Communication method and device based on federated learning

Abstract

A communication method and device based on federated learning are described. A communication method of a user equipment (UE) therefor comprises: receiving, from a network, configuration information for a UE-side artificial intelligence (AI) / machine learning (ML) model in which federated learning is performed by the network and the UE; and providing a UE-side parameter for a network-side AI / ML model corresponding to the UE-side AI / ML model, wherein the providing of the UE-side parameter does not involve providing training data of the UE-side AI / ML model, and the configuration information includes configuration information based on parameters collected from a plurality of UEs.

Description

Communication method and device based on federated learning

[0001] The following description relates to a mobile communication system utilizing an AI (Artificial Intelligence) / ML (Machine Learning) model, specifically to a federated learning-based communication method and apparatus for multiple user devices (UEs).

[0002] Various technologies such as LTE, LTE-Advanced, and WiFi are used in wireless communication systems, and 5G is also included here.

[0003] Figure 1 shows the structure of a system for 5G communication.

[0004] Referring to FIG. 1, the NG-RAN (Next Generation - Radio Access Network) may include a base station (20) that provides user plane and control plane protocol termination to the UE (10). For example, the base station (20) may include a gNB (next generation-Node B) and / or an eNB (evolved-Node B). For example, the UE (10) may be fixed or mobile and may be referred to by other terms such as terminal, MS (Mobile Station), UT (User Terminal), SS (Subscriber Station), MT (Mobile Terminal), or Wireless Device. For example, the base station may be a fixed station communicating with the UE (10) and may be referred to by other terms such as BTS (Base Transceiver System) or Access Point.

[0005] The example in FIG. 1 illustrates a case including only gNB. Base stations (20) can be connected to each other via Xn interfaces. Base stations (20) can be connected to a 5th generation core network (5G Core Network: 5GC) via NG interfaces. More specifically, base stations (20) can be connected to an access and mobility management function (AMF) (30) via an NG-C interface and to a user plane function (UPF) (30) via an NG-U interface.

[0006]

[0007] Meanwhile, starting from 5G Release 19, communication methods incorporating AI / ML are being discussed as a Work Item (WI).

[0008] Specifically, regarding the physical layer, beam management, positioning improvements, and Channel Status Information (CSI) enhancements are being discussed; however, there is still a lack of discussion regarding specific combined learning methods, such as federated learning, and the resulting signaling.

[0009] In order to solve the problem described above, one aspect of the present invention proposes a method and apparatus for a plurality of UEs to communicate based on federated learning with a network.

[0010] According to an embodiment, federated learning between a plurality of UEs and a network may be a multi-user beam management situation for a plurality of UEs, and in such a situation, we propose a signaling and operation method for federated learning between a UE-side AI / ML model and a network-side AI / ML model.

[0011] According to an embodiment, we propose configuration information for the collaborative learning of UE-side AI / ML models deployed on a plurality of UEs, as well as signaling and operation methods for their Life Cycle Management (LCM).

[0012] The technology proposed below is assumed to be applicable not only to current 5G systems but also to 6G and subsequent mobile communication systems; therefore, although the term 'AI / ML model' used in 5G may be referred to by other terms such as 'Function' or 'Agent' that perform specific functions, for the convenience of the following explanation, 5G terminology will be used.

[0013] The problems to be solved by the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art to which the present invention belongs from the description below.

[0014] In one aspect of the present invention for solving the problem described above, a communication method is proposed in which a user device (UE) in a mobile communication system communicates with a network, comprising: receiving configuration information for a UE-side AI (Artificial Intelligence) / ML (Machine Learning) model from the network, wherein federated learning is performed by the network and the UE; and providing UE-side parameters for a network-side AI / ML model corresponding to the UE-side AI / ML model, wherein providing the UE-side parameters does not provide training data for the UE-side AI / ML model, and the configuration information includes configuration information based on parameters collected from a plurality of UEs.

[0015] The above configuration information can be received through UE-specific RRC (Radio Resource Control) signaling or group-specific RRC signaling including the plurality of UEs.

[0016] In addition, the above plurality of UEs can correspond to a UE group for multi-user beam management.

[0017] Meanwhile, the UE group for the above-mentioned multi-user beam management can be configured based on one or more of the following: Channel Status Information (CSI) based, spatial separation based, Quality of Service (QoS) based, Signal-to-Noise Ratio (SNR) and Modulation and Coding Scheme (MCS) based, dynamic grouping based, or Coordinated Multipoint (CoMP) based.

[0018] Additionally, receiving the above configuration information may include transmitting model structure information that each of the plurality of UEs, including the UE, can support to the network; and receiving information from the network to configure a UE group for multi-user beam management based on the model structure information that the plurality of UEs can support.

[0019] In addition, the above-mentioned UE-side AI / ML model can be trained through federated learning by the above-mentioned network and a plurality of UEs in the UE group for multi-user beam management.

[0020] Additionally, the above setting information may include setting information for a CSI (Channel Status Information) report, and the setting information for the CSI report may include one or more of information for a first type beam set (Set B) or information for a second type beam set (Set A).

[0021] At this time, if the setting information for the above CSI report includes only information for the second type beam set (Set A), the UE can transmit the UE-side parameters based on the fact that the measurement resources for the first type beam set (Set B) exist within the range of measurement resources for the second type beam set (Set A).

[0022] Meanwhile, if the setting information for the above CSI report includes information for the first type beam set (Set B) and information for the second type beam set (Set A), the UE can transmit UE-side parameters based on the measurement resources corresponding to the information for the first type beam set (Set B) and the measurement resources corresponding to the information for the second type beam set (Set A).

[0023] Additionally, the above UE-side parameters may include one or more of the neural network's weights, biases, split points, or gradients.

[0024] In addition, the above configuration information may additionally include configuration information for the Life Cycle Management (LCM) of the UE-side AI / ML model; and may additionally include reporting to the network if the UE-side AI / ML model becomes unusable according to the LCM configuration information.

[0025] In addition, update, retraining, or disable instruction information configured in consideration of reports of unusable UE-side AI / ML models received from each of the plurality of UEs from the above network may also be additionally received.

[0026] In addition, the above-mentioned UE-side AI / ML model may include an AI / ML model deployed within the UE and an AI / ML model deployed on a UE vendor server.

[0027] Meanwhile, in another aspect of the present invention, a communication method is proposed in which a network in a mobile communication system communicates with a plurality of user devices (UEs), wherein the network transmits configuration information for UE-side AI (Artificial Intelligence) / ML (Machine Learning) models, which undergo federated learning by the network and the plurality of UEs, to the plurality of UEs; and receives UE-side parameters for a network-side AI / ML model corresponding to the UE-side AI / ML model from the plurality of UEs, wherein receiving the UE-side parameters does not involve receiving training data for the UE-side AI / ML model, and the configuration information includes configuration information based on parameters collected from the plurality of UEs.

[0028] Meanwhile, in another aspect of the present invention, a user device (UE) that communicates with a network in a mobile communication system comprises: at least one processor; and at least one computer memory that can be operably connected to the at least one processor and stores instructions that cause the at least one processor to perform operations when executed, wherein the operations include receiving configuration information for a UE-side AI (Artificial Intelligence) / ML (Machine Learning) model in which federated learning is performed by the network and the UE; and providing UE-side parameters for a network-side AI / ML model corresponding to the UE-side AI / ML model, wherein providing the UE-side parameters does not provide training data for the UE-side AI / ML model, and the configuration information includes configuration information based on parameters collected from a plurality of UEs.

[0029] Meanwhile, in another aspect of the present invention, a network for communicating with a plurality of user devices (UEs) in a mobile communication system comprises: at least one processor; and at least one computer memory that can be operably connected to the at least one processor and stores instructions that cause the at least one processor to perform operations when executed, wherein the operations include transmitting configuration information for UE-side AI (Artificial Intelligence) / ML (Machine Learning) models, which undergo federated learning by the network and the plurality of UEs, to the plurality of UEs; and receiving UE-side parameters for a network-side AI / ML model corresponding to the UE-side AI / ML model from the plurality of UEs, wherein receiving the UE-side parameters does not receive training data for the UE-side AI / ML model, and the configuration information includes configuration information based on parameters collected from the plurality of UEs.

[0030] According to the embodiments of the present invention as described above, a plurality of UEs can efficiently communicate with a network based on federated learning.

[0031] According to an embodiment, federated learning between a plurality of UEs and a network may be a multi-user beam management situation for a plurality of UEs, and in this situation, communication efficiency can be increased by predicting a beam set with excellent transmission performance through federated learning between a UE-side AI / ML model and a network-side AI / ML model.

[0032] In addition, according to the embodiment, a multi-user beam management method utilizing network-UE integrated learning enables users to efficiently utilize resources while reducing interference that may occur in the same resources, thereby enabling high-speed data transmission.

[0033] In addition, according to the embodiment, configuration information for federated learning of UE-side AI / ML models deployed on a plurality of UEs and signaling for their LCM are implemented to establish an efficient federated learning environment between a network and multiple UEs.

[0034] The effects obtainable from the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art from the description below.

[0035] Figure 1 shows the structure of a system for 5G communication.

[0036] FIG. 2 is a diagram illustrating a method for a plurality of UEs to communicate with a network based on federated learning according to an embodiment of the present invention.

[0037] FIG. 3 is a diagram illustrating a method for performing multi-user beam management using federated learning according to an embodiment of the present invention.

[0038] FIGS. 4 and 5 are drawings for explaining the concept of associative learning according to one embodiment of the present invention, distinguishing it from other learning methods.

[0039] FIG. 6 is a diagram illustrating an example in which a UE-side model and a network-side model form a two-sided model according to an embodiment of the present invention.

[0040] FIGS. 7 and 8 are drawings for explaining the concept that setting information according to one embodiment of the present invention represents setting information for a CSI report.

[0041] FIGS. 9 and 10 are drawings for explaining a method in which a UE operates a first type beam set according to network configuration information according to an embodiment of the present invention.

[0042] FIG. 11 illustrates a wireless device that can be applied to the present technology.

[0043] Hereinafter, embodiments of the present invention are described in detail with reference to the attached drawings so that those skilled in the art can easily implement the invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals.

[0044] Throughout the specification, when a part is described as "including" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.

[0045]

[0046] Network - Federated learning among multiple UEs

[0047] FIG. 2 is a diagram illustrating a method for a plurality of UEs to communicate with a network based on federated learning according to an embodiment of the present invention.

[0048] Among a plurality of UEs (110a - 110k) according to one embodiment of the present invention, a specific UE (110) can receive (S110a - S110k) configuration information for a UE-side AI / ML model in which federated learning is performed by the network (NW) and the UE (110) from the network (120).

[0049] In the following description, it is assumed that the learning methods of AI / ML models can be classified into centralized learning, distributed learning, and federated learning, depending on the batch relationships and the location where learning is performed.

[0050] Specifically, federated learning assumes a method in which data is trained in a distributed manner on each user (client) device, while only the parameters of the trained model are shared with the server. This approach is characterized by its ability to improve the model while protecting data privacy, and examples in mobile communication scenarios will be explained in more detail below.

[0051] Additionally, configuration information received from the network (120) may be received through upper-layer signaling such as RRC (Radio Resource Control) signaling, or through lower-layer signaling such as MAC-CE (Medium Access Control - Control Element) or DCI (Downlink Control Information). However, in a preferred embodiment of the present invention, configuration information of the initial network (120) is received through upper-layer signaling such as RRC signaling, and it is proposed to use lower-layer signaling such as MAC-CE / DCI when changing the settings of the AI / ML model of a specific UE (110) as part of LCM operation, etc.

[0052] Generally, RRC signaling is performed in a UE-specific or cell-specific manner. In one embodiment of the present invention, since it is for federated learning of AI / ML models of a plurality of UEs (110a - 110k), it is proposed to use UE-specific RRC signaling for the UEs (110a - 110k) that are the subject of federated learning, and in some cases, group-specific RRC signaling for a group of UEs including these UEs (110a - 110k) may also be used.

[0053]

[0054] In FIG. 2, the UE (110) can train a UE-side AI / ML model (S120a - S120k) based on this configuration information and provide UE-side parameters for a network-side AI / ML model corresponding to the UE-side AI / ML model to the network (120) (S130a - S130l). At this time, regarding the provision of UE-side parameters, it is proposed that training data for the UE-side AI / ML model is not provided, as explained in relation to the concept of federated learning described above. However, if necessary, the inference results of the UE-side AI / ML model and / or other additional UE-side data may be transmitted to the network (120), but the data itself used for training the UE-side AI / ML model is not provided to the network (120). Through this, the amount of data transmitted to the network (120) can be reduced, while enabling the network-side AI / ML model to efficiently perform federated learning with multiple UE-side AI / ML models.

[0055] Additionally, in one embodiment of the present invention, the UE-side AI / ML model may include not only the AI / ML deployed within the UE but also the AI / ML model deployed on the UE vendor server. That is, some of the UE-side AI / ML models may be deployed on the UE vendor server rather than the UE to be trained (S120a - S120k) and / or may support the training (S120a - S120k) of the AI / ML model deployed within the UE.

[0056] In federated learning, AI / ML models located at the UE vendor and AI / ML models located within the UE can exchange training data to enable combined learning, which can be distinguished from federated learning between UE-side AI / ML models and network-side AI / ML models.

[0057] Accordingly, unless explicitly stated otherwise in this specification, the 'UE-side AI / ML model' is assumed to include the concept of an AI / ML model deployed within the UE as described above and an AI / ML model deployed on a UE vendor server.

[0058]

[0059] Meanwhile, the configuration information transmitted by the network (120) to the plurality of UEs (110a-110k) may include configuration information based on parameters collected from the plurality of UEs (110a-110k) (S140a - S140k).

[0060] For example, the UE (110) receiving configuration information (S140a-S140k) from the network (120) may be performed by transmitting model structure information (or performance information) that can be supported by each of the multiple UEs (110a-110k) to the network (120) in steps S130a-S130k, setting up target UEs (110a-110k) for federated learning based on the model structure information (or performance information) that can be supported by the multiple UEs (110a-110k) from the network (120), and providing configuration information for federated learning of their UE-side AI / ML models.

[0061] In FIG. 2, steps S110a - S110k and steps S140a - S140k are shown separately, but in a continuous operation between the network (120) and the UEs (110a - 110k), the steps may be corresponding steps.

[0062]

[0063] Multi-user Beam Management

[0064] Below, an example of performing multi-user (MU) beam management through federated learning between a network and multiple UEs is described based on the above description.

[0065] FIG. 3 is a diagram illustrating a method for performing multi-user beam management using federated learning according to an embodiment of the present invention.

[0066] As illustrated in FIG. 3, it is proposed that a plurality of UEs each include a UE-side model (210a - 210k), and that each UE-side model (210a - 210k) can derive a model output using a dataset for learning and model parameters. Specifically, in FIG. 3, it is proposed that each UE-side AI / ML model (210a - 210k) is learned through federated learning by a plurality of UEs (110a - 110k) of a UE group for network and multi-user (MU) beam management.

[0067] Meanwhile, the network (120) may include a network-side model (220) corresponding to the UE-side models (210a-210k) as illustrated in FIG. 3. The network-side model (220) can perform federated learning by receiving parameters of a plurality of UE-side models (210a-210k) as described above in relation to FIG. 2, and FIG. 3 illustrates an example in which the top K beams (Top-K beams) (230) for each UE are predicted and provided to the UEs (110a-110k). Additionally, the network-side model (220) may derive configuration information (240) for a 'first type beam set (Set B)' that the UEs (110a-110k) must measure for beam management, and transmit this to the UEs as configuration information. The concept of a beam set will be explained in more detail below.

[0068]

[0069] FIGS. 4 and 5 are drawings for explaining the concept of associative learning according to one embodiment of the present invention, distinguishing it from other learning methods.

[0070] As described above in relation to FIG. 2, when UEs (110a-110k) transmit parameters to the network (120), it is preferable not to transmit the training data of the UE-side models (210a-210k).

[0071] The UE-side parameters transmitted at this time may include one or more of neural network / model weights, biases, split points, or gradients. These parameters may be defined differently depending on the type of model.

[0072] For example, linear regression and logistic regression models can include weights and biases as parameters.

[0073] Support Vector Machines (SVMs) include weights and biases, and XGBoost models may include weights.

[0074] Decision tree models include split points as parameters, and random forests can include the split points of each decision tree as parameters.

[0075] Neural network models include weights and biases, and convolutional neural networks may include convolutional weights.

[0076] Recurrent neural network (RNN) and Long Short-Term Memory (LSTM) structures may include weights, including gate-specific weights, as parameters.

[0077] The gradient in the model described above can be obtained by calculating the gradient for each weight and bias through the backpropagation of the loss function. The change in the parameters is derived by multiplying the calculated gradient by the learning rate, and the parameters can be updated by reflecting this change.

[0078] Meanwhile, in distributed learning or federated learning environments, the method of sharing the gradient mentioned above can be classified as follows.

[0079] (i) Method of sharing updated parameters after calculating the gradient,

[0080] (ii) A method of sharing only the change in parameters after calculating the slope,

[0081] (iii) There may be a method to merge the gradients of multiple models by directly exchanging gradients and update the parameters at once based on this.

[0082] Among these, (i) and (ii) correspond to federated learning and can be configured to apply different learning rates to each model.

[0083]

[0084] Meanwhile, regarding neural network / model weights, weight values ​​in each layer of the neural network model are transmitted from the client (UE (110)) to the server (NW (120)), and the weights are calculated while training the model based on the data held by each client (UE (110)). Clients (UE (110)) send the weights learned from their local data to the server (NW (120)), and the server (NW (120)) can update the global model by averaging them.

[0085] It is desirable that the bias values ​​present in each layer of the neural network also be shared along with the weights. Bias is a value added to the output of each neuron, which helps the model learn more complex patterns. This can also be transmitted to the server (NW(120)) and used for global model updates.

[0086] The transmission of gradients may be optional. In some methods, the client (UE (110)) may send gradient information of the loss function for local data. The server (NW (120)) can use this gradient to update the weights of the global model. However, sharing the model parameters (weights and biases) themselves is more commonly seen than sending gradient information directly.

[0087]

[0088] Meanwhile, in a preferred embodiment of the present invention, model architecture information and / or learning rate information may be additionally transmitted.

[0089] In most cases, the client (UE (110)) and the server (NW (120)) use the same model architecture, but in certain cases, the server (NW (120)) may provide model architecture information to the client (UE (110)). Since the shared parameters are based on the same architecture, the parameters learned by the client (UE (110)) can be integrated into the server (NW (120)) without any issues.

[0090] Additionally, since each client (UE (110)) may use a different learning rate during federated learning, learning rate information may be shared to help optimize. However, in most cases, the learning rate is set in the local environment of each individual client (UE (110)) and is generally not shared.

[0091]

[0092] The embodiment of FIG. 4 illustrates a case where each UE (110a-110k) transmits only parameters (310a-310k) to the network (120), and the embodiment of FIG. 5 illustrates a case where a specific UE (110a) transmits a model output value (410a) along with parameters (310a) to the network. For example, as described later in relation to FIG. 6, the models of the UE (110) and the network (120) may form a two-sided model, and the output (410a) of the UE-side model may be used as an input to the network-side model. Additionally, the embodiment of FIG. 5 illustrates a case where a specific UE (410k) transmits only the model output value (410k).

[0093] In summary, in one embodiment of the present invention, it is proposed that the training data itself is not shared. It is preferable that the data of each client (UE (110)) be securely stored on a local device and not transmitted to a central server (NW (120)). It is also preferable that intermediate results or detailed data information of the client (UE (110)) are not transmitted, thereby protecting personal information. In this way, in federated learning, model parameters (weights, biases, etc.) rather than data are shared, enabling collaborative learning while protecting user privacy.

[0094]

[0095] Meanwhile, in one embodiment of the present invention, the UE group (110a-110k) for the multi-user beam management described above may be configured based on one or more of the following: Channel Status Information (CSI) based, spatial separation based, Quality of Service (QoS) based, Signal-to-Noise Ratio (SNR) and Modulation and Coding Scheme (MCS) based, dynamic grouping based, or Coordinated Multipoint (CoMP) based.

[0096] The explanation for each method is as follows.

[0097] First, a grouping method based on CSI is described. CSI is information indicating the channel status of the UE (110), and the UE (110) can feed back this information to the network (120; e.g., base station) based on CSI-RS. The base station (120) analyzes the received feedback to evaluate the channel correlation between users and can reduce spatial interference by grouping users with low correlation. Even if users with different spatial channel characteristics are placed in the same time and frequency resources, interference can be minimized because independent spatial beamforming is possible.

[0098] Secondly, a grouping method based on spatial separation is described. This is based on the difference in physical location between users, and since users who are physically far apart have low channel correlation and do not cause interference with each other, simultaneous transmission is possible through independent beams. The base station (120) can group users expected to have less interference into the same group based on user location information or channel characteristics.

[0099] Third, there is a grouping method based on Quality of Service (QoS). UEs (110) may have different QoS requirements, and for example, if users requiring high-speed data transmission and users requiring only low bandwidth are grouped together, inefficiency in resource utilization may occur. Accordingly, efficient resource allocation is possible by grouping users with similar data speed requirements into the same group.

[0100] Fourth, there is a grouping method based on Signal-to-Noise Ratio (SNR) and Modulation and Coding System (MCS). Grouping users with similar SNRs allows the same MCS to be applied, improving the efficiency of scheduling and resource allocation. On the other hand, if users with significant differences in SNR are grouped together, resource allocation will be tailored to users with lower SNRs, which may result in performance degradation for high-efficiency users.

[0101] Fifth, there is a dynamic grouping method. This is a method that tracks user status in real time and reorganizes groups, taking into account that user location or channel status may change over time. When channel correlation changes or new users are added, the base station (120) can update user groups in real time to maintain MU-MIMO performance.

[0102] Sixth, there is the CoMP-based grouping method. CoMP is a technology in which multiple base stations cooperate to provide a beam to a single user or group of users, minimizing interference between base stations and enabling the efficient utilization of resources. CoMP-based grouping can maximize MU-MIMO performance by performing cooperative scheduling and beamforming among multiple base stations.

[0103]

[0104] FIG. 6 is a diagram illustrating an example in which a UE-side model and a network-side model form a two-sided model according to an embodiment of the present invention.

[0105] The two-sided model shown in Fig. 6 itself is a two-sided model based on the Qualcomm contribution submitted to 3GPP TR 38.843 and R-2307347. For CSI compression, federated learning methods in the UE-side model and the network-side model can be performed, such as Type 1, 2, 3 mentioned in TR 38.843 and Option 1, 2, 3, 4, 5 mentioned in the R-2307347 contribution.

[0106] Specifically, to perform the two main functions of CSI generation and CSI reconstruction, a two-dimensional model operating at both ends of the network-side model and the UE-side model may be required.

[0107] Such a two-sided model may be composed of a plurality of UE-side models (210a-210k) and a network-side model (220) of the embodiments of the present invention described above. That is, the two-sided model may be configured such that, as shown in FIG. 6, not only is one UE-side model and one network-side model connected, but a plurality of UE-side models (210a-210k) and a network-side model (220) are also connected. This may also be implemented by inputting the model output values ​​of the plurality of UE-side models (210a-210k) into the network-side model (220), as described above in relation to FIG. 5.

[0108]

[0109] CSI settings related to SetA / SetB

[0110] In a communication method using an AI / ML model, 'Set A' can represent a beam predicted by the AI / ML model, and 'Set B' can represent a measurement beam used for training the AI / ML model, etc.

[0111] For example, in case 1, which predicts the DL Tx beam in the spatial domain, a case in which the second type beam set (Set A beam) is predicted based on the measurement results of the first type beam set (Set B beam) can be implemented, and in case 2, which predicts the DL Tx beam in the time domain, a case in which the second type beam set (Set A) beam is predicted based on the past measurement results of the first type beam set (Set B beam) can be implemented.

[0112] The terms Set A / B described above may be replaced with different terms depending on the standardization process, and all embodiments of the invention proposed in this document do not exclude the use of other terms used to refer to the 'AI / ML-based prediction target beam set' and 'measurement target beam set' as described above, and may sequentially refer to them as the 'first type beam set (Set B)' and the 'second type beam set (Set A)'.

[0113] FIGS. 7 and 8 are drawings for explaining the concept that setting information according to one embodiment of the present invention represents setting information for a CSI report.

[0114] Specifically, the setting information in the embodiment described above in relation to FIG. 2 may include setting information for CSI reporting, and in one embodiment of the present invention, the setting information for CSI reporting is proposed to include one or more of information for a first type beam set (Set B) or information for a second type beam set (Set A).

[0115]

[0116] Specifically, in FIG. 7, when each beam is arranged in a horizontal / vertical direction, reference numeral 710 indicates a case where sets of first type beams (Set B1-set B3) are set within the second type beam (Set A) without overlapping each other, and reference numeral 720 indicates a case where sets of second type beams (Set B1-Set B3) are set within the second type beam (Set A) with some overlap.

[0117] This configuration information may include the following information.

[0118] (1) CSI-Resource configuration

[0119] It specifies which reference signal (nzp-CSI-RS-SSB, csi-IM-Resource) to transmit from the network (120) to the UE (110) and may be included as information configuring the transmission type (periodic, aperiodic, semipersistent).

[0120] This CSI- Resource configuration may include csi-ResouceConfigID, bwp-ID, csi-IM-ResourceSetList, csi-SSB-ResourceSetList, csi-SSB-ResourceSetListExt, nzp-CSI-RS-ResourceSetList, and resourceType.

[0121] This information allows you to determine which SSB and nzp-CSI-RS are used for CSI-RS measurement and reporting, and to perform UE-specific CSI-resource configuration.

[0122] (2) CSI-ResourceConfigID

[0123] CSI-ResourceConfigId may be included as an ID used to uniquely identify and reference the CSI-Resource configuration.

[0124] Since the first type beam (Set B) and the second type beam (Set A) are also used among the entire CSI-RS (or SSB) beams, the CSI-ResourceConfigID corresponding to the first type beam (Set B) and the second type beam (Set A) can be set.

[0125] Although the description of FIG. 7 above was explained with the focus on the CSI-RS beam, as shown in FIG. 8, the first type beam (Set B) and the second type beam (Set A) can also be set separately for the SSB beams (810a-810b), just as with the CSI-RS beams (820a-820c).

[0126]

[0127] FIGS. 9 and 10 are drawings for explaining a method in which a UE operates a first type beam set according to network configuration information according to an embodiment of the present invention.

[0128] As described above, the network (120) may provide one or more of the configuration information for the first type beam (Set B) and the configuration information for the second type beam (Set A) through the configuration information. If the network (120) provides information for the second type beam (Set A) and information for a specific first type beam (Set B1) among the first type beams, the UEs (110a-110k) may derive an estimation result for the second type beam (Set A) based on the measurement result of this specific first type beam (Set B1).

[0129] However, UEs (110a-110k) may receive setting information for a subsequent first type beam (Set B) from the network (120) as a change to the first type beam (Set B) is required during the learning process, and accordingly, may perform measurements on the beam by changing the first type beam (Set B) to Set B2 or Set B3. In particular, as illustrated by reference numeral 910 in FIG. 9, a specific first type beam (Set B3) may be set outside the range of the initially set second type beam (Set A) according to information on the first type beam (Set B) additionally received from the network (120).

[0130] In addition, as shown by reference numeral 920 in FIG. 9, the first type beam set (Set B) can be modified and used in a form that overlaps each other.

[0131] In FIG. 9, the first type beam (Set B) and the second type beam (Set A) are illustrated from the perspective of beams, but they can be similarly described from the perspective of time-frequency domain resources for measuring them, and FIG. 10 is illustrated from the perspective of measurement resources for the first type beam (Set B) and the second type beam (Set A).

[0132]

[0133] Meanwhile, in one embodiment of the present invention, when the setting information for the CSI report includes only information for the second type beam set (Set A) (e.g., Set A resource ID (1010)), the UE (110) measures the beam based on the fact that the measurement resource for the first type beam set (Set B) exists within the measurement resource range for the second type beam set (Set A), and proposes transmitting UE-side parameters based thereon to the network (120).

[0134] However, if the configuration information for the CSI report includes information for a first type beam set (Set B) (e.g., Set B Resource ID (1020)) and information for a second type beam set (Set A) (e.g., Set A Resource ID (1010)), the UE (110) measures the beam based on the measurement resource corresponding to the information for the first type beam set (Set B) (1020) and the measurement resource corresponding to the information for the second type beam set (Set A) (1010), and transmits the UE-side parameters based thereon to the network (120).

[0135] That is, in one embodiment of the present invention, information (1020) for a first type beam (Set B) is explicitly received by the UE (110), and as long as the received information (1020) does not exceed the range of a second type beam (Set A), it is proposed that the measurement resources for the first type beam (Set B) are assumed to be within the measurement resource range for the second type beam (Set A).

[0136] As described above in relation to FIG. 9, this suggests that while the network (110) may explicitly set the first type beam (Set B) outside the range of the second type beam (Set A) because a change in the first type beam (Set B) is required during the process of federated learning, without receiving such explicit setting information, the UE operates by assuming that the first type beam (Set B) is within the range of the second type beam (Set A) according to the received setting information, thereby improving the federated learning efficiency of multiple UEs (110a - 110k).

[0137]

[0138] LCM (Life Cycle Management)

[0139] Meanwhile, AI / ML models are managed through LCM (Life Cyclic Management), and it is desirable for each UE (110) to immediately report to the network (120) when the status of the model changes, such as when it becomes unavailable or available.

[0140] However, the UE (110) cannot automatically disable the model even if the model is changed to be unusable and reports this to the network (120), but the UE (110) can disable the model when it receives a deactivation instruction from the network (120).

[0141] In the case of NW-UE federated learning of the above-described embodiment, the network (120) transmits configuration information for LCM to a plurality of UEs (110a-110k) and, based on this configuration information, receives a report from the plurality of UEs (110a-110k) that are the target of NW-UE federated learning that the model is unusable.

[0142] Even if a specific model is unavailable to a specific UE (110) in a group of UEs that are the subject of federated learning, the UE (110) may maintain parameter reports for federated learning to the network (120) until it receives a deactivation instruction from the network (120) that manages the federated learning, provided that the network (120) utilizes the parameters by taking into account that they are parameters transmitted by the specific UE (110) that received such unavailable reports.

[0143] The network (120) may change the group of UEs to perform integrated learning based on reports from these UEs.

[0144] Additionally, the network (120) may optionally transmit update, retraining, or disable instruction information to the UEs (110a-110k) in consideration of reports of unusable UE-side AI / ML models received from each of the plurality of UEs.

[0145]

[0146] Device configuration

[0147] FIG. 11 illustrates a wireless device that can be applied to the present technology.

[0148] Referring to FIG. 11, the first wireless device (100) and the second wireless device (200) can transmit and receive wireless signals through various wireless access technologies (e.g., LTE, NR). Here, the first wireless device (100) and the second wireless device (200) can correspond to the UE (110) and network (120) of FIG. 2, respectively.

[0149] The first wireless device (100) includes one or more processors (102) and one or more memories (104), and may additionally include one or more transceivers (106) and / or one or more antennas (108). The processor (102) controls the memory (104) and / or transceivers (106) and may be configured to implement the descriptions, functions, procedures, proposals, methods and / or flowcharts of operation disclosed in this document. For example, the processor (102) may process information within the memory (104) to generate a first information / signal and then transmit a wireless signal containing the first information / signal through the transceiver (106). Additionally, the processor (102) may receive a wireless signal containing a second information / signal through the transceiver (106) and then store information obtained from the signal processing of the second information / signal in the memory (104). Memory (104) may be connected to the processor (102) and may store various information related to the operation of the processor (102). For example, memory (104) may store software code containing instructions for performing some or all of the processes controlled by the processor (102) or for performing the descriptions, functions, procedures, proposals, methods, and / or operation sequence diagrams disclosed in this document. Here, the processor (102) and memory (104) may be part of a communication modem / circuit / chip designed to implement wireless communication technology (e.g., LTE E-UTRA, 5G NR). A transceiver (106) may be connected to the processor (102) and may transmit and / or receive wireless signals through one or more antennas (108). The transceiver (106) may include a transmitter and / or receiver. The transceiver (106) may be combined with an RF (Radio Frequency) unit. In the present invention, the wireless device may refer to a communication modem / circuit / chip.

[0150] The second wireless device (200) includes one or more processors (202) and one or more memories (204), and may additionally include one or more transceivers (206) and / or one or more antennas (208). The processor (202) controls the memory (204) and / or transceivers (206) and may be configured to implement the descriptions, functions, procedures, proposals, methods and / or operation sequences disclosed in this document. For example, the processor (202) may process information within the memory (204) to generate a third information / signal and then transmit a wireless signal containing the third information / signal through the transceiver (206). Additionally, the processor (202) may receive a wireless signal containing a fourth information / signal through the transceiver (206) and then store information obtained from the signal processing of the fourth information / signal in the memory (204). The memory (204) may be connected to the processor (202) and may store various information related to the operation of the processor (202). For example, the memory (204) may store software code containing instructions for performing some or all of the processes controlled by the processor (202) or for performing the descriptions, functions, procedures, proposals, methods, and / or operation sequence diagrams disclosed in this document. Here, the processor (202) and the memory (204) may be part of a communication modem / circuit / chip designed to implement wireless communication technology (e.g., LTE E-UTRA, 5G NR). The transceiver (206) may be connected to the processor (202) and may transmit and / or receive wireless signals through one or more antennas (208). The transceiver (206) may include a transmitter and / or receiver. The transceiver (206) may be used in combination with an RF unit. In the present invention, the wireless device may refer to a communication modem / circuit / chip.

[0151] Hereinafter, hardware elements of the wireless device (100, 200) will be described in more detail. Although not limited thereto, one or more protocol layers may be implemented by one or more processors (102, 202). For example, one or more processors (102, 202) may implement one or more layers (e.g., functional layers such as PHY, MAC, RLC, PDCP, RRC, SDAP). One or more processors (102, 202) may generate one or more Protocol Data Units (PDUs) and / or Service Data Units (SDUs) according to the descriptions, functions, procedures, proposals, methods, and / or flowcharts of operation disclosed in this document. One or more processors (102, 202) may generate messages, control information, data, or information according to the descriptions, functions, procedures, proposals, methods, and / or flowcharts of operation disclosed in this document. One or more processors (102, 202) may generate a signal (e.g., baseband signal) containing a PDU, SDU, message, control information, data, or information according to the functions, procedures, proposals, and / or methods disclosed in this document and provide it to one or more transceivers (106, 206). One or more processors (102, 202) may receive a signal (e.g., baseband signal) from one or more transceivers (106, 206) and may obtain a PDU, SDU, message, control information, data, or information according to the descriptions, functions, procedures, proposals, methods, and / or flowcharts disclosed in this document.

[0152] One or more processors (102, 202) may be referred to as a controller, microcontroller, microprocessor, or microcomputer. One or more processors (102, 202) may be implemented by hardware, firmware, software, or a combination thereof. For example, one or more Application Specific Integrated Circuits (ASICs), one or more Digital Signal Processors (DSPs), one or more Digital Signal Processing Devices (DSPDs), one or more Programmable Logic Devices (PLDs), or one or more Field Programmable Gate Arrays (FPGAs) may be included in one or more processors (102, 202). The descriptions, functions, procedures, proposals, methods, and / or flowcharts disclosed in this document may be implemented using firmware or software, and the firmware or software may be implemented to include modules, procedures, functions, etc. Firmware or software configured to perform the descriptions, functions, procedures, proposals, methods, and / or operation sequences disclosed in this document may be contained in one or more processors (102, 202) or stored in one or more memories (104, 204) and driven by one or more processors (102, 202). The descriptions, functions, procedures, proposals, methods, and / or operation sequences disclosed in this document may be implemented using firmware or software in the form of code, instructions, and / or sets of instructions.

[0153] One or more memories (104, 204) may be connected to one or more processors (102, 202) and may store various forms of data, signals, messages, information, programs, code, instructions, and / or commands. One or more memories (104, 204) may be composed of ROM, RAM, EPROM, flash memory, hard drive, registers, cache memory, computer read storage media, and / or combinations thereof. One or more memories (104, 204) may be located inside and / or outside of one or more processors (102, 202). Additionally, one or more memories (104, 204) may be connected to one or more processors (102, 202) through various technologies such as wired or wireless connections.

[0154] One or more transceivers (106, 206) may transmit user data, control information, wireless signals / channels, etc., as mentioned in the methods and / or operation flowcharts, etc., of this document to one or more other devices. One or more transceivers (106, 206) may receive user data, control information, wireless signals / channels, etc., as mentioned in the descriptions, functions, procedures, proposals, methods and / or operation flowcharts, etc., disclosed in this document from one or more other devices. For example, one or more transceivers (106, 206) may be connected to one or more processors (102, 202) and may transmit and receive wireless signals. For example, one or more processors (102, 202) may control one or more transceivers (106, 206) to transmit user data, control information, or wireless signals to one or more other devices. Additionally, one or more processors (102, 202) may control one or more transceivers (106, 206) to receive user data, control information, or wireless signals from one or more other devices. Additionally, one or more transceivers (106, 206) may be connected to one or more antennas (108, 208), and one or more transceivers (106, 206) may be configured to transmit and receive user data, control information, wireless signals / channels, etc., as described in the descriptions, functions, procedures, proposals, methods, and / or flowcharts of operation disclosed in this document through one or more antennas (108, 208). In this document, one or more antennas may be multiple physical antennas or multiple logical antennas (e.g., antenna ports). One or more transceivers (106, 206) can convert the received wireless signal / channel, etc. from an RF band signal to a baseband signal in order to process the received user data, control information, wireless signal / channel, etc. using one or more processors (102, 202).One or more transceivers (106, 206) can convert user data, control information, wireless signals / channels, etc. processed using one or more processors (102, 202) from baseband signals to RF band signals. To this end, one or more transceivers (106, 206) may include (analog) oscillators and / or filters.

[0155]

[0156] The detailed description of the preferred embodiments of the present invention disclosed above is provided to enable those skilled in the art to implement and practice the present invention. Although the present invention has been described with reference to preferred embodiments, those skilled in the art will understand that various modifications and changes can be made to the present invention without departing from the scope of the invention. For example, those skilled in the art may utilize each configuration described in the above embodiments in a manner that combines with one another.

[0157] Accordingly, the present invention is not intended to be limited to the embodiments shown herein, but to be given the broadest scope consistent with the principles and novel features disclosed herein.

[0158] The communication method and device based on federated learning according to the embodiments of the present invention as described above are suitable for use in a communication environment utilizing AI / ML of 5G systems under 3GPP, but can also be used in the same way in 6G and subsequent next-generation mobile communication systems under 3GPP.

Claims

1. In a method for a user device (UE) to communicate with a network in a mobile communication system, Receive configuration information for a UE-side AI (Artificial Intelligence) / ML (Machine Learning) model, in which federated learning is performed by the network and the UE from the above network; and It includes providing UE-side parameters for a network-side AI / ML model corresponding to the above-mentioned UE-side AI / ML model, wherein Providing the above UE-side parameters does not provide training data for the above UE-side AI / ML model, and The above configuration information includes configuration information based on parameters collected from a plurality of UEs. Communication method.

2. In Paragraph 1, The above configuration information is received through UE-specific RRC (Radio Resource Control) signaling or group-specific RRC signaling including the plurality of UEs, Communication method.

3. In Paragraph 1, The above plurality of UEs correspond to UE groups for multi-user beam management, Communication method.

4. In Paragraph 3, The UE group for the above-mentioned multi-user beam management is, Established based on one or more of CSI (Channel Status Information), spatial separation, QoS (Quality of Service), SNR (Signal-to-Noise Ratio) and MCS (Modulation and Coding Scheme), dynamic grouping, or CoMP (Coordinated Multipoint) Communication method.

5. In Paragraph 3, Receiving the above setting information is, The plurality of UEs, including the above UE, each transmit model structure information that can be supported to the network; and Includes receiving information for setting a UE group for multi-user beam management based on the supportable model structure information of the plurality of UEs from the above network. Communication method.

6. In Paragraph 3, The above UE-side AI / ML model is learned through federated learning by the network and a plurality of UEs in the UE group for multi-user beam management. Communication method.

7. In Paragraph 1, The above configuration information includes configuration information for CSI (Channel Status Information) reporting, and The setting information for the above CSI report includes one or more of the information for a first type beam set (Set B) or the information for a second type beam set (Set A). Communication method.

8. In Paragraph 7, If the configuration information for the above CSI report includes only information for the second type beam set (Set A), the UE transmits the UE-side parameters based on the fact that the measurement resources for the first type beam set (Set B) exist within the range of measurement resources for the second type beam set (Set A). Communication method.

9. In Paragraph 7, If the configuration information for the above CSI report includes information for the first type beam set (Set B) and information for the second type beam set (Set A), the UE transmits the UE-side parameters based on the measurement resources corresponding to the information for the first type beam set (Set B) and the measurement resources corresponding to the information for the second type beam set (Set A). Communication method.

10. In Paragraph 1, The above UE-side parameters include one or more of the neural network's weights, biases, split points, or gradients. Communication method.

11. In Paragraph 1, The above configuration information additionally includes configuration information for the Life Cycle Management (LCM) of the UE-side AI / ML model; additionally including reporting to the network when the UE-side AI / ML model is unavailable according to the above LCM configuration information, Communication method.

12. In Paragraph 11, Additionally receiving update, retraining, or disable instruction information configured in consideration of reports of unusable UE-side AI / ML models received from each of the plurality of UEs from the above network, Communication method.

13. In Paragraph 11, The above UE-side AI / ML model is, including AI / ML models deployed within the UE and AI / ML models deployed on UE vendor servers, Communication method.

14. A method in which a network in a mobile communication system communicates with multiple user devices (UEs), Transmitting configuration information for UE-side AI (Artificial Intelligence) / ML (Machine Learning) models, for which federated learning is performed by the above network and the above plurality of UEs, to the above plurality of UEs; and The method includes receiving UE-side parameters for a network-side AI / ML model corresponding to the above UE-side AI / ML model from the plurality of UEs, Receiving the above UE-side parameters does not receive training data of the above UE-side AI / ML model, and The above configuration information includes configuration information based on parameters collected from the plurality of UEs. Communication method.

15. In a user device (UE) that communicates with a network in a mobile communication system, At least one processor; and It includes at least one computer memory that can be operably connected to the at least one processor and stores instructions that cause the at least one processor to perform operations when executed, and The above operations are, Receive configuration information for a UE-side AI (Artificial Intelligence) / ML (Machine Learning) model, in which federated learning is performed by the network and the UE from the above network; and It includes providing UE-side parameters for a network-side AI / ML model corresponding to the above-mentioned UE-side AI / ML model, wherein Providing the above UE-side parameters does not provide training data for the above UE-side AI / ML model, and The above configuration information includes configuration information based on parameters collected from a plurality of UEs. User device.

16. In a network that communicates with multiple user devices (UEs) in a mobile communication system, At least one processor; and It includes at least one computer memory that can be operably connected to the at least one processor and stores instructions that cause the at least one processor to perform operations when executed, and The above operations are, Transmitting configuration information for UE-side AI (Artificial Intelligence) / ML (Machine Learning) models, for which federated learning is performed by the above network and the above plurality of UEs, to the above plurality of UEs; and The method includes receiving UE-side parameters for a network-side AI / ML model corresponding to the above UE-side AI / ML model from the plurality of UEs, Receiving the above UE-side parameters does not receive training data of the above UE-side AI / ML model, and The above configuration information includes configuration information based on parameters collected from the plurality of UEs. network.

Images