Method and apparatus for reporting channel state information in wireless communication system

Abstract

A 5th-Generation (5G) or 6th-Generation (6G) communication system is provided for supporting higher data rates following a 4th-Generation (4G) communication system, such as Long Term Evolution (LTE). According to an embodiment of the present disclosure, a user equipment (UE) may compress estimated channel state information (CSI) data between a base station (BS) and the UE through an artificial intelligence (AI) model for encoding, where the AI model for encoding is trained in association with an AI model for decoding in the BS; transmitting the compressed CSI data to the BS; sending the CSI data before compression to the BS based on a preset condition; and receiving information for updating the AI model for encoding when the BS determines to update the AI model for encoding based on the CSI data.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to reporting channel state information (CSI) in a wireless communication system, and more particularly, to a method and apparatus for reporting CSI based on artificial intelligence (AI). Background Art

[0002] Looking back at the development of radio communications generation after generation, technologies for human services such as voice, multimedia, data, etc. have been developed. After the commercialization of the fifth generation (5G) communication system, an explosive increase in connected devices is expected to be connected to the communication network. As examples of things connected to the network, there may be cars, robots, drones, home appliances, displays, smart sensors installed in various infrastructures, construction machinery, factory equipment, etc. Mobile devices are expected to evolve into various form factors such as augmented reality (AR) glasses, virtual reality (VR) headsets, hologram devices, etc. In order to provide various services by connecting hundreds of billions of devices and things in the sixth generation (6G) era, efforts are being made to develop better 6G communication systems. For these reasons, the 6G communication system is called a super-5G system.

[0003] In the 6G communication system expected to be realized around 2030, the maximum transmission rate is 1 terabit per second (bps), that is, 1000 gigabits per second, and the maximum wireless delay is 100 microseconds (μsec). In other words, in the 6G communication system, the transmission rate is accelerated by 50 times, while the wireless delay is reduced to one-tenth of that of the 5G communication system.

[0004] In order to obtain these high data transmission rates and ultra-low latency, the 6G communication system is considered to be implemented in the terahertz (THz) band (e.g., from 95 gigahertz (GHz) to 3THz). Since the path loss and atmospheric absorption phenomena in the THz band are more serious compared to the millimeter wave (mmWave) band introduced in the 5G system, the importance of technologies for ensuring signal range, i.e., coverage is expected to increase. As the main technologies for ensuring coverage, it is necessary to develop radio frequency (RF) components, antennas, new waveforms that are superior to orthogonal frequency division multiplexing (OFDM) in terms of coverage, beamforming, and massive multiple input and multiple output (massive MIMO), full-dimensional MIMO (FFD-MIMO), array antennas, multi-antenna transmission technology (e.g., massive antenna technology), etc. In addition, new technologies for increasing THz band signal coverage are being discussed, such as metamaterial-based lenses and antennas, high-dimensional spatial multiplexing technology using orbital angular momentum (OAM), reconfigurable smart surfaces (RIS), etc.

[0005] In addition, in order to improve frequency efficiency and system network, in the 6G communication system, the following are being developed: full-duplex technology that uses the same frequency resources for both uplink and downlink transmissions, network technology that uses satellites and high altitude platform stations (HAPS) etc. in combination, network structure innovation technology that supports mobile base stations and allows network operation optimization and automation, dynamic spectrum sharing technology that avoids collisions based on spectrum usage prediction, artificial intelligence-based communication technology that achieves system optimization by using AI (AI) from the design stage and internalizes end-to-end AI support functions, and next-generation distributed computing technology that realizes services whose complexity exceeds the computing power limit of terminals by using ultra-high-performance communication and computing resources (e.g., mobile edge computing (MEC) cloud). In addition, by designing new protocols for the 6G communication system, the following are being developed: mechanisms for realizing a hardware-based secure environment and secure use of data and technologies for protecting privacy, as well as attempts to strengthen connectivity between devices, further optimize networks, promote softwareization of network entities, and increase the openness of wireless communications.

[0006] With such research and development of 6G communication systems, it is expected that the next level of hyper-connected experience will become possible through the hyper-connectivity of 6G communication systems (including not only the connection between things but also the connection between people and things). Specifically, it is predicted that services such as true immersive extended reality (true immersive XR), high-fidelity mobile holograms, digital replicas, etc. can be provided. In addition, services such as remote surgery, industrial automation, and emergency response with enhanced security and reliability can be provided through 6G communication systems for application in various fields such as industry, medical care, vehicles, and home appliances. Summary of the invention

[0007] [Technical issues]

[0008] Embodiments of the present disclosure provide a method and apparatus for performing channel state information (CSI) reporting based on an artificial intelligence (AI) model.

[0009] [Technical solution]

[0010] A fifth generation (5G) or sixth generation (6G) communication system is provided for supporting a higher data rate after a fourth generation (4G) communication system such as long term evolution (LTE). According to an embodiment of the present disclosure, a UE can compress channel state information (CSI) data estimated between a base station (BS) and the UE by using an AI model for encoding trained in relation to an AI model for decoding in the BS, send the compressed CSI data to the BS, send the CSI data before compression to the BS based on a preset condition, and receive information for updating the AI ​​model for encoding when the BS determines to update the AI ​​model for encoding based on the CSI data. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 is a diagram describing an autoencoder-based compression and reconstruction process for channel state information (CSI) data.

[0012] Figure 2 is a diagram for describing an example of a method in which a base station (BS) reconstructs compressed CSI data.

[0013] Figure 3 is a flowchart for describing a method of performing CSI profiling according to an embodiment.

[0014] Figure 4 is a flowchart for describing a method of performing CSI profiling according to a request of a BS according to an embodiment.

[0015] Figure 5a is a diagram for describing a method of specifying a portion of CSI data according to an embodiment.

[0016] Figure 5b is a diagram for describing a method of specifying a portion of CSI data according to an embodiment.

[0017] Figure 6 is a flowchart for describing the operation of a BS for CIS profiling according to an embodiment.

[0018] Figure 7 is a flow chart for describing the operation of a user equipment (UE) for CSI profiling according to an embodiment.

[0019] Figure 8 is a flowchart for describing a method of performing CSI profiling at preset intervals according to an embodiment.

[0020] Fig. 9 is a flowchart for describing a method of performing CSI profiling according to a request of a UE according to an embodiment.

[0021] Fig.10 is a flowchart for describing a method of performing CSI profiling at preset intervals according to an embodiment.

[0022] Fig.11 is a block diagram of a UE according to an embodiment.

[0023] Fig.12 is a block diagram of a BS according to an embodiment. DETAILED DESCRIPTION

[0024] Best Mode

[0025] According to an embodiment, a method for a user equipment (UE) to report channel state information (CSI) in a wireless communication system may include: compressing CSI data estimated between a base station (BS) and the UE through an artificial intelligence (AI) model for encoding, which artificial intelligence (AI) model is jointly trained with the AI ​​model for decoding of the base station; sending compressed CSI data to the BS; based on preset conditions, sending the CSI data before compression to the BS; and, when the base station determines to update the AI ​​model for encoding based on the CSI data, receiving information for updating the AI ​​model for encoding.

[0026] According to an embodiment, a method for a BS for reporting CSI in a wireless communication system may include: receiving CSI data compressed by an AI model for encoding from a UE, which AI model is jointly trained with an AI model for decoding of the BS; based on preset conditions, receiving CSI data before compression from the UE; determining to update the AI ​​model for encoding based on the CSI data; and sending information for updating the AI ​​model for encoding.

[0027] According to an embodiment, a UE for reporting CSI in a wireless communication system includes a transceiver; and a processor connected to the transceiver, and the processor can be configured to: compress CSI data estimated between the BS and the UE through an AI model for encoding jointly trained with the AI ​​model for decoding of the BS, send the compressed CSI data to the BS, send the CSI data before compression to the BS based on preset conditions, and, when the BS determines to update the AI ​​model for encoding based on the CSI data, receive information for updating the AI ​​model for encoding.

[0028] According to an embodiment, a BS for reporting CSI in a wireless communication system includes a transceiver; and a processor connected to the transceiver, and the processor can be configured to: receive CSI compressed by an AI model for encoding jointly trained with an AI model for decoding of the BS from a UE, receive CSI data before compression from the UE based on preset conditions, determine to update the AI ​​model for encoding based on the CSI data, and send information for updating the AI ​​model for encoding.

[0029] According to an embodiment of the present disclosure, there is provided a computer-readable recording medium storing a program on a computer to execute at least one embodiment of the present disclosure.

[0030] According to an embodiment of the present disclosure, CSI reporting is performed by using an AI model that adapts to environmental changes such as network characteristics, channel characteristics, or data characteristics, thereby avoiding performance degradation of the communication system.

[0031] A person skilled in the art can clearly understand other technical features by referring to the drawings, the description and the appended claims.

[0032] Before starting "detailed description", the specific terms and phrases used in the entire specification will now be defined first. The term "connection" and its derivatives refer to direct or indirect communication between two or more elements, regardless of whether these elements are physically in contact with each other. The terms "send", "receive" and "communication" and their derivatives not only include direct communication and indirect communication. The terms "include" and "comprise" and their derivatives refer to including but not limited to. The term "or" is inclusive and refers to "and / or". The phrase "related to A" and its derivatives refer to "including A", "included in A", "interconnected with A", "including A", "included in A", "connected to A or connected with A", "connected to A or connected with A", "communicating with A", "collaborating with A", "interlaced A", "parallel A", "close to A", "associated A or associated with A", "having A", "characterized by A", "having a relationship with A", etc.

[0033] When the term "connected" or "coupled" is used, a component may be directly connected or coupled to another component. However, unless otherwise defined, it should also be understood that the component may be indirectly connected or coupled to another component via another new component.

[0034] Invention Mode

[0035] Various modifications can be made to the embodiments of the present disclosure, and the embodiments of the present disclosure will be described more fully below with reference to the accompanying drawings. It should be understood that the present disclosure is not limited to a specific embodiment, but includes all modifications, equivalents and replacements belonging to the technical scope and concept of the present disclosure.

[0036] In the description of the embodiments, when it is considered that the detailed explanation of the related art may unnecessarily obscure the essence of the present disclosure, the detailed explanation of the related art is omitted. The serial numbers (e.g., first, second, etc.) used in the explanation of this specification are only used to distinguish components from each other.

[0037] When the term "connected" or "coupled" is used, a component may be directly connected or coupled to another component. However, unless otherwise defined, it should also be understood that the component may be indirectly connected or coupled to another component via another new component.

[0038] Throughout the specification, components represented by "... unit", "... module", etc. may be a combination of two or more components, or may be divided into two or more by function. Each component may perform its main function, and further perform a part or all of the functions served by another component. In this way, a part of the main function served by each component may be dedicated and may be performed by another component.

[0039] A base station (BS) as used herein may refer to an entity for performing resource allocation for a user equipment (UE), and may be at least one of a gNode B, an eNode B, a Node B (or an xNode B, where x represents any letter including 'g' and 'e'), a radio access unit, a BS controller, a satellite, an aircraft, or a node in a network. A UE may be used interchangeably with a terminal, and may include a mobile station (MS), a vehicle, a satellite, an aircraft, a cellular phone, a smart phone, a computer, or a multimedia system with a communication function. In the present disclosure, a downlink (DL) may refer to a radio transmission path for a signal sent from a BS to a UE, and an uplink (UL) may refer to a radio transmission path for a signal sent from a UE to a BS.

[0040] Although the following embodiments of the present disclosure will focus on long term evolution (LTE), LTE-Advanced (LTE-A) or fifth generation (5G) systems as examples, they can be applied to other communication systems with similar technical backgrounds or channel types. For example, other communication systems may include 5G-Advanced, New Radio (NR)-Advanced or sixth generation (6G) mobile communication technology developed after 5G mobile communication technology (or NR), and the term 5G may be a concept that includes existing LTE, LTE-A and other similar services. In addition, the embodiments of the present disclosure will also be applied to different communication systems with some modifications to the extent that they do not significantly deviate from the scope of the present disclosure when judged by those skilled in the art.

[0041] It should be understood that each box and combination of boxes of the flowchart can be executed by computer program instructions. Computer program instructions can be loaded onto a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, and thus they generate a device for performing the functions described in the (one or more) boxes of the flowchart when executed by the processor of the computer or other programmable data processing device. Computer program instructions can also be stored in a computer executable or computer readable memory, which can guide a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer executable or computer readable memory can produce a product including an instruction device for performing the functions specified in the (one or more) flowchart boxes. Computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are performed on the computer or other programmable device, thereby generating a computer-implemented process, so that the instructions executed on the computer or other programmable device provide operations for implementing the functions specified in the (one or more) flowchart boxes.

[0042] In addition, each box can represent a module, segment or part of code including one or more executable instructions for performing (one or more) specific logical functions. It should be noted that in some alternative embodiments, the functions described in the box can occur out of order. For example, two consecutive boxes can be executed substantially simultaneously, or in reverse order according to the corresponding functions.

[0043] In the following description, for the convenience of explanation, terms related to broadcast information, control information, state changes (e.g., events), network entities, components of messages and devices, terms related to communication coverage, etc. are mentioned. The present disclosure is not limited to the terms to be used in the following description, and different terms having the same meaning may be used in a technical sense.

[0044] In the following description, for ease of explanation, terms and definitions used in the latest standards in the current existing communication standards (i.e., LTE and NR standards defined in the 3rd Generation Partnership Project (3GPP)) will be used. However, the present disclosure is not limited to the terms and definitions and can be equally applied to any system conforming to other standards.

[0045] As a future communication system after LTE, the 5G communication system needs to freely reflect various demands from users and service providers, and thus support services that simultaneously meet various demands. Services considered for the 5G communication system may include enhanced mobile broadband (eMBB), massive machine type communication (mMTC), ultra-reliable low latency communication (URLLC), etc.

[0046] The purpose of eMBB is to provide more enhanced data rates than LTE, LTE-A or LTE-Pro can support. For example, in a 5G communication system, eMBB is required to provide a peak data rate of 20Gbps in DL and a peak data rate of 10Gbps in UL for a single BS. In addition, the 5G communication system needs to provide an increased user-perceived data rate while providing the peak data rate. In order to meet these requirements, it is required to enhance various technologies for transmission or reception including multiple-input multiple-output (MIMO) transmission technology. Although the current LTE uses a transmission bandwidth of up to 20MHz in the 2GHz band for signal transmission, the 5G communication system can use a frequency bandwidth wider than 20MHz in the 3 to 6GHz band or in the 6GHz or higher band, thereby meeting the data rate required by the 5G communication system.

[0047] Meanwhile, in the 5G communication system, mMTC is considered to support application services such as Internet of Things (IoT) application services. In order for mMTC to effectively provide IoT, support for access from a large number of terminals in a cell, enhanced terminal coverage, extended battery time, reduced terminal prices, etc. are required. Because IoT is equipped in various sensors and devices to provide communication functions, it can be assumed that it supports a large number of UEs in a cell (e.g., 1,000,000 terminals / km 2 ). In addition, by the nature of the service, mMTC-enabled UEs are more likely to be located in shadowed areas such as underground in buildings, which may not be covered by cells, and therefore mMTC requires even greater coverage than expected for other services provided by 5G communication systems. mMTC-enabled UEs need to be low-cost UEs, and since batteries in UEs are difficult to replace frequently, a fairly long battery life, such as 10-15 years, is required.

[0048] Finally, URLLC is a mission-critical cellular-based wireless communication service. For example, URLLC can provide services for remote control of robots or machines, industrial automation, drones, remote health care, emergency alerts, etc. Therefore, the communication provided by URLLC requires very low latency and very high reliability. For example, URLCC services need to meet sub-millisecond (less than 0.5 milliseconds) air interface latency and at the same time require a reliability of equal to or lower than 10 -5 Therefore, for URLLC services, the 5G system needs to provide a smaller transmission time interval (TTI) than other services, and at the same time requires a design that allocates a wide range of resources to the frequency band to ensure the reliability of the communication link.

[0049] In order to meet various requirements of corresponding services, system operation methods such as adjusting beams or changing frequency bands may be considered. In this case, there may be various channel environments for corresponding beams / frequency bands.

[0050] Channel estimation is an important factor in system performance. The UE can estimate the channel based on a reference signal (RS) sent from the BS. There may be various RSs sent from the BS to the UE for channel estimation, but in this disclosure, for ease of explanation, a channel state information (CSI) RS (CSI-RS) is taken as an example. The UE can estimate the channel between the BS and the UE based on the CSI-RS. In addition, the UE can report information about the estimated channel (i.e., CSI) to the BS.

[0051] Traditionally, codebook-based CSI reporting is supported. The codebook approach causes information loss in the process of quantizing information about the estimated channel state. A method of reducing information loss is to configure the codebook differently, but this increases the size of the codebook to be sent, resulting in an increase in the size of the time-frequency resources used for CSI reporting.

[0052] To this end, a method of reporting compressed CSI based on an autoencoder is discussed. The autoencoder is a neural network model consisting of an encoder for compressing input data and a decoder for obtaining output data by reconstructing the compressed data to be equal to the input data.

[0053] In the present specification, an encoder may include an AI model for encoding, the AI ​​model including a neural network that performs compression on input data. A decoder may include an AI model for decoding, the AI ​​model including a neural network that performs reconstruction on compressed data.

[0054] Furthermore, in this specification, a neural network (NN) is an AI model that simulates cranial nerves, and is not limited to an AI model that uses a specific algorithm.

[0055] In addition, in this specification, a parameter is a value used in an operation process on each layer constituting a NN, and may include, for example, a weight used when an input value is applied to a certain operation expression. The parameter may be represented in the form of a matrix. The parameter is a value set as a training result, which may be updated with additional training data as needed.

[0056] In this specification, NN configuration information is information about elements constituting NN and includes parameters. Using the NN configuration information, an AI model for encoding or an AI model for decoding can be configured. The NN configuration information may include information about at least one of the number of convolutional layers included in the NN, the number of filter kernels of each convolutional layer, and the parameters of each filter kernel.

[0057] Figure 1 is a diagram for describing a CSI compression and reconstruction process using an autoencoder.

[0058] See also Figure 1 , the autoencoder 100 may include an encoder 110 and a decoder 120. The encoder 110 may include an AI model for encoding, and the decoder 120 may include an AI model for decoding. However, this is merely an example, and in addition to the AI ​​model for encoding and the AI ​​model for decoding, the encoder 110 and the decoder 120 may also include a pre-processor or a post-processor for performing a function of enhancing CSI compression and reconstruction performance.

[0059] The encoder 110 may be implemented in a UE. The UE may estimate a channel between the UE and the BS based on a CSI-RS received from the BS. The UE may recognize a CSI-RS transmitted by the BS and thereby estimate a channel through which the CSI-RS has passed based on a received signal and the CSI-RS. The channel estimated by the UE may be expressed as a channel matrix as shown in the following equation 1 express:

[0060] [Equation 1]

[0061]

[0062] Where, the channel matrix in Equation 1 In, N r Represents the number of receiving antennas, N t represents the number of transmit antennas, i represents the symbol index, and k represents the subcarrier index. r，t (i, k) is the channel matrix The elements of represent the channel state in the kth subcarrier and the i-th symbol between the rth receiving antenna and the tth transmitting antenna.

[0063] The UE can use singular value decomposition (SVD) to calculate the channel matrix Decompose. Through SVD, the channel matrix It can be expressed as the product of three matrices as shown in Equation 2:

[0064] [Equation 2]

[0065]

[0066] In Equation 2, U and V indicate regular matrices, and D is a diagonal matrix including non-negative singular values.

[0067] In addition, the UE can decompose the channel matrix by using eigenvalue decomposition (EVD) as in Equation 3 Through EVD, channel matrix It can be expressed as the following equation 3:

[0068] [Equation 3]

[0069]

[0070] In Equation 3, U and V denote regular matrices, and D is a diagonal matrix including non-negative singular values.

[0071] Since the channel matrix can be determined when information about V is obtained Traditional CSI reporting is performed by the UE sending the index of the codebook most similar to V to the BS.

[0072] In case of sending compressed V using an autoencoder, it can provide information about the channel state to the BS by using fewer resources than sending V while avoiding information loss that would occur when sending a codebook.

[0073] The UE may input V to the encoder 110 and obtain an output z, i.e., compressed V. The UE may send z to the BS on the time-frequency resource. The decoder 120 including the AI ​​model for decoding may run in the BS 120. The BS may input the received z to the decoder 120 and obtain As output, in this specification, V may be described as CSI data, and z may be described as compressed CSI data.

[0074] The AI ​​model for encoding of encoder 110 and the AI ​​model for decoding of decoder 120 can be jointly trained so that can have similar values ​​to the input V. For example, the loss information can be used to reconstruct The difference between the input V and the back propagation process is performed. The training can be performed by updating the NN configuration information of the AI ​​model for encoding and the NN configuration information of the AI ​​model for decoding so that the reconstructed has a similar value to the input V. For example, the BS may pre-acquire CSI data by, for example, field testing or acquiring CSI data received from a plurality of UEs in real time, and perform training based on the acquired CSI data.

[0075] The NN configuration information of the AI ​​model for encoding may include information about at least one of the number of convolutional layers included in the AI ​​model for encoding, the number of filter kernels of each convolutional layer, the size of the filter kernel of each convolutional layer, the parameters of each filter kernel, or the parameters of the activation layer. The NN configuration information of the AI ​​model for decoding may include information about at least one of the number of convolutional layers included in the AI ​​model for decoding, the number of filter kernels of each convolutional layer, the size of the filter kernel of each convolutional layer, the parameters of each filter kernel, or the parameters of the activation layer.

[0076] The AI ​​model for encoding and the AI ​​model for decoding of the autoencoder can be trained in the BS or a separate server. Further, the BS or the server can store the NN configuration information of the AI ​​model for encoding and the NN configuration information of the AI ​​model for decoding. The UE can receive the NN configuration information of the AI ​​model for encoding from the BS or the server.

[0077] When the characteristics or data of the previously trained dataset are changed due to changes in network or channel conditions, the AI ​​model suffers from performance degradation. In this case, in order to improve the performance of the AI ​​model, a different AI model suitable for the environment is used, or a fine-tuning process of additional training of some layers is required.

[0078] At the same time, the structure of the autoencoder for CSI compression includes an AI model for encoding on the UE side and an AI model for decoding on the BS side. Even when the performance of the autoencoder deteriorates due to changes in network or channel conditions or data characteristics, it is difficult to determine whether V is correctly reconstructed because the BS's AI model for decoding only receives z. In an embodiment of the present disclosure, in order to minimize the performance degradation of the autoencoder, a CSI profiling method is provided, by which the BS collects V values ​​from the UE at a certain time / situation, and uses the collected V to analyze the performance of the AI ​​model in the current environment. In the present disclosure, through CSI profiling, the BS can change or fine-tune the AI ​​model for encoding running in the UE.

[0079] Figure 2 is a diagram for describing an example of a method in which a BS reconstructs compressed CSI data.

[0080] Reference Figure 2 , the BS may input the compressed CSI data z received from the UE into the first convolution layer 210 of the AI ​​model for decoding. In the convolution operation flow of the first convolution layer 210, a multiplication operation may be performed on each element of the compressed CSI data z and each filter kernel value, and a combined value (e.g., addition operation) of the result of the multiplication operation may be output as a feature map. The feature map output from the first convolution layer 210 is input to the first activation layer 220.

[0081] The first activation layer 220 may give nonlinear characteristics to each feature map. The first activation layer 220 may include a Sigmoid function, a Tanh function, a rectified linear unit (ReLU) function, etc., but is not limited thereto. Giving nonlinear characteristics to the first activation layer 220 means changing some sample values ​​of the feature map output by the first convolutional layer 210 and outputting the result. In this case, the change is performed by applying nonlinear characteristics.

[0082] The first activation layer 220 determines whether to forward the sample values ​​of the feature map output from the first convolution layer 210 to the second convolution layer 230. For example, some sample values ​​of the feature map are activated by the first activation layer 220 and forwarded to the second convolution layer 230, and some sample values ​​are deactivated by the first activation layer 220 and not forwarded to the second convolution layer 230. The first activation layer 220 emphasizes the unique characteristics of the compressed CSI data z represented by the feature map.

[0083] The feature map 225 output from the first activation layer 220 is input to the second convolution layer 230. The "3X3x4" marked on the second convolution layer 230 indicates, for example, that the convolution on the input feature map is performed by using four filter kernels, each of which has a size of 3x3. The output of the second convolution layer 230 is input to the second activation layer 240. The second activation layer 240 can give nonlinear characteristics to the input data.

[0084] The feature map output by the second activation layer 240 is input to the third convolutional layer 250 . Figure 3 The '3x3x1' marked on the third convolutional layer 250 shown indicates that, for example, a filter kernel of size 3x3 is used to perform convolution to obtain an output. The third convolutional layer 250 is used to output the reconstructed CSI data. A layer that uses a filter kernel to generate an output.

[0085] although Figure 2 The AI ​​model 200 for decoding is illustrated, including three convolutional layers 210, 230 and 250 and two activation layers 220 and 240, but it is only an example, and the number of convolutional layers and the number of activation layers may vary depending on the implementation. In addition, the AI ​​model 200 for decoding may include other layers in addition to the convolutional layers, such as a fully connected layer, a pooling layer, etc. In addition, depending on the implementation, the AI ​​model 200 for decoding may be implemented by a recursive neural network (RNN), in which case the CNN structure of the AI ​​model 200 for decoding according to the present disclosure is changed to an RNN structure.

[0086] As combined Figure 2As described, the AI ​​model for encoding (not shown) may have a symmetrical structure with the AI ​​model for decoding 200. However, it is merely an example, and at least one of the number of layers, layer types, or parameters between the AI ​​model for encoding (not shown) and the AI ​​model for decoding 200 may be modified according to implementation.

[0087] Figure 3 is a flowchart for describing a method of performing CSI profiling according to an embodiment.

[0088] In operation S310, the BS 120 may transmit a CSI-RS to the UE 110. The CSI-RS may be classified into a periodic CSI-RS, a semi-persistent CSI-RS, and an aperiodic CSI-RS by a transmission method. The BS 120 may set a method of transmitting the CSI-RS to the UE 110 in a high-layer signaling (e.g., a radio resource control (RRC) signaling), a MAC-CE signaling, or an L1 signaling, or a combination of two or more of the above signalings.

[0089] In operation S320, UE 110 may perform compression on CSI data estimated based on CSI-RS. UE 110 may estimate a channel between UE and BS based on CSI-RS received from BS 120. The channel estimated by UE may be expressed as channel matrix UE 100 can obtain the channel matrix by using SVD or EVD as described above. V is obtained from . V is now described as CSI data.

[0090] The AI ​​model for encoding jointly trained with the AI ​​model for decoding may be run in the UE 110. The UE 110 may receive NN configuration information of the AI ​​model for encoding from the BS 120 or an external server. For example, the UE 110 may receive NN configuration information through high-layer signaling, and configure parameters for the AI ​​model for encoding based on the received NN configuration information. The UE 110 may obtain compressed CSI data by inputting the CSI data into the AI ​​model for encoding.

[0091] In operation S330, the UE 110 may transmit the compressed CSI data to the BS 120. The UE 110 may transmit the compressed CSI data to the BS 120 periodically, semi-persistently, or aperiodically.

[0092] In operation S340, UE 110 may send CSI data before compression to BS 120 based on a preset condition. In an embodiment, UE 110 may send CSI data before compression to BS 120 based on a request of BS 120. In another embodiment, UE 110 may send CSI data before compression to BS 120 based on a preset interval. The interval for sending CSI data may be set in high-layer signaling, MAC-CE signaling, or L1 signaling, or a combination of two or more signalings. For example, UE 110 may receive estimation configuration information including information about channel periodicity from BS through high-layer signaling.

[0093] In another embodiment, when a change in the channel state is detected or a degradation in decoding performance is detected, UE 110 may send CSI data to BS 120. For example, UE 110 may send CSI data corresponding to M compressed CSI data most recently sent to the BS to BS 120. In this case, the value of M may be set for UE 110 according to higher layer signaling, MAC-CE or L1 signaling, or a combination of two or more signalings.

[0094] In another embodiment, when a change in the channel state is detected or a degradation in decoding performance is detected, UE 110 may send a request to BS 120 to update the AI ​​model for encoding, which triggers CSI analysis. In response to BS 120 receiving a request to update the AI ​​model for encoding from UE 110, UE 110 may send CSI data to BS 120 via a physical uplink shared channel (PUSCH) or a physical uplink control channel (PUCCH).

[0095] In operation S350, the BS 120 may determine to update the AI ​​model for encoding. The BS 120 may compare the CSI data received from the UE 110 with the CSI data reconstructed based on the compressed CSI data from the UE 110 by using the AI ​​model for decoding.

[0096] When the difference between the reconstructed CSI data and the received CSI data is equal to or greater than a preset threshold, the BS 120 may identify that the compressed CSI data has an error. In the case of identifying that the compressed CSI data has an error, the BS 120 may determine to update the AI ​​model used for encoding.

[0097] In operation S360 , the BS 120 may transmit information for updating the AI ​​model for encoding to the UE 110 .

[0098] In an embodiment, BS120 may request UE 110 to change the AI ​​model used for encoding using information for updating the AI ​​model used for encoding. BS120 may store NN configuration information for each AI model used for encoding. The NN configuration information for optimizing the compression and reconstruction performance of CSI data may differ according to channel characteristics. Therefore, BS120 may request UE 110 to change the AI ​​model used for encoding to improve the compression and reconstruction performance of CSI data. In this case, the request to change the AI ​​model used for encoding may include the NN configuration information of the AI ​​model used for encoding to be changed or an indicator for indicating the NN configuration information.

[0099] In another embodiment, the BS 120 may instruct the UE 110 to train the AI ​​model for encoding using the information for updating the AI ​​model for encoding. The instruction to train the AI ​​model for encoding may include an instruction to perform weighted training on specific CSI data.

[0100] In another embodiment, when it is determined that there is an error in a specific portion of the compressed CSI data, BS120 may indicate an increase in the number of layers in a specific area in which the error-related CSI data or a portion of the error-related CSI data is input for the AI ​​model for encoding. For example, when the mean square error (MSE) received in a specific subband is equal to or greater than a certain level, it may be indicated to increase the number of layers in the area related to the specific subband of the AI ​​model for encoding. To this end, BS120 may send an indicator to UE 110 to indicate the CSI data related to the error. By indicating the CSI data related to the error, fine-tuning can be performed by updating some layers or some parameters of the AI ​​model for encoding associated with the indication.

[0101] In another embodiment, BS 120 may train the AI ​​model for encoding and the AI ​​model for decoding so that the error between the received CSI data and the reconstructed CSI data is less than a preset threshold. When receiving CSI data of each of the multiple compressed CSI data from UE 110, BS 120 may train the AI ​​model for encoding and the AI ​​model for decoding based on the reception. BS 120 may send NN configuration information of the updated AI model for encoding as a training result to UE 110.

[0102] Figure 4 is a flowchart describing a method of performing CSI profiling at the request of a BS according to an embodiment.

[0103] In operation S410 , the BS 120 may transmit a CSI-RS to the UE 110 .

[0104] In operation S420, the UE 110 may perform compression on the CSI data estimated based on the CSI-RS. The UE 110 may obtain a channel matrix representing a channel between the UE and the BS based on the received CSI-RS. The CSI data can be obtained from the channel matrix by using SVD or EVD. V obtained in. UE 110 can obtain compressed CSI data by inputting the CSI data into the AI ​​model for encoding.

[0105] In operation S430 , the UE 110 may transmit the compressed CSI data to the BS 120 .

[0106] In operation S440, the UE 110 may transmit a UL signal to the BS 120. The UE 110 may transmit the UL signal in resources allocated from the BS.

[0107] In operation S450, BS120 may detect a change in channel state or degradation of decoding performance based on a UL signal. The UL signal may be sent from the UE to the BS on a PUSCH or a PUCCH. In an embodiment, when the decoding performance of the data received by the UL signal fails to meet a preset threshold, the BS120 may determine that the decoding performance has degraded. For example, when the bit error rate (BER) of the received data is equal to or greater than a threshold, the BS120 may determine that the decoding performance has degraded. In another embodiment, the BS120 may detect a change in channel state as a result of receiving a UL signal. For example, the BS120 may detect a change in channel state by estimating the UL channel based on a sounding reference signal (SRS) received from the UE 110. The change in channel state may refer to a change in at least one property of Doppler spread, Doppler shift, average delay, or delay spread, but this is only an example and the channel properties are not limited thereto.

[0108] In operation S460, BS 120 may send a request for CSI data to UE 110. When BS 120 receives a plurality of compressed CSI data from UE 110, the request for CSI data may include an indicator for specifying CSI data corresponding to at least one of the plurality of compressed CSI numbers. The indicator may include, for example, N bits indicating a value of the number of CSI data. This means that the most recently transmitted 2 N +1 compressed CSI data All CSI data or some CSI data (1~2 N +1). In another example, the indicator may include a timestamp that may specify the CSI data. In this way, the CSI data corresponding to the compressed CSI data reported at the time marked by the timestamp may be specified. In another example, the indicator may include an identifier that may specify the CSI data.

[0109] In addition, the request for CSI data may include an indicator for specifying a portion of the CSI data. The CSI data may be a matrix whose dimensions are determined according to the number of BS antennas and the number of subbands. The indicator may include information specifying at least one of the rows or columns of the matrix.

[0110] In operation S470, UE 110 may transmit the requested CSI data to BS 120. UE 110 may identify at least one of the plurality of CSI data based on the indicator included in the request for CSI data. In addition, UE 110 may identify a portion of the CSI data based on the indicator included in the request for CSI data. UE 110 may transmit the CSI data to BS 120 based on the identification.

[0111] In operation S480, BS120 may determine to update the AI ​​model for encoding based on the received CSI data. When the difference between the reconstructed CSI data and the received CSI data is equal to or greater than a preset threshold, BS120 may identify that there is an error in the compressed CSI data. In the case of identifying that there is an error in the compressed CSI data, BS120 may determine to update the AI ​​model for encoding.

[0112] In operation S490, BS120 may send information for updating the AI ​​model for encoding to UE 110. In an embodiment, BS120 may request UE 110 to change the AI ​​model for encoding. In another embodiment, BS120 may instruct UE 110 to train the AI ​​model for encoding using information for updating the AI ​​model for encoding. In another embodiment, when it is determined that there is an error in a specific part of the compressed CSI data, BS120 may indicate to increase the number of layers in a specific area in which the CSI data related to the error or the part of the CSI data related to the error is input for the AI ​​model for encoding. In another embodiment, BS120 may train the AI ​​model for encoding and the AI ​​model for decoding through a training device in BS120 so that the error between the received CSI data and the reconstructed CSI data is less than a preset threshold. BS120 may send NN configuration information for updating the AI ​​model for encoding as a training result to UE 110.

[0113] Figure 5a is a diagram for describing a method of specifying a portion of CSI data according to an embodiment.

[0114] Reference Figure 5a, the CSI data may be a matrix V consisting of Nt x K complex numbers, where Nt may be the number of antennas of the BS, and K may be the number of subcarriers or subbands.

[0115] The BS may transmit an indicator 'Partition V' to the UE, which may specify a portion of V. In the case where the partition V is deactivated, the entire matrix V consisting of Nt×K complex numbers may be transmitted. In the case where the partition V is activated and includes information about a partial range of Nt and a partial range of K, the UE may transmit a portion of V corresponding to the information to the BS.

[0116] For example, for V including 8 x 13 complex numbers, data transmission of certain subbands may be indicated in the form of indices. In the case of receiving indexes 4, 5, 6, and 7 of K through partition V, the UE may send a portion 510 corresponding to indexes 4, 5, 6, and 7 of K in the complex numbers constituting V to the BS.

[0117] Figure 5a is a diagram for describing a method of specifying a portion of CSI data according to an embodiment.

[0118] Reference Figure 5a , the CSI data may be a matrix V consisting of Nt x K complex numbers, where Nt may be the number of antennas of the BS, and K may be the number of subcarriers or subbands.

[0119] The BS may transmit an indicator 'Partition V' to the UE, which may specify a portion of V. In the case where the partition V is deactivated, the entire matrix V consisting of Nt×K complex numbers may be transmitted. In the case where the partition V is activated and includes information about a partial range of Nt and a partial range of K, the UE may transmit a portion of V corresponding to the information to the BS.

[0120] For example, for V including 8 x 13 complex numbers, data transmission of certain subbands may be indicated in the form of indices. In the case of receiving indexes 4, 5, 6, and 7 of K through partition V, the UE may send a portion 510 corresponding to indexes 4, 5, 6, and 7 of K in the complex numbers constituting V to the BS.

[0121] Figure 5b is a diagram for describing a method of specifying a portion of CSI data according to an embodiment.

[0122] Reference Figure 5b , the CSI data may be a matrix V consisting of Nt x K complex numbers, where Nt may be the number of antennas of the BS, and K may be the number of subcarriers or subbands.

[0123] The BS may send an indicator 'Partition V' to the UE, which may specify a portion of V. For example, for V including 8 x 13 complex numbers, data transmission of certain subbands may be indicated in the form of indices. In the case of receiving indexes 3 and 4 of Nt through partition V, the UE may send a portion 520 corresponding to indexes 3 and 4 of Nt in the complex numbers constituting V to the BS.

[0124] However, it is only an example, and the partition V may include an index in the form of coordinates. For example, in the case where the partition V includes an index (3, 4), a complex number in which Nt is equal to 3 and K is equal to 4 may be transmitted to the BS.

[0125] Figure 6 is a flow chart describing the operation of a BS for CIS profiling according to an embodiment.

[0126] In operation S605, the BS may monitor a change in decoding performance or a channel of a received signal.

[0127] In operation S610, the BS may determine whether CSI analysis is required according to the monitoring result.

[0128] In operation S615, the BS may maintain the AI ​​model for encoding. When the decoding performance of the received signal satisfies a preset threshold or the channel state change is within a preset range, the BS may determine that CSI analysis is not required. The BS may maintain the AI ​​model for encoding.

[0129] In operation S620, the BS may determine whether part of the CSI data is needed. When the decoding performance of the received signal does not meet the preset threshold or the channel state changes beyond the preset range, the BS may determine that CSI analysis is needed. The BS may determine whether part of the CSI data is needed for CSI analysis.

[0130] In operation S625, when it is determined that part of the CSI data is not needed, the BS may send an indicator specifying the CSI data to the UE. When the BS 120 receives a plurality of compressed CSI data from the UE, it may send an indicator to the UE to specify CSI data corresponding to at least one compressed CSI data among the plurality of compressed CSI data.

[0131] In operation S630, the BS may receive a portion of the CSI data corresponding to the indicator.

[0132] In operation S635, when determining that a portion of the CSI data is required, the BS may send an indicator specifying the portion of the CSI data to the UE. For example, when determining that a channel state of a specific subband or antenna has changed, the BS may determine that a portion of the CSI data corresponding to the subband or antenna is required.

[0133] In operation S640, the BS may receive a portion of the CSI data corresponding to the indicator.

[0134] In operation S645 , the BS may determine to update the AI ​​model used for encoding based on the received CSI data or part of the received CSI data.

[0135] In operation S650, the BS may transmit information for updating an AI model for encoding to the UE.

[0136] Figure 7 is a flowchart for describing the operation of a UE for CSI profiling according to an embodiment.

[0137] In operation S710, the UE may transmit compressed CSI data to the BS. The UE may transmit compressed CSI data to the BS periodically, semi-persistently, or aperiodically according to a setting.

[0138] In operation S720, the UE may identify whether a request for CSI data is received from the BS.

[0139] In operation S730, when a request for CSI data is not received from the BS, the UE may maintain the AI ​​model for encoding.

[0140] In operation S740, when receiving the request for CSI data from the BS, the UE may determine whether an indicator indicating a portion of the CSI data is received. For example, the UE may determine whether the request for CSI data includes a partition V indicator.

[0141] In operation S750, when the indicator indicating the part of the CSI data is not received, the UE may send the CSI data to the BS. The request for the CSI data may include specifying an index or a timestamp of at least one of the plurality of CSI data or an identifier of the CSI data. The UE may send the CSI data corresponding to the index or the timestamp or the identifier of the CSI data to the BS.

[0142] In operation S760, when receiving an indicator indicating a portion of CSI data, the UE may send the portion of CSI data. In the request for CSI data, a partition V indicator may be activated. In addition, the partition V indicator may include an index for specifying at least some of the complex numbers constituting V. The index may be an integer value, an integer range, or a coordinate consisting of integer values ​​to specify at least one of each row or column of V. The UE may send the portion of CSI data corresponding to the partition V indicator to the BS.

[0143] In operation S770 , the UE may receive information for updating an AI model for encoding.

[0144] In operation S780, the UE may update the AI ​​model used for encoding based on the received information.

[0145] Figure 8 is a flowchart for describing a method of performing CSI profiling at preset intervals according to an embodiment.

[0146] In operation S810, BS120 may send channel estimation configuration information to UE110. Channel estimation configuration information is information for CSI profiling, which includes information about a method of sending CSI data to the BS. For example, it may include information indicating that the transmission of CSI data is periodic, semi-persistent, or aperiodic. Further, when the transmission of CSI data is periodic, semi-persistent, or aperiodic, the channel estimation configuration information may include information about an interval for sending CSI data. When the transmission of CSI data is semi-persistent and activated by a medium access control (MAC) control unit (CE), CSI data may be sent according to a set interval.

[0147] Furthermore, in an embodiment, channel estimation configuration information may be configured for UE 110 during an initial access or handover process.

[0148] In operation S820 , the BS 120 may transmit a CSI-RS to the UE 110 .

[0149] In operation S830, the UE 110 may perform compression on the CSI data estimated based on the CSI-RS. The UE 110 may obtain a channel matrix representing a channel between the UE and the BS based on the received CSI-RS. The CSI data can be obtained from the channel matrix by using SVD or EVD V obtained in. UE 110 can obtain compressed CSI data by inputting the CSI data into the AI ​​model for encoding.

[0150] In operation S840 , the UE 110 may transmit the compressed CSI data to the BS 120 .

[0151] In operation S850, the UE 110 may determine to send the CSI data based on the interval setting according to the channel estimation configuration information. For example, when the transmission of the CSI data is set to be periodic according to the channel estimation configuration information, the UE 110 may determine to send the CSI data at a set interval. For example, when the interval is set to 4 ms, the UE 110 may determine to send the CSI data to the BS 120 at an interval of 4 ms. However, this is merely an example, and the interval may be set based on various time units such as a time slot, a symbol, etc.

[0152] In another example, when the transmission of CSI data is set to semi-persistent according to the channel estimation configuration information, the UE 110 may identify whether the transmission of CSI data is activated through the MAC CE. When the transmission of CSI data is activated through the MAC CE, the UE 110 may determine to send the CSI data to the BS 120 at the interval set by the channel estimation configuration information.

[0153] The interval set for UE 110 may be reset through higher layer signaling, MAC CE, or L1 signaling.

[0154] In operation S860 , the UE 110 may transmit CSI data to the BS 120 .

[0155] In operation S870, BS 120 may determine to update the AI ​​model for encoding based on the received CSI data. Operation S870 may be similar to the above-mentioned Figure 3 The described operation S350 corresponds.

[0156] In operation S880, BS 120 may send information for updating the AI ​​model for encoding to UE 110. Operation S880 may be similar to the above-mentioned Figure 3 The described operation S360 corresponds.

[0157] Fig. 9 is a flowchart for describing a method of performing CSI profiling according to a request of a UE according to an embodiment.

[0158] In operation S905 , the BS 120 may transmit a CSI-RS to the UE 110 .

[0159] In operation S910, the UE 110 may perform compression on the CSI data estimated based on the CSI-RS. The UE 110 may obtain a channel matrix representing a channel between the UE and the BS based on the received CSI-RS. The CSI data can be obtained from the channel matrix by using SVD or EVD V obtained in. UE 110 can obtain compressed CSI data by inputting the CSI data into the AI ​​model for encoding.

[0160] In operation S915 , the UE 110 may transmit the compressed CSI data to the BS 120 .

[0161] In operation S920, the BS 120 may transmit a DL signal to the UE 110. The DL signal 110 may be transmitted to the UE 110 through the PDSCH or the PDCCH.

[0162] In operation S925, UE 110 may detect a change in channel state or degradation of decoding performance based on a DL signal. In an embodiment, when the decoding performance of data received via a DL signal fails to meet a preset threshold, UE 110 may determine that the decoding performance has degraded. For example, when the BER of the received data is equal to or greater than a threshold, UE 110 may determine that the decoding performance has degraded. In another embodiment, as a result of receiving a DL signal, UE 110 may detect a change in channel state. For example, UE 110 may detect a change in channel state via a demodulation reference signal (DMRS). A change in channel state may refer to a change in at least one property of Doppler spread, Doppler shift, average delay, or delay spread, but this is only an example and channel properties are not limited thereto.

[0163] In operation S930, UE 110 may request BS 120 to update the AI ​​model for encoding. When it is determined that CSI profiling of BS 120 is required, UE 110 may request BS 120 to update the AI ​​model for encoding. For example, UE 110 may determine that CSI profiling is required when the decoding performance of the received signal does not meet a preset threshold or the change of the channel state exceeds a preset range.

[0164] In addition, the request to update the AI ​​model for encoding may include a cause value for the update request. For example, the occasion when the decoding performance of the received signal does not meet the preset threshold can be defined as event Ax, and the occasion when the change of the channel state exceeds the preset range can be defined as event Ay. The update request for the AI ​​model for encoding may include an indicator indicating that event Ax is the cause value or an indicator indicating that event Ay is the cause value. This is just an example, and the cause value that may be included in the update request of the AI ​​model is not limited to the above examples.

[0165] In operation S935, BS 120 may request CSI data from UE 110. Upon receiving the update request for the AI ​​model for encoding, BS 120 may determine the CSI data to be requested from UE 110. For example, when the update request for the AI ​​model for encoding includes an indicator indicating that event Ax has occurred, BS 120 may request a portion of CSI data associated with decoding performance degradation. In another example, when the update request for the AI ​​model for encoding includes an indicator indicating that event Ay has occurred, BS 120 may request the most recent 2 n +1 or some CSI data.

[0166] In operation S940, UE 110 may transmit CSI data to BS 120. UE 110 may transmit CSI data requested from BS 120. For example, UE 110 may transmit a portion of the requested CSI data based on the request of BS 120. In another example, UE 110 may transmit the most recent 2 CSI data based on the request of BS 120. n +1 CSI data.

[0167] In operation S945, BS 120 may determine to update the AI ​​model for encoding. Operation S945 may be similar to the above-mentioned Figure 3 The described operation S350 corresponds.

[0168] In operation S950, BS 120 may send information for updating the AI ​​model for encoding to UE 110. Operation S950 may be similar to the above-mentioned Figure 3 The described operation S360 corresponds.

[0169] Fig.10 is a flowchart for describing a method of performing CSI profiling at preset intervals according to an embodiment.

[0170] In operation S1010, the BS may receive compressed CSI data and CSI data before compression. The BS may receive compressed CSI data from the UE. In addition, the BS may receive CSI data before compression. In an embodiment, the UE may send CSI data before compression to the BS based on a request of the BS. In another embodiment, the UE may send CSI data before compression to the BS based on a preset interval. In another embodiment, the UE may send CSI data to the BS when a change in channel state or degradation of decoding performance is detected.

[0171] In operation S1020, the BS may identify whether the error has been resolved after the CSI analysis and retraining process. For example, the BS may identify whether the difference between the reconstructed CSI data and the received CSI data is equal to or greater than a preset threshold. When the difference between the reconstructed CSI data and the received CSI data is equal to or greater than the preset threshold, the BS may identify that there is an error in the compressed CSI data. Since the BS identifies that there is an error in the compressed CSI data, the BS may train an AI model for encoding based on at least one CSI data received from the UE. By training the AI ​​model for encoding, the AI ​​model for encoding may be retrained based on at least one CSI data.

[0172] In operation S1030, the BS may send a signal to the UE indicating that the error is resolved. When it is determined that the error is resolved as a result of training the AI ​​model for encoding, the BS may send a signal with a parsing result IE value of 0 to the UE to notify that the error is resolved. For example, when the difference between the reconstructed CSI data and the received CSI data is less than a preset threshold as a result of training the AI ​​model for encoding, the BS may determine that the error is resolved.

[0173] In operation S1040, the BS may send an indicator indicating the CSI data associated with the error to the UE. The UE may configure parameters for the region of the AI ​​model for encoding to which the CSI data specified by the indicator is applied so that a high weight is reflected on the CSI data. In addition, the UE may trigger online training of the AI ​​model for encoding based on the specified CSI data. For example, the UE may trigger online training of the AI ​​model for encoding based on the specified CSI data to obtain z using which CSI data similar to the input CSI data is reconstructed.

[0174] Fig.11 is a block diagram of a UE according to an embodiment of the present disclosure; and

[0175] Reference Fig.11 , UE 1100 may include a transceiver 1110, a processor 1120, and a memory 1130. The transceiver 1110, the processor 1120, and the memory 1130 of UE 1100 may operate according to the communication method of UE 1100 described above. However, the elements of UE 1100 are not limited thereto. For example, UE 1100 may include more or less elements than described above. In addition, the transceiver 1110, the processor 1120, and the memory 1130 may be implemented in a single chip. The processor 1120 may include one or more processors.

[0176] The receiver and transmitter of UE 1100 are collectively referred to as a transceiver 1110, which can send or receive signals to a BS or a network entity. The signals sent to or received from a BS or a network entity may include control information and data. To this end, the transceiver 1110 may include an RF transmitter for up-converting the frequency of a signal to be sent and amplifying the signal and an RF receiver for low-noise amplifying a received signal and down-converting the frequency of the received signal. It is only an example of the transceiver 1110, and the elements of the transceiver 1110 are not limited to the RF transmitter and the RF receiver.

[0177] The transceiver 1110 may perform functions of transmitting and receiving signals on a wireless channel. For example, the transceiver 1110 may receive a signal on a wireless channel and output it to the processor 1120, and transmit a signal output from the processor 1120 on a wireless channel.

[0178] The memory 1130 may store programs and data required for the operation of the UE 1100. Further, the memory 1130 may store control information or data included in the signal obtained by the BS. The memory 1130 may include a storage medium, such as a read-only memory (ROM), a random access memory (RAM), a hard disk, a compact disk ROM (CD-ROM) and a digital versatile disk (DVD), or a combination of storage media. Alternatively, the memory 1130 may not exist alone, but may be integrated in the processor 1120. The memory 1130 may include a volatile memory, a non-volatile memory, or a combination of a volatile memory and a non-volatile memory. The memory 1130 may also provide stored data according to a request of the processor 1120.

[0179] The processor 1120 can control a series of processes of the UE 1100 according to the above-mentioned embodiments of the present disclosure. For example, the processor 1120 can receive a control signal and a data signal through the transceiver 1110, and process the received control signal and data signal. The processor 1120 can send the processed control signal and data signal through the transceiver 1110. The processor 1120 can record data in the memory 1130, or read data from the memory 1330. The processor 1120 can perform the functions of the protocol stack requested by the communication standard. To this end, the processor 1120 may include at least one processor or microprocessor. In the embodiments of the present disclosure, a part of the transceiver 1110 and the processor 1120 may be referred to as a CP.

[0180] The processor 1120 may include one or more processors. The one or more processors may include a general-purpose processor such as a central processing unit (CPU), an application processor (AP), a digital signal processor (DSP), a graphics processing unit (GPU), a visual processing unit (VPU), etc., or a dedicated artificial intelligence (AI) processor such as a neural processing unit (NPU). For example, when one or more processors are dedicated AI processors, the dedicated AI processor may be designed in a hardware structure dedicated to processing a specific AI model.

[0181] In an embodiment, the processor 1120 may be configured to compress the estimated CSI data between the BS and the UE 1100 through an AI model for encoding jointly trained with an AI model for decoding of the base station. The processor 1120 may send the compressed CSI data to the BS. The processor 1120 may send the CSI data before compression to the BS based on a preset condition. When the BS determines to update the AI ​​model for encoding based on the CSI data, the processor 1120 may receive information for updating the AI ​​model for encoding.

[0182] Fig.12is a block diagram of a BS according to an embodiment.

[0183] Reference Fig.12 , BS1200 may include a transceiver 1210, a processor 1220, and a memory 1230. The transceiver 1210, the processor 1220, and the memory 1230 of BS1200 may operate according to the communication method of BS1200 described above. However, the elements of BS1200 are not limited thereto. For example, the base station 1200 may include more or less elements than those described above. In addition, the transceiver 1210, the processor 1220, and the memory 1230 may be implemented in a single chip. The processor 1220 may include one or more processors.

[0184] The receiver and transmitter of BS1200 are collectively referred to as transceiver 1210, which can send or receive signals to UE or network entity. The signals sent to or received from UE or network entity may include control information and data. To this end, transceiver 1210 may include an RF transmitter for up-converting the frequency of the signal to be sent and amplifying the signal and an RF receiver for low-noise amplification of the received signal and down-converting the frequency of the received signal. It is only an example of transceiver 1210, and the elements of transceiver 1210 are not limited to RF transmitter and RF receiver.

[0185] The transceiver 1210 may perform functions of transmitting and receiving signals on a wireless channel. For example, the transceiver 1210 may receive a signal on a wireless channel and output it to the processor 1220, and transmit a signal output from the processor 1220 on a wireless channel.

[0186] The memory 1230 may store programs and data required for the operation of the base station 1200. Further, the memory 1230 may store control information or data included in the signal obtained by the base station 1200. The memory 1230 may include a storage medium such as ROM, RAM, hard disk, CD-ROM, DVD, or a combination of storage media. Alternatively, the memory 1230 may not exist alone, but may be integrated in the processor 1220. The memory 1230 may include a volatile memory, a non-volatile memory, or a combination of a volatile memory and a non-volatile memory. The memory 1230 may also provide the stored data according to the request of the processor 1220.

[0187] The processor 1220 can control a series of processes in which the BS 1200 operates according to an embodiment of the present disclosure. For example, the processor 1220 can receive a control signal and a data signal through the transceiver 1210, and process the received control signal and data signal. The processor 1220 can send the processed control signal and data signal through the transceiver 1210. The processor 1220 can record data in the memory 1230, or read data from the memory 1330. The processor 1220 can perform the functions of the protocol stack requested by the communication standard. To this end, the processor 1220 may include at least one processor or microprocessor. In an embodiment of the present disclosure, a part of the transceiver 1210 and the processor 1220 may be referred to as a CP.

[0188] The processor 1220 may include one or more processors. The one or more processors may include a general-purpose processor such as a central processing unit (CPU), an application processor (AP), a digital signal processor (DSP), a graphics processing unit (GPU), a visual processing unit (VPU), or a dedicated artificial intelligence (AI) processor such as a neural processing unit (NPU). For example, when one or more processors are dedicated AI processors, the dedicated AI processor may be designed in a hardware structure dedicated to processing a specific AI model.

[0189] In an embodiment, the processor 1220 may receive CSI data compressed by an AI model for encoding from a UE, the AI ​​model for encoding being jointly trained with an AI model for decoding of a BS. Further, the processor 1220 may receive CSI data before compression from a UE based on a preset condition. The processor 1220 may determine to update the AI ​​model for encoding based on the CSI data. Further, the processor 1220 may send information for updating the AI ​​model for encoding.

[0190] The AI-related functions according to the embodiments of the present disclosure are operated by a processor and a memory. There may be one or more processors. One or more processors may include general-purpose processors such as a central processing unit (CPU), an application processor (AP), a digital signal processor (DSP), a graphics processing unit (GPU), a visual processing unit (VPU), etc., or a dedicated artificial intelligence (AI) processor such as a neural processing unit (NPU). One or more processors may control the processing of input data according to predefined operating rules or AI models stored in the memory. When one or more processors are dedicated AI processors, they may be designed in a hardware structure dedicated to processing a specific AI model.

[0191] Predefined operating rules or AI models can be formulated through learning. Specifically, predefined operating rules or AI models formulated through learning refer to predefined operating rules or AI models established to execute desired features (or objects) when a basic AI model (or deep learning model) is trained using a large amount of training data through a learning algorithm. This learning can be performed by the device itself that executes AI according to the present disclosure, or by a separate server and / or system. Examples of learning algorithms may include supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, but are not limited to this.

[0192] The AI ​​model (or deep learning model) may be composed of multiple neural network layers. Each of the multiple neural network layers may have multiple weight values, and perform neural network operations through operations between the operation results of the previous layer and the multiple weight values. The multiple weight values ​​possessed by the multiple neural network layers may be optimized by the learning results of the AI ​​model. For example, multiple weight values ​​may be updated during the training process to reduce or minimize the loss value or cost value obtained by the AI ​​model. The artificial neural network may include a deep neural network (DNN), such as a convolutional neural network (CNN), a recursive neural network (RNN), a restricted Boltzmann machine (RBM), a deep belief network (DBN), a bidirectional recursive deep neural network (BRDNN), or a deep Q network, but is not limited thereto.

[0193] The specified examples used to describe the embodiments of the present disclosure are merely a combination of one of the standards, methods, detailed methods, and operations, and through a combination of two or more of the above schemes, the BS and the UE can update the autoencoder for obtaining compressed CSI data through CSI analysis. The CSI analysis can be performed according to a method determined by a combination of one, two, or more of the above schemes. For example, a part of the operations in an embodiment can be performed in combination with a part of the operations in another embodiment.

[0194] The machine-readable storage medium may be provided in the form of a non-transitory storage medium. The term "non-transitory storage medium" may refer to a tangible device that does not include signals such as electromagnetic waves, and may not distinguish between semi-permanent storage of data and temporary storage of data in the storage medium. For example, a non-transitory storage medium may include a cache that temporarily stores data.

[0195] In an embodiment of the present disclosure, the above-mentioned methods according to various embodiments of the present disclosure may be provided in a computer program product. The computer program product may be a commercial product that can be traded between a seller and a buyer. The computer program product may be distributed in the form of a machine-readable storage medium (e.g., a CD-ROM), or distributed directly between two user devices (e.g., a smartphone) or distributed online (e.g., downloaded or uploaded). In the case of online distribution, at least a portion of the computer program product (e.g., a downloadable application) may be at least temporarily stored or arbitrarily created in a storage medium that may be readable by a device such as a manufacturer's server, an application store's server, or a relay server.

[0196] The machine-readable storage medium may be provided in the form of a non-transitory storage medium. The term "non-transitory storage medium" may refer to a tangible device that does not include signals such as electromagnetic waves, and may not distinguish between semi-permanent storage of data and temporary storage of data in the storage medium. For example, a non-transitory storage medium may include a cache that temporarily stores data.

[0197] In the embodiments of the present disclosure, the above methods according to various embodiments of the present disclosure may be provided in a computer program product. The computer program product may be a commercial product that can be traded between a seller and a buyer. The computer program product may be distributed in the form of a machine-readable storage medium (e.g., a CD-ROM), downloaded through an application store (e.g., a play store), or downloaded from a computer program store. TM ), directly between two user devices (e.g., smartphones), or online (e.g., downloading or uploading). In the case of online distribution, at least a portion of the computer program product (e.g., a downloadable application) may be at least temporarily stored or arbitrarily created in a storage medium that may be readable by a device such as a manufacturer's server, an application store's server, or a relay server.

Claims

1. A method for reporting channel state information (CSI) of a user equipment (UE) in a wireless communication system, the method comprising: Compressing the estimated CSI data between a base station BS and the UE by an artificial intelligence AI model for encoding, wherein the AI ​​model for encoding is jointly trained with an AI model for decoding of the BS; sending compressed CSI data to the BS; Based on a preset condition, sending the CSI data before compression to the BS; as well as When the BS determines to update the AI ​​model for encoding based on the CSI data, information for updating the AI ​​model for encoding is received.

2. A method for reporting channel state information (CSI) in a wireless communication system by a base station (BS), the method comprising: receiving, from the UE, CSI data compressed by an AI model for encoding, the AI ​​model for encoding being jointly trained with an AI model for decoding of the BS; Based on a preset condition, receiving CSI data before compression from the UE; Determining to update the AI ​​model for encoding based on the CSI data; and Send information for updating the AI ​​model for encoding.

3. A user equipment UE for reporting channel state information CSI in a wireless communication system, the UE comprising: Transceiver; as well as A processor is connected to the transceiver and is configured to: Compressing the estimated CSI data between a base station BS and the UE by an artificial intelligence AI model for encoding, wherein the AI ​​model for encoding is jointly trained with an AI model for decoding of the BS; sending compressed CSI data to the BS; Based on a preset condition, sending the CSI data before compression to the BS; as well as When the BS determines to update the AI ​​model for encoding based on the CSI data, information for updating the AI ​​model for encoding is received.

4. The UE according to claim 3, wherein: The processor is configured to: When the decoding performance of the uplink (UL) signal received by the BS does not meet a preset threshold or a change in the channel state is detected based on the UL signal, a request for the CSI data is received from the BS.

5. The UE according to claim 4, wherein: The CSI data includes a matrix representing a channel state, The request for the CSI data includes at least one of the following: the number of matrices that the BS requests to transmit; or a partial indicator that specifies at least some of the matrices that the BS requests to transmit.

6. The UE according to claim 3, wherein: The processor is configured to: receiving channel estimation configuration information from the BS through high-layer signaling, and The CSI data is transmitted based on an interval set according to the channel estimation configuration information.

7. The UE according to claim 3, wherein: The processor is configured to: When the decoding performance of the downlink DL signal received by the UE does not meet a preset threshold or a change in the channel state is detected based on the DL signal, sending a request for updating the AI ​​model for encoding to the BS; as well as A request for the CSI data is received from the BS, the request for the CSI data being based on the request for updating the AI ​​model for encoding.

8. The UE according to claim 3, wherein: The information for updating the AI ​​model for encoding includes the neural network NN configuration information of the AI ​​model for encoding updated in the BS, and Among them, the NN configuration information includes: information related to at least one of the type of layers, the number of layers, the number of filter cores, the size of the filter cores, or the parameters of the filter cores included in the updated AI model for encoding.

9. The UE according to claim 3, wherein: When the decoding performance of the UL signal received by the BS does not meet a preset threshold, the compressed CSI is identified as having an error, and The information for updating the AI ​​model for encoding includes: an indicator indicating CSI data associated with the error among a plurality of CSI data transmitted to the BS, or an indicator indicating a portion of the CSI data associated with the error.

10. A base station BS for reporting channel state information CSI in a wireless communication system, the BS comprising: Transceiver; as well as A processor is connected to the transceiver and is configured to: Receiving, from a user equipment UE, CSI compressed by an AI model for encoding, wherein the AI ​​model for encoding is jointly trained with an AI model for decoding of the BS; Based on a preset condition, receiving CSI data before compression from the UE; Determining to update the AI ​​model for encoding based on the CSI data; and Send information for updating the AI ​​model for encoding.

11. The BS according to claim 10, wherein: The processor is configured to: When the decoding performance of the uplink UL signal received by the BS does not meet a preset threshold or a change in the channel state is detected based on the UL signal, a request for the CSI data is sent to the UE.

12. The BS according to claim 10, wherein: The processor is configured to: Sending channel estimation configuration information to the UE via high-layer signaling, and The CSI data is received based on an interval set according to the channel estimation configuration information.

13. The BS according to claim 10, wherein: The processor is configured to: When the decoding performance of the downlink DL signal received by the UE does not meet a preset threshold or a change in the channel state is detected based on the DL signal, receiving a request for updating the AI ​​model for encoding from the UE; and Based on the request for updating the AI ​​model for encoding, a request for the CSI data is sent to the UE.

14. The BS according to claim 10, wherein: The information for updating the AI ​​model for encoding includes the neural network NN configuration information of the AI ​​model for encoding updated in the BS, and Among them, the NN configuration information includes: information related to at least one of the type of layers, the number of layers, the number of filter kernels, the size of the filter kernels, or the parameters of the filter kernels included in the updated AI model for encoding.

15. The BS according to claim 10, wherein: The processor is configured to: When the decoding performance of the UL signal received by the BS does not meet a preset threshold, it is identified that there is an error in the compressed CSI, and The information used to update the AI ​​model for encoding includes: an indicator indicating CSI data associated with the error among multiple CSI data sent to the BS, or an indicator indicating a portion of the CSI data associated with the error.

Images