Method and device for measuring and reporting channel state information in wireless communication system
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- KOREA ADVANCED INST OF SCI & TECH
- Filing Date
- 2025-12-17
- Publication Date
- 2026-06-25
Smart Images
Figure KR2025021946_25062026_PF_FP_ABST
Abstract
Description
Method and device for measuring and reporting channel state information in a wireless communication system
[0001] The present disclosure relates to a method and apparatus for measuring and reporting channel state information in a wireless communication system. Specifically, it relates to a method and apparatus for predicting channel state information (CSI) based on artificial intelligence / machine learning (AI / ML), and measuring and reporting channel state information based on the predicted CSI.
[0002]
[0003] Massive MIMO (multi-input multi-output) has been introduced in wireless communication systems to support users resource-efficiently, thereby improving system performance. While Massive MIMO-based systems can concentrate signals in a specific direction using tens or hundreds of antennas, this can significantly increase the amount of channel feedback information reported by user terminals to the base station. Consequently, measures may be required to reduce the overhead generated during the process of user terminals transmitting channel state information to the base station. Recently, various forms of artificial intelligence (AI)-based communication technologies can be applied to wireless communication systems (e.g., 3GPP, IEEE), and AI technologies for improving channel state information feedback are also being considered. AI technology is being discussed as a technique to reduce overhead and improve throughput during the transmission of channel state information; the following describes a method for measuring and reporting channel state information based on AI / ML.
[0004] The technical problems to be solved in this disclosure are not limited to those mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art to which this disclosure belongs from the description below.
[0005]
[0006] This specification relates to a method and apparatus for measuring and reporting channel state information in a wireless communication system.
[0007] This specification relates to a method and apparatus for measuring and reporting channel state information based on AI / ML-based CSI prediction in a wireless communication system.
[0008] This specification relates to a method and apparatus for performing learning and inference operations for an AI / ML-based CSI prediction model by considering uplink control information (UCI) loss information in a wireless communication system.
[0009] This specification relates to a method and apparatus for performing learning and inference operations on an AI / ML-based CSI prediction model by considering data transmission success probability information in a wireless communication system.
[0010]
[0011] According to one embodiment of the present invention, a method for operating a base station in a wireless communication system may include the steps of: transmitting a plurality of CSI-RS to a user terminal; receiving a plurality of CSI measurement reports corresponding to each time instance and a plurality of UCIs corresponding to each of the plurality of CSI measurement reports based on the measurement of the user terminal for each of the plurality of CSI-RS, wherein the base station stores the plurality of CSI measurement reports in a CSI buffer, provides the plurality of CSI measurement reports corresponding to each time instance as input to an AI / ML-based CSI prediction model, and obtains a plurality of predicted CSIs corresponding to each time instance as output based on the inference operation of the AI / ML-based CSI prediction model; generating a UCI loss probability value for each of the plurality of UCIs corresponding to each time instance and comparing the generated UCI loss probability value for each of the plurality of UCIs with a UCI loss threshold; and, when the UCI loss probability value for each of the plurality of UCIs is smaller than the UCI loss threshold, generating a data transmission success probability value and comparing the generated data transmission probability value with the data transmission threshold to determine whether to retrain the AI / ML-based CSI prediction model.
[0012] In addition, according to one embodiment of the present invention, a base station device in a wireless communication system comprises a transceiver for transmitting and receiving signals, a processor for controlling the transceiver, and a memory for storing instructions for specific operations executed by the processor, wherein the specific operations include: transmitting a plurality of CSI-RS to a user terminal, and receiving a plurality of CSI measurement reports corresponding to each time instance and a plurality of UCIs corresponding to each of the plurality of CSI measurement reports based on the user terminal's measurement for each of the plurality of CSI-RS; wherein the base station stores the plurality of CSI measurement reports in a CSI buffer, provides the plurality of CSI measurement reports corresponding to each time instance as input to an AI / ML-based CSI prediction model, obtains a plurality of predicted CSIs corresponding to each time instance as output based on the inference operation of the AI / ML-based CSI prediction model, generates a UCI loss probability value for each of the plurality of UCIs corresponding to each time instance, compares the generated UCI loss probability value for each of the plurality of UCIs with a UCI loss threshold, and, if the UCI loss probability value for each of the plurality of UCIs is smaller than the UCI loss threshold, a data transmission success probability Values are generated, and the generated data transmission probability values are compared with the data transmission threshold value to determine whether to retrain the AI / ML-based CSI prediction model.
[0013] In addition, the following points may apply in common.
[0014] According to one embodiment of the present invention, if there is a UCI having a UCI loss probability value greater than a threshold value among the UCI loss probability values for each of the plurality of UCIs, the base station may initialize a CSI measurement report corresponding to the UCI having a UCI loss probability value greater than the threshold value in the CSI buffer.
[0015] According to one embodiment of the present invention, a base station can perform retraining of an AI / ML-based CSI prediction model by reflecting CSI measurement reports initialized in a CSI buffer.
[0016] In addition, according to one embodiment of the present invention, the data transmission success probability value may be determined based on the number of multiple UCIs corresponding to each of the multiple CSI measurement reports and the ratio of the number of UCIs containing a positive response among the multiple UCIs.
[0017] In addition, according to one embodiment of the present invention, if the data transmission success probability value is smaller than the data transmission threshold, the CSI measurement report of a specific time instance derived based on the data transmission success probability value and the data transmission threshold is deleted from the CSI buffer to perform an update to the CSI buffer, and retraining of an AI / ML-based CSI prediction model can be performed based on the updated CSI buffer.
[0018] In addition, according to one embodiment of the present invention, the prediction accuracy of a retrained AI / ML-based CSI prediction model is measured, and if the prediction accuracy is smaller than a preset value, the data in the CSI buffer is initialized, and the data re-collection and AI / ML-based CSI prediction model update can be repeated until the data transmission success probability value becomes greater than the data transmission threshold.
[0019]
[0020] According to the present disclosure, a method for measuring and reporting channel status information in a wireless communication system can be provided.
[0021] According to the present disclosure, a method for measuring and reporting channel state information based on AI / ML-based CSI prediction in a wireless communication system can be provided.
[0022] According to the present disclosure, a method for performing learning and inference operations on an AI / ML-based CSI prediction model by considering UCI loss information in a wireless communication system can be provided.
[0023] According to the present disclosure, a method for performing learning and inference operations on an AI / ML-based CSI prediction model by considering data transmission success probability information in a wireless communication system can be provided.
[0024] The effects obtainable from the present disclosure are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art to which the present disclosure pertains from the description below.
[0025]
[0026] FIG. 1 is a drawing showing a wireless communication system applicable to the present disclosure.
[0027] FIG. 2 is a drawing showing an apparatus to which the present disclosure is applicable.
[0028] FIG. 3 is a diagram showing a network configuration to which the present disclosure is applicable.
[0029] FIG. 4 is a diagram showing a resource block and a resource element to which the present disclosure is applicable.
[0030] FIG. 5 is a diagram illustrating an AI / ML operation to which the present disclosure is applicable.
[0031] FIG. 6 is a diagram showing an AI / ML-based CSI prediction model applicable to the present disclosure.
[0032] FIG. 7 is a diagram showing the input and output of a CSI prediction model located on the terminal side applicable to the present disclosure.
[0033] FIG. 8 is a diagram showing the input and output of a CSI prediction model located on the network side applicable to the present disclosure.
[0034] FIG. 9 is a diagram illustrating a method for a base station and a user terminal applicable to the present disclosure to determine channel state information.
[0035] FIG. 10 is a diagram showing a case where an AI / ML-based CSI prediction model applicable to the present disclosure is located at a terminal.
[0036] FIG. 11 is a diagram showing a case where an AI / ML-based CSI prediction model applicable to the present disclosure is located at a terminal.
[0037] FIG. 12 is a diagram illustrating the operation method of an AI / ML-based CSI prediction model located on the network side applicable to the present disclosure.
[0038] FIG. 13 is a flowchart of the operation method of an AI / ML-based CSI prediction model located on the network side applicable to the present disclosure.
[0039]
[0040] Hereinafter, embodiments of the present disclosure are described in detail with reference to the attached drawings so that those skilled in the art can easily implement them. However, the present disclosure may be embodied in various different forms and is not limited to the embodiments described herein.
[0041] In describing the embodiments of the present disclosure, if it is determined that a detailed description of known configurations or functions may obscure the essence of the present disclosure, such detailed description is omitted. Furthermore, parts of the drawings unrelated to the description of the present disclosure have been omitted, and similar parts are denoted by similar reference numerals.
[0042] In the present disclosure, when a component is described as being "connected," "combined," or "joined" with another component, this may include not only a direct connection but also an indirect connection in which another component exists in between. Furthermore, when a component is described as "comprising" or "having" another component, this means that, unless specifically stated otherwise, it does not exclude the other component but may include additional components.
[0043] In the present disclosure, terms such as first, second, etc. are used solely for the purpose of distinguishing one component from another and do not limit the order or importance of the components unless specifically stated otherwise. Accordingly, within the scope of the present disclosure, a first component in one embodiment may be referred to as a second component in another embodiment, and likewise, a second component in one embodiment may be referred to as a first component in another embodiment.
[0044] In this disclosure, distinct components are intended to clearly describe their respective features and do not imply that the components are separate. That is, multiple components may be integrated to form a single hardware or software unit, or a single component may be distributed to form multiple hardware or software units. Accordingly, such integrated or distributed embodiments are included within the scope of this disclosure, unless otherwise noted.
[0045] In the present disclosure, the components described in various embodiments do not necessarily mean essential components, and some may be optional components. Accordingly, embodiments consisting of a subset of the components described in one embodiment are also included within the scope of the present disclosure. Furthermore, embodiments including additional components in addition to the components described in various embodiments are also included within the scope of the present disclosure.
[0046] The present disclosure describes a wireless communication network, and operations performed in the wireless communication network may be performed in the process of controlling the network and transmitting or receiving signals by a system (e.g., a base station) governing the wireless communication network, or in the process of transmitting or receiving signals by a terminal connected to the wireless network.
[0047] FIG. 1 is a diagram illustrating a wireless communication system applicable to the present disclosure. The wireless communication system may include 4G, 5G, and 6G communication systems. Additionally, the wireless communication system refers to a system in which communication is performed based on a network and is not limited to a specific system but can be applied to various systems.
[0048] In the wireless communication system described below, uplink and downlink transmissions can be performed between the base station (110) and the terminals (121, 122, 123). The base station (110) may refer to a network end node that communicates with the terminals (121, 122, 123), and the base station (110) may be connected to a data network (140) through a core network (130) to support communication between terminals. The core network (130) may be composed of a plurality of network nodes (or network functions) including the base station (110), and wireless communication may be performed based on the network nodes. For example, the base station (110) may be referred to as node B, and may be referred to as eNB, gNB, etc., depending on each wireless communication system. Additionally, depending on the wireless communication system, the base station may be in the form of xNB, where x can be any alphabet and is not limited to a specific form. Additionally, the base station may be referred to as at least one of a CU (central unit), DU (distributed unit), RU (radio unit), or RRH (radio remote head). Additionally, the base station may be referred to as at least one of a TP (transmission point), TRP (transmission and reception point), and relay node. Furthermore, the base station may be referred to as an access point in a WiFi system or an RSU (road side unit) in a V2X (vehicle to everything) system. The base station may also take various forms as an entity that communicates with the terminal and may not be limited to a specific form.
[0049] A terminal may refer to a fixed device or a mobile device, and may refer to a device that exchanges data by performing uplink and downlink transmissions through communication with a base station. For example, a terminal may be a smartphone, smart pad, smart watch, smart glasses, computer, laptop, and other communication-capable device. Additionally, a terminal may include an IoT (Internet of Things) device or a V2X device, and may not be limited to a specific form.
[0050] For the sake of convenience, the following description focuses on the uplink and downlink between base stations and terminals.
[0051] Furthermore, embodiments of the present disclosure may be supported by documents based on wireless access systems, such as 3GPP (3rd Generation Partnership Project) systems and IEEE 802.xx systems. Additionally, embodiments of the present disclosure may be applied to other wireless access systems and are not limited to the systems described above. Hereinafter, preferred embodiments according to the present disclosure will be described in detail with reference to the accompanying drawings. The detailed description disclosed below in conjunction with the accompanying drawings is intended to describe exemplary embodiments of the present disclosure and is not intended to represent the only embodiment in which the technical configuration of the present disclosure can be implemented.
[0052] FIG. 2 is a drawing showing an apparatus to which the present disclosure may be applied. The apparatus (200) may include a processor (210), a memory (220), an antenna (230), and a transceiver (240).
[0053] The processor (210) can perform baseband-related signal processing and execute commands stored in memory (220). Additionally, memory (220) can store information regarding commands and operations executed by the processor (210). Furthermore, memory (220) can store information processed by the processor (210), software related to the operation of the device (200), operating systems, applications, etc., and may include components such as buffers.
[0054] Additionally, the antenna (230) may include one or more physical antennas, and if it includes multiple antennas, it may support MIMO (Multiple Input Multiple Output) transmission and reception. The transceiver (230) may include a radio frequency (RF) transmitter and an RF receiver. The processor (210) of the device (200) may be configured to implement the operation in the embodiments described below.
[0055] Additionally, the device (200) may communicate with another device (250). Here, the other device (250) may also include a processor (260), memory (270), antenna (230), and transceiver (240). In one example, the device (200) and the other device (250) may be a base station and a terminal. In another example, the device (200) and the other device (250) may both be terminals. In yet another example, the device (200) and the other device (250) may be the aforementioned RIS, repeater, IoT device, vehicle, and other devices, and are not limited to specific embodiments. That is, the device (200) and the other device (250) may refer to devices that communicate based on the aforementioned embodiments, and are not limited to specific embodiments.
[0056] FIG. 3 is a diagram showing a network configuration to which the present disclosure can be applied.
[0057] Referring to FIG. 3, devices (340, 350) can communicate through a network (Internet, 310). Additionally, devices (340, 350) can be connected to an edge computing server (330). Here, a cloud data center (or big data server, 320) can be connected to the network (310). The network (310) can transmit information obtained directly from the devices (340, 350) or obtained through the edge computing server (330) to the cloud center (320). The cloud center (320) can build a model by performing training through an AI / ML (artificial intelligence / machine learning) or deep learning model (321). For example, the built model can provide inferences related to the mutual operation of the devices (340, 350), thereby improving communication efficiency. Additionally, each of the devices (340, 350) may include at least one of a processor (341, 351), a network device (342, 352), a sensor unit (343, 353), a memory (344, 354), a digital signal processing unit (345, 355), an RF transceiver (346, 356), and an antenna (347, 357), as shown in FIG. 3. Specifically, the network device (342, 352) may provide functions necessary for connecting to the network (310). Additionally, the sensor unit (343, 353) may perform the operation of sensing the aforementioned environment and network information. Additionally, the digital signal processing unit (345, 355) can perform signal processing for information transmission and can perform communication by forming a beam to the antenna (347, 357) through the RF transceiver (346, 356).
[0058] FIG. 4 is a diagram illustrating a resource block and a resource element to which the present disclosure may be applied. In a wireless communication system, the resource block and resource element may be determined based on the time domain and the frequency domain. Frames for uplink and downlink transmission may consist of 10 subframes having a time length of 1 ms, and the subframes may consist of consecutive orthogonal frequency division multiplexing (OFDM) symbols. Uplink and downlink communication are basically performed based on the frequency division duplex (FDD) method, but may operate in the time division duplex (TDD) method based on uplink and downlink switching. For example, referring to FIG. 4, resource elements may be indexed based on subcarrier spacing (SCS), and 12 resource elements may constitute a resource block. Wireless communication between a base station and a terminal may be performed based on the resource block. Additionally, the SCS may vary depending on the numerology applied in the wireless communication system. For example, the number of SCS, CP (cyclic prefix), and OFDM symbols may vary depending on the numerology, and Tables 1 and 2 below may represent the number of SCS, CP (cyclic prefix), and OFDM symbols according to the numerology. However, this is merely one example and is not limited thereto. Referring to Table 1, the SCS and CP applied may vary depending on the value of u. Here, referring to Table 2, the number of slots per frame and the number of slots per subframe may vary depending on u, and the basic unit of the resource may be determined based on this.
[0059] [Table 1]
[0060]
[0061] [Table 2]
[0062]
[0063] FIG. 5 is a diagram illustrating the operation of an AI / ML-based model applicable to the present disclosure. Referring to FIG. 5, the AI / ML-based model can be configured in various forms, and an AI / ML-based model corresponding to each function can be derived. For example, the AI / ML-based model may include data collection, model training, management, inference, and model storage functions. Here, data collection may be a function that provides data to model training, management, and inference. For example, input data may include at least one of values measured by another network, terminal feedback values, and output feedback values of the AI / ML model, but the data may not be limited to a specific form. Here, the training data provided by data collection to model training may be data provided for the AI / ML model training function, and the data provided to inference may be inference data. The AI / ML model may perform an inference operation based on the inference data. Model training can be a function that performs training, validation, and testing of AI / ML models, thereby providing performance metrics for the models. An AI / ML model trained based on model training can be provided as an inference, and results can be output by performing inference based on the inference data. For example, the output results can be provided back to model training as model performance feedback. Model training can perform updates to the AI / ML model using the results based on the inference, and the inference can be reused to repurpose the updated AI / ML model.
[0064] In addition, the inference receives inference data from data collection, can generate an output through the provided AI / ML model, and can perform operations based on the output. Below, a method for measuring and reporting channel status information based on the AI / ML model function of FIG. 5 is described.
[0065] FIG. 6 is a diagram illustrating a CSI prediction model to which the present disclosure is applicable. Referring to FIG. 6, the CSI prediction model receives historic CSI as an input value and performs inference based thereon to obtain a predicted CSI as an output. That is, the CSI prediction model can predict a future CSI based on a previous CSI. For example, the CSI prediction model can be trained. The CSI prediction model can be trained based on previously measured CSIs. Additionally, the CSI prediction model can be trained by utilizing other data and may not be limited to a specific form. Previous CSI information may be provided as an input value to the trained CSI prediction model. The CSI prediction model performs an inference operation using the association and degree of correlation between previous CSI measurement information, and thereby derives a predicted CSI as an output value. Here, the CSI prediction model may be located on at least one of the terminal side and the network side. For example, the CSI prediction model may operate by being located on only one side, either the terminal or the network, or by being located on both sides, and may not be limited to a specific form.
[0066] For example, AI / ML-based CSI feedback technology can maintain or improve throughput while reducing the overhead that occurs when a user terminal transmits channel state information to a base station. Through this, the base station and the user terminal can predict appropriate channel data and ensure system safety.
[0067] AI / ML models can learn channel change patterns during the training process to reduce the error between actual channel information and predicted channel information, and the AI / ML model can be updated based on this. The trained AI / ML model can receive previous CSI information as input and perform an inference (or inference) operation to derive the predicted CSI as output. Here, the inference process may be the operation by which the trained AI / ML model derives an output, and it may be important for the AI / ML model to learn channel change patterns to predict channel information that can improve performance in terms of throughput or overhead based on the inference process. The AI / ML model can improve the prediction accuracy of current or future channel data by utilizing channel data from multiple instances, including past time instances, to learn channel change patterns according to time. For example, to utilize channel data from multiple time instances, it is necessary to consider the prediction error of each time instance and potential error factors that may occur in the actual communication environment.
[0068] For example, consider a situation in which non-ideal uplink control information loss (UCI) is fed back in a time-domain channel state information prediction technique. The UCI may be uplink control information transmitted from a user terminal to a base station, and may include ACK / NACK, CSI information, scheduling request information, and other information, but is not limited to a specific form. The UCI may be transmitted from the user terminal to the base station via the uplink control channel, specifically through a physical uplink control channel (PUCCH). As another example, the UCI may also be transmitted from the user terminal to the base station along with data via the data channel, specifically through a physical uplink shared channel (PUSCH), and is not limited to a specific form.
[0069] For example, a non-ideal UCI feedback situation may involve cases of significant UCI loss. In other words, CSI information may not be properly transmitted from the user terminal to the base station, which can lead to reduced accuracy when training and inference operations are performed on AI / ML models based on such CSI information. However, even in an ideal UCI feedback environment, prediction errors can occur during the model training process, and these errors can affect the accuracy of channel predictions for future time instances. Therefore, a method may be required to improve the prediction accuracy of time-domain channel state information prediction technology while reducing the overhead incurred during the transmission and reception of channel state information; this is described below.
[0070] For example, channel change patterns over time may be information necessary to predict channel state information in the time domain. Considering the above, an AI / ML model may be trained by utilizing channel state information data from multiple time instances, including past channel state information data, to predict channel state information in the time domain. Previous CSI information may be provided as input to the trained AI / ML model, and the predicted CSI of the channel state information of the current time instance or the channel state information of a future time instance may be provided as output.
[0071] FIG. 7 is a diagram illustrating the input and output of a CSI prediction model located at the terminal side applicable to the present disclosure. Referring to FIG. 7, the AI / ML-based CSI prediction model (hereinafter, CSI prediction model) may be a model located at the user terminal side. The CSI prediction model can derive a prediction for a channel data set of a future time instance as an output (720) using a channel data set of a past time instance as an input (710). Subsequently, the user terminal can transmit the predicted CSI to the base station using the output (720) derived through the CSI prediction model. Here, the CSI prediction model can derive the CSI based on N predictions. Specifically, time instances Channel data is provided as input (710) to a CSI prediction model, and an inference operation is performed in the CSI prediction model to produce a future time instance as output (720). Future channel data can be predicted. Here, channel state information can be derived based on the prediction for each time instance, and the derived predicted CSI can be transmitted to the base station. The base station can recognize the channel state using the predicted CSI information obtained from the user terminal, determine the modulation coding scheme (MCS), the number of ranks, and other matters necessary for data transmission, and instruct the user terminal.
[0072] FIG. 8 is a diagram illustrating the input and output of a CSI prediction model located on the network side applicable to the present disclosure. Referring to FIG. 8, the AI / ML-based CSI prediction model (hereinafter, CSI prediction model) may be a model located on the network side. The CSI prediction model receives a channel data set of a past time instance, which is feedback received from a user terminal, as an input (710), and can derive a prediction for a channel data set of a present or future time instance as an output (720) through an inference operation. Here, the CSI prediction model can derive channel state information based on N predictions. Specifically, time instances Channel data is provided as input (710) to the CSI prediction model, and future time instances are provided as output (720) through the inference operation of the CSI prediction model. Future channel data can be predicted. Here, channel state information can be derived based on the prediction for each time instance, and the base station can recognize the channel state using the predicted CSI information, determine the MCS, rank number, and other matters necessary for data transmission, and instruct the user terminal.
[0073] The CSI prediction model can derive channel state information for current or future time instances by performing inference operations through an input dataset constructed based on channel state information of past time instances, as described above. Here, if uncertain data is included in the input dataset constructed based on channel state information of past time instances, the prediction accuracy may decrease. For example, if the channel dataset is constructed by including channel state information data from a time instance with a large UCI loss, the base station may use inaccurate channel state information data for current or future time instances.
[0074] Considering the above points, the base station can calculate the UCI loss and compare the calculated UCI loss with a preset threshold value. For example, the base station can use the probability value of the calculated UCI loss as the threshold value. It can be compared with, but is not limited to. Here, by comparing the calculated UCI with a pre-set threshold value, it can be determined whether the UCI loss affects the CSI prediction. For example, the probability value of the calculated UCI loss is as follows: In larger cases, UCI loss may affect CSI prediction.
[0075] The calculated probability value of UCI loss is the threshold value If the value is greater, the base station can transmit UCI loss information to the user terminal. The user terminal can check the UCI loss information transmitted from the base station and the input information of the CSI prediction model stored in the buffer. Using the input stored in the buffer, the CSI prediction model can reset the channel data of the time instance with the larger UCI loss among the channel state information of multiple past time instances.
[0076] As another example, the user terminal can initialize channel data of the time instance with a large UCI loss among the channel state information of multiple time instances stored in a buffer as input to the CSI prediction model, and perform retraining or updating of the CSI prediction model using the new channel data set.
[0077] The operation of resetting the channel data of the time instance with a large UCI loss among the channel state information of multiple past time instances, and the operation of resetting the channel data of the time instance with a large UCI loss among the channel state information of multiple past time instances and retraining or updating the CSI prediction model using a new channel data set may be performed optionally and may not be limited to a specific form.
[0078] FIG. 9 is a diagram illustrating a method for a base station and a user terminal applicable to the present disclosure to determine channel state information.
[0079] Referring to FIG. 9, the network (910) can transmit at least one CSI-RS (channel state information-reference signal) to a user terminal (920), and the user terminal (920) can perform measurements based on the received at least one CSI-RS. For example, the network (910) may refer to a base station operation, and for convenience of explanation below, it is referred to as a network operation, but it may not be limited to a specific term.
[0080] The user terminal (920) can generate at least one CSI report based on channel measurement information and transmit it to the network (910). Here, the CSI report can be transmitted to the network (910) along with the UCI. The network (910) can derive UCI loss information based on the UCI transmitted from the user terminal (920). For example, the network (910) can transmit UCI loss information to the user terminal (920) when certain conditions are satisfied, and this will be described later. The user terminal (920) can generate the CSI report by further considering the UCI loss information.
[0081] As a more specific example, FIG. 10 is a diagram illustrating a case where an AI / ML-based CSI prediction model applicable to the present disclosure is located at a terminal. Referring to FIG. 10, a network (1010) may transmit at least one CSI-RS to a user terminal (1020), and the user terminal (1020) may perform measurements based on the received at least one CSI-RS. As an example, the network (1010) may transmit N CSI-RS to the user terminal (1020) based on time instances, but is not limited thereto. The user terminal (1020) may generate each CSI measurement information through channel estimation based on time instances. The user terminal (1020) may store each CSI measurement information based on time instances in a CSI buffer. An AI / ML-based CSI prediction model (hereinafter, CSI prediction model) may be located at the user terminal (1010), and each CSI measurement information stored in the CSI buffer may be provided as input to the CSI prediction model. The CSI prediction model can perform inference operations based on input and derive predicted CSI information as output. Here, the predicted CSI information may be time-instance-based predicted CSI information corresponding to each time-instance-based CSI measurement information provided as input. Each time-instance-based predicted CSI information can be transmitted to the network (1010) through a CSI report. The user terminal (1020) can transmit the CSI report to the network (1010) along with the UCI. As a specific example, the time-instance-based predicted CSI information may be included in each UCI, and the user terminal (1020) can transmit N CSI reports to the network (1010).
[0082] The network (1010) can store each time instance-based predicted CSI information in a predicted CSI buffer based on the CSI report received from the user terminal (1020). Additionally, the network (1010) can generate UCI loss information using the UCI received from the user terminal (1020) and compare the generated UCI loss information with a threshold value. As a specific example, a UCI loss probability can be calculated, and the UCI loss probability is compared with the aforementioned threshold value. It can be compared to. Here, the probability of UCI loss If the value is greater than the threshold, the network (1010) can transmit UCI loss information to the user terminal (1020). For example, each UCI can correspond to each CSI report. That is, the UCI corresponding to each CSI report is the threshold value The threshold value compared to It can be determined whether it exceeds. Based on a plurality of UCIs received from a user terminal (1020), the network (1010) determines whether the UCI loss is at a threshold value. In the case of greater loss, UCI loss information can be reported to the user terminal (1020). For example, the network (1010) may not transmit loss information to the user terminal (1020) if the UCI loss probability for all of the multiple UCIs does not exceed a threshold value. On the other hand, if the loss probability of some UCIs exceeds the threshold value, the network (1010) may transmit information regarding the UCI among the multiple UCIs for which the UCI loss probability exceeds the threshold value to the user terminal (1020) by including this information in the UCI loss information. As another example, the UCI loss information may include information for each of the multiple UCIs, and for each of the multiple UCIs, the UCI for which the UCI loss probability exceeds the threshold value may be indicated in the UCI loss information and transmitted to the user terminal (1020). As a specific example, an index is assigned to each UCI corresponding to each CSI report, and the UCI loss information may include information regarding whether the UCI for which the UCI loss probability exceeds the threshold value corresponds to the index. For example, each index can be composed of 1 bit within the loss information and may indicate whether a threshold value has been exceeded.
[0083] As another example, if the UCI loss probability exceeds a threshold, the network (1010) may transmit a CSI buffer initialization request to the user terminal (1020). Here, the CSI buffer initialization request may be directed for each UCI. That is, the network (1010) may transmit a CSI buffer initialization request to the user terminal (1020) based on the CSI report corresponding to the UCI. Based on the CSI buffer initialization request, the user terminal (1020) may initialize the corresponding CSI measurement information in the CSI buffer using the time instance of the CSI report corresponding to the UCI. Afterward, the user terminal (1020) may perform CSI prediction based on the CSI measurement information stored in the CSI buffer and transmit the predicted CSI to the network (1010). Through the above, by reflecting the UCI loss information and removing previous CSI measurement information containing unclear data, the accuracy of the predicted CSI derived through the inference of the CSI prediction model can be improved.
[0084] As another example, if the UCI loss probability exceeds a threshold, the network (1010) may transmit a request for CSI prediction model retraining to the user terminal (1020). Here, the request for CSI prediction model retraining may be directed based on the CSI report corresponding to the UCI. That is, the network (1010) may transmit CSI report information corresponding to the UCI for which the UCI loss probability exceeds the threshold to the user terminal (1020). The user terminal (1020) may perform retraining of the CSI prediction model in accordance with the request for CSI prediction model retraining. As an example, the user terminal (1020) may initialize CSI measurement information corresponding to the time instance of the CSI report for the UCI for which the loss probability is greater than the threshold in the CSI buffer, and perform retraining of the CSI prediction model based on the remaining CSI measurement information. Afterward, the user terminal (1020) can perform an inference operation by providing CSI measurement information stored in the CSI buffer as input to the retrained CSI prediction model, thereby deriving the predicted CSI as output. The user terminal (1020) can transmit a CSI report containing the predicted CSI to the network (1010). That is, the accuracy of the CSI prediction model can be improved by removing previous CSI measurement information containing unclear data by reflecting UCI loss information.
[0085] FIG. 11 is a diagram showing a case where an AI / ML-based CSI prediction model applicable to the present disclosure is located at a terminal.
[0086] Referring to FIG. 11, the network (1110) may transmit at least one CSI-RS to a user terminal (1120), and the user terminal (1120) may perform measurements based on the received at least one CSI-RS. For example, the network (1110) may transmit N CSI-RS to the user terminal (1120) based on time instances, but is not limited thereto. The user terminal (1120) may generate each CSI measurement information through channel estimation based on time instances. The user terminal (1120) may store each CSI measurement information based on time instances in a CSI buffer, and each CSI information measured based on time instances may be transmitted to the network (1110) through a CSI report. The user terminal (1120) may transmit the CSI report to the network (1110) together with a UCI. As a specific example, time instance-based measured CSI information may be included in each UCI, and the user terminal (1120) may transmit N CSI reports to the network (1110).
[0087] The network (1110) can store each time instance-based measured CSI information in a CSI buffer based on the CSI report received from the user terminal (1120). The network (1110) can provide each CSI measurement information stored in the CSI buffer as input to the CSI prediction model through an AI / ML-based CSI prediction model on the network side (hereinafter, CSI prediction model). The CSI prediction model can perform an inference operation based on the input and derive predicted CSI information as output. Here, the predicted CSI information may be time instance-based predicted CSI information corresponding to each time instance-based CSI measurement information provided as input, and the predicted CSI information may be stored in the predicted CSI buffer.
[0088] Additionally, the network (1110) can generate UCI loss information using the UCI received from the user terminal (1120) and compare the generated UCI loss information with a threshold value. As a specific example, a UCI loss probability can be calculated, and the UCI loss probability is compared with the aforementioned threshold value. It can be compared to. Here, the probability of UCI loss If the value is greater than the threshold, the network (1110) can transmit UCI loss information to the user terminal (1120). For example, each UCI can correspond to each CSI report. That is, the UCI corresponding to each CSI report is the threshold value The threshold value compared to It can be determined whether it exceeds. Based on a plurality of UCIs received from a user terminal (1120), the network (1110) determines whether the UCI loss is at a threshold value. In the case of greater loss, UCI loss information can be reported to the user terminal (1120). For example, the network (1110) may not transmit loss information to the user terminal (1120) if the UCI loss probability for all of the multiple UCIs does not exceed a threshold value. On the other hand, if the loss probability of some UCIs exceeds the threshold value, the network (1110) may transmit information regarding the UCI among the multiple UCIs for which the UCI loss probability exceeds the threshold value to the user terminal (1120) by including this information in the UCI loss information. As another example, the UCI loss information may include information for each of the multiple UCIs, and for each of the multiple UCIs, the UCI for which the UCI loss probability exceeds the threshold value may be indicated in the UCI loss information and transmitted to the user terminal (1120). As a specific example, an index is assigned to each UCI corresponding to each CSI report, and the UCI loss information may include information regarding whether the UCI for which the UCI loss probability exceeds the threshold value corresponds to the index. For example, each index can be composed of 1 bit within the loss information and may indicate whether a threshold value has been exceeded.
[0089] As another example, if the UCI loss probability exceeds a threshold, the network (1110) may perform an initialization operation on the predicted CSI buffer. Here, the initialization operation on the predicted CSI buffer may be performed for each predicted CSI based on a time instance. That is, the network (1110) may perform an initialization operation on the predicted CSI information of the corresponding time instance based on the CSI report corresponding to the UCI. By removing previous CSI measurement information containing unclear data by reflecting the UCI loss information as described above, the accuracy of the predicted CSI derived through the inference of the CSI prediction model can be improved.
[0090] As another example, if the UCI loss probability exceeds a threshold, the network (1110) can retrain the CSI prediction model. Here, the CSI prediction model retraining operation can be performed based on the CSI report corresponding to the UCI. That is, the network (1110) can initialize the CSI report information corresponding to the UCI for which the UCI loss probability exceeded the threshold, and perform the CSI prediction model retraining operation using the remaining CSI report information. As an example, the network (1110) can then perform an inference operation by providing the CSI measurement information stored in the CSI buffer as input to the retrained CSI prediction model, thereby deriving the predicted CSI as output. That is, the accuracy of the CSI prediction model can be improved by removing previous CSI measurement information containing unclear data by reflecting the UCI loss information.
[0091] As described above, in the case of a non-ideal UCI feedback communication environment, AI / ML models can be managed based on UCI loss probability values. However, even in ideal UCI feedback situations, there may be a possibility that the base station incorrectly transmits channel information, such as the MCS level or the number of ranks (number of MIMO streams), to the user terminal based on channel prediction errors, channel estimation errors, and other errors. In other words, downlink transmission may be performed without the base station properly reflecting the actual channel environment, and in such cases, the base station may continuously receive negative acknowledgment (e.g., HARQ NACK) messages. Here, the base station can perform channel prediction using an input dataset consisting of time instances of the received messages. Since the AI / ML model may output prediction results with low accuracy in such cases, it may be necessary to have a method for configuring the channel dataset to accurately reflect changes in time patterns in the current channel environment and a method for modifying the AI / ML model based on this.
[0092] Considering the points mentioned above, the following describes a method to improve time-domain channel prediction accuracy in wireless communication systems while reducing or maintaining channel state information feedback overhead.
[0093] For example, a base station can periodically monitor the channel prediction performance of an AI / ML-based CSI prediction model (hereinafter referred to as the CSI prediction model) by considering both the UCI loss probability value and the data transmission success probability. That is, the performance of the CSI prediction model can be evaluated by considering not only the UCI loss probability value but also the data transmission success probability value. For example, the data transmission success probability may be information indicating whether transmission is successful based on HARQ (hybrid automatic repeat and request) ACK / NACK information. Through the above, the prediction error that may occur even in an ideal UCI feedback situation can be reduced.
[0094] Specifically, when CSI prediction model performance monitoring and model updates are performed considering UCI loss probability values and data transmission success probability values, the base station periodically ( ) can be calculated, and this is the threshold value( It can be compared with ). Here, In that case, it could be an unideal UCI feedback situation, and CSI measurement information corresponding to the time instance of the corresponding UCI can be deleted from the CSI buffer, as described above.
[0095] the other side, In this case, it could be an ideal UCI feedback situation. Here, the base station uses HARQ ACK / NACK to determine the probability of successful data transmission ( ) can be further considered, and the probability of successful data transmission ( ) can be equal to the following mathematical formula 1.
[0096] [Mathematical Formula 1]
[0097]
[0098]
[0099] Here, is the aforementioned in the case of a network-side model, and may be the aforementioned in the case of a terminal-side model. That is, the probability of successful data transmission ( ) may be the ratio of the number of UCIs indicating transmission success via ACK to the number of multiple UCIs reported from the user terminal to the base station. For example, the probability of data transmission success ( ) is a threshold value related to the probability of successful data transmission( It can be compared with ). Here, In this case, time instance based on the following mathematical formula 2 can be derived. In mathematical formula 2 can be a threshold value. For example, Based on the case, based on continuous time instances A case can be derived. As a specific example, the number of multiple UCIs may be N past multiple UCIs based on the current time instance, and among such multiple UCIs, the number of UCIs containing ACKs is determined. ...can be derived. Therefore, if the time instance changes, the population in the denominator may change, and accordingly Whether or not may vary, and time instances based on mathematical formula 2 This can be derived.
[0100] [Mathematical Formula 2]
[0101]
[0102] Based on the above Data corresponding to a specific time instance may be deleted from the CSI buffer or the predicted CSI buffer to update the CSI buffer or the predicted CSI buffer. Subsequently, retraining of the CSI prediction model may be performed based on the updated CSI buffer. Afterward, the accuracy of the retrained CSI prediction model can be measured. For example, the accuracy of the retrained CSI prediction model can be evaluated by comparing the predicted CSI information derived from the CSI measurement information provided as input to the CSI prediction model with the actual CSI information generated based on the CSI measurement information. In other words, the prediction accuracy of the model can be determined by comparing the CSI information predicted by the CSI prediction model with the CSI information derived from the actual measurement information. Here, the prediction accuracy is a specific value ( If it is less than ), the trained CSI prediction model may be difficult to use. Therefore, it is necessary to retrain or update the CSI prediction model, and considering this, the data in the CSI buffer or the predicted CSI buffer can be initialized. After that, Data recollection and model updates can be performed until the condition is satisfied.
[0103] FIG. 12 is a diagram illustrating the operation method of an AI / ML-based CSI prediction model located on the network side applicable to the present disclosure.
[0104] Referring to FIG. 12, the network (1210) may transmit at least one CSI-RS to a user terminal (1220), and the user terminal (1220) may perform measurements based on the received at least one CSI-RS. For example, the network (1210) may transmit N CSI-RS to the user terminal (1220) based on time instances, but is not limited thereto. The user terminal (1220) may generate each CSI measurement information through channel estimation based on time instances. The user terminal (1220) may store each CSI measurement information based on time instances in a CSI buffer, and each CSI information measured based on time instances may be transmitted to the network (1210) through a CSI report. The user terminal (1220) may transmit the CSI report to the network (1210) together with the UCI. As a specific example, time instance-based measured CSI information may be included in each UCI, and the user terminal (1220) may transmit N CSI reports to the network (1210).
[0105] The network (1210) can store each time instance-based measured CSI information in a CSI buffer based on the CSI report received from the user terminal (1220). The network (1210) can provide each CSI measurement information stored in the CSI buffer as input to the CSI prediction model through an AI / ML-based CSI prediction model on the network side (hereinafter, CSI prediction model). The CSI prediction model can perform an inference operation based on the input and derive predicted CSI information as output. Here, the predicted CSI information may be time instance-based predicted CSI information corresponding to each time instance-based CSI measurement information provided as input, and the predicted CSI information may be stored in the predicted CSI buffer.
[0106] Additionally, the network (1210) can generate UCI loss information using the UCI received from the user terminal (1220) and compare the generated UCI loss information with a threshold value. As a specific example, a UCI loss probability can be calculated, and the UCI loss probability is compared with the aforementioned threshold value. It can be compared to. Here, the probability of UCI loss If the value is greater than the threshold, the network (1210) can transmit UCI loss information to the user terminal (1220). For example, each UCI can correspond to each CSI report. That is, the UCI corresponding to each CSI report is the threshold value The threshold value compared to It can be determined whether it exceeds. Based on a plurality of UCIs received from a user terminal (1220), the network (1210) determines whether the UCI loss is at a threshold value. In the case of greater loss, UCI loss information can be reported to the user terminal (1220). For example, the network (1210) may not transmit loss information to the user terminal (1220) if the UCI loss probability for all of the multiple UCIs does not exceed a threshold value. On the other hand, if the loss probability of some UCIs exceeds the threshold value, the network (1210) may transmit information regarding the UCI among the multiple UCIs for which the UCI loss probability exceeds the threshold value to the user terminal (1220) by including this information in the UCI loss information. As another example, the UCI loss information may include information for each of the multiple UCIs, and for each of the multiple UCIs, the UCI for which the UCI loss probability exceeds the threshold value may be indicated in the UCI loss information and transmitted to the user terminal (1220). As a specific example, an index is assigned to each UCI corresponding to each CSI report, and the UCI loss information may include information regarding whether the UCI for which the UCI loss probability exceeds the threshold value corresponds to the index. For example, each index can be composed of 1 bit within the loss information and may indicate whether a threshold value has been exceeded.
[0107] As another example, if the UCI loss probability exceeds a threshold, the network (1210) may perform an initialization operation on the predicted CSI buffer. Here, the initialization operation on the predicted CSI buffer may be performed for each predicted CSI based on a time instance. That is, the network (1210) may perform an initialization operation on the predicted CSI information of the corresponding time instance based on the CSI report corresponding to the UCI. By removing previous CSI measurement information containing unclear data by reflecting the UCI loss information as described above, the accuracy of the predicted CSI derived through the inference of the CSI prediction model can be improved.
[0108] As another example, if the UCI loss probability exceeds a threshold, the network (1210) can retrain the CSI prediction model. The retraining of the CSI prediction model can be performed based on the CSI report corresponding to the UCI. The network (1210) can initialize the CSI measurement information corresponding to the UCI whose UCI loss probability exceeded the threshold in the CSI buffer, and perform retraining of the CSI prediction model based on the remaining CSI measurement information.
[0109] As another example, even if the loss probability of the UCI based on the time instance does not exceed a threshold value, the network (1210) has a data transmission probability value Calculate, is the threshold value It can be compared with. Here, the network (1210) is based on Equation 2. Time instance based on the case You can verify that the network (1210) is in time instance Data corresponding to can be deleted from the CSI buffer or the predicted CSI buffer, and retraining of the CSI prediction model can be performed based on the new buffer. Then, the network (1210) measures the accuracy of the retrained CSI prediction model, and the prediction accuracy is a specific value ( If less than ), the data in the CSI buffer or the predicted CSI buffer can be initialized. The network (1210) Data recollection and model updates can be performed until the condition is satisfied.
[0110] That is, the network (1210) Time instance corresponding to Derived based on the aforementioned mathematical formula 2, and the time instance Data corresponding to the CSI can be deleted from the CSI buffer or the predicted CSI buffer to perform retraining of the CSI prediction model. However, if the prediction accuracy of the retrained CSI prediction model is less than a specific value, the CSI buffer and the predicted CSI buffer can be initialized. Through the above description, the network (1210) can improve the learning accuracy of the CSI prediction model, and thereby increase the accuracy of the output value according to the inference operation.
[0111] For example, when performing channel prediction in a network as described above, the network and the user terminal may not transmit or receive channel state information data while performing channel prediction. Therefore, if channel prediction is performed based on FIG. 12, the overhead generated during the process of transmitting and receiving channel state information data can be reduced.
[0112] For example, as described above, the CSI prediction model can be configured to reflect current channel environment patterns by considering UCI loss probability values and data transmission success probability values. In other words, by considering data transmission success probability values, the CSI prediction model can improve learning accuracy to reflect the actual channel environment, and can improve prediction accuracy performance while maintaining or reducing overhead by utilizing existing data without additional data transmission or reception for channel measurement.
[0113] As described above, the CSI prediction model can perform channel prediction by being trained based on historical channel data with high accuracy. In other words, the channel data used for the training and inference operations of the CSI prediction model can be selected from historical channel data with a low probability of error by considering the probability of successful data transmission, thereby improving the accuracy of the CSI prediction model. For example, the CSI prediction model can improve accuracy based on UCI loss information by considering non-ideal UCI feedback environments, and even in ideal UCI feedback environments, the accuracy of the CSI prediction model can be improved by reducing the probability of data error by further considering the probability of successful data transmission.
[0114] FIG. 13 is a flowchart of the operation method of an AI / ML-based CSI prediction model located on the network side applicable to the present disclosure.
[0115] Referring to FIG. 13, a base station can transmit multiple CSI-RS to a user terminal (S1310). Here, the user terminal can generate multiple CSI measurement reports corresponding to each time instance based on measurements for each of the multiple CSI-RS. The base station can receive multiple CSI measurement reports and multiple UCIs corresponding to each of the multiple CSI measurement reports from the user terminal (S1320). The base station can store the multiple CSI measurement reports in a CSI buffer and provide the multiple CSI measurement reports corresponding to each time instance as input to an AI / ML-based CSI prediction model. An AI / ML-based CSI prediction model can obtain multiple predicted CSIs corresponding to each time instance as output based on an inference operation (S1330). Subsequently, the base station can generate a UCI loss probability value for each of the multiple UCIs corresponding to each time instance and compare the generated UCI loss probability value for each of the multiple UCIs with a UCI loss threshold (S1340). Subsequently, if the UCI loss probability value for each of the multiple UCIs is smaller than the UCI loss threshold, the base station can generate a data transmission success probability value and compare the generated data transmission probability value with the data transmission threshold to determine whether to retrain the AI / ML-based CSI prediction model (S1350). Additionally, as an example, the base station may include a transceiver, a process, and a memory, and can perform the above-described operation based thereon.
[0116] Here, for example, if there exists a UCI among the UCI loss probability values for each of multiple UCIs that has a UCI loss probability value greater than a threshold, the base station can initialize the CSI measurement report corresponding to the UCI with the UCI loss probability value greater than the threshold in the CSI buffer. Additionally, the base station can perform retraining of the AI / ML-based CSI prediction model by reflecting the CSI measurement report initialized in the CSI buffer.
[0117] As another example, the data transmission success probability value can be determined based on the ratio of the number of multiple UCIs corresponding to each of the multiple CSI measurement reports to the number of UCIs containing positive responses among the multiple UCIs. Here, if the data transmission success probability value is smaller than the data transmission threshold, the CSI measurement report for a specific time instance derived based on the data transmission success probability value and the data transmission threshold can be deleted from the CSI buffer to update the CSI buffer, which may be as shown in Equation 2. Subsequently, the base station can retrain the AI / ML-based CSI prediction model based on the updated CSI buffer.
[0118] In addition, as an example, the prediction accuracy of a retrained AI / ML-based CSI prediction model can be measured, and if the prediction accuracy is lower than a preset value, the data in the CSI buffer can be initialized, and the data re-collection and AI / ML-based CSI prediction model update can be repeated until the data transmission success probability value becomes greater than the data transmission threshold, as described above.
[0119]
[0120] The various embodiments of the present disclosure are not intended to list all possible combinations but to describe representative aspects of the present disclosure, and the matters described in the various embodiments may be applied independently or in combination of two or more.
[0121] In addition, various embodiments of the present disclosure may be implemented by hardware, firmware, software, or a combination thereof. In the case of implementation by hardware, it may be implemented by one or more ASICs (Application Specific Integrated Circuits), DSPs (Digital Signal Processors), DSPDs (Digital Signal Processing Devices), PLDs (Programmable Logic Devices), FPGAs (Field Programmable Gate Arrays), general processors, controllers, microcontrollers, microprocessors, etc.
[0122] The scope of the present disclosure includes software or machine-executable instructions (e.g., operating system, application, firmware, program, etc.) that enable an operation according to a method of various embodiments to be executed on a device or computer, and a non-transitory computer-readable medium on which such software or instructions, etc. are stored and executable on a device or computer.
[0123]
[0124] The present disclosure is applicable to a wireless communication system comprising a method and apparatus for measuring and reporting channel state information.
Claims
1. In a method of operating a base station in a wireless communication system, A step of transmitting multiple CSI-RS (channel state information-reference signals) to a user terminal; A step of receiving a plurality of CSI measurement reports corresponding to each time instance and a plurality of UCIs (uplink control information) corresponding to each of the plurality of CSI measurement reports based on the measurement of the user terminal for each of the plurality of CSI-RSs, wherein the base station stores the plurality of CSI measurement reports in a CSI buffer; A step of providing the plurality of CSI measurement reports corresponding to each of the above time instances as input to an AI / ML (artificial intelligence / machine learning) based CSI prediction model, and obtaining a plurality of predicted CSIs corresponding to each of the above time instances as output based on the inference operation of the AI / ML based CSI prediction model; A step of generating a UCI loss probability value for each of the plurality of UCIs corresponding to each of the above time instances, and comparing the generated UCI loss probability value for each of the plurality of UCIs with a UCI loss threshold; and A base station operation method comprising the step of generating a data transmission success probability value when the UCI loss probability value for each of the plurality of UCIs is smaller than the UCI loss threshold, and determining whether to retrain the AI / ML-based CSI prediction model by comparing the generated data transmission probability value with the data transmission threshold.
2. In Paragraph 1, A base station operation method in which, if there is a UCI having a UCI loss probability value greater than the threshold value among the UCI loss probability values for each of the plurality of UCIs, the base station initializes a CSI measurement report corresponding to the UCI having a UCI loss probability value greater than the threshold value in the CSI buffer.
3. In Paragraph 1, A base station operation method in which the base station performs retraining of the AI / ML-based CSI prediction model by reflecting the CSI measurement report initialized in the CSI buffer.
4. In Paragraph 1, A base station operation method in which the above data transmission success probability value is determined based on the ratio of the number of the plurality of UCIs corresponding to each of the plurality of CSI measurement reports and the number of UCIs containing a positive response among the plurality of UCIs.
5. In Paragraph 1, If the above data transmission success probability value is smaller than the above data transmission threshold, an update to the CSI buffer is performed by deleting the CSI measurement report of a specific time instance derived based on the above data transmission success probability value and the above data transmission threshold from the CSI buffer, and A base station operation method that performs retraining of the AI / ML-based CSI prediction model based on the updated CSI buffer.
6. In Paragraph 5, A base station operation method comprising measuring the prediction accuracy of the retrained AI / ML-based CSI prediction model, and if the prediction accuracy is smaller than a preset value, initializing the data in the CSI buffer, and repeating data re-collection and updating the AI / ML-based CSI prediction model until the data transmission success probability value becomes greater than the data transmission threshold value.
7. In a base station device in a wireless communication system, A transceiver that transmits and receives signals; A processor controlling the above transceiver; and It includes memory that stores instructions for specific operations executed by the above-mentioned processor, The above specific operation is: Transmit multiple CSI-RS to the user terminal, and Based on the measurement of the user terminal for each of the plurality of CSI-RSs, the base station receives a plurality of CSI measurement reports corresponding to each time instance and a plurality of UCIs corresponding to each of the plurality of CSI measurement reports, wherein the base station stores the plurality of CSI measurement reports in a CSI buffer. The plurality of CSI measurement reports corresponding to each of the above time instances are provided as input to an AI / ML-based CSI prediction model, and a plurality of predicted CSIs corresponding to each of the above time instances are obtained as output based on the inference operation of the AI / ML-based CSI prediction model. A UCI loss probability value is generated for each of the plurality of UCIs corresponding to each of the above time instances, and the generated UCI loss probability value for each of the plurality of UCIs is compared with a UCI loss threshold, and A base station that, when the UCI loss probability value for each of the plurality of UCIs is smaller than the UCI loss threshold, generates a data transmission success probability value and compares the generated data transmission probability value with the data transmission threshold to determine whether to retrain the AI / ML-based CSI prediction model.
8. In Paragraph 7, A base station that, when there is a UCI having a UCI loss probability value greater than the threshold value among the UCI loss probability values for each of the plurality of UCIs, initializes a CSI measurement report corresponding to the UCI having a UCI loss probability value greater than the threshold value in the CSI buffer.
9. In Paragraph 7, The base station performs retraining on the AI / ML-based CSI prediction model by reflecting the CSI measurement report initialized in the CSI buffer.
10. In Paragraph 7, A base station, wherein the data transmission success probability value is determined based on the ratio of the number of UCIs corresponding to each of the plurality of CSI measurement reports and the number of UCIs containing a positive response among the plurality of UCIs.
11. In Paragraph 7, If the above data transmission success probability value is smaller than the above data transmission threshold, an update to the CSI buffer is performed by deleting the CSI measurement report of a specific time instance derived based on the above data transmission success probability value and the above data transmission threshold from the CSI buffer, and A base station that performs retraining on the AI / ML-based CSI prediction model based on the updated CSI buffer.
12. In Paragraph 11, A base station that measures the prediction accuracy of the retrained AI / ML-based CSI prediction model, initializes the data in the CSI buffer when the prediction accuracy is smaller than a preset value, and repeats data recollection and updating of the AI / ML-based CSI prediction model until the data transmission success probability value becomes greater than the data transmission threshold value.