Apparatus and method for data generator
By using tuning parameters to adjust a data generator that represents physical conditions, the apparatus efficiently generates training data within communication network entities, addressing the challenge of transmitting large datasets and optimizing machine learning model training.
Patent Information
- Application Number
- PCT/EP2024/061882
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-12
- Filing Date
- 2024-04-30
- Publication Date
- 2024-11-21
AI Technical Summary
The training of machine learning models in communication networks requires large datasets, making it difficult and costly to transmit these datasets between User Equipment (UE) and gNBs, especially for channel conditions and geometrical environments.
An apparatus and method that use tuning parameters based on physical condition measurements to adjust a data generator, allowing it to recreate training or monitoring data without transferring large datasets, by using a model that represents the physical conditions, such as channel or signal path models, and selecting appropriate prototypes for different radio channel scenarios.
Enables efficient generation of training or monitoring data within the network entities, reducing data transfer overhead and maintaining data integrity, allowing for optimized machine learning model training and updates without exchanging large datasets.
Smart Images

Figure EP2024061882_21112024_PF_FP_ABST
Abstract
Description
[0001] TITLE
[0002] APPARATUS AND METHOD FOR DATA GENERATOR
[0003] TECHNOLOGICAL FIELD
[0004] Examples of the disclosure relate to a data generator. Some relate to a data generator for generating training data or monitoring data for use in communication networks
[0005] BACKGROUND
[0006] Machine learning (ML) models have many use cases. In communications networks they can be used in gNBs and / or User Equipment (UE). The training of these ML models requires huge datasets. It is difficult to transmit these huge datasets between respective entities in a network.
[0007] BRIEF SUMMARY
[0008] According to various, but not necessarily all, examples of the disclosure there is provided an apparatus comprising means for: receiving one or more tuning parameters where the tuning parameters are based, at least in part, on data obtained from measurements of physical conditions such that the tuning parameters represent, at least in part, the physical conditions; using the tuning parameters to adjust a data generator where the data generator comprises a model that represents the physical conditions; using the adjusted data generator to generate data for a machine learning model.
[0009] The measurements may comprise measurements of physical conditions of a channel such that the tuning parameters represent, at least in part, the physical conditions of the channel and the data generator comprises a channel model that represents the channel. The measurements may comprise measurements of reference signal received power estimations for a set of such that the tuning parameters represent, at least in part, the signal paths for a geometrical environment and the data generator comprises the signal paths.
[0010] The means may be for using the data generated by the data generator to train the machine learning model.
[0011] The data generated by the adjusted data generator may comprise measurement data from which the tuning parameters were based.
[0012] The means may be for configuring the data generator to generate data for a selected functionality of a machine learning model.
[0013] A plurality of prototypes may be available for the data generator wherein respective prototypes are configured for different radio channel scenarios and the means are for selecting a prototype for use in the data generator based on a radio channel scenario.
[0014] The tuning parameters may be received in a hierarchy wherein tuning parameters from different levels of the hierarchy are configured to be used to adjust different levels of the data generator.
[0015] The hierarchy may comprise at least one of: a level of tuning parameters indicating a prototype for the data generator; a level of tuning parameters indicating information specific to a prototype; a level of tuning parameters comprising data obtained from or derived from measurements.
[0016] The means may be for performing a check of the generated data by comparing at least part of the generated data to measurement data.
[0017] The means may be for receiving data samples affected by anomalies and extracting an impact of the anomalies to the data generator and adjusting the data generator based on the impact. The apparatus may be a network entity.
[0018] The apparatus may be a user equipment.
[0019] According to various, but not necessarily all, examples of the disclosure there is provided a method comprising: receiving one or more tuning parameters where the tuning parameters are based, at least in part, on data obtained from measurements of physical conditions such that the tuning parameters represent, at least in part, the physical conditions; using the tuning parameters to adjust a data generator where the data generator comprises a model that represents the physical conditions; using the adjusted data generator to generate data for a machine learning model.
[0020] According to various, but not necessarily all, examples of the disclosure there is provided a computer program comprising instructions which, when executed by an apparatus, cause the apparatus to perform at least: receiving one or more tuning parameters where the tuning parameters are based, at least in part, on data obtained from measurements of physical conditions such that the tuning parameters represent, at least in part, the physical conditions; using the tuning parameters to adjust a data generator where the data generator comprises a model that represents the physical conditions; using the adjusted data generator to generate data for a machine learning model.
[0021] While the above examples of the disclosure and optional features are described separately, it is to be understood that their provision in all possible combinations and permutations is contained within the disclosure. It is to be understood that various examples of the disclosure can comprise any or all of the features described in respect of other examples of the disclosure, and vice versa. Also, it is to be appreciated that any one or more or all of the features, in any combination, may be implemented by / comprised in / performable by an apparatus, a method, and / or computer program instructions as desired, and as appropriate. BRIEF DESCRIPTION
[0022] Some examples will now be described with reference to the accompanying drawings in which:
[0023] FIG. 1 shows an example network;
[0024] FIG. 2 shows an example method;
[0025] FIG. 3 shows an example data generator;
[0026] FIG. 4 shows an example data generator;
[0027] FIG. 5 shows an example data generator:
[0028] FIG. 6 shows an example method;
[0029] FIG. 7 shows an example method;
[0030] FIG. 8 shows an example signaling chart;
[0031] FIG. 9 shows an example signaling chart; and
[0032] FIG. 10 shows an example apparatus.
[0033] The figures are not necessarily to scale. Certain features and views of the figures can be shown schematically or exaggerated in scale in the interest of clarity and conciseness. For example, the dimensions of some elements in the figures can be exaggerated relative to other elements to aid explication. Corresponding reference numerals are used in the figures to designate corresponding features. For clarity, all reference numerals are not necessarily displayed in all figures.
[0034] DEFINITIONS
[0035] BVDM Building Vector Data Maps
[0036] CSI Channel State Information
[0037] CSI RS Channel State Information Reference Signal
[0038] E-LITRA Evolved Universal Mobile Telecommunications System (UMTS)
[0039] Terrestrial Radio Access
[0040] DoA Direction of Arrival gNB NR base station
[0041] LCM Life Cycle Management
[0042] LO Local Oscillator
[0043] LOS Line of Sight ML Machine learning
[0044] MPC Model Predictive Control
[0045] NLOS Non-Line of Sight
[0046] NR New Radio
[0047] O-RAN Open Radio Access Network
[0048] PRB Physical Resource Block
[0049] RAN Radio Access Network
[0050] RRC Radio Resource Control
[0051] RSRP Received Strength Signal Power
[0052] SIM Subscriber Identity Module
[0053] UE User Equipment
[0054] DETAILED DESCRIPTION
[0055] Fig. 1 illustrates an example of a network 100 such as a 5G NR network. The network 100 comprises a plurality of terminal apparatus 110 and a plurality of network entities 120, 130. The network entities 120, 130 can comprise one or more node apparatus 120 that can be configured to communicate with the terminal apparatus 110. The network entities can also comprise one or more core network apparatus 130. The one or more core network apparatus 130 communicate with the node apparatus 120. In some examples the one or more core network apparatus 130 communicate with the terminal apparatus 110.
[0056] The one or more core network apparatus 130 can, in some examples, communicate with each other. The one or more node apparatus 120 can, in some examples, communicate with each other.
[0057] The network 100 can be a cellular network comprising a plurality of cells 122 each served by a node apparatus 120. In this example, the interface between the terminal apparatus 110 and a node apparatus 120 defining a cell 122 is a wireless interface 124.
[0058] The node apparatus 120 comprises one or more cellular radio transceivers. The terminal apparatus 110 comprises one or more cellular radio transceivers. In the example illustrated the cellular network 100 is a third generation Partnership Project (3GPP) network in which the terminal apparatus 110 are user equipment (UE) and the node apparatus 120 can be access nodes such as base stations (gNB).
[0059] The term ‘user equipment’ is used to designate mobile equipment comprising a smart card for authentication / encryption etc. such as a Subscriber Identity Module (SIM). In some examples the term ‘user equipment’ is used to designate mobile equipment comprising circuitry embedded as part of the user equipment for authentication / encryption such as software SIM.
[0060] The node apparatus 120 can be any suitable base station. A base station is an access node. It can be a network entity responsible for radio transmission and reception in one or more cells to or from the UE 110. The node apparatus 120 can be a network element in a Radio Access Network (RAN), an Open-Radio Access Network (O-RAN), a E-UTRA (Evolved Universal Mobile Telecommunications System (UMTS) Terrestrial Radio Access) network, or any other suitable type of network.
[0061] The core network apparatus 130 can be part of a core network. The core network apparatus 130 can be configured to manage functions relating to connectivity for the UEs 110. For example, the core network apparatus 130 can be configured to manage functions such as connectivity, mobility, authentication, authorization and / or other suitable functions.
[0062] In the example of Fig. 1 the core network apparatus 130 is shown as a single entity. In some examples the core network apparatus 130 could be distributed across a plurality of entities. For example, the core network apparatus 130 could be cloud based or distributed in any other suitable manner.
[0063] The network 100 can be a 4G or 5G network, for example. It can for example be a New Radio (NR) network that uses gNB or eNB as access nodes 120. New Radio is the 3GPP name for 5G technology. In such cases the node apparatus 120 can comprise gNodeBs (gNBs) 120 configured to provide user plane and control plane protocol terminations towards the UE 110 and / or to perform any other suitable functions. The gNBs 120 are interconnected with each other by means of an X2 / Xn interface 126. The gNBs are also connected by means of the N2 interface 128 to the core network apparatus 130. Other types of networks and interfaces could be used in other examples. Other types of networks could comprise next generation mobile and communication network, for example, a 6G network.
[0064] ML (Machine Learning) models can be used in communication networks 100 such as the network of Fig. 1. The ML models can be used for different functionalities. The ML models need huge data sets to train and maintain the models for their intended functionality. It can be difficult and costly to transmit such datasets between devices such as UEs 110 and gNBs 120. Examples of the disclosure enable data such as training data or monitoring data to be obtained as required without transferring huge datasets between UEs 110 and gNBs 120.
[0065] Fig. 2 shows an example method that can be used to enable training data or monitoring data to be obtained as required without transferring huge datasets between UEs 110 and gNBs120. The method could be implemented by a UE 110 or by a network entity such as a gNB 120.
[0066] The method comprises, at block 200 receiving one or more tuning parameters. The tuning parameters are based, at least in part, on data obtained from measurements of physical conditions such that the tuning parameters represent, at least in part, the physical conditions. The tuning parameters are based on the data obtained from measurements in that they can be obtained by processing data obtained from the measurements. The processing can ensure that a relationship between the tuning parameters and the physical conditions is maintained.
[0067] The measurements of physical conditions that the tuning parameters are obtained from can comprise any suitable measurements. The measurements can be performed by the apparatus that implements the method. For example, the measurements can be performed by a UE 110 or gNB 120.
[0068] The measurements of physical conditions can comprise any measurements that enable information about the physical conditions to be obtained. The physical conditions can comprise any physical factors that affect the signaling between a gNB 120 and a UE 110.
[0069] In some examples the measurements can comprise measurements of physical conditions of a channel such that the tuning parameters represent, at least in part, the physical conditions of the channel. In other examples the tuning parameters represent the physical conditions of a geometrical environment. For instance, the measurements could comprise measurements of reference signal received power (RSRP) estimations for a set of such that the tuning parameters represent, at least in part, the signal paths for a geometrical environement.
[0070] The method also comprises, at block 202, using the tuning parameters to adjust a data generator. The data generator comprises a model that represents the physical conditions. For example, if the tuning parameters are obtained from measurements of physical conditions of a channel the data generator can comprise a channel model that represents the channel.
[0071] In some examples the data generator can be configured to generate training data or monitoring data for a selected functionality of a machine learning model. The configuration of the data generator for a selected functionality could be achieved through the use of prototypes. A prototype can be a model which relates to a specific functionality of an ML model. A prototype can comprise a stochastic model or any other suitable type of model like a 3GPP channel model.
[0072] Different prototypes or combinations of prototypes can be available for different functionalities. Configuring the data generator for a particular functionality can comprise selecting an appropriate prototype or combination or prototypes. For instance, if the ML model is to be used for estimating channel conditions then the respective prototypes available for the data generator can be configured for different radio channel scenarios. The different radio channel scenarios can comprise different functionalities of the ML model, different channel conditions, types of channels, antenna configurations, or other scenarios. The different functionalities of the ML could be the intended use for the ML outputs, for instance, whether the outputs are to be used for channel prediction or channel compression or traffic scheduling or any other suitable use. The different channel conditions could comprise location of a UE 110, mobility of a UE 110 or any other relevant conditions. The different types of channels could comprise Line of Sight (LOS), Non-Line of Sight (NLOS), indoor channels, outdoor channels or any other relevant type of channels.
[0073] A prototype can be selected for use in the data generator based on a radio channel scenario or based on any other suitable factor. The factors used for the selection of the prototype can depend upon the use case of the ML model and / or any other suitable factor.
[0074] The method also comprises, at block 204, using the adjusted data generator to generate data for a machine learning model. The data can comprise training data or monitoring data or any other suitable type of data. Once the data has been generated it can be used to train or monitor the machine learning model.
[0075] The data that is generated by the adjusted data generator can be a large data set. The data that is generated can be large enough so that it can be used to train or update or monitor an ML model.
[0076] The data that is generated by the adjusted data generator can comprise measurement data from which the tuning parameters were based. That is, the adjusted data generator can enable lossless recreation of the measurement data, or at least a subset of the measurement data.
[0077] The tuning parameters can be received in any suitable format. In some examples the tuning parameters can be received in a hierarchy. The hierarchy can comprise multiple levels and different tuning parameters can be assigned to different levels of the hierarchy. Tuning parameters from different levels of the hierarchy can be configured to be used to adjust different levels of the data generator. For example, the hierarchy could comprise, a level of tuning parameters indicating a prototype for the data generator, a level of tuning parameters indicating information specific to a prototype, a level of tuning parameters comprising data obtained from or derived from measurements, and / or any other suitable parameters. The information specific to a prototype could comprise the number of multipath components comprising a current radio channel or any other suitable information, the current mobile speed, the direction of arrival (DoA) of the main Doppler components, beamforming gains, a noise floor, well defined diffuse scatterers, or any other suitable information.
[0078] In some examples the tuning parameters from the different levels of the hierarchy can be received at different times. This can enable tuning parameters from a given level to be sent more frequently than tuning parameters from a different level. For instance, tuning parameters comprising data obtained from or derived from measurements could change more often that the type of prototype that is needed. Therefore, a tuning parameter that indicates a type of protype could be assigned to a level of the hierarchy so that it is sent less frequently than the tuning parameters comprising data obtained from or derived from measurements.
[0079] In some examples the method could comprise additional blocks that are not shown in Fig. 2. For instance, the method could comprise performing a check of the generated data by comparing at least part of the generated data to measurement data. The check could be performed by transmitting a subset of the generated data to enable comparison of the subset of the generated data with a subset of the original data.
[0080] In some examples the method can comprise receiving data samples affected by anomalies and extracting an impact of the anomalies to the data generator. The data generator can then be adjusted based on the impact.
[0081] Fig. 3 schematically shows an example data generator 302 that can be used in examples of the disclosure. The data generator 302 could be a training data generator, a monitoring data generator or any other suitable type of generator.
[0082] The data generator 302 is configured to receive tuning parameters 300 as an input and provide data 304 as an output. The data 304 provided as an output could be training data or monitoring data. The data generator 302 can be implemented in a UE 110 or a gNB 120 or any other suitable device within a network 100. The data generator 302 can be implemented in a device that uses a ML program. The data generator 302 can be implemented in a device that train or updates an ML program.
[0083] The tuning parameters 300 are based, at least in part, on data obtained from measurements of physical conditions. The measurements can be performed by the gNB 120 and / or the UE 110. The tuning parameters 300 provide a representation of prior knowledge of the physical conditions.
[0084] If the data generator 302 is implemented by a gNB 120 then one or more UEs 110 can generate the tuning parameters and send the tuning parameters 300 to the gNB 120. If the data generator 302 is implemented by a UE 110 then one or more gNBs120 can generate the tuning parameters and send the tuning parameters 300 to the UE 110. In some examples the tuning parameters could be sent in both directions between a UE 110 and a gNB 120. A gNB 120 could generate tuning parameters for multiple UEs 110 and could send these to multiple UEs 110. Similarly, a gNB 120 could generate data that is relevant for multiple UEs 110.
[0085] The tuning parameters 300 are used to update the data generator 302 so that the data generator 302 provides a model of the measured physical condition. The data 304 that is output by the data generator 302 therefore takes into account the prior knowledge of the physical conditions.
[0086] The use of the data generator 302 can enable the data 304 to be generated in the same device as will make use of the data 304. Therefore, instead of exchanging large volumes of training data or monitoring data only the tuning parameters need to be exchanged. The use of the data generator 302 comprising a model of the physical condition can enable the original measured data to be recreated. In some examples the data generator 302 can enable the original measured data to be recreated without loss of data.
[0087] Fig. 4 schematically shows another example data generator 302 that can be used in examples of the disclosure. The data generator 302 is configured to receive tuning parameters 300 as an input and provide data 304 as an output. The data 304 could be training data or monitoring data or any other suitable type of data. In the example of Fig. 4 a plurality of prototypes 400 are available for the data generator 302.
[0088] The prototypes 400 can comprise models where the models of the respective prototypes 400 relate to different functionalities of an ML model. The respective prototypes 400 can comprise stochastic models or any other suitable types of models.
[0089] In the schematic example of Fig. 4 three different prototypes 400A, 400B, 400C are shown. Any number of prototypes 400 could be used in actual implementations of the disclosure. The actual number of prototypes 400 that are available could be much larger than three, for instance in some examples over 100 prototypes 400 could be available. The number of prototypes 400 can be determined by the number of functionalities that are available for the ML model or by any other relevant factor.
[0090] The prototypes can be predefined for a specific functionality, for instance they could be designed for channel compression or channel prediction or any other suitable function. The different prototypes 400A, 400B, 400C can then be designed for a specific scenario. For instance, in the example of Fig. 4 a first prototype 400A could be LOS channels, a second prototype 400B could be NLOS channels and a third prototype 400C could be for indoor channels. Additional prototypes 400 for other scenarios could also be available.
[0091] In some examples, for the functionality of beam management the prototypes 400 might comprise different predefined building vector data maps with specific shadowing objects, street crossings and other obstacles. These affect the strength of the received beams in a specific manner, which can be used at the data generator 302 to recreate the related spatial and temporal variations of the receive power of the strongest beams.
[0092] The use of the prototypes 400 can enable the data generation to be narrowed down to a subspace of the known scenario characteristics. For instance, if the specific scenario is known then a prototype 400 corresponding to that specific scenario can be selected and used in the data generator 302. The apparatus that implements the data generator 302 only needs to be receive an indication of the prototype 400 that is to be used and not the data that is used to select the prototype. This can reduce the data that needs to be transferred.
[0093] The prototype 400 that is needed can change if the scenario changes. However, the main scenario characteristics might change seldomly and so an update to the prototype 400 might only be needed on seldom occasions. For example, it might only be needed to update a prototype 400 every minute or even less frequently.
[0094] Fig. 5 schematically shows another example data generator 302. The data generator 302 is configured to receive tuning parameters 300 as an input and provide data 304 as an output and a plurality of prototypes 400 are available for the data generator 302. In the example of Fig. 5 the data generator 302 can be provided in a hierarchy.
[0095] The hierarchy of tuning parameters 300 can comprise multiple levels 500 and different tuning parameters 300 can be assigned to different levels 500 of the hierarchy. Tuning parameters 300 from different levels 500 of the hierarchy can be configured to be used to adjust different levels of the data generator 302.
[0096] In the example of Fig. 5 the hierarchy of tuning parameters 300 comprises three different levels 500A, 500B, 500C. The hierarchy of tuning parameters 300 can comprise a different number of levels 500 in other examples.
[0097] In the example of Fig. 5 a first level 500A of tuning parameters 300 can be used to indicate a protype for the data generator 302. For instance, the first level 500A of tuning parameters 300 could indicate which of the available prototypes 400 should be selected for use in the data generator 302. In the schematic example of Fig. 5 this could indicate if the first protype 400A, the second prototype 400B, or the third prototype 400C should be selected.
[0098] The second level 500B of tuning parameters 300 can indicate information specific to the prototype 400 that has been selected in the first level 500A. The second level 500B of tuning parameters 300 can indicate information that fine tunes parameters of the prototype 400 that has been selected in the first level 500A. The information specific to a protype 400 could comprise the number of multipath components comprising a current radio channel or any other suitable information, the current mobile speed, the DoA of the main Doppler components, beamforming gains, a noise floor, well defined diffuse scatterers, or any other suitable information.
[0099] In some examples data from a predefined library could be used to report the tuning parameters 300 in the second level 500B of the hierarchy. The predefined library could comprise specific noise floors, colored diffuse scatterer, set of Model Predictive Control (MPC) parameter evolutions, or any other suitable information.
[0100] The third level 500C of the hierarchy of tuning parameters 300 comprises data obtained from or derived from measurements. The third level 500C of the hierarchy can comprise compressed information extracted from measurements. The measurements could be any suitable measurements. In examples where the ML model is being used for channel compression or prediction the measurements could be CSI estimates from the UE 110. In examples where the ML model is being used for beam management the measurements could be the reference signal received power (RSRP) estimations for a set of spatial and / or temporal beams.
[0101] The data samples that used in the third level 500C of the hierarchy can be selected so that they have the biggest impact or add the most value to the information in the other levels of the hierarchy.
[0102] The data generator 302 can be implemented using any suitable means. In some examples the data generator 302 can be implemented as a specific ML model but with respect to relevant levels of the hierarchy of tuning parameters 300. For instance, the data generator 302 can be implemented as a specific ML model with respect to a first level 500A and a second level 500B of the hierarchy of tuning parameters 300 also as a rule based PHY layer block. For example, the first level 500A and a second level 500B of the hierarchy of tuning parameters 300 could be implemented by some more of less modified 3GPP channel models. A data generator 302 implemented this way, without the tuning parameters 300 from the third level 500C would be completely synthetic and would miss dynamic factors. The dynamic factors that would be missed could comprise the birth and death of multipath components, timing advance related phase jumps and / or shifts, time varying phase slopes in case of Local Oscillator (LO) frequency offsets, interfering signals from adjacent cells and other potential real-world effects. This additional information can then be provided by the tuning parameters 300 in the third level 500C of the hierarchy or other lower levels.
[0103] Any suitable means could be used to extract the tuning parameters 300 for the third level 500C. In some examples a ML model could be used. The ML model could be trained to identify anomalies in the estimated data relative to synthetic data that is generated from the data generator 302 using just the first level 500A and the second level 500B of the hierarchy of tuning parameters 300. These anomalies can then be used to determine the tuning parameters 300 for the third level 500C. tuning parameters 300 for the third level 500C can be extracted so that similar anomalies are reproduced by the data generator 302 more often or less often.
[0104] In some examples the entity that performs the measurements can identify data samples that are affected by anomalies. The entity can then report the subset of the data samples. The sub set of data samples can be reported in a compressed form. The apparatus that implements the data generator 302 can then extract the impact of the anomalies on the data generator 302.
[0105] In some examples the data generator 302 can be deployed using a canary release strategy such that updates are initially released only to a subset of a few affected users before a full roll out of the update. For instance, the data generated by the data generator 302 can be used to train a suitable model, such as a 3GPP model first. Then the trained ML model can be used to make decisions that affect a small number of entities. For instance, if the ML model is implemented by a gNB 120 it could be used to make decisions for a small group of UEs 110, or a small number of Physical Resource Blocks (PRBs) of one UE 110 (for instance, precoders). These UEs 110 could then report if the assumption of the data generator 302 was good. For example, UE 110 could report if the selected precoder was good or could report the rate that has been achieved or could report any other suitable information. If the report indicates that the assumptions of the data generator 302 were not good then this could trigger some more measurements to update the tuning parameters 300 and enable the data generator 302 to be further adjusted. Any suitable means or methods can be used to generate the tuning parameters 300. in some example constant data retrieval could be used which comprises collecting all relevant information over time without the need to store all data. This can be similar to data retrieval used for Kalman filtering. Generating the tuning parameters 300 in this way can enable large data sets such as long- term channel estimates to be handled on effectively by UEs 110 and / or by any other suitable entities within a network 100.
[0106] Examples of the disclosure therefore enable a gNB 120 or UE 110 to generate data 304. This avoids the need to exchange large data sets or identifications of the ML models. The tuning parameters 300 and the model used for the data generator 302 contain sufficient information that they can be used to generate sufficient data for any scenario. This enables the gNB 120 and / or UE 110 to train and / or update an ML model to any current scenarios or conditions.
[0107] As an example, use case a gNB 120 can be configured to provide tuning parameters 300 to UEs 110 entering its cell. In such cases the data generator 302 can be configured for the use case of channel prediction. For this use case it can be assumed that a UE 110 enters the cell at a specific location with specific channel conditions such as LOS, NLOS, indoor in car, or any other suitable condition. The UE 110 reports its location to the gNB 120 and requests the best fitting ML model for the given position and functionality. In this case the functionality is channel prediction. The gNB 120 can be assumed to know the channel conditions in its cell. For example, the gNB 120 can know the channel conditions in its cell from previous UE 110 feedback. Based on these known channel conditions the gNB 120 selects the best fitting data set for the given position of the UE 110. The gNB 120 can also identify the tuning parameters 300 for the given position of the UE 110. The tuning parameters 300 could be determined online or off line. The tuning parameters 300 can comprise a hierarchy of parameters or could be in any other suitable format.
[0108] The tuning parameters 300 can then be reported from the gNB 120 to the UE 110. The UE 110 uses the tuning parameters 300 to adjust the data generator 302 at the UE 110. The UE 110 can then use the adjusted data generator 302 to generate data that can then be used to update its own ML model. The ML model that is used by the UE 110 can be vendor specific. That is, different UEs 110 can use different ML models. The gNB 120 does not need to know details of the ML model used by the UE 110 because the gNB 120 only needs to send the tuning parameters 300 and not the data configured for use by the ML model.
[0109] This use case scenario enables the UE 110 to have an ML model which is optimized, or substantially optimized, for a specific scenario. This can be achieved without the need for long-term and time-consuming observations of the radio environment by the UE 110.
[0110] Use cases that use the gNB 120 to make ML based predictions have the benefits over use cases where the UE 110 makes the ML based predictions because the gNB 120 or cloud can implement more complex neural networks or other types of models, have more processing power and energy, can access ML model libraries over the core network more easily, have better control over processing latency, and / or for any other relevant factors. However, to provide a high-performance level the gNB 120 needs to use an ML model that is adapted to the current UE 110 radio channel conditions.
[0111] As an example use case where the gNB 120 makes the ML based predictions, the UE 110 can be configured to perform measurements and provide the tuning parameters 300 to the gNBs 120. For instance, in some use cases the gNB 120 could be configured to perform channel prediction. In such cases a UE 110 could track radio channel evolution and report tuning parameters 300 generated from the tracking to the gNB 120. The UE 110 could track the evolution of the channel based on regularly transmitted CSI RS. The UE 110 can then generate the tuning parameters 300 from the CSI and report these back to the gNB 120.
[0112] The gNB 120 can use the received tuning parameters 300 to adjust the data generator 302 at the gNB 120 and use the adjusted data to update an ML model at the gNB 120. The updated ML model therefore takes into account the CSI received from the UE 110.
[0113] The UEs 110 can be configured so that they transmit updates to the tuning parameters 300. For instance, if a UE 110 is moving then the UE could send the updates to the tuning parameters 300. The updates could be sent when one or more trigger conditions are satisfied. The trigger conditions could be a timer, a movement threshold of the UE 110 or any other suitable factor or combination of factors.
[0114] As another example use case the ML model could be used for beam managements. In this case the data generator 302 could be configured to provide a set of relevant spatial beams and their evolution in the time domain as the output data 304. The ML model can be trained to learn the spatially most relevant beams as well as the set of most likely beams for the next time steps. This can be learned from observed RSRP measurements and the current position of the UE 110.
[0115] The information about the most relevant beams can then be used to reduce the search space for finding the best beams. For example, only the most relevant beams need to be measured. If the number of most relevant beams is small compared to the overall number of possible beams then the beam search process is significantly faster and with less overhead.
[0116] In such use cases the prototypes 400 of a data generator 302 could comprise different building vector data maps (BVDM). These can be similar to BVDMs that could be used for ray tracing simulations. A BVDM that best fits the observed set of relevant beams over multiple time steps can then be selected from the available BVDMs and used as a prototype 400 for the data generator 302. The UE 110 (or other entity in the network 100) can provide further tuning parameters 300 for finer tuning of the data generator 302. The further tuning parameters 300 can comprise additional objects and / or reports from observed data sets to cover anomalies or any other suitable information.
[0117] Fig. 6 shows an example method that could be implemented by a network entity such as a gNB 120 in examples of the disclosure.
[0118] The method comprises, at block 600, configuring a data generator 302. In this example the data generator 302 can be implemented by the gNB 120. The configuring of the data generator 302 can comprise ensuring that the same type of data generator 302 is being referred to by both the gNB 120 and the UE 110. For instance, it can indicate that the same functionality or purpose of the data generator 302 is referred to by both the UE 110 and the gNB 120. Any suitable type of signaling can be used to configure the data generator 302. In some examples Radio Resource Control (RRC) signaling could be used.
[0119] At block 602 the method comprises aligning the prototypes 400 of the data generator 302 with a UE 110. This makes sure that the same prototypes 400 are referred to by both the UE 110 and the gNB 120. This could involve signaling to the UE 110 which protypes 400 are available to the gNB 120 for a given functionality or a scenario. For example, a gNB 120 could indicate which prototypes 400 are available for their own cell. The signaling could be RRC signaling or any other suitable type of signaling.
[0120] At block 604 the gNB 120 transmits a CSI-RS. The gNB can transmit a Non-Zero Power (NZP) CSI RS at predefined intervals. This can enable the UE 110 to make CSI measurements. Other types of measurements could be made in other examples.
[0121] At block 606 the gNB 120 receives the tuning parameters 300. In this example the tuning parameters 300 are derived from the CSI measurements. The tuning parameters 300 could be derived from other measurements in other examples.
[0122] At block 608 the gNB 120 uses the received tuning parameters 300 to adjust the data generator 302. This could comprise using the tuning parameters 300 to select a prototype 400 for use in the data generator 302 and / or for making any other suitable adjustments.
[0123] The adjusted data generator 302 can be used to generate the data. The generated data is then used, at block 610, to train an ML model or to monitor and ML model. The ML model can be trained or monitored for the specific functionality.
[0124] Fig. 7 shows an example method that could be implemented by a network entity such as a UE 110 in examples of the disclosure. The method of Fig. 7 corresponds to the method of Fig. 6 in that Fig. 7 shows a method that could be implemented by a UE 110 in communication with the gNB 120 of Fig. 6. The method comprises, at block 700, configuring a data generator 302. In this example the data generator 302 can be implemented by the gNB 120. The configuring of the data generator 302 can comprise ensuring that the same type of data generator 302 is being referred to by both the gNB 120 and the UE 110. For instance, it can indicate that the same functionality or purpose of the data generator 302 is referred to by both the UE 110 and the gNB 120.
[0125] Any suitable type of signaling can be used to configure the data generator 302. In some examples Radio Resource Control (RRC) signaling could be used.
[0126] At block 702 the method comprises aligning the prototypes 400 of the data generator 302 with the gNB 120. This makes sure that the same prototypes 400 are referred to by both the UE 110 and the gNB 120. This could involve receiving signals from the gNB 120 indicating which protypes 400 are available to the gNB 120 for a given functionality or a scenario. For example, a gNB 120 could indicate which prototypes 400 are available for their own cell. The signaling could be RRC signaling or any other suitable type of signaling.
[0127] At block 704 the UE 110 estimates CSI. This can be estimated over multiple time slots. The CSI can be estimated from CSI RS transmitted by the gNB 120 at regular intervals over multiple time slots. Other types of measurements could be made in other examples.
[0128] At block 706 the UE 110 infers tuning parameters 300 and reports these to the gNB 120. Any suitable means or process can be used to infer the tuning parameters 300. The tuning parameters 300 can be inferred from the CSI measurements and / or form any other suitable measurements. The tuning parameters 300 can indicate a prototype 400 that is to be used for the data generator 302 and / or can indicate and other suitable information.
[0129] At block 708 the UE 110 monitors the performance of the gNB 120. In this particular scenario the UE 110 could monitor the precoder performance. If the performance drops below a given threshold the UE 110 could infer and report further tuning parameters 300 to enable the more up to date data 304 to be generated by the gNB 120 and used for updating the ML model.
[0130] In the examples of Figs. 6 and 7 the ML model is implemented by the gNB 120 and the tuning parameters 300 are generated by the UE 110 and transmitted to the gNB120. In other examples the ML model could be implemented by the UE 110. In which case tuning parameters could be generated by the gNB 120 and transmitted to the UE 110 as appropriate.
[0131] For the example use case of beam management the tuning parameters 300 of Figs. 6 and 7 are related to BVDM prototypes instead of channel model prototypes and the measurements are related to beam specific RSRP values instead of the estimated radio channels.
[0132] Fig. 8 shows an example signaling chart that could be used to implement some examples of the disclosure. In the example of Fig. 8 the ML model and the data generator 302 can be implemented by the UE 110.
[0133] At block 800 the gNB 120 selects or creates an object for a data generator 302.
[0134] At block 802 the UE 110 capability is exchanged with the gNB 120. This can be exchanged within RRC signaling or any other suitable type of signaling. This can enable the UE 110 to indicate the capabilities that the UE 110 has in terms of a data generator 302 and / or ML models and / or any other relevant capability.
[0135] At block 804 the data generator 302 is sent from the gNB 120 to the UE 110. This can be the data generator 302 that was selected or created at block 800.
[0136] The data generator 302 can be sent from the gNB 120 to the UE 110 using any suitable means. In some examples the data generator 302 can be sent from the gNB 120 to the UE 110 in an offline manner. For instance, the data generator 302 could be sent via a specific cloud server or a cloud service. In other examples the data generator 302 could be sent to the UE 110 in an online manner. For instance, the data generator 302 could be sent via 3GPP signaling such as RRC. At block 806 the gNB 120 identifies an ML feature and / or a functionality to the UE 110. The functionality can comprise a use case for the ML model. This indication can be transmitted using any suitable signaling.
[0137] At block 808 tuning parameters 300 are sent from the gNB 120 to the UE 110. The tuning parameters 300 can be sent as a hierarchy or in any other suitable format.
[0138] In response to receiving the tuning parameters 300 the UE 110 transmits an acknowledge signal at block 810.
[0139] At block 812 the gNB 120 can generate training data for the indicated ML feature and functionality. Other types of data such as monitoring data could be generated in other examples. The purpose of the generation of this data can be to enable a cross check of the data generated by the data generator 302 in the UE 110.
[0140] At block 814 the UE 110 uses the tuned data generator 302 to generate the training data 304. The data generator 302 is tuned using the tuning parameters 300 that were received at block 808.
[0141] In some examples the UE 110 can also compute one or more cross-checking criteria. The cross-checking criteria can define circumstances in which the UE 110 will perform a cross check of the data generated by the data generator 302 against the data generated by the gNB 120 at block 812. The cross-checking criteria can comprise some data statistics or any other suitable criteria.
[0142] If a cross-check of the data is to be used, then at block 816 the data to be checked is sent from the UE 110 to the gNB 120 for evaluation by the gNB 120. The gNB 120 can then perform the evaluation of the cross-checking data. If the data passes the cross-check then, at block 818 the gNB 120 will send an acknowledgement to the UE 110. At block 820 the UE 110 can start to train the ML model using the training data 304 generated by the training date generator 302. The training data 304 may have been checked by the gNB 120 but this check need not be performed in all implementations.
[0143] The ML model can be trained in an online or offline manner. In some examples both online and offline training could be used. For instance, in some examples off line training could be used for training data that has been generated for only tuning parameters 300 in upper levels of a hierarchy. For instance, a first level of tuning parameters 300 could indicate a prototype 400 that is to be used for a data generator. The data generator 302 tuned at this level could be used offline to generate synthetic training data for training an ML model. A further fine-tuned training data generator 302 could be used online for further training data that could be used to update the ML model for a more specific use case.
[0144] At block 822 the UE 110 sends a measurement report to the gNB 120. The measurement report is based on the ML model inference. The measurement report is specific to the use case.
[0145] Fig. 9 shows an example signaling chart that could be used to implement some examples of the disclosure. In the example of Fig. 9 the ML model and the data generator 302 can be implemented by the gNB 120.
[0146] At block 900 the UE110 selects an object for a data generator 302. This could be vendor specific. Different UEs 110 could have different data generators 302 and / or different objects.
[0147] At block 902 the UE 110 capability is exchanged with the gNB 120. This can be exchanged within RRC signaling or any other suitable type of signaling. This can enable the UE 110 to indicate the capabilities that the UE 110 has in terms of a data generator 302 and / or ML models and / or any other relevant capability.
[0148] At block 904 the data generator 302 is sent from the UE 110 to the gNB120. This can be the data generator 302 that was selected at block 900. The data generator 302 can be sent from the UE 110 to the gNB 120 using any suitable means. In some examples the data generator 302 can be sent from the UE 110 to the gNB120 in an offline manner. For instance, the data generator 302 could be sent via a specific cloud server or a cloud service. In other examples the data generator 302 could be sent UE 110 to the gNB 120 in an online manner. For instance, the data generator 302 could be sent via 3GPP signaling such as RRC.
[0149] At block 906 the UE 110 identifies an ML feature and / or a functionality to the gNB 120. The functionality can comprise a use case for the ML model. This indication can be transmitted using any suitable signaling.
[0150] At block 908 the UE 110 generates the tuning parameters 300. The UE 110 can obtain or derive the tuning parameters 300 from measurements made by the UE 110.
[0151] At block 910 tuning parameters 300 are sent from the UE 110 to the gNB 120. The tuning parameters 300 can be sent as a hierarchy or in any other suitable format.
[0152] In response to receiving the tuning parameters 300 the gNB 120 transmits an acknowledge signal at block 912.
[0153] At block 914 the UE 110 can generate training data for the indicated ML feature and functionality. The purpose of the generation of this data can be to enable a cross check of the data generated by the data generator 302 in the gNB 120.
[0154] At block 916 the gNB 120 uses the tuned data generator 302 to generate the training data 304. Other types of data such as monitoring data could be used in other examples. The data generator 302 is tuned using the tuning parameters 300 that were received at block 910.
[0155] In some examples the gNB 120 can also compute one or more cross-checking criteria. The cross-checking criteria can define circumstances in which the gNB 120 will perform a cross check of the data generated by the data generator 302 against the data generated by the UE 110 at block 914. The cross-checking criteria can comprise some data statistics or any other suitable criteria. If a cross-check of the data is to be used, then at block 918 the data to be checked is sent from the gNB 120 to the UE 110 for evaluation by the UE 110. The UE 110 can then perform the evaluation of the cross-checking data. If the data passes the crosscheck then, at block 920 the UE 110 will send an acknowledgement to the gNB 120.
[0156] At block 922 the gNB 120 can start to train the ML model using the training data 304 generated by the training date generator 302. The training data 304 may have been checked by the UE 110 but this check need not be performed in all implementations.
[0157] The ML model can be trained in an online or offline manner. In some examples both online and offline training could be used as described above.
[0158] At block 924 the UE 110 sends a measurement report to the gNB 120. At block 926 the gNB 120 uses the trained ML model to make inferences based on the received measurement reports. The ML model has been trained to the specific scenario of the UE 110.
[0159] Examples of the disclosure provide multiple advantages. The data generators 302 can be specific to particular functions and / or scenarios. This can enable them to be used to replace Life Cycle management (LCM) of ML models such as activation, deactivation, monitoring, updating, retraining, or any other relevant functions.
[0160] Examples of the disclosure also avoid the reporting of large data sets, or the reporting of smaller data sets and then performing augmentation of the smaller data set. The reporting of this data can be avoided because the tuning parameters 300 and the data generator 302 contain sufficient information to enable the data 302 to be generated at the device that uses the data. This results in a huge overhead reduction and can also provide increased privacy.
[0161] Further there is no need to exchange ML models between respective entities in the network 100. This can provide for improved privacy. Also, some vendors might want to avoid exchanging ML models through third party networks 100. The use of prototypes 400 and / or a hierarchy of tuning parameters 300 also provides further benefits. For instance, selecting a particular protype 400 might provide a large amount of information for the generation of the data but only requires a small amount of data to be exchanged. Also, the selection of a prototype only needs to take place on seldom occasions. Tuning parameters 300 within respective levels of a hierarchy can then be used to tune the data generator 302 to an appropriate level. The hierarchy can provide an efficient format for exchanging the tuning parameters 300 and the information needed to generate the data 304.
[0162] The data generators 302 can be used for multiple different uses. For example, a data generator 302 could be used for channel prediction, CSI compression, positioning, beam management, and / or any other suitable use. A data generator 302 can be adapted for a specific use case by selecting an appropriate prototype 400 or by any other suitable process.
[0163] In some examples predefined rules can be used for combining ML models with specific prototypes 400 for the data generator 302. This can help to provide fast training and effective implementation.
[0164] In some examples the training of the ML model with the data 304 generated by the data generator 302 can lead to an ML model which is the same as, or substantially the same as, an ML model that has been trained with the data obtained directly from the measurements. In this case the use of the tuning parameters 300 and the data generator 302 effectively provide lossless compression.
[0165] Fig. 10 illustrates an example apparatus 1000. The apparatus 1000 could be provided within an entity such as a UE 110 or a node apparatus such as a gNB 120. The apparatus 1000 can comprise a controller for of a controller such entities. Implementation of the apparatus 1000 may be as controller circuitry. The apparatus 1000 may be implemented in hardware alone, have certain aspects in software including firmware alone or can be a combination of hardware and software (including firmware). As illustrated in Fig. 10 the apparatus 1000 can be implemented using instructions that enable hardware functionality, for example, by using executable instructions of a computer program 1006 in a general-purpose or special-purpose processor 1002 that may be stored on a computer readable storage medium (disk, memory etc.) to be executed by such a processor 1002.
[0166] The processor 1002 is configured to read from and write to the memory 1004. The processor 1002 may also comprise an output interface via which data and / or commands are output by the processor 1002 and an input interface via which data and / or commands are input to the processor 1002.
[0167] The memory 1004 stores a computer program 1006 comprising computer program instructions (computer program code) that controls the operation of the apparatus when loaded into the processorlO. The computer program instructions, of the computer program 1006, provide the logic and routines that enables the apparatus to perform the methods illustrated in the Figs. The processor 1002 by reading the memory 1004 is able to load and execute the computer program 1006.
[0168] The apparatus 1000 therefore comprises: at least one processor 1002; and at least one memory 1004 storing instructions that, when executed by the at least one processor 1002, cause an entity 110, 120 at least to perform: receiving 200 one or more tuning parameters 300 where the tuning parameters 300 are based, at least in part, on data obtained from measurements of physical conditions such that the tuning parameters 300 represent, at least in part, the physical conditions; using the tuning parameters 300 to adjust 202 a data generator 302 where the data generator 302 comprises a model that represents the physical conditions; using the adjusted data generator to generate 204 data 304 for a machine learning model.
[0169] The computer program 1006 may arrive at the UE 110 or gNB 120 via any suitable delivery mechanism 1008. The delivery mechanism 1008 may be, for example, a machine-readable medium, a computer-readable medium, a non-transitory computer- readable storage medium, a computer program product, a memory device, a record medium such as a Compact Disc Read-Only Memory (CD-ROM) or a Digital Versatile Disc (DVD) or a solid-state memory, an article of manufacture that comprises or tangibly embodies the computer program 1006. The delivery mechanism may be a signal configured to reliably transfer the computer program 1006. The apparatus may propagate or transmit the computer program 1006 as a computer data signal.
[0170] The computer program 1006 can comprise computer program instructions for causing an entity 110, 120 to perform at least the following or for performing at least the following: receiving 200 one or more tuning parameters 300 where the tuning parameters 300 are based, at least in part, on data obtained from measurements of physical conditions such that the tuning parameters 300 represent, at least in part, the physical conditions; using the tuning parameters 300 to adjust 202 a data generator 302 where the data generator 302 comprises a model that represents the physical conditions; using the adjusted data generator to generate 204 data 304 for a machine learning model.
[0171] The computer program instructions may be comprised in a computer program, a non- transitory computer readable medium, a computer program product, a machine- readable medium. In some but not necessarily all examples, the computer program instructions may be distributed over more than one computer program.
[0172] Although the memory 1004 is illustrated as a single component / circuitry it may be implemented as one or more separate components / circuitry some or all of which may be integrated / removable and / or may provide permanent / semi-permanent / dynamic / cached storage.
[0173] Although the processor 1002 is illustrated as a single component / circuitry it may be implemented as one or more separate components / circuitry some or all of which may be integrated / removable. The processor 1002 may be a single core or multi-core processor. References to ‘computer-readable storage medium’, ‘computer program product’, ‘tangibly embodied computer program’ etc. or a ‘controller’, ‘computer’, ‘processor’ etc. should be understood to encompass not only computers having different architectures such as single / multi- processor architectures and sequential (Von Neumann) / parallel architectures but also specialized circuits such as field-programmable gate arrays (FPGA), application specific circuits (ASIC), signal processing devices and other processing circuitry. References to computer program, instructions, code etc. should be understood to encompass software for a programmable processor or firmware such as, for example, the programmable content of a hardware device whether instructions for a processor, or configuration settings for a fixed-function device, gate array or programmable logic device etc.
[0174] As used in this application, the term ‘circuitry’ may refer to one or more or all of the following:
[0175] (a) hardware-only circuitry implementations (such as implementations in only analog and / or digital circuitry) and
[0176] (b) combinations of hardware circuits and software, such as (as applicable):
[0177] (i) a combination of analog and / or digital hardware circuit(s) with software / firmware and
[0178] (ii) any portions of hardware processor(s) with software (including digital signal processor(s)), software, and memory(ies) that work together to cause an apparatus, such as a mobile phone or server, to perform various functions and
[0179] (c) hardware circuit(s) and or processor(s), such as a microprocessor(s) or a portion of a microprocessor(s), that requires software (e.g. firmware) for operation, but the software may not be present when it is not needed for operation.
[0180] This definition of circuitry applies to all uses of this term in this application, including in any claims. As a further example, as used in this application, the term circuitry also covers an implementation of merely a hardware circuit or processor and its (or their) accompanying software and / or firmware. The term circuitry also covers, for example and if applicable to the particular claim element, a baseband integrated circuit for a mobile device or a similar integrated circuit in a server, a cellular network device, or other computing or network device. The stages illustrated in the Figs, can represent steps in a method and / or sections of code in the computer program 1006. The illustration of a particular order to the blocks does not necessarily imply that there is a required or preferred order for the blocks and the order and arrangement of the block may be varied. Furthermore, it can be possible for some blocks to be omitted.
[0181] The term ‘comprise’ is used in this document with an inclusive not an exclusive meaning. That is any reference to X comprising Y indicates that X may comprise only one Y or may comprise more than one Y. If it is intended to use ‘comprise’ with an exclusive meaning then it will be made clear in the context by referring to “comprising only one...” or by using “consisting”.
[0182] In this description, the wording ‘connect’, ‘couple’ and ‘communication’ and their derivatives mean operationally connected / coupled / in communication. It should be appreciated that any number or combination of intervening components can exist (including no intervening components), i.e., so as to provide direct or indirect connection / coupling / communication. Any such intervening components can include hardware and / or software components.
[0183] As used herein, the term "determine / determining" (and grammatical variants thereof) can include, not least: calculating, computing, processing, deriving, measuring, investigating, identifying, looking up (for example, looking up in a table, a database or another data structure), ascertaining and the like. Also, "determining" can include receiving (for example, receiving information), accessing (for example, accessing data in a memory), obtaining and the like. Also, " determine / determining" can include resolving, selecting, choosing, establishing, and the like.
[0184] In this description, reference has been made to various examples. The description of features or functions in relation to an example indicates that those features or functions are present in that example. The use of the term ‘example’ or ‘for example’ or ‘can’ or ‘may’ in the text denotes, whether explicitly stated or not, that such features or functions are present in at least the described example, whether described as an example or not, and that they can be, but are not necessarily, present in some of or all other examples. Thus ‘example’, ‘for example’, ‘can’ or ‘may’ refers to a particular instance in a class of examples. A property of the instance can be a property of only that instance or a property of the class or a property of a sub-class of the class that includes some but not all of the instances in the class. It is therefore implicitly disclosed that a feature described with reference to one example but not with reference to another example, can where possible be used in that other example as part of a working combination but does not necessarily have to be used in that other example.
[0185] Although examples have been described in the preceding paragraphs with reference to various examples, it should be appreciated that modifications to the examples given can be made without departing from the scope of the claims.
[0186] Features described in the preceding description may be used in combinations other than the combinations explicitly described above.
[0187] Although functions have been described with reference to certain features, those functions may be performable by other features whether described or not.
[0188] Although features have been described with reference to certain examples, those features may also be present in other examples whether described or not.
[0189] The term ‘a’, ‘an’ or ‘the’ is used in this document with an inclusive not an exclusive meaning. That is any reference to X comprising a / an / the Y indicates that X may comprise only one Y or may comprise more than one Y unless the context clearly indicates the contrary. If it is intended to use ‘a’, ‘an’ or ‘the’ with an exclusive meaning then it will be made clear in the context. In some circumstances the use of ‘at least one’ or ‘one or more’ may be used to emphasis an inclusive meaning but the absence of these terms should not be taken to infer any exclusive meaning.
[0190] The presence of a feature (or combination of features) in a claim is a reference to that feature or (combination of features) itself and also to features that achieve substantially the same technical effect (equivalent features). The equivalent features include, for example, features that are variants and achieve substantially the same result in substantially the same way. The equivalent features include, for example, features that perform substantially the same function, in substantially the same way to achieve substantially the same result.
[0191] In this description, reference has been made to various examples using adjectives or adjectival phrases to describe characteristics of the examples. Such a description of a characteristic in relation to an example indicates that the characteristic is present in some examples exactly as described and is present in other examples substantially as described.
[0192] The above description describes some examples of the present disclosure however those of ordinary skill in the art will be aware of possible alternative structures and method features which offer equivalent functionality to the specific examples of such structures and features described herein above and which for the sake of brevity and clarity have been omitted from the above description. Nonetheless, the above description should be read as implicitly including reference to such alternative structures and method features which provide equivalent functionality unless such alternative structures or method features are explicitly excluded in the above description of the examples of the present disclosure.
[0193] Whilst endeavoring in the foregoing specification to draw attention to those features believed to be of importance it should be understood that the Applicant may seek protection via the claims in respect of any patentable feature or combination of features hereinbefore referred to and / or shown in the drawings whether or not emphasis has been placed thereon. l / we claim:
Claims
CLAIMS1 . An apparatus comprising means for: receiving one or more tuning parameters where the tuning parameters are based, at least in part, on data obtained from measurements of physical conditions such that the tuning parameters represent, at least in part, the physical conditions; using the tuning parameters to adjust a data generator where the data generator comprises a model that represents the physical conditions; using the adjusted data generator to generate data for a machine learning model.
2. An apparatus as claimed in claim 1 wherein the measurements comprise measurements of physical conditions of a channel such that the tuning parameters represent, at least in part, the physical conditions of the channel and the data generator comprises a channel model that represents the channel.
3. An apparatus as claimed in claim 1 wherein the measurements comprise measurements of reference signal received power estimations for a set of such that the tuning parameters represent, at least in part, the signal paths for a geometrical environment and the data generator comprises the signal paths.
4. An apparatus as claimed in any preceding claim wherein the means are for using the data generated by the data generator to train the machine learning model.
5. An apparatus as claimed in any preceding claim wherein the data generated by the adjusted data generator comprises measurement data from which the tuning parameters were based.
6. An apparatus as claimed in any preceding claim wherein the means are for configuring the data generator to generate data for a selected functionality of a machine learning model.
7. An apparatus as claimed in any preceding claim wherein a plurality of prototypes are available for the data generator wherein respective prototypes areconfigured for different radio channel scenarios and the means are for selecting a prototype for use in the data generator based on a radio channel scenario.
8. An apparatus as claimed in any preceding claim wherein the tuning parameters are received in a hierarchy wherein tuning parameters from different levels of the hierarchy are configured to be used to adjust different levels of the data generator.
9. An apparatus as claimed in any of claims 7 to 8 wherein the hierarchy comprises at least one of: a level of tuning parameters indicating a prototype for the data generator; a level of tuning parameters indicating information specific to a prototype; a level of tuning parameters comprising data obtained from or derived from measurements.
10. An apparatus as claimed in any preceding claim wherein the means are for performing a check of the generated data by comparing at least part of the generated data to measurement data.
11. An apparatus as claimed in any preceding claim wherein the means are for receiving data samples affected by anomalies and extracting an impact of the anomalies to the data generator and adjusting the data generator based on the impact.
12. An apparatus as claimed in any preceding claim wherein the apparatus is a network entity.
13. An apparatus as claimed in any of claims 1 to 11 wherein the apparatus is a user equipment.
14. A method comprising: receiving one or more tuning parameters where the tuning parameters are based, at least in part, on data obtained from measurements of physical conditions such that the tuning parameters represent, at least in part, the physical conditions; using the tuning parameters to adjust a data generator where the data generator comprises a model that represents the physical conditions;using the adjusted data generator to generate data for a machine learning model.
15. A computer program comprising instructions which, when executed by an apparatus, cause the apparatus to perform at least: receiving one or more tuning parameters where the tuning parameters are based, at least in part, on data obtained from measurements of physical conditions such that the tuning parameters represent, at least in part, the physical conditions; using the tuning parameters to adjust a data generator where the data generator comprises a model that represents the physical conditions; using the adjusted data generator to generate data for a machine learning model.
Citation Information
Patent Citations
Beam management using adaptive learning
US10666342B1
Automated and intelligent data channel optimizers with sensitivity assessment
US20220171727A1
Channel state information values-based estimation of reference signal received power values for wireless networks
WO2021209145A1
Measurement configuration for local area machine learning radio resource management
WO2021224705A1
