analytics Function node.

[0092] In some embodiments, the method 700 may further comprise providing, to the update function node, one or more parameters that represent a state of the network.

[0093] As discussed above, the methods 400 and 500 may be performed by an update function node, and the present disclosure provides an update function node that is adapted to perform any or all of the steps of methods 400 and 500. The update function node may be a physical or virtual node, and may for example comprise a virtualised function that is running in a cloud, edge cloud or fog deployment. The update function node may for example comprise or be instantiated in any part of a logical core network node, network management centre, network operations centre, radio access node, etc. Any such communication network node may itself be divided between several logical and / or physical functions.

[0094] Figure 8 is a block diagram illustrating an example update function node 800 which may implement the method 400 and / or 500, as illustrated in Figures 4 and 5, according to examples of the present disclosure, for example on receipt of suitable instructions from a computer program 850. Referring to Figure 8, the update function node 800 comprises a processor or processing circuitry 802, and may comprise a memory 804 and interfaces 806. The processing circuitry 802 is operable to perform some or all of the steps of the method 400 and / or 500 as discussed above with reference to Figures 4 and 5. The memory 804 may contain instructions executable by the processing circuitry 802 such that the update function node 800 is operable to perform some or all of the steps of the method 400 and / or 500, as illustrated in Figures 4 and 5. The instructions may also include instructions for executing one or more telecommunications and / or data communications protocols. The instructions may be stored in the form of the computer program 850. In some examples, the processor or processing circuitry 802 may include one or more microprocessors or microcontrollers, as well as other digital hardware, which may include digital signal processors (DSPs), special-purpose digital logic, etc. The processor or processing circuitry 802 may be implemented by any type of integrated circuit, such as an Application Specific Integrated Circuit (ASIC), Field Programmable Gate Array (FPGA), etc. The memory 804 may include one or several types of memory suitable for the processor, such as read-only memory (ROM), random-access memory, cache memory, flash memory devices, optical storage devices, solid state disk, hard disk drive, etc.

[0095] As discussed above, the method 600 may be performed by a first network node, and the present disclosure provides a first network node that is adapted to perform any or all of the steps of the method 600. The first network node may be a physical or virtual node, and may for example comprise a virtualised function that is running in a cloud, edge cloud or fog deployment. The first network node may for example comprise or be instantiated in any part of a logical core network node, network management centre, network operations centre, radio access node, etc. Any such communication network node may itself be divided between several logical and / or physical functions.

[0096] Figure 9 is a block diagram illustrating an example first network node 900 which may implement the method 600 as illustrated in Figure 6, according to examples of the present disclosure, for example on receipt of suitable instructions from a computer program 950. Referring to Figure 9, the first network node 900 comprises a processor or processing circuitry 902, and may comprise a memory 904 and interfaces 906. The processing circuitry 902 is operable to perform some or all of the steps of the method 600 as discussed above with reference to Figure 6. The memory 904 may contain instructions executable by the processing circuitry 902 such that the first network node 900 is operable to perform some or all of the steps of the method 600, as illustrated in Figure 6. The instructions may also include instructions for executing one or more telecommunications and / or data communications protocols. The instructions may be stored in the form of the computer program 950. In some examples, the processor or processing circuitry 902 may include one or more microprocessors or microcontrollers, as well as other digital hardware, which may include digital signal processors (DSPs), special-purpose digital logic, etc. The processor or processing circuitry 902 may be implemented by any type of integrated circuit, such as an Application Specific Integrated Circuit (ASIC), Field Programmable Gate Array (FPGA), etc. The memory 904 may include one or several types of memory suitable for the processor, such as read-only memory (ROM), random-access memory, cache memory, flash memory devices, optical storage devices, solid state disk, hard disk drive, etc.

[0097] As discussed above, the method 700 may be performed by a second network node, and the present disclosure provides a second network node that is adapted to perform any or all of the steps of the method 700. The second network node may be a physical or virtual node, and may for example comprise a virtualised function that is running in a cloud, edge cloud or fog deployment. The second network node may for example comprise or be instantiated in any part of a logical core network node, network management centre, network operations centre, radio access node, etc. Any such communication network node may itself be divided between several logical and / or physical functions.

[0098] Figure 10 is a block diagram illustrating an example second network node 1000 which may implement the method 700 as illustrated in Figure 7, according to examples of the present disclosure, for example on receipt of suitable instructions from a computer program 1050. Referring to Figure 10, the second network node 1000 comprises a processor or processing circuitry 1002, and may comprise a memory 1004 and interfaces 1006. The processing circuitry 1002 is operable to perform some or all of the steps of the method 700 as discussed above with reference to Figure 7. The memory 1004 may contain instructions executable by the processing circuitry 1002 such that the second network node 1000 is operable to perform some or all of the steps of the method 700, as illustrated in Figure 7. The instructions may also include instructions for executing one or more telecommunications and / or data communications protocols. The instructions may be stored in the form of the computer program 1050. In some examples, the processor or processing circuitry 1002 may include one or more microprocessors or microcontrollers, as well as other digital hardware, which may include digital signal processors (DSPs), special-purpose digital logic, etc. The processor or processing circuitry 1002 may be implemented by any type of integrated circuit, such as an Application Specific Integrated Circuit (ASIC), Field Programmable Gate Array (FPGA), etc. The memory 1004 may include one or several types of memory suitable for the processor, such as read-only memory (ROM), random-access memory, cache memory, flash memory devices, optical storage devices, solid state disk, hard disk drive, etc. Figures 4 to 7 discussed above provide an overview of methods which may be performed according to different examples of the present disclosure. These methods may be performed by a first network node, an update function node and / or a second network node as illustrated in Figures 8 to 10, and enable updated parameter values for a digital twin to be acquired in a manner that preserves normal network operation and that allows a digital twin to remain synchronised with the network regardless of whether the updated parameter values could be provided by the network or not.

[0099] There now follows a detailed discussion of how different process steps illustrated in Figures 4 to 7 and discussed above may be implemented, for example by an update function node 800, a first network node 900, and / or a second network node 1000.

[0100] Figure 11 is an example signalling exchange between an update function node and a second network node for training a second machine-learning model to predict a future state of the network. As noted above, the second machine-learning model may be a time-series forecaster, such as an RNN or a regressor, for example.

[0101] It will be appreciated that different conditions may trigger the training of the second machine-learning model. The triggering conditions may be time bounded (for example, periodical and / or event based). For example, the training of the second machine learning model may be triggered upon detection of a degradation of the performance of the second machine learning model.

[0102] At step 1102, the update function node sends, to the second network node, a request for parameters that represent the network state.

[0103] The parameters that represent the network state may vary depending on the use case (for example, the type of network, or the type of digital twin).

[0104] In some embodiments, the parameters may represent the load of the network. For example, in a radio access network, these parameters may indicate the CPU usage of the baseband board, the power / energy consumption in the network, the number of active UEs, the aggregate throughput, etc. In some embodiments, the parameters may represent the configuration of the network. For example, these parameters may indicate semi-static data such as the model and software revision of the nodes in the network, configuration data such as the spectrum range, the number of bands, the azimuth, and the tilt of the antennas in the network, traffic routes, etc. In some embodiments, the parameters may represent a characterization of the environment that can affect network operation. For example, these parameters may indicate interference in wireless channels in a RAN based on measurement reports from UEs, or a status of a cloud infrastructure in a core network, such as compute node load.

[0105] At step 1104, the update function node receives, from the second network node, parameters that represent the network state.

[0106] At step 1106, the update function node characterizes the received parameters.

[0107] The update function node splits the received parameters into a set of input parameters, and a set of output parameters, depending on the time at which the parameters were received. For example, the received parameters may be split such that the set of timestamped input parameters would produce the timestamped output parameters as an output.

[0108] The update function node may also analyse the received parameters to determine a measure of completeness of the received parameters. The measure of completeness may indicate how many of the requested network parameters have been received. Additionally or alternatively, the update function node may analyse the received parameters to determine a time period over which the parameters were received.

[0109] At step 1108, the update function node stores the set of input parameters and the set of output parameters. The set of input parameters and the set of output parameters may be stored in a temporary buffer at the update function node.

[0110] In response to a criterion being met (for example, in response to a particular time period expiring, or in response to a particular number of received parameters being stored), at step 1110, the update function node retrieves the set of input parameters and the set of output parameters. At step 1112, the update function node trains a second machine-learning model to predict a future network state, based on a current network state, using the set of input parameters and the set of output parameters.

[0111] Figure 12 is an example signalling exchange between an update function node and a second network node for training a first machine learning model to generate updated values of parameters of a digital twin. The first machine learning model may be a transformer, or use another generative algorithm such as a Variational Autoencoder or a RNN.

[0112] At step 1202, the update function node sends, to the second network node, a request for parameters that represent the network state and for updated parameter values for a digital twin. At step 1204, the update function node receives, from the second network node, parameters that represent the network state and updated parameter values for a digital twin.

[0113] At step 1206, the update function node analyses the received parameters to determine a measure of completeness of the received parameters. The measure of completeness may indicate how many of the requested network parameters have been received. The update function node also analyses the received parameters to determine a time period over which the parameters were received.

[0114] At step 1208, in response to the measure of completeness meeting a completeness criterion, and in response to the obtained parameters having been obtained within a particular time period, the update function node splits the obtained parameters into a prompt tuple and an output tuple. The prompt tuple is a vector of values, and the output tuple is also a vector of values that is generated based on the input tuple.

[0115] At step 1210, the update function node stores the input tuple and the output tuple. The input tuple and the output tuple may be stored in a temporary buffer at the update function node.

[0116] That is, in step 1210, only datasets that are complete and timely (in which all the parameters have been collected within a certain time) are stored as training data for the first machine learning model. In response to a criterion being met (for example, in response to a particular time period expiring or in response to a particular number of received parameters being stored), at step 1212, the update function node retrieves the input tuple and the output tuple. At step 1214, the update function node trains a first machine learning model to generate updated values of parameters of a digital twin, using the input tuple and the output tuple.

[0117] Figure 13 is an example signalling exchange between a first network node, an update function node and a second network node for acquiring updated parameter values for a digital twin.

[0118] In step 1301 , the first network node sends, to an update function node of a digital twin, an update request for the digital twin. The update request may explicitly define which parameters of the digital twin are to be updated. Additionally or alternatively, the update request may comprise an identifier of the digital twin. The identifier of the digital twin may identify the type of the digital twin, for example.

[0119] The update request for the digital twin may optionally specify one or more constraints relating to the acquisition of updated parameter values. For example, the update request may specify a time period within which the update function node is to provide a response to the request. It will be appreciated that this may impact how the update function node determines which updated parameter values of the digital twin should be collected from the real network, and which updated parameter values of the digital twin should be generated using a first machine learning model. For example, if the update function node may determine to only obtain updated parameter values from the network that can be obtained within the time period, and then determine to generate the remaining updated parameters using the first machine learning model.

[0120] Additionally or alternatively, the update request may specify a minimum confidence level associated with the first machine learning model. This may also impact the determination made by the update function node described above. For example, if the confidence level associated with the first machine learning model does not exceed the minimum confidence level specified in the update request, the update function node may determine to obtain all of the updated parameter values for the digital twin from the network.

[0121] Additionally or alternatively, the update request may specify a validity period which is to apply to the update provided by update function node.

[0122] At step 1304, in response to receiving an update request that comprises an identifier of the digital twin, the update function node determines one or more parameters of the digital twin that are to be updated based on the identifier of the digital twin.

[0123] If a confidence level associated with the first machine learning model and a confidence level associated with the machine learning model for predicting a future network state exceed respective thresholds, the signalling flow may proceed to step 1306, 1310 or 1314. If the confidence level associated with the first machine learning model fails to exceed its associated threshold, or the confidence level associated with the machine learning model fails to exceed its associated threshold, the signalling flow proceeds to step 1322.

[0124] At step 1306, the update function node infers a future network state using the machine learning model.

[0125] At step 1308, the update function node generates the updated values of the one or more parameters using the first machine learning model, based on the inferred network state. The signalling flow then proceeds to step 1318.

[0126] That is, in steps 1306 and 1308, the update function node determines which updated parameter values of the digital twin should be collected from the real network, and which updated parameter values of the digital twin should be generated using a first machine learning model, based on the confidence associated with the first machine learning model and the second machine-learning model. In this example, as both the confidence associated with the first machine learning model, and the confidence associated with the machine learning model are determined to be acceptable, the update function node determines that all of updated parameter values for the digital twin should be generated using a first machine learning model. It will be appreciated that basing this determination on the confidence associated with the first machine learning model and the second machine-learning model ensures the quality of the updated parameter values that are provided to the first network node.

[0127] Alternatively, the signalling flow may proceed from step 1304 to step 1310, if the update request comprises a constraint specifying a time period in which the update function node is to provide a response to the request.

[0128] In step 1310, the update function node determines to obtain updated values for as many of the one or more parameters from the network as permitted by the network load and as can be obtained from the network within the time period. The network load may be determined from a future network state that has been predicted using the second machine-learning model.

[0129] At step 1312, the update function node determines to generate updated values for the remaining one or more parameters using the first machine learning model. The signalling flow then proceeds to step 1318.

[0130] That is, in steps 1310 and 1312, the update function node determines which updated parameter values of the digital twin should be collected from the real network, and which updated parameter values of the digital twin should be generated using a first machine learning model, based on the constraint provided in the update request, and based on the state of the network. Basing this determination on the constraint provided in the update request and the state of the network, enables a response to the request to be provided that fulfils the constraint, ensures normal network operation (by virtue of not overloading the network during the process of acquiring the updated parameters values) and ensures that all the updated parameter values are provided in the response to the request (by virtue of generating the updated values that are unable to be obtained from the network).

[0131] That is, generating some or all of the updated parameter values using a first machine learning model as part of the digital twin update, enables the digital twin to remain synchronised with the network regardless of whether the network is able to provide the requested updated parameter values or not.

[0132] Alternatively, the signalling flow may proceed from step 1304 to step 1314, if the update request does not comprise a constraint specifying a time period in which the update function node is to provide a response to the request. At step 1314, the update function node determines to obtain updated values for as many of the one or more parameters from the network as permitted by the network load. As noted above, the network load may be determined from a future network state that has been predicted using the second machine-learning model.

[0133] At step 1316, the update function node determines to generate updated values for the remaining one or more parameters using the first machine learning model. The signalling flow then proceeds to step 1318.

[0134] That is, in steps 1314 and 1316, the update function node determines which updated parameter values of the digital twin should be collected from the real network, and which updated parameter values of the digital twin should be generated using a first machine learning model, based the state of the network. Basing this determination on the state of the network ensures normal network operation (by virtue of not overloading the network during the process of acquiring the updated parameters values) and ensures that all the updated parameter values are provided in the response to the request (by virtue of generating the updated values that are unable to be obtained from the network).

[0135] At step 1318, the update function node aggregates the generated updated values and / or the obtained updated values, to form a list of updated parameter values for the digital twin.

[0136] At step 1320, the update function node sends, to the first network node, a response to the request, the response comprising the obtained updated values and / or the generated updated values. The first network node may then use the obtained updated values and / or the generated updated values to update the state of the digital twin, such that the digital twin remains in sync with the network.

[0137] As noted above, the signalling flow may, in some embodiments, proceed from step 1304 to step 1322. At step 1322, the update function node obtains updated parameter values for the digital twin from the second network node.

[0138] That is, in step 1322, the update function node determines which updated parameter values of the digital twin should be collected from the real network, and which updated parameter values of the digital twin should be generated using a first machine learning model, based on the confidence associated with the first machine learning model and the second machine-learning model. In this example, as the confidence associated with both the first machine learning model and / or the confidence associated with the machine learning model are determined to be unacceptable, the update function node determines that all of updated parameter values for the digital twin should be obtained from the real network. Basing this determination on the confidence associated with the first machine learning model and the second machine-learning model ensures the quality of the updated parameter values that are provided to the first network node. If the update request comprises a constraint specifying a time period in which the update function node is to provide a response to the request, upon this time period elapsing, the signalling flow proceeds to step 1324. At step 1324, the update function node stops obtaining updated parameter values from the second network node. It will be appreciated that, in these embodiments, the update function node may not have obtained updated values for all of the parameters that were requested to be updated by the first network node.

[0139] In step 1326, the update function node sends, to the first network node, a response to the request comprising the obtained updated values of parameters of the digital twin. The response to the request may be a partial response to the request that does not comprise updated values for all of the parameters that were requested to be updated.

[0140] Alternatively, if the update request does not comprise a constraint specifying a time period in which the update function node is to provide a response to the request, the signalling flow proceeds from step 1322 to step 1328.

[0141] At step 1328, in response to obtaining updated values for all the parameters indicated in the update request, the update function node sends, to the first network node, a response to the request comprising the obtained updated values.

[0142] In some embodiments, a response to the update request may comprise an indication of the proportion of obtained updated values and the proportion of generated updated values comprised in the response. For example, the response sent to the first network node at step 1326 or step 1328 may comprise an indication that all of the updated values comprised in the response have been obtained from the real network.

[0143] In some embodiments, the update request may comprise a constraint on a proportion of obtained updated values and a proportion of generated updated values to be provided in the response to the request. It will be appreciated that, in these embodiments, this would impact the determination of which updated parameter values of the digital twin should be collected from the real network, and which updated parameter values of the digital twin should be generated using a first machine learning model, as the update function node will perform this split in a manner that meets this constraint.

[0144] In some embodiments, the update request may comprise a constraint on a maximum throughput that is to apply to providing the response to the request. For example, the constraint may indicate a throughput budget that is to apply to digital twin update operations performed by the update function node and the second network node. It will be appreciated that, in these embodiments, this would impact the determination of which updated parameter values of the digital twin should be collected from the real network, and which updated parameter values of the digital twin should be generated using a first machine learning model. The update function node may take into account the computational cost of generating updated parameter values, and the computational cost of obtaining updated parameter values when performing the split, in order to ensure that the total cost of these operations do not exceed the throughput budget.

[0145] Examples of the present disclosure thus propose an update function node that determines whether to, and how to, "split” between obtaining updated values of parameters of a digital twin from a network and generating synthetic updated values of parameters of a digital twin, given factors such as the state of the network and constraints relating to the acquisition the updated values. The update function node may also then generate synthetic updated values of parameters of a digital twin using machine-learning techniques.

[0146] Advantages afforded by example methods according to the present disclosure include that normal network operation is guaranteed, regardless of the demands for parameter update (as not all the updated parameter values for the digital twin need be obtained from the network should this result in the network becoming overloaded). Advantages also include ensuring that a digital twin will remain synchronized with the network, regardless of whether the updated parameter values can be obtained from the network or not (within any constraints of the update request), as these parameters can instead be generated using a first machine learning model.

[0147] The methods of the present disclosure may be implemented in hardware, or as software modules running on one or more processors. The methods may also be carried out according to the instructions of a computer program, and the present disclosure also provides a computer readable medium having stored thereon a program for carrying out any of the methods described herein. A computer program embodying the disclosure may be stored on a computer readable medium, or it could, for example, be in the form of a signal such as a downloadable data signal provided from an Internet website, or it could be in any other form.

[0148] It should be noted that the above-mentioned examples illustrate rather than limit the disclosure, and that those skilled in the art will be able to design many alternative embodiments without departing from the scope of the appended claims. The word "comprising” does not exclude the presence of elements or steps other than those listed in a claim, "a” or "an” does not exclude a plurality, and a single processor or other unit may fulfil the functions of several units recited in the claims. Any reference signs in the claims shall not be construed so as to limit their scope. Abbreviations

[0149] DT Digital Twin

[0150] DUF DT Update Function

[0151] 3P Third Party

[0152] OSS Operations Support System

[0153] RAN Radio Access Network

[0154] O-RAN Open RAN

[0155] PCF Policy Control Function

[0156] UPF User Plane Function

[0157] SDN Software-Defined Network

[0158] QoS Quality of Service

[0159] UE User Equipment

[0160] TTI Transmission Time Interval

[0161] DTI Decision Time Interval

[0162] PRB Physical Resource Block

[0163] OFDM Orthogonal frequency-division multiplexing

[0164] RRM Radio Resource Management

[0165] RF Radio Frequency

[0166] RU Radio Unit

[0167] RL Reinforcement Learning

[0168] NDC Network Data Collector

[0169] Al Artificial Intelligence

[0170] NWDAF Network Data and Analytics Function

[0171] NEF Network Exposure Function

[0172] NF Network Function

Claims

CLAIMS1. A method (400) performed by an update function node of a digital twin, wherein the digital twin is associated with a network, the method (400) comprising: receiving (402), from a first network node, a request indicating that one or more parameters of the digital twin are to be updated; based on one or more parameters that represent a state of the network and / or one or more constraints, determining (404), for each of the one or more parameters: whether to obtain an updated value of the parameter from a second network node, or whether to generate an updated value of the parameter; for one or more of the one or more parameters: in response to determining to obtain the updated value of the parameter from the second network node, obtaining (406) the updated value of the parameter from the second network node, or in response to determining to generate the updated value of the parameter, generating (406) the updated value of the parameter; and sending (408), to the first network node, a response to the request, the response comprising the obtained updated values and / or the generated updated values.

2. The method (400) of claim 1 , wherein generating (406) the updated value of the parameter comprises generating the updated value of the parameter using a first machine learning model.

3. The method (400) of claim 2, wherein the first machine learning model is a generative model or a discriminative model.

4. The method (400) of claim 3, wherein the generative model is one of: a generative adversarial network (GAN), a transformer, a variational Autoencoder (VAE).

5. The method (400) of claim 3, wherein the discriminative model is a regression-based model or a recurrent neural network.

6. The method (400) of any preceding claim, wherein the request comprises an identifier of the digital twin.

7. The method (400) of claim 6, the method (400) further comprising: determining, based on the identifier of the digital twin, the one or more parameters of the digital twin.

8. The method (400) of any of claims 1 to 6, wherein the request identifies the one or more parameters.

9. The method (400) of any preceding claim, wherein the request identifies the one or more constraints.

10. The method (400) of any of claims 2 to 9, wherein the first machine learning model has been trained to generate updated values of parameters of the digital twin based on one or more parameters that represent the state of the network.

11. The method (400) according to any preceding claim, wherein the one or more parameters that represent the state of the network comprise at least one of: one or more parameters that represent a predicted future state of the network; and one or more parameters that represent a current state of the network.

12. The method (400) of claim 11 , wherein the method (400) further comprises: using a second machine learning model to generate the one or more parameters that represent the predicted future state of the network.

13. The method (400) of claim 12, wherein the second machine learning model has been trained to generate, based on the one or more parameters that represent the current state of the network, the one or more parameters that represent the predicted future state of the network.

14. The method (400) of any preceding claim, wherein the one or more parameters that represent the state of the network describe at least one of: a load of the network, a configuration of the network, a radio environment of the network, and a physical environment of the network.

15. The method (400) of any of claims 2 to 14, wherein the one or more constraints comprise at least one of: a threshold value to be exceeded by a confidence level associated with the first machine learning model; a time period in which the update function node is to provide, to the first network node, the response to the request; a proportion of obtained updated values and a proportion of generated updated values to be provided in the response to the request;a validity period that is to apply to the updated values provided in the response to the request; and a maximum throughput that is to apply to providing the response to the request.

16. The method (400) of any of claims 2 to 15, wherein the step of determining (404) comprises, for each of the one or more parameters: determining, based on the one or more parameters that represent the state of the network, whether the second network node is able to provide an updated value of the parameter; in response to determining that the second network node is able to provide the updated value of the parameter, determining to obtain the updated value of the parameter from the second network node; and in response to determining the second network node is not able to provide the updated value of the parameter, determining to generate the updated value of the parameter using the first machine learning model.

17. The method (400) of claim 15, wherein the one or more constraints comprise the threshold value, and wherein the step of determining (404) comprises: in response to the confidence level associated with the first machine learning model exceeding the threshold value, determining, for each of the one or more parameters, to generate an updated value of the parameter using the first machine learning model; and in response to a confidence level associated with the first machine learning model failing to exceed the threshold value, determining, for each of the one or more parameters, to obtain an updated value of the parameter from the second network node.

18. The method (400) of claim 15, wherein the one or more constraints comprise the time period, and wherein the step of determining (404) comprises, for each of the one or more parameters: determining whether an updated value of the parameter can be obtained from the second network node within the time period; determining, based on the one or more parameters that represent the state of the network, whether the second network node is able to provide an updated value of the parameter; in response to determining that the updated value of the parameter can be obtained from the second network node within the time period and in response to determining that the second network node is able to provide the updated value of the parameter, determining to obtain the updated value of the parameter from the second network node; and in response to determining that the updated value of the parameter cannot be obtained from the second network node within the time period, or in response to determining the second network node is not able to provide the updated value of the parameter, determining to generate the updated value of the parameter using the first machine learning model.

19. The method (400) of any preceding claim, wherein the response to the request further comprises an indication of the proportion of obtained updated values and the proportion of generated updated values provided in the response.

20. The method (400) of any preceding claim, wherein the first network node comprises the digital twin, or wherein the first network node comprises a management function of the digital twin.

21. The method (400) of any preceding claim, wherein the second network node comprises a network data collector function.

22. The method (400) of any preceding claim, wherein the update function node is a non-Real Time Radio Intelligence Controller, or wherein the update function node is a Network Data and Analytics Function node.

23. A method (500), performed by an update function node of a digital twin, of training a first machine learning model to generate updated values of parameters of a digital twin, wherein the digital twin is associated with a network, the method (500) comprising: sending (502), to a second network node, a request for one or more parameters that represent a state of the network and one or more updated values of parameters of the digital twin; receiving (504), from the second network node, one or more parameters that represent a state of the network and one or more updated values of parameters of the digital twin; determining (506) whether the received parameters and the received updated values were received within a particular time period; in response to determining that the received parameters and the received updated values were received within the particular time period, forming (508) a training data set comprising the received parameters and the received updated values; and using (510) the training dataset to train the first machine learning model to generate updated values of parameters of the digital twin based on one or more parameters that represent a state of the network.

24. The method (500) of claim 23, wherein the first machine learning model is a generative model or a discriminative model.

25. The method (500) of claim 24, wherein the generative model is one of: a generative adversarial network (GAN), a transformer, a variational Autoencoder (VAE).

26. The method (500) of claim 24, wherein the discriminative model is a regression-based model or a recurrent neural network.

27. The method (500) of any of claims 23-26, wherein the training dataset comprises: a set of input training data comprising the received parameters; and a set of output training data comprising the received updated values.

28. The method (500) of any of claims 23 to 27, wherein the second network node comprises a network data collector function.

29. The method (500) of any of claims 23-28, wherein the update function node is a non-Real Time Radio Intelligence Controller, or wherein the update function node is a Network Data and Analytics Function node.

30. The method (500) of any of claims 23-29, wherein the one or more parameters that represent a state of the network describe at least one of: a load of the network, a configuration of the network, a radio environment of the network, or a physical environment of the network.31 . The method (500) of any of claims 23-30, wherein the method (500) further comprises: generating an updated value of a parameter of the digital twin using the first machine learning model.

32. A method (600) performed by a first network node, the method (600) comprising: sending (602), to an update function node of a digital twin, a request indicating that one or more parameters of the digital twin are to be updated, wherein the digital twin is associated with a network; and receiving (604), from the update function node, a response to the request, the response comprising one or more updated values of parameters of the digital twin that have been obtained from a second network node, and / or one or more updated values of parameters of the digital twin that have been generated by the update function node.

33. The method (600) of claim 32, wherein the one or more updated values of parameters of the digital twin that have been generated by the update function node have been generated using a first machine learning model.

34. The method (600) of claim 33, wherein the first machine learning model is a generative model or a discriminative model.

35. The method (600) of claim 34, wherein the generative model is one of: a generative adversarial network (GAN), a transformer, a variational Autoencoder (VAE).

36. The method (600) of claim 34, wherein the discriminative model is a regression-based model or a recurrent neural network.

37. The method (600) of any of claims 32-36, wherein the request comprises an identifier of the digital twin.

38. The method (600) of any of claims 32-37, wherein the request identifies the one or more parameters.

39. The method (600) of any of claims 32-38, wherein request identifies one or more constraints.

40. The method (600) of claim 39, when dependent on any of claims 33 to 38, wherein the one or more constraints comprise at least one of: a threshold value to be exceeded by a confidence level associated with the first machine learning model; a time period in which the update function node is to provide, to the first network node, the response to the request; a proportion of obtained updated values and a proportion of generated updated values to be provided in the response to the request; a validity period that is to apply to the updated values provided in the response to the request; and a maximum throughput that is to apply to providing the response to the request.41 . The method (600) of any of claims 32-40, wherein the response to the request further comprises an indication of the proportion of obtained updated values and the proportion of generated updated values provided in the response.

42. The method (600) of any of claims 32-41 , wherein the first network node comprises the digital twin, or wherein the first network node comprises a management function of the digital twin.

43. The method (600) of any of claims 32-42, wherein the update function node is a non-Real Time Radio Intelligence Controller, or wherein the update function node is a Network Data and Analytics Function node.

44. The method (600) of any of claims 32-43, wherein the second network node comprises a network data collector function.

45. A method (700) performed by a second network node, the method (700) comprising: providing (702), to an update function node of a digital twin, one or more updated values of parameters of the digital twin, wherein the digital twin is associated with a network.

46. The method (700) of claim 45, wherein the method (700) further comprises: providing, to the update function node, one or more parameters that represent a state of the network.

47. The method (700) of claim 45 or 46, wherein the second network node comprises a network data collector function.

48. The method (700) of any of claims 45-47, wherein the update function node is a non-Real Time Radio Intelligence Controller, or wherein the update function node is a Network Data and Analytics Function node.

49. A computer program product comprising a computer readable medium, the computer readable medium having computer readable code embodied therein, the computer readable code being configured such that, on execution by a suitable computer or processor (802, 902, 1002), the computer or processor ((802, 902, 1002) is caused to perform a method (400, 500, 600, 700) of any one of claims 1 to 48.

50. An update function node (304, 800) of a digital twin associated with a network, the update function node (304, 800) configured to: receive, from a first network node (302, 900), a request indicating that one or more parameters of the digital twin are to be updated; based on one or more parameters that represent a state of the network and / or one or more constraints, determine, for each of the one or more parameters: whether to obtain an updated value of the parameter from a second network node (306, 1000), or whether to generate an updated value of the parameter using a generative model; for one or more of the one or more parameters: in response to determining to obtain the updated value of the parameter from the second network node (306, 1000), obtain the updated value of the parameter from the second network node (306, 1000), orin response to determining to generate the updated value of the parameter using a generative model, generate the updated value of the parameter using a generative model; and send, to the first network node (302, 900), a response to the request, the response comprising the obtained updated values and / or the generated updated values.51 . The update function node (304, 800) of claim 50, wherein the update function node (304, 800) is further configured to perform the method of any of claims 2-22.

52. An update function node (304, 800) of a digital twin associated with a network, the update function node (304, 800) comprising a processor (802) and a memory (804), said memory (804) containing instructions executable by said processor (802), whereby said update function node (304, 800) is operative to: receive, from a first network node (302, 900), a request indicating that one or more parameters of the digital twin are to be updated; based on one or more parameters that represent a state of the network and / or one or more constraints, determine, for each of the one or more parameters: whether to obtain an updated value of the parameter from a second network node (306, 1000), or whether to generate an updated value of the parameter using a generative model; for one or more of the one or more parameters: in response to determining to obtain the updated value of the parameter from the second network node (306, 1000), obtain the updated value of the parameter from the second network node (306, 1000), or in response to determining to generate the updated value of the parameter using a generative model, generate the updated value of the parameter using a generative model; and send, to the first network node (302, 900), a response to the request, the response comprising the obtained updated values and / or the generated updated values.

53. The update function node (304, 800) of claim 52, wherein the update function node (304, 800) is further operative to perform the method of any of claims 2-22.

54. An update function node (304, 800) of a digital twin associated with a network, the update function node (304, 800) configured to: send, to a second network node (306, 1000), a request for one or more parameters that represent a state of the network and one or more updated values of parameters of the digital twin;receive, from the second network node (306, 1000), one or more parameters that represent a state of the network and one or more updated values of parameters of the digital twin; determine whether the received parameters and the received updated values were received within a particular time period; in response to determining that the received parameters and the received updated values were received within the particular time period, form a training data set comprising the received parameters and the received updated values; and use the training dataset to train the first machine learning model to generate updated values of parameters of the digital twin based on one or more parameters that represent a state of the network.

55. The update function node (304, 800) of claim 54, wherein the update function node (304, 800) is further configured to perform the method of any of claims 24-31 .

56. An update function node (304, 800) of a digital twin associated with a network, the update function node (304, 800) comprising a processor (802) and a memory (804), said memory (804) containing instructions executable by said processor (802), whereby said update function node (304, 800) is operative to: send, to a second network node (306, 1000), a request for one or more parameters that represent a state of the network and one or more updated values of parameters of the digital twin; receive, from the second network node (306, 1000), one or more parameters that represent a state of the network and one or more updated values of parameters of the digital twin; determine whether the received parameters and the received updated values were received within a particular time period; in response to determining that the received parameters and the received updated values were received within the particular time period, form a training data set comprising the received parameters and the received updated values; and use the training dataset to train the first machine learning model to generate updated values of parameters of the digital twin based on one or more parameters that represent a state of the network.

57. The update function node (304, 800) of claim 56, wherein the update function node (304, 800) is further operative to perform the method of any of claims 24-31 .

58. A first network node (302, 900), the first network node (302, 900) configured to: send, to an update function node (304, 800) of a digital twin, a request indicating that one or more parameters of the digital twin are to be updated, wherein the digital twin is associated with a network; andreceive, from the update function node (304, 800), a response to the request, the response comprising one or more updated values of parameters of the digital twin that have been obtained from a second network node (306, 1000), and / or one or more updated values of parameters of the digital twin that have been generated by the update function node (304, 800).

59. The first network node (302, 900) of claim 58, wherein the first network node (302, 900) is further configured to perform the method of any of claims 33-44.

60. A first network node (302, 900), the first network node (302, 900) comprising a processor (902) and a memory (904), said memory (904) containing instructions executable by said processor (902), whereby said first network node (302, 900) is operative to: send, to an update function node (304, 800) of a digital twin, a request indicating that one or more parameters of the digital twin are to be updated, wherein the digital twin is associated with a network; and receive, from the update function node (304, 800), a response to the request, the response comprising one or more updated values of parameters of the digital twin that have been obtained from a second network node (306, 1000), and / or one or more updated values of parameters of the digital twin that have been generated by the update function node (304, 800).61 . The first network node (302, 900) of claim 60, wherein the first network node (302, 900) is further operative to perform the method of any of claims 33-44.

62. A second network node (306, 1000), the second network node (306, 1000) configured to: provide, to an update function node (304, 800) of a digital twin, one or more updated values of parameters of the digital twin, wherein the digital twin is associated with a network.

63. The second network node (306, 1000) of claim 62, wherein the second network node (306, 1000) is further configured to perform the method of any of claims 46-48.

64. A second network node (306, 1000), the second network node (306, 1000) comprising a processor (1002) and a memory (1004), said memory (1004) containing instructions executable by said processor (1002), whereby said second network node (306, 1000) is operative to: provide, to an update function node (304, 800) of a digital twin, one or more updated values of parameters of the digital twin, wherein the digital twin is associated with a network.

65. The second network node (306, 1000) of claim 64, wherein the second network node (306, 1000) is further operative to perform the method of any of claims 46-48.

Images