Training Data Collection

Abstract

1. An apparatus comprising: a coordinator function configured to send a first configuration message to a first collector function, the first configuration message including information for configuring the first collector function to collect first training data, transform the first training data to generate transformed first training data, and report the transformed first training data.

Description

[Technical Field]

[0001] Various example embodiments relate to collecting training data. [Background technology]

[0002] In wireless telecommunications networks, network nodes collect and process training data to train machine learning (ML) models to improve network operation. While techniques exist for collecting and processing training data, they can produce unexpected results. Therefore, it is desirable to provide improved techniques for collecting and processing training data. Summary of the Invention

[0003] The scope of protection sought for various example embodiments of the invention is set forth in the independent claims. Example embodiments and features described herein that are not within the scope of the independent claims, if any, should be interpreted as useful examples for understanding various embodiments of the invention.

[0004] According to various, but not all, example embodiments of the present invention, an apparatus is provided that includes a coordinator function configured to send a first configuration message to a first collector function, the first configuration message including information for configuring the first collector function to collect first training data, transform the first training data to generate transformed first training data, and report the transformed first training data.

[0005] The first training data may include an N-dimensional matrix of values ​​collected by a first collector function.

[0006] The value may include channel information. The channel information may be about a wireless link between the entity hosting the collector function and the transmitter. The channel information may include beamforming and / or channel values.

[0007] The first configuration message can include information for configuring the first collector function to transform the first training data by removing specified vendor-specific data and / or artifacts to generate transformed first training data. The specified vendor-specific data and / or artifacts can include radio frequency receiver delays, number of radio frequency chains, and the like.

[0008] The first configuration message can include information for configuring the first collector function to transform the first training data to generate transformed first training data with a target reconstruction.

[0009] The first configuration message may include information for configuring the first collector function to report the transformed first training data in a transformed training data format and / or at a specified reporting period.

[0010] The transformed first training data may include an N-dimensional matrix of values ​​collected by the first collector function.

[0011] The first configuration message may include information for configuring the first collector function to report the transformed first training data to the coordinator function and / or the training function.

[0012] The first configuration message may include information for configuring the first collector function to report the transformed first training data to both the first training function and the second training function.

[0013] The coordinator function may be configured to send a second configuration message to the second collector function, the second configuration message including information for configuring the second collector function to collect second training data, reconstruct the second training data to generate transformed second training data, and report the transformed second training data.

[0014] The coordinator function may be configured to combine the received transformed training data to form combined transformed training data. The received transformed training data may be from multiple collector functions and / or from the same collector function at different times.

[0015] The coordinator function may be configured to combine the received transformed training data by performing overlapping, averaging, filtering, pruning, puncturing, scaling, and / or normalization.

[0016] The coordinator function may be configured to send the combined transformed training data to the training function.

[0017] The coordinator function may be configured to send a configuration message to a collector function provided by a vendor common to the coordinator function and the collector function, the configuration message including instructions to configure the collector function to collect training data and report the training data.

[0018] The coordinator function may be configured to send a conversion configuration message to the training function, the conversion configuration message including information for configuring the training function to convert the received transformed training data and / or the received combined transformed training data into converted training data.

[0019] The conversion configuration message may include information for configuring the training function to convert the received transformed training data and / or the received combined transformed training data into converted training data by performing dilation, overlap, averaging, filtering, pruning, puncturing, normalization, scaling, and / or translation. The conversion may be by an appropriate filtering or projection operation.

[0020] The coordinator function may be configured to send a first conversion configuration message to the first training function, the first conversion configuration message including information for configuring the first training function to convert the received transformed training data and / or the received combined transformed training data into the first converted training data, and to send a second conversion configuration message to the second training function, the second conversion configuration message including information for configuring the second training function to convert the received transformed training data and / or the received combined transformed training data into the second converted training data.

[0021] The messages and / or data may be transmitted on a physical sidelink shared channel, a physical downlink shared channel, a physical uplink shared channel, and / or any wireless medium channel. The messages and / or data may be transmitted on a physical sidelink control channel, a physical downlink common control channel, and / or a physical uplink common channel. The messages and / or data may be transmitted other than on the air interface, such as via the F1, Xn, and / or NG interfaces. The messages and / or data may be transmitted via the O-RAN A1, E1, E2, and F1 interfaces. It will be appreciated that these are applicable to 4G, 5G, and 6G systems.

[0022] The training data, transformed training data, and / or transformed training data may include a data format specified by a combination of parameters associated with the training data, such as sampling resolution, array shape and / or length, e.g., scalar, vector, matrix, and the like.

[0023] According to various, but not all, example embodiments of the present invention, a method is provided that includes sending a first configuration message to a first collector function, the first configuration message including information for configuring the first collector function to collect first training data, transform the first training data to generate transformed first training data, and report the transformed first training data.

[0024] The first training data may include an N-dimensional matrix of values ​​collected by a first collector function.

[0025] The value may include channel information. The channel information may be about a wireless link between the entity hosting the collector function and the transmitter. The channel information may include beamforming and / or channel values.

[0026] The first configuration message can include information for configuring the first collector function to transform the first training data by removing specified vendor-specific data and / or artifacts to generate transformed first training data. The specified vendor-specific data and / or artifacts can include radio frequency receiver delays, number of radio frequency chains, and the like.

[0027] The first configuration message can include information for configuring the first collector function to transform the first training data to generate transformed first training data with a target reconstruction.

[0028] The first configuration message may include information for configuring the first collector function to report the transformed first training data in a transformed training data format and / or at a specified reporting period.

[0029] The transformed first training data may include an N-dimensional matrix of values ​​collected by the first collector function.

[0030] The first configuration message may include information for configuring the first collector function to report the transformed first training data to the coordinator function and / or the training function.

[0031] The first configuration message may include information for configuring the first collector function to report the transformed first training data to both the first training function and the second training function.

[0032] The method may include sending a second configuration message to a second collector function, the second configuration message including information for configuring the second collector function to collect second training data, reconstruct the second training data to generate transformed second training data, and report the transformed second training data.

[0033] The method can include combining the received transformed training data to form combined transformed training data. The received transformed training data can be from multiple collector functions and / or from the same collector function at different times.

[0034] The method may include combining the received transformed training data by overlapping, averaging, filtering, pruning, puncturing, scaling, and / or normalizing.

[0035] The method can include sending the combined transformed training data to a training function.

[0036] The sending may be performed by a coordinator function, and the method may include sending a configuration message to a collector function provided by a vendor common to the coordinator function and the collector function, the configuration message including instructions to configure the collector function to collect training data and report the training data.

[0037] The method may include sending a conversion configuration message to a training function, the conversion configuration message including information for configuring the training function to convert the received transformed training data and / or the received combined transformed training data into converted training data.

[0038] The conversion configuration message may include information for configuring the training function to convert the received transformed training data and / or the received combined transformed training data into converted training data by performing dilation, overlap, averaging, filtering, pruning, puncturing, normalization, scaling, and / or translation. The conversion may be by an appropriate filtering or projection operation.

[0039] The method may include sending a first conversion configuration message to a first training function, the first conversion configuration message including information for configuring the first training function to convert the received transformed training data and / or the received combined transformed training data into first converted training data, and sending a second conversion configuration message to a second training function, the second conversion configuration message including information for configuring the second training function to convert the received transformed training data and / or the received combined transformed training data into second converted training data.

[0040] The messages and / or data may be transmitted on a physical sidelink shared channel, a physical downlink shared channel, a physical uplink shared channel, and / or any wireless medium channel. The messages and / or data may be transmitted on a physical sidelink control channel, a physical downlink common control channel, and / or a physical uplink common channel. The messages and / or data may be transmitted other than on the air interface, such as via the F1, Xn, and / or NG interfaces. The messages and / or data may be transmitted via the O-RAN A1, E1, E2, and F1 interfaces. It will be appreciated that these are applicable to 4G, 5G, and 6G systems.

[0041] The training data, transformed training data, and / or transformed training data may include a data format specified by a combination of parameters associated with the training data, such as sampling resolution, array shape and / or length, e.g., scalar, vector, matrix, and the like.

[0042] According to various, but not all, example embodiments of the present invention, an apparatus is provided that includes at least one processor and at least one memory that stores instructions that, when executed by the at least one processor, cause the apparatus to perform at least the methods and example embodiments thereof described above.

[0043] In accordance with various, but not all, example embodiments of the present invention, a non-transitory computer-readable medium is provided that includes stored program instructions for implementing the above-described methods and example embodiments thereof.

[0044] According to various, but not all, example embodiments of the present invention, an apparatus is provided that includes a collector function configured to receive a configuration message from a coordinator function, the configuration message including information for configuring the collector function to collect training data, transform the training data to generate transformed training data, and report the transformed training data.

[0045] The training data may include an N-dimensional matrix of values ​​collected by a collector function.

[0046] The value may include channel information. The channel information may be about a wireless link between the entity hosting the collector function and the transmitter. The channel information may include beamforming and / or channel values.

[0047] The configuration message may include information for configuring the collector function to transform the training data to generate transformed training data by removing specified vendor-specific data and / or artifacts, which may include radio frequency receiver delays, several radio frequency chains, and the like.

[0048] The configuration message may include information for configuring the collector function to transform the training data to generate transformed training data with a target reconstruction.

[0049] The configuration message may include information for configuring the collector function to report the transformed training data in a transformed training data format and / or at a specified reporting interval.

[0050] The transformed training data may include an N-dimensional matrix of values ​​collected by the collector function.

[0051] The configuration message may include information for configuring the collector function to report the transformed training data to the coordinator function and / or the training function.

[0052] The configuration message may include information for configuring the collector function to report the transformed training data to both the first training function and the second training function.

[0053] The collector function may be configured to receive a configuration message from a coordinator function provided by a vendor common to the collector function, the configuration message including instructions to configure the collector function to collect training data and report the training data.

[0054] The messages and / or data may be transmitted on a physical sidelink shared channel, a physical downlink shared channel, a physical uplink shared channel, and / or any wireless medium channel. The messages and / or data may be transmitted on a physical sidelink control channel, a physical downlink common control channel, and / or a physical uplink common channel. The messages and / or data may be transmitted other than on the air interface, such as via the F1, Xn, and / or NG interfaces. The messages and / or data may be transmitted via the O-RAN A1, E1, E2, and F1 interfaces. It will be appreciated that these are applicable to 4G, 5G, and 6G systems.

[0055] The training data, transformed training data, and / or transformed training data may include a data format specified by a combination of parameters associated with the training data, such as sampling resolution, array shape and / or length, e.g., scalar, vector, matrix, and the like.

[0056] According to various, but not all, example embodiments of the present invention, a method is provided that includes receiving a configuration message from a coordinator function, the configuration message including information for configuring a collector function to collect training data, transform the training data to generate transformed training data, and report the transformed training data.

[0057] The training data may include an N-dimensional matrix of values ​​collected by a collector function.

[0058] The value may include channel information. The channel information may be about a wireless link between the entity hosting the collector function and the transmitter. The channel information may include beamforming and / or channel values.

[0059] The configuration message may include information for configuring the collector function to transform the training data to generate transformed training data by removing specified vendor-specific data and / or artifacts, which may include radio frequency receiver delays, several radio frequency chains, and the like.

[0060] The configuration message may include information for configuring the collector function to transform the training data to generate transformed training data with a target reconstruction.

[0061] The configuration message may include information for configuring the collector function to report the transformed training data in a transformed training data format and / or at a specified reporting interval.

[0062] The transformed training data may include an N-dimensional matrix of values ​​collected by the collector function.

[0063] The configuration message may include information for configuring the collector function to report the transformed training data to the coordinator function and / or the training function.

[0064] The configuration message may include information for configuring the collector function to report the transformed training data to both the first training function and the second training function.

[0065] The method may include receiving a configuration message from a coordinator function provided by a vendor common to the collector function, the configuration message including instructions to configure the collector function to collect training data and report the training data.

[0066] The messages and / or data may be transmitted on a physical sidelink shared channel, a physical downlink shared channel, a physical uplink shared channel, and / or any wireless medium channel. The messages and / or data may be transmitted on a physical sidelink control channel, a physical downlink common control channel, and / or a physical uplink common channel. The messages and / or data may be transmitted other than on the air interface, such as via the F1, Xn, and / or NG interfaces. The messages and / or data may be transmitted via the O-RAN A1, E1, E2, and F1 interfaces. It will be appreciated that these are applicable to 4G, 5G, and 6G systems.

[0067] The training data, transformed training data, and / or transformed training data may include a data format specified by a combination of parameters associated with the training data, such as sampling resolution, array shape and / or length, e.g., scalar, vector, matrix, and the like.

[0068] According to various, but not all, example embodiments of the present invention, an apparatus is provided that includes at least one processor and at least one memory that stores instructions that, when executed by the at least one processor, cause the apparatus to perform at least the methods and example embodiments thereof described above.

[0069] In accordance with various, but not all, example embodiments of the present invention, a non-transitory computer-readable medium is provided that includes stored program instructions for implementing the above-described methods and example embodiments thereof.

[0070] According to various, but not all, example embodiments of the present invention, an apparatus is provided that includes a training function configured to receive a conversion configuration message from a coordinator function, the conversion configuration message including information for configuring the training function to convert received transformed training data and / or received combined transformed training data into converted training data.

[0071] The conversion configuration message may include information for configuring the training function to convert the received transformed training data and / or the received combined transformed training data into converted training data by performing dilation, overlap, averaging, filtering, pruning, puncturing, normalization, scaling, and / or translation. The conversion may be by an appropriate filtering or projection operation.

[0072] The training data may include an N-dimensional matrix of values ​​collected by a collector function.

[0073] The value may include channel information. The channel information may be about a wireless link between the entity hosting the collector function and the transmitter. The channel information may include beamforming and / or channel values.

[0074] The transformed first training data may include an N-dimensional matrix of values ​​collected by a collector function.

[0075] The training function may be configured to combine the received transformed training data to form combined transformed training data.

[0076] The training function may be configured to combine the received transformed training data by performing overlapping, averaging, filtering, pruning, puncturing, scaling, and / or normalization.

[0077] The messages and / or data may be transmitted on a physical sidelink shared channel, a physical downlink shared channel, a physical uplink shared channel, and / or any wireless medium channel. The messages and / or data may be transmitted on a physical sidelink control channel, a physical downlink common control channel, and / or a physical uplink common channel. The messages and / or data may be transmitted other than on the air interface, such as via the F1, Xn, and / or NG interfaces. The messages and / or data may be transmitted via the O-RAN A1, E1, E2, and F1 interfaces. It will be appreciated that these are applicable to 4G, 5G, and 6G systems.

[0078] The training data, transformed training data, and / or transformed training data may include a data format specified by a combination of parameters associated with the training data, such as sampling resolution, array shape and / or length, e.g., scalar, vector, matrix, and the like.

[0079] According to various, but not all, example embodiments of the present invention, a method is provided that includes receiving a conversion configuration message from a coordinator function, the conversion configuration message including information for configuring a training function to convert received transformed training data and / or received combined transformed training data into converted training data.

[0080] The conversion configuration message may include information for configuring the training function to convert the received transformed training data and / or the received combined transformed training data into converted training data by performing dilation, overlap, averaging, filtering, pruning, puncturing, normalization, scaling, and / or translation. The conversion may be by an appropriate filtering or projection operation.

[0081] The training data may include an N-dimensional matrix of values ​​collected by a collector function.

[0082] The value may include channel information. The channel information may be about a wireless link between the entity hosting the collector function and the transmitter. The channel information may include beamforming and / or channel values.

[0083] The transformed first training data may include an N-dimensional matrix of values ​​collected by a collector function.

[0084] The method can include combining the received transformed training data to form combined transformed training data.

[0085] The method may include combining the received transformed training data by overlapping, averaging, filtering, pruning, puncturing, scaling, and / or normalizing.

[0086] The messages and / or data may be transmitted on a physical sidelink shared channel, a physical downlink shared channel, a physical uplink shared channel, and / or any wireless medium channel. The messages and / or data may be transmitted on a physical sidelink control channel, a physical downlink common control channel, and / or a physical uplink common channel. The messages and / or data may be transmitted other than on the air interface, such as via the F1, Xn, and / or NG interfaces. The messages and / or data may be transmitted via the O-RAN A1, E1, E2, and F1 interfaces. It will be appreciated that these are applicable to 4G, 5G, and 6G systems.

[0087] The training data, transformed training data, and / or transformed training data may include a data format specified by a combination of parameters associated with the training data, such as sampling resolution, array shape and / or length, e.g., scalar, vector, matrix, and the like.

[0088] According to various, but not all, example embodiments of the present invention, an apparatus is provided that includes at least one processor and at least one memory that stores instructions that, when executed by the at least one processor, cause the apparatus to perform at least the methods and example embodiments thereof described above.

[0089] In accordance with various, but not all, example embodiments of the present invention, a non-transitory computer-readable medium is provided that includes stored program instructions for implementing the above-described methods and example embodiments thereof.

[0090] Further particular and preferred aspects are set out in the accompanying independent and dependent claims. Features of the dependent claims may be combined with features of the independent claims as appropriate, and in combinations other than those explicitly set out in the claims.

[0091] Where features of an apparatus are described as operable to perform a function, it will be understood that the features of the apparatus include features of an apparatus that provide this function or that are adapted or configured to perform this function.

[0092] Some example embodiments will now be described with reference to the accompanying drawings. [Brief explanation of the drawings]

[0093] [Figure 1] This is a diagram of insufficient training data. [Figure 2] 1 is a signal chart in which data from Vendor 2 is used to enhance Vendor 1's data. [Figure 3] 1 is a signal chart in which data from Vendor 2 is used to enhance data from Vendor 1, and a controller function (NR-C) performs a combination of vendor-agnostic / domain-invariant data (DID1 and DID2). [Figure 4] 2 and / or 3. FIG. 2 is a signaling chart when multiple NR_Es report DIDs to multiple NR_Ts—this approach can be combined with the approach in FIG. 2 and / or FIG. [Figure 5] 2 is a signal chart when multiple NR_Es report DIDs to multiple NR_Ts—this approach can be combined with the approaches in FIG. 2, FIG. 3, and / or FIG. 4. [Figure 6] 10 is a signal chart for an example of SL positioning. [Figure 7] 1 is a diagram of an enhanced positioning use case. The entries in the DVD (or DID) matrix, DVD(a,b), represent UE-specific positioning measurements taken at frequency resource a*Fs and time resource b*Ts, where Fs and Ts are UE-specific sampling frequency and sampling time, respectively. DETAILED DESCRIPTION OF THE INVENTION

[0094] Before discussing example embodiments in more detail, an overview will first be provided. Some example embodiments provide techniques in which network nodes in a wireless telecommunications network are provided with functionality to coordinate, collect, and use training data to train ML models to perform various network- and / or device-specific tasks, commonly referred to as radio resource management (RRM). Typically, a collection function in a network node whose training function uses the training data is provided by the same vendor as the network node can receive the training data along with values ​​and in a format known to the training function. In some embodiments, even a collection function in a network node whose training function uses the training data is provided by the same vendor as the network node provides the training data in an agnostic or immutable format. A collector function in a network node whose training function uses the training data is provided by a different vendor than the network node provides the training data in an agnostic or immutable format that does not disclose vendor-specific information about the capabilities of the entities that collected the data. This training data can be provided to a coordinator function that combines the received data or to a training function for combining the received data. The training function is typically provided with information such as transformation details that allow the combined data to be subsequently transformed or converted into a format that can then be used by the training function to train its (vendor-specific) ML model. This approach is useful for collecting a diverse range of training data from network nodes provided by other vendors in a consistent manner.

[0095] Some example embodiments relate to the Artificial Intelligence (AI) / Machine Learning (ML) Study Item (SI) [3GPP RP-213599] for the Rel-18 New Radio (NR) Air Interface. The SI aims to explore the benefits of enhancing the air interface with features that enable support for AI / ML-based algorithms for enhanced performance and / or reduced complexity / overhead. The goal of this SI is to lay the foundation for future air interface use cases to leverage AI / ML techniques. An initial set of use cases to be covered includes channel state information (CSI) feedback enhancements (e.g., overhead reduction, improved accuracy, prediction, etc.), beam management (e.g., beam prediction in time and / or spatial domains for overhead and latency reduction, improved beam selection accuracy, etc.), positioning accuracy enhancements, and the like. For these use cases, the benefits must be evaluated (using developed methodologies and defined key performance indicators (KPIs)), and the potential impact on the specification must be assessed, including physical (PHY) layer aspects, protocol aspects, etc. One of the major expected consequences of the SI is that "AI / ML approaches for selected sub-use cases must be diverse enough to support various requirements at the gNode B-User Equipment (gNB-UE) collaboration level." It should be noted that additional use cases may be addressed in the "AI / ML for Air Interface" work item (WI) phase. Starting with Release 18, it is highly likely that a wide variety of use cases and applications for ML in gNBs and UEs will be proposed. Rel. 17 Positioning Reference Unit (PRU) - A PRU is a 5G network entity that can be designated by a 5G NR network to assist in one or more positioning sessions.A PRU is a location-aware device or network node (e.g., a roadside unit, another UE, etc.) that can be activated on demand by a location management function (LMF) to perform specific positioning operations, such as measuring and / or transmitting specific positioning signals. The Liaison Document (LS) [R2-2106920] on RAN1 PRUs evaluates the use of location-aware Positioning Reference Units (PRUs) for positioning and observes improvements in the use of PRUs to enhance positioning performance. Note: The term "Positioning Reference Unit (PRU)" is used only as a term of art in this discussion. PRU does not necessarily imply the introduction of a new network node. The PRU, if agreed upon, can support at least some of the UE's Rel-16 positioning capabilities, which is up to RAN2. Positioning functions may include, but are not limited to, providing positioning measurements (e.g., Reference Signal Time Difference (RSTD), Reference Signal Received Power (RSRP), Receive-Transmit (Rx-Tx) Time Difference), transmitting an uplink (UL) sounding reference signal (SRS) for positioning—the PRU may be requested by the LMF to provide its own known location coordinate information to the LMF. If the PRU's antenna direction information is known, this information may also be requested by the LMF. ML models for Rel. 18 radio resource management (RRM) are expected to be vendor-specific and therefore trained with vendor-specific data. A foreseeable RAN consequence is that companies agree that vendor-specific ML models are trained for the same RRM function using only vendor-specific training data, so that training data is not exchanged between vendors. This is one reason why vendors (UE and / or gNB) may not want to share their data. That is, the data is UE-specific and often confidential, and the data gives vendors a competitive advantage by enabling them to create and deploy their ML-based solutions that overperform competitors' solutions.

[0096] As illustrated in Figure 1, data collection for training a vendor-specific ML model is therefore expected to be a lengthy process that will most likely result in a suboptimal training dataset that is unbalanced, i.e., there remains a large imbalance between the minority and majority labels, and sparse, i.e., the collected data does not well characterize all scenarios of interest.

[0097] To mitigate at least some of the above limitations and ensure that robust but vendor-specific ML models are trained, vendor-specific data would benefit from being artificially diversified and augmented for each vendor before being used to train vendor-based ML models. The process of artificially enhancing training data is called data augmentation, and the success of the procedure depends on two main factors: the quantity and quality of the initial training data, as well as the augmentation algorithm and its design assumptions. Nevertheless, there are no concrete proposals at this stage on how the required training data should be collected from different UEs in the network to enable a vendor-independent ML-enabled solution.

[0098] Some example embodiments provide techniques in which vendor-specific training data (hereafter referred to as domain-variable data) is diversified by using data from other vendors without exposing / sharing the domain-variable data set between vendors. To this end, the domain-variable data is initially relied upon, i.e., stripped of its vendor-specific nature. Three types of NR elements are included, and the combination of functions can be performed by one NR element.

[0099] ML Coordinator Function (NR-C) - An NR network element responsible for aggregating training data collected by different UEs and / or from multiple UE vendors and for defining the format of the vendor-independent or converted training data format that each UE must transport back to the NR-C. The NR-C may be a gNB-CU, NRT-RIC, NWDAF, etc.

[0100] ML Data Collector Function (NR-E) - An NR network element responsible for collecting / modifying raw data as commanded by the NR-C in a first or vendor-specific format and transporting the data to the NR-T or NR-C using a vendor-independent or converted format. An NR-E may be a UE, gNB-CU, RT-RIC, RSU, etc. Let NR-Ek denote an NR-E that collects training data specific to vendor k.

[0101] ML Training Function (NR-T)—An NR network element that combines training data from different sources, e.g., multiple NR-Es, and trains vendor-specific ML functions. The NR-T may be an NWDAF, LMF, serving gNB, or UE. The NR-T may reside within the same network element as the corresponding NR-E, e.g., a gNB or UE. Let NR-Tm represent the NR-T that trains the ML model for vendor m.

[0102] Provide training data from Vendor 2 to Vendor 1 An example embodiment is shown in Figure 2, in which a coordinator function configures data collector functions to provide training data to training functions. One of the data collector functions is from the same vendor as the training function and therefore can provide its training data in a format expected by the training function. The other data collector functions are from a different vendor and therefore are instructed to collect specified training data, convert this training data to a specified format, and provide this converted training data to the training function. The training function then converts the converted training data to match the format of the training data provided by the data collector functions from the same vendor as the training function, combines the training data, and uses this combined training data to train an ML model.

[0103] NR-C configures an element NR-Ek, k=1..K, for each vendor k to provide its training data in a given format.

[0104] Thus, in step S10, NR_C instructs NR_E1 to provide training data to NR_T1 in a vendor-specific format called domain variable data (DVD) when the training data and ML training are for the same vendor. In step S20, NR_E1 collects training data DVD1 in the vendor-specific format and reports this training data DVD1 to NT_T1 in step S30.

[0105] In step S40, NR_C instructs NR_E2 to provide training data to NR_T2 in a vendor-independent format called Domain Invariant Data (DID) format if the vendor for which training is required is different from the vendor for which the data is collected. NR_C defines how the DID is obtained at each NR-Ek side by providing details of a Vendor-Invariant Conversion (VIC) or transformation, as described in more detail below.

[0106] In step S50, each NR-Ek, k=1...K, converts domain variable data k (DVD(k)) into a DID using the VIC provided by the NR_C. The DID has a unique format (format, size, type, e.g., integer, real number, etc.) and may be scenario-specific; for example, a DID for beamforming may be a real 2D or 3D matrix where each entry is the RSRP of an RS at a given (time, frequency) or (time, frequency, space) location, and a DID for positioning enhancement may be a complex 3D matrix where each entry is the channel gain at a given (time, delay, space) location, generating a universally understood / agreed DID format that can be used across different UE vendors / domains. DVD-to-DID conversion (VIC) is configured by the NR-C and typically has the goal of stripping the DVD from sensitive information (UE-specific data payload, symbols, identifiers) and vendor-specific artifacts, such as UE-specific TX / RX delays and beam offsets. Note that the exact form of the VIC conversion to be applied to the DVD to obtain the DID is derived in the NR_E, which performs the conversion. The VIC parameterization may be fully / totally configured by the NR-C and can constrain the DID format, DID reporting periodicity, and target VIC performance; performance metrics are up to the use case.

[0107] In step S60, the NR-Ek sends its DIDk to the NR-Tm. Alternatively, the DID may be sent first to the NR-C, which then forwards the DID to each NR-Tm.

[0108] In step S70, NR-Tm diversifies DVDm using DIDs from domains k, k = 1:K, where m ≠ k, and each DID(k), k = 1...K, is used to reconstruct DID(k) into DVD(m). The reconstruction or conversion process consists of inputting DID(k) into a module that applies a domain-m-specific transformation to the DID and outputs an approximate DVD(m), called the reconstructed DVD: R-DVD(k → m). The DID-to-R-DVD conversion (invariant-to-variable conversion (IVC)) is the inverse of VIC and involves transforming the DID into the DVD format and also including domain-specific information, if available (depending on the use case). This conversion ensures that the R-DVD has the same properties and format as DVDm for a particular UE vendor m. The exact form of the IVC transformation needs to be known only to the vendor-specific functions (NR_T and / or NR_E).

[0109] In step S80, NR-Tm uses all R-DVD(k→m), k=1..K, and combines them with the original DVD(m) to form a superset C-DVD(m)=Combine{R-DVD(k→m),∀k=1..K,DVD(m)}. The function Combine can consist of various operations such as overlapping data sets, averaging, filtering, etc.

[0110] In step S90, C-DVD(m) is augmented to obtain a final training dataset for domain m, and in step S100, a domain m-specific ML model is trained. Such augmentation typically involves concatenation of datasets, random mixing of datasets, and the like.

[0111] In other words, Vendor 1 requests training of the ML module. NR_T1 is a function that trains Vendor 1's specific ML model. NR_T1 is configured by NR_C to collect DVD1 from NR_E1, where NR_E1 is from Vendor 1—where both training and data are from the same vendor and therefore data can be directly shared—and collect DID2 from NR_E2, where NR_E2 is from Vendor 2 and therefore data needs to be vendor-dependent before sharing. NR_E1 and NR_E2 are configured by NR_C to collect training data and share it with NR_T1. NR_E2 is configured by NR_C to apply its specific VIC2 to transform its own DVD2 into DID2. NR_T1 is configured by NR_C to apply IVC1 to DID2, thereby converting DID2 into a reconstructed DVD, R-DVD(2→1), i.e., data available to Vendor 1 for training and derived from the NR element of Vendor 2. NR_T1 then combines DVD1 and R-DVD(2→1), and such a combining function is generally denoted Combine1. The function Combine1 can perform any of the following operations on DVD1 and R-DVD(2→1): overlapping, averaging, filtering, pruning, puncturing, scaling, normalizing, or a combination of the above operations.

[0112] An example embodiment is shown in Figure 3, in which a coordinator function configures data collector functions to provide training data to the coordinator function. One of the data collector functions is from the same vendor as the training function and can therefore provide its training data in a format expected by the training function, but is instructed to collect specified training data, convert this training data to a specified format, and provide this converted training data to the coordinator function. The other data collector function is from a different vendor and is instructed to collect specified training data, convert this training data to a specified format, and provide this converted training data to the coordinator function. The coordinator function then combines the training data and provides it to the training function. The training function converts the converted training data to match the format of the training data provided by the data collector function from the same vendor as the training function and uses this training data to train an ML model.

[0113] In particular, the NR-C function collects DID(k) and combines them into a combined DID (C-DID), which is then sent to the specific vendor's NR-T. The NR_C configures the target DID format for all NR_Es for which data will be collected. Each NR_E is expected to be able to derive the corresponding VIC transformation knowing the DID format and its own DVD format. The NR_T knows the inverse transform IVC corresponding to the vendor for which training data will be generated.

[0114] Therefore, NR_C instructs NR_E1 to provide training data to NR_T1 in a vendor-specific format called domain variable data (DVD) when the training data and ML training are for the same vendor. In step S20, NR_E1 collects training data DVD1 in the vendor-specific format and reports this training data DVD1 to NT_T1 in step S30.

[0115] In steps S110 and S140, NR_C instructs NR_E1 and NR_E2 to provide training data to NR_C in a vendor-independent format called domain-invariant data (DID) format. NR_C defines how DID is obtained at each NR-Ek side by providing details of a vendor-to-invariant conversion (VIC) or transformation, as described in more detail below.

[0116] In steps S120 and S150, each NR-Ek, k=1...K, converts domain variable data k (DVD(k)) into a DID using the VIC provided by the NR_C. The DID has a unique format (format, size, type, e.g., integer, real number, etc.) and may be scenario-specific; for example, a DID for beamforming may be a real 2D or 3D matrix where each entry is the RSRP of an RS at a given (time, frequency) or (time, frequency, space) location, and a DID for positioning enhancement may be a complex 3D matrix where each entry is the channel gain at a given (time, delay, space) location, generating a universally understood / agreed DID format that can be used across different UE vendors / domains. DVD-to-DID conversion (VIC) is configured by the NR-C and typically has the goal of stripping the DVD from sensitive information (UE-specific data payload, symbols, identifiers) and vendor-specific artifacts, such as UE-specific TX / RX delays and beam offsets. Note that the exact form of the VIC conversion to be applied to the DVD to obtain the DID is derived in the NR_E, which performs the conversion. The VIC parameterization may be fully / totally configured by the NR-C and can constrain the DID format, DID reporting periodicity, and target VIC performance; performance metrics are up to the use case.

[0117] In steps S130 and S160, each NR-Ek sends its DIDk to the NR-C.

[0118] In step S170, NR-C combines all DIDs into a superset C-DID(m) = Combine{DID(k)}. The function Combine can consist of various operations such as overlapping data sets, averaging, filtering, etc.

[0119] In step S180, the C-DID is reported to NR_T1.

[0120] In step S190, using the C-DID, the NR_T1 diversifies the C-DID to reconstruct it into an R-DVD1. The reconstruction or conversion process consists of inputting the C-DID into a module that applies a domain-specific transformation to the DID and outputs an approximation of a DVD, called a reconstructed DVD: R-DVD. The DID-to-R-DVD conversion (Invariant-to-Variable Conversion (IVC)) is the inverse of VIC and involves transforming the DID into a DVD format and also including domain-specific information, if available (depending on the use case). This conversion ensures that the R-DVD has the same properties and format as a DVDm for a particular UE vendor m. The exact form of the IVC transformation needs to be known only to the vendor-specific functions (NR_T and / or NR_E).

[0121] In step S200, R-DVD(m) is augmented to obtain a final training dataset for domain m, and in step S210, a domain m-specific ML model is trained.

[0122] Vendor 2 provides training data to Vendor 1 and vice versa An example embodiment is shown in Figure 4, in which a coordinator function configures a data collector function to provide training data from multiple vendors to a training function. This example embodiment can be combined with the example embodiments described above. The data collector function is instructed to collect specified training data, convert the training data to a specified format, and provide the converted training data to the training function. The training function then combines the training data to match the format of the vendor's training data and uses the training data to train its ML model.

[0123] Therefore, NR_C instructs NR_E1 and NR_E2 to provide training data to NR_T1 and NR_T2 in a vendor-specific format called domain variable data (DVD). In step S20, NR_E1 collects training data DVD1 in the vendor-specific format and reports this training data DVD1 to NT_T1 in step S30. In step S220, NR_E2 collects training data DVD2 in the vendor-specific format and reports this training data DVD1 to NT_T2 in step S230.

[0124] In steps S240 and S280, NR_C instructs NR_E1 and NR_E2 to provide training data to NR_T1 and NR_T2 in a vendor-independent format called Domain Invariant Data (DID) format. NR_C defines how DID is obtained at each NR-Ek side by providing details of a Vendor-Invariant Conversion (VIC) or transformation, as described in more detail below.

[0125] In steps S250 and S290, each NR-Ek, k=1...K, converts domain variable data k (DVD(k)) into a DID using the VIC provided by the NR_C. The DID has a unique format (format, size, type, e.g., integer, real number, etc.) and may be scenario-specific; for example, a DID for beamforming may be a real 2D or 3D matrix where each entry is the RSRP of an RS at a given (time, frequency) or (time, frequency, space) location, and a DID for positioning enhancement may be a complex 3D matrix where each entry is the channel gain at a given (time, delay, space) location, generating a universally understood / agreed DID format that can be used across different UE vendors / domains. DVD-to-DID conversion (VIC) is configured by the NR-C and typically has the goal of stripping the DVD from sensitive information (UE-specific data payload, symbols, identifiers) and vendor-specific artifacts, such as UE-specific TX / RX delays and beam offsets. Note that the exact form of the VIC conversion to be applied to the DVD to obtain the DID is derived in the NR_E, which performs the conversion. The VIC parameterization may be fully / totally configured by the NR-C and can constrain the DID format, DID reporting periodicity, and target VIC performance; performance metrics are up to the use case.

[0126] In steps S260, S270, S300, and S310, each NR-Ek sends its DIDk to both NR_T1 and NR_T2.

[0127] In steps S320 and S330, NR_T1 and NR_T2 combine all DIDs k into a superset C-DID(m) = Combine{DID(k)}. The function Combine can consist of various operations such as overlapping data sets, averaging, filtering, etc.

[0128] In steps S340 and S350, C-DID(m) is used to train a domain m-specific ML model.

[0129] Alternatively, C-DID, NR-T1, and NR_T2 are optionally used to diversify C-DID to reconstruct it into R-DVD1 and R-DVD1. The reconstruction or conversion process consists of inputting the C-DID to a module that applies domain-specific transformations to the DID and outputs an approximation of the DVD, called the reconstructed DVD: R-DVD. The conversion from DID to R-DVD (invariant-to-variable conversion (IVC)) is the inverse of VIC and involves transforming the DID into a DVD format and also including domain-specific information, if available (depending on the use case). This conversion ensures that the R-DVD has the same properties and format as DVDm for a specific UE vendor m. The exact form of the IVC transformation needs to be known only to the vendor-specific functionality (NR_T and / or NR_E). R-DVD(m) is extended to obtain the final training dataset for domain m and to train a domain-m-specific ML model.

[0130] Vendor Y provides training data to vendor A An example embodiment is shown in Figure 5, in which a coordinator function from one vendor configures a data collector function from another vendor to provide training data to the training function of the other vendor. This example embodiment can be combined with the example embodiments described above. The data collector function is instructed to collect specified training data, convert this training data to a specified format, and provide this converted training data to the training function. The training function then combines training data to match the format of the vendor's training data and uses this training data to train its ML model.

[0131] In step S20, NR_EY collects training data DVDY in a vendor-specific format.

[0132] In step S370, NR_C instructs NR_EY to provide training data to NR_TA in a vendor-independent format called Domain Invariant Data (DID) format. NR_C defines how DID is obtained at each NR-Ek side by providing details of the Vendor-Invariant Conversion (VIC) or transformation, as described above.

[0133] In step S380, NR_C reports the conversion from DID to R-DVD (Invariant to Variable Conversion (IVC)), including domain-specific information if available (depending on the use case). This conversion ensures that the R-DVD has the same properties and format as DVDm for a specific UE vendor m. In this example, IVC_A provides the conversion from DID to R-DVD_A.

[0134] In step S390, each NR-Ek, k=1...K, converts domain variable data k (DVD(k)) into a DID using the VIC provided by the NR_C. The DID has a unique format (format, size, type, e.g., integer, real number, etc.) and may be scenario-specific; for example, a DID for beamforming may be a real 2D or 3D matrix where each entry is the RSRP of an RS at a given (time, frequency) or (time, frequency, space) location, and a DID for positioning enhancement may be a complex 3D matrix where each entry is the channel gain at a given (time, delay, space) location, generating a universally understood / agreed DID format that can be used across different UE vendors / domains. DVD-to-DID conversion (VIC) is configured by the NR-C and typically has the goal of stripping the DVD from sensitive information (UE-specific data payload, symbols, identifiers) and vendor-specific artifacts, such as UE-specific TX / RX delays and beam offsets. Note that the exact form of the VIC conversion to be applied to the DVD to obtain the DID is derived in the NR_E, which performs the conversion. The VIC parameterization may be fully / totally configured by the NR-C and can constrain the DID format, DID reporting periodicity, and target VIC performance; performance metrics are up to the use case.

[0135] In step S400, each NR-Ek sends its DIDk to the NR_TA.

[0136] In step S410, using the DID, the NR-TA diversifies the DID in order to reconstruct it into an R-DVD_A. The reconstruction or conversion process consists of inputting the DID into a module that applies domain-specific transformations to the DID and outputs an approximation of the DVD, called a reconstructed DVD:R-DVD. This conversion ensures that the R-DVD has the same properties and format as DVDm for a particular UE vendor m. R-DVD_A is augmented to obtain a final training dataset for domain m, and in step S400, a domain-m-specific ML model is trained.

[0137] Sidelink (SL) positioning 6, in one example embodiment, for SL positioning, the Vendor 1 UE uses its own data DVD1 and DID2 collected from the Vendor 2 UE to train an ML-based position estimator. While this example is related to positioning, it will be understood that this technique is applicable to other use cases.

[0138] In step S430, the vendor 1 UE collects time-frequency measurements of the DL PRS and stores them in matrix DVD1.

[0139] In step S440, the vendor 2 UE is instructed by the vendor 1 UE to collect time-frequency measurements of the DL PRS and store them in matrix DVD2, where: DVD2(i,j) = Channel frequency response at frequency i*Fs2 and time j*Ts2, where Fs2 and Ts2 are sampling frequency and time, both specific to the vendor 2 UE, and i≠a, j≠b.

[0140] Using the new SL PSSCH IE, the vendor 1 UE configures the vendor 2 UE to report DID2, which is produced by converter VIC2 for all vendor 2 UEs.

[0141] The vendor 1 UE configures VIC2 parameters. For example, VIC2 should output DID2, where: DID2(i,j)=i*Fs1 and CFR at time i*Ts1, in other words DVD2 should be resampled at rate Fs1 and its resolution changed from Ts2 to Ts1. The RX beam response of vendor 2 UE, i.e. W2, is subtracted from DVD2. For example, VIC2 should apply the transformation of DVD2 as follows: DID2=inv(W2)*DVD2.

[0142] In step S450, the vendor 2 UE collects DVD2, applies VIC2 as instructed, and reports DID2 to the LMF in step S460.

[0143] In step S470, the vendor 1 UE uses DID2 and converter IVC2 to reconstruct the R-DVD (2 → 1). In other words, the vendor 1 UE applies its own response to DID2 to artificially generate what the vendor 1 UE data should look like at resource index (i,j).

[0144] Next, in step 480, the Vendor 1 UE combines DVD1 with R-DVD(2→1) to form C-DVD(1). For example, the Vendor 1 UE can overlap two matrices or calculate two average responses.

[0145] In step S490, the Vendor 1 UE uses the C-DVD1 and the preferred latest enhancement method (scaling, translation, etc.) to produce the enhanced DVD1.

[0146] In step S500, the expanded set is then used to train a preferred state-of-the-art ML-based position estimator (e.g., a deep neural network (DNN) with a rectified linear unit (ReLU) activation function).

[0147] UE-assisted DL positioning As illustrated in FIG. 7, one example embodiment includes UE-assisted DL positioning, in which the NR_T (or an associated Network Data Analysis Function (NWDAF) function) trains an ML-based position estimator for UE vendor 1 using the DVD of the vendor 1 UE and DID2 collected from vendor 2 generally using the techniques described in FIG. 3.

[0148] The vendor 1 UE is instructed by the NR-C to collect time-frequency measurements of the DL PRS and store them in matrix DVD1, where: DVD1(a,b) = Channel frequency response at frequency a*Fs1 and time b*Ts1, where Fs1 and Ts1 are sampling frequency and time, both specific to vendor 1 UE.

[0149] A new LPP IE is used for Vendor 1 UE to report DVD to NR_T.

[0150] The vendor 2 UE is instructed by the NR_C to collect time-frequency measurements of the DL PRS and store them in matrix DVD2, where: DVD2(i,j) = Channel frequency response at frequency i*Fs2 and time j*Ts2, where Fs2 and Ts2 are sampling frequency and time, both specific to the vendor 2 UE, and i≠a, j≠b.

[0151] A new LPP IE is used, so the NR_C configures the vendor 2 UEs to report DID2, where DID2 is generated by the converter VIC2 for all vendor 2 UEs.

[0152] NR_C configures VIC2 parameters. For example, VIC2 should output DID2, where: DID2(i,j)=i*Fs1 and CFR at time i*Ts1, in other words DVD2 should be resampled at rate Fs1 and its resolution changed from Ts2 to Ts1. The RX beam response of vendor 2 UE, i.e. W2, is subtracted from DVD2. For example, VIC2 should apply the transformation of DVD2 as follows: DID2=inv(W2)*DVD2.

[0153] The Vendor 2 UE collects DVD2, applies VIC2 as instructed, and reports DID2 to the NR_C.

[0154] NR_T uses DID2 and converter IVC2 to reconstruct the R-DVD (2 → 1). In other words, NR_T applies a vendor 1 UE-specific response to DID2 to artificially generate what the vendor 1 UE's data should look like at resource index (i,j).

[0155] NR_T combines DVD1 with R-DVD(2→1) to form C-DVD(1). For example, NR_T can superimpose two matrices or calculate two average responses.

[0156] NR_T uses C-DVD1 and the preferred modern enhancement methods (scaling, translation, etc.) to produce an enhanced DVD1.

[0157] The expanded set is then used to train a preferred state-of-the-art ML-based position estimator (e.g., a DNN with a ReLU activation function).

[0158] Those skilled in the art will readily recognize that the steps of the various above-described methods can be performed by a programmed computer. As used herein, some embodiments are also intended to cover a program storage device, such as a digital data storage medium, that is machine- or computer-readable and encodes a machine-executable or computer-executable program of instructions, the instructions performing some or all of the steps of the above-described methods. The program storage device may be, for example, a digital memory, a magnetic storage medium such as a magnetic disk or tape, a hard drive, or an optically readable digital data storage medium. Embodiments are also intended to cover a computer programmed to perform the steps of the above-described methods. As used herein, the term non-transitory refers to the medium itself (i.e., not tangible, not a signal), as opposed to the persistence of the data storage (e.g., RAM vs. ROM).

[0159] As used in this application, the term "circuit equipment" means (a) hardware-only circuit embodiments (e.g., embodiments in analog and / or digital circuitry only); (b) (where applicable), (i) a combination of analog and / or digital hardware circuitry and software / firmware; and (ii) Any portion of a hardware processor together with software (including a digital signal processor, software, and memory that work together to cause a device such as a mobile phone or server to perform various functions). A combination of hardware circuits and software, such as: (c) Hardware circuitry and / or processors, such as microprocessors or portions of microprocessors, that require software (e.g., firmware) for operation but may not be present when software is not needed for operation. represents one or more or all of the following:

[0160] This definition of circuit equipment applies to all uses of the term in this application, including in any claims. By way of further example, as used in this application, the term circuit equipment also covers embodiments of a hardware circuit or processor (or multiple processors) alone, or portions of a hardware circuit or processor and its (or their) accompanying software and / or firmware. The term circuit equipment also covers, for example, and where applicable to particular claim elements, baseband or processor integrated circuits for mobile devices, or similar integrated circuits in servers, cellular network devices, or other computing or network devices.

[0161] While example embodiments of the present invention have been described in the preceding paragraphs with reference to various examples, it should be understood that modifications to the examples shown can be made without departing from the scope of the present invention as claimed.

[0162] Features described in the preceding description may be used in combinations other than those explicitly described.

[0163] Although functions are described with reference to particular features, these functions may be enabled by other features, whether or not described.

[0164] Although features are described with reference to particular embodiments, these features may also be present in other embodiments whether or not described.

[0165] While an effort has been made in the foregoing specification to draw attention to those features of the invention which are believed to be of particular importance, it is to be understood that the applicant claims protection in respect of any patentable feature or combination of features hereinabove mentioned and / or shown in the drawings, whether or not specifically emphasized.

Claims

1. a coordinator function configured to send a first configuration message to a first collector function, the first configuration message including information for configuring the first collector function to collect first training data, transform the first training data to generate transformed first training data, and report the transformed first training data; An apparatus comprising:

2. the first configuration message: transforming the first training data by removing specified vendor-specific data and / or artifacts to generate the transformed first training data; and / or reporting the transformed first training data in a transformed training data format and / or at a specified reporting interval.

2. The apparatus of claim 1, further comprising information for configuring the first collector function to:

3. the first configuration message: reporting said transformed first training data to said coordinator function and / or training function; and / or reporting the transformed first training data to both a first training function and a second training function.

3. The apparatus of claim 1, further comprising information for configuring the first collector function to:

4. 4. The apparatus of claim 1, wherein the coordinator function is configured to send a second configuration message to a second collector function, the second configuration message including information for configuring the second collector function to collect second training data, reconstruct the second training data to generate transformed second training data, and report the transformed second training data.

5. 5. Apparatus according to any preceding claim, wherein the coordinator function is configured to combine received transformed training data to form combined transformed training data, and preferably to send the combined transformed training data to the training function.

6. 6. The apparatus of claim 1, wherein the coordinator function is configured to send a configuration message to a collector function provided by a vendor common to the coordinator function and the collector function, the configuration message including instructions for configuring the collector function to collect training data and report the training data.

7. 7. The apparatus of claim 1, wherein the coordinator function is configured to send a conversion configuration message to the training function, the conversion configuration message including information for configuring the training function to convert received transformed training data and / or received combined transformed training data into converted training data.

8. 8. The apparatus of claim 7, wherein the conversion configuration message includes information for configuring the training function to convert the received transformed training data and / or the received combined transformed training data into converted training data by performing expanding, averaging, filtering, pruning, puncturing, normalizing, scaling, and / or translating.

9. 9. The apparatus of claim 1, wherein the coordinator function is configured to: send a first conversion configuration message to the first training function, the first conversion configuration message including information for configuring the first training function to convert received transformed training data and / or received combined transformed training data into first converted training data; and send a second conversion configuration message to a second training function, the second conversion configuration message including information for configuring the second training function to convert received transformed training data and / or received combined transformed training data into second converted training data.

10. The apparatus according to any one of claims 1 to 9, wherein the messages and / or data are transmitted on a physical sidelink shared channel, a physical downlink shared channel, a physical uplink shared channel, and / or any wireless medium channel.

11. sending a first configuration message to a first collector function, the first configuration message including information for configuring the first collector function to collect first training data, transform the first training data to generate transformed first training data, and report the transformed first training data; A method comprising:

12. a collector function configured to receive a configuration message from a coordinator function, the configuration message including information for configuring the collector function to collect training data, transform the training data to generate transformed training data, and report the transformed training data. An apparatus comprising:

13. 1. A method comprising: receiving a configuration message from a coordinator function, the configuration message including information for configuring a collector function to collect training data, transform the training data to generate transformed training data, and report the transformed training data.

14. a training function configured to receive a conversion configuration message from a coordinator function, the conversion configuration message including information for configuring the training function to convert received transformed training data and / or received combined transformed training data into converted training data; An apparatus comprising:

15. receiving a conversion configuration message from a coordinator function, said conversion configuration message including information for configuring a training function to convert the received transformed training data and / or the received combined transformed training data into converted training data; A method comprising:

Images