Method and apparatus for model transfer in wireless communication system

Abstract

According to one embodiment of the present disclosure, a method performed by a first device in a wireless communication system comprises the steps of: receiving information related to one or more models from a second device; and determining a model application time (MAT) associated with the one or more models. The MAT is determined based on at least one of i) a model transfer type, ii) a number of model transfers, iii) a number of one or more models, and / or iv) a model transfer unit.

Description

Technical Field

[0001] The present disclosure relates to a method and apparatus for model transfer in a wireless communication system. Background Art

[0002] Mobile communication systems have been developed to provide voice services while ensuring user mobility. However, the scope of mobile communication systems has expanded to include data services in addition to voice. Due to the current explosive increase in traffic, resource shortages have emerged, leading to user demands for higher-speed services. Therefore, a more advanced mobile communication system is needed.

[0003] The requirements for next-generation mobile communication systems need to support the adaptation to explosive data traffic, a significant increase in data rates per user, an adaptation to a significant increase in the number of connected devices, very low end-to-end latency, and high energy efficiency. To this end, various technologies have been studied, such as dual connectivity, massive multiple-input multiple-output (MIMO), in-band full-duplex, non-orthogonal multiple access (NOMA), ultra-wideband support, and device networking.

[0004] Communication functionality can be implemented based on models. In this regard, model transfer / delivery can be performed. For example, a model (or information about a model) can be transferred from a first device (e.g., a terminal, a base station, or a first terminal) to a second device (e.g., a base station, a terminal, or a second terminal). Summary of the Invention

[0005] Technical issues

[0006] When the time points at which operations based on transferred / delivered models can be performed are consistently determined / defined and used, this can be inefficient in terms of reliability / latency of communication functions. For example, after performing a specific model transfer, the time required to apply / use the model (e.g., the time required to compile the transferred model) can vary depending on the model structure, the number of models, the capabilities of the entity receiving the model, etc. That is, if a consistently defined time point is used for all models (and / or all model transfers / deliveries), the following problems may arise.

[0007] As an example, it can be assumed that a time point consistently defined for all models (and / or all model transfers / deliveries) is earlier than the time required to apply / use the specific model being delivered. In this case, an operation based on a specific model may be triggered at a time point when it is impossible to use the specific model (a time point when preparations for application / use of the specific model are not complete). As a result, an operation based on a specific model may not be executed normally or may cause additional signaling after the time point when the specific model can be applied.

[0008] As an example, it can be assumed that the time point consistently defined for all models (and / or all model transfers / deliveries) is later than the time required to apply / use a specific, concrete model being delivered. In this case, the time point at which an operation based on a specific model is executed may be later than the time point at which preparations for use / application of the specific model are completed. In other words, unnecessary delays may occur before initiating an operation based on the specific model.

[0009] An object of the present disclosure is to provide a method for solving the above-mentioned problems.

[0010] The technical objectives to be achieved by the present disclosure are not limited to those described above merely by way of example, and other technical objectives not mentioned can be clearly understood from the following description by those skilled in the art to which the present disclosure pertains.

[0011] Technical Solution

[0012] A method performed by a first device in a wireless communication system according to an embodiment of the present disclosure includes receiving information for at least one model from a second device, and determining a model application time (MAT) associated with the at least one model.

[0013] The MAT is determined based on at least one of: i) a model transfer type, ii) a model transfer number, iii) a quantity of at least one model, and / or iv) a model transfer unit.

[0014] The MAT may be a time related to a time from a reference time point to a time point at which verification, testing, application, deployment, or compilation of one or more models is completed by the first device.

[0015] The reference time point may be i) the time point at which the model transfer process is triggered or initiated by the second device, ii) the time point at which the first device sends a response to the triggering of the model transfer process, iii) the time point at which the transmission of information is initiated, iv) the time point at which the transmission of information is completed, or v) the time point at which the first device sends a response to the receipt of information.

[0016] The MAT may be the time required for the first device to activate, apply, or prepare the model or a function associated with the model.

[0017] The method may further include reporting information regarding a minimum MAT supported by the first device to the second device, and receiving information related to the MAT from the second device.

[0018] The MAT can be greater than or equal to the minimum MAT.

[0019] The model transfer type may be defined based on whether the information includes information for a structure associated with one or more models.

[0020] Based on the model transfer type being a first type, the information may include information on parameters associated with one or more models.

[0021] The model-based transfer type is the second type, and the information may include i) information on parameters and ii) information on structures.

[0022] The MAT associated with the second type may be greater than the MAT associated with the first type.

[0023] The model transfer number may be related to whether the transfer of at least one model based on the information is an initial transfer.

[0024] The MAT associated with the initial transition may be greater than the MAT associated with 2 or more model transitions.

[0025] Based on the number of at least one model being greater than the defined number, the MAT may be determined based on a priority associated with each model.

[0026] The priority may be defined based on at least one of: i) a start time point of model transfer associated with each model, ii) a function associated with each model, iii) a model transfer type, iv) a number of model transfers, and / or v) an ID of each model.

[0027] The model transfer unit may be related to the number of models that can be simultaneously transferred to the first device.

[0028] The MAT may be determined as a first MAT based on the number of at least one model that is less than or equal to the model transfer unit. The MAT may be determined as a second MAT based on the number of at least one model that is greater than the model transfer unit. The second MAT may be greater than the first MAT.

[0029] A first device operating in wireless communication according to another embodiment of the present disclosure includes one or more transceivers, one or more processors, and one or more memories connected to the one or more processors and storing instructions.

[0030] The instructions configure the one or more processors to perform all the steps of any one of the methods upon execution by the one or more processors.

[0031] A device according to another embodiment of the present disclosure includes one or more memories and one or more processors operatively connected to the one or more memories.

[0032] The one or more memories store instructions that, upon execution by the one or more processors, configure the one or more processors to perform all the steps of any one of the methods.

[0033] One or more non-transitory computer-readable media according to another embodiment of the present disclosure store instructions. The instructions executable by one or more processors configure the one or more processors to perform all the steps of any one of the methods.

[0034] A method performed by a second device in a wireless communication system according to another embodiment of the present disclosure includes sending information related to at least one model to a first device, and determining a model application time (MAT) related to the at least one model.

[0035] The MAT is determined based on at least one of: i) a model transfer type, ii) a model transfer number, iii) a quantity of at least one model, and / or iv) a model transfer unit.

[0036] A second device operating in wireless communication according to another embodiment of the present disclosure includes one or more transceivers, one or more processors, and one or more memories connected to the one or more processors and storing instructions.

[0037] The instructions configure the one or more processors to perform all the steps of the method upon execution by the one or more processors.

[0038] Beneficial effects

[0039] According to the embodiments of the present disclosure, when a specific model is transferred / delivered, the application time point of the specific model becomes clear. It can be expected that the operation / communication function based on the delivered model will be performed normally based on the application time point, and the reliability of the operation / communication function can be ensured.

[0040] Furthermore, unnecessary delays / signaling overheads in the execution of operations / communication functions based on the delivered model may be reduced.

[0041] Effects that can be achieved by the present disclosure are not limited to those described above by way of example, and other effects and advantages of the present disclosure will be more clearly understood by those skilled in the art to which the present disclosure pertains from the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Illustrate the functional framework of AI / ML models.

[0043] Figure 2 A signaling process according to an embodiment of the present disclosure is illustrated.

[0044] Figure 3 is a flowchart illustrating a method performed by a user equipment according to an embodiment of the present disclosure.

[0045] Figure 4is a flowchart illustrating a method performed by a base station according to another embodiment of the present disclosure.

[0046] Figure 5 The configurations of the first device and the second device according to the embodiment of the present disclosure are illustrated. DETAILED DESCRIPTION

[0047] The following will be Figure 1 The detailed description disclosed above will describe exemplary embodiments of the present disclosure rather than describing unique embodiments for carrying out the present disclosure. The following detailed description includes details that provide a complete understanding of the present disclosure. However, those skilled in the art will appreciate that the present disclosure can be carried out without these details.

[0048] In some cases, in order to prevent the concepts of the present disclosure from being obscure, known structures and devices may be omitted, or may be illustrated in a block diagram format based on the core functions of each structure and device.

[0049] In the following, downlink (DL) means communication from a base station to a terminal, and uplink (UL) means communication from a terminal to a base station. In the downlink, the transmitter may be part of the base station, and the receiver may be part of the terminal. In the uplink, the transmitter may be part of the terminal, and the receiver may be part of the base station. The base station may be represented as a first communication device, and the terminal may be represented as a second communication device. The base station (BS) may be replaced with terms including a fixed station, a node B, an evolved node B (eNB), a next-generation node B (gNB), a base transceiver system (BTS), an access point (AP), a network (5G network), an AI system, a roadside unit (RSU), a vehicle, a robot, an unmanned aerial vehicle (UAV), an AR (augmented reality) device, a VR (virtual reality) device, and the like. In addition, the terminal may be fixed or mobile and may be replaced with terms including user equipment (UE), mobile station (MS), user terminal (UT), mobile subscriber station (MSS), subscriber station (SS), advanced mobile station (AMS), wireless terminal (WT), machine type communication (MTC) device, machine-to-machine (M2M) device and device-to-device (D2D) device, vehicle, robot, AI module, unmanned aerial vehicle (UAV), AR (augmented reality) device, VR (virtual reality) device, etc.

[0050] AIML related description

[0051] With the advancement of artificial intelligence / machine learning (AI / ML) technologies, the nodes and UEs that constitute wireless communication networks are becoming increasingly intelligent / advanced.

[0052] In particular, due to the intelligence of the network / base station, it is expected that various network / base station decision parameters can be quickly optimized and derived / applied based on various environmental parameters.

[0053] Environmental parameters may include at least one of the distribution / location of base stations, the distribution / location / material of buildings / furniture, the location / movement direction / speed of the UE, or climate information. However, the above parameters are merely examples, and in addition to the listed parameters, environmental parameters may also include other environmental parameters related to network / base station decision parameters.

[0054] The network / base station decision parameters may include at least one of the transmit / receive power of each base station (BS), the transmit power of each UE, the precoder / beam of the BS / UE, the time / frequency resource allocation for each UE, or the duplex method of each BS. However, the above parameters are only examples, and in addition to the listed parameters, the network / base station decision parameters may also include other parameters determined by the network / base station.

[0055] In line with this trend, many standardization organizations (e.g., 3GPP, O-RAN) are considering the introduction of AI / ML, and related research is also actively underway.

[0056] In a narrow sense, AI / ML can be simply referred to as artificial intelligence based on deep learning, but conceptually, it can be categorized as follows.

[0057] - Artificial Intelligence: It is any automation that allows machines to do work that would otherwise be done by humans.

[0058] -Machine Learning: It refers to a technology in which machines learn patterns from data for decision making without explicit programming rules.

[0059] Deep learning: It is a model based on artificial neural networks that allows machines to simultaneously perform feature extraction and decision-making based on unstructured data. The algorithm relies on a multi-layer network of interconnected nodes for feature extraction and transformation, which is inspired by biological neural systems (i.e., neural networks). Common deep learning network architectures include deep neural networks (DNNs), recurrent neural networks (RNNs), and convolutional neural networks (CNNs).

[0060] As mentioned above, artificial intelligence (AI) is the broadest concept of AI / ML, and deep learning is the narrowest concept of AI / ML. Machine learning (ML) can be interpreted as a concept narrower than artificial intelligence and broader than deep learning.

[0061] Types of AI / ML based on various criteria

[0062] -Offline and online

[0063] Offline learning

[0064] Offline learning faithfully follows the sequential process of database collection, learning, and prediction. That is, collection and learning can be performed offline, and the completed program can be installed on-site and used for prediction. This offline learning approach is used in most cases.

[0065] Online Learning

[0066] Online learning refers to a method that gradually improves performance by performing incremental learning using additional data generated, taking advantage of the fact that data that can be used for recent learning is continuously generated via the Internet.

[0067] Classification based on AI / ML framework concepts

[0068] -Focused learning

[0069] In centralized learning, all data resources / storage / learning (e.g., supervised learning, unsupervised learning, reinforcement learning, etc.) are performed in one centralized node while training data collected from multiple different nodes is reported to the centralized node.

[0070] - Federated Learning

[0071] Federated learning is built on data, where a collective model exists across distributed data owners. Instead of collecting data into the model, the AI / ML model is imported into the data source, allowing local nodes / individual devices to collect data and train their own copies of the model, thus eliminating the need to report source data to a centralized node.

[0072] In federated learning, the parameters and weights of AI / ML models are retransmitted to centralized nodes to support general model training. Federated learning offers advantages in terms of improved computing speed and information security. Specifically, uploading personal data to a centralized server is unnecessary, preventing the leakage and misuse of personal information.

[0073] -Distributed learning

[0074] Distributed learning refers to the concept of scaling and distributing the machine learning process across a cluster of nodes. The trained model is shared across multiple nodes, which are split up and operated simultaneously to accelerate model training.

[0075] Classification according to learning methods

[0076] -Supervised learning

[0077] Supervised learning is a machine learning task that aims to learn a mapping function from input to output given a labeled dataset. The input data is called training data and has known labels or outcomes. An example of supervised learning is shown below.

[0078] 1) Regression: linear regression, logistic regression

[0079] 2) Instance-based algorithm: k-nearest neighbor (KNN)

[0080] 3) Decision Tree Algorithm: CART

[0081] 4) Support Vector Machine: SVM

[0082] 5) Bayesian algorithm: Naive Bayes

[0083] 6) Ensemble Algorithms: Extreme Gradient Boosting, Bagging: Random Forest

[0084] Supervised learning can be further grouped into regression and classification problems, where classification is predicting a label and regression is predicting a quantity.

[0085] -Unsupervised learning

[0086] Unsupervised learning is a machine learning task that aims to learn a function that describes the hidden structure in unlabeled data. The input data is unlabeled, and the outcome is unknown. Some examples of unsupervised learning include K-means clustering, principal component analysis (PCA), nonlinear independent component analysis (ICA), and long short-term memory (LSTM).

[0087] -Reinforcement Learning

[0088] In reinforcement learning (RL), an agent aims to optimize a long-term goal by interacting with an environment based on a trial-and-error process. It is goal-oriented learning based on interactions with the environment. Examples of RL algorithms are as follows.

[0089] 1) Q-learning

[0090] 2) Multi-arm probabilistic machine (bandit) learning

[0091] 3) Deep Q Network

[0092] 4) State-Action-Reward-State-Action (SARSA)

[0093] 5) Time difference learning

[0094] 6) Executor-Critic Reinforcement Learning

[0095] 7) Deep Deterministic Policy Gradient

[0096] 8) Monte Carlo Tree Search

[0097] Reinforcement learning can be further grouped into model-based reinforcement learning and model-free reinforcement learning.

[0098] Model-based reinforcement learning: It refers to RL algorithms that use predictive models and use a model of the various dynamic states of the environment and the rewards these states bring to the environment to derive the probabilities of transitions between states.

[0099] Model-free reinforcement learning: It refers to RL algorithms that are based on values ​​or policies that maximize future rewards. The multi-agent environment / state is less computationally complex and does not require an accurate representation of the environment.

[0100] RL algorithms can also be classified into value-based RL and policy-based RL, policy-based RL and non-policy RL, etc.

[0101] Representative models of deep learning

[0102] 1. Feedforward Neural Network (FFNN)

[0103] FFNN consists of an input layer, a hidden layer, and an output layer.

[0104] 2. Recurrent Neural Network (RNN)

[0105] RNN is a type of artificial neural network in which hidden nodes are connected to directed edges to form a directed loop. It is a model suitable for processing sequential data such as speech and text.

[0106] 3. Convolutional Neural Network (CNN)

[0107] CNN is used for two purposes: reducing model complexity and extracting good features by applying convolution operations commonly used in the fields of video processing or image processing.

[0108] - Kernel or Filter: It refers to a unit / structure that applies weights to the input of a specific range / unit.

[0109] - Stride: It refers to the range of movement of the kernel within the input.

[0110] - Feature map: It refers to the result of applying the kernel to the input.

[0111] -Padding: It refers to the value that is added to resize the feature map.

[0112] - Pooling: It refers to the operation of reducing the size of feature maps by downsampling them (e.g., max pooling, average pooling).

[0113] 4. Autoencoder

[0114] An autoencoder is a neural network that takes a feature vector x as input and outputs the same or a similar vector x'. The input nodes and output nodes of the autoencoder have the same features.

[0115] Figure 1 Illustrate the functional framework of AI / ML models.

[0116] exist Figure 1 In the functional framework shown, the definition of each term and the operation of each function can be based on Table 1 below.

[0117] [Table 1]

[0118]

[0119]

[0120]

[0121] Dataset

[0122] The datasets used in AI / ML are classified into training data, validation data, and test data, and their definitions are as follows.

[0123] - Training data

[0124] Dataset used to train the model

[0125] - Verify data

[0126] Dataset used to validate the trained model

[0127] Validation data is a dataset that is often used to prevent overfitting of the training dataset.

[0128] Validation data is a dataset used to select the best model among the various models trained during the training process. Therefore, validation data can be considered as a dataset related to training.

[0129] -Test data

[0130] The test data is the dataset used for final evaluation. The test data is independent of the training data.

[0131] The data set may use a training set including the above data in a predetermined proportion.

[0132] For example, a training set consisting of training data and validation data in a ratio of 8:2 or 7:3 can be used.

[0133] For example, a training set including training data, validation data, and test data in a ratio of 6:2:2 may be used.

[0134] Collaboration Level

[0135] Depending on whether the AI / ML function between the base station and the UE is capable, the cooperation level can be defined as shown in Table 2 below.

[0136] [Table 2]

[0137]

[0138] The collaboration levels shown in Table 2 are examples and can be modified and utilized differently from the illustrated examples based on implementation methods. For example, a collaboration level that combines two or more of the illustrated collaboration levels can be defined / utilized. For example, a collaboration level that does not include one or more of the illustrated collaboration levels can be utilized.

[0139] In recent years, efforts to apply artificial intelligence / machine learning (AI / ML) technologies to wireless communication networks have been active. Specifically, in 3GPP Rel-18, research began on applying AI / ML technologies to the air interface between the UE and the network. Beam management (BM), CSI acquisition, and positioning were considered as the main use cases for integrating AI / ML into the air interface in this research, and this disclosure proposes a method for determining the application time point of the corresponding model when transferring the AI / ML model.

[0140] The embodiments described below are not limited to the case where the AI / ML model is transferred. As an example, the embodiments described below can also be applied to the case where the model for implementing the communication function is transferred. As an example, the embodiments described below can also be extended and applied to the case where the model related to the communication function is transferred. In the following, "AI / ML model" can be interpreted / replaced with "model", "model related to (specific) communication function", "model for (specific) telecommunication function", "model related to (specific) communication process" or "model for (specific) telecommunication process".

[0141] In the present disclosure, “ / ” means “and”, “or”, or “and / or”, depending on the context. In the present disclosure, from the perspective of the air interface, “terminal” and “UE” can be used interchangeably in the same / similar sense, and “base station”, “network” and “TRP” can also be used interchangeably in the same / similar sense.

[0142] In the 3GPP Rel-18 AI / ML study project, two terms related to the model transfer operation between devices / entities, “model transfer” and “model delivery”, are defined as shown in Table 3 below.

[0143] Specifically, both terms refer to the operation of transferring models between devices / entities. Model transfer is defined as a transfer over an air interface. On the other hand, "model delivery" is a term that covers model delivery by means other than an air interface.

[0144] The embodiments of the present disclosure are primarily described with the "model transfer" operation as the target, but this is for ease of description only and is not intended to limit the scope of application of the embodiments of the present disclosure. The embodiments of the present disclosure can also be applied to the "model delivery" operation. Therefore, hereinafter, "model transfer" may be interpreted / replaced with "model delivery."

[0145] [Table 3]

[0146]

[0147]

[0148]

[0149] Problems to be solved by the present disclosure are as follows.

[0150] When the AI / ML model training and inference entities differ, a model transfer from the training entity to the inference entity may occur. In this case, a commitment must be made as to the point in time at which the transferred model can be applied. Specifically, only when the applicable point in time of the transferred model between the transferring entity and the receiving entity is committed can the point in time at which the relevant operations / functions can be enabled be identified.

[0151] Hereinafter, in the present disclosure, the following assumptions will be described: the subject of the transfer model is the base station / network (NW) (any entity in the group), and the subject of the reception model is the terminal. However, this is for ease of description, and as described above, the embodiments of the present disclosure can also be applied to transfer operations based on other device / entity combinations (e.g., UE→BS(NW), UE2→UE1). That is, in the following embodiments, the "terminal" can be replaced by the base station (NW) or the first terminal, and the "base station" can be replaced by the terminal or the second terminal.

[0152] Model transfer operations can be classified into several types. Referring to the definition of "model transfer", model transfer operations can be classified as follows.

[0153] As an example, the model transfer operation can be divided into i) a transfer operation for a "complete model" and ii) a transfer operation for a "partial model".

[0154] As an example, the model transfer operation can be divided into a) a transfer operation on model parameters in a specific model structure (hereinafter referred to as parameter transfer for convenience) and b) a transfer operation on model parameters and model structure.

[0155] A more detailed classification of three or more steps is also possible, instead of the two-step classification described above.

[0156] As an example, the model transfer operation can be distinguished based on at least one of the above examples (at least one of i), ii), a), and / or b). As a specific example, the model transfer operation can be a transfer operation for model parameters and / or model structure related to the complete model / partial model.

[0157] As described above, the type associated with model transfer (model transfer type) can be based on at least one of i) the type of model (e.g., complete model / partial model) and / or ii) the type of information associated with the model (e.g., model parameters / model structure / model parameters and model structure).

[0158] As for the application start time point of the transferred model to be addressed in the present disclosure (hereinafter referred to as model application time (MAT)), the above-mentioned various model transfer types can be considered.

[0159] Here, MAT may refer to the time it takes / required for the device receiving the model transfer to actually apply / deploy the transferred model based on a specific time point associated with the model transfer. The base station may anticipate / expect the terminal to apply / deploy the transferred model after the corresponding time point of the defined / configured MAT.

[0160] As an example, the specific time point may be i) the time point at which the base station triggers / initiates a transfer to the terminal, ii) the time point at which the terminal sends an ACK in response to a transfer indication, iii) the time point at which the transmission of model-related data from the base station to the terminal starts / is completed, or iv) the time point at which the terminal sends model data to the base station and reception is normally completed (e.g., an ACK for the corresponding data).

[0161] After the time point based on the above example, the terminal can report the model deployment completion related message to the base station (for example, allocate corresponding report-related UL resources).

[0162] The reason for needing MAT is as follows. The reason is that after decoding, interpreting, validating, and / or testing / compiling the transferred model data (e.g., structure and parameters), the corresponding model should be able to be applied / deployed. The time required for this process will vary depending on the model type described above. For example, when defining the two types of models described below, the MAT required for type 2 is inevitably greater than the MAT for type 1. Therefore, the following method 1 is proposed.

[0163] Type 1: Model parameter transfer

[0164] Type 2: Model structure + parameter transfer

[0165] Method 1

[0166] Multiple model transfer types can be defined. A separate / different MAT can be defined / configured for each model transfer type.

[0167] For example, different MATs may be defined / configured depending on whether a transfer of model structure information is included (whether model structure information is transferred) (e.g., classification into Type 1 and Type 2). That is, MAT1 associated with Type 1 and MAT2 associated with Type 2 may be defined / configured. As a specific example, the MAT (MAT1 or MAT2) may be determined based on the model transfer type (Type 1 / Type 2).

[0168] Here, the model structure information may include one or more of the following: i) model structure information such as RNN, CNN, and autoencoder, ii) the number of hidden layers for the same structure, iii) node number information for each layer, iv) input / output parameter configuration information, and / or v) preprocessing / post-processing related information.

[0169] Model transfers that include model structures (e.g., Type 2) preferably employ a larger MAT with a larger model data size, time spent interpreting and deploying the structure for the corresponding model, larger model verification / testing / compiling time, and / or additional verification / testing processes, etc. As an example, MAT2 associated with Model Transfer Type 2 may be defined / configured to be larger than MAT1 associated with Model Transfer Type 1.

[0170] The required MAT may differ between the initial transfer of a model and its type and the transfer of an updated model after one or more transfers. This is because the first model transfer between a particular device may contain more information about the model than subsequent model transfers. Furthermore, even with the same amount of information, the first model transfer may require more time to verify / test / compile the corresponding model in the receiving model's main body. Therefore, the following is proposed.

[0171] Method 2

[0172] Different model application times (MATs) may be defined / configured according to the number of model transfers. For example, separate / different MATs may be defined / configured according to the first model transfer or the second or subsequent model transfers for a device.

[0173] A terminal / device can simultaneously drive multiple AI / ML models for the same or multiple functions / purposes / use cases. For example, the multiple AI / ML models may include i) a model for CSI prediction, ii) a model for beamforming prediction, and / or iii) a model for CSI compression.

[0174] The transfer of models can also be performed from the same or different base stations at the same time / overlapping time. In this case, model transfer-related processing (e.g., interpretation, verification, compilation, etc.) can be performed sequentially depending on the terminal implementation. Different MATs can be defined / configured according to the corresponding model transfer-related processing.

[0175] Due to lack of memory / processing power, the terminal may not be able to receive the transfer for (some) models. Therefore, the following approach is proposed.

[0176] Method 3

[0177] Transfers for multiple models may occur simultaneously. In this case, different MATs may be defined / configured based on the number of models transferred simultaneously (and the priority of the models). As an example, different MATs may be defined / configured based on the number of models transferred simultaneously and / or the priority of each model transferred (e.g., the priority of each model in the multiple models transferred simultaneously).

[0178] And / or, it is possible to prevent model transfers for a specific number or more models from being received or expected to be received. As an example, the terminal may not receive some models of multiple models transferred simultaneously (based on the priority of the models). As an example, the terminal may not expect a predetermined number or more of models to be transferred simultaneously. The base station may simultaneously send only fewer than a predetermined number of models to the terminal.

[0179] According to Approach 3, a MAT for all models can be added, or a different MAT can be applied to each model.

[0180] For example, when multiple models are transferred simultaneously, a MAT for all models can be added so that the terminal can sequentially receive / deploy the models. For example, when multiple models are transferred simultaneously, a different MAT can be applied to each model by prioritizing which model to receive / deploy from. A small MAT value can be used to define / configure high-priority models and models that are deployed / compiled first. A large MAT value can be used to define / configure low-priority models and models that are deployed / compiled later.

[0181] Hereinafter, embodiments for reducing the implementation burden of a terminal will be described. The following embodiments may be applied in addition to the above embodiments, or may be applied independently of the above embodiments.

[0182] According to an embodiment, when a certain number or more of model transfers occur, a rule may be defined / configured such that the terminal performs model transfers up to a predetermined number and does not receive transfers for the remaining models.

[0183] Depending on the implementation, the terminal may not expect a certain number or more of model transfers to occur.

[0184] Depending on the embodiment, when a certain number or more of model transfers occur, a rule may be defined to not receive transfers for any of these models.

[0185] According to an embodiment, when a certain number or more of model transfers occur, a rule may be specified to not expect to receive transfers for all models.

[0186] The "priority of the model" can be configured / defined by considering one or more of the following factors:

[0187] - Model transfer start time point (for example, when transfer for model 1 starts earlier than transfer for model 2, the MAT for model 2 is defined / configured to be a value greater than the MAT for model 1)

[0188] -Related features (importance / urgency)

[0189] For example, when a transition to multiple models occurs, the model for the function to be deployed / compiled first can be determined according to a predefined priority rule or according to the order configured / indicated by the base station. In this case, the MAT for function 1 and the MAT for function 2 can be configured / defined differently. As an example, even if they have the same function, the model for PCell / PScell ​​can be defined / configured to take precedence over the model for SCell.

[0190] - Model transfer type (e.g., model transfer type based on whether the model transfer type includes model structure transfer)

[0191] - Model transfer number (e.g., whether the model transfer is the initial model transfer)

[0192] - Model ID order (e.g. make lower / higher IDs first)

[0193] -Model priority set by the base station

[0194] Depending on the implementation, the MAT for a model may be defined, configured, or increased in proportion to the number of models being transferred simultaneously. For example, the receiving device may optionally determine which model to deploy first. For example, the MAT may be increased without regard to the model's priority. For example, the MAT may be increased based on the model's priority.

[0195] Depending on the embodiment, a terminal may be specified / defined so as not to receive or not to expect to receive transfers of partial / full models.

[0196] Depending on the terminal implementation, model transfer-related processing for multiple models can be performed simultaneously. For such terminals, as long as the number of models transferred simultaneously does not exceed a certain limit, processing can be performed without applying the relaxed MAT (i.e., increased MAT) as in Method 3. Therefore, in addition to or instead of Method 3, the following method is proposed.

[0197] Method 4

[0198] Model Transfer Units (MTU) or Model Application Units (MAU) may be defined (to allow for devices / terminals that can receive multiple models simultaneously).

[0199] The maximum number of MTUs / MAUs that a terminal / device can receive simultaneously / within a predetermined time (or transfer-related processing is possible) can be defined / set. The same MAT can be applied to transfer operations for models with a number less than the maximum number of MTUs / MAUs.

[0200] As an example, a longer MAT may be applied to model transfers exceeding the maximum number of MTUs / MAUs (similar to method 3 above). Alternatively, for model transfers exceeding the maximum number of MTUs / MAUs, some / all of the transfers may not be performed (depending on priority).

[0201] For each terminal, the maximum number of MTUs / MAUs that the terminal / device can receive simultaneously / within a predetermined time (or transfer-related processing is possible) may be different. In this case, the following embodiments may be considered.

[0202] According to an embodiment, the information on the maximum number of MTUs / MAUs may be transferred to the base station based on the UE capability report. Specifically, the terminal may send UE capability information including the maximum number of MTUs / MAUs to the base station.

[0203] Depending on the implementation, the maximum number of MTUs / MAUs may be defined as a specific value for all terminals or for each terminal type / category. In this case, the maximum number of MTUs / MAUs may be defined for each function / purpose / use case or for all functions / purposes / use cases. As an example, the maximum number of MTUs / MAUs for all functions / purposes / use cases may be defined as 2. As an example, the maximum number of MTUs / MAUs for each function / purpose / use case may be defined as 1.

[0204] In addition to the above method 4, (for a terminal / device supporting a specific number or more of the maximum number of MTUs / MAUs), the following implementations may be considered when there is no currently occupied number of MTUs / MAUs or the number is less than a specific number or more.

[0205] Depending on the implementation, a specific type, a specific ID, a specific number of times, and / or a specific number of simultaneous transmissions (e.g., type 1, ID=0, two or more times, or one simultaneous transmission) may be assumed. This case may be defined / configured to apply a shorter MAT than other cases. The reason is that MTU / MAU related resources (e.g., memory, hardware / software resources responsible for processing power (e.g., DSP, modem, FPGA, etc.)) may be allocated to the model transfer.

[0206] According to an embodiment, in the application of the proposed method (e.g., at least one of methods 1 to 4), a process in which the terminal additionally transfers / reports to the base station whether the model transfer for the model is complete by considering the case where one / more transfers for a specific / all models are not performed can be defined / configured. For example, the terminal may send information indicating whether the model transfer is complete to the base station. As an example, based on not performing the transfer for a specific model / all models, the terminal may send information for an incomplete model transfer to the base station.

[0207] Depending on the implementation, a transfer / reporting process for information related to model transfer may be defined.

[0208] As an example, the terminal may transfer / report the following i) and / or ii) to the base station.

[0209] i) Information on the transfer model for the current terminal / device (e.g., the number of MTUs / MAUs occupied / to be occupied, information on the model being sent / sent / to be sent (e.g., size, type, etc.))

[0210] ii) Hardware / software status information of the terminal / device related to the model transfer (e.g., information on the remaining memory / buffer occupied / to be occupied)

[0211] As an example, the transfer / reporting process can be defined by being included in a model transfer process. As a specific example, the transfer / reporting process can be performed in conjunction with the model transfer process. The transfer / reporting process can be performed before the model transfer process is initiated, after the model transfer process is initiated, and / or during the model transfer process.

[0212] As an example, the transfer / reporting process described above may be defined as being separate from the model transfer process. As a specific example, the transfer / reporting process may be performed based on a time period / time point separate from the model transfer process. As another specific example, the transfer / reporting process may be performed independently, regardless of whether the model transfer process is initiated.

[0213] This report may be a report initiated by the NW via configuration / instruction of the base station or a report initiated by the UE based on a specific event. For example, a specific event may refer to a situation where information related to the transfer model or hardware / software of the terminal / device changes. For example, a specific event may refer to a situation where a value / quantity related to information related to the transfer model or hardware / software of the terminal / device is equal to or greater than / equal to or less than a specific threshold.

[0214] In the application of the proposed methods (e.g., method 1, method 2, method 3, and method 4), the minimum required application time for each type / number / number of simultaneous transmissions may be different for each device. For example, in the case of the MAT of the terminal, a process of transmitting the minimum required MAT for each type / number / number of simultaneous transmissions of the terminal to the base station as a UE capability report may be performed. The base station may set a related value (e.g., an application time value for each type / number / number of simultaneous transmissions) based on the minimum required MAT for each type / number / number of simultaneous transmissions of the terminal, and the related value may be set in the terminal.

[0215] In the present disclosure, model data / information (e.g., structure, parameters) used for model transfer can be defined as control plane information or user plane information. In addition, model data / information (e.g., structure and parameters) used for model transfer can be defined as physical layer (L1) / MAC layer (L2) / RRC layer (L3) messages.

[0216] In the above embodiments, the term "model application time (MAT)" is used for convenience of description, but is not intended to limit the technical concept of the above embodiments to this term. For example, "MAT" can be replaced by time associated with model transfer, minimum time associated with model transfer, minimum time for a model, minimum time for application of a model, processing time associated with a model, minimum processing time associated with a model, time for compiling a model, minimum time for compiling a model, or time associated with compiling a model.

[0217] The proposed methods (e.g., method 1, method 2, method 3, method 4) can be used / applied together with one or more methods. As an example, operations related to model transfer can be performed by a terminal / base station based on at least one of method 1, method 2, method 3, and / or method 4. For example, operations related to model transfer can be performed by a terminal / base station based on a combination of two or more of method 1, method 2, method 3, and / or method 4.

[0218] In terms of implementation, the operation of the base station / terminal according to the above embodiment (eg, the operation based on at least one of methods 1 to 4) can be performed by the following method. Figure 5 Devices in (e.g. Figure 5 Processed by processors 110 and 210).

[0219] In addition, the operation of the base station / terminal according to the above embodiment (eg, the operation based on at least one of methods 1 to 4) can be used to drive at least one processor (eg, Figure 5 110 and 210) in the form of instructions / programs (eg, instructions or executable code) stored in a memory (eg, Figure 5 140 and 240).

[0220] Hereinafter, a signaling process based on the above-mentioned embodiment will be described.

[0221] Figure 2 A signaling process according to an embodiment of the present disclosure is illustrated.

[0222] Figure 2 An example of signaling between a user equipment (UE) and a network (NW) based on the above-mentioned proposed method is illustrated. Here, UE / NW are merely examples and can be replaced and applied with various devices. Figure 2 It is only for the convenience of description and does not limit the scope of the present disclosure. In addition, depending on the situation and / or configuration, it may be omitted. Figure 2 Some steps are shown.

[0223] belong Figure 2 The base station of the NW in may correspond to any entity, such as a base station (BS), a node B, a TRP, etc., and the UE may also be replaced with an entity / server that is responsible for UE-related AI / ML operations instead of the UE.

[0224] The UE may perform a process S205 of reporting the terminal capability value for the AI / ML model and / or features / functions related thereto to the base station. In this case, the terminal capability value including the reported terminal capability value related to MAT / MAU / MTU may be reported based on the proposed methods (e.g., Method 1, Method 2, Method 3, and Method 4).

[0225] The base station may perform configuration (S210) for related functions / parameters and / or AI / ML models based on the terminal capability value reported in S205. In this process, configuration values ​​related to MAT / MAU / MTU may be included based on the proposed methods (e.g., Method 1, Method 2, Method 3, and Method 4).

[0226] Then, the model transfer process between the base station and the UE can be performed. The detailed processing of the model transfer process can be diversified, and Figure 2 An embodiment of the process is illustrated. A process of initiating a model transfer from the base station to the UE or from the UE to the base station may be performed (S215). In addition, this process may be omitted. Then, a transfer of the model (data / information) from the base station to the UE may be performed (S220). Although not shown in this example, a process of the UE responding to the model transfer initiation message of the base station and / or a process of the UE responding to the model transfer of the base station may be included in the model transfer process.

[0227] After the model transfer, the UE can assume that the deployment / activation / application of the model is completed after the MAT time point defined / configured based on the proposed methods (e.g., Method 1, Method 2, Method 3, and Method 4) of the present disclosure, and after this time point, the base station can perform the process of activating / triggering functions (e.g., CSI reporting, beam reporting, and positioning-related signaling) to the UE (S225).

[0228] As mentioned above, the above NW / UE signaling and operations can be performed by the following (in Figure 5 For example, the NW (or base station) may correspond to a first wireless device, and the UE may correspond to a second wireless device, and in some cases, the opposite case may also be considered.

[0229] For example, the above NW / UE signaling and operations can be performed by Figure 5 One or more processors 110 and 210 are processed, and the above-mentioned NW / UE signaling and operations can be used to drive Figure 5 The instructions / programs (eg, instructions, executable codes) of at least one processor 110 or 210 are stored in the memories 140 and 240 .

[0230] In the following, reference will be made to Figure 3 and Figure 4The above-mentioned embodiments are described in detail from the perspective of the operation of the first device and the second device. The methods to be described below are distinguished only for the convenience of description, and it goes without saying that some components of any one method can be replaced by some components of another method, or can be applied in combination with each other. In the following, the first device may refer to the subject that receives the model in the process related to the model transfer, and the second device may refer to the subject that transfers the model in the process related to the model transfer. The first device may be a UE, a base station or a first UE, and the second device may be a base station, a UE or a second UE.

[0231] As an example, a base station may transfer a model (and / or information specific to the model) to a UE. As an example, a UE may transfer a model (and / or information specific to the model) to a base station. As an example, a second UE may transfer a model (and / or information specific to the model) to a first UE.

[0232] Figure 3 is a flowchart for describing a method performed by a first device according to an embodiment of the present disclosure.

[0233] Reference Figure 3 , a method performed by a first device according to an embodiment of the present disclosure includes receiving information for at least one model ( S310 ) and determining a model application time point ( S320 ).

[0234] In S310 , the first device receives information regarding at least one model from the second device.

[0235] The information about the at least one model may be based on the above-mentioned information related to model transfer.

[0236] In S320 , the first device determines a model application time (MAT) associated with at least one model.

[0237] A point in time at which an operation / function based on at least one model may be triggered / initiated / executed may be determined by the MAT.

[0238] As an example, the first device may expect to receive information related to a trigger / indication of an operation / function based on at least one model from the second device after the MAT.

[0239] MAT can refer to the time associated with model application / activation / preparation. This will be described in detail below.

[0240] According to an embodiment, MAT is time related to time from a reference time point (e.g., a first time point) to a time point (e.g., a second time point) when the first device completes verification, testing, application, deployment, or compilation of at least one model.

[0241] As an example, MAT may be defined / configured / indicated as being equal to or greater than a time from a first time point to a second time point. As an example, MAT may be defined / configured / indicated as a time including a time from a first time point to a second time point.

[0242] As an example, the reference time point may be i) a time point at which the model transfer process is triggered or initiated by the second device, ii) a time point at which the first device sends a response to the triggering of the model transfer process, iii) a time point at which the transmission of information is initiated, iv) a time point at which the transmission of information is completed, or v) a time point at which the first device sends a response to the receipt of information.

[0243] According to an embodiment, the MAT may be the time required for the first device to activate, apply, or prepare the model or a function related to the model.

[0244] The MAT may be determined based on at least one of the above-described methods 1 to 4.

[0245] Specifically, the MAT may be determined based on at least one of: i) a model transfer type, ii) a model transfer number, iii) a number of at least one model, and / or iv) a model transfer unit.

[0246] As an example, the model transfer type may be defined based on whether the information includes information on a structure related to at least one model.

[0247] Based on the model transfer type being the first type, the information may include information on parameters associated with at least one model.

[0248] The model-based transfer type is the second type, and the information may include i) information on parameters and ii) information on structures.

[0249] The MAT associated with the second type may be greater than the MAT associated with the first type. This embodiment may be based on method 1.

[0250] As an example, the number of model transitions may be related to whether the transition of at least one model based on the information is an initial transition. The MAT associated with the initial transition may be greater than the MAT associated with two or more model transitions. This embodiment may be based on method 2.

[0251] As an example, if the number of at least one model is greater than a defined number, the MAT may be determined based on a priority associated with each model. This embodiment may be based on method 3. The priority may be defined based on at least one of the following: i) a start time point of a model transfer associated with each model, ii) a function associated with each model, iii) a model transfer type, iv) a number of model transfers, and / or v) an ID of each model.

[0252] As an example, a model transfer unit may be related to the number of models that can be simultaneously transferred to the first device. A MAT may be determined as a first MAT based on the number of at least one model being less than or equal to the model transfer unit. A MAT may be determined as a second MAT based on the number of at least one model being greater than the model transfer unit. The second MAT may be greater than the first MAT. This embodiment may be based on Method 4.

[0253] The MAT may be defined / configured / determined / indicated based on the capability of the first device. Implementations related thereto will be described in detail.

[0254] According to an embodiment, the method may further include a minimum MAT reporting step and a MAT receiving step, which will be described in detail below.

[0255] In the minimum MAT reporting step, the first device reports information on the minimum MAT supported by the first device to the second device. The minimum MAT reporting step may be performed before S310.

[0256] As an example, information for the minimum MAT may be reported based on an existing UE capability transfer procedure. Specifically, the first device may send capability information including information for the minimum MAT supported by the first device to the second device.

[0257] In the MAT receiving step, the first device receives information related to the MAT from the second device. The MAT receiving step may be performed before S310 or S320. The MAT may be greater than or equal to the minimum MAT.

[0258] As an example, the first device may determine the MAT based on information related to the MAT. Specifically, the MAT determined based on at least one of the following may be configured / indicated by the information related to the MAT: i) a model transfer type, ii) a number of model transfers, iii) a number of at least one model, and / or iv) a model transfer unit may be configured / indicated by the information related to the MAT.

[0259] As an example, one or more MATs may be configured / indicated based on information related to the MAT. The first device may determine the MAT among the one or more MATs.

[0260] That is, the MAT among the one or more MATs may be determined based on at least one of: i) model transfer type, ii) model transfer number, iii) number of at least one model, and / or iv) model transfer unit.

[0261] Specific examples of one or more MATs are as follows.

[0262] 1) MAT for each model transfer type

[0263] 2) MAT for each number of models transferred simultaneously (e.g., MAT when the number of at least one model is less than or equal to the model transfer unit, MAT when the number of at least one model is greater than the model transfer unit, etc.)

[0264] 3) MAT based on the model transfer type and the number of models transferred simultaneously (e.g., MAT according to the number of models transferred simultaneously in type 1 (or type 2) model transfer)

[0265] 4) MAT based on the type of model transferred relative to the initial transfer and / or the number of models transferred simultaneously

[0266] 5) MAT based on the model transfer type associated with two or more transfers (e.g., model updates) and / or the number of models transferred simultaneously

[0267] One or more MATs may be based on at least one of the above examples 1) to 5). The above operations based on S310 and 320, the minimum MAT reporting step and the MAT receiving step may be performed by Figure 5 For example, the first device 100 / 200 may control one or more transceivers 130 / 230 and / or one or more memories 140 / 240 to perform operations based on S310, S320, the minimum MAT reporting step, and the MAT receiving step.

[0268] Hereinafter, the above-mentioned embodiment will be described in detail from the perspective of the operation of the second device.

[0269] S410, S420, the minimum MAT receiving step and the MAT sending step described below correspond to the reference Figure 3 The S310, S320, minimum MAT reporting step and MAT receiving step described in the preceding text are omitted by taking into account the corresponding relationship. That is, the detailed description of the operation of the second device described below can be replaced by the corresponding operation. Figure 3 The description / implementation method is replaced by .

[0270] Figure 4 is a flowchart for describing a method performed by a second device according to another embodiment of the present disclosure.

[0271] Reference Figure 4 , a method performed by a second device according to another embodiment of the present disclosure includes sending information for at least one model ( S410 ) and determining a model application time point ( S420 ).

[0272] In S410 , the second device sends information for at least one model to the first device.

[0273] In S420, the second device determines a model application time (MAT) associated with at least one model.

[0274] A point in time at which an operation / function based on at least one model may be triggered / initiated / executed may be determined by the MAT.

[0275] As an example, the second device may send information related to triggering / indication of an operation / function based on at least one model to the first device after the MAT.

[0276] The method may further include a minimum MAT receiving step and a MAT sending step.

[0277] In the minimum MAT receiving step, the second device receives information on the minimum MAT supported by the first device from the first device. The minimum MAT receiving step may be performed after S410.

[0278] In the MAT sending step, the second device sends information related to the MAT to the first device. The MAT sending step can be performed before S410 or S420.

[0279] The above operations based on S410 and S420, the minimum MAT receiving step and the MAT sending step can be performed by Figure 5 For example, the second device 100 / 200 may control one or more transceivers 130 / 230 and / or one or more memories 140 / 240 to perform operations based on S410, S420, the minimum MAT receiving step, and the MAT sending step.

[0280] Refer to the following Figure 5 The following describes devices to which embodiments of the present disclosure are applicable (devices that implement methods / operations according to embodiments of the present disclosure).

[0281] Figure 5 The configurations of the first device and the second device according to the embodiment of the present disclosure are illustrated.

[0282] The first device 100 may include a processor 110 , an antenna unit 120 , a transceiver 130 , and a memory 140 .

[0283] The processor 110 can perform signal processing related to the baseband, and includes a high-level processing unit 111 and a physical layer processing unit 115. The high-level processing unit 111 can process operations of the MAC layer, the RRC layer, or a higher layer. The physical layer processing unit 115 can process operations of the PHY layer. For example, if the first device 100 is a base station (BS) device in BS-UE communication, the physical layer processing unit 115 can perform uplink received signal processing, downlink transmitted signal processing, etc. For example, if the first device 100 is a first UE device in inter-UE communication, the physical layer processing unit 115 can perform downlink received signal processing, uplink transmitted signal processing, sidelink transmitted signal processing, etc. In addition to performing signal processing related to the baseband, the processor 110 can also control the overall operation of the first device 100.

[0284] The antenna unit 120 may include one or more physical antennas, and if the antenna unit 120 includes multiple antennas, MIMO transmission / reception is supported. The transceiver 130 may include a radio frequency (RF) transmitter and an RF receiver. The memory 140 may store information processed by the processor 110 and software, an operating system, and applications related to the operation of the first device 100. The memory 140 may also include components such as a buffer.

[0285] In the embodiments described in the present disclosure, the processor 110 of the first device 100 may be configured to implement operations of a BS in BS-UE communication (or operations of a first UE device in inter-UE communication).

[0286] The second device 200 may include a processor 210 , an antenna unit 220 , a transceiver 230 , and a memory 240 .

[0287] The processor 210 can perform signal processing related to the baseband and includes a high-level processing unit 211 and a physical layer processing unit 215. The high-level processing unit 211 can process operations of the MAC layer, the RRC layer, or higher layers. The physical layer processing unit 215 can process operations of the PHY layer. For example, if the second device 200 is a UE device in BS-UE communication, the physical layer processing unit 215 can perform downlink received signal processing, uplink transmitted signal processing, etc. For example, if the second device 200 is a second UE device in inter-UE communication, the physical layer processing unit 215 can perform downlink received signal processing, uplink transmitted signal processing, sidelink received signal processing, etc. In addition to performing signal processing related to the baseband, the processor 210 can also control the overall operation of the second device 210.

[0288] The antenna unit 220 may include one or more physical antennas, and if the antenna unit 220 includes multiple antennas, MIMO transmission / reception is supported. The transceiver 230 may include an RF transmitter and an RF receiver. The memory 240 may store information processed by the processor 210 and software, an operating system, and applications related to the operation of the second device 200. The memory 240 may also include components such as a buffer.

[0289] In the embodiments described in the present disclosure, the processor 210 of the second device 200 may be configured to implement operations of a UE in BS-UE communication (or operations of a second UE device in inter-UE communication).

[0290] The description of the BS and UE in BS-UE communication (or the first UE device and the second UE device in inter-UE communication) in the examples of the present disclosure can be equally applied to the operations of the first device 100 and the second device 200, and redundant description is omitted.

[0291] In addition to LTE, NR, and 6G, the wireless communication technology implemented in the apparatus 100 and the apparatus 200 according to the present disclosure may also include narrowband Internet of Things (NB-IoT) for low-power communication. For example, NB-IoT technology may be an example of low-power wide area network (LPWAN) technology and may be implemented in standards such as LTE Cat NB1 and / or LTE Cat NB2. NB-IoT technology is not limited to the above names.

[0292] Additionally or alternatively, the wireless communication technology implemented in the apparatus 100 and the apparatus 200 according to the present disclosure may perform communication based on the LTE-M technology. For example, the LTE-M technology may be an example of an LPWAN technology and may be referred to by various names, such as enhanced machine type communication (eMTC). For example, the LTE-M technology may be implemented using at least one of various standards, such as 1) LTE CAT0, 2) LTE Cat M1, 3) LTE Cat M2, 4) LTE non-BL (non-bandwidth limited), 5) LTE-MTC, 6) LTE machine type communication, and / or 7) LTE M. The LTE-M technology is not limited to the above names.

[0293] Additionally or alternatively, considering low-power communication, the wireless communication technology implemented in the apparatus 100 and the apparatus 200 according to the present disclosure may include at least one of ZigBee, Bluetooth, and a low-power wide area network (LPWAN), and is not limited to the above names. For example, ZigBee technology can create a personal area network (PAN) related to small / low-power digital communication based on various standards such as IEEE 802.15.4, and can be referred to by various names.

Claims

1. A method performed by a first device in a wireless communication system, the method comprising the following steps: receiving information for at least one model from a second device; as well as determining a model application time MAT associated with the at least one model, The MAT is determined based on at least one of: i) a model transfer type, ii) a number of model transfers, iii) a quantity of the at least one model, and / or iv) a model transfer unit.

2. The method according to claim 1, wherein The MAT is a time related to a time from a reference time point to a time point when verification, testing, application, deployment, or compilation of one or more models is completed by the first device.

3. The method according to claim 2, wherein: The reference time point is i) the time point at which the model transfer process is triggered or initiated by the second device, ii) the time point at which the first device sends a response to the triggering of the model transfer process, iii) the time point at which the transmission of the information is initiated, iv) the time point at which the transmission of the information is completed, or v) the time point at which the first device sends a response to the reception of the information.

4. The method according to claim 1, wherein The MAT is the time required for the first device to activate, apply or prepare a model or a function related to the model.

5. The method according to claim 1, further comprising the steps of: reporting information about a minimum MAT supported by the first device to the second device; as well as Information related to the MAT is received from the second device.

6. The method according to claim 5, wherein: The MAT is greater than or equal to the minimum MAT.

7. The method according to claim 1, wherein The model transfer type is defined based on whether the information includes information for a structure associated with one or more models.

8. The method according to claim 7, wherein: Based on the model transfer type being the first type, the information includes information on parameters associated with the one or more models, and Wherein, based on the model transfer type being the second type, the information includes i) information on the parameters and ii) information on the structure.

9. The method according to claim 8, wherein The MAT associated with the second type is greater than the MAT associated with the first type.

10. The method according to claim 1, wherein The model transfer number is related to whether the transfer of the at least one model based on the information is an initial transfer.

11. The method according to claim 10, wherein: The MAT associated with the initial transition is greater than the MAT associated with two or more of the model transitions.

12. The method according to claim 1, wherein The MAT is determined based on a priority associated with each model based on the number of the at least one model being greater than a defined number.

13. The method according to claim 12, wherein: The priority is defined based on at least one of: i) a start time point of a model transfer associated with each model, ii) a function associated with each model, iii) the model transfer type, iv) the number of model transfers, and / or v) an ID of each model.

14. The method according to claim 1, wherein The model transfer unit is related to the number of models that can be simultaneously transferred to the first device.

15. The method according to claim 14, wherein determining the MAT as a first MAT based on the number of the at least one model being less than or equal to the model transfer unit, wherein the MAT is determined as a second MAT based on the number of the at least one model being greater than the model transfer unit, and The second MAT is greater than the first MAT.

16. A first device operating in a wireless communication system, the first device comprising: one or more transceivers; one or more processors; as well as one or more memories connected to the one or more processors and storing instructions, The instructions, upon being executed by the one or more processors, configure the one or more processors to perform all the steps of the method according to any one of claims 1 to 15.

17. A device comprising: one or more memories; as well as one or more processors operatively connected to the one or more memories, The one or more memories store instructions that, upon being executed by the one or more processors, configure the one or more processors to perform all the steps of the method according to any one of claims 1 to 15.

18. One or more non-transitory computer-readable media storing instructions, in, The instructions executable by one or more processors configure the one or more processors to perform all the steps of the method according to any one of claims 1 to 15.

19. A method performed by a second device in a wireless communication system, the method comprising the following steps: sending information related to at least one model to the first device; as well as determining a model application time MAT associated with the at least one model, The MAT is determined based on at least one of: i) a model transfer type, ii) a number of model transfers, iii) a quantity of the at least one model, and / or iv) a model transfer unit.

20. A second device operating in a wireless communication system, the second device comprising: one or more transceivers; one or more processors; as well as one or more memories connected to the one or more processors and storing instructions, The instructions, upon being executed by the one or more processors, configure the one or more processors to perform all the steps of the method according to claim 19.

Images