Artificial intelligence-based communication method and apparatus in wireless communication system

Abstract

A method for operation of a network entity, according to an embodiment of the present disclosure, comprises the steps of: receiving, from a user equipment, information indicating the capability of the user equipment, the information indicating the capability of the user equipment including information indicating a data type that may be processed by the user equipment; determining the data type on the basis of the information indicating the capability of the user equipment and a communication condition; and transmitting, to the user equipment, data type information indicating the determined data type.

Description

Artificial intelligence-based communication method and device in a wireless communication system

[0001] The present disclosure relates to a communication method and device in a wireless communication system, and relates to a technology that can effectively perform communication using an artificial intelligence model in a network entity (e.g., a base station, etc.) and a terminal.

[0002] Looking back at the evolution of wireless communication over successive generations, technologies have primarily been developed for human-facing services such as voice, multimedia, and data. With the commercialization of the 5G (5th Generation) communication system, an explosive increase in connected devices is expected to be connected to communication networks. Examples of networked objects include vehicles, robots, drones, home appliances, displays, smart sensors installed in various infrastructures, construction equipment, and factory equipment. Mobile devices are also expected to evolve into diverse form factors, such as augmented reality glasses, virtual reality headsets, and holographic devices. In the 6G (6th Generation) era, efforts are being made to develop improved 6G communication systems to connect hundreds of billions of devices and objects and provide diverse services. For this reason, 6G communication systems are often referred to as "beyond 5G."

[0003] The 6G communication system, expected to be realized around 2030, will have a maximum transmission speed of terabytes (i.e., 1,000 gigabits) per second (bps) and a wireless latency of 100 microseconds (μsec). In other words, compared to 5G, the transmission speed in a 6G communication system will be 50 times faster and the wireless latency will be reduced to one-tenth.

[0004] To achieve these high data rates and ultra-low latency, 6G communication systems are being considered for implementation in the terahertz (THz) band (e.g., from 95 gigahertz (GHz) to 3 terahertz (THz)). Compared to the millimeter wave (mmWave) band introduced in 5G, the terahertz band is expected to have more severe path loss and atmospheric absorption, making it more important to develop technologies that can guarantee signal reach, or coverage. Key technologies to ensure coverage include Radio Frequency (RF) components, antennas, new waveforms that offer better coverage than Orthogonal Frequency Division Multiplexing (OFDM), beamforming, and multiple antenna transmission technologies such as massive Multiple-Input and Multiple-Output (MIMO), Full Dimensional MIMO (FD-MIMO), array antennas, and large-scale antennas. In addition, new technologies such as metamaterial-based lenses and antennas, high-dimensional spatial multiplexing using Orbital Angular Momentum (OAM), and Reconfigurable Intelligent Surface (RIS) are being discussed to improve the coverage of terahertz band signals.

[0005] In addition, in order to improve frequency efficiency and system network, 6G communication systems are developing full duplex technology that utilizes the same frequency resources at the same time for uplink and downlink; network technology that integrates satellites and HAPS (High-Altitude Platform Stations); network structure innovation technology that supports mobile base stations and enables optimization and automation of network operation; dynamic spectrum sharing technology through collision avoidance based on spectrum usage prediction; AI-based communication technology that utilizes AI (Artificial Intelligence) from the design stage and internalizes end-to-end AI support functions to realize system optimization; and next-generation distributed computing technology that realizes services with complexity that exceeds the limits of terminal computing capabilities by utilizing ultra-high-performance communication and computing resources (Mobile Edge Computing (MEC), cloud, etc.). In addition, efforts are being made to further strengthen connectivity between devices, further optimize networks, promote softwareization of network entities, and increase the openness of wireless communications through the design of new protocols to be used in 6G communication systems, the implementation of hardware-based security environments, the development of mechanisms for the safe use of data, and the development of technologies for maintaining privacy.

[0006] Research and development of these 6G communication systems are expected to enable a new level of hyper-connected experience through the hyper-connectivity of 6G communication systems, which encompass not only connections between things but also connections between people and things. Specifically, 6G communication systems are expected to enable services such as truly immersive eXtended Reality (XR), high-fidelity mobile holograms, and digital replicas. Furthermore, services such as remote surgery, industrial automation, and emergency response, which are provided through 6G communication systems through enhanced security and reliability, will be applied in diverse fields such as industry, medicine, automobiles, and home appliances.

[0007] The present disclosure may have as its primary purpose a method and device for reducing overhead associated with using an artificial intelligence model in communication systems such as 5G, 5G-Advanced, and 6G, and improving communication performance between networks and terminals.

[0008] A method performed by a network entity in a wireless communication system, comprising: a step of receiving information indicating a capability of a terminal from a terminal, wherein the information indicating the capability of the terminal includes information indicating a type of data that can be processed by the terminal; a step of determining a data type based on the information indicating the capability of the terminal and a communication condition; and a step of transmitting data type information indicating the determined data type to the terminal.

[0009] A method performed by a terminal in a wireless communication system, comprising: a step of transmitting information indicating a capability of the terminal to a network entity, wherein the information indicating the capability of the terminal includes information indicating a data type that can be processed by the terminal; a step of receiving data type information indicating a data type determined based on the information indicating the capability of the terminal and a communication condition; a step of generating an output value of an artificial intelligence model based on the determined data type; and a step of transmitting the generated output value to the network entity.

[0010] A method and device according to one embodiment of the present disclosure can provide an effective artificial intelligence model-based transmission and reception operation in a wireless communication system.

[0011] Specifically, the method and device according to one embodiment of the present disclosure can perform efficient data processing according to network communication conditions by performing a quantization operation of an artificial intelligence model or using a quantized artificial intelligence model.

[0012] The effects that can be obtained from the present disclosure are not limited to the effects mentioned above, and other effects that are not mentioned can be clearly understood by a person having ordinary skill in the art to which the present disclosure belongs from the description below.

[0013] The features and advantages of one embodiment of the present disclosure will become more apparent from the following description taken in conjunction with the accompanying drawings.

[0014] FIG. 1 illustrates a wireless communication system according to one embodiment of the present disclosure.

[0015] Figure 2 is a drawing for explaining the structure of a terminal according to one embodiment.

[0016] FIG. 3 is a diagram illustrating the structure of a network entity according to one embodiment.

[0017] FIG. 4 is a diagram for explaining an artificial intelligence model-based communication method according to one embodiment of the present disclosure.

[0018] FIG. 5 is a diagram for explaining an artificial intelligence model according to one embodiment of the present disclosure.

[0019] FIG. 6 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0020] FIG. 7 illustrates information for indicating an artificial intelligence model quantization capability of a user equipment (UE) according to one embodiment of the present disclosure.

[0021] FIG. 8A and FIG. 8B illustrate information for indicating an artificial intelligence model-related capability of a user equipment (UE) according to one embodiment of the present disclosure.

[0022] Figure 9 shows setting information related to an artificial intelligence model according to one embodiment of the present disclosure.

[0023] FIG. 10 illustrates setting information regarding reporting of a terminal according to one embodiment of the present disclosure.

[0024] FIG. 11 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0025] FIG. 12 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0026] FIG. 13 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0027] FIG. 14 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0028] FIG. 15 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0029] FIG. 16 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0030] FIG. 17 illustrates information for a data type change request according to one embodiment of the present disclosure.

[0031] FIG. 18 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0032] FIG. 19 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0033] FIG. 20 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0034] FIG. 21 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0035] FIG. 22 is a diagram for explaining the operation of a network entity according to one embodiment of the present disclosure.

[0036] FIG. 23 is a diagram for explaining the operation of a user terminal according to one embodiment of the present disclosure.

[0037] Embodiments of the present disclosure may address the problems and / or disadvantages described above and provide the advantages described below. One aspect of the present disclosure may provide a network entity (or node) and a communication method thereof in a wireless communication system.

[0038] The terms used in this disclosure are used only to describe specific embodiments and may not be intended to limit the scope of other embodiments. The singular expression may include plural expressions unless the context clearly indicates otherwise. Terms used herein, including technical or scientific terms, may have the same meaning as commonly understood by those of ordinary skill in the art described in this disclosure. Terms defined in general dictionaries among the terms used in this disclosure may be interpreted as having the same or similar meaning in the context of the relevant technology, and shall not be interpreted in an idealized or overly formal sense unless explicitly defined in this disclosure. In some cases, even if a term is defined in this disclosure, it cannot be interpreted to exclude embodiments of the present disclosure.

[0039] The various embodiments of the present disclosure described below illustrate hardware-based approaches. However, since the various embodiments of the present disclosure encompass techniques utilizing both hardware and software, the various embodiments of the present disclosure do not exclude software-based approaches.

[0040] Additionally, although various embodiments of the present disclosure describe various embodiments using terminology used in certain communication standards (e.g., 3rd generation partnership project (3GPP)), this is merely an example for illustrative purposes. Various embodiments of the present disclosure can be easily modified and applied to other communication systems.

[0041] Hereinafter, various embodiments of the present disclosure will be described.

[0042] Meanwhile, the core of AI-based beam management is that the base station performs SSB (synchronization signal and physical broadcast channel block) transmission using some beams from the entire beam grid pattern. The UE can feed back the reference signals received power (RSRP) value of the received SSB beam to the base station. The base station can perform beam prediction using the RSRP values ​​received from the terminal as inputs to the AI ​​model. The base station can predict the RSRP values ​​corresponding to the untransmitted beams through beam prediction. The base station can sweep the beams corresponding to the top k RSRP values ​​among the RSRP values ​​of the entire beam grid obtained through beam prediction (k is a hyper-parameter related to beam operation). The terminal can report to the base station the beam corresponding to the largest RSRP value among the k received beams.

[0043] For AI-based beam management, beam prediction can be performed using an AI model at the base station using RSRP values ​​reported from the terminal as input. This type of inference performed by a single entity is referred to as "one-sided model inference." The input to the base station AI model can be the RSRP values ​​reported from the terminal.

[0044] When seeking to improve the performance of AI-based beam management (e.g., reducing beam sweeping overhead or improving beam prediction accuracy), leveraging only one-sided model inference is unlikely to yield significant performance gains. In particular, the hardware complexity of mobile communication systems is limited, making it difficult to increase the size of AI models or the amount of data processed indiscriminately. Consequently, operating beam management solely with a one-sided model presents limitations.

[0045] A method and device according to one embodiment of the present disclosure are a method and device for performing beam operation by applying two-sided model inference to reduce overhead and improve beam prediction accuracy compared to beam operation based on one-sided model inference.

[0046] In general, UEs can utilize raw values ​​for reporting information and additional UE-oriented information. For example, signal-related information such as L1-RSRP, received signal strength indicator (RSSI), and reference signal received quality (RSRQ); UE-side beam information such as Rx beam shape / direction, Rx beam angle, Rx beam-width, and Rx beam boresight; UE position information such as UE location and UE moving direction; and channel information such as historical CIR, SINRs, and CQIs. Since this information is available to the UE, performing UE-side model inference using this information can improve beam management efficiency. In the case of the existing one-sided model inference based on the NW-side model, since inference is performed only on the NW-side, UE-oriented information and raw data cannot be used for AI inference unless the UE reports all information. Therefore, we plan to utilize two-sided model inference that performs inference on both the UE-side and the NW-side.

[0047] FIG. 1 illustrates a wireless communication system according to one embodiment of the present disclosure.

[0048] FIG. 1 illustrates some of the nodes utilizing a wireless channel in a wireless communication system, including a base station (110), a first terminal (120), and / or a second terminal (130). Although FIG. 1 illustrates only one base station, this is merely an example. The wireless communication system of FIG. 1 may further include other base stations identical or similar to the base station (110).

[0049] The base station (110) is a network infrastructure that provides wireless access to terminals (120, 130). The base station (110) has coverage defined as a certain geographical area based on the distance at which a signal can be transmitted. In addition to the base station, the base station (110) may be referred to as an 'access point (AP)', 'evolved Node B (eNB)', 'next generation node B (gNB)', '5G node (5th generation node)', 'wireless point', 'transmission / reception point (TRP)', or other terms having equivalent technical meanings.

[0050] The first terminal (120) and the second terminal (130) are each devices used by a user and can communicate with the base station (110) via a wireless channel. At least one of the first terminal (120) or the second terminal (130) can be operated without the user's intervention. For example, at least one of the first terminal (120) or the second terminal (130) may be a device that performs machine type communication (MTC) and may not be carried by the user. Each of the first terminal (120) and the second terminal (130) may be referred to as a terminal, or other terms having equivalent technical meanings, such as 'user equipment (UE),' 'mobile station,' 'subscriber station,' 'customer premises equipment (CPE),' 'remote terminal,' 'wireless terminal,' 'electronic device,' or 'user device.'

[0051] The base station (110), the first terminal (120), and the second terminal (130) can transmit and / or receive wireless signals in the millimeter wave (mmWave) band (e.g., 28 GHz, 30 GHz, 38 GHz, 60 GHz). At this time, in order to improve channel gain, the base station (110), the first terminal (120), and / or the second terminal (130) can perform beamforming.

[0052] Beamforming may include transmit beamforming and / or receive beamforming. That is, the base station (110), the first terminal (120), and / or the second terminal (130) may impart directionality to the transmit signal or the receive signal. To impart directionality to the receive signal, the base station (110) and / or the terminals (120, 130) may select serving beams (112, 113, 121, 131) through a beam search or beam management procedure. After the serving beams (112, 113, 121, 131) are selected, subsequent communication may be performed through resources that are in a quasi-co-located (QCL) relationship with the resources that transmitted the serving beams (112, 113, 121, 131).

[0053] The base station (110), the first terminal (120), and the second terminal (130) of the present disclosure may each be a transmitting apparatus, a transmitting node, a receiving apparatus, and / or a receiving node. For example, the base station (110) may transmit an RF (radio frequency) signal to the first terminal (120). The base station (110) may receive the RF signal from the first terminal (120). As another example, the first terminal (120) may transmit an RF signal to the base station (110) or the second terminal (130). The first terminal (120) may receive the RF signal from the base station (110) or the second terminal (130).

[0054] Figure 2 is a drawing for explaining the structure of a terminal according to one embodiment.

[0055] Referring to FIG. 2, a terminal (200) according to one embodiment may include a transceiver (transmitting and receiving unit) (210), a memory (220), and / or a processor (230). Although the terminal (200) is described in the present disclosure as including the transceiver (210), the memory (220), and / or the processor (230), this is merely an example. For example, the terminal (200) may further include other components in addition to the transceiver (210), the memory (220), and the processor (230).

[0056] According to one embodiment, the transceiver (210), memory (220), and processor (230) may be implemented or formed as separate chips. However, this is merely an example, and the transceiver (210), memory (220), and / or processor (230) may be implemented or formed as a single chip.

[0057] According to one embodiment, the transceiver (210) may include at least one transmitter and / or at least one receiver. For example, the transceiver (210) may include an RF transmitter for amplifying and up-converting the frequency of a transmitted signal. The transceiver (210) may include an RF receiver for down-converting the frequency of a received signal and amplifying low-noise.

[0058] The configurations of the transceiver (210) described in the present disclosure are merely examples, and the configuration of the transceiver (210) is not limited to an RF transmitter and an RF receiver. For example, the transceiver (210) may further include a coupler to ensure isolation between the RF transmitter and the RF receiver.

[0059] In one embodiment, the transceiver (210) may transmit or receive signals to the processor (230). For example, the transceiver (210) may transmit or deliver an RF signal received via a wireless communication channel to the processor (230). The transceiver (210) may receive or deliver an RF signal from the processor (230).

[0060] In one embodiment, the transceiver (210) may be referred to as a UE transmitter or a UE receiver.

[0061] According to one embodiment, the transceiver (210) may transmit signals to or receive signals from a base station (e.g., base station (110) of FIG. 1) or a network entity (e.g., access and mobility management function (AMF) entity). In one embodiment, the transmitted or received signals may include control signals and data.

[0062] According to one embodiment, the memory (220) may include or store programs and data necessary for the operations of the terminal (200). For example, the memory (220) may be a non-transitory memory, and a program stored in the non-transitory memory may be organically combined with a hardware configuration of the terminal (200) (e.g., a processor (230) or a transceiver (210)). The memory (220) may store control information or data including a signal acquired by the terminal (200). In one embodiment, the memory (220) may include a read-only memory (ROM), a random access memory (RAM), a hard disk, a CD-ROM, a DVD, and / or a storage medium.

[0063] According to one embodiment, the processor (230) may include one processor or multiple processors. For example, the processor (230) may include a communication processor. For example, the processor (230) may include a communication processor and / or an application processor.

[0064] In one embodiment, the processor (230) may control a series of processes performed by the terminal (200). For example, the transceiver (210) may receive a data signal containing control information transmitted by a base station or network entity. The processor (230) may process the received control signal and data signal.

[0065] The term "processor" in the present disclosure may be replaced with various terms referring to a configuration that executes or performs operations of the terminal (200). For example, the term "processor" may be replaced with a controller or a computing circuit.

[0066] The terminal (200) of the present disclosure may correspond to the first terminal (120) and / or the second terminal (130) of FIG. 1.

[0067] FIG. 3 is a diagram illustrating the structure of a network entity according to one embodiment.

[0068] Referring to FIG. 3, a network entity (300) according to one embodiment may include a transceiver (transmitter / receiver) (310), a memory (320), and / or a processor (330). Although the network entity (300) is described in the present disclosure as including the transceiver (310), the memory (320), and / or the processor (330), this is merely an example. For example, the network entity (300) may further include other components in addition to the transceiver (310), the memory (320), and the processor (330). The network entity (300) may represent network functions included in a base station or other core network.

[0069] According to one embodiment, the transceiver (310), memory (320), and processor (330) may be implemented or formed as separate chips. However, this is merely an example, and the transceiver (310), memory (320), and / or processor (330) may be implemented or formed as a single chip.

[0070] According to one embodiment, the transceiver (310) may include at least one transmitter and / or at least one receiver. For example, the transceiver (310) may include an RF transmitter for amplifying and up-converting the frequency of a transmitted signal. The transceiver (310) may include an RF receiver for down-converting the frequency of a received signal and amplifying low-noise.

[0071] The configurations of the transceiver (310) described in the present disclosure are merely examples, and the configuration of the transceiver (310) is not limited to an RF transmitter and an RF receiver. For example, the transceiver (310) may further include a coupler to ensure isolation between the RF transmitter and the RF receiver.

[0072] In one embodiment, the transceiver (310) may transmit or receive signals to the processor (330). For example, the transceiver (310) may transmit or deliver an RF signal received via a wireless communication channel to the processor (330). The transceiver (310) may receive or deliver an RF signal from the processor (230).

[0073] According to one embodiment, the transceiver (310) may be referred to as a network entity transmitter or a network entity receiver.

[0074] In one embodiment, the transceiver (310) may transmit a signal to the terminal (200) or receive a signal from the terminal (200). In one embodiment, the transmitted or received signal may include a control signal and data.

[0075] According to one embodiment, the memory (320) may include programs and data necessary for the operations of the network entity (300). For example, the memory (320) may be a non-transitory memory, and the programs stored in the non-transitory memory may be organically combined with the hardware configuration of the network entity (300) (e.g., the processor (330) or the transceiver (310)). The memory (320) may store control information or data including signals acquired by the network entity (300). In one embodiment, the memory (320) may include a read-only memory (ROM), a random access memory (RAM), a hard disk, a CD-ROM, a DVD, and / or storage media.

[0076] According to one embodiment, the processor (330) may include one processor or multiple processors. For example, the processor (330) may include a communication processor. For example, the processor (330) may include a communication processor and / or an application processor.

[0077] In one embodiment, the processor (330) may control a series of processes performed by the network entity (300). For example, the transceiver (310) may receive a data signal including control information transmitted by the network entity. The processor (330) may process the received control signal and data signal.

[0078] The term "processor" in the present disclosure may be replaced with various terms referring to a configuration that executes or performs operations of a network entity (300). For example, the term "processor" may be replaced with a controller or a computing unit.

[0079] The network entity (300) of the present disclosure may correspond to the base station (110) of FIG. 1.

[0080] The devices described in FIGS. 2 and 3 may correspond to devices of a transmitter or receiver. A terminal or network entity according to an embodiment of the present disclosure may be a transmitter if it is a transmitter, and may be a receiver if it is a receiver.

[0081] Hereinafter, the transmitter and receiver may each refer to the terminals or network entities described in FIGS. 1 to 3 above. When describing downlink signals, the network entity will be the transmitter and the terminal will be the receiver, and when describing uplink signals, the terminal will be the transmitter and the network entity will be the receiver.

[0082] FIG. 4 is a diagram for explaining an artificial intelligence model-based communication method according to one embodiment of the present disclosure.

[0083] Referring to FIG. 4, an artificial intelligence model-based communication method (hereinafter, referred to as “communication method”) according to an embodiment of the present disclosure can be distinguished into ONE-SIDED MODEL and TWO-SIDED MODEL depending on whether a network entity (e.g., a base station) or a user equipment (UE) uses an artificial intelligence model. For example, when a base station (BS) and a user equipment (UE) each use an artificial intelligence model during a transmission and reception process between the BS and the UE, the method is referred to as TWO-SIDED MODEL. When only one of the BS and the UE uses an artificial intelligence model, the method is referred to as ONE-SIDED MODEL. When only the UE (420) uses an artificial intelligence model (ONE-SIDED MODEL), the UE (420) can transmit output information of the artificial intelligence model to the BS (410). When both the base station (410) and the terminal (420) use an artificial intelligence model (TWO-SIDED MODEL), the base station (410) or the terminal (420) can each receive output information of the other's artificial intelligence model and input it into its own artificial intelligence model. The communication method according to the embodiments of the present disclosure can use the artificial intelligence model for various cases such as CSI (channel state indicator) prediction, CSI compression, and beam prediction based on the ONE-SIDED MODEL or TWO-SIDED MODEL.

[0084] The communication method according to the embodiments of the present disclosure relates to a technology that can improve the efficiency of the artificial intelligence-based communication method described above, and can achieve the purpose of the technology based on quantization of an artificial intelligence model.

[0085] FIG. 5 is a diagram for explaining an artificial intelligence model according to one embodiment of the present disclosure.

[0086] Referring to FIG. 5, the structure of an artificial intelligence model (500) according to one embodiment is illustrated. The artificial intelligence model (500) can apply weights (520) and / or biases (530) to input information (510) to produce output information (540). Alternatively, the artificial intelligence model (500) can apply an activation function to the input information (510) to produce output information (540). The input information (510) can be used to produce output information (540) based on weights, biases, or activation functions in an operation layer (layer) (550). The produced output information can then be used as input information in another layer to produce further output information. The artificial intelligence model (500) can include multiple layers that produce output information using input information, and various weights, biases, or activation functions can be applied to each layer. The artificial intelligence model (500) can be delivered or transferred between the transmitter and receiver through its structure information and weight information.

[0087] According to embodiments of the present disclosure, the type of data processed by an AI model in the process of calculating input information and generating output information may be set. For example, the data type processed by the AI ​​model may be an integer or a real number. An integer data type may be expressed as "int," and a real number data type may be expressed as "float," "double," or "long double." The AI ​​model may perform calculations by applying specific data types to numbers such as input information, weight information, bias information, and / or output information required in the calculation process. A user terminal or network entity may modify the AI ​​model so that the AI ​​model performs calculations by expressing data in a specific data type during the calculation process. This modification of the AI ​​model to perform calculations with a specific data type may be referred to as "quantization of the AI ​​model." If a network entity or a user terminal has AI model quantization capabilities, the network entity or user terminal may modify its AI model to perform calculations by applying a specific data type. If a network entity or a user terminal does not have the capability of quantizing an artificial intelligence model, the network entity can receive a quantized model from the user terminal and the user terminal can receive a quantized model from the network entity and operate the quantized artificial intelligence model.

[0088] The integer data type is a data type for representing integers, and the real number data type is a data type for representing real numbers. There may be various representations for representing integers or real numbers. For example, the integer or real number data types can be classified in various ways based on the number of bits that the data type uses to represent a single number. If the first data type is a method for representing integers with 16 bits, the first data type can be represented as "INT16". Also, if the second data type is a method for representing integers with 8 bits, the second data type can be represented as "INT8". If the third data type is a method for representing real numbers with 32 bits, the third data type can be represented as "FP32". If the fourth data type is a method for representing real numbers with 16 bits, the fourth data type can be represented as "FP16".

[0089] In the embodiments of the present disclosure, the types of data types and the manner of data types are not limited to the examples described above, and the embodiments can be applied to various known data types.

[0090] User terminals or network entities can reduce the computational burden of AI models by converting the data types they process. For example, an AI model may experience less computational burden when processing "FP16" data types than when processing "FP32" data types. However, the accuracy of the predicted results output by the AI ​​model may be better when using "FP32" data types.

[0091] FIG. 6 is a drawing for explaining a communication method (hereinafter, “communication method”) according to one embodiment of the present disclosure.

[0092] Referring to FIG. 6, transmission and reception operations between a network entity (610) and a user equipment (UE) (620) are illustrated. The network entity (610) of FIG. 6 may correspond to the network entity of FIG. 3. The UE (620) of FIG. 6 may correspond to the terminal of FIG. 2.

[0093] Below, the operation of the communication method of Fig. 6 is described.

[0094] Operation 630 represents an operation in which a network entity (e.g., a base station) (610) receives terminal capability information indicating the capability of the terminal from the terminal. The terminal capability information includes the AI ​​model quantization capability of the terminal of FIG. 7 and the NW entity model data type information of FIG. 8, and may indicate at least one of information on whether the terminal uses an AI model, whether the terminal can quantize an AI model, or a data type that can be processed by the terminal-side AI model. In addition, the terminal capability information may indicate information on the data type that the terminal can process by quantizing an AI model. The terminal capability information includes AI model information for operation in a network entity owned by the terminal. This information includes quantized AI model information for each use case (whether a model exists for each quantization type). In addition, it may also include model performance (e.g., accuracy) information for each quantized model.

[0095] In one embodiment, terminal capability information may include information about factors that may impact the performance of an AI model. For example, terminal capability information may include information about the total and available storage capacity, the total and available memory capacity used for computation, and so on.

[0096] Terminal capability information regarding the artificial intelligence model is exemplified in FIGS. 7 and 8.

[0097] Operation 632 represents an operation in which the network entity (610) determines the data type to be applied to the operation of the artificial intelligence model based on terminal capability information and communication conditions. The network entity (610) can determine whether the terminal can use the artificial intelligence model through the terminal capability information, information on the data types that the terminal-side artificial intelligence model can process, and / or whether the terminal's artificial intelligence model can be quantized, and thus can determine an appropriate data type accordingly. In addition, the network entity (610) can determine the data type based on the communication conditions. At this time, the communication conditions may include at least one of an accuracy condition, a latency condition, and / or a data throughput condition for transmitted and received information.

[0098] For example, according to a communication method according to one embodiment of the present disclosure, an artificial intelligence model of a network entity or a terminal may apply a data type of FP32 type when predicting or compressing a channel state indicator (CSI) based on a threshold value of an average data throughput (or total data throughput). Alternatively, under URLLC (Ultra reliable low latency communication) (latency < 1 ms, Reliability < 10^(-5)) communication conditions, the artificial intelligence model of a network entity or a terminal may apply a data type of FP32 type when predicting or compressing a channel state indicator (CSI). In addition, a specific data type may be applied to the operation of the artificial intelligence model for various purposes, such as beam prediction and / or positioning.

[0099] Operation 634 represents an operation of transmitting configuration information from the network entity (610) to the user terminal (620). The configuration information may include information indicating the data type determined in operation 632. The configuration information may represent a data type (or AI model-related settings) applied to the artificial intelligence model of the terminal (620) and / or an output (or result) report data type (or AI model output report settings) of the artificial intelligence model of the terminal (620). The output report data type of the artificial intelligence model of the terminal (620) may represent a data type received from the terminal (620) by the network entity (610). The terminal (620) may convert (or quantize) an existing artificial intelligence model into an artificial intelligence model applicable to a specific data type based on the received configuration information, and may perform inference using the converted (or quantized) artificial intelligence model. In addition, the terminal (620) may transmit information output from the artificial intelligence model as a specific data type to the network entity (610) based on the received configuration information. Configuration information may be information that specifies data types for each use case based on an AI model. For example, configuration information may indicate the data types applied to AI models for CSI compression, CSI prediction, beam management, and / or positioning. In other words, configuration information may be information that maps the quantization level for each use case based on an AI model.

[0100] In one embodiment, when the quantization level changes, i.e., when the data type to be applied to the artificial intelligence model changes, the network entity (610) may indicate the changed data type to the terminal (620) using a low layer message. (The low layer message may be in the form of a MAC CE or DCI message.)

[0101] In one embodiment, the configuration information may indicate a predefined reporting format for a specific case related to reporting the output (or result) of the AI ​​model. For example, the configuration information may indicate that, in the case of a CSI prediction case, the terminal reports to the network entity in a predefined reporting format.

[0102] The exchange of configuration information between the network entity (610) and the terminal (620) may be performed based on RRC messages. Configuration information related to operation 634 may be exemplified in FIGS. 9 and 10.

[0103] Operation 636 represents an operation in which the terminal (620) converts (or quantizes) an artificial intelligence model based on the received configuration information and outputs a test result using the converted (or quantized) artificial intelligence model. The terminal (620) can convert (or quantize) an existing artificial intelligence model on the terminal side into an artificial intelligence model capable of processing a specific data type based on the received configuration information. For example, if the data type that can be processed by the existing artificial intelligence model on the terminal side is FP32, the terminal can convert the existing artificial intelligence model into an artificial intelligence model capable of processing an INT16 data type based on the received configuration information (if the terminal has a quantization capability).

[0104] Operation 638 may represent an operation in which the terminal (620) performs a test on the artificial intelligence model converted in operation 636 and transmits the test results for information output from the converted artificial intelligence model to the network entity (610). That is, the network entity (610) receives the test results of the output values ​​of the artificial intelligence model based on the data type information from the terminal (620). The terminal (620) may store a test data set for performing the test of the artificial intelligence model. The test results may be a message including the output metrics of the artificial intelligence model. The test results may include results regarding the accuracy, computational speed, and reliability of predicted data.

[0105] Operation 640 represents an operation in which the network entity (610) determines whether to continue applying the configuration information (or data type) based on the test result received in operation 638, and transmits the determination result to the terminal (620). The network entity (610) determines whether the test result (accuracy, speed, reliability, etc.) satisfies the reference condition, and may transmit indication information (e.g., a 1-bit message) (DCI, MAC CE, or RRC message) indicating whether to use the configuration information to the terminal based on the determination result. The indication information indicating whether to use the configuration information may correspond to the indication information indicating whether to use the converted artificial intelligence model (or the quantized artificial intelligence model).

[0106] In one embodiment, if the test result does not satisfy the reference condition, the network entity (610) may change the configuration information (or data type information) and transmit it to the terminal. The terminal may convert the artificial intelligence model based on the changed configuration information and generate a test result again based on the output of the converted artificial intelligence model. Then, the terminal may transmit the test result to the network entity. That is, in the communication method according to one embodiment of the present disclosure, if the test result for the converted artificial intelligence model does not satisfy the reference condition, the network entity may change the configuration information and transmit it to the terminal, and the terminal may test the artificial intelligence model based on the changed configuration information and transmit the result to the network entity. These operations may be repeatedly performed until a test result that satisfies the reference condition is derived.

[0107] Additionally, in one embodiment, if the test result does not satisfy the reference condition, the network entity (610) may transmit information to the terminal indicating that the test result does not satisfy the reference condition, and the terminal may utilize an artificial intelligence model of a previously used data type.

[0108] In one embodiment, if the network entity (610) can directly transmit the artificial intelligence model to the terminal (620), operations 636 to 640 may be omitted.

[0109] Action 642 is an example of a prediction or management-related action based on an artificial intelligence model (e.g., a CSI prediction action).

[0110] The 6421 operation represents an operation of transmitting a CSI-RS (reference signal) from a network entity (610) to a terminal (620).

[0111] Action 6422 represents an action in which the terminal (620) predicts CSI using an artificial intelligence model based on the received CSI-RS.

[0112] Action 6423 represents an action in which the terminal (620) transmits the prediction result in action 6422 to the network entity (610).

[0113] Actions 6421 and 6423 illustrate CSI prediction actions, but various other actions utilizing artificial intelligence models, such as CSI compression, beam prediction, and / or positioning, may be applied.

[0114] FIG. 7 illustrates information for indicating an artificial intelligence model quantization capability of a user equipment (UE) according to one embodiment of the present disclosure.

[0115] The information illustrated in FIG. 7 may correspond to terminal capability information related to operation 630 of FIG. 6 . The terminal capability information may indicate at least one of the following: whether the terminal uses an AI model, whether the terminal can quantize the AI ​​model, or information regarding the types of data that the terminal-side AI model can process. Furthermore, the terminal capability information may indicate information regarding the types of data that the terminal can process by quantizing the AI ​​model.

[0116] Referring to FIG. 7, index 0 indicates that the terminal does not have the quantization capability of the AI ​​model. Index 1 indicates that the terminal can quantize the AI ​​model and apply the FP32 data type. Index 2 indicates that the terminal can quantize the AI ​​model and apply the FP16 data type. Index 3 indicates that the terminal can quantize the AI ​​model and apply the INT32 data type. Index 4 indicates that the terminal can quantize the AI ​​model and apply the INT8 data type. Index 5 indicates that the terminal can quantize the AI ​​model and apply the INT4 data type. As the capabilities of the terminals are distinguished, the number of indexes and the number of information bits for indicating the terminal capabilities may increase.

[0117] In one embodiment, the terminal capability information may indicate whether the terminal can use an artificial intelligence model, whether the terminal can quantize the artificial intelligence model if the terminal can use the artificial intelligence model, and / or, if the terminal can quantize the artificial intelligence model, information on the data types applicable to the quantized artificial intelligence model.

[0118] FIG. 8a illustrates information for indicating an artificial intelligence model-related capability of a user equipment (UE) according to one embodiment of the present disclosure.

[0119] The information shown in FIG. 8a may correspond to the setting information related to operation 630 of FIG. 6.

[0120] Referring to FIG. 8A, terminal capability information may include data type information corresponding to an AI-based operation case. The AI-based operation cases may include CSI compression, CSI prediction, beam operation, and / or positioning. In addition, data type information may be mapped to each case. For example, the terminal capability information may indicate that an AI model based on the FP16 data type is used in an AI-based CSI compression operation, and an AI model based on the INT16 data type is used in an AI-based CSI prediction operation. In addition, the terminal capability information may further include performance information regarding the corresponding data type for each case. For example, the terminal capability information may include information indicating the performance of a first data type corresponding to a first case when applied to an AI model. The performance information may include accuracy, loss, etc. Based on the aforementioned terminal capability information, the network entity may determine the data type for a specific case, quantize the network entity-side AI model, or request the AI ​​model from the terminal.

[0121] FIG. 8B illustrates information indicating AI model-related capabilities of a user equipment (UE) according to one embodiment of the present disclosure. Specifically, FIG. 8B may indicate information regarding an AI model that can be transmitted from a user equipment (UE) to a network entity when the network entity requests an AI model from the UE. In other words, FIG. 8B may indicate information regarding AI models available to the network entity. FIG. 8B may indicate the AI ​​model data type of the network entity.

[0122] The information illustrated in FIG. 8b may correspond to the configuration information related to operation 630 of FIG. 6. That is, the terminal capability information may include the information illustrated in FIG. 8b.

[0123] Referring to FIG. 8B, terminal capability information may include information about AI models available to a network entity, and may include data type information corresponding to AI-based operation cases. AI-based operation cases may include CSI compression, CSI prediction, beam operation, and / or positioning. In addition, data type information may be mapped to each case. For example, the user terminal capability information may indicate that an AI model based on the FP16 data type is used in an AI-based CSI compression operation, and an AI model based on the INT16 data type is used in an AI-based CSI prediction operation. In addition, the user terminal capability information may further include performance information of the AI ​​model for the corresponding data type for each case. For example, the user terminal capability information may include information indicating the performance of a first data type corresponding to a first case when applied to an AI model. The performance information may include accuracy, loss, and the like. The network entity may determine the data type for a specific case, quantize the network entity-side AI model, or request the AI ​​model from the terminal based on the AI ​​model quantization capability of the terminal in FIG. 7, the AI ​​model-related capability of the terminal in FIG. 8a, and / or information about the AI ​​model available to the network entity in FIG. 8b.

[0124] Figure 9 shows setting information related to an artificial intelligence model according to one embodiment of the present disclosure.

[0125] The information illustrated in FIG. 9 may correspond to the setting information related to operation 634 of FIG. 6.

[0126] The configuration information may include information indicating the data type determined in operation 632 of FIG. 6. The configuration information may include the data type (or AI model-related settings) applied to the terminal's artificial intelligence model.

[0127] Referring to Figure 9, “AI-based operation cases” and “data types” and “whether critical data types” are illustrated.

[0128] An "AI-based operation case" can represent a set of operations utilizing an AI model, such as index-based CSI compression, CSI prediction, beam operation, and positioning. Configuration information can indicate the data type corresponding to each case. For example, configuration information can indicate that the AI ​​model applies the FP16 data type for a CSI compression operation, and the AI ​​model applies the INT16 data type for a CSI prediction operation.

[0129] "Whether Critical Data Type" indicates whether the data type indicated by the data type information is a threshold. For example, if the configuration information indicates that the INT8 data type is applied to the AI ​​model (index 3) when the CSI compression operation is in progress (index 0), and "Whether Critical Data Type" is "O" (index 0), the configuration information may indicate that any one of INT8, INT16, FP16, and / or FP32 data types can be applied to the AI ​​model when the CSI compression operation is in progress. Alternatively, conversely, the configuration information may indicate that any one of INT8 and / or INT4 data types can be applied to the AI ​​model when the CSI compression operation is in progress. That is, if "Whether Critical Data Type" is "O", since the data type indicated in the configuration information is a threshold, the configuration information may indicate that the data type corresponding to the index before or after the index of the indicated data type can be applied to the AI ​​model. When the AI ​​model is transmitted directly from the network entity to the terminal, the "Whether Critical Data Type" field may be omitted. "Whether it is a critical data type" can be indicated by a 1-bit field. If "Whether it is a critical data type" is 'X', the terminal can operate the terminal artificial intelligence model with the data type indicated by the index of the 'data type'.

[0130] FIG. 10 illustrates setting information regarding reporting of a user terminal according to one embodiment of the present disclosure.

[0131] The information shown in Fig. 10 may correspond to the setting information related to operation 634 of Fig. 6.

[0132] The configuration information may include the output (or result) reporting data type (or, output reporting settings of the AI ​​model) of the terminal's AI model. That is, the configuration information may indicate data type information for the results output from the terminal-side AI model. When the output value of the terminal-side AI model is used as the input of the network entity-side AI model, the configuration information may set the data type information for the results output from the terminal-side AI model, thereby setting the data type for the input of the network entity-side AI model.

[0133] The output report data type of the terminal-side AI model can indicate the data type received by the network entity from the terminal. Based on the received configuration information, the terminal can transmit information output from the terminal-side AI model in a specific data type to the network entity. The configuration information may be information in which the data type is set for each use case based on the AI ​​model. For example, the configuration information may indicate information on the data types applied to the AI ​​model for each of CSI compression, CSI prediction, beam management, and / or positioning operations.

[0134] Configuration information can specify a predefined reporting format for specific cases related to reporting the output (or results) of an AI model. For example, configuration information can indicate that a terminal reports to a network entity in a predefined reporting format during CSI prediction operations.

[0135] The types of artificial intelligence-based operation cases and types of data types related to terminal capability information or setting information described in FIGS. 7 to 10 are not limited to those described in the drawings, and may include more cases and data types than those depicted in the drawings.

[0136] FIG. 11 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0137] The embodiment described in FIG. 11 features an operation of transmitting an artificial intelligence model from a network entity (1110) to a terminal (1120).

[0138] Referring to FIG. 11, transmission and reception operations between a network entity (1110) and a user equipment (UE) (1120) are illustrated. The network entity (1110) of FIG. 11 may correspond to the network entity of FIG. 3. The UE (1120) of FIG. 11 may correspond to the terminal of FIG. 2.

[0139] Below, the operation of the communication method of Fig. 11 is described.

[0140] The embodiment of FIG. 11 is an embodiment of transmitting an artificial intelligence model from a network entity (1110) to a terminal (1120).

[0141] Operation 1130 represents an operation in which a network entity (e.g., a base station) (1110) receives terminal capability information indicating the capabilities of the terminal from the terminal. The terminal capability information may indicate whether the terminal uses an AI model, whether the terminal can quantize the AI ​​model, and / or information regarding the types of data that the terminal-side AI model can process. Furthermore, the terminal capability information may indicate information regarding the types of data that the terminal can process by quantizing the AI ​​model. Furthermore, the terminal capability information may include data types and model performance information regarding the AI ​​model that the terminal can transmit to the network entity. (See FIG. 8b)

[0142] The terminal capability information regarding the artificial intelligence model can be applied to the contents exemplarily described in FIGS. 7 and 8.

[0143] Operation 1132 represents an operation in which a network entity (1110) determines the data type to be applied to the calculation of an artificial intelligence model based on terminal capability information and communication conditions. Alternatively, operation 1132 may determine a terminal-side artificial intelligence model based on terminal capability information and communication conditions in the network entity (1110).

[0144] The network entity (1110) can determine an appropriate data type based on the terminal capability information, such as whether the terminal can use an artificial intelligence model, information on the type of data that can be processed by the terminal-side artificial intelligence model, whether the terminal's artificial intelligence model can be quantized, and / or information on the type of data and model performance of the artificial intelligence model that the terminal can transmit to the network entity. In addition, the network entity (1110) can determine the data type based on the communication condition. In this case, the communication condition may be an accuracy condition, a latency condition, and / or a data throughput condition for the transmitted and received information. For example, according to a communication method according to an embodiment of the present disclosure, the network entity or the artificial intelligence model of the terminal can apply an FP32 type data type in a specific artificial intelligence-based operation case based on a threshold value of the average data throughput (or, the total data throughput). Alternatively, under URLLC (Ultra reliable low latency communication) (latency < 1ms, Reliability < 10^(-5)) communication conditions, the AI ​​model of a network entity or terminal can apply a data type of FP32 type for CSI (channel state indicator) prediction (as an example). In addition, specific data types can be applied to the operation of the AI ​​model for various purposes such as beam prediction and / or positioning.

[0145] Operation 1134 represents an operation of transmitting configuration information from the network entity (1110) to the user terminal (1120). The configuration information may include information indicating the data type determined in operation 1132. The configuration information may represent a data type (or AI model-related settings) applied to the artificial intelligence model of the terminal (1120) and an output (or result) report data type (or AI model output report settings) of the artificial intelligence model of the terminal (1120). The output report data type of the artificial intelligence model of the terminal (1120) may represent the type of data received from the terminal (1120) by the network entity (1110). The terminal (1120) may receive an artificial intelligence model based on the received configuration information from the network entity (1810) and perform inference using the received artificial intelligence model. In addition, the terminal (1120) may transmit information output from the artificial intelligence model as a specific data type to the network entity (1110) based on the received configuration information. Configuration information may be information that specifies data types for use cases (e.g., CSI prediction) based on AI models. For example, configuration information may indicate data types applied to AI models for CSI compression, CSI prediction, beam management, and / or positioning. In other words, configuration information may be information that maps the quantization level for each use case based on the AI ​​model.

[0146] The exchange of configuration information between a network entity (1110) and a terminal (1120) may be performed based on RRC messages. Configuration information related to operation 1134 may be exemplified in FIGS. 9 and 10.

[0147] Operation 1136 represents an operation of transmitting artificial intelligence model information from a network entity (1110) to a terminal (1120). If the terminal does not have an artificial intelligence model or if the terminal cannot process data of a specific data type by converting an existing artificial intelligence model, the network entity (1110) may transmit an artificial intelligence model capable of processing data of a specific data type to the terminal (1120). For example, if the network entity (1110) determines to apply the FP16 data type to the artificial intelligence model in a CSI prediction operation, the network entity (1110) may transmit an artificial intelligence model capable of processing the FP16 data type to the terminal (1120). At this time, the network entity (1110) may transmit the artificial intelligence model by transmitting structural information and weight information of the artificial intelligence model capable of processing the FP16 data type to the terminal (1120). The terminal (1120) may generate and operate the artificial intelligence model based on the structural information and weight information of the received artificial intelligence model.

[0148] In the embodiment of FIG. 11, since the network entity (1110) transmits the AI ​​model to the terminal (1120), a separate AI model performance test procedure may be omitted. However, if the AI ​​model performance test procedure is performed, the terminal may perform a test on the received AI model and transmit the test results for information output from the received AI model to the network entity. In other words, the network entity may receive the AI ​​model performance test results from the terminal. The test results may include results regarding the accuracy, computation speed, and reliability of predicted data.

[0149] Based on the test results, the network entity can determine whether to continue using the configuration information (or data type) or the AI ​​model transmitted to the terminal, and transmit the determination result to the terminal. The network entity can determine whether the test results (such as accuracy, speed, or reliability) satisfy the reference conditions, and based on the determination result, transmit instruction information (e.g., a 1-bit message) (DCI, MAC CE, or RRC message) indicating whether to use the configuration information or AI model to the terminal.

[0150] In one embodiment, if the test result does not satisfy the reference condition, the network entity can change the configuration information (or data type information) and transmit a different AI model to the terminal. The terminal can additionally generate a test result based on the output of the received AI model, and this test result may be different from the result of a previously performed test. That is, in the communication method according to one embodiment of the present disclosure, if the test result for the AI ​​model received by the terminal does not satisfy the reference condition, the network entity can transmit a different (or updated) AI model to the terminal, and the terminal can test the AI ​​model based on the newly received AI model information and transmit the result to the network entity. These operations can be repeatedly performed until a test result satisfying the reference is derived.

[0151] Action 1138 is an example of a prediction or management-related action based on an artificial intelligence model (e.g., a CSI prediction action).

[0152] Action 11381 represents an action of transmitting a CSI-RS (reference signal) from a network entity (1110) to a terminal (1120).

[0153] Action 11382 represents an action in which the terminal (1120) predicts CSI using an artificial intelligence model based on the received CSI-RS.

[0154] Action 11383 represents an action in which the terminal (1120) transmits the prediction result in action 11382 to the network entity (1110).

[0155] Actions 11381 and 11383 illustrate CSI prediction actions, but actions for various cases such as CSI compression, beam prediction, and / or positioning may be applied.

[0156] FIG. 12 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0157] The embodiment described in FIG. 12 illustrates a case where quantization of an artificial intelligence model is possible in a network entity (1210). That is, the network entity (1210) can convert an existing artificial intelligence model to create an artificial intelligence model capable of processing data of a specific data type.

[0158] Referring to FIG. 12, transmission and reception operations between a network entity (1210) and a user equipment (UE) (1220) are illustrated. The network entity (1210) of FIG. 12 may correspond to the network entity of FIG. 3. The UE (1220) of FIG. 12 may correspond to the terminal of FIG. 2.

[0159] Below, the operation of the communication method of Fig. 12 is described.

[0160] Operation 1230 represents an operation in which a network entity (e.g., a base station) (1210) receives terminal capability information indicating the capabilities of the terminal from the terminal. The terminal capability information may indicate whether the terminal uses an AI model, whether the terminal can quantize the AI ​​model, and / or information regarding the types of data that the terminal-side AI model can process. Furthermore, the terminal capability information may indicate information regarding the types of data that the terminal can process by quantizing the AI ​​model. Furthermore, the terminal capability information may include data types and model performance information regarding the AI ​​model that the terminal can transmit to the network entity. (See FIG. 8b)

[0161] In one embodiment, terminal capability information may include information about factors that may impact the performance of an AI model. For example, terminal capability information may include information about the total and available storage capacity, the total and available memory capacity used for computation, and so on.

[0162] Terminal capability information regarding the artificial intelligence model is exemplified in FIGS. 7 and 8.

[0163] Operation 1232 represents an operation in which a network entity (1210) determines a data type to be applied to the operation of an artificial intelligence model based on terminal capability information and communication conditions. The network entity (1210) can determine an appropriate data type based on the terminal capability information, such as whether the terminal can use the artificial intelligence model, information on the data type that the artificial intelligence model on the terminal side can process, whether the artificial intelligence model of the terminal can be quantized, and / or data type or performance information about the artificial intelligence model that can be transmitted from the terminal to the network entity. In addition, the network entity (1210) can determine the data type based on the communication conditions. In this case, the communication conditions may be accuracy conditions, latency conditions, and / or data throughput conditions for transmitted and received information. For example, according to a communication method according to one embodiment of the present disclosure, an artificial intelligence model of a network entity or a terminal may apply a data type of FP32 type when predicting or compressing a channel state indicator (CSI) based on a threshold value of an average data throughput (or total data throughput). Alternatively, under URLLC (Ultra reliable low latency communication) (latency < 1 ms, Reliability < 10^(-5)) communication conditions, the artificial intelligence model of a network entity or a terminal may apply a data type of FP32 type when predicting or compressing a channel state indicator (CSI). In addition, a specific data type may be applied to the operation of the artificial intelligence model for various purposes, such as beam prediction and / or positioning.

[0164] Operation 1234 represents an operation of transmitting configuration information from a network entity (1210) to a user terminal (1220). The configuration information may include information indicating the data type determined in operation 1232. The exchange of configuration information between the network entity (1210) and the terminal (1220) may be performed based on an RRC message. Configuration information related to operation 1234 may be exemplified in FIGS. 9 and 10 .

[0165] Action 1236 is an example of a prediction or management-related action based on an artificial intelligence model (e.g., a CSI prediction action).

[0166] Action 12361 represents an action of transmitting a CSI-RS (reference signal) from a network entity (1210) to a terminal (1220).

[0167] Operation 12362 represents an operation in which the terminal (1220) transmits information about the CSI-RS received in operation 12361 to the network entity (1210).

[0168] Action 12363 represents an action in which a network entity (1210) predicts CSI using an artificial intelligence model on the network entity side based on information received in action 12362.

[0169] Actions 12361 and 12363 illustrate CSI prediction actions, but actions for various cases such as CSI compression, beam prediction, and / or positioning may be applied.

[0170] FIG. 13 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0171] The embodiment described in FIG. 13 is an embodiment in which a network entity (1310) receives an artificial intelligence model from a terminal (1320). That is, the network entity (1310) requests the terminal (1320) to transmit an artificial intelligence model, receives the requested artificial intelligence model, and performs inference using the artificial intelligence model.

[0172] Referring to FIG. 13, transmission and reception operations between a network entity (1310) and a user equipment (UE) (1320) are illustrated. The network entity (1310) of FIG. 13 may correspond to the network entity of FIG. 3. The UE (1320) of FIG. 13 may correspond to the terminal of FIG. 2.

[0173] Below, the operation of the communication method of Fig. 13 is described.

[0174] Operation 1330 represents an operation in which a network entity (e.g., a base station) (1310) receives terminal capability information indicating the terminal's capabilities from the terminal. The terminal capability information may indicate whether the terminal uses an AI model, whether the terminal can quantize the AI ​​model, and / or information regarding the types of data that the terminal's AI model can process. Furthermore, the terminal capability information may indicate information regarding the types of data that the terminal can process by quantizing the AI ​​model.

[0175] In one embodiment, terminal capability information may include information about factors that may impact the performance of an AI model. For example, terminal capability information may include information about the total and available storage capacity, the total and available memory capacity used for computation, and so on.

[0176] Terminal capability information regarding the artificial intelligence model is exemplified in FIGS. 7 and 8.

[0177] Operation 1332 represents an operation in which the network entity (1310) determines the data type to be applied to the calculation of the artificial intelligence model based on terminal capability information and / or communication conditions. Alternatively, operation 1332 may represent an operation in which the network entity (1310) determines the artificial intelligence model to be requested from the terminal based on terminal capability information and / or communication conditions. The network entity (1310) may determine the artificial intelligence model to be requested from the terminal based on the information described above in FIG. 8B .

[0178] The network entity (1310) can determine whether the terminal can use an artificial intelligence model, whether the terminal-side artificial intelligence model can process data types, and / or whether the terminal's artificial intelligence model can be quantized, etc., through the terminal capability information, and thus can determine an appropriate data type (or an artificial intelligence model capable of processing the corresponding data type). In addition, the network entity (1310) can determine an appropriate data type (or an artificial intelligence model capable of processing the corresponding data type) based on communication conditions. At this time, the communication conditions may be accuracy conditions, latency conditions, and / or data throughput conditions for transmitted and received information.

[0179] An AI model of a network entity or terminal may apply a data type of FP32 type when predicting or compressing a channel state indicator (CSI) based on a threshold value of average data throughput (or total data throughput). Alternatively, under URLLC (Ultra reliable low latency communication) (latency < 1 ms, Reliability < 10^(-5)) communication conditions, the AI ​​model of a network entity or terminal may apply a data type of FP32 type when predicting or compressing a channel state indicator (CSI). In addition, specific data types may be applied to the operation of the AI ​​model for various purposes, such as beam prediction and / or positioning.

[0180] Operation 1334 represents an operation of transmitting configuration information from a network entity (1310) to a user terminal (1320). The configuration information may include information indicating a data type (or AI model) determined in operation 1332. The configuration information may indicate a data type (or AI model-related settings) applied to the AI ​​model. The configuration information may be information in which a data type is set for each use case based on the AI ​​model. For example, the configuration information may indicate information on a data type applied to an AI model for CSI compression, CSI prediction, beam management, and / or positioning. In other words, the configuration information may be information in which a quantization level for each use case based on the AI ​​model is mapped.

[0181] Operation 1336 represents an operation in which a network entity (1310) requests the terminal (1320) to transmit the AI ​​model determined in operation 1332. The request for transmission of the AI ​​model may include requests for multiple AI models. For example, the network entity (1310) may request the terminal to transmit multiple AI models corresponding to multiple use cases.

[0182] Action 1338 represents an action in which a network entity (1310) receives an artificial intelligence model requested in action 1336 from a terminal (1320).

[0183] If a network entity does not have an AI model or if the network entity cannot process data of a specific data type by converting an existing AI model, the network entity (1310) may request the terminal (1320) to receive an AI model capable of processing data of the specific data type. For example, if the network entity (1310) decides to apply the FP16 data type to the AI ​​model in the CSI prediction operation, the network entity (1310) may request the terminal (1320) to receive an AI model capable of processing the FP16 data type. The terminal (1320) may transmit the AI ​​model by transmitting structural information and weight information of the AI ​​model capable of processing the FP16 data type to the network entity (1310). The network entity (1310) may generate and operate the AI ​​model based on the structural information and weight information of the received AI model.

[0184] In the embodiment of FIG. 13, since the network entity (1310) receives the AI ​​model from the terminal (1320), a separate AI model performance testing procedure may be omitted. However, if the AI ​​model performance testing procedure is performed, the network entity may perform a test on the received AI model and generate test results for information output from the received AI model. The test results may include results regarding the accuracy, computational speed, and reliability of predicted data.

[0185] In one embodiment, the network entity (1310) may perform a test on the AI ​​model received in operation 1338. The network entity (1310) may store a test data set for performing the test on the received AI model. In addition, the network entity (1310) may determine whether to continue using the received AI model based on the test result. The network entity (1310) may determine whether the test result (such as accuracy, speed, or reliability) satisfies a reference condition, and if the test result does not satisfy the reference condition, may transmit a request to transmit another AI model to the terminal. In a communication method according to one embodiment of the present disclosure, if the test result for the AI ​​model received by the network entity does not satisfy the reference condition, the network entity may transmit a request to transmit another AI model to the terminal, receive another AI model from the terminal, and perform the test again. These operations may be repeatedly performed until a test result that satisfies the reference condition is derived.

[0186] Action 1340 is an example of a prediction or management-related action based on an artificial intelligence model (e.g., a CSI prediction action).

[0187] Action 13401 represents an action of transmitting a CSI-RS (reference signal) from a network entity (1310) to a terminal (1320).

[0188] Action 13402 represents an action in which a network entity (1310) receives a CSI report from a terminal (1320).

[0189] Action 13403 represents an action in which a network entity (1310) predicts CSI through an artificial intelligence model based on CSI report information.

[0190] Actions 13401 to 13403 illustrate CSI prediction actions, but actions for various cases such as CSI compression, beam prediction, and / or positioning may be applied.

[0191] FIG. 14 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0192] Referring to FIG. 14, transmission and reception operations between a network entity (1410) and a user equipment (UE) (1420) are illustrated. The network entity (1410) of FIG. 14 may correspond to the network entity of FIG. 3. The UE (1420) of FIG. 14 may correspond to the terminal of FIG. 2.

[0193] The embodiment of FIG. 14 may represent an embodiment in which both the network entity (1410) and the user terminal (1420) are capable of quantizing the artificial intelligence model.

[0194] Below, the operation of the communication method of Fig. 14 is described.

[0195] Operation 1430 represents an operation in which a network entity (e.g., a base station) (1410) receives terminal capability information indicating the capabilities of the terminal from a terminal (1420). The terminal capability information may indicate whether the terminal uses an AI model, whether the terminal can quantize the AI ​​model, and / or information regarding the types of data that the terminal-side AI model can process. Furthermore, the terminal capability information may indicate information regarding the types of data that the terminal can process by quantizing the AI ​​model. Furthermore, the terminal capability information may include information regarding the data types and performance of the AI ​​model that the terminal can transmit to the network entity. (See FIG. 8b)

[0196] In one embodiment, terminal capability information may include information about factors that may impact the performance of an AI model. For example, terminal capability information may include information about the total and available storage capacity, the total and available memory capacity used for computation, and so on.

[0197] Terminal capability information regarding the artificial intelligence model is exemplified in FIGS. 7 and 8.

[0198] Operation 1432 represents an operation in which the network entity (1410) determines a data type to be applied to the operation of the artificial intelligence model based on terminal capability information and communication conditions. The network entity (1410) can determine whether the terminal can use the artificial intelligence model through the terminal capability information, information on the data types that the terminal-side artificial intelligence model can process, and / or whether the terminal's artificial intelligence model can be quantized, and thus can determine an appropriate data type accordingly. In addition, the network entity (1410) can determine the data type based on the communication conditions. In this case, the communication conditions may be accuracy conditions, latency conditions, and / or data throughput conditions for transmitted and received information. For example, according to a communication method according to one embodiment of the present disclosure, the artificial intelligence model of the network entity or the terminal can apply a data type of FP32 type when predicting CSI (channel state indicator) or compressing CSI based on a threshold value of average data throughput (or total data throughput). Alternatively, under URLLC (Ultra reliable low latency communication) (latency < 1ms, Reliability < 10^(-5)) communication conditions, the AI ​​model of a network entity or terminal may apply the FP32 type data type when predicting or compressing CSI (channel state indicator). In addition, specific data types may be applied to the operation of the AI ​​model for various purposes, such as beam prediction and / or positioning.

[0199] Operation 1434 represents an operation of transmitting configuration information from a network entity (1410) to a user terminal (1420). The configuration information may include information indicating a data type determined in operation 1432. The configuration information may represent a data type (or AI model-related settings) applied to an artificial intelligence model of the terminal (1420) and / or an output (or result) reporting data type (or AI model output reporting settings) of the artificial intelligence model of the terminal (1420). The output reporting data type of the artificial intelligence model of the terminal (1420) may represent a type of data received from the terminal (1420) by the network entity (1410). The terminal (1420) may convert (or quantize) an existing artificial intelligence model into an artificial intelligence model applicable to a specific data type based on the received configuration information, and may perform inference using the converted (or quantized) artificial intelligence model. Additionally, the terminal (1420) may transmit information output from the AI ​​model as a specific data type to the network entity (1410) based on the received configuration information. The configuration information may be information in which the data type is set for each use case based on the AI ​​model. For example, the configuration information may indicate information on the data type applied to the AI ​​model for CSI compression, CSI prediction, beam management, and / or positioning. In other words, the configuration information may be information in which the quantization level for each use case based on the AI ​​model is mapped.

[0200] In one embodiment, when the quantization level changes, i.e., when the data type to be applied to the artificial intelligence model changes, the network entity (1410) can instruct the terminal (1420) about the changed data type using a low layer message.

[0201] In one embodiment, the configuration information may indicate a predefined reporting format for a specific case related to reporting the output (or result) of the AI ​​model. For example, the configuration information may indicate that, in the case of a CSI prediction case, the terminal reports to the network entity in a predefined reporting format.

[0202] The exchange of configuration information between a network entity (1410) and a terminal (1420) may be performed based on RRC messages. Configuration information related to operation 1434 may be exemplified in FIGS. 9 and 10.

[0203] Operation 1436 represents an operation in which the terminal (1420) converts (or quantizes) an artificial intelligence model based on the received configuration information, and outputs a test result using the converted (or quantized) artificial intelligence model. The terminal (1420) can convert (or quantize) an existing artificial intelligence model on the terminal side into an artificial intelligence model that can process a specific data type based on the received configuration information. For example, if the data type that can be processed by the existing artificial intelligence model on the terminal side is FP32, the terminal can convert the existing artificial intelligence model into an artificial intelligence model that can process the INT16 data type based on the received configuration information (if the terminal has a quantization capability).

[0204] Operation 1438 may represent an operation in which the terminal (1420) performs a test on the artificial intelligence model converted in operation 1436 and transmits the test results for information output from the converted artificial intelligence model to the network entity (1410). That is, the network entity (1410) receives the test results of the output values ​​of the artificial intelligence model based on the data type information from the terminal (1420). The terminal (1420) may store a test data set for performing the test of the artificial intelligence model. The test results may be a message including the output metrics of the artificial intelligence model. The test results may include results regarding the accuracy, computational speed, and reliability of predicted data.

[0205] Operation 1440 represents an operation in which a network entity (1410) determines whether to continue applying configuration information (or data type) based on the test result received in operation 1438, and transmits the determination result to the terminal (1420). The network entity (1410) determines whether the test result (accuracy, speed, reliability, etc.) satisfies a reference condition, and may transmit indication information (e.g., a 1-bit message) (DCI, MAC CE, or RRC message) indicating whether to use the configuration information to the terminal based on the determination result. The indication information indicating whether to use the configuration information may correspond to the indication information indicating whether to use the converted artificial intelligence model (or quantized artificial intelligence model).

[0206] In one embodiment, if the test result does not satisfy the reference condition, the network entity (1410) may change the configuration information (or data type information) and transmit it to the terminal. The terminal may convert the artificial intelligence model based on the changed configuration information and generate a test result again based on the output of the converted artificial intelligence model. Then, the terminal may transmit the test result to the network entity. That is, in the communication method according to one embodiment of the present disclosure, if the test result for the converted artificial intelligence model does not satisfy the reference condition, the network entity may change the configuration information and transmit it to the terminal, and the terminal may test the artificial intelligence model based on the changed configuration information and transmit the result to the network entity. These operations may be repeatedly performed until a test result that satisfies the reference condition is derived.

[0207] In one embodiment, if the network entity (1410) can directly transmit the artificial intelligence model to the terminal (1420), operations 1436 to 1440 may be omitted.

[0208] Action 1442 is an example of a prediction or management-related action (e.g., a positioning action) based on an artificial intelligence model.

[0209] Action 14421 represents an action of transmitting a PRS (position reference signal) from a network entity (1410) to a terminal (1420).

[0210] Action 14422 represents an action in which the terminal (1420) generates output information in the terminal-side artificial intelligence model based on the received PRS.

[0211] Action 14423 represents an action in which the terminal (1420) transmits output information from action 14422 to the network entity (1410).

[0212] The 14424 operation represents an operation in which the network entity (1410) drives an artificial intelligence model on the network entity side based on output information of the terminal-side artificial intelligence model and outputs information.

[0213] Actions 14421 and 14424 illustrate positioning actions, but various other actions may be applied, such as CSI prediction, CSI compression, and / or beam prediction.

[0214] FIG. 15 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0215] Referring to FIG. 15, transmission and reception operations between a network entity (1510) and a user equipment (UE) (1520) are illustrated. The network entity (1510) of FIG. 15 may correspond to the network entity of FIG. 3. The UE (1520) of FIG. 15 may correspond to the terminal of FIG. 2.

[0216] The embodiment of FIG. 15 may represent an embodiment in which a network entity (1510) and a user terminal (1520) mutually transmit an artificial intelligence model. That is, FIG. 15 may represent an embodiment in which the network entity (1510) and the user terminal (1520) cannot quantize the artificial intelligence model. The network entity (1510) and the user terminal (1520) may mutually transmit and receive artificial intelligence models through their respective connected server devices. That is, even if the network entity (1510) and the user terminal (1520) cannot use and quantize the artificial intelligence model, they may receive it from another node (or server) and mutually transmit and receive artificial intelligence models.

[0217] Below, the operation of the communication method of Fig. 15 is described.

[0218] Operation 1530 represents an operation in which a network entity (e.g., a base station) (1510) receives terminal capability information indicating the capabilities of the terminal from a terminal (1520). The terminal capability information may indicate whether the terminal uses an AI model, whether the terminal can quantize the AI ​​model, and / or information regarding the types of data that the terminal-side AI model can process. Furthermore, the terminal capability information may indicate information regarding the types of data that the terminal can process by quantizing the AI ​​model. Furthermore, the terminal capability information may include information regarding the data types and performance of the AI ​​model that the terminal can transmit to the network entity. (See FIG. 8b)

[0219] In one embodiment, terminal capability information may include information about factors that may impact the performance of an AI model. For example, terminal capability information may include information about the total and available storage capacity, the total and available memory capacity used for computation, and so on.

[0220] Terminal capability information regarding the artificial intelligence model is exemplified in FIGS. 7 and 8.

[0221] Operation 1532 represents an operation in which a network entity (1510) determines a data type (or an AI model capable of processing the corresponding data type) to be applied to the operation of an AI model based on terminal capability information and communication conditions. The network entity (1510) can determine an appropriate data type based on the terminal capability information, such as whether the terminal can use the AI ​​model, information on the data types that the AI ​​model on the terminal side can process, whether the AI ​​model of the terminal can be quantized, and / or data types or performance information about the AI ​​model that can be transmitted from the terminal to the network entity. In addition, the network entity (1510) can determine a data type (or an AI model capable of processing the corresponding data type) based on the communication conditions. At this time, the communication conditions may be accuracy conditions, latency conditions, and / or data throughput conditions for transmitted and received information. For example, according to a communication method according to one embodiment of the present disclosure, an artificial intelligence model of a network entity or a terminal may apply a data type of FP32 type when predicting or compressing a channel state indicator (CSI) based on a threshold value of an average data throughput (or total data throughput). Alternatively, under URLLC (Ultra reliable low latency communication) (latency < 1 ms, Reliability < 10^(-5)) communication conditions, the artificial intelligence model of a network entity or a terminal may apply a data type of FP32 type when predicting or compressing a channel state indicator (CSI). In addition, a specific data type may be applied to the operation of the artificial intelligence model for various purposes, such as beam prediction and / or positioning.

[0222] Operation 1534 represents an operation of transmitting configuration information from a network entity (1510) to a user terminal (1520). The configuration information may include information indicating a data type (or an artificial intelligence model capable of processing the data type) determined in operation 1532. The configuration information may be information in which a data type is set for each use case based on an artificial intelligence model. For example, the configuration information may represent information on a data type applied to an artificial intelligence model for CSI compression, CSI prediction, beam management, and / or positioning. In other words, the configuration information may be information in which a quantization level for each use case based on an artificial intelligence model is mapped.

[0223] The exchange of configuration information between a network entity (1510) and a terminal (1520) may be performed based on RRC messages. Configuration information related to operation 1534 may be exemplified in FIGS. 9 and 10.

[0224] Operation 1536 represents an operation of transmitting artificial intelligence model information from a network entity (1510) to a terminal (1520). If the terminal does not have an artificial intelligence model or if the terminal cannot process data of a specific data type by converting an existing artificial intelligence model, the network entity (1510) may transmit an artificial intelligence model capable of processing data of a specific data type to the terminal (1520). For example, if the network entity (1510) decides to apply the FP16 data type to the artificial intelligence model in a CSI prediction operation, the network entity (1510) may transmit an artificial intelligence model capable of processing the FP16 data type to the terminal (1520). At this time, the network entity (1510) may transmit the artificial intelligence model by transmitting structural information and weight information of the artificial intelligence model capable of processing the FP16 data type to the terminal (1520). The terminal (1520) may generate and operate the artificial intelligence model based on the structural information and weight information of the received artificial intelligence model.

[0225] In the embodiment of FIG. 15, since the network entity (1510) transmits the AI ​​model to the terminal (1520), a separate AI model performance test procedure may be omitted. However, if the AI ​​model performance test procedure is performed, the terminal may perform a test on the received AI model and transmit the test results for information output from the received AI model to the network entity. In other words, the network entity may receive the AI ​​model performance test results from the terminal. The test results may include results regarding the accuracy of predicted data, computational speed, and reliability.

[0226] Based on the received test results, the network entity can determine whether to continue using the configuration information (or data type) or the AI ​​model transmitted to the terminal, and transmit the determination result to the terminal. The network entity can determine whether the test results (such as accuracy, speed, or reliability) satisfy the reference conditions, and based on the determination result, transmit instruction information (e.g., a 1-bit message) (DCI, MAC CE, or RRC message) indicating whether to use the configuration information or AI model to the terminal.

[0227] Operation 1538 represents an operation in which a network entity (1510) requests the terminal (1520) to transmit the AI ​​model determined in operation 1532. The request for transmission of the AI ​​model may include requests for multiple AI models. For example, the network entity (1510) may request the terminal to transmit multiple AI models corresponding to multiple use cases.

[0228] Action 1540 represents an action in which a network entity (1510) receives an artificial intelligence model requested in action 1538 from a terminal (1520).

[0229] If a network entity does not have an AI model or if the network entity cannot process data of a specific data type by converting an existing AI model, the network entity (1510) may request the terminal (1520) to receive an AI model capable of processing data of the specific data type. For example, if the network entity (1510) decides to apply the FP16 data type to the AI ​​model in the CSI prediction operation, the network entity (1510) may request the terminal (1520) to receive an AI model capable of processing the FP16 data type. The terminal (1520) may transmit the AI ​​model by transmitting structural information and weight information of the AI ​​model capable of processing the FP16 data type to the network entity (1510). The network entity (1510) may generate and operate the AI ​​model based on the structural information and weight information of the AI ​​model received from the terminal (1520).

[0230] In the embodiment of FIG. 15, since the network entity (1510) receives the AI ​​model from the terminal (1520), a separate AI model performance testing procedure may be omitted. However, if the AI ​​model performance testing procedure is performed, the network entity may perform a test on the received AI model and generate test results for information output from the received AI model. The test results may include results regarding the accuracy, computational speed, and reliability of predicted data. In addition, the network entity may determine whether to use the received AI model based on the generated test results.

[0231] In one embodiment, the network entity (1510) may perform a test on the AI ​​model received in operation 1540. The network entity (1510) may store a test data set for performing the test on the received AI model. In addition, the network entity (1510) may determine whether to continue using the received AI model based on the test result. The network entity (1510) may determine whether the test result (such as accuracy, speed, or reliability) satisfies a reference condition, and if the test result does not satisfy the reference condition, the network entity may transmit a transmission request for another AI model to the terminal. In a communication method according to one embodiment of the present disclosure, if the test result for the AI ​​model received by the network entity does not satisfy the reference condition, the network entity may transmit a transmission request for another AI model to the terminal, receive another AI model, and perform the test again. These operations may be repeatedly performed until a test result that satisfies the reference condition is derived.

[0232] Action 1542 is an example of a prediction or management-related action (e.g., a positioning action) based on an artificial intelligence model.

[0233] Action 15421 represents an action of transmitting a PRS (position reference signal) from a network entity (1510) to a terminal (1520).

[0234] Action 15422 represents an action in which the terminal (1520) generates output information in the terminal-side artificial intelligence model based on the received PRS.

[0235] Action 15423 represents an action in which the terminal (1520) transmits output information from action 15422 to the network entity (1510).

[0236] The 15424 operation represents an operation in which a network entity (1510) drives an artificial intelligence model on the network entity side based on output information of the terminal-side artificial intelligence model and outputs information.

[0237] Actions 15421 to 15424 illustrate positioning actions, but actions for various cases such as CSI prediction, CSI compression, and / or beam prediction may be applied.

[0238] FIG. 16 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0239] Referring to FIG. 16, transmission and reception operations between a network entity (1610) and a user equipment (UE) (1620) are illustrated. The network entity (1610) of FIG. 16 may correspond to the network entity of FIG. 3. The UE (1620) of FIG. 16 may correspond to the terminal of FIG. 2.

[0240] The embodiment of FIG. 16 is an embodiment for a case where a user terminal (1620) requests a change in data type from a network entity (1610), i.e., a change in quantization level.

[0241] Below, the operation of the communication method of Fig. 16 is described.

[0242] Operation 1630 represents an operation in which a network entity (e.g., a base station) (1610) receives a data type change request from a terminal. The data type change request may include terminal capability information indicating the capabilities of the terminal. The terminal capability information may indicate whether the terminal uses an AI model, whether the terminal can quantize the AI ​​model, and / or information regarding the types of data that the terminal-side AI model can process. Furthermore, the terminal capability information may indicate information regarding the types of data that the terminal can process by quantizing the AI ​​model. Furthermore, the terminal capability information may include information regarding the data types and performance of the AI ​​model that the terminal can transmit to the network entity. (See FIG. 8b)

[0243] A request for a data type change in operation 1630 may include information indicating data types that can be changed for each AI-based operation case of the terminal, information regarding cases requiring a data type change, etc. For example, the request for a data type change may include data types applicable to AI-based CSI prediction operations on the terminal. Furthermore, for example, the request for a data type change may include a request for a data type change for AI-based CSI compression operations. The request for a data type change may be information transmitted and received in a bitmap (one-hot encoding) format.

[0244] A request to change the data type of the 1630 operation can be sent and received as an RRC message. The request to change the data type can be an on-demand request from the terminal side, an on-demand request from the network entity side, or a periodic request.

[0245] In one embodiment, terminal capability information may include information about factors that may impact the performance of an AI model. For example, terminal capability information may include information about the total and available storage capacity, the total and available memory capacity used for computation, and so on.

[0246] Information included in a request to change the data type of a terminal can be exemplified in Fig. 17.

[0247] Operation 1632 represents an operation in which a network entity (1810) determines a data type to be applied to the operation of an artificial intelligence model based on a data type change request, terminal capability information, and / or communication conditions. The network entity (1610) can determine an appropriate data type based on the terminal capability information, such as whether the terminal can use the artificial intelligence model, information on data types that can be processed by the terminal-side artificial intelligence model, and / or whether the terminal's artificial intelligence model can be quantized. Furthermore, the network entity (1610) can determine the data type based on the information included in the data type change request.

[0248] In addition, the network entity (1610) may determine the data type based on the communication condition. At this time, the communication condition may be an accuracy condition, a latency condition, and / or a data throughput condition for the transmitted and received information. For example, according to a communication method according to an embodiment of the present disclosure, an artificial intelligence model of a network entity or a terminal may apply an FP32 type data type when predicting or compressing a channel state indicator (CSI) based on a threshold value of an average data throughput (or, an overall data throughput). Alternatively, under an ultra-reliable low latency communication (URLLC) (latency < 1 ms, reliability < 10^(-5)) communication condition, the artificial intelligence model of a network entity or a terminal may apply an FP32 type data type when predicting or compressing a channel state indicator (CSI). In addition, a specific data type may be applied to the operation of the artificial intelligence model for various purposes, such as beam prediction and / or positioning.

[0249] Operation 1634 represents an operation of transmitting configuration information from a network entity (1610) to a user terminal (1620). The configuration information may include information indicating a data type determined in operation 1632. The configuration information may represent a data type (or AI model-related settings) applied to an artificial intelligence model of the terminal (1620) and / or an output (or result) reporting data type (or AI model output reporting settings) of the artificial intelligence model of the terminal (1620). The output reporting data type of the artificial intelligence model of the terminal (1620) may represent a data type received from the terminal (1620) by the network entity (1610). The terminal (1620) may convert (or quantize) an existing artificial intelligence model into an artificial intelligence model applicable to a specific data type based on the received configuration information, and may perform inference using the converted (or quantized) artificial intelligence model. Additionally, the terminal (1620) may transmit information output from the AI ​​model as a specific data type to the network entity (1610) based on the received configuration information. The configuration information may be information in which the data type is set for each use case based on the AI ​​model. For example, the configuration information may indicate information on the data type applied to the AI ​​model for CSI compression, CSI prediction, beam management, and / or positioning. In other words, the configuration information may be information in which the quantization level for each use case based on the AI ​​model is mapped.

[0250] In one embodiment, when the quantization level changes, i.e., when the data type to be applied to the artificial intelligence model changes, the network entity (1610) can instruct the terminal (1620) about the changed data type using a low layer message.

[0251] In one embodiment, the configuration information may indicate a predefined reporting format for a specific case related to reporting the output (or result) of the AI ​​model. For example, the configuration information may indicate that, in the case of a CSI prediction case, the terminal reports to the network entity in a predefined reporting format.

[0252] The exchange of configuration information between a network entity (1610) and a terminal (1620) may be performed based on RRC messages. Configuration information related to operation 1634 may be exemplified in FIGS. 9 and 10.

[0253] Operation 1636 represents an operation in which the terminal (1620) converts (or quantizes) an artificial intelligence model based on the received configuration information, and outputs a test result using the converted (or quantized) artificial intelligence model. The terminal (1620) can convert (or quantize) an existing artificial intelligence model on the terminal side into an artificial intelligence model that can process a specific data type based on the received configuration information. For example, if the data type that can be processed by the existing artificial intelligence model on the terminal side is FP32, the terminal can convert the existing artificial intelligence model into an artificial intelligence model that can process the INT16 data type based on the received configuration information (if the terminal has a quantization capability).

[0254] Operation 1638 may represent an operation in which the terminal (1620) performs a test on the artificial intelligence model converted in operation 1636 and transmits the test results for information output from the converted artificial intelligence model to the network entity (1610). That is, the network entity (1610) receives the test results of the output values ​​of the artificial intelligence model based on the data type information from the terminal (1620). The terminal (1620) may store a test data set for performing the test of the artificial intelligence model. The test results may be a message including the output metrics of the artificial intelligence model. The test results may include results regarding the accuracy, computational speed, and reliability of predicted data.

[0255] Operation 1640 represents an operation in which a network entity (1610) determines whether to continue applying configuration information (or data type) based on the test result received in operation 1638, and transmits the determination result to the terminal (1620). The network entity (1610) determines whether the test result (accuracy, speed, reliability, etc.) satisfies a reference condition, and may transmit indication information (e.g., a 1-bit message) (DCI, MAC CE, or RRC message) indicating whether to use the configuration information to the terminal based on the determination result. The indication information indicating whether to use the configuration information may correspond to the indication information indicating whether to use the converted artificial intelligence model (or quantized artificial intelligence model).

[0256] In one embodiment, if the test result does not satisfy the reference condition, the network entity (1610) may change the configuration information (or data type information) and transmit it to the terminal. The terminal may convert the artificial intelligence model based on the changed configuration information and generate a test result again based on the output of the converted artificial intelligence model. Then, the terminal may transmit the test result to the network entity. That is, in the communication method according to one embodiment of the present disclosure, if the test result for the converted artificial intelligence model does not satisfy the reference condition, the network entity may change the configuration information and transmit it to the terminal, and the terminal may test the artificial intelligence model based on the changed configuration information and transmit the result to the network entity. These operations may be repeatedly performed until a test result that satisfies the reference condition is derived.

[0257] In one embodiment, if the network entity (1610) can directly transmit the artificial intelligence model to the terminal (1620), operations 1636 to 1640 may be omitted.

[0258] Action 1642 is an example of a prediction or management-related action based on an artificial intelligence model (e.g., a CSI prediction action).

[0259] The 16421 operation represents an operation of transmitting a CSI-RS (reference signal) from a network entity (1610) to a terminal (1620).

[0260] Action 16422 represents an action in which the terminal (1620) predicts CSI using an artificial intelligence model based on the received CSI-RS.

[0261] Action 16423 represents an action in which the terminal (1620) transmits the prediction result in action 16422 to the network entity (1610).

[0262] Actions 16421 and 16423 illustrate CSI prediction actions, but actions for various cases such as CSI compression, beam prediction, and / or positioning may be applied.

[0263] FIG. 17 illustrates information for a data type change request according to one embodiment of the present disclosure.

[0264] The information illustrated in FIG. 17 may be information included in a data type change request related to operation 1630 of FIG. 16.

[0265] Referring to FIG. 17, a data type change request may include changeable data type information corresponding to an AI-based operation case. The AI-based operation cases may include CSI compression, CSI prediction, beam operation, and / or positioning. Furthermore, data type information may be mapped to each case. For example, the data type change request may indicate that an AI model based on the FP16 data type is available for an AI-based CSI compression operation. Furthermore, the data type change request may indicate an AI-based operation (e.g., CSI prediction) requiring a data type change. Furthermore, the data type change request may further include terminal performance information regarding the corresponding data type for each case. For example, the data type change request may include information indicating the performance of the terminal when the first data type corresponding to the first case is applied to the AI ​​model. The performance information may include accuracy, loss, and the like. A network entity may determine a data type for a specific case or quantize an AI model on the network entity's side based on the information included in the aforementioned data type change request.

[0266] FIG. 18 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0267] The embodiment described in FIG. 18 features an operation of transmitting an artificial intelligence model from a network entity (1810) to a terminal (1820) when a data type change request is made.

[0268] Referring to FIG. 18, transmission and reception operations between a network entity (1810) and a user equipment (UE) (1820) are illustrated. The network entity (1810) of FIG. 18 may correspond to the network entity of FIG. 3. The UE (1820) of FIG. 18 may correspond to the terminal of FIG. 2.

[0269] The embodiment of FIG. 18 is an embodiment for a case where a user terminal (1820) requests a change in data type from a network entity (1810), i.e., a change in quantization level.

[0270] Below, the operation of the communication method of Fig. 18 is described.

[0271] Operation 1830 represents an operation in which a network entity (e.g., a base station) (1810) receives a data type change request from a terminal. The data type change request may include terminal capability information indicating the capabilities of the terminal. The terminal capability information may indicate whether the terminal uses an AI model, whether the terminal can quantize the AI ​​model, and / or information regarding the types of data that the terminal-side AI model can process. Furthermore, the terminal capability information may indicate information regarding the types of data that the terminal can process by quantizing the AI ​​model. Furthermore, the terminal capability information may include information regarding the data types and performance of the AI ​​model that the terminal can transmit to the network entity. (See FIG. 8b)

[0272] A request for a data type change in operation 1830 may include information indicating data types that can be changed for each AI-based operation case of the terminal, information regarding cases requiring a data type change, etc. For example, the request for a data type change may include data types applicable to AI-based CSI prediction operations on the terminal. Furthermore, for example, the request for a data type change may include a request for a data type change for AI-based CSI compression operations. The request for a data type change may be information transmitted and received in a bitmap (one-hot encoding) format.

[0273] A request to change the data type of an 1830 operation can be sent and received as an RRC message. The request to change the data type can be an on-demand request from the terminal side, an on-demand request from the network entity side, or a periodic request.

[0274] In one embodiment, terminal capability information may include information about factors that may impact the performance of an AI model. For example, terminal capability information may include information about the total and available storage capacity, the total and available memory capacity used for computation, and so on.

[0275] Information included in a request to change the data type of a terminal can be exemplified in Fig. 17.

[0276] Operation 1832 represents an operation in which the network entity (1810) determines the data type to be applied to the operation of the artificial intelligence model based on a data type change request, terminal capability information, and / or communication conditions. Alternatively, operation 1832 may determine the terminal-side artificial intelligence model based on terminal capability information and communication conditions in the network entity (1810).

[0277] The network entity (1810) can determine an appropriate data type based on terminal capability information, such as whether the terminal can use an AI model, the data types that the terminal-side AI model can process, and / or whether the terminal's AI model can be quantized. Furthermore, the network entity (1810) can determine the data type based on information included in the data type change request.

[0278] In addition, the network entity (1810) can determine the data type based on the communication condition. At this time, the communication condition may be an accuracy condition, a latency condition, and / or a data throughput condition for the transmitted and received information. For example, according to a communication method according to an embodiment of the present disclosure, the AI ​​model of the network entity or the terminal can apply the FP32 type data type in a specific AI-based operation case based on a threshold value of the average data throughput (or, the total data throughput). Alternatively, in the URLLC (Ultra reliable low latency communication) (latency < 1 ms, Reliability < 10^(-5)) communication condition, the AI ​​model of the network entity or the terminal can apply the FP32 type data type for CSI (channel state indicator) prediction (as an example). In addition, a specific data type can be applied to the operation of the AI ​​model for various purposes such as beam prediction and / or positioning.

[0279] Operation 1834 represents an operation of transmitting configuration information from the network entity (1810) to the user terminal (1820). The configuration information may include information indicating the data type determined in operation 1832. The configuration information may represent a data type (or AI model-related settings) applied to the artificial intelligence model of the terminal (1820) and an output (or result) reporting data type (or AI model output reporting settings) of the artificial intelligence model of the terminal (1820). The output reporting data type of the artificial intelligence model of the terminal (1820) may represent the type of data received from the terminal (1820) by the network entity (1810). The terminal (1820) may receive an artificial intelligence model based on the received configuration information from the network entity (1810) and perform inference using the received artificial intelligence model. In addition, the terminal (1820) may transmit information output from the artificial intelligence model as a specific data type to the network entity (1810) based on the received configuration information. Configuration information may be information that specifies data types for use cases (e.g., CSI prediction) based on AI models. For example, configuration information may indicate data types applied to AI models for CSI compression, CSI prediction, beam management, and / or positioning. In other words, configuration information may be information that maps the quantization level for each use case based on the AI ​​model.

[0280] The exchange of configuration information between a network entity (1810) and a terminal (1820) may be performed based on RRC messages. Configuration information related to operation 1834 may be exemplified in FIGS. 9 and 10.

[0281] Operation 1836 represents an operation of transmitting artificial intelligence model information from a network entity (1810) to a terminal (1820). If the terminal does not have an artificial intelligence model or if the terminal cannot process data of a specific data type by converting an existing artificial intelligence model, the network entity (1810) may transmit an artificial intelligence model capable of processing data of a specific data type to the terminal (1820). For example, if the network entity (1810) decides to apply the FP16 data type to the artificial intelligence model in the CSI prediction operation, the network entity (1810) may transmit an artificial intelligence model capable of processing the FP16 data type to the terminal (1820). At this time, the network entity (1810) may transmit the artificial intelligence model by transmitting structural information and weight information of the artificial intelligence model capable of processing the FP16 data type to the terminal (1820). The terminal (1820) may generate and operate the artificial intelligence model based on the structural information and weight information of the received artificial intelligence model.

[0282] In the embodiment of FIG. 18, since the network entity (1810) transmits the AI ​​model to the terminal (1820), the AI ​​model performance testing procedure may be omitted. However, if the AI ​​model performance testing procedure is performed, the terminal may perform a test on the received AI model and transmit the test results for information output from the received AI model to the network entity. In other words, the network entity may receive the AI ​​model performance test results from the terminal. The test results may include results regarding the accuracy, computational speed, and reliability of predicted data.

[0283] Based on the test results, the network entity can determine whether to continue using the configuration information (or data type) or the AI ​​model transmitted to the terminal, and transmit the determination result to the terminal. The network entity can determine whether the test results (such as accuracy, speed, or reliability) satisfy the reference conditions, and based on the determination result, transmit instruction information (e.g., a 1-bit message) (DCI, MAC CE, or RRC message) indicating whether to use the configuration information or AI model to the terminal.

[0284] In one embodiment, if the test result does not satisfy the reference condition, the network entity can change the configuration information (or data type information) and transmit a different AI model to the terminal. The terminal can then generate a test result again based on the output of the received AI model. That is, in the communication method according to one embodiment of the present disclosure, if the test result for the AI ​​model received by the terminal does not satisfy the reference condition, the network entity can transmit a different AI model to the terminal, and the terminal can test the AI ​​model again and transmit the result to the network entity. These operations can be repeatedly performed until a test result that satisfies the reference condition is derived.

[0285] Action 1838 is an example of a prediction or management-related action based on an artificial intelligence model (e.g., a CSI prediction action).

[0286] Action 18381 represents an action of transmitting a CSI-RS (reference signal) from a network entity (1810) to a terminal (1820).

[0287] Action 18382 represents an action in which a terminal (1820) predicts CSI using an artificial intelligence model based on the received CSI-RS.

[0288] Action 18383 represents an action in which the terminal (1820) transmits the prediction result from action 18382 to the network entity (1810).

[0289] Actions 18381 and 18383 illustrate CSI prediction actions, but actions for various cases such as CSI compression, beam prediction, and / or positioning may be applied.

[0290] FIG. 19 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0291] According to the embodiment described in Fig. 19, when the data type to be used in the artificial intelligence model is predefined, the network entity (1910) can transmit the artificial intelligence model to the terminal (1920).

[0292] Referring to FIG. 19, transmission and reception operations between a network entity (1910) and a user equipment (UE) (1920) are illustrated. The network entity (1910) of FIG. 19 may correspond to the network entity of FIG. 3. The UE (1920) of FIG. 19 may correspond to the terminal of FIG. 2.

[0293] The embodiment of Fig. 19 is an embodiment of a transmission and reception operation between a network entity (1910) and a terminal (1920) in a case where the data type of an artificial intelligence model for each artificial intelligence-based operation case (CSI prediction, CSI compression, etc.) is predefined.

[0294] Below, the operation of the communication method of Fig. 19 is described.

[0295] Operation 1930 represents an operation in which the network entity (1910) transmits configuration information to the terminal (1920). In the embodiment of FIG. 19, since it is assumed that the data type of the artificial intelligence model for each artificial intelligence-based operation case is predefined, the configuration information does not need to include such information. However, the configuration information may include the output report data type of the artificial intelligence model on the terminal side. The configuration information may indicate the type of data that the network entity (1910) receives from the terminal (1920). The terminal (1920) may transmit information to the network entity (1910) in a predefined data type, and may also transmit output information of the artificial intelligence model to the network entity (1910) based on the configuration information received in operation 1930. The exchange of configuration information between the network entity (1910) and the terminal (1920) may be performed based on an RRC message (e.g., an RRC reconfiguration message).

[0296] Action 1932 represents an action of transmitting an artificial intelligence model from a network entity (1910) to a terminal (1920). The network entity (1910) can transmit an appropriate artificial intelligence model to the terminal (1920) according to a predefined data type.

[0297] Operation 1934 represents operations based on a terminal-side artificial intelligence model. For example, it may represent transmission and reception operations between a network entity (1910) and a terminal (1920) for CSI prediction, CSI compression, beam operation (or prediction), and / or positioning. The terminal (1920) may use the artificial intelligence model received in operation 1932 and transmit the output information of the artificial intelligence model to the network entity (1910) based on the configuration information.

[0298] In the embodiment of FIG. 19, operation 1934 exemplifies operations based on a terminal-side artificial intelligence model (one-sided model), but may also exemplify operations using both the terminal and network entity artificial intelligence models (two-sided model) or operations based on the network entity artificial intelligence model.

[0299] FIG. 20 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0300] Referring to FIG. 20, transmission and reception operations between a network entity (2010) and a user equipment (UE) (2020) are illustrated. The network entity (2010) of FIG. 20 may correspond to the network entity of FIG. 3. The UE (2020) of FIG. 20 may correspond to the terminal of FIG. 2.

[0301] The embodiment of Fig. 20 is an embodiment of a transmission and reception operation between a network entity (2010) and a terminal (2020) in a case where the data type of an artificial intelligence model for each artificial intelligence-based operation case (CSI prediction, CSI compression, etc.) is predefined. The embodiment of Fig. 20 shows an embodiment in a case where a change in the report data type of a terminal (2020) is required during communication between the network entity (2010) and the terminal (2020).

[0302] Below, the operation of the communication method of Fig. 20 is described.

[0303] Operation 2030 represents an operation in which the network entity (2010) transmits configuration information indicating a report data type to the terminal (2020). The configuration information may indicate a type of data that the network entity (2010) receives from the terminal (2020). The terminal (2020) may transmit output information of the artificial intelligence model to the network entity (2010) based on the changed report data type information received in operation 2030.

[0304] Action 2032 represents an action of transmitting an artificial intelligence model from a network entity (2010) to a terminal (2020). The network entity (2010) may transmit an appropriate artificial intelligence model to the terminal (2020) based on the changed report data type information. Alternatively, the network entity (2010) may transmit an appropriate artificial intelligence model to the terminal (2020) based on a predefined data type.

[0305] Operation 2034 represents operations based on the terminal-side artificial intelligence model. For example, it may represent transmission and reception operations between the network entity (2010) and the terminal (2020) for CSI prediction, CSI compression, beam operation (or prediction), and / or positioning. The terminal (2020) may use the artificial intelligence model received in operation 2032 and transmit the output information of the artificial intelligence model to the network entity (2010) based on the reported data type information.

[0306] In the embodiment of FIG. 20, operation 2034 exemplifies operations based on a terminal-side artificial intelligence model (one-sided model), but may also exemplify operations using both the terminal and network entity artificial intelligence models (two-sided model) or operations based on the network entity artificial intelligence model.

[0307] FIG. 21 is a drawing for explaining a communication method according to one embodiment of the present disclosure.

[0308] Referring to FIG. 21, transmission and reception operations between a network entity (2110) and a user equipment (UE) (2120) are illustrated. The network entity (2110) of FIG. 21 may correspond to the network entity of FIG. 3. The UE (2120) of FIG. 21 may correspond to the terminal of FIG. 2.

[0309] Below, the operation of the communication method of Fig. 21 is described.

[0310] Operation 2130 represents an operation in which a network entity (e.g., a base station) (2110) receives terminal capability information indicating the capabilities of the terminal from the terminal. The terminal capability information may indicate whether the terminal uses an AI model, whether the terminal can quantize the AI ​​model, and / or information regarding the types of data that the terminal-side AI model can process. Furthermore, the terminal capability information may indicate information regarding the types of data that the terminal can process by quantizing the AI ​​model. Furthermore, the terminal capability information may include information regarding the data types and performance of the AI ​​model that the terminal can transmit to the network entity. (See FIG. 8b)

[0311] In one embodiment, terminal capability information may include information about factors that may impact the performance of an AI model. For example, terminal capability information may include information about the total and available storage capacity, the total and available memory capacity used for computation, and so on.

[0312] Terminal capability information regarding the artificial intelligence model is exemplified in FIGS. 7 and 8.

[0313] Operation 2132 represents an operation in which a network entity (2110) determines a data type to be applied to the operation of an artificial intelligence model based on terminal capability information and communication conditions. At this time, the network entity (2110) may determine a critical data type. Once a critical data type is determined, data types with lower complexity or higher complexity may be permitted based on the critical data type. For example, if the critical data type is determined to be "INT16," data types with higher complexity, including INT16, may be permitted together. Alternatively, if the critical data type is determined to be "INT16," data types with lower complexity, including INT16, may be permitted. Therefore, the terminal may select which data type to apply to the artificial intelligence model. The critical data type was described above with reference to FIG. 9.

[0314] The network entity (2110) can determine an appropriate data type based on the terminal capability information, such as whether the terminal can use an artificial intelligence model, information on the data types that can be processed by the terminal-side artificial intelligence model, and / or whether the terminal's artificial intelligence model can be quantized. In addition, the network entity (2110) can determine the data type based on the communication conditions. In this case, the communication conditions may be accuracy conditions, latency conditions, and / or data throughput conditions for transmitted and received information. For example, according to a communication method according to one embodiment of the present disclosure, the network entity or the artificial intelligence model of the terminal can apply a data type of FP32 type when predicting CSI (channel state indicator) or compressing CSI based on a threshold value of average data throughput (or total data throughput). Alternatively, under URLLC (Ultra reliable low latency communication) (latency < 1ms, Reliability < 10^(-5)) communication conditions, the AI ​​model of a network entity or terminal may apply the FP32 type data type when predicting or compressing CSI (channel state indicator). In addition, specific data types may be applied to the operation of the AI ​​model for various purposes, such as beam prediction and / or positioning.

[0315] Operation 2134 represents an operation of transmitting configuration information from a network entity (2110) to a user terminal (2120). The configuration information may include information indicating a data type determined in operation 2132. The configuration information may represent a data type (or AI model-related settings) applied to an artificial intelligence model of the terminal (2120) and / or an output (or result) report data type (or AI model output report settings) of the artificial intelligence model of the terminal (2120). The output report data type of the artificial intelligence model of the terminal (2120) may represent a data type received from the terminal (2120) by the network entity (2110). The terminal (2120) may convert (or quantize) an existing artificial intelligence model into an artificial intelligence model applicable to a specific data type based on the received configuration information, and may perform inference using the converted (or quantized) artificial intelligence model. Additionally, the terminal (2120) may transmit information output from the AI ​​model in a specific data type to the network entity (2110) based on the received configuration information. The configuration information may be information in which the data type is set for each use case based on the AI ​​model. For example, the configuration information may indicate information on the data type applied to the AI ​​model for CSI compression, CSI prediction, beam management, and / or positioning.

[0316] The exchange of configuration information between a network entity (2110) and a terminal (2120) may be performed based on RRC messages. Configuration information related to operation 2134 may be exemplified in FIGS. 9 and 10.

[0317] Operation 2136 represents an operation in which the terminal (2120) transmits data type information to the network entity (2110). Since the terminal (2120) has received critical data type information in operation 2134, the terminal (2120) can determine which data type among the available multiple data types to apply to the artificial intelligence model and transmit the determined data type information to the network entity (2110). Operation 2136 can be transmitted and received via low layer signaling such as UCI / MAC CE or RRC signaling.

[0318] In one embodiment, the network entity (2110) may transmit information about the data types that can be processed by the artificial intelligence model of the network entity (2110) to the terminal (2120). Accordingly, the terminal (2120) may determine the data type information by considering the performance of the artificial intelligence model of the network entity (2110) prior to operation 2136.

[0319] Operation 2138 exemplifies operations for transmitting and receiving AI models, transforming (or quantizing) AI models, and / or transmitting and receiving AI model test results. Details of these operations are included in the descriptions of FIGS. 6, 11 to 16, and 18 to 21.

[0320] FIG. 22 is a diagram for explaining a method of a network entity according to one embodiment of the present disclosure.

[0321] Referring to Fig. 22, the operations of a network entity are illustrated. The network entity of Fig. 22 may correspond to the network entity of Fig. 3.

[0322] Operation 2210 represents an operation in which a network entity receives information indicating the capabilities of a terminal from a terminal. The terminal capability information may indicate whether the terminal uses an AI model, whether the terminal can quantize the AI ​​model, information on the types of data that the AI ​​model on the terminal side can process, and / or information on the AI ​​models that the terminal can transmit to the network entity. Furthermore, the terminal capability information may indicate information on the types of data that the terminal can process by quantizing the AI ​​model.

[0323] In one embodiment, terminal capability information may include information about factors that may impact the performance of an AI model. For example, terminal capability information may include information about the total and available storage capacity, the total and available memory capacity used for computation, and so on.

[0324] In one embodiment, operation 2210 may represent an operation in which a network entity receives a request from a terminal to change a previously established data type.

[0325] Terminal capability information regarding the artificial intelligence model is exemplified in FIGS. 7 and 8.

[0326] The 2210 operation may correspond to the 630 operation of FIG. 6, the 1130 operation of FIG. 11, the 1230 operation of FIG. 12, the 1330 operation of FIG. 13, the 1430 operation of FIG. 14, the 1530 operation of FIG. 15, the 1630 operation of FIG. 16, the 1830 operation of FIG. 18, or the 2130 operation of FIG. 21.

[0327] Operation 2220 represents an operation in which a network entity determines a data type based on information indicating the capabilities of a terminal and communication conditions. The network entity can determine an appropriate data type based on the terminal capability information, such as whether the terminal can use an artificial intelligence model, information on the types of data that can be processed by the terminal-side artificial intelligence model, whether the terminal's artificial intelligence model can be quantized, and / or data types or model performance information of the artificial intelligence model that the terminal can transmit to the network entity. In addition, the network entity can determine the data type based on the communication conditions. In this case, the communication conditions may be accuracy conditions, latency conditions, and / or data throughput conditions for transmitted and received information. For example, according to a communication method according to an embodiment of the present disclosure, the network entity or the artificial intelligence model of the terminal can apply a data type of FP32 type when predicting a channel state indicator (CSI) or compressing CSI based on a threshold value of average data throughput (or total data throughput). Alternatively, under URLLC (Ultra reliable low latency communication) (latency < 1ms, Reliability < 10^(-5)) communication conditions, the AI ​​model of a network entity or terminal may apply the FP32 type data type when predicting or compressing CSI (channel state indicator). In addition, specific data types may be applied to the operation of the AI ​​model for various purposes, such as beam prediction and / or positioning.

[0328] The 2220 operation may correspond to the 632 operation of FIG. 6, the 1132 operation of FIG. 11, the 1232 operation of FIG. 12, the 1332 operation of FIG. 13, the 1432 operation of FIG. 14, the 1532 operation of FIG. 15, the 1632 operation of FIG. 16, the 1832 operation of FIG. 18, or the 2132 operation of FIG. 21.

[0329] Operation 2230 represents an operation in which a network entity transmits data type information indicating a determined data type to a terminal. The data type information may indicate a data type (or AI model-related settings) applied to the terminal's artificial intelligence model and / or an output (or result) reporting data type (or AI model output reporting settings) of the terminal's artificial intelligence model. The output reporting data type of the terminal's artificial intelligence model may indicate the type of data received by the network entity from the terminal. Based on the received data type information, the terminal may convert (or quantize) an existing artificial intelligence model into an artificial intelligence model applicable to a specific data type, and perform inference using the converted (or quantized) artificial intelligence model. Furthermore, the terminal may transmit information output from the artificial intelligence model as a specific data type to the network entity based on the received data type information. The data type information may be information in which a data type is set for each use case based on the artificial intelligence model. For example, the data type information may indicate information on a data type applied to an artificial intelligence model for CSI compression, CSI prediction, beam management, and / or positioning. That is, data type information can be information mapped to the quantization level for each case based on the artificial intelligence model.

[0330] The 2230 operation may correspond to the 634 operation of FIG. 6, the 1134 operation of FIG. 11, the 1234 operation of FIG. 12, the 1334 operation of FIG. 13, the 1434 operation of FIG. 14, the 1534 operation of FIG. 15, the 1634 operation of FIG. 16, the 1834 operation of FIG. 18, or the 2134 operation of FIG. 21.

[0331] Operation 2240 represents an operation in which a network entity receives output information of an artificial intelligence model from a terminal based on data type information. For example, if the data type of data to be reported from the terminal to the network entity is set to INT 16, the terminal may transmit data output from the artificial intelligence model to the network entity in the INT16 type. At this time, the output data of the terminal-side artificial intelligence model may be a test result indicating the performance of the terminal-side artificial intelligence model, and may be prediction information, vectors, etc. according to artificial intelligence-based operation cases (CSI prediction, CSI compression, beam prediction, etc.).

[0332] In one embodiment, the network entity may further perform an operation of receiving a test result of an output value of an artificial intelligence model based on data type information from a terminal, an operation of determining whether to use the determined data type based on the test result, and an operation of transmitting information indicating whether to use the determined data type to the terminal. These operations may correspond to operations 638 to 640 of FIG. 6 , operations 1438 to 1440 of FIG. 14 , and operations 1638 to 1640 of FIG. 16 .

[0333] In one embodiment, a network entity may perform operations such as requesting information about an AI model applicable to a determined data type from a terminal and receiving structure and weight information about the AI ​​model applicable to the determined data type from the terminal. These operations may correspond to operations 1336 through 1338 of FIG. 13 .

[0334] In one embodiment, a network entity may transmit to a terminal the structure and weight information of a first AI model applicable to a first data type, and receive from the terminal the structure and weight information of a second AI model applicable to a second data type. These operations may correspond to operations 1536 to 1540 of FIG. 15 .

[0335] While the operations of the communication method according to the embodiments of the present disclosure have been described separately for each embodiment, the operations included in each embodiment can be combined with operations of other embodiments to form a new embodiment. Accordingly, combined embodiments not described as a single embodiment in the present disclosure can also be understood as being described by the present disclosure.

[0336] FIG. 23 is a diagram for explaining a method of a user terminal according to one embodiment of the present disclosure.

[0337] Referring to Fig. 23, the operations of the terminal are illustrated. The terminal of Fig. 23 can correspond to the user terminal of Fig. 2.

[0338] Operation 2310 refers to an operation in which a terminal transmits information indicating the terminal's capabilities to a network entity. The terminal capability information may indicate whether the terminal uses an AI model, whether the terminal can quantize the AI ​​model, information on the types of data that the AI ​​model on the terminal side can process, and / or information on the AI ​​models that the terminal can transmit to the network entity. Furthermore, the terminal capability information may indicate information on the types of data that the terminal can process by quantizing the AI ​​model.

[0339] In one embodiment, terminal capability information may include information about factors that may impact the performance of an AI model. For example, terminal capability information may include information about the total and available storage capacity, the total and available memory capacity used for computation, and so on.

[0340] In one embodiment, operation 2310 may represent an operation in which a terminal transmits a request to a network entity to change a previously established data type.

[0341] Terminal capability information regarding the artificial intelligence model is exemplified in FIGS. 7 and 8.

[0342] The 2310 operation may correspond to the 630 operation of FIG. 6, the 1130 operation of FIG. 11, the 1230 operation of FIG. 12, the 1330 operation of FIG. 13, the 1430 operation of FIG. 14, the 1530 operation of FIG. 15, the 1630 operation of FIG. 16, the 1830 operation of FIG. 18, or the 2130 operation of FIG. 21.

[0343] Operation 2320 represents an operation in which a terminal receives data type information indicating a data type determined from a network entity. The data type information may indicate a data type (or AI model-related settings) applied to an artificial intelligence model of the terminal and / or an output (or result) reporting data type (or AI model output reporting settings) of the artificial intelligence model of the terminal. The output reporting data type of the artificial intelligence model of the terminal may indicate the type of data that the network entity receives from the terminal. Based on the received data type information, the terminal may convert (or quantize) an existing artificial intelligence model into an artificial intelligence model applicable to a specific data type, and perform inference using the converted (or quantized) artificial intelligence model. Furthermore, based on the received data type information, the terminal may transmit information output from the artificial intelligence model as a specific data type to the network entity. The data type information may be information in which a data type is set for each use case based on the artificial intelligence model. For example, the data type information may indicate information on a data type applied to an artificial intelligence model for CSI compression, CSI prediction, beam management, and / or positioning. That is, data type information can be information mapped to the quantization level for each case based on the artificial intelligence model.

[0344] The 2320 operation may correspond to the 634 operation of FIG. 6, the 1134 operation of FIG. 11, the 1234 operation of FIG. 12, the 1334 operation of FIG. 13, the 1434 operation of FIG. 14, the 1534 operation of FIG. 15, the 1634 operation of FIG. 16, the 1834 operation of FIG. 18, or the 2134 operation of FIG. 21.

[0345] Action 2330 represents an action in which the terminal generates an output of an artificial intelligence model based on the data type information received in action 2320.

[0346] Operation 2340 represents an operation in which a terminal transmits output information of an artificial intelligence model to a network entity based on data type information. For example, if the data type of data to be reported from the terminal to the network entity is set to INT 16, the terminal may transmit data output from the artificial intelligence model to the network entity in the INT16 type. At this time, the output data of the artificial intelligence model on the terminal side may be a test result indicating the performance of the artificial intelligence model on the terminal side, and may be prediction information, vectors, etc. according to artificial intelligence-based operation cases (CSI prediction, CSI compression, beam prediction, etc.).

[0347] In one embodiment, the terminal may perform an operation of transmitting a test result of an output value of an artificial intelligence model based on data type information to a network entity and an operation of receiving information indicating whether to use the determined data type from the network entity. These operations may correspond to operations 638 to 640 of FIG. 6 , operations 1438 to 1440 of FIG. 14 , and operations 1638 to 1640 of FIG. 16 .

[0348] In one embodiment, the terminal may receive a request from a network entity for an AI model capable of applying a determined data type, and perform operations for transmitting structure and weight information for the AI ​​model capable of applying the determined data type to the network entity. These operations may correspond to operations 1336 through 1338 of FIG. 13 .

[0349] In one embodiment, the terminal may receive, from a network entity, the structure and weight information of a first AI model applicable to a first data type, and transmit, to the network entity, the structure and weight information of a second AI model applicable to a second data type. These operations may correspond to operations 1536 to 1540 of FIG. 15 .

[0350] While the operations of the communication method according to the embodiments of the present disclosure have been described separately for each embodiment, the operations included in each embodiment can be combined with operations of other embodiments to form a new embodiment. Accordingly, combined embodiments not described as a single embodiment in the present disclosure can also be understood as being described by the present disclosure.

[0351] The communication method according to embodiments of the present disclosure can be universally used by utilizing a quantized artificial intelligence model in various types of artificial intelligence model-based communication environments, regardless of the form in which the artificial intelligence model is used in communication.

[0352] In addition, the communication method according to the embodiments of the present disclosure can effectively utilize the artificial intelligence model in a communication environment in which there are various problems such as performance degradation of the artificial intelligence model, differences in the artificial intelligence model depending on the device manufacturer, cell-specific artificial intelligence model, and artificial intelligence model support methods depending on the base station.

[0353] The communication method according to embodiments of the present disclosure determines the appropriate data type to be applied to the AI ​​model by considering the terminal's performance and communication conditions, thereby improving transmission and reception efficiency under various conditions. The communication method according to embodiments of the present disclosure can achieve benefits such as improved computational speed, improved low-power performance, and optimized storage space through quantization of the AI ​​model.

[0354] The various embodiments of the present disclosure and the terminology used therein are not intended to limit the technical features described in the present disclosure to specific embodiments, but should be understood to include various modifications, equivalents, or substitutes of the embodiments. In connection with the description of the drawings, similar reference numerals may be used for similar or related components. The singular form of a noun corresponding to an item may include one or more items, unless the relevant context clearly indicates otherwise. In the present disclosure, each of the phrases "A or B," "at least one of A and B," "at least one of A or B," "A, B, or C," "at least one of A, B, and C," and "at least one of A, B, or C" can include any one of the items listed together in the corresponding phrase among the phrases, or all possible combinations thereof. Terms such as "first," "second," or "first" or "second" may be used merely to distinguish one component from another and do not limit the components in any other respect (e.g., importance or order). When a component (e.g., a first component) is referred to as "coupled" or "connected" to another component (e.g., a second component), with or without the terms "functionally" or "communicatively," it means that the component can be connected to the other component directly (e.g., wired), wirelessly, or through a third component.

[0355] The term "module" as used herein may include a unit implemented in hardware, software, or firmware, and may be used interchangeably with terms such as logic, logic block, component, or circuit. A module may be an integrally formed component or a minimum unit or part of a component that performs one or more functions. For example, according to one embodiment, a module may be implemented in the form of an application-specific integrated circuit (ASIC).

[0356] Various embodiments of the present disclosure may be implemented as software (e.g., a program (140)) including one or more commands stored in a storage medium (e.g., an internal memory (136) or an external memory (138)) readable by a machine (e.g., an electronic device (101)). For example, a device (e.g., a processor (e.g., processor (120) of an electronic device (101)) can call at least one command from among one or more commands stored from a storage medium and execute it. This enables the device to operate to perform at least one function according to the called at least one command. The one or more commands may include code generated by a compiler or code executable by an interpreter. A storage medium readable by the device may be provided in the form of a non-transitory storage medium. Here, 'non-transitory' only means that the storage medium is a tangible device and does not contain a signal (e.g., EM wave), and this term does not distinguish between cases where data is stored semi-permanently and cases where it is stored temporarily in the storage medium.

[0357] According to one embodiment, the method according to various embodiments disclosed in the present disclosure may be provided as included in a computer program product. The computer program product may be traded as a product between a seller and a buyer. The computer program product may be distributed in the form of a device-readable storage medium (e.g., a compact disc read-only memory (CD-ROM)), or may be distributed online (e.g., downloaded or uploaded) via an application store (e.g., Play Store™) or directly between two user devices (e.g., smartphones). In the case of online distribution, at least a portion of the computer program product may be temporarily stored or temporarily generated in a device-readable storage medium, such as the memory of a manufacturer's server, an application store's server, or an intermediary server.

[0358] According to various embodiments, each component (e.g., a module or a program) of the described components may include one or more entities. According to various embodiments, one or more components or operations of the aforementioned components may be omitted, or one or more other components or operations may be added. Alternatively or additionally, a plurality of components (e.g., a module or a program) may be integrated into a single component. In such a case, the integrated component may perform one or more functions of each of the plurality of components identically or similarly to those performed by the corresponding component among the plurality of components prior to integration. According to various embodiments, the operations performed by a module, program, or other component may be executed sequentially, in parallel, iteratively, or heuristically, or one or more of the operations may be executed in a different order, omitted, or one or more other operations may be added.

Claims

1. A method performed by a base station in a wireless communication system, A step of receiving, from a terminal, first information including a data type processable by the terminal; A step of determining a data type based on the first information and communication conditions; and Comprising a step of transmitting second information including the determined data type to the terminal, The second information includes one of an integer type and a real number type, and the integer type and the real number type are classified based on the number of bits for expressing data. The above communication conditions are related to at least one of data throughput, latency and accuracy. method.

2. In claim 1, Further comprising a step of receiving data output from an artificial intelligence model of the terminal from the terminal, The above second information is, Including the type of data applied to the artificial intelligence model of the terminal and the type of data output from the artificial intelligence model of the terminal. method.

3. In claim 1, The above method, A step of receiving a test result of an output value of an artificial intelligence model based on the second information from the terminal; A step of determining whether to use the determined data type based on the test results; and Further comprising a step of transmitting information indicating whether the determined data type is used to the terminal; method.

4. In claim 1, The above method, Further comprising a step of transmitting structure and weight information of an artificial intelligence model capable of applying the determined data type to the terminal. method.

5. In claim 1, The above method, A step of requesting information about an artificial intelligence model that can apply the determined data type to the terminal; and Further comprising a step of receiving structure and weight information for the artificial intelligence model capable of applying the determined data type from the terminal. method.

6. In claim 1, The above method, A step of transmitting structure and weight information of a first artificial intelligence model capable of applying a first data type to the terminal; Further comprising a step of receiving structure and weight information of a second artificial intelligence model capable of applying a second data type from the terminal. method.

7. In claim 1, The step of receiving the first information from the terminal comprises: Comprising a step of receiving a request for changing a previously set data type from the terminal, method.

8. As a base station of a wireless communication system, Transmitter and receiver; and Including a processor connected to the above transceiver, The above processor: An operation of receiving first information from a terminal, the first information including a data type processable by the terminal; An operation for determining a data type based on the first information and communication conditions; and It is set to perform an operation of transmitting second information including the determined data type to the terminal, The second information includes one of an integer type and a real number type, and the integer type and the real number type are classified based on the number of bits for expressing data. The above communication conditions are related to at least one of data throughput, latency and accuracy. Base station.

9. In claim 8, The above processor, It is further set to perform an operation of receiving data output from the artificial intelligence model of the terminal from the terminal, The above second information is, Including the type of data applied to the artificial intelligence model of the terminal and the type of data output from the artificial intelligence model of the terminal. Base station.

10. In claim 8, The above processor, An operation for receiving a test result of an output value of an artificial intelligence model based on the second information from the terminal; An action for determining whether to use the determined data type based on the test results above; and It is set to further perform an operation of transmitting information indicating whether the determined data type is used to the terminal. Base station.

11. In claim 8, The above processor, It is further set to perform an operation of transmitting structure and weight information of an artificial intelligence model that can apply the determined data type to the terminal. Base station.

12. In claim 8, The above processor, An operation of requesting information about an artificial intelligence model that can apply the determined data type to the terminal; and It is further set to perform an operation of receiving structure and weight information for the artificial intelligence model that can apply the determined data type from the terminal. Base station.

13. In claim 8, The above processor, An operation of transmitting structure and weight information of a first artificial intelligence model capable of applying a first data type to the terminal; and It is further set to perform an operation of receiving structure and weight information of a second artificial intelligence model that can apply a second data type from the terminal. Base station.

14. A method performed by a terminal in a wireless communication system, A step of transmitting first information including a data type processable by the terminal to the base station; A step of receiving second information indicating a data type determined based on the first information and communication conditions; A step of generating an output value of an artificial intelligence model based on the determined data type; and comprising a step of transmitting the generated output value to the base station; The second information includes one of an integer type and a real number type, and the integer type and the real number type are classified based on the number of bits for expressing data. The above communication conditions are related to at least one of data throughput, latency and accuracy. method.

15. As a user terminal of a wireless communication system, Transmitter and receiver; and Including a processor connected to the above transceiver, The above processor: Transmitting first information including a data type processable by the terminal to the base station; Receive second information indicating a data type determined based on the first information and communication conditions; Generate an output value of an artificial intelligence model based on the determined data type; and The generated output value is set to be transmitted to the base station, The second information includes one of an integer type and a real number type, and the integer type and the real number type are classified based on the number of bits for expressing data. The above communication conditions are related to at least one of data throughput, latency and accuracy. User terminal.

Images