User equipment, network node and methods performed therein for handling a computational model

Abstract

According to some embodiments herein it is herein provided a UE (10) for handling communication in a wireless communications network. The UE is configured to transmit to a network node (120) information related to one or more UE-sided conditions and / or configurations and receive a first subset of data samples from the network node (120) to be used in one or more computational models at the UE (10), which first subset is labelled based on the transmitted information.

Description

[0001] USER EQUIPMENT, NETWORK NODE AND METHODS PERFORMED THEREIN

[0002] TECHNICAL FIELD

[0003] Embodiments herein relate to a user equipment (UE), a network node and methods performed therein for communication. Furthermore, a computer program and a computer readable storage medium are also provided herein. In particular, embodiments herein relate to handling communication, such as handling a computational model, in a wireless communications network.

[0004] BACKGROUND

[0005] In a typical wireless communications network, UEs, also known as wireless communication devices, mobile stations, stations (STA) and / or wireless devices, communicate via for example a Radio Access Network (RAN) with one or more core networks (CN). The RAN covers a geographical area which is divided into service areas or cell areas, with each service area or cell area being served by radio network node such as an access node e.g. a Wi-Fi access point or a radio base station (RBS), which in some networks may also be called, for example, a NodeB, a gNodeB, or an eNodeB. The service area or cell area is a geographical area where radio coverage is provided by the radio network node. The radio network node operates on radio frequencies to communicate over an air interface with the UEs within range of the radio network node. The radio network node communicates over a downlink (DL) to the UE and the UE communicates over an uplink (UL) to the radio network node.

[0006] A Universal Mobile Telecommunications System (UMTS) is a third-generation telecommunications network, which evolved from the second generation (2G) Global System for Mobile Communications (GSM). The UMTS terrestrial radio access network (UTRAN) is essentially a RAN using wideband code division multiple access (WCDMA) and / or High-Speed Packet Access (HSPA) for communication with user equipment. In a forum known as the Third Generation Partnership Project (3GPP), telecommunications suppliers propose and agree upon standards for present and future generation networks and UTRAN specifically and investigate enhanced data rate and radio capacity. In some RANs, e.g. as in UMTS, several radio network nodes may be connected, e.g., by landlines or microwave, to a controller node, such as a radio network controller (RNC) or a base station controller (BSC), which supervises and coordinates various activities of the plural radio network nodes connected thereto. The RNCs are typically connected to one or more core networks.

[0007] Specifications for the Evolved Packet System (EPS) have been completed within the 3GPP, and this work continues in the coming 3GPP releases, such as 5G, for example New Radio (NR), and beyond networks. The EPS comprises the Evolved Universal Terrestrial Radio Access Network (E-UTRAN), also known as the Long-Term Evolution (LTE) radio access network, and the Evolved Packet Core (EPC), also known as System Architecture Evolution (SAE) core network. E-UTRAN / LTE is a 3GPP radio access technology wherein the radio network nodes are directly connected to the EPC core network. As such, the Radio Access Network (RAN) of an EPS has an essentially “flat” architecture comprising radio network nodes connected directly to one or more core networks.

[0008] With the 5G technologies such as NR, focus is on a set of features such as the use of very many transmit- and receive-antenna elements that makes it possible to utilize beamforming, such as transmit-side and receive-side beamforming. Transmit-side beamforming means that the transmitter can amplify the transmitted signals in a selected direction or directions, while suppressing the transmitted signals in other directions. Similarly, on the receive-side, a receiver can amplify signals from a selected direction or directions, while suppressing unwanted signals from other directions.

[0009] Artificial Intelligence (Al) and Machine Learning (ML) have been investigated, both in academia and industry, as promising tools to optimize the design of the air interface in wireless communications networks. Example use cases include using autoencoders for Channel State Information (CSI) compression to reduce the feedback overhead and improve channel prediction accuracy; using deep neural networks for classifying Line-of- Sight (LOS) and Non-LOS (NLOS) conditions to enhance the positioning accuracy; and using reinforcement learning for beam selection at the network side and / or the User Equipment (UE) side to reduce the signalling overhead and beam alignment latency; and using deep reinforcement learning to learn an optimal precoding policy for complex Multiple Input Multiple Output (MIMO) precoding problems.

[0010] In the 3rd Generation Partnership Project (3GPP) New Radio (NR) standardization work for release 18 (Rel-18), a study item (SI) on AI / ML for the NR air interface was started in May 2022. It has been agreed that the works will continue in Rel. 19. The works will explore the benefits of augmenting the air interface with features enabling improved support of AI / ML-based algorithms for enhanced performance and / or reduced complexity / overhead. Through studying and specifying a few selected use cases, such as CSI feedback, beam management, and positioning, the works aim to design the mechanisms to accommodate AI / ML into the 3GPP standard.

[0011] General aspects for NR Rel-18 AI / ML for NR air interface:

[0012] An important part of Al development and operation is the lifecycle management (LCM) of the AI / ML model, e.g., model training, model deployment, model inference, model monitoring, model updating, and AI / ML functionality.

[0013] In NR Rel-18 AI / ML for NR air interface study item, the LCM procedure is studied for the case that an AI / ML model has a model identity (ID) with associated information and / or for the case that a given functionality is provided by some AI / ML operations.

[0014] Two types of LCM operations were studied in NR Rel-18, functionality-based LCM, and model-l D-based LCM.

[0015] Functionality-based LCM: functionality refers to an AI / ML-enabled Feature- and / or feature group- (FG) enabled by configuration(s), where configuration(s) is(are) supported based on conditions indicated by UE capability. Correspondingly, functionality-based LCM operates based on at least one configuration of an AI / ML-enabled feature and / or FG or specific configurations of an AI / ML-enabled feature and / or FG. In functionality-based LCM, the network indicates activation, deactivation, fallback and / or switching of AI / ML functionality via 3GPP signalling, e.g., radio resource control (RRC), medium access control (MAC)-control element (CE), and / or downlink control information (DCI). Models may not be identified at the Network, and the UE may perform model-level LCM. Whether and how much awareness and / or interaction network (NW) should have about model-level LCM requires further study. For functionality identification, there may be either one or more than one functionality defined within an AI / ML-enabled feature, whereby AI / ML-enabled feature refers to a feature where AI / ML may be used.

[0016] In model-l D-based LCM, ML models are identified at the Network, and Network and / or UE may activate, deactivate, select, and / or switch individual AI / ML models via model ID. A model may be associated with specific configurations and / or conditions associated with the UE capability of an AI / ML-enabled feature and / or FG and additional conditions, e.g., scenarios, sites, and datasets, as determined / identified between the UE side and NW side. An AI / ML model identified by a model ID may be logical, and how it maps to physical AI / ML model(s) may be up to implementation. Fig. 1 shows a functional framework for AI / ML for NR air interface that can be used for studying model LCM aspects for different Al for physical layer (PHY) use cases. The general framework consists of the following:

[0017] Data Collection is a function that provides input data to the Model Training, Management, and Inference functions. o Training Data: data needed as input for the AI / ML Model Training function. o Monitoring Data: data needed as input for the Management of AI / ML models or AI / ML functionalities. o Inference Data: data needed as input for the AI / ML Inference function. Model Training is a function that performs AI / ML model training, validation, and testing which may generate model performance metrics that can be used as part of the model testing procedure. The Model Training function is also responsible for data preparation, e.g., data pre-processing and cleaning, formatting, and transformation, based on Training Data delivered by a Data Collection function, if required. o Trained / Updated Model: In case of having a Model Storage function, this is used to deliver trained, validated, and tested AI / ML models to the Model Storage function, or to deliver an updated version of a model to the Model Storage function.

[0018] Management is a function that oversees the operation, e.g., selection, (de)activation, switching, fallback, and monitoring, e.g., performance, of AI / ML models or AI / ML functionalities. This function is also responsible for making decisions to ensure the proper inference operation based on data received from the Data Collection function and the Inference function. o Management Instruction: information needed as input to manage the Inference function. Concerning information may include selection, (de)activation, and / or switching of AI / ML models or AI / ML-based functionalities, fallback to non-AI / ML operation, i.e. , not relying on inference process, etc. o Model Transfer and / or Delivery Request: used to request model(s) to the Model Storage function. o Performance Feedback and / or Retraining Request: information needed as input for the Model Training function, e.g., for model (re)training or updating purposes. Inference is a function that provides outputs from the process of applying AI / ML models or AI / ML functionalities, using the data that is provided by the Data Collection function, i.e. , Inference Data, as an input. The Inference function is also responsible for data preparation, e.g., data pre-processing and cleaning, formatting, and transformation, based on Inference Data delivered by a Data Collection function, if required. o Inference Output: data used by the Management function to monitor the performance of AI / ML models or AI / ML functionalities.

[0019] Model Storage is a function responsible for storing trained and / or updated models that can be used to perform the Inference function. o Note: The Model Storage function in Fig. 1 is only intended as a reference point (if any) when applicable for protocol terminations, model transfer and / or delivery, and related processes. It should be stressed that its purpose does not encompass restricting the actual storage locations of models. Therefore, the specification impact of all data, information, and / or instruction flows, i.e., the arrows in Figure 1 , to / from this function should be studied case by case. o Model Transfer and / or Delivery: used to deliver an AI / ML model to the Inference function.

[0020] UE-NW collaboration levels for one-and two-sided AI / ML models:

[0021] The AI / ML models being discussed in the Rel-18 study item on AI / ML for the NR air interface can be categorized into the following two types:

[0022] One-sided AI / ML model, which can be a UE-sided AI / ML model whose inference is performed entirely at the UE, or a NW-sided AI / ML model whose inference is performed entirely at the NW.

[0023] Two-sided AI / ML model, which refers to a paired AI / ML Model(s) over which joint inference is performed across the UE and the NW, i.e., the first part of the inference is firstly performed by UE and then the remaining part is performed by base station (BS), or vice versa. As an example of a two-sided AI / ML model, Fig. 2 shows a use case of an autoencoder (AE)-based CSI feedback / report, where an encoder, e.g., UE-part of the two-sided AE model, is operated at a UE to compress the estimated wireless channel, and the output of the encoder, such as the compressed wireless channel information estimates, is reported from the UE to a BS. The Bs uses a decoder, e.g., NW- part of the two-sided AE model, to reconstruct the estimated wireless channel information. Here the two-sided AI / ML model is composed of the encoder at the UE side and the decoder at the base station, e.g., a gNB, side. Note that in the case of a two-sided model, the code is generated by the encoder and only interpretable by a jointly trained decoder. The situation is different from running an AI / ML model in the UE, reporting the output over the air in a fully standardized format, and running a separate AI / ML model at the base station. Thus, Fig. 2 shows an AE-based CSI compression using a two-sided AI / ML model use case.

[0024] When applying AI / ML on air interface use cases, different levels of collaboration between network nodes and UEs can be considered:

[0025] No collaboration between network nodes and UEs. In this case, a proprietary ML model operating with the existing standard air-interface is applied at one end of the communication chain, e.g., at the UE side, and the model life cycle management, e.g., model selection / training, model monitoring, model retraining, model update, is done at this node without inter-node assistance, e.g., assistance information provided by the network node.

[0026] Limited collaboration between network nodes and UEs for one-sided models.

[0027] In this case, an ML model is operating at one end of the communication chain, e.g., at the UE side, but this node gets assistance from the node(s) at the other end of the communication chain, e.g., a next generation Node B (gNB), for its Al model life cycle management to some extent, e.g., for training / retraining the Al model, model update, model monitoring, model selection, fallback, and / or switching.

[0028] - Joint ML operation between network nodes and UEs for two-sided models. In this case, it is assumed that the Al model is split with one part located at the NW side and the other part located at the UE side. Hence, the Al model requires joint inference between the NW and UE, and the Al model life cycle management involves both ends of a communication chain.

[0029] For two-sided models, the model training process may require sharing of data from one side to the other side since the input and output of a two-sided model reside within different vendor’s domain. Different model training types can be considered for two-sided model training between M network vendors and N UE / chip-set vendors, where M>=1 and N>=1 :

[0030] - Type 1 : Joint training of the two-sided model at a single side and / or entity, e.g., the UE-side or the NW-side. For instance, a two-sided model, UE-part model and NW-part model, is trained at the NW side, e.g., by a NW vendor, then, the UE-part of the trained model, e.g., encoder for the AE-based CSI compression use case, is transferred and / or delivered from the NW-side to the UE-side, and vice versa.

[0031] - Type 2: Joint training of the two-sided model at network side and UE side, respectively. Joint training can be done simultaneously at the network and UE sides or be performed in a sequential way. In case of Type 2 simultaneous joint training, the UE-part model, i.e., trained at the UE side, and the NW-part model, i.e., trained at the NW side, are jointly trained in the same loop through exchanging forward propagation values and backward propagation values between NW and UE. In the case of Type 2 sequential joint training, one side, i.e., UE-side or NW-side, starts its model training first, it then opens an application programming interface (API) to facilitate the other side to do the model training. For instance, the NW side trains its model first, thus, also obtaining what is sometimes known as a nominal encoder, but that is not used at the UE, and then the UE side can train its encoder by using an API. The API would accept, e.g., a CSI report and a target CSI, both of which are derived by the UE side based on the data. Note that the CSI report is generated, at least partially, by the UE encoder under training and may thus not be an efficient CSI report at each step in the training. The API would return gradients of the decoder and a loss function, with respect to the variables in the CSI report. Thus, allowing the UE to train an encoder that is matched to the decoder.

[0032] - Type 3: Sequential training starting with UE side training or sequential training starting with NW side training, where the UE-part model and the NW-part model are trained by UE side and network side, respectively. Take sequential training with NW-first training approach for AE-based CSI compression as an example. The NW can firstly train the UE-part and NW-part models jointly using training data, e.g., target CSI samples, and then share a dataset consisting of UE-part model output, e.g., latent space variables, associated with the ground-truth / labels, e.g., target CSI, for the UE-side to train its UE-part model, e.g., an encoder. Alternatively, the NW can share a dataset consisting of gradients of the NW-part model, e.g., the gradients of the decoder, together with loss function value indicating the discrepancy of the NW-part model output, e.g., the decoder output, and the ground-truth and / or labels, e.g., target CSI, with respect to the UE-part model output, e.g., latent space variables, based on which the UE-side trains its UE-part model, e.g., an encoder.

[0033] SUMMARY

[0034] As part of developing embodiments, herein one or more problems were first identified. Currently, although the dataset delivery for NW-first training type 3 has been discussed in 3GPP for two-sided AI / ML model cases, there is no clear mechanism on how the NW should provide the dataset to a UE side to train a UE-part model for a specific group of UEs with a specific set of UE-sided conditions and / or configurations. Examples of UE-sided conditions and / or configurations include specific UE antenna configurations, radio frequency (RF) design, Singular value decomposition (SVD) operations, channel estimation algorithms, hardware impairments, etc. A specific group of UEs may be associated with a specific UE vendor / chipset, a specific software / hardware version of a device release, a specific UE type, a specific UE, etc.

[0035] Although not discussed in 3GPP yet, dataset delivery from NW to UE may also be relevant for one-sided UE-sided model use cases, where instead of letting the NW to train a UE-side model and delivering and / or transferring the trained model to the UE-side / UEs, the NW could deliver a training dataset for the UE-side / UEs to train / retrain / update its UE- sided model by itself.

[0036] Different groups of UEs with different UE-sided conditions and / or UE-specific configurations can have different dataset distributions and / or characteristics. Training dataset delivery from a NW side to a UE side without considering UE-sided conditions and / or configurations can result in mismatched data distribution and / or characteristics between training dataset, used for training a UE-side / UE-part model, and inference dataset, used to generate input data for the UE-side / UE-part model inference, leading to model performance degradation, see R1-2307916, Evaluation on AI / ML for CSI feedback enhancement, Qualcomm, 3GPP TSG RAN WG1 #1142023-08-11. Therefore, it is important to have mechanisms to support dataset delivery that can provide matched training datasets to the UE side.

[0037] In addition, when a new group of UEs with a new set of UE-sided conditions and / or configurations is introduced, these new UEs can have a different inference data distribution and / or characteristic as compared to the legacy UEs, implying that the stored training dataset at the NW for legacy UEs may not be valid for these new UEs. Mechanisms to incorporate the new UE-sided conditions and / or configurations toward dataset delivery, therefore, become important for providing a matched dataset for the UE side to train a UE-part model for these new UEs.

[0038] The object of embodiments herein is to provide a mechanism for handling communication, such as handling and / or training computational models, in an efficient manner.

[0039] According to an aspect of embodiments herein the object is achieved by providing a method performed by a UE for handling communication in a wireless communications network. The UE transmits to a network node information related to one or more UE-sided conditions and / or configurations. The UE further receives from the network node, a first subset of data samples to be used in one or more computational models at the UE, which first subset is labelled based on the transmitted information.

[0040] According to another aspect of embodiments herein the object is achieved by providing a method performed by a network node for handling communication in a wireless communications network. The network node receives from a UE, information related to one or more UE-sided conditions and / or configurations. The network node further transmits a first subset of data samples to the UE to be used in one or more computational models at the UE, which first subset is labelled based on the received information.

[0041] It is furthermore provided herein a computer program product comprising instructions, which, when executed on at least one processor, cause the at least one processor to carry out the methods herein, as performed by the UE and the network node, respectively. It is additionally provided herein a computer-readable storage medium, having stored thereon a computer program product comprising instructions which, when executed on at least one processor, cause the at least one processor to carry out the methods herein, as performed by the UE and the network node, respectively.

[0042] The object is further achieved by providing a UE and a network node configured to perform the methods herein.

[0043] According to an aspect of embodiments herein the object is achieved by providing a UE for handling communication in a wireless communications network. The UE is configured to transmit to a network node, information related to one or more UE-sided conditions and / or configurations. The UE further receives from the network node, a first subset of data samples to be used in one or more computational models at the UE, which first subset is labelled based on the transmitted information. According to another aspect of embodiments herein the object is achieved by providing a method performed by a network node for handling communication in a wireless communications network. The network node receives from a UE, information related to one or more UE-sided conditions and / or configurations. The network node further transmits a first subset of data samples to the UE to be used in one or more computational models at the UE, which first subset is labelled based on the received information.

[0044] Embodiments herein focus on methods for training dataset delivery for a specific or a group of UEs with a specific set of one or more UE-sided conditions and / or configurations, also referred to as UE configurations.

[0045] For the two-sided AI / ML model cases, the dataset delivery method covers one or more of the following:

[0046] - the dataset delivery from the UE to the network node, i.e. , used for the network node to train network node's actual network-part model and network's nominal UE-part model and to generate another dataset to be delivered to the UE side, and

[0047] - the dataset delivery from the network node to the UE, i.e., for the UE-side to train, retrain, and / or update UE's actual UE-part model.

[0048] For the one-sided UE-sided model cases, the dataset delivery method covers: dataset delivery from the network node to the UE, i.e., for the UE-sided to train, retrain, and / or update its UE-sided model.

[0049] Information on the UE-sided conditions and / or configurations may be used to label the training dataset with differentiating distribution and / or characteristics of the dataset.

[0050] The proposed dataset delivery mechanism enables a UE / UE-side to obtain a more accurate training dataset by matching data distribution and / or characteristic of the training dataset with the UE-side condition and / or configuration. Hence, this dataset delivery mechanism may support UE / UE-side to train UE / NW-side condition and / or configuration specific UE-side / UE-part model(s), giving better model performance, and / or lower computational complexity, and / or smaller model sizes.

[0051] Thereby embodiments herein handle handling communication, such as handling or training computational models, in an efficient manner.

[0052] BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Embodiments will now be described in more detail in relation to the enclosed drawings, in which: Fig. 1 shows an overview according to prior art;

[0054] Fig. 2 shows an overview according to prior art;

[0055] Fig. 3 shows an overview depicting a wireless communications network according to embodiments herein;

[0056] Fig. 4 shows a combined signalling scheme and flowchart according to embodiments herein;

[0057] Fig. 5 shows a schematic flowchart depicting a method performed by a UE according to embodiments herein;

[0058] Fig. 6 shows a schematic flowchart depicting a method performed by a network node according to embodiments herein;

[0059] Fig. 7 shows an overview depicting a wireless communications network according to some embodiments herein;

[0060] Fig. 8 shows an overview depicting a wireless communications network according to some embodiments herein;

[0061] Fig. 9 shows an overview depicting a wireless communications network according to some embodiments herein;

[0062] Fig. 10 shows an overview depicting a wireless communications network according to some embodiments herein;

[0063] Fig. 11 shows an overview depicting a wireless communications network according to some embodiments herein;

[0064] Fig. 12 shows an overview depicting a wireless communications network according to some embodiments herein;

[0065] Fig. 13 shows an overview depicting a wireless communications network according to some embodiments herein;

[0066] Fig. 14 is a block diagram depicting a UE according to embodiments herein; and Fig. 15 is a block diagram depicting a network node according to embodiments herein.

[0067] DETAILED DESCRIPTION

[0068] Embodiments herein relate to communication networks in general. Fig. 3 is a schematic overview depicting a wireless communications network 1. The wireless communications network 1 comprises one or more RANs and one or more CNs. The wireless communications network 1 may use a number of different technologies, such as Wi-Fi, Long Term Evolution (LTE), LTE-Advanced, NR, Wideband Code Division Multiple Access (WCDMA), Global System for Mobile communications / Enhanced Data rate for GSM Evolution (GSM / EDGE), Worldwide Interoperability for Microwave Access (WiMax), or Ultra Mobile Broadband (UMB), just to mention a few possible implementations.

[0069] In the wireless communications network 1, wireless devices e.g. a user equipment (UE) 10 such as a mobile station, a non-access point (non-AP) STA, a STA, a wireless device and / or a wireless terminal, communicate via one or more Access Networks (AN), e.g. a RAN, to one or more core networks (CN). It should be understood by those skilled in the art that “UE” is a non-limiting term which means any terminal, wireless communication terminal, internet of things (loT) capable device, Machine Type Communication (MTC) device, Device to Device (D2D) terminal, or node e.g. smart phone, laptop, mobile phone, sensor, relay, mobile tablets or even a base station communicating within a cell.

[0070] The wireless communications network 1 comprises a radio network node 12 providing radio coverage over a geographical area, e.g. a first service area, of a first radio access technology (RAT), such as NR, LTE, UMTS, Wi-Fi or similar. The radio network node 12 may be a radio access network node such as radio network controller or an access point such as a wireless local area network (WLAN) access point or an Access Point Station (AP STA), an access controller, a base station, e.g. a radio base station such as a NodeB, an evolved Node B (eNB, eNodeB), a base transceiver station, Access Point Base Station, base station router, a transmission arrangement of a radio base station, a stand-alone access point or any other network unit capable of serving a UE within the service area served by the radio network node 12 depending e.g. on the first radio access technology and terminology used.

[0071] According to embodiments herein, a network node (NW) 120, such as the radio network node 12 or a core network node, or a computational model node, acquires UE- side condition and / or configuration-related information from the UE 10, based on which the network node 120 labels the data samples within a training dataset and may deliver to the UE / UE-side a matched training dataset to be used for training a one-sided UE-side model or a UE-part of a two-sided model.

[0072] • The concept “network” may refer to one of a generic network node, such as the radio network node 12, e.g., a gNB, or the corresponding node in a 6G network,, a base station, transmission and reception point (TRP), a unit within the base station to handle at least some ML operation, a relay node, a core network node, a core network node that handle at least some ML operations, or a device supporting device to device (D2D) communication. • A computational model may refer to an ML-based model, a configuration of an ML-based model, a non-ML-based functionality, or a configuration of a non- ML-based functionality.

[0073] • The terms “ML-model” and “Al-model” are interchangeable. An AI / ML model can be defined as a functionality or be part of a functionality that is deployed / implemented in a first node. This first node can receive a message from a second node indicating that the functionality is not performing correctly. Further, an AI / ML model can be defined as a feature or part of a feature that is implemented / supported in a first node. This first node can indicate the feature version to a second node. If the ML-model is updated, the feature version maybe changed by the first node.

[0074] Fig. 4 is a combined flowchart and signalling scheme according to some embodiments herein.

[0075] Action 401. The network node 120 may transmit a configuration indication to the UE 10. The configuration indication may indicate one or more parameters to perform the method herein.

[0076] Action 402. The UE 10 transmits information related to its UE-sided condition and / or configuration to the network node 120.

[0077] Action 403. The network node 120 may thus obtain UE-side conditions and / or configurations-related information from the UE 10 and / or UE-side, based on which the network node 120 may label one or more data samples, i.e. , a first subset, within a training dataset based on the information.

[0078] Action 404. The network node 120 delivers to the UE 10 and / or UE-side a matched training dataset to be used for training the UE-side. Thus, the UE 10 receives the first subset of data samples from the network node 120, e.g., a gNB, to be used in the one or more computational models, which first subset is labelled based on the obtained information.

[0079] Action 405. The UE 10 may train one or more computational models using the first subset of data samples.

[0080] Example embodiments of a method performed by the UE 10 for handling communication in the wireless communications network will now be described with reference to a flowchart depicted in Fig. 5. The actions do not have to be taken in the order stated below but may be taken in any suitable order. Optional actions are marked in dashed boxes.

[0081] Action 501. The UE 10 may obtain configuration and / or condition from a network node or be preconfigured with the configuration and / or condition. The UE 10 may receive a configuration from the network node 120 to apply for the training of the one or more computational models according to the first subset of data samples

[0082] Action 502. The UE 10 transmits to the network node 120 information related to one or more UE-sided conditions and / or configurations, such as one or more of the following: UE-specific conditions under which the data are collected, such as any of UE vendor ID, UE chipset ID, UE release number, Software / hardware version, UE antenna configuration, UE RF impairments, UE power states, Al complexity levels, and Al model structures, UE memory status, UE processing time, means of UE classification information, UE mobility status such as speed, data collection information, NW configuration, such as RRC configuration, applied by the UE at the moment of data collection; data and / or parameters related to one or more computational models. The UE 10 may transmit the information by reporting collected measurements as data samples to the network node 120. The one or more UE-sided conditions and / or configurations may comprise one or more of: a UE-specific condition under which the data are collected; and a network configuration applied by the UE 10 at the moment of data collection.

[0083] Action 503. The UE 10 further receives the first subset of data samples from the network node 120, e.g., a gNB, to be used in one or more computational models at the UE 10, which first subset is labelled based on the transmitted information. For example, receives a matched training dataset to be used for training the UE-side. Thus, the first subset may comprise a training dataset matched to information related to the one or more computational models to be used for training the one or more computational models on the UE-side.

[0084] Action 504. The UE 10 may train the one or more computational models using the first subset of data samples.

[0085] Action 505. The UE 10 may receive from a third node, such as a UE server, one or more computational models for a UE-sided condition and / or configuration.

[0086] Action 506. The UE 10 may transfer the received first subset of data samples to the third node.

[0087] Action 507. The UE 10 may transfer the one or more UE-sided conditions and / or configurations and applied received configuration to apply for the first subset of data samples to the third node. Action 508. The UE 10 may receive from the third node one or more computational models for the one or more UE-sided conditions and / or configurations.

[0088] Thus, it is herein disclosed a method, implemented in the UE 10 capable of running one or more AI / ML models for an AI / ML-enabled feature. The UE 10 transmits the information related to its UE-sided condition and / or configuration to the network node 120.

[0089] The UE 10 may receive the first subset of data samples from the network node 120 in response of requesting a first subset of data samples from the network node 120.

[0090] The UE 10 may: a) receive configurations related to data collection for the AI / ML-enabled feature. b) perform channel / radio measurements based on the received configuration. c) report collected measurements together with other information (if needed) as data samples to the network node 120.

[0091] The one or more computational models may comprise one or more AI / ML models being one or more one-sided UE-side models; and / or being one or more UE-part of two- sided models.

[0092] The UE 10 may in response of reporting the collected dataset to the network node 120 perform one or more of the following:

[0093] - Receiving a first subset of data samples from the network node 120

[0094] - Receiving a configuration from the network node 120 to apply for the training of the one or more AI / ML models according to the first subset of data samples.

[0095] - Using the received first subset of data samples and the applied configuration to train one or more AI / ML models for its UE-sided condition and / or configuration for the AI / ML enabled feature.

[0096] The UE 10 may perform one or more of the following:

[0097] - Receiving a first subset of data samples from the network node 120

[0098] - Receiving a configuration from the network node 120 to apply for the training of the one or more AI / ML models according to the first subset of data samples

[0099] - transferring the received first subset of data samples to a third node, e.g., a UE server. transferring the UE-sided condition and / or configuration and the applied received configuration to apply for the first subset of data samples to a third node, e.g., a UE server. receiving from the third node one or more AI / ML models for its UE-sided condition / configuration for the AI / ML enabled feature

[0100] The third node may collect multiple different first subsets of data samples from multiple UEs associated to the same UE-sided condition / configuration and may use these aggregated data samples to train one or more AI / ML models for this UE-sided condition and / or configuration for the AI / ML enabled feature.

[0101] The first subset of data samples may be selected from a first dataset, e.g., UE- part / UE-side model training dataset, by the network node 120 based on the received UE- sided condition and / or configuration of this UE 10.

[0102] The first dataset, e.g., UE-part model training dataset, may be generated by a fourth node, e.g., a gNB, a CN node, a network server, an operation, administration and management (OAM), using one or more two-sided models, e.g., one or more pairs actual NW-part model and nominal UE-part model, trained at the fourth node for the AI / ML- enabled feature together with a second dataset, e.g., a NW-part model training dataset.

[0103] The second dataset may be created at the fourth node, e.g., a gNB, a CN node, a server, an OAM, using the UE report (see c) above) from multiple UEs collected from one or more NW nodes.

[0104] The fourth node may be a different node from the NW node.

[0105] The fourth node may be the network node 120.

[0106] The UE 10 may receive from the third node, e.g., a UE server, one or more AI / ML models for its UE-sided condition and / or configuration for the AI / ML enabled feature.

[0107] The third node may receive one or more first datasets, e.g., UE-part / UE-side model training dataset, from the fourth node and may use the received one or more first datasets to train one or more AI / ML models for one or more UE-sided conditions and / or configurations for the AI / ML enabled feature.

[0108] The one or more first datasets may be created by the fourth node based on UE- sided conditions and / or configurations, e.g., each first dataset is associated to a specific UE-sided condition and / or configuration.

[0109] The one or more first datasets may be generated by the fourth node using one or more two-sided models trained at the fourth node for the AI / ML-enabled feature together with the second dataset.

[0110] The second dataset may be created at the fourth node using the UE report (see c) above) from multiple UEs collected from one or more NW nodes.

[0111] The fourth node may be a different node from the network node 120.

[0112] The fourth node may be the network node 120. The information related to the one or more UE-sided conditions and / or configurations may comprise any of:

[0113] UE-specific conditions under which the data are collected, such as any of UE vendor ID, UE chipset ID, UE release number, Software / hardware version, UE antenna configuration, UE RF impairments, UE power states, Al complexity levels, and Al model structures, UE memory status, UE processing time, means of UE classification information, UE mobility status such as speed, data collection information,

[0114] NW configuration, such as RRC configuration, applied by the UE at the moment of data collection.

[0115] The information related to the one or more UE-sided conditions and / or configurations may be implicitly indicated in forms of IDs, e.g., UE-configuration-ID, Dataset ID.

[0116] The information related to the one or more UE-sided conditions and / or configurations may be transmitted via UE capability report.

[0117] The information related to the one or more UE-sided conditions and / or configurations may be transmitted together with the measurements in the UE report (see c) above).

[0118] The UE may indicate the one or more UE-sided conditions and / or configurations when reporting the measurements to the network node 120 (see c) above).

[0119] The configuration related to data collection (see a) above) may indicate information related to the NW-side condition and / or configuration of the NW node.

[0120] The information related to NW-side condition and / or configuration may include at least one of the scenarios, e.g., NLOS / LOS percentages per cell / site / area, single user (SU) or multi user (MU) scheduling percentages per cell / site, level of interference and source of interference, NW vendor ID, Software / hardware version, the NW antenna configuration related info, NW RF impairments, NW node power states, Al complexity levels, and Al model structures.

[0121] The information related to NW-sided condition / configuration may be implicitly indicated in forms of IDs, e.g., NW-configuration-ID, the second dataset ID, that is indicated to the UE by the NW when providing the associated NW configuration.

[0122] The UE indicates the first set of additional conditions when transmitting the first dataset. Example embodiments of a method performed by the network node 120, such as the radio network node 12, for handling communication in the wireless communications network will now be described with reference to a flowchart depicted in Fig. 6. The actions do not have to be taken in the order stated below but may be taken in any suitable order. Optional actions are marked in dashed boxes.

[0123] Action 601. The network node 120 may transmit a configuration to the UE 10. The configuration may indicate one or more parameters to perform the method herein. The network node 120 may transmit the configuration to the UE 10 to apply for the training of the one or more models according to the first subset of data samples. The configuration may comprise configuration for data collection indicating information related to the network-side condition and / or configuration of the network node. The information related to network-side condition and / or configuration of the network node may include one or more of the following: Line of site (LOS) or non-line of site (NLOS) percentages per cell / site and / or area; single user (SU) or multi user (MU) scheduling percentages per cell / site; level of interference and source of interference; network vendor ID; Software or hardware version, network antenna configuration related information, network radio frequency impairments, network node power states, computational model complexity level, and computational model structure.

[0124] Action 602. The network node 120 receives from the UE 10, information related to one or more UE-sided conditions and / or configurations, such as information of UE-specific conditions under which the data are collected, such as any of UE vendor ID, UE chipset ID, UE release number, Software / hardware version, UE antenna configuration, UE RF impairments, UE power states, Al complexity levels, and Al model structures, UE memory status, UE processing time, means of UE classification information, UE mobility status such as speed, data collection information, NW configuration, such as RRC configuration, applied by the UE at the moment of data collection; data and / or parameters related to one or more computational models.

[0125] Action 603. The network node 120 may collect the first subset of data samples by selecting from the first dataset based on the received UE-sided condition and / or configuration of this UE 10 or collected from the fourth node using one or more two-sided models trained at the fourth node. The network node 120 may train a pair of a NW-part model and nominal UE-part model using a second dataset.

[0126] Action 604. The network node 120 may thus obtain UE-side condition and / or configuration-related information from the UE / UE-side, based on which the network node 120 may label one or more data samples, i.e. , the first subset, within a training dataset based on the received information. Thus, the network node 120 may label the first subset within a training dataset based on the received information. It should be noted that the network node 120 may label data samples within each dataset based on UE-side conditions and / or configurations, as well as a corresponding NW side conditions and / or configurations.

[0127] Action 605. The network node 120 transmits to the UE 10 the first subset of data samples to be used in the one or more computational models at the UE 10, which first subset is labelled based on the received information. The network node 120 delivers to the UE / UE-side the matched training dataset to be used for training the UE-side. Thus, the network node 120 transmits the first subset of data samples to the UE 10 to be used (training) in the one or more computational models. The first subset may comprise the training dataset matched to information related to the one or more computational models to be used for training the one or more computational models on the UE-side.

[0128] Thus, embodiments may provide one or more of the following remarks:

[0129] - The proposed dataset delivery mechanism enables the UE 10 to obtain a more accurate training dataset by matching data distribution / characteristic of the training dataset with the UE-side condition / configuration and the NW-side condition / configuration. Hence, this dataset delivery mechanism can support UE / UE-side to train UE / NW-side condition / configuration specific UE-side / UE- part model(s), giving better model performance, and / or lower computational complexity, and / or smaller model sizes.

[0130] In addition, by using the disclosed mechanism, training data collection can be done more efficiently. For example, the data collection may only be conducted when necessary for a specific group of UEs, i.e. , when a new group of UEs with new UE-side conditions / configurations or when

[0131] In the text herein, the concept of ‘network (NW)’, network node 120, and / or a gNB can be understood as a generic network node, gNB, base station, unit within the base station, relay node, core network node, a core network node, or a device supporting D2D communication. The network node 120 may be deployed in a 5G network, or a 6G network. Moreover, although the term AI / ML model uses a single form, it should be well understood that it should not prevent the implementation of more than one AI / ML model. The UE 10 may either be configured or autonomously switch between the models depending on certain conditions and / or proprietary implementations. Herein two-sided CSI compression is used as an example of a feature that benefits and / or requires a dataset delivery between entities to conduct the AI / ML model training. Note that the disclosure should also be applicable for other AI / ML features that require and / or benefit from a dataset delivery between entities.

[0132] In embodiments herein the UE 10, and the network node 120 for two-sided model cases, is assumed to have the capability of running AI / ML models supporting the AI / ML- enabled features and its respected data collection procedures. In addition, the term “training” may need to also be understood as a generic term and may represent training, retraining, or fine-tuning.

[0133] In some examples, the dataset transmitted by the UE 10 to the network node 120 may have the same size (number) as the reference signals transmitted by the network node 120 for data collection. In the implementation, the size between these two may be different. For example, the UE 10 may not transmit all of the measurement results, and / or part of the transmitted dataset is not received by the network node 120. In addition, some of the examples also assume that the total dataset transmitted by the UE 10 to the network node 120 is the same as the total dataset received by the UE 10 from the network node 120. In the implementation, the network node 120 may discard some of the received dataset, e.g., some of the elements of the dataset have the same and / or similar values, etc.

[0134] Different dataset delivering methods can be considered for training one or more UE-side models or UE-part of two-sided models for an AI / ML-enabled feature.

[0135] In the methods 1a, 1b, and 3a described below, the network node 120, e.g., a gNB, is described as a single entity for simplicity. In general, the network node 120 may be composed of numerous components, and further signalling may be required between the components of the network node 120.

[0136] In one example, the network node 120 may be connected with multiple TRPs or radio units. In this case, the uplink and downlink transmission, e.g., NW-side data collection, over the air (OTA) dataset delivery, between the UEs and the network node 120 are via the multiple TRPs. For instance, a first group of UEs perform UL and / or DL transmission to the network node 120 via a first TRP of the network node 120, a second group of UEs perform UL and / or DL transmission via a second TRP of the given network node 120, a third group of UEs perform UL and / or DL transmission via both the first and second TRPs of the given network node 120, etc. In another example, the network node 120 may be implemented with split architecture, where the network node 120 comprises a gNB- control or central unit (CU) and one or more gNB- distributed unit (DU), and a gNB-DU is connected to the gNB-CU via the F1 interface, i.e. , F1AP. In this case, the AI / ML related processing, e.g., model training, dataset processing, can be performed by a gNB-DU. Alternatively, if the gNB-DU has limited processing power and / or storage capability, then AI / ML related processing, e.g., model training, dataset processing, can be performed by gNB-CU, which requires sending the data samples or (sub)dataset between gNB-DU and gNB-CU via the F1 interface.

[0137] In the methods 2a, 2b and 3b described below, the network node 120, e.g., gNB, is described as a single entity for simplicity. In general, the network node 120 may be composed of numerous components, and further signalling is required between the components of the network node 120.

[0138] In one example, the network node 120 may be connected with multiple TRPs. In this case, the uplink and downlink transmission, e.g., NW-side data collection, OTA dataset delivery, between the UEs and the network node 120 are via the multiple TRPs. For instance, a first group of UEs perform UL and / or DL transmission to the gNB via a first TRP of the network node 120, a second group of UEs perform UL and / or DL transmission via a second TRP of the given network node 120, a third group of UEs perform UL and / or DL transmission via both the first and second TRPs of the given network node 120, etc.

[0139] In another example, the network node 120 may be implemented with split architecture, where the network node 120 comprises a gNB-CU and one or more gNB-DUs, and a gNB-DU is connected to the gNB-CU via the F1 interface. In this case, the interaction between network node 120 and the central entity 2, e.g., a NW-side training centre, involves traversing the F1 interface, aka, F1AP, between gNB-DU and gNB-CU. For example, the collected data samples from the TRPs of the network node 120 are packaged by the connected gNB-DU, then sent from gNB-DU to gNB-CU via F1AP, after that gNB-CU forwards the data samples to the central entity 2. Similarly, when a subDataset is to be sent from the central entity 2 to gNB, the subDataset is first sent to gNB-CU, then passed on to the gNB-DU, the one connected over the air with the UE, via F1AP, after which the subDataset is transmitted on the downlink from a TRP connected to the given gNB-Dll.

[0140] Method group 1 : OTA dataset delivery from NW-side to UE-side:

[0141] Method 1a (Fig. 7)

[0142] In a short summary method 1a includes: UE-side / UE-part model training dataset, called as the first dataset herein, is created by the NW nodes for a UE-side condition and / or configuration; over the air dataset delivery from NW nodes, e.g., gNBs, to UEs; on-device model training.

[0143] Method 1a has one or more of the following actions:

[0144] Action 1 : A NW node, i.e., the network node 120 such as a gNB, collects data samples, e.g., channel measurements, from multiple UEs 10, 10’, and labels the collected data samples with the associated UE-condition and / or configuration information. The network node 120 creates a dataset, called as the second dataset herein, using these collected data samples. Some of the UE-condition and / or configuration used as label may be provided by the UE 10 at the time of reporting the collected data measurements, wherein the said UE-condition / configuration represents the UE-condition and / or configuration at the time of performing the data measurement. In another embodiment some of the UE- condition and / or configuration to be used as label are instead derived by the NW node, such as the RRC configuration of the UE 10 at the time of collecting the measurements. For example, the network node 120 upon receiving the collected data samples from the UE 10 will label them with the UE-specific RRC configuration the UE applied during the data measurement collection. In one embodiment, the labels used by the network node 120 for the collected data samples may be associated with UE-condition / configuration provided by the UE 10 and UE-condition and / or configuration derived by the network node 120.

[0145] Action 2: If no NW-part model and nominal UE-part model is available at the network node 120, the network node 120 trains a pair of the NW-part model and nominal UE-part model using the second dataset. The network node 120 creates another dataset, called as the first dataset herein, using the second dataset and the trained and / or available NW- part model and nominal UE-part model pair. The network node 120 divides the first dataset into sub datasets based on the associated UE-side condition / configuration of each data sample. For instance, subDataset 1 consists of data samples that are associated with UE-side condition and / or configuration 1, and subDataset 2 consists of data samples that are associated with UE-side condition and / or configuration 2. Action 3: The network node 120 delivers part or all of the data samples within a subDataset to one or more of the Al-feature capable UEs 10, 10’, based on the UEs’ associated UE-side condition / configuration. The network node 120 may also indicate as part of the subDataset a configuration that the UE 10 should apply when training according to the provided subDataset. For example, this configuration may be used by the UE 10 to retrain / finetune / update the said subDataSet according to the new provided configuration, e.g., on a different set of beams, or frequencies or cells.

[0146] Additionally, or alternatively, the network node 120 delivers part or all of the data samples within a subDataset to one or more of the Al-feature capable UEs 10, 10’, based on the UEs’ associated UE-side condition and / or configuration. The dataset delivery can be done on demand, i.e. , in response to the UE's request. Or initiated by the NW, e.g., indicates to the UE 10 the availability of the dataset, and confirmed by the UE 10, i.e., the UE 10 is willing to train the model and ready to receive the dataset.

[0147] Action 4: The UE 10, 10’ uses the received data samples to train, retrain, finetune and / or update one or more of its UE-part models for this UE-sided condition and / or configuration for the AI / ML enabled feature.

[0148] Method 1b (Fig. 8)

[0149] In a short summary Method 1b includes:

[0150] • UE-side / UE-part model training dataset, called as the first dataset herein, is created by the NW nodes for a UE-side condition and / or configuration;

[0151] • Over the air dataset delivery from one or more NW nodes to UEs 10, 10’, the UEs forward received data samples to a central entity 1, e.g., a UE training center; being an example of the third node herein, the central entity 1 trains one or more AI / ML models, and delivers the model to the UEs with the UE-side condition and / or configuration.

[0152] Method 1b has one or more of the following steps:

[0153] Actions 1-3: the same as for Method 1a. Additionally, or alternatively, the UE 10 may request the dataset for the NW instead of providing it blindly to all the UEs with same conditions. It could be that the over-the-top (OTT) server designate one or more UEs to fetch this data instead of collecting it from all the UEs.

[0154] Action 4: The UE 10, 10’ transfers the received data samples to a central entity 1, e.g., its corresponding UE-side training center. Action 5: The central entity 1 collects the data samples from multiple UEs 10, 10’ associated to the same UE-sided condition and / or configuration and uses part or all of these aggregated data samples to training one or more UE-part AI / ML models for this UE-sided condition and / or configuration for the AI / ML enabled feature.

[0155] Action 6: The central entity 1 delivers the trained one or more UE-part AI / ML models to the Al-feature capable UEs 10, 10’ with the same UE-side condition and / or configuration.

[0156] Method 2a (Fig. 9)

[0157] In a short summary Method 2a includes:

[0158] • UE-side / UE-part model training dataset, called the first dataset herein, is created by a central entity 2, e.g., a NW server, a CN node, an OAM, called as the fourth node herein, the central entity 2 distributes the data samples in the first dataset to one or more NW nodes;

[0159] • over the air dataset delivery from NW nodes, e.g., gNBs, to UEs;

[0160] • on-device model training.

[0161] Method 2a has one or more of the following steps:

[0162] Action 1 : Network node 120, e.g., a gNB, collects data samples, e.g., channel measurements, from multiple UEs, and labels the collected data samples with the associated UE-condition / configuration information. The network node 120 forwards the collected data samples to a central entity 2, e.g., a NW-side training center, called as the fourth node herein. The central entity 2 creates a dataset, referred to as the second dataset herein, using all or part of these data samples collected from one or more NW nodes.

[0163] Action 2: If no NW-part model and nominal UE-part model are available at the central entity 2, the central entity 2 trains a pair of the NW-part model and nominal UE-part model using the second dataset. The central entity 2 creates another dataset, referred to as the first dataset herein, using the second dataset and the trained and / or available NW-part model and nominal UE-part model pair. The central entity 2 divides the first dataset into sub datasets based on the associated UE-side condition and / or configuration of each data sample. For instance, subDataset 1 consists of data samples that are associated with UE-side condition and / or configuration 1 , and subDataset 2 consists of data samples that are associated with UE-side condition and / or configuration 2. Action 3: The central entity 2 delivers part or all of the data samples within a subDataset to one or more of NW nodes. The NW nodes send the part or all of the received data samples within a subDataset to one or more Al-feature capable UEs, based on the UEs’ associated UE-side condition and / or configuration.

[0164] Action 4: The UE 10 uses the received data samples to train, retrain, finetune, and / or update one or more of its UE-part models for this UE-sided condition and / or configuration for the AI / ML enabled feature.

[0165] Method 2b (Fig. 10)

[0166] Short summary:

[0167] • UE-side / UE-part model training dataset, also called the first dataset herein, is created a central entity 2, e.g., a NW server, a CN node, an OAM, also called the fourth node herein, for a UE-side condition and / or configuration;

[0168] • the central entity 2 distributes the data samples in the first dataset to one or more NW nodes;

[0169] • Over the air dataset delivery from NW nodes, e.g., gNBs, to UEs, the UE forwards data to a central entity 1 , e.g., a UE server, a CN node, an OAM; also called as the third node herein, the central entity 1 trains the AI / ML models and delivers the model to the UEs with the UE-side condition and / or configuration. Method 2b has one or more of the following steps:

[0170] Actions 1-3: the similar actions as for Method 2a. That is, the network node 120, e.g., a gNB, collects data samples. If no NW-part model and nominal UE-part model are available at the central entity 2, the central entity 2 trains a pair of the NW-part model and nominal UE-part model using the second dataset. The central entity 2 delivers part or all of the data samples within a subDataset to one or more of NW nodes.

[0171] Actions 4-6: the same as for Method 1b. That is, the UE transfers the received data samples to a central entity 1 , e.g., its corresponding UE-side training center: The central entity 1 collects the data samples from multiple UEs 10, 10’ associated to the same UE-sided condition / configuration and uses part or all of these aggregated data samples to training one or more UE-part AI / ML models for this UE-sided condition / configuration for the AI / ML enabled feature. The central entity 1 delivers the trained one or more UE-part AI / ML models to the Al-feature capable UEs 10, 10’ with the same UE-side condition and / or configuration.

[0172] Method group 3: over-the-top (OTT) dataset delivery from NW-side to UE-side In the methods 3a and 3b described below, the network node 120, e.g., gNB, is described as a single entity for simplicity. In general, the network node 120 may be composed of numerous components, and further signalling is required between the components of the gNB. For example, the network node 120 may be implemented with split architecture. In this case, the subDatasets are sent from gNB-Cll to UE vendor training centers, while the subDatasets may be produced by either gNB-Dll or gNB- Cll. If the subDatasets are produced by gNB-Dll, the subDatasets are sent from gNB- Dll to gNB-Cll via F1AP, after which gNB-Cll may forward the subDatasets to UE vendor training centers.

[0173] Method 3a (Fig. 11)

[0174] Short summary:

[0175] • UE-side / UE-part model training dataset, also called the first dataset, is created by the NW nodes for a UE-side condition and / or configuration;

[0176] • the NW nodes deliver the first dataset to a central entity 1 , e.g., a UE server, a CN node, an OAM; called the third node herein, the central entity 1 trains the AI / ML models and delivers the model to the UEs with the UE-side condition and / or configuration.

[0177] Method 3a has one or more of the following steps:

[0178] Actions 1-2: the same as for Method 1a. That is, the NW node, i.e. , the network node 120 such as a gNB, collects data samples, e.g., channel measurements, from multiple UEs 10, 10’, and labels the collected data samples with the associated UE-condition and / or configuration information. If no NW-part model and nominal UE-part model is available at the network node 120, the network node 120 trains a pair of the NW-part model and nominal UE-part model using the second dataset.

[0179] Action 3: The NW node 120 delivers part or all of the data samples within a subDataset to a central entity 1, e.g., a UE side training center, called as the third node herein, based on the associated UE-side condition and / or configuration.

[0180] Additionally, the central entity 1 may request the NW to report the dataset. Or the procedure can be initiated by the NW by indicating the availability of new dataset, or some updates to an earlier dataset. The central entity 1 request the NW to report the dataset.

[0181] Action 4: The central entity 1 collects the data samples associated to the same UE- sided condition and / or configuration from one or more NW nodes and uses part or all of these aggregated data samples to train one or more UE-part AI / ML models for this UE- sided condition and / or configuration for the AI / ML enabled feature.

[0182] Action 5: The central entity 1 delivers the trained one or more UE-part AI / ML models to the Al-feature capable UEs with the same UE-side condition and / or configuration.

[0183] Method 3b (Fig. 12)

[0184] Short summary:

[0185] • A UE-side / UE-part model training dataset, called the first dataset herein, is created at a central entity 2, e.g., a NW-side training center, called as the fourth node herein, for each UE-side condition and / or configuration;

[0186] • the central entity 2 delivers the corresponding first dataset to a central entity 1 , e.g., a UE-side training center; called as the third node herein, based on the UE-side condition and / or configuration, the central entity 1 trains the AI / ML models and delivers the trained models to the UEs with the associated UE-side condition and / or configuration.

[0187] Method 3b has one or more of the following steps:

[0188] Actions 1-2: the same as for Method 2a. That is, the network node 120, e.g., a gNB, collects data samples. If no NW-part model and nominal UE-part model are available at the central entity 2, the central entity 2 trains a pair of the NW-part model and nominal UE-part model using the second dataset.

[0189] Action 3: The central entity 2, e.g., the NW side training center, called as the fourth node herein, delivers part or all of the data samples within a subDataset to a central entity 1, e.g., a UE side training center, called as the third node herein, based on the associated UE-side condition and / or configuration.

[0190] Action 4: The central entity 1 uses part or all of the received data samples within the subDataset to training one or more UE-part AI / ML models for this UE-sided condition / configuration for the AI / ML enabled feature.

[0191] Action 5: The central entity 1 delivers the trained models one or more UE-part AI / ML models to the Al-feature capable UEs with the same UE-side condition and / or configuration.

[0192] Note that even though the examples above are for two-sided model use cases. The dataset delivery methods described above can also be simplified and applied for one-sided UE-sided model cases. For example, the first dataset to be delivered from a NW node (or a central entity 2) to a UE (or a central entity 1) can be a subset of the second dataset, which includes the measurements collected from multiple UEs in the network together with other assistance information if needed.

[0193] The UE 10 indicating its UE-side condition / configuration to the NW, see action 502 in Fig. 5.

[0194] For all the methods discussed above, the action 1 may require the network node 120 or the central entity 2, e.g., NW side training center, to be able to categorize and / or label the data samples within the first dataset based on UE-side conditions and / or configurations, so that it can divide the dataset into sub datasets and deliver the correct subDataset to the correct UEs or central entity 1, e.g., UE side training center, based on the UE-side condition and / or configuration.

[0195] To support the data sample labelling and categorizing at the network node 120, or the central entity 2, the UE 10 indicates its UE-side condition and / or configuration to the network node 120.

[0196] In one embodiment, the UE 10 indicates its UE-side condition and / or configuration related information to the network node 120. The UE-side condition and / or configuration related information may include at least one of: UE vendor ID, UE chipset ID, UE release number, Software and / or hardware version, UE antenna configuration, UE RF impairments, UE power states, Al complexity levels, and Al model structures, etc. The UE 10 can report the explicit UE-side condition and / or configuration related parameters to the network node 120. The UE 10 may also report the UE-side condition / configuration related information implicitly, e.g., by using a form of IDs. In one dependent embodiment, the UE-side conditions and / or configurations are included in the UE capability report. This information may also be included when the UE 10 reports its capability to support data collection, or when the UE 10 reports its UE-side additional conditions associated to this Al-enabled feature to the NW, or when the UE 10 reports the collected training data samples to the NW, e.g., the UE-side condition and / or configuration related information is reported together with the data samples transmission from the UE 10 to the network node 120.

[0197] Extension with NW-side condition / configuration for dataset delivery.

[0198] In some scenarios the NW side condition and / or configuration may change, e.g. NW changing antenna tilt, beam pattern, transmit power, etc., would require fine- tuning, updating, and / or retraining the models at the NW and the corresponding UE part model in the two-sided model case. The same is applicable for the UE sided model if the changes at the NW side impact the measurements collected for training.

[0199] In other scenarios, the network node 120 might train multiple NW sided models each corresponding to different NW side conditions and / or configurations.

[0200] This may require the network node 120 or the central entity 2, e.g., NW side training center, to be able to categorize and / or label the data samples within each dataset based on UE-side conditions and / or configurations, as well as the corresponding NW side conditions and / or configurations.

[0201] In this case, the network node 120 may request the UE 10 to report the NW- side condition and / or configuration used for training the model at the UE 10 to confirm that the model is applicable to the NW-side conditions and / or configurations that the network node 120 is operating with. If not applicable, the network node 120 may deliver another dataset that is associated with both the NW and UE conditions. Or alternatively, configure the UE 10 to measure and report new data samples to create a new dataset associated with the NW and UE conditions. See Fig. 13.

[0202] Action 1. The NW side training center: creates a second dataset using data samples collected from multiple gNBs; divides second dataset into one or more subDatasets based on NW / sided additional conditions. For example, subDataset 1 for NW-side condition / configuration ID=1, and subDataset 2 for NW-side condition and / or configuration I D=2.

[0203] Action 2. The NW side training center: Trains NW-part model x+(nominal UE-part model x) using subDataset x of the second Dataset; Creates subDataset x of the first dataset using the trained NW-part and nominal UE-part model pair x together with the corresponding SubDataset x; divides the subDatasets x into one or more subDatasets based on the UE-sided additional conditions. E.g., subDataset x-y for NW-side condition and / or configuration! D=x and UE-side condition and / or configuration I D=y. Action 3: The central entity 2, e.g., the NW side training center, called as the fourth node herein, delivers part or all of the data samples within a subDataset to a central entity 1, e.g., a UE side training center, called as the third node herein, based on the associated UE-side condition and / or configuration.

[0204] Action 4: The central entity 1 uses part or all of the received data samples within the subDataset to training one or more UE-part AI / ML models for this UE-sided condition / configuration for the AI / ML enabled feature.

[0205] Action 5: The central entity 1 delivers the trained models one or more UE-part AI / ML models to the Al-feature capable UEs with the same UE-side condition / configuration. Data collection configuration related aspects.

[0206] In action 1 of the proposed methods in Figs. 7-13, the network node 120 or the central entity 2, e.g., NW side training centre, will collect data samples from UEs to create a second dataset, which will then be used together with a trained model pair at the NW or the central entity 2 to create a first dataset. To enable the second dataset collection at the network node 120 or the central entity 2, some data collection related configuration and / or procedures need to be designed.

[0207] Receiving the UE-side condition and / or configuration from the UE 10, the network node 120 or a central entity 2 may then check whether dataset associated to the UE-side condition / configuration already exists in the network node 120. The network node 120 or the central entity 2 may also optionally or additionally check whether there are already models available to support the UE 10 with the associated UE-side condition and / or configuration.

[0208] In one embodiment, for the case of the network node 120 or the central entity 2, e.g., NW-side training centre, does not have dataset associated with the UE-side condition / configuration, the network node 120 may initiate data collection procedure, e.g., by sending configurations and or indication for data collection to the UE 10.

[0209] In one embodiment, a change in the NW side condition and / or configuration, requires collecting dataset samples to create a new dataset with the UE-side condition and / or configuration. In this case, the network node 120 indicates the availability of a new dataset with the associated UE-side condition and / or configuration and the NW side condition and / or configuration to the UEs and / or central entity 1 , depending on the training method 1 , 2 or 3. If the NW side conditions and / or configuration does not match the NW-side condition labelled in the dataset used for training the UE model. The UE 10 / central entity 1 may request the network node 120 to provide the new dataset. The UE 10 / central entity 1 uses the new dataset to retrain / finetune / update the model.

[0210] In another embodiment, the data collection procedure may be initiated by the UE 10, i.e., by sending an indication that the UE 10 requires data collection procedure for a specific UE-side condition / configuration, i.e., to in return obtain the dataset from the network node 120 that may be used to train the UE-part model.

[0211] For data collection, the UE 10 may be configured by the network node 120 with some configurations related to data collection. For the case of two-sided CSI compression, the configurations may include the configuration of the reference signals used for data collection. The configurations may also include the configuration on how to quantize and / or compress the measured signals. In one example, scalar quantization may be used. In another example, the quantization, or compression, may use the Rel. 16 codebook or the alike, e.g., Rel. 16 with enhanced parameter values of the codebook. In addition, the reference signal used for data collection may be regular reference signals or may be reference signals specifically used for data collection purposes. The reference signal specifically designed for data collection, may for example, has denser time and frequency domain resources and / or higher power reference signals, etc.

[0212] After receiving the configuration, the UE 10 may then receive the reference signals and conduct measurement for the mentioned reference signals.

[0213] The UE 10 reports the data samples including measurement results to the network node 120.

[0214] In action 1 of the proposed methods, the UE 10 may report the measurement results together with other information if needed as data samples to the network node 120. The measured results can be compressed and / or quantized according to a predetermined method in the standard text or may be NW configurable.

[0215] In one embodiment, the UE report carrying data samples for NW-sided data collection contains the UE-side condition and / or configuration related information. As an example, a subset of UE-side condition and / or configuration related parameters is not reported in the UE capability report, and the rest of the related parameters are carried in the UE report carrying data samples. As another example, the UE-side condition and / or configuration related information may be defined as an ID, and this ID is reported together with the data samples.

[0216] In another embodiment, to make sure that the dataset has a correct identification on which scenario, NW configuration, etc., the dataset is collected in, the UE report carrying data samples for NW-sided data collection may also contain NW- side condition and / or configuration related information.

[0217] Triggering events / conditions for dataset delivery and the first dataset generation.

[0218] Receiving at least the data samples from the UE 10, the network node 120 may determine whether to reuse the already stored model pair or retrain its stored model pair or train a new model pair, and whether it needs to conduct a dataset delivery to assist UE-side / UE-part model training, re-training, and / or fine-tuning. In one embodiment, the collected data may need to be obtained from M different UEs with the same additional conditions before the training / retraining / fine- tuning can be started, e.g., to guarantee that the trained model pair will generalize well across UEs with the same additional conditions. The value of M may be specified in the standard text, e.g., as part of data quality assurance requirements for the associated Al-enabled feature, or may be left to the NW proprietary implementation. The maximum number of data samples reported from each UE may also be further considered and / or specified, e.g., to avoid bias.

[0219] In another embodiment, the network node 120 may additionally, or optionally, first consider the data distribution of the collected dataset. If the data distribution of the collected dataset does not match with the dataset, or subset of dataset, currently available at the network node 120, the network node 120 may start the training procedures and dataset delivery. If the data distribution of the collected dataset matches with the dataset, or subset of dataset, that is currently available at the network node 120, and used to train one or more NW-part model(s), e.g., the decoder(s), the NW may omit the training and dataset delivering procedure.

[0220] In another embodiment, the network node 120 may first consider the performance of the pair of nominal UE-part model, e.g., nominal encoder, and NW-part model, e.g., actual decoder, when applying the dataset collected from the UE 10. The performance is, for example, measured using intermediate key performance indicators (KPI), e.g., normalized mean square error (NMSE), squared generalized cosine similarity (SGCS), etc.. If the performance of the mentioned pair is acceptable, e.g., above a certain threshold, the network node 120 may omit training, retraining, and / or fine tuning and dataset delivery procedure.

[0221] If the network node 120 determines to conduct training, retraining, and / or fine- tuning, the network node 120 may then conduct the training based on at least (part of) the second dataset to obtain the nominal UE-part model and the actual NW-part model. Note that the training may also be done by using mixed dataset, i.e. , mixed dataset from the newly collected data from the UE 10 with new additional conditions and the already existing dataset available at the network node 120.

[0222] Using the nominal encoder, either via training, retraining, and / or fine-tuning with at least (part of) second dataset or using the existing models, the network node 120 can generate the first dataset by using the second dataset as the inputs of the nominal UE-part model. The first dataset may consist of the output of the nominal UE-part model or the processed version of the nominal UE-part model output, e.g., the quantized version of the UE-part model outputs, and the associated second dataset.

[0223] The UE or UE-side receives the first dataset from the NW / NW-side, see action 503 in Fig. 5.

[0224] After the network node 120 or the central entity 2, e.g., NW side training centre, has created a first dataset for a specific UE-side condition and / or configuration. It will deliver the dataset to the UE 10 or a central entity 1 , e.g., UE-side training centre.

[0225] In one example, if the first dataset is used to train a single pair of nominal UE- part model and NW-part model. This, for example, may be because the model, in particular the NW-part model, is expected to perform well for all of configurations, deployments, scenarios, etc.

[0226] In one embodiment, the network node 120 may not give any additional label for the first dataset. In another embodiment, one ID may be used to represent the case of the first dataset may be used for all NW additional condition. For example, the first dataset may be labelled as Dataset-0. Here, 0 represents information that this Dataset delivered to the UE 10 may be used for all NW-side conditions and / or configurations.

[0227] In another example, the second dataset may be used by the network node 120 to train multiple pairs of nominal UE-part models and the actual NW-part models. In one example of implementation, the network node 120 may use the dataset to train more than one UE-part models associated to one NW-part model. In this example, each UE-part model is designed to perform well for a specific NW-side condition and / or configuration.

[0228] In one embodiment, the transmission of the first dataset from the network node 120 to the UE 10 may contain information, e.g., one or more IDs, of the NW-side condition / configuration. For example, the transmission of the first dataset with NW-side condition / configuration x may be labelled as Dataset x. Here, x represents the associated NW-side condition and / or configuration.

[0229] In some scenarios, all data in the first dataset can be used to train different NW- side additional conditions. In this case, the UE 10 may receive the first dataset with the size of X times the second dataset used by the network node 120 to train actual NW- part + nominal UE-part model pair(s).

[0230] In other scenarios, part of the first dataset may only be used to train one, or a set of, NW-side conditions and / or configuration. In this case, the UE 10 may receive the first dataset with the same size to the second dataset. E.g., the reference signals during the collection of the second dataset may already be transmitted under different NW-side additional conditions. In this case, the UE 10 will receive the first dataset with the same size to the second dataset. I.e., the first dataset may consist of Dataset 1 , Dataset 2, ... , Dataset X where the sum of the size of Dataset 1 + Dataset 2 + ... + Dataset X equals to the size of the second dataset. The size of each Dataset x may or may not be the same.

[0231] Note that the above two scenarios serve as an example. Combination of the above scenarios may also be applied.

[0232] In one embodiment, the first dataset may further contain information of the UE- side conditions. For example, the second dataset may be labelled as Dataset x-y where y represents a UE-side condition and / or configuration.

[0233] Transmission of the first dataset with a label of x-y is most compatible with UEs with UE-side condition and / or configuration y, e.g., as the dataset matches with the actual UE implementation. Therefore, in a dependent embodiment, the transmission of the first Dataset labelled with x-y may be transmitted only to the UE 10 with additional conditions y.

[0234] Fig. 14 shows a block diagram depicting the UE 10 for handling communication in the wireless communications network.

[0235] The UE 10 may comprise processing circuitry 1401 , e.g. one or more processors, configured to perform the methods herein.

[0236] The UE 10 and / or the processing circuitry 1401 is configured to transmit to the network node 120, information related to the one or more UE-sided conditions and / or configurations.

[0237] The UE 10 and / or the processing circuitry 1401 is configured to receive the first subset of data samples from the network node 120, e.g., a gNB, to be used in the one or more computational models at the UE 10, which first subset is labelled based on the transmitted information. The first subset may comprise a training dataset matched to information related to the one or more computational models to be used for training the one or more computational models on the UE-side. For example, the UE 10 and / or the processing circuitry may be configured to receive a matched training dataset to be used for training the UE-side. The UE 10 and / or the processing circuitry may be configured to train the one or more computational models using the first subset of data samples. The UE 10 and / or the processing circuitry 1401 may be configured to report collected measurements as data samples to the network node 120. The UE 10 and / or the processing circuitry may be configured to receive the configuration from the network node 120 to apply for the training of the one or more computational models according to the first subset of data samples.

[0238] The UE 10 and / or the processing circuitry may be configured to transfer the received first subset of data samples to the third node.

[0239] The UE 10 and / or the processing circuitry may be configured to transfer the one or more UE-sided conditions and / or configurations and the applied received configuration to apply for the first subset of data samples to the third node.

[0240] The UE 10 and / or the processing circuitry may be configured to receive from the third node, one or more computational models for the one or more UE-sided conditions and / or configurations.

[0241] The UE 10 and / or the processing circuitry may be configured to receive from the third node, one or more computational models for a UE-sided condition and / or configuration.

[0242] The one or more UE-sided conditions and / or configurations may comprise one or more of: a UE-specific condition under which the data are collected; and a network configuration applied by the UE at the moment of data collection.

[0243] The UE 10 and / or the processing circuitry 1401 may be configured to obtain configuration and / or condition from a network node, internally, preconfigured.

[0244] The UE 10 and / or the processing circuitry 1401 may be configured to train one or more computational models using the first subset of data samples.

[0245] The UE 10 further comprises a memory 1405. The memory comprises one or more units to be used to store data on, such as indications, computational model, performance metrics, reconfiguration, applications to perform the methods disclosed herein when being executed, and similar. The UE 10 comprises a communication interface 1406 comprising transmitter, receiver, transceiver and / or one or more antennas. Thus, it is herein provided the UE 10 for handling communication in a wireless communications network, wherein the UE 10 comprises processing circuitry and a memory, said memory comprising instructions executable by said processing circuitry whereby said UE 10 is operative to perform any of the methods herein.

[0246] The methods according to the embodiments described herein for the UE 10 are respectively implemented by means of e g. a computer program product 1407 or a computer program product, comprising instructions, i.e. , software code portions, which, when executed on at least one processor, cause the at least one processor to carry out the actions described herein, as performed by the UE 10. The computer program product 1407 may be stored on a computer-readable storage medium 1408, e g. a universal serial bus (USB) stick, a disc or similar. The computer-readable storage medium 1408, having stored thereon the computer program product, may comprise the instructions which, when executed on at least one processor, cause the at least one processor to carry out the actions described herein, as performed by the UE 10. In some embodiments, the computer-readable storage medium may be a non-transitory or transitory computer- readable storage medium.

[0247] Fig. 15 shows a block diagram depicting the network node 120 for handling communication in the wireless communications network.

[0248] The network node 120 may comprise processing circuitry 1501 , e.g. one or more processors, configured to perform the methods herein.

[0249] The network node 120 and / or the processing circuitry 1501 is configured to receive from the UE 10, information related to one or more UE-sided conditions and / or configurations.

[0250] The network node 120 and / or the processing circuitry 1501 is configured to transmit the first subset of data samples to the UE to be used in the one or more computational models at the UE 10, which first subset is labelled based on the received information. The first subset may comprise a training dataset matched to information related to the one or more computational models to be used for training the one or more computational models on the UE-side. Thus, the network node 120 and / or the processing circuitry 1501 may be configured to transmit a matched training dataset to be used for training the UE-side. The network node 120 and / or the processing circuitry 1501 may be configured to label one or more data samples, i.e. , the first subset, within the training dataset based on the received information.

[0251] The network node 120 and / or the processing circuitry 1501 may be configured to transmit the configuration to the UE 10 to apply for the training of the one or more models according to the first subset of data samples. The configuration may comprise configuration for data collection indicating information related to the network-side condition and / or configuration of the network node. The information related to networkside condition and / or configuration of the network node may include one or more of the following: Line of site or non-line of site percentages per cell / site / area; single user or multi user scheduling percentages per cell / site; level of interference and source of interference; network vendor ID; Software or hardware version, network antenna configuration related information, network radio frequency impairments, network node power states, computational model complexity level, and computational model structure.

[0252] The network node 120 and / or the processing circuitry 1501 may be configured to collect the first subset of data samples by selecting from a first dataset based on the received UE-sided condition / configuration of this UE or collected from a fourth node using one or more two-sided models trained at the fourth node.

[0253] The network node 120 further comprises a memory 1505. The memory comprises one or more units to be used to store data on, such as indications, data, datasets, computational model, performance metrics, reconfiguration, applications to perform the methods disclosed herein when being executed, and similar. The network node 120 comprises a communication interface 1506 comprising transmitter, receiver, transceiver and / or one or more antennas. Thus, it is herein provided the network node 120 for handling communication in a wireless communications network, wherein the network node 120 comprises processing circuitry and a memory, said memory comprising instructions executable by said processing circuitry whereby said network node 120 is operative to perform any of the methods herein.

[0254] The methods according to the embodiments described herein for the network node 120 are respectively implemented by means of e.g. a computer program product 1507 or a computer program product, comprising instructions, i.e. , software code portions, which, when executed on at least one processor, cause the at least one processor to carry out the actions described herein, as performed by the network node 120. The computer program product 1507 may be stored on a computer-readable storage medium 1508, e.g. a USB stick, a disc or similar. The computer-readable storage medium 1508, having stored thereon the computer program product, may comprise the instructions which, when executed on at least one processor, cause the at least one processor to carry out the actions described herein, as performed by the network node 120. In some embodiments, the computer-readable storage medium may be a non- transitory or transitory computer-readable storage medium.

[0255] As will be readily understood by those familiar with communications design, that functions means or modules may be implemented using digital logic and / or one or more microcontrollers, microprocessors, or other digital hardware. In some embodiments, several or all of the various functions may be implemented together, such as in a single application-specific integrated circuit (ASIC), or in two or more separate devices with appropriate hardware and / or software interfaces between them. Several of the functions may be implemented on a processor shared with other functional components of a radio network node, for example.

[0256] Alternatively, several of the functional elements of the processing means discussed may be provided through the use of dedicated hardware, while others are provided with hardware for executing software, in association with the appropriate software or firmware. Thus, the term “processor” or “controller” as used herein does not exclusively refer to hardware capable of executing software and may implicitly include, without limitation, digital signal processor (DSP) hardware, read-only memory (ROM) for storing software, random-access memory for storing software and / or program or application data, and non-volatile memory. Other hardware, conventional and / or custom, may also be included. Designers of communications receivers will appreciate the cost, performance, and maintenance trade-offs inherent in these design choices.

[0257] In certain embodiments, some or all of the functionality described herein may be provided by processing circuitry executing instructions stored on in memory, which in certain embodiments may be a computer program product in the form of a non- transitory computer-readable storage medium. In alternative embodiments, some or all of the functionalities may be provided by the processing circuitry without executing instructions stored on a separate or discrete device-readable storage medium, such as in a hard-wired manner. In any of those particular embodiments, whether executing instructions stored on a non-transitory computer-readable storage medium or not, the processing circuitry can be configured to perform the described functionality. The benefits provided by such functionality are not limited to the processing circuitry alone or to other components of the computing device but are enjoyed by the computing device as a whole, and / or by end users and a wireless network generally.

[0258] It will be appreciated that the foregoing description and the accompanying drawings represent non-limiting examples of the methods and apparatus taught herein. As such, the apparatus and techniques taught herein are not limited by the foregoing description and accompanying drawings. Instead, the embodiments herein are limited only by the following claims and their legal equivalents.

Claims

CLAIMS1. A method performed by a user equipment, UE, (10) for handling communication in a wireless communications network (1), the method comprising: transmitting (502) to a network node (120) information related to one or more UE-sided conditions and / or configurations, and receiving (503) a first subset of data samples from the network node (120) to be used in one or more computational models at the UE (10), which first subset is labelled based on the transmitted information.

2. The method according to claim 1, wherein the first subset comprises a training dataset matched to information related to the one or more computational models to be used for training the one or more computational models on the UE-side.

3. The method according to any of the claims 1-2, further comprising training (504) the one or more computational models using the first subset of data samples.

4. The method according to any of the claims 1-3, wherein transmitting (502) the information comprises reporting collected measurements as data samples to the network node (120).

5. The method according to any of the claims 1-4, further comprising: receiving (501) a configuration from the network node (120) to apply for the training of the one or more computational models according to the first subset of data samples.

6. The method according to any of the claims 1-5, further comprising one or more of the following: transferring (506) the received first subset of data samples to a third node; transferring (507) the one or more UE-sided conditions and / or configurations and an applied received configuration to apply for the first subset of data samples to the third node; and receiving (508) from the third node one or more computational models for the one or more UE-sided conditions and / or configurations.

7. The method according to any of the claims 1-6, further comprising receiving (505) from a third node, one or more computational models for a UE- sided condition and / or configuration.

8. The method according to any of the claims 1-7, wherein the one or more UE-sided conditions and / or configurations comprise one or more of: a UE-specific condition under which the data are collected; and a network configuration applied by the UE at the moment of data collection.

9. A method performed by a network node (120) for handling communication in a wireless communications network, the method comprising receiving (602) from a user equipment, UE (10), information related to one or more UE-sided conditions and / or configurations, and transmitting (605) to the UE (10) a first subset of data samples to be used in one or more computational models at the UE (10), which first subset is labelled based on the received information.

10. The method according to claim 9, wherein the first subset comprises a training dataset matched to information related to the one or more computational models to be used for training the one or more computational models on the UE-side.

11. The method according to any of the claims 9-10, further comprising labelling (604) the first subset within a training dataset based on the received information.

12. The method according to any of the claims 9-11 , further comprising transmitting (601) a configuration to the UE (10) to apply for the training of the one or more models according to the first subset of data samples.

13. The method according to claim 12, wherein configuration comprises configuration for data collection indicating information related to the network-side condition and / or configuration of the network node.

14. The method according to claim 13, wherein the information related to network-side condition and / or configuration of the network node includes one or more ofthe following: Line of site or non-line of site percentages per cell / site / area; single user or multi user scheduling percentages per cell / site; level of interference and source of interference; network vendor ID; Software or hardware version, network antenna configuration related information, network radio frequency impairments, network node power states, computational model complexity level, and computational model structure.

15. The method according to any of the claims 9-14, further comprising collecting (603) the first subset of data samples by selecting from a first dataset based on the received UE-sided condition / configuration of this UE or collected from a fourth node using one or more two-sided models trained at the fourth node.

16. A user equipment, UE, (10) for handling communication in a wireless communications network (1), wherein the UE (10) is configured to: transmit to a network node (120), information related to one or more UE- sided conditions and / or configurations, and receive a first subset of data samples from the network node (120) to be used in one or more computational models at the UE (10), which first subset is labelled based on the transmitted information.

17. The UE according to claim 16, wherein the UE is configured to perform the method according to any of the claims 2-8.

18. A network node (120) for handling communication in a wireless communications network (1), wherein the network node is configured to: receive from a user equipment, UE (10), information related to one or more UE-sided conditions and / or configurations, and transmit to the UE (10) a first subset of data samples to be used in one or more computational models at the UE (10), which first subset is labelled based on the received information.

19. The network node according to claim 18, wherein the network node is configured to perform the method according to any of the claims 10-15.

20. A computer program product comprising instructions, which, when executed on at least one processor, cause the at least one processor (1401 , 1501) to carry out the method according to any of the claims 1-15, as performed by the UE (10) and the network node (120), respectively.

21. A computer-readable storage medium, having stored thereon a computer program product (1407, 1507) comprising instructions which, when executed on at least one processor, cause the at least one processor to carry out the method according to any of the claims 1-15, as performed by the UE (10) and the network node (120), respectively.

Images