Method for synchronization in wireless communication system and apparatus therefor

Abstract

An operating method of a base station in a wireless communication system according to an embodiment of the present disclosure may comprise the operations of: determining whether synchronization for input data of at least one of a plurality of first artificial intelligence models of a terminal is required; and on the basis of the determination that the synchronization is required, transmitting, to the terminal, an artificial intelligence synchronization indication message for synchronizing the input data of the at least one of the plurality of first artificial intelligence models with input data of a second artificial intelligence model in the associated base station, wherein the first artificial intelligence models have the same neural network structure as the second artificial intelligence model.

Description

Method for synchronization in a wireless communication system and device therefor

[0001] The present disclosure relates to a method for synchronizing input data of an artificial intelligence (AI) model in a wireless communication system and a device therefor.

[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 5G (5th-generation) communication systems, 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 expected to evolve into diverse form factors, including augmented reality glasses, virtual reality headsets, and holographic devices. In the 6th-generation (6G) 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] In a 6G communication system, the maximum transmission speed is terabytes per second (i.e., 1,000 gigabits per second), and the wireless latency is 100 microseconds (μsec). In other words, compared to a 5G communication system, the transmission speed in a 6G communication system is 50 times faster, and the wireless latency is 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 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 experience more severe path loss and atmospheric absorption, making it more important to ensure signal reach, or coverage, in this band. 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 surfaces (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 for uplink and downlink at the same time; network technology that integrates satellites and high-altitude platform stations (HAPS); 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 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 (the next 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 (Truly Immersive 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 find application in diverse fields such as industry, medicine, automobiles, and home appliances.

[0007] Meanwhile, in wireless communication systems such as 5G or 6G, the introduction of AI models into devices (e.g., base stations and / or terminals) is expected to reduce overhead between each entity. In wireless communication systems, AI models based on AI algorithms such as machine learning or predictive analytics can dynamically adjust wireless communication parameters based on changing network conditions, user behavior, or other requirements. These AI models can reduce complex signal transmission and reception procedures in wireless communication environments. To achieve this, the AI ​​models applied to each function on both devices performing wireless communication must be able to produce substantially identical or similar inference results.

[0008] One embodiment of the present disclosure provides a method for synchronizing input data of an AI model applied to devices in a wireless communication system and a device therefor.

[0009] One embodiment of the present disclosure provides a method for synchronizing input data for an AI model provided in a base station and a terminal in a wireless communication system and a device therefor.

[0010] One embodiment of the present disclosure provides a method and a device therefor for identifying input mismatch in an AI model applied to an end device performing wireless communication in a wireless communication system and synchronizing input data for the AI ​​model.

[0011] One embodiment of the present disclosure provides a method and device for synchronizing input data for an AI model by dynamically adjusting wireless communication parameters to be applied to the AI ​​model in a wireless communication system.

[0012] In a wireless communication system according to one aspect of the present disclosure, a method of operating a first device may include an operation of determining whether synchronization is required for first input data of a first artificial intelligence model shared with a second device for one function, and an operation of transmitting, in response to a determination that synchronization is required, a first synchronization instruction message to the second device, the first synchronization instruction message instructing synchronization of second input data of a second artificial intelligence model having substantially the same neural network structure as the first artificial intelligence model with the first input data.

[0013] A method of operating a terminal in a wireless communication system according to one aspect of the present disclosure may include receiving a synchronization instruction message from a base station for synchronizing input data of at least one first artificial intelligence model among a plurality of first artificial intelligence models with input data of a second artificial intelligence model of the base station, and initializing, newly defining, or modifying input data for at least one of the plurality of first artificial intelligence models based on the received synchronization instruction message.

[0014] By synchronizing input data of an AI model used in wireless communication devices in a wireless communication system according to one embodiment of the present disclosure, it is possible to reduce deterioration of wireless communication quality and / or power consumption performance due to input mismatch.

[0015] The effects that can be obtained from the exemplary embodiments of this specification are not limited to the effects mentioned above, and other effects not mentioned can be clearly derived and understood by a person having ordinary skill in the technical field to which the exemplary embodiments of this specification belong, from the description below. In other words, unintended effects that come from practicing the exemplary embodiments of this specification can also be derived by a person having ordinary skill in the technical field from the exemplary embodiments of this specification.

[0016] FIG. 1 is a configuration diagram of a wireless communication system that supports wireless communication based on an AI model according to one embodiment of the present disclosure.

[0017] FIG. 2 is a structural diagram of a wireless communication system for supporting wireless communication based on an AI model according to one embodiment of the present disclosure.

[0018] FIG. 3 illustrates an example of a signaling procedure performed between wireless communication devices in a wireless communication system according to one embodiment of the present disclosure.

[0019] FIG. 4 illustrates an example of a neural network model for implementing a wireless communication system based on an AI model according to one embodiment of the present disclosure.

[0020] FIG. 5 is a flowchart of an operation method of a wireless communication system according to one embodiment of the present disclosure.

[0021] FIG. 6 is a flowchart of an operation method of a wireless communication system according to one embodiment of the present disclosure.

[0022] FIG. 7 is a flowchart of an operation method of a wireless communication system according to one embodiment of the present disclosure.

[0023] FIG. 8 is a flowchart of an operation method of a wireless communication system according to one embodiment of the present disclosure.

[0024] Figure 9 is a flowchart of an operating method according to one embodiment of the present disclosure.

[0025] FIGS. 10A and 10B illustrate an example of a base station-driven synchronization procedure using a synchronization indication message (e.g., a synchronization input reset (AIR) message) according to one embodiment of the present disclosure.

[0026] FIGS. 11A and 11B illustrate an example of a base station-driven synchronization procedure using a synchronization indication message (e.g., a synchronization input definition (AID) message) according to one embodiment of the present disclosure.

[0027] FIG. 12 illustrates an example of a terminal-driven synchronization procedure using a synchronization instruction message (e.g., a synchronization input definition (AID) message) according to one embodiment of the present disclosure.

[0028] FIG. 13 illustrates an example of a terminal-driven synchronization procedure using a synchronization instruction message (e.g., a synchronization input definition (AID) message) according to one embodiment of the present disclosure.

[0029] FIGS. 14A and 14B illustrate an example of a base station-driven synchronization procedure using a synchronization confirmation message (AC (AIR / AID confirmation) message) according to one embodiment of the present disclosure.

[0030] FIG. 15 illustrates an example of a terminal-driven synchronization procedure using a synchronization request message (AR (AIR / AID request) message) according to one embodiment of the present disclosure.

[0031] FIG. 16 illustrates an example of a base station-driven synchronization procedure using an AI input gather (AIG) message according to one embodiment of the present disclosure.

[0032] Figure 17a shows an experimental data graph regarding the rate at which decoding failures occur according to normalized throughput in a wireless communication system.

[0033] Figure 17b illustrates the relationship between input mismatch and normalized throughput in a wireless communication system over time.

[0034] The following description refers to the attached drawings, and specific examples of implementations are illustrated within the drawings. Furthermore, other examples may be utilized and structural changes may be made without departing from the scope of the various examples.

[0035] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings so that those skilled in the art can easily implement the present disclosure. However, the present disclosure may be implemented in various different forms and is not limited to the embodiments described herein. In connection with the description of the drawings, the same or similar reference numerals may be used for identical or similar components. Furthermore, in the drawings and related descriptions, descriptions of well-known functions and configurations may be omitted for clarity and conciseness.

[0036] FIG. 1 is a configuration diagram of a wireless communication system (100) that supports wireless communication based on an AI model according to one embodiment of the present disclosure.

[0037] Referring to FIG. 1, as an example, a wireless communication system (100) may support wireless communication based on a cellular network, which may include a plurality of cells. Each of the plurality of cells may, for example, be a service area for wireless communication. The illustrated wireless communication system (100) may include a base station (110) and / or a plurality of terminals (120a, 120b, 120c, 120d) corresponding to one cell. The base station (110) and / or the plurality of terminals (120a, 120b, 120c, 120d) may be a wireless communication device or an end device that performs wireless communication.

[0038] A plurality of terminals (120a, 120b, 120c, 120d) included in a wireless communication system (100) may be in the same or different operating states (e.g., active state, inactive state, or idle state). The plurality of terminals (120a, 120b, 120c, 120d) may perform operations according to the operating states. For example, the plurality of terminals (120a, 120b, 120c, 120d) may include one or more terminals in the same operating state. Hereinafter, the term "terminal (120)" may be used to refer to all of the plurality of terminals (120a, 120b, 120c, 120d), or to refer to at least one specific or target terminal. The at least one specific or target terminal may be a terminal having substantially the same or similar operating state.

[0039] A base station (110) is a network infrastructure that provides wireless access to terminals (120). The base station (110) has coverage defined as a specific geographic area based on the distance at which it can transmit signals. The coverage may be a service area corresponding to the corresponding cell. In addition to the base station, the base station (110) may be referred to as an 'access point (AP)', 'eNodeB (eNB)', '5th generation node', 'next Generation NodeB (gNB)', '5G NodeB (5gNB)', 'wireless point', 'transmission / reception point (TRP)', 'digital unit (DU)', 'radio unit (RU), remote radio head (RRH)' or other terms having equivalent technical meanings.

[0040] The terminal (120) is a device used by a user and performs communication with the base station (110) via a wireless channel. In some cases, the terminal (120) may be operated without the user's involvement. That is, the terminal (120) is a device that performs machine type communication (MTC) and may not be carried by the user. The terminal (120) may be referred to as a 'user equipment (UE)', a 'mobile station', a 'subscriber station', a 'remote terminal', a 'wireless terminal', an 'electronic device', a 'user device', or other terms having an equivalent technical meaning. A terminal (e.g., terminal 120) may include at least one of, for example, a cellular phone, a smart phone, a computer, a tablet PC, a mobile phone, a video phone, an e-book reader, a desktop PC, a laptop PC, a netbook computer, a workstation, a server, a PDA, a portable multimedia player (PMP), an MP3 player, a medical device, a camera, a wearable device, or a multimedia system capable of performing a communication function. Furthermore, the type of the terminal is not limited to the above examples.

[0041] To improve channel gain, the base station (110) or the terminal (120) may perform beamforming. Beamforming may include transmit beamforming or receive beamforming. For example, the base station (110) may perform transmit beamforming for signal transmission via downlink. For example, the base station (110) may perform receive beamforming for signal reception via uplink. For example, the terminal (120) may perform receive beamforming for signal reception via downlink. For example, the terminal (120) may perform transmit beamforming for signal transmission via uplink. That is, the base station (110) or the terminal (120) may provide directionality to the transmit signal or the receive signal. For this purpose, the base station (110) or the terminal (120) may select serving beams through a beam search procedure or a beam management procedure. After serving beams are selected, the base station (110) or terminal (120) can perform communication through resources that are in a quasi-co-located (QCL) relationship with the resources that transmitted the serving beams. For example, the base station (110) or terminal (120) can transmit and receive wireless signals in a millimeter wave band (e.g., 28 GHz, 30 GHz, 38 GHz, 60 GHz).

[0042] The terminal (120) can perform a synchronization process and / or a cell search process through a primary synchronization signal (PSS), a secondary synchronization signal (SSS), and a physical broadcast channel (PBCH). The terminal (120) can perform an access process to complete connection to a network through a base station (110). The terminal (120) can transmit a preamble through a physical random access channel (PRACH) and receive a response message to the preamble. The terminal that has performed the above-described process can receive a downlink control channel (e.g., a physical downlink control channel (PDCCH)) or a downlink data channel (e.g., a physical downlink shared channel (PDSCH)) as an uplink / downlink signal transmission process, or can transmit an uplink control channel (e.g., a physical uplink control channel (PUCCH)) or an uplink data channel (e.g., a physical uplink shared channel (PUSCH)). These access procedures can be used for various purposes, such as initial connection, uplink synchronization coordination, resource allocation, or handover.

[0043] The wireless communication system (100) may employ an AI model to perform at least one function among the functions of a wireless communication service. For example, the AI ​​model applied to the wireless communication system (100) may be trained using input data at a base station (110). For example, the AI ​​model applied to the wireless communication system (100) may be trained using input data at terminals (120). The AI ​​model may evolve through learning through training. The AI ​​model may be defined for each of the various functions implemented in the wireless communication system (100). The AI ​​model applied to the wireless communication system (100) may, for example, aim to improve network performance and user experience. For example, the AI ​​model applied to the wireless communication system (100) may use a machine learning algorithm to predict network traffic. For example, the AI ​​model applied to the wireless communication system (100) may optimize resource allocation. As an example, the AI ​​model applied to the wireless communication system (100) may dynamically adjust communication parameters using reinforcement learning.

[0044] The wireless communication system (100) can perform efficient operations without complex signal transmission / reception in a wireless communication environment through AI technology. For example, in the wireless communication system (100), the base station (110) and the terminal (120) can use AI models that provide specific functions together to perform operations corresponding to the same output data through the same input data, thereby reducing overhead due to signal transmission. In other words, the wireless communication system (100) using a synchronous AI model can perform corresponding operations through the same output data resulting from the same input data, thereby reducing overhead due to signal transmission.

[0045] As an example, in a wireless communication system (100), an AI model may be implemented in each of a base station (110) and / or a terminal (120) for each function to obtain downlink control information (DCI) or uplink control information (UCI).

[0046] The AI ​​model provided for acquiring the DCI may use channel quality indication (CQI) information that may be acquired or provided by the terminal (120) as input data. The AI ​​model provided for acquiring the DCI may use scheduling information that may be acquired or provided by the base station (120) as input data. The AI ​​model provided for acquiring the DCI may use CQI information and scheduling information as input data. The AI ​​model may evolve to service the corresponding function through learning by input data including CQI information and / or scheduling information corresponding to the input data. The evolution means that the probability of acquiring a valid DCI may be gradually increased through iterative learning through training. The result data by the iterative learning and / or new input data acquired for the iterative learning may be accumulated as input data for the corresponding AI model.

[0047] The AI ​​model for obtaining the above DCI can be applied to each of the base station (110) and the terminal (120). The base station (110) and the terminal (120) must be able to share input data for the AI ​​model. For example, the terminal (120) can measure channel quality and report CQI information corresponding to the measured channel quality to the base station (110). The PF (proportional fair) scheduler of the base station (110) can obtain scheduling information using the CQI information. The base station (110) can transmit the obtained scheduling information to the terminal (110).

[0048] The base station (110) can input CQI information and scheduling information of each terminal (120) into an AI model activated for the terminal to obtain DCI for the terminal. The terminal (120) can input its own CQI information and scheduling information into the AI ​​model activated to perform the corresponding function to obtain DCI for itself. If the AI ​​models of the base station (110) and the terminal (120) have evolved through the same training process and the input data is synchronized, the base station (110) and the terminal (120) will obtain substantially the same DCI. The fact that the AI ​​models have evolved through the same training process may mean, for example, that newly acquired or changed data are accumulated as input data during the training process. For example, the DCI output from the AI ​​model may indicate whether a DCI for the terminal exists in a specific slot (e.g., '1' if a DCI exists or '0' if a DCI does not exist).

[0049] The above-described operation can be applied substantially similarly or identically to an AI model for obtaining UCI, although the type of input data may be different.

[0050] For example, in a wireless communication system (100), an AI model may be implemented for each function for allocating uplink resources or allocating downlink resources.

[0051] The AI ​​model designed to allocate the above uplink resources can use as input data information such as the type of service or traffic volume that the terminal (120) intends to use via uplink, as well as information such as the uplink resource usage amount at the base station (120). The AI ​​model can evolve to provide the corresponding function through learning based on information that can be considered when allocating uplink resources corresponding to the input data. The evolution means that the probability of effective uplink resource allocation can be gradually increased through repeated learning through training.

[0052] The AI ​​model for allocating the above uplink resources can be applied to each of the base station (110) and the terminal (120). The base station (110) and the terminal (120) must be able to share input data for the AI ​​model. For example, the terminal (120) can report information such as the type of service it wants to use or the amount of traffic to the base station (110). The base station (110) can transmit information regarding the amount of uplink resources used or the extent of allocable uplink resources to the terminal (110).

[0053] The base station (110) can input information about uplink resource allocation of each terminal (120) and its own uplink resource-related information into an AI model activated for each terminal to obtain uplink resources to allocate for the terminal. The terminal (120) can input information about its own uplink resource allocation and the uplink resource-related information of the base station (110) into an AI model activated to perform the corresponding function to obtain uplink resources that can be allocated for the terminal. If the AI ​​models in the base station (110) and the terminal (120) have evolved through the same training process and the input data is synchronized, the base station (110) and the terminal (120) will allocate substantially the same or similar uplink resources.

[0054] The above-described operation can be applied substantially similarly or identically to an AI model for acquiring downlink resources, although the type of input data may be different.

[0055] In addition, in the wireless communication system (100), an AI model for controlling uplink transmission power or a control function for discontinuous reception (DRX) may be implemented for each function for controlling downlink transmission power. The AI ​​model provided for the control function for DRX may provide a DRX cycle, duration length, or inactive length based on input data such as the operating state of the terminal (120) and learned data, for example.

[0056] As described above, in the wireless communication system (100), it is necessary to assume that the input data to be used for the same AI model between the base station (110) and the terminal (120). However, due to the characteristics of wireless communication, the transmission / reception of signals may not be perfect due to channel changes and / or various factors (e.g., see FIG. 17a). This may cause a mismatch in the input data (e.g., input mismatch) between the base station (110) and the terminal (120). The input mismatch may cause a phenomenon in which the input data entering each AI model of the base station (110) and the terminal (120) are mismatched. The input mismatch accumulates over time, leading to a phenomenon in which the operation according to the output data of the AI ​​model is mismatched, and as a result, a malfunction may occur in the AI ​​model (e.g., see FIG. 17b).

[0057] Figure 17a shows an experimental data graph regarding the rate at which decoding failures occur according to normalized throughput (w) in a wireless communication system (100).

[0058] Figure 17b illustrates the relationship in which input mismatch corresponding to normalized throughput changes over time in a wireless communication system (100).

[0059] Accordingly, in the wireless communication system (100), a synchronization procedure can be provided to synchronize input data for the same AI model applied to the base station (110) and the terminal (120) so that input mismatch does not occur.

[0060] For example, in order for a base station (110) and a terminal (120) in a wireless communication system (100) to independently apply an AI model to perform a specific function, the input data at the base station (110) and the input data at the terminal (120) for the corresponding AI model must be substantially identical. However, in an actual wireless communication environment, a situation may arise in which the accumulated input data is inconsistent due to reasons such as a transmission error or an error in internal information processing during the process in which the base station (110) and the terminal (120) share information. The accumulated input data may correspond to input data that has been changed or modified due to continued use of the target AI model in the base station (110) or the terminal (120) to which the target AI model is applied, for example. If the input data for a specific AI model at the base station (110) and the terminal (120) do not match, even if the specific AI model has evolved through the same training process, the result data obtained through the AI ​​model of the base station (110) may be different from the result data obtained through the AI ​​model of the terminal (120).

[0061] According to one example, the at least one specific or target terminal (120) can process a corresponding function (e.g., identification of DCI presence, etc.) with a counterpart device (e.g., base station (110)) based on an AI model. The at least one specific or target terminal (120) and the counterpart device (e.g., base station (110)) can use input data when processing the corresponding function through the AI ​​model. The input data may change due to learning or evolution of the corresponding AI function. The input data that changes may be referred to as 'accumulated input data'. As the accumulated input data continuously processes the corresponding function by the corresponding AI function, a situation in which a mismatch occurs between the at least one specific or target terminal (120) and the counterpart device (e.g., base station (110)). This may be referred to as 'input mismatch'.

[0062] Therefore, the base station (110) and / or the terminal (120) need to perform a procedure for synchronizing input data for each AI model they share. For convenience, in the following description, the device that identifies the need for input data synchronization among the base station (110) or the terminal (120) and initiates the input data synchronization procedure based on this will be referred to as the 'first device'. The counterpart device of the first device that initiated the input data synchronization procedure will be referred to as the 'second device'. However, it should be noted that in certain embodiments, the roles of the first device and the second device may be reversed. In addition, in the description to be described below, the AI ​​model will be used to mean an AI model activated in the terminal (120) for the purpose of performing a specific function (e.g., a function of predicting whether DCI exists) or an AI model activated in the base station (110) for the terminal (120).

[0063] In one example, the first device and the second device can perform a synchronization procedure to match accumulated input data for the corresponding AI model, either periodically or aperiodically. The synchronization procedure can resolve input mismatch. The synchronization procedure can be performed between the first and second devices (110, 120), for example, to reset the accumulated input data to the initial input data. The synchronization procedure can be performed between the first and second devices (110, 120), for example, to change or modify the accumulated input data to predetermined input data.

[0064] More specifically, the first device can identify whether synchronization of input data for a specific AI model is required, and based on this, initiate a procedure for synchronizing the input data for the specific AI model. A situation in which synchronization of the input data for the specific AI model is required may correspond to a situation in which an input mismatch occurs in which the input data do not match. A situation in which synchronization of the input data for the specific AI model is required may correspond to the arrival of a preset cycle in preparation for the occurrence of an input mismatch in which the input data do not match. The input mismatch may cause information acquired through the AI ​​model applied to the first device to be inconsistent with information acquired through the AI ​​model applied to the second device.

[0065] The first device may perform a synchronization procedure when it identifies a situation in which synchronization of input data is required. In one example, the synchronization procedure may be performed to ensure that a first AI model in the first device and a second AI model in the second device use substantially the same input data. In one example, the synchronization procedure may be performed in response to a discrepancy between first input data accumulated corresponding to the first AI model and second input data accumulated corresponding to the second AI model. In one example, the synchronization procedure may be performed in response to a discrepancy between processing result data by the first AI model and processing result data by the second AI model.

[0066] The above synchronization procedure may be a procedure for synchronizing the first input data accumulated corresponding to the first AI model and the second input data accumulated corresponding to the second AI model to substantially the same input data. As an example, the synchronization procedure may be a procedure for initializing the first input data accumulated corresponding to the first AI model and the second input data accumulated corresponding to the second AI model. As an example, the synchronization procedure may be a procedure for changing or modifying the first input data accumulated corresponding to the first AI model and the second input data accumulated corresponding to the second AI model to preset input data.

[0067] Various types of messages can be used in the above synchronization procedure. Table 1 below defines examples of messages that can be used in the synchronization procedure.

[0068] Message Classification FunctionSynchronization Input Reset Message (AIR (AI input reset) message) A message that instructs to reset the accumulated input data corresponding to the target AI model for input data synchronizationSynchronization Input Define Message (AID (AI input define) message) A message that instructs to define the accumulated input data corresponding to the target AI model as input data appropriate for the situation for input data synchronizationSynchronization Approval Request Message (AUR (AIR / AID uplink request) message) A message that requests approval for transmitting a synchronization instruction message to the uplinkSynchronization Approval Response Message (AUG (AIR / AID uplink grant) message) A message that grants approval for transmitting a synchronization instruction message to the uplink in response to the synchronization approval request messageSynchronization Confirmation Message (AC (AIR / AID confirmation) message) A message that confirms that input data synchronization will be performed in response to the synchronization instruction messageSynchronization Request Message (AR (AIR / AID request) message) A message that requests to instruct to synchronize the input data for the target AI modelSynchronization Collection A message to inform the period (e.g., time offset) for data gathering (gathering) for sending a synchronization instruction message that instructs the synchronization of input data for the target AI model (AIG (AI input gather) message)

[0069] In the above , the 'synchronization input reset message' or 'synchronization input definition message' may be collectively referred to as a 'synchronization instruction message'. The synchronization input instruction message may be a message transmitted by a specific device (e.g., a first device) to instruct a counterpart device (e.g., a second device) to synchronize accumulated input data for a target AI model.

[0070] If the first device identifies a situation requiring synchronization of input data, it may transmit a synchronization instruction message to the second device. For example, the first device may transmit a synchronization instruction message to the second device to instruct it to perform synchronization of input data for a target AI model that has encountered one or multiple consecutive input mismatches.

[0071] In one example, a first device may transmit a first synchronization instruction message (e.g., a synchronization input reset message) to a second device, which instructs the second device to initialize accumulated input data for a target AI model with initialization data (e.g., see FIGS. 10A and 10B ). The first device may, for example, periodically transmit the first synchronization instruction message to the second device considering a preset cycle (e.g., see FIG. 10A ). The first device may, for example, aperiodically transmit the first synchronization instruction message to the second device in response to acquiring an event regarding an input mismatch (e.g., see FIG. 10B ). The event regarding an input mismatch may be acquired by detecting a single error or a predetermined number of consecutive errors in the output of a specific AI model.

[0072] As an example, a first device (e.g., terminal (120)) may perform a procedure to obtain permission from a second device (e.g., base station (110)) before transmitting a first synchronization instruction message (e.g., see FIG. 13). For example, the first device may transmit a message (e.g., a synchronization approval request message) to the second device inquiring about performing a synchronization procedure of input data. The second device may transmit a message (e.g., a synchronization approval response message) to the first device in response to the inquiry from the first device inquiring about performing a synchronization procedure of input data. The first device may transmit the first synchronization instruction message to the second device in response to receiving a message from the second device inquiring about performing a synchronization procedure of input data.

[0073] As an example, a first device (e.g., base station (110)) may perform a procedure of transmitting a first synchronization instruction message to a second device (e.g., terminal (120)) in response to a request from a second device (e.g., see FIG. 15). This is based on the premise that the second device determines that synchronization of input data for a specific AI model is required. For example, the second device may transmit a synchronization request message to the first device in response to determining that synchronization of input data for a specific AI model is required. The synchronization request message may be an AR message (AIR / AID request message). The first device may transmit the first synchronization instruction message to the second device in response to receiving the synchronization request message.

[0074] As an example, a first device (e.g., a base station (110)) may perform a procedure for collecting input data for a target AI model and / or data for transmitting a synchronization instruction message in response to receiving a synchronization request message from a second device (e.g., a terminal (120)) (e.g., see FIG. 16). This is based on the premise that the second device determines that synchronization of input data for the target AI model is necessary. For example, the first device may transmit a synchronization collection message to the second device in response to receiving the synchronization request message. The synchronization collection message may be an artificial intelligence gather (AIG) message. The synchronization collection message may include time period information for collecting data for synchronizing the input data. The time period information may be set in units of slots, for example. The first device may transmit a first synchronization instruction message to the second device in response to the elapse of a time period set for data collection.

[0075] The initialization input data for the above reset can be shared by the first device and the second device. The initialization input data for the reset can be, for example, agreed upon in advance between the first device and the second device. In this case, the first synchronization instruction message can be an AIR message (Artificial Intelligence Input Reset Message) that includes a model identifier (ID) indicating the target AI model or an identifier indicating a reset to the initialization input data.

[0076] The initialization input data for the above reset can be provided by the first device to the second device. The initialization input data for the reset can be transmitted by the first device by including it in a first synchronization instruction message, for example. In this case, the first synchronization instruction message can be an AIR message including a model identifier (ID) indicating the target AI model or initialization input data.

[0077] The first device may transmit a first synchronization instruction message including index information indicating the initialization input data in order to share the initialization input data with the second device. In this case, the first synchronization instruction message may be an AIR message including a model identifier (ID) indicating a target AI model or index information of initial input data that can identify the initialization input data.

[0078] According to the above-described procedure, the first device transmitting the first synchronization instruction message and the second device receiving the first synchronization instruction message can reset the accumulated input data corresponding to the target AI model to initialization input data.

[0079] For example, the first device may receive a synchronization confirmation message from the second device in response to the first synchronization instruction message. The synchronization confirmation message may be a first AC message (AIR confirmation message). After receiving the synchronization confirmation message, the first device may reset accumulated input data for the target AI model to initialization input data (e.g., see FIGS. 14A and 14B). After transmitting the synchronization confirmation message, the second device may reset accumulated input data for the target AI model to initialization input data.

[0080] In one example, a first device may transmit a second synchronization instruction message (e.g., a synchronization input definition message) to a second device, which instructs the second device to change or modify accumulated input data for a target AI model into predetermined input data (e.g., FIGS. 11A, 11B, and 12). The first device may, for example, periodically transmit the second synchronization instruction message to the second device considering a preset cycle. The first device may, for example, aperiodically transmit the second synchronization instruction message to the second device in response to acquiring an event regarding an input mismatch. The event regarding the input mismatch may be acquired by detecting a single error or a predetermined number of consecutive errors in the output of a specific AI model.

[0081] As an example, a first device (e.g., terminal (120)) may perform a procedure to obtain permission from a second device (e.g., base station (110)) before transmitting a second synchronization instruction message (e.g., see FIG. 13). For example, the first device may transmit a message (e.g., a synchronization approval request message) to the second device inquiring about performing a synchronization procedure for input data. The second device may transmit a message (e.g., a synchronization approval response message) to the first device in response to the inquiry from the first device inquiring about performing a synchronization procedure for input data. The first device may transmit a second synchronization instruction message to the second device in response to receiving a message from the second device inquiring about performing a synchronization procedure for input data.

[0082] As an example, a first device (e.g., base station (110)) may perform a procedure of transmitting a second synchronization instruction message to a second device (e.g., terminal (120)) in response to a request from a second device (e.g., see FIG. 15). This is based on the premise that the second device determines that synchronization of input data for a specific AI model is required. For example, the second device may transmit a synchronization request message to the first device in response to determining that synchronization of input data for a specific AI model is required. The synchronization request message may be an AR message. The first device may transmit a second synchronization instruction message to the second device in response to receiving the synchronization request message.

[0083] As an example, a first device (e.g., base station (110)) may perform a procedure for collecting input data for a target AI model and / or data for transmitting a synchronization instruction message in response to receiving a synchronization request message from a second device (e.g., terminal (120)) (e.g., see FIG. 16). This is based on the premise that the second device determines that synchronization of input data for the target AI model is necessary. For example, the first device may transmit a synchronization collection message to the second device in response to receiving the synchronization request message. The synchronization collection message may be an AIG message. The synchronization collection message may include information regarding a time period for collecting data for synchronizing the input data. The information regarding the time period may be set in units of slots, for example. The first device may transmit a second synchronization instruction message to the second device in response to the elapse of a time period set for data collection.

[0084] The input data to be changed or modified (hereinafter referred to as “new input data”) may be shared by the first device and the second device. The new input data may, for example, be agreed upon in advance between the first device and the second device. In this case, the second synchronization instruction message may be an AID message (Artificial Intelligence Input Define Message) that includes a model identifier (ID) indicating a target AI model or an identifier (or index) indicating new input data.

[0085] The new input data may be provided by the first device to the second device. For example, the new input data may be transmitted by the first device by including it in a second synchronization instruction message. In this case, the second synchronization instruction message may be a model identifier (ID) indicating the target AI model or an AID message containing the new input data.

[0086] The first device may transmit a second synchronization instruction message including index information indicating the new input data to share the new input data with the second device. In this case, the second synchronization instruction message may be an AID message including a model identifier (ID) indicating the target AI model or index information that can identify the new input data.

[0087] According to the above-described procedure, the first device transmitting the second synchronization instruction message and the second device receiving the second synchronization instruction message can change or modify the accumulated input data corresponding to the target AI model into new input data.

[0088] For example, the first device may receive a synchronization confirmation message from the second device in response to the second synchronization instruction message (e.g., see FIGS. 14A and 14B ). The synchronization confirmation message may be a second AC message (AID confirmation message). After receiving the synchronization confirmation message, the first device may change or modify accumulated input data for the target AI model to new input data. After transmitting the synchronization confirmation message, the second device may change or modify accumulated input data for the target AI model to new input data.

[0089] FIG. 2 is a structural diagram of a wireless communication system (200) for supporting wireless communication based on an AI model according to one embodiment of the present disclosure.

[0090] Referring to FIG. 2, as an example, a wireless communication system (200) may include one or more base stations (210a, 210b) (e.g., base station (110) of FIG. 1) or a core network (CN) (220) (e.g., evolved packet core (EPC) or New Generation Core (NGC)). One or more base stations (210a, 210b) may be connected to the CN (220). For example, the base stations (210a, 210b) may include an eNB corresponding to a base station in an LTE network or a gNB corresponding to a base station in an NR network. For example, the base stations (210a, 210b) may also be devices unrelated to a radio access technology (RAT), such as an AP in a WiFi network.

[0091] One or more base stations (210a, 210b) may be configured as a single unit or may be divided into various units. The various units constituting a single base station may support mobile communication functions corresponding to the role of the base station. The mobile communication functions may include, for example, functions for each PDCP / RLC / MAC / PHY / RF layer. One unit may support one or more functions. Multiple units may support one or more functions in a distributed manner.

[0092] One or more base stations (210a, 210b) may be interconnected with other base stations via an interface such as X2 or Xn. For example, the base stations (210a, 210b) may be interconnected with each other via an X2 or Xn interface. In one example, the base stations (210a, 210b) may be interconnected with a CN (220) (e.g., NGC) via an interface such as S1 or NG. The interface such as S1 or NG may be an interface between the base station and the core network. More specifically, the one or more base stations (210a, 210b) may be interconnected with an Access and Mobility Management Function (AMF) via an N2 interface. The one or more base stations (210a, 210b) may be interconnected with a User Plane Function (UPF) via an N3 interface.

[0093] FIG. 3 illustrates an example of a signaling procedure performed between wireless communication devices in a wireless communication system (e.g., the wireless communication system (100) of FIG. 1) according to one embodiment of the present disclosure.

[0094] Referring to FIG. 3, in a wireless communication system (100), a terminal (UE) (120) (e.g., one of the terminals 120a, 120b, 120c, and 120d of FIG. 1) can receive information from a base station (BS) (110) (e.g., the base station (110) of FIG. 1) via a downlink. The terminal (120) can transmit information to the base station (110) via an uplink. The information transmitted and received by the base station (110) and / or the terminal (120) may include data and / or various control information. The physical channels in the base station (110) and / or the terminal (120) may exist in various ways depending on the type and / or purpose of the information transmitted and received.

[0095] In operation 301, the terminal (120) may perform an initial cell search to synchronize with the base station (110). The terminal (120) may perform the initial cell search, for example, in response to being powered on or entering a new cell. During the initial cell search process, the terminal (120) may receive a synchronization signal (e.g., a primary synchronization signal (PSS) or a secondary synchronization signal (SSS)) from the base station (110) to synchronize with the base station. During the initial cell search process, the terminal (120) may obtain information such as a cell identifier that can identify the base station (110). The terminal (120) may receive a physical broadcast channel (PBCH) from the base station (110) identified based on the synchronization signal. The terminal (120) may obtain broadcast information (e.g., a master information block: MIB) of the cell transmitted by the base station (110) through the PBCH. During initial cell search, the terminal (120) can receive a downlink reference signal (DL RS) to identify the downlink channel status. As a result, the terminal (120) can complete downlink synchronization with the base station (110).

[0096] In operation 302, the terminal (120) may receive system information (SI) via a downlink physical channel (e.g., PDCCH or PDSCH). For example, the terminal (120) may identify a radio resource on which a PDSCH is to be transmitted via the PDCCH, and receive system information via the PDSCH that uses the identified radio resource. The system information may include, for example, a lot of information that the terminal (120) requires for wireless communication with the base station (110).

[0097] When the terminal (120) acquires system information through synchronization for the downlink, it can perform a random access procedure for the base station (110) using the system information. The random access procedure may include multiple operations (e.g., operations 303, 304, 305, and 306) for implementing random access.

[0098] Specifically, in operation 303, the terminal (120) can transmit a preamble corresponding to a specific sequence via a physical random access channel (PRACH) (msg1). The terminal (120) can obtain resources for the PRACH from a first system information block (SIB1: system information block 1) received via a PDSCH. The preamble is a unique signal used for the initial connection between the terminal (120) and the base station (110) in wireless communication. Based on the preamble, the terminal (120) can notify the base station (110) of its presence and the start of wireless communication.

[0099] In operation 304, the terminal (120) may receive a random access response (RAR) for the preamble signal via the PDCCH or its corresponding PDSCH (msg2). The RAR may be a response that notifies the terminal (120) that the base station (110) is ready to start communication in response to recognizing the preamble of the terminal (120).

[0100] In operation 305, the terminal (120) may transmit an RRC connection request message (msg3) to the base station (110) using scheduling information in the RAR. The terminal (120) may receive a response message (msg4) corresponding to the RRC connection request message (msg3) from the base station (110). The terminal (120) may transmit a PUSCH to the base station (110) using information obtained from the response message (msg4). The scheduling information may include, for example, information indicating when or which frequency band the base station (110) will use to transmit data to the terminal (120). As a result, the terminal (120) may complete uplink synchronization with the base station (110).

[0101] In the case of a contention-based RACH, the terminal (120) may additionally perform a contention resolution procedure at operation 306. Specifically, the terminal (120) may perform a contention resolution procedure, such as receiving a PDCCH and its corresponding PDSCH. This may be a process for determining whether to grant priority to a specific terminal when multiple terminals attempt to access the base station (110) simultaneously using the same preamble. This process is necessary for efficient use of resources in wireless communications.

[0102] The terminal (120) that has performed at least some of operations 301 to 306 may perform a procedure for receiving or transmitting a signal via uplink / downlink. For example, the terminal (120) may perform a downlink signal reception procedure for receiving a signal via PDCCH and / or PDSCH in operation 307. For example, the terminal (120) may perform an uplink signal transmission procedure for transmitting a signal via PUSCH and / or PUCCH in operation 308.

[0103] According to an example, the terminal (120) can receive downlink control information (DCI) via a PDCCH. The terminal (120) can obtain control information such as resource allocation information from the DCI. The DCI can be configured in different formats depending on the intended use. The intended use may include, but is not limited to, PUSCH scheduling within a cell, one or multiple PUSCH scheduling within a cell, or indication of cell group downlink feedback information for the terminal, PUSCH scheduling within a cell, PDSCH scheduling within a downlink cell, and / or PDCSH scheduling within a cell. In addition, the terminal (120) may also transmit uplink control information (UCI) to the base station (110). The UCI may include at least one of HARQ-ACK / NACK (hybrid automatic repeat and request acknowledgement / negative-ACK), SR (scheduling request), CQI (channel quality indication), PMI (precoding matrix indication), RI (rank indication), and / or BI (beam indication).

[0104] In the signaling method of the wireless communication system (100) described above, an AI model may be used. The AI ​​model may be used in at least part of the signaling procedure. At least one function of the base station (110) and / or terminal (120) of the wireless communication system (100) may be performed on behalf of or supported by the AI ​​model. Below, a neural network model applicable to the wireless communication system (100) is exemplarily illustrated.

[0105] FIG. 4 illustrates an example of a neural network model for implementing a wireless communication system (e.g., the wireless communication system (100) of FIG. 1) based on an AI model according to one embodiment of the present disclosure.

[0106] A wireless communication system (100) according to one embodiment of the present disclosure may support an AI function. The AI ​​function may be implemented by an AI model (415) provided in a base station (410) (e.g., the base station (110) of FIG. 1) or a terminal (420a, 420b) (e.g., at least one terminal (120) among the terminals (120a, 120b, 120c, 120d) of FIG. 1). Hereinafter, the term “terminal (420)” may be used to refer to all of the terminals (420a, 420b) or to refer to at least one specific terminal. The AI ​​model (415) may perform learning, such as deep learning, based on information (413, 423) input by the base station (410) and / or multiple terminals (420). The above AI model (415) can provide fast and accurate processing results for input data (413, 423) by evolving through learning.

[0107] For example, a base station (410) or a terminal (420) may operate an AI model (415) for each function, such as handover, network selection, or resource scheduling. The base station (410) or the terminal (420) may perform the corresponding function through the AI ​​model (415) that supports tasks for wireless communication.

[0108] The base station (410) may include a plurality of AI models (415a, 415b) (hereinafter referred to as “first AI models”) that perform tasks according to functions. The first AI models (415a, 415b) may be provided in the base station (410) according to functions for performing wireless communication. The base station (410) may activate the first AI models (415a, 415b) corresponding to the relevant functions in response to events that occur when performing wireless communication. The first AI models (415a, 415b) may correspond to AI models (425a, 425b) (hereinafter referred to as “second AI models”) that are provided to perform the same function for each terminal (420).

[0109] The terminal (420) may include one or more AI models (425a, 425b) (hereinafter referred to as "second AI models") that perform tasks according to their functions. The second AI models (425a, 425b) may be provided in the terminal (420) according to their functions for performing wireless communication. The terminal (420) may activate the second AI models (425a, 425b) corresponding to the relevant function in response to an event that occurs while performing wireless communication.

[0110] For example, a base station (410) or a terminal (420) may provide AI models that are mutually compatible with each other by function. For example, a base station (410) or a terminal (420) may provide AI models for one or more functions for the purpose of resource scheduling, and when an event requiring resource scheduling occurs, the AI ​​model for processing the corresponding function may be activated.

[0111] In one example, the first or second AI model (415a, 415b, 425a, 425b) can be applied to various layers. For example, the first or second AI model (415a, 415b, 425a, 425b) can be applied to the application layer, the network layer, the MAC layer, or the physical (PHY) layer. In particular, as the AI ​​model is applied to the physical layer, new signal processing or communication mechanisms such as deep learning-based channel coding and / or decoding, deep learning-based MIMO control, and deep learning-based resource scheduling and / or allocation can be introduced.

[0112] According to an example, the first or second AI models (415a, 415b, 425a, 425b) may be provided in the base station (410) and the terminal (420), respectively. For example, the base station (410) may include the first AI model (415a, 415b). The terminal (420) may include the second AI model (425a, 425b). Each of the terminals (420) may include the second AI model (425a, 425b) to perform tasks or functions for the same purpose or different purposes. The first terminal (420a) may include, for example, the second AI model (425a). The second terminal (420b) may include, for example, the second AI model (425b). Part or all of the first AI model (415a, 415b) may be trained from an AI model (415) that is designed to perform substantially the same task or function as the second AI model (425a, 425b). The AI ​​model (415) may be the first AI model to perform the task. The training may be performed by the base station (410) and / or the terminal (420). The base station (410) may obtain training data according to training from the first AI model (415a, 415b). The terminal (420) may obtain training data according to training from the second AI model (425a, 425b).

[0113] The base station (410) can transmit the first AI model (415a, 415b) or parameters specifying the first AI model (415a, 415b) to the terminal(s) (420). The terminal(s) (420) can store the received first AI model (415a, 415b) in memory or update a pre-stored AI model (425a, 425b) based on the parameters specifying the first AI model (415a, 415b).

[0114] For example, since the first AI model (415a, 415b) and the second AI model (425a, 425b) are trained from the same AI model (415), the first AI model (415a, 415b) and the second AI model (425a, 425b) may have substantially the same neural network structure. Specifically, the first AI model (415a, 415b) or the second AI model (425a, 425b) may have substantially the same input layer, hidden layer(s), and / or output layer, respectively. Nodes of each layer constituting the first AI model (415a, 415b) or the second AI model (425a, 425b) and / or correlations between nodes may be substantially the same in both AI models.

[0115] In one example, part or all of the first AI model (415a, 415b) of the base station (410) may be substantially identical to or similar to the second AI model (425a, 425b) of the terminal (420). The first AI model (415a, 415b) may be, for example, a neural network model that is substantially identical to the second AI model (425a, 425b). In this case, the first AI model (415a, 415b) and the second AI model (425a, 425b) may be trained to output substantially identical or similar results based on the same input being applied.

[0116] If the first AI model (415a, 415b) and the second AI model (425a, 425b) have substantially the same or similar neural network structures, and thus the input data is synchronized, the first AI model (415a, 415b) and the second AI model (425a, 425b) can generate substantially the same or similar outputs. If the first AI model (415a, 415b) and the second AI model (425a, 425b) generate substantially the same or similar outputs, it can be said that the first input data (413a, 413b) of the first AI model (415a, 415b) is synchronized with the second input data (423a, 423b) of the second AI model (425a, 425b). For example, synchronization of the first AI model (415a, 415b) and the second AI model (425a, 425b) may require that the first and second input data (413a, 413b, 423a, 423b) match. That is, in order to obtain substantially identical or similar output data (417a, 417b, 427a, 427b) from the first AI model (415a, 415b) and the second AI model (425a, 425b), the second input data (413a, 413b) for the first AI model (415a, 415b) and the second input data (423a, 423b) for the second AI model (425a, 425b) must be maintained at substantially identical or similar values. In this way, maintaining the first input data (413a, 413b) for the first AI model (415a, 415b) and the second input data (423a, 423b) for the second AI model (425a, 425b) as substantially the same or similar values ​​may be referred to as a 'synchronization procedure' or an 'AI synchronization procedure'.For example, if the accumulated first input data (413a, 413b) for the first AI model (415a, 415b) and the accumulated second input data (415a, 415b) for the second AI model (425a, 425b) are the same, the first AI model (415a, 415b) and the second AI model (425a, 425b) can obtain first and second output data (417a, 417b, 427a, 427b) that are substantially the same or similar.

[0117] According to an example, the base station (410) or the terminal (420) can perform a synchronization procedure to make the first input data (413a, 413b) of the first AI model (415a, 415b) and the second input data (423a, 423b) of the second AI model (425a, 425b) substantially identical or similar, thereby obtaining identical or similar first and second output data (417a, 417b, 427a, 427b). The first input data (413a, 413b) may be input data accumulated corresponding to the first AI model (415a, 415b). The second input data (423a, 423b) may be input data accumulated corresponding to the second AI model (425a, 425b).

[0118] In one embodiment, a plurality of first AI models (415a, 415b) may be distinguished based on a model ID. A plurality of second AI models (425a, 425b) may be distinguished based on a model ID. The first AI models (415a, 415b) and the second AI models (425a, 425b) may be associated with each other. The second AI models (425a, 425b) associated with the first AI models (415a, 415b) may have the same model ID as the first AI models (415a, 415b). The second AI models (425a, 425b) associated with the first AI models (415a, 415b) may have a model ID that is pre-associated with the model ID of the first AI models (415a, 415b). The first AI models (415a, 415b) and the second AI models (425a, 425b), which are related to each other, may be referred to as having input data synchronized with respect to each other. This synchronization with respect to each other may indicate that the input data (413a, 413b, 423a, 423b) applied to the first AI models (415a, 415b) and the second AI models (425a, 425b) are substantially identical.

[0119] In one embodiment, the model ID for the second AI model (425a, 425b) may be generated in response to the base station (410) transmitting the second AI model (425a, 425b) based on the first AI model (415a, 415b) to the terminal (420a, 420b). For example, the model ID for the second AI model may be generated in response to the base station (410) transmitting the second AI model (425a, 425b) based on the first AI model (415a, 415b) to the terminal (420a, 420b).

[0120] In one embodiment, the model ID for the second AI model (425a, 425b) may be generated in response to the initial connection between the base station (410) and the terminal (420a, 420b). For example, in response to the initial connection between the base station (410) and the terminal (420a, 420b), the base station (410) may generate the model ID for the first AI model. Although not limited thereto, the base station (410) may transmit the second AI model to the terminal (420a, 420b) through an initial connection procedure with the terminal (420a, 420b). Specifically, the second AI model may be transmitted to the terminal (420a, 420b) during any one of the operations of the initial connection procedure.

[0121] An input mismatch may be identified by at least one of a first device (e.g., a base station (410)) or a second device (e.g., a terminal (420)). The input mismatch may indicate that first input data (413a, 413b) accumulated corresponding to a first AI model (415a, 415b) and second input data (423a, 423b) accumulated corresponding to a second AI model (425a, 425b) do not match. As an example, the input mismatch may be estimated or obtained based on an output mismatch in which output data (417a, 417b) from the first AI model (415a, 415b) and output data (427a, 427b) from the second AI model (425a, 425b) do not match. For that reason, it can be inferred that the reason why the output data (417a, 417b) of the first AI model (415a, 415b) and the output data (427a, 427b) of the second AI model (425a, 425b) do not match is likely due to an input mismatch in which the first input data (413a, 413b) and the second input data (423a, 423b) do not match.

[0122] An input mismatch may be identified, for example, based on a signal based on a first AI model (415a, 415b) not being received with a configuration or timing inferred by a second AI model (425a, 425b). An input mismatch may be identified, for example, based on a signal based on a second AI model (425a, 425b) not being received with a configuration or timing inferred by the first AI model (415a, 415b). A signal based on a first AI model (415a, 415b) (or a second AI model (425a, 425b)) not being received with a configuration or timing inferred by the second AI model (425a, 425b) (or a first AI model (415a, 415b)) may be caused by a mismatch of input data for each AI model. Accordingly, the wireless communication system according to various embodiments of the present disclosure proposes a base station (410), a terminal (420a, 420b), and an operation method for resolving input mismatch.

[0123] Specifically, according to one embodiment of the present disclosure, in an environment where the same input data (413a, 413b, 423a, 423b) is provided, the output data (417a, 417b, 427a, 427b) of the first AI model (415a, 415b) and the second AI model (425a, 425b) substantially match, and an operation corresponding to the corresponding output data (417a, 417b, 427a, 427b) is performed, thereby reducing overhead due to signal transmission. However, due to the nature of wireless communication, a mismatch (i.e., input mismatch) may occur in input data (413a, 413b, 423a, 423b) between the base station (410) and the terminal (420a, 420b) due to channel changes and / or various factors, and accordingly, output data (417a, 417b, 427a, 427b) of the first AI model (415a, 415b) and the second AI model (425a, 425b) may mismatch. Since a mismatch in the output data (417a, 417b, 427a, 427b) may cause a malfunction, it is necessary to synchronize the input data (413a, 413b, 423a, 423b) for each of the first AI model (415a, 415b) and the second AI model (425a, 425b). The input data (413a, 413b, 423a, 423b) may be input data accumulated as each of the first AI model (415a, 415b) and the second AI model (425a, 425b) performs operations.

[0124] As described above, when using an AI model with synchronized input data, even if the devices to which the AI ​​model is applied (e.g., the base station (410) and the terminal (420)) are different, the same result data can be obtained, which allows the heterogeneous devices to perform the operation according to the corresponding function in the same manner. For example, in heterogeneous devices (e.g., the base station (410) and the terminal (420)) in which the function of predicting the presence of DCI (DCI presence prediction) is implemented as an AI model, the presence of DCI can be predicted in the same manner, and thus the operation of transmitting or receiving a signal at the same time can be performed. In this case, in heterogeneous devices (e.g., the base station (410) and the terminal (420)), an error due to an output mismatch can be determined when reception of a response signal (ACK / NACK) fails, or when DCI does not exist after waking up. For example, when one device (e.g., the base station (110) or the terminal (120)) receives necessary information from the counterpart device (e.g., the terminal (120) or the base station (110)), If this fails, the CSI report may fail to be received, or the response signal (ACK / NACK) may fail to be decoded. In this case, the device (e.g., the base station (410) and the terminal (420)) may acquire, identify, or recognize that an input mismatch has occurred in the AI ​​model.

[0125] FIG. 5 is a flowchart of an operation method of a wireless communication system according to one embodiment of the present disclosure.

[0126] Referring to FIG. 5, a base station (e.g., base station (110) of FIG. 1) and a terminal (e.g., terminal (120) of FIG. 1) of a wireless communication system may each have an AI model (e.g., first AI models (415a, 415b) or second AI models (425a, 425b) of FIG. 4). The base station (110) may have, for example, a first AI model (e.g., first AI models (415a, 415b) of FIG. 4). The first AI model of the base station (110) may be provided for each terminal. The terminal (120) may have, for example, a second AI model (e.g., second AI models (425a, 425b) of FIG. 4). The first AI model and the second AI model may have substantially the same neural network structure. When substantially identical input data is applied to the first and second AI models, their respective outputs can be substantially identical. Accordingly, the overhead associated with signal transmission between the base station (110) and the terminal (120) can be reduced.

[0127] In one embodiment, the terminal (120) may include multiple second AI models associated with various communication functions. At least one of the multiple second AI models may support one or more communication functions. For example, at least one of the multiple second AI models may support at least one of various communication functions, such as channel estimation, resource allocation, or beamforming.

[0128] In one embodiment, the second AI models can be distinguished based on the communication functions they each support. Furthermore, at least one of the second AI models can have substantially the same neural network structure as the first AI model. The second AI model having substantially the same neural network structure as the first AI model can have communication functions that are substantially the same as or corresponding to the first AI model. For example, a first AI model for identifying the presence of a DCI based on scheduling information can have substantially the same neural network structure as a second AI model for identifying the presence of a DCI based on the same scheduling information. In the present disclosure, an operating method of a wireless communication system including these second AI models and the first AI models is described.

[0129] Each operation of the above operating method can be performed by a first device or a second device. The first device can be either a base station or a terminal. The second device can be either a terminal or a base station. If the first device is a base station, the second device can be a terminal. If the second device is a base station, the first device can be a terminal. The following description will assume that the relevant operations are performed by the first device, but the same operations can also be performed by the second device.

[0130] In operation 501, a first device may determine whether synchronization of input data for at least one of a plurality of first AI models is required. At least one of the plurality of first AI models may be included in a plurality of second AI models of a second device. For example, the first device may determine whether input data for a first AI model provided by the first device needs to be synchronized with input data for a second AI model provided in the second device. The first AI model and the second AI model that determine whether synchronization of the input data is required may be AI models provided for substantially the same function (e.g., a task such as adaptive modulation, channel estimation, power control, resource allocation, or beamforming). The first AI model and / or the second AI model may adjust wireless communication parameters according to network conditions and / or user requirements using a mathematical model and / or an algorithm. For example, the wireless communication parameters may be adjusted by input data for the first AI model and / or the second AI model.

[0131] In one embodiment, the first device may determine whether synchronization of input data for at least one of the plurality of first AI models is required based on at least one of (a) a preset period or (b) one or more input mismatches.

[0132] The above (a) preset cycle may be preset between the two devices, for example, so that the first device can actually perform wireless communication with the second device. For example, upon initial connection to the base station, the base station may provide information regarding the cycle for performing synchronization by including it in the system information (SI) transmitted to the terminal. For example, the first device may monitor the preset cycle and determine that synchronization is necessary based on the arrival of the cycle. The first device may determine that synchronization is necessary whenever the preset cycle is reached.

[0133] Above (b) one or more input mismatches can be determined to have occurred at a point in time when a predetermined type offset has elapsed from the detection time when at least one input mismatch is detected. The predetermined time offset can be set, for example, in units of slots. For example, the first device can obtain an input mismatch from one or more signals received from the second device, and can determine that synchronization for input data is necessary based on the acquisition of one or more input mismatches. In particular, the first device can obtain input mismatches that are received continuously. In one example, it can be determined that synchronization for input data of an AI model in which the input mismatch is identified is necessary based on the acquisition of an input mismatch at least N times (where N is a natural number greater than or equal to 2) or more consecutive times. The above input mismatch may occur when the input data in the first device and the input data in the second device are different due to failure to receive or identify input data required to support various functions, such as, for example, channel state information (CSI), HARQ information, downlink control indicator (DCI) information, discontinuous reception (DRX) configuration information, power control information, resource allocation information, or various scheduling information. The difference in the input data makes it possible to predict that the output data obtained through the AI ​​model in the first device and the output data obtained through the AI ​​model in the second device are different. The AI ​​model in the first device may have substantially the same neural network structure as the AI ​​model in the second device.

[0134] For example, a first device may determine that synchronization of input data is required based on receiving a synchronization request message from a second device. Furthermore, for example, the first device may determine that synchronization of input data is required based on receiving an uplink request for a synchronization indication message (e.g., a synchronization authorization request message) from the second device. The synchronization request message may be transmitted based on a first physical channel. The uplink request may be transmitted based on a second physical channel. The first physical channel and the second physical channel may be different channels. The synchronization request message and the uplink request may be configured in different formats.

[0135] In operation 502, the first device may transmit a synchronization instruction message to the second device to synchronize input data to be used for at least one of the plurality of second AI models with input data to be used for an associated first AI model based on determining that synchronization for input data is required. The first AI model may have substantially the same neural network structure as the second AI model. The first AI model may have a communication function corresponding to the second AI model. For example, a second AI model for identifying the presence of a DCI based on scheduling information may have substantially the same neural network structure as a first AI model for identifying the presence of a DCI based on the same scheduling information.

[0136] In one embodiment, the synchronization instruction message may include at least one of the plurality of fields of [Table 2]. However, the synchronization instruction message according to one embodiment of the present disclosure is not limited to the plurality of fields of [Table 2], and may further include other fields or information.

[0137] Description Mismatch Category Field Indicates at least one of a plurality of preset input mismatch categories. Model ID Field Indicates at least one of a plurality of second AI models. Input Index Field Indicates input data to be synchronized among a plurality of input data for the second AI model. Time Offset Field Indicates a time offset (e.g., number of slots) from the time of receiving the synchronization instruction message to the time of initiating synchronization for at least one of a plurality of input data for the first AI model.

[0138] In one embodiment, the model ID field may include a model ID associated with each of a plurality of second AI models. For example, a second AI model A associated with a first communication function may have a first model ID, and a second AI model B associated with a second communication function may have a second model ID. A second device receiving a synchronization instruction message may determine the second AI model to which synchronization of input data is to be performed by identifying the model ID recorded in the model ID field.

[0139] As defined in the above , the unit of the time offset recorded in the time offset field may be, for example, a slot. The input index recorded in the input index field may indicate input data requiring synchronization among input data required to support various functions, such as channel state information (CSI) information, HARQ information, downlink control indicator (DCI) information, discontinuous reception (DRX) configuration information, power control information, resource allocation information, or various scheduling information. The HARQ information may include information regarding whether a response message (e.g., an ACK message or a NACK message) corresponding to transmission / reception traffic has been received. The DRX configuration information may include information regarding a DRX cycle, a DRX maintenance interval, or an inactive interval, for example. In addition, the input index may also indicate information regarding adaptive modulation or information regarding channel estimation.

[0140] In one embodiment, the model ID for the second AI model may be generated in response to the first device transmitting a second AI model based on the first AI model to the second device. For example, the model ID for the second AI model may be generated in response to the base station transmitting a second AI model based on the first AI model to the terminal.

[0141] In one embodiment, the model ID for the second AI model may be generated in response to the initial connection between the first device and the second device. For example, in response to the initial connection between the base station and the terminal, the base station may generate the model ID for the first AI model. Although not limited thereto, the base station may transmit the second AI model to the terminal through an initial connection procedure with the terminal. Specifically, the second AI model may be transmitted to the terminal during any one of the operations of the initial connection procedure.

[0142] According to one embodiment, the input index recorded in the input index field included in the synchronization instruction message may indicate input data to be synchronized among various input data as defined above. The input index field may also serve to indicate that the input data to be used for the first AI model and the input data to be used for the second AI model be reset to initial input data. If the initial input data for the reset is already shared by the first device and the second device, the synchronization instruction message does not need to include the input data for synchronization. However, if the initial input data for the reset is not shared by the first device and the second device, the synchronization instruction message needs to include the initial input data for synchronizing the input data. In this case, although not defined in [Table 2], the synchronization instruction message may further include a field for recording the initial input data for synchronizing the input data. As an example, if the initial input data for the reset is shared by the first device and the second device, the first device and the second device may perform a procedure for sharing the initial input data via RRC.

[0143] In the present disclosure, a synchronization instruction message in the format of [Table 2] may be referred to as an AIR message. The first device may, for example, initialize input data before transmitting the AIR message. The first device may, for example, initialize input data after transmitting the AIR message. The second device may, for example, initialize input data after receiving the AIR message.

[0144] In one embodiment, the synchronization instruction message may include at least one of the plurality of fields of [Table 3]. However, the synchronization instruction message according to one embodiment of the present disclosure is not limited to the plurality of fields of [Table 3], and may further include other fields or information.

[0145] Category Mismatch Category Field Indicates at least one of a plurality of preset input mismatch categories. Model ID Field Indicates at least one of a plurality of second AI models. Input Index Field Indicates input data to be synchronized among a plurality of input data for the second AI model. Time Offset Field Indicates a time offset (e.g., number of slots) from the time of receiving the synchronization instruction message to the time of initiating synchronization of at least one of the plurality of input data for the first AI model. Input Index Field Index information for indicating input data for newly defining or modifying input information to be synchronized

[0146] In one embodiment, the synchronization instruction message may further include an input data field. The second device receiving the synchronization instruction message may redefine or modify the input data for the second AI model based on the synchronization instruction message. For example, the second device may modify the input data for the second AI model to correspond to the input data recorded in the input data field. For example, the input data for the first AI model or the parameters (e.g., weights) of the second AI model may be adjusted based on the input data.

[0147] In one embodiment, the synchronization instruction message may not include an input data field. In this case, the first device must establish a procedure for transmitting the input data for synchronization to the second device. For example, if the first device maintains an RRC connection with the second device, the first device can transmit the input data for synchronization to the second device using an RRC message.

[0148] In the present disclosure, a synchronization instruction message having a format such as [Table 3] may be referred to as an AID message. For example, after transmitting an AID message, a first device may update current input data (e.g., accumulated input data) with input data (e.g., new input data) included in the AID message. A second device may, for example, after receiving an AID message, update accumulated input data with new input data included in the AID message.

[0149] In one embodiment, the synchronization instruction message may include a first synchronization instruction message (e.g., an AIR message) in a first format (e.g., a format such as Table 2), or a second synchronization instruction message (e.g., an AID message) in a second format (e.g., a format such as Table 3). The first synchronization instruction message is a synchronization instruction message that instructs to reset input data that is a synchronization target. The second synchronization instruction message is a synchronization instruction message that instructs to newly define or modify input data that is a synchronization target.

[0150] A first device can transmit a synchronization instruction message in either the first format or the second format to a second device. By transmitting a synchronization instruction message in either the first format or the second format to the second device, the first device can instruct the second device to reset, redefine, or modify input data that is the target of synchronization of the second device.

[0151] In one embodiment, the synchronization instruction message may include at least one of a mismatch category field, a model ID field, an input index field, or a time interval field. The synchronization instruction message may include at least one of the mismatch category field, the model ID field, the input index field, or the time interval field, and may further include an input data field for newly defining or modifying the input data to be synchronized.

[0152] In one embodiment, the synchronization instruction message may be substantially identical to or similar to the synchronization instruction message described above with reference to [Table 2]. Furthermore, the synchronization instruction message may be substantially identical to or similar to the synchronization instruction message described above with reference to [Table 3].

[0153] In one embodiment, the first device may select either the first format or the second format and transmit a synchronization instruction message of the selected format to the second device. However, one embodiment of the present disclosure is not limited thereto, and the first device may use only one of the synchronization instruction messages or only one of the synchronization instruction messages for synchronization purposes. In this case, the synchronization instruction message may include an identifier field indicating the type of message.

[0154] In one embodiment, the synchronization instruction message may include at least one of the plurality of fields of [Table 4]. However, the synchronization instruction message according to one embodiment of the present disclosure is not limited to the plurality of fields of [Table 4], and may further include other fields or information.

[0155] Synchronization method field: First type information, second type information Mismatch category field: Indicates at least one of a plurality of preset input mismatch categories. Model ID field: Indicates at least one of a plurality of second AI models. Input index field: Indicates input data to be synchronized among a plurality of input data for the second AI model. Time offset field: Indicates the time offset (number of slots) from the time of receiving the synchronization instruction message to the time of starting synchronization of at least one of the plurality of input data for the first AI model. Input data field: New input data for newly defining or modifying the input data to be synchronized

[0156] In one embodiment, the synchronization instruction message may further include a synchronization method field. The synchronization method field may include either first type information or second type information. The first type information may indicate a message type that performs synchronization on input data by resetting the input data, as suggested in [Table 2], for example. The second type information may indicate a message type that performs synchronization on input data by changing or modifying the input data, as suggested in [Table 3], for example.

[0157] For example, if the synchronization method field includes first type information, the synchronization instruction message may instruct the input data to be synchronized to be reset. In this case, the input data may be reset to initial input data. The initial input data may be pre-arranged or shared through an agreed-upon procedure.

[0158] For example, if the synchronization method field includes second type information, the synchronization instruction message may instruct the synchronization target input data to be newly defined or modified. In this case, the synchronization target input data may be recorded and transmitted in the input data field of the synchronization instruction message or transmitted through a predetermined procedure.

[0159] As described above, a synchronization instruction message including a synchronization method field can have both the functions of the first synchronization instruction message and the second synchronization instruction message described above with reference to [Table 2] and / or [Table 3], respectively, by setting the synchronization method field to either the first type information or the second type information.

[0160] When synchronization of input data for a second AI model is performed, synchronization of input data for a first AI model may also be required. According to one embodiment of the present disclosure, a first device may determine whether synchronization of input data for the first AI model is required. Based on determining that synchronization of the input data is required, the first device may perform a reset to synchronize the input data for the first AI model with the input data for the second AI model. The reset to synchronize the input data may include, for example, initializing accumulated input data. The reset to synchronize the input data may include, for example, changing the accumulated input data to new input data. The reset to synchronize the input data may include, for example, modifying a portion of the accumulated input data. In the following description, the terms reset or set may be used to mean any operation that updates the accumulated input data by initializing, changing, or modifying the accumulated input data.

[0161] In operation 503, the first device may perform synchronization of the input data in response to transmitting the synchronization instruction message. The synchronization of the input data may be an operation in which the first device makes the input data for the first AI model substantially identical to the input data for the second AI model of the second device. Accordingly, in response to the first device transmitting the synchronization instruction message to the second device, the input data for the first AI model may be synchronized to be substantially identical or similar to the input data for the second AI model. The synchronization of the input signal may be achieved by resetting to initial input data, changing to new input data, or modifying the input data.

[0162] In one embodiment, input data for the first AI model may be initialized, redefined, or modified based on at least one of the types of synchronization instruction messages, or input data included in the synchronization instruction messages.

[0163] The first device may be instructed to perform synchronization on the input data, for example, by resetting the input data via a synchronization instruction message. In this case, the first device may reset the accumulated input data corresponding to the first AI model to initial input data. If the second device is instructed to reset the input data via the synchronization instruction message, the second device may reset the accumulated input data corresponding to the second AI model to initial input data. The initial input data to be reset may be shared in advance between the first device and the second device, for example. The initial input data to be reset may be provided via a synchronization instruction message, for example. The initial input data to be reset may be transmitted from the first device to the second device, for example, via a separately provided procedure.

[0164] By resetting the input data through the above-described operation, the first AI model and the second AI model can have substantially identical input data. Accordingly, the input data for the first AI model and the input data for the second AI model can be substantially synchronized.

[0165] A first device may, for example, transmit input data for synchronization to a second device while instructing synchronization of input data via a synchronization instruction message. In this way, when instructing to newly define or modify input data for synchronization, the first device may change or modify the input data to be applied to the first AI model based on the input data transmitted to the second device for synchronization. The second device may change or modify the input data to be applied to the second AI model based on the input data received from the first device for synchronization.

[0166] The above-described operation allows the first AI model and the second AI model to have substantially identical input data when the input data is changed or modified. Accordingly, the input data for the first AI model and the input data for the second AI model can be substantially synchronized.

[0167] In one embodiment, the synchronization indication message may be transmitted based on the PDCCH. For example, the synchronization indication message may be transmitted via DCI. In this case, the information to be included in the synchronization indication message needs to be defined in a format suitable for transmission via the DCI. For example, if the model ID is transmitted via DCI, the index may be explicitly included in the synchronization indication message. However, if the synchronization indication message is transmitted via RRC, the identifier may be included in the synchronization indication message.

[0168] FIG. 6 is a flowchart of an operation method of a wireless communication system according to one embodiment of the present disclosure.

[0169] In addition to synchronizing the input data of the first AI model in response to the first device sending a synchronization instruction message, an operation may be required to determine whether synchronization of the input data of the first AI model is necessary to improve the stability or reliability of the first AI model.

[0170] According to one embodiment of the present disclosure, a first device may determine whether synchronization of input data for a first AI model is required, and based on the determination that synchronization is required, may reset to synchronize input data of the first AI model and input data of a second AI model. The reset may be resetting, newly defining, or modifying input data for the first and / or second AI models. The reset may be first input data, which is accumulated input data for the first AI model, and second input data, which is accumulated input data for the second AI model having the same index as the first input data. The number of input data to be synchronized may be one or more.

[0171] In operation 601, the first device may determine whether synchronization of input data for at least one of a plurality of second AI models of the second device is required. In operation 602, based on the determination that synchronization of the input data is required, the first device may transmit a synchronization instruction message to the second device to synchronize the input data of at least one of the plurality of first AI models with input data of an associated second AI model. Operations 601 and 602 may correspond to operations 501 and 502 of FIG. 5 , respectively.

[0172] In operation 603, the first device may determine that synchronization is required for input data of the first AI model in response to receiving a synchronization confirmation message for a synchronization instruction message from the second device. The synchronization confirmation message may include indicator information indicating which synchronization instruction message it is in response to. The indicator information may be a timestamp corresponding to a transmission time of the synchronization instruction message, an index, or a sequence for distinguishing the synchronization instruction message. Based on the determination, the first device may reset, redefine, or modify the input data that is a synchronization target of the first AI model to be substantially identical to the input data that is a synchronization target of the second AI model.

[0173] Specifically, the first device can initialize, redefine, or modify the plurality of input data for the first AI model to correspond to the initialized, newly defined, or modified input data among the plurality of input data for the second AI model based on the synchronization instruction message.

[0174] In one embodiment, the synchronization confirmation message may include at least one of the plurality of fields of [Table 5]. However, the synchronization confirmation message according to one embodiment of the present disclosure is not limited to the plurality of fields of [Table 5], and may further include other fields or information.

[0175] Distinctive Content Model ID Field Indicates at least one of a plurality of first AI models. Time Offset Field Indicates a time offset (e.g., number of slots) from the time of receipt of the verification message to the time at which synchronization is to be initiated for at least one of a plurality of input data for the first AI model. Message Index Field Indicates the synchronization instruction message that is the target of verification. It can be expressed as an index.

[0176] In one embodiment, the model ID included in the model ID field of the synchronization confirmation message may be related to the model ID included in the model ID field of the synchronization instruction message. For example, the second AI model for the first communication function and the first AI model for the first communication function may be related to each other. When it is necessary to reset, redefine, or modify at least one input data of the second AI model for the first communication function and to reset, redefine, or modify at least one input data of the first AI model for the first communication function, it is necessary to indicate the first AI model as a synchronization target. Accordingly, the model ID field included in the model ID field may be determined based on the model ID field included in the synchronization instruction message. For example, when the first model ID was included in the synchronization instruction message, the model ID field of the synchronization confirmation message may also include the model ID associated with the first model ID. The model ID associated with the first model ID may include a model ID substantially identical to the first model ID, or a pre-mapped model ID.

[0177] In one embodiment, the model ID for the first AI model may be generated in response to the first device transmitting a second AI model based on the first AI model to the second device. For example, the model ID for the first AI model may be generated in response to the base station transmitting a second AI model based on the first AI model to the terminal.

[0178] In one embodiment, the model ID for the first AI model may be generated in response to the initial connection between the first device and the second device. For example, in response to the initial connection between the base station and the terminal, the terminal may generate the model ID for the first AI model. Although not limited thereto, the base station may transmit the second AI model to the terminal through an initial connection procedure with the terminal. Specifically, the second AI model may be transmitted to the terminal during any one of the operations of the initial connection procedure.

[0179] In one embodiment, the time offset field may include slot information. For example, the slot information may indicate the number of slots from the time of receiving the synchronization confirmation message until at least one of the plurality of input data for the first AI model is synchronized. For example, the slot information may indicate the number of slots from the time of receiving the synchronization confirmation message until the input data for the first AI model is initialized, newly defined, or modified.

[0180] In one embodiment, the time offset field may include symbol information. For example, the symbol information may indicate the number of symbols from the time of receiving the synchronization confirmation message until at least one of the plurality of input data for the first AI model is synchronized. For example, the symbol information may indicate the number of symbols from the time of receiving the synchronization confirmation message until the input data for the first AI model is initialized, newly defined, or modified.

[0181] In one embodiment, the time offset field may include subframe information. For example, the subframe information may indicate the number of subframes from the time of receiving the synchronization confirmation message until at least one of the plurality of input data for the first AI model is synchronized. For example, the subframe information may indicate the number of subframes from the time of receiving the synchronization confirmation message until the input data for the first AI model is initialized, newly defined, or modified.

[0182] In one embodiment, the time offset field may include time offset information. For example, the time offset information may indicate a time offset from the time of receiving the synchronization confirmation message until at least one of the plurality of input data for the first AI model is synchronized. For example, the time offset information may indicate a time offset from the time of receiving the synchronization confirmation message until the time of initializing, redefining, or modifying the input data for the first AI model.

[0183] In one embodiment, the synchronization confirmation message may include a message index field. The message index field may include information for identifying the synchronization instruction message that formed the basis for transmitting the synchronization confirmation message. This information may be a message index. Two or more synchronization instruction messages may be distinguished based on the message index.

[0184] In one embodiment, the message index field may further include information for identifying the format of the synchronization instruction message that formed the basis for transmitting the synchronization confirmation message. For example, the message may further include information for distinguishing the synchronization instruction message as either a synchronization instruction message of a first format (e.g., an AIR message) or a synchronization instruction message of a second format (e.g., an AID message).

[0185] Additionally, in one embodiment, the message index field may further include information for identifying the synchronization method field of the synchronization instruction message that is the basis for transmitting the synchronization confirmation message, and for identifying whether the information included in the synchronization method field is first type information or second type information.

[0186] For example, the first device may not receive a synchronization confirmation message for a synchronization instruction message from the second device. In this case, the first device may retransmit the synchronization instruction message to the second device. If the first device does not receive a synchronization confirmation message from the second device even after transmitting the synchronization instruction message a set number of times, the first device may no longer perform the synchronization procedure for the input data. In this case, the first device may perform synchronization for the input data again when a synchronization request is received from the second device.

[0187] FIG. 7 is a flowchart of an operation method of a wireless communication system according to one embodiment of the present disclosure.

[0188] In a wireless communication system according to one embodiment of the present disclosure, not only a base station but also a terminal can transmit a synchronization instruction message, but when synchronizing input data for a first AI model operated at the base station through a terminal, it is necessary to introduce an additional operation in consideration of stability and reliability of the first AI model.

[0189] In operation 701, the terminal may determine whether synchronization of input data for a specific AI model is required. The specific AI model may be, for example, one of the first AI models or one of the second AI models. For example, the terminal may determine whether at least one second AI model among the multiple second AI models operated by the terminal requires synchronization of input data. Operation 701 may correspond to operation 501 of FIG. 5 .

[0190] In operation 702, if the terminal determines that synchronization of input data of a specific AI model is necessary, the terminal may transmit a synchronization approval request message to the base station. The synchronization approval request message may be, for example, a message requesting approval for synchronizing input data of a specific AI model for which synchronization of input data is determined to be necessary with the base station.

[0191] As an example, a synchronization authorization request message may include information requesting authorization to transmit a synchronization indication message (AIR message) in a first format via uplink, or requesting authorization to transmit a synchronization indication message (AID message) in a second format via uplink.

[0192] In one embodiment, the synchronization authorization request message may include an identifier that requests uplink transmission authorization for a synchronization instruction message having first field information or for uplink transmission authorization for a synchronization instruction message having second field information.

[0193] In operation 703, the terminal may receive a synchronization approval response message from the base station in response to the synchronization approval request message, and may transmit a synchronization instruction message (AIR message or AID message) to the base station in response to receiving the synchronization approval response message. Operation 703 may correspond to operation 502.

[0194] In operation 704, the terminal may, in response to transmitting the synchronization instruction message, reset the input data of the second AI model to the initial input data, or change or modify the input data to new input data. Operation 704 may correspond to operation 503.

[0195] FIG. 8 is a flowchart of an operation method of a wireless communication system according to one embodiment of the present disclosure.

[0196] Referring to FIG. 8, the first device (810) may be either a terminal or a base station, and the second device (820) may be either a base station or a terminal. If the first device (810) is a base station, the second device (820) may be a terminal. If the second device (820) is a base station, the first device (810) may be a terminal. The first device (810) may include a first AI model, and the second device (820) may include a second AI model.

[0197] In operation 801, the first device (810) may determine whether synchronization of input data of at least one first AI model among the plurality of first AI models operated by the first device (810) is required. Based on determining that synchronization of the input data is required, the first device (810) may transmit a synchronization request message to the second device.

[0198] In one embodiment, the synchronization request message may include at least one of the plurality of fields of [Table 6]. However, the synchronization request message according to one embodiment of the present disclosure is not limited to the plurality of fields of [Table 6], and may further include other fields or information.

[0199] Mismatch Category Field Indicates at least one of a plurality of preset input mismatch categories. Model ID Field Indicates at least one of a plurality of first AI models.

[0200] In one embodiment, the synchronization request message may include a mismatch category field. The mismatch category field may include at least one of a plurality of input mismatch categories. The mismatch category may include, but is not limited to, mismatch causes such as, for example, a control channel error or a sudden environmental change.

[0201] In one embodiment, the Model ID field may include a Model ID for identifying the first AI model whose input data is to be synchronized and needs to be reset, redefined, or modified. For example, if a Model ID indicating the first AI model is included, the second device (820) may identify the first AI model to be synchronized based on the included Model ID.

[0202] In one embodiment, the second device (820) may receive a synchronization request message. Based on the synchronization request message, the second device (820) may identify a mismatch category or identify a first AI model as a synchronization target for synchronizing input data.

[0203] In one embodiment, the second device (820) may determine that synchronization of input data of the first AI model is required based on receiving a synchronization request message. Specifically, the second device (820) may determine that synchronization of the input data is required based on receiving the synchronization request message, and at this time, the first AI model for which synchronization of the input data is required may be identified based on a model ID included in a model ID field. The second device (820) may obtain a second AI model, which is its own AI model corresponding to the first AI model identified based on the model ID.

[0204] In operation 802, the second device (820) may transmit a synchronization instruction message to the first device (810) in response to determining that synchronization is required. Operation 802 may correspond to operation 502. In one embodiment, the synchronization instruction message may be a synchronization instruction message having a format of any one of [Table 2], [Table 3], or [Table 4].

[0205] In one embodiment, the second device (820) may transmit a synchronization instruction message to the first device (810) when an event requiring synchronization of input data occurs based on preset requirements, without receiving a synchronization request message.

[0206] In operation 803, the first device (820) may synchronize input data of the first AI model based on the received synchronization instruction message. Specifically, the first AI model may be reset, newly defined, or modified based on a plurality of fields of the received synchronization instruction message. In operation 803, in response to transmitting the synchronization instruction message, the second device (820) may synchronize input data of the second AI model based on the transmitted synchronization instruction message. Specifically, the input data of the second AI model may be reset, newly defined, or modified based on a plurality of fields of the transmitted synchronization instruction message.

[0207] Accordingly, the first AI model and the second AI model can have their input data synchronized with each other. In other words, if substantially identical input data is provided to the first AI model and the second AI model, the first AI model and the second AI model can produce substantially identical outputs.

[0208] In one embodiment, when the first device (810) is a terminal and the second device (820) is a base station, the terminal may transmit a synchronization request message to the base station based on determining that synchronization is required for input data. The base station may transmit a synchronization instruction message to the terminal in response to receiving the synchronization request message. The input data for each of the first AI model and / or the second AI model having a model ID identical to or related to the first AI model may be reset, newly defined, or modified based on a plurality of fields of the synchronization instruction message.

[0209] As the synchronization request message is used, the first device (810) determines that synchronization is required and controls the second device (820) capable of transmitting a synchronization instruction message to generate and transmit a synchronization instruction message.

[0210] Figure 9 is a flowchart of an operating method according to one embodiment of the present disclosure.

[0211] Referring to FIG. 9, the first device (910) may be either a terminal or a base station, and the second device (920) may be either a base station or a terminal. If the first device (910) is a base station, the second device (920) may be a terminal. If the second device (920) is a base station, the first device (910) may be a terminal. The first device (910) may include a first AI model, and the second device (920) may include a second AI model.

[0212] Referring to FIG. 9, in operation 901, a first device (910) may determine whether synchronization is required for input data of at least one first AI model among a plurality of first AI models. Based on determining that synchronization of the input data is required, the first device (910) may transmit a synchronization request message to a second device (920). Operation 901 may correspond to operation 801. The synchronization request message transmitted by the first device (910) in operation 901 may be substantially identical to the synchronization request message transmitted by the second device (920) in operation 801.

[0213] In one embodiment, the second device (920) may determine, before transmitting the synchronization instruction message, at least one field information among a plurality of fields to be included in the synchronization instruction message, or input data for newly defining or modifying input data to be synchronized. The second device (920) may transmit a synchronization instruction message including at least one field information or input data to the first device (910) to synchronize input data of at least one first AI model among a plurality of first AI models with input data of an associated second AI model. The second AI model associated with the first AI model may have a model ID substantially the same as the first AI model, or may be a second AI model having a model ID that is associated in advance.

[0214] In one embodiment, the second device (920) may perform operation 902 in the process of determining the field information or the input data. In operation 902, the second device (920) may transmit a synchronization collection message to the first device (910). Based on transmitting the synchronization collection message to the first device (910), the second device (920) may collect data during a time offset allocated for data collection from the time at which the synchronization collection message is transmitted. The data may be base data for determining the field information or the input data. The second device (920) may determine at least one of the field information or the input data based on the data collected during the time period for data collection.

[0215] In one embodiment, the synchronization collection message may include at least one of the plurality of fields of [Table 7]. However, the synchronization collection message according to one embodiment of the present disclosure is not limited to the plurality of fields of [Table 7], and may further include other fields or information.

[0216] Description Model ID field Indicates at least one of a plurality of first AI models. Model On / Off field Indicates whether the first AI model indicated by the Model ID field is activated or deactivated. The first AI model is activated or deactivated based on the information in the Model On / Off field. Time Offset field Indicates the time offset (e.g., number of slots) allocated to collect base data from the time the AI ​​synchronization collection message is received.

[0217] In one embodiment, the second device (920) may transmit a synchronization collection message to the first device (910) based on preset requirements without receiving a synchronization request message.

[0218] In one embodiment, the first device (910) can deactivate the first AI model based on receiving a synchronization collection message. Additionally, the second device (920) can deactivate the second AI model based on transmitting the synchronization collection message. For example, the second device (920) can deactivate the second AI model for a predetermined time offset based on transmitting the synchronization collection message. The time offset for data collection can be preset. The first device (910) can determine the time offset for data collection based on the time offset field of the synchronization collection message.

[0219] In one embodiment, the first device (910) may deactivate at least one of the first AI models based on a Model On / Off field. The first AI model to be deactivated may be determined based on a Model ID field.

[0220] In operation 904, the second device (920) may transmit a synchronization instruction message to the first device (910). The second device (920) may transmit the synchronization instruction message to the first device (910) based on the expiration of a time offset set for data collection. The synchronization instruction message of operation 904 may be substantially identical to the synchronization instruction message of operation 502.

[0221] In operation 905, the second device (920) may synchronize input data of the first AI model based on the received synchronization instruction message. Specifically, the input data of the first AI model may be reset, newly defined, or modified based on a plurality of fields of the received synchronization instruction message. In operation 905, in response to transmitting the synchronization instruction message, the second device (920) may synchronize input data of the second AI model based on the transmitted synchronization instruction message. Specifically, the input data of the second AI model may be reset, newly defined, or modified based on a plurality of fields of the transmitted synchronization instruction message.

[0222] FIGS. 10A and 10B illustrate an example of a base station-driven synchronization procedure using a synchronization indication message (e.g., a synchronization input reset (AIR) message) according to one embodiment of the present disclosure.

[0223] Referring to FIG. 10A, the base station (1010) can determine whether synchronization is required for input data. For example, the base station (1010) can transmit a synchronization indication message (e.g., an AIR message) to the terminal (1020) based on the arrival of a preset period (e.g., an AIR period) (1011, 1015). The synchronization indication message can have, for example, a first format. The preset period can be defined as, for example, a predetermined time period. The preset period can be defined by, for example, the number of slots, the number of symbols, the number of subframes, or a time offset. That is, the period for transmitting the synchronization indication message can be set based on information that can identify a time interval (e.g., a slot, a symbol, a subframe, a time offset, etc.). The base station (1010) may transmit a synchronization instruction message to the terminal (1020), for example, whenever a preset period is reached (1011, 1015).

[0224] The terminal (1020) may reset the input data for the second AI model based on receiving the synchronization instruction message. In addition, the base station (1010) may reset the input data for the first AI model in response to transmitting the synchronization instruction message to the terminal (1020) (1013, 1017). As the input data for each of the first AI model and the second AI model are reset to the initial input data, the inference values ​​based on the first AI model and the second AI model may substantially match. The input data for each of the first AI model and the second AI model may be, for example, input data accumulated in each of the first AI model and the second AI model.

[0225] Referring to FIG. 10B, the base station (1010) can determine whether synchronization for input data is required. For example, the base station (1010) can identify an input mismatch (1021, 1023). The mismatch category illustrated in FIG. 10B may be, but is not limited to, a channel error. In one embodiment, the base station (1010) can transmit a synchronization indication message (e.g., an AIR message) to the terminal (1020) based on the detection of two or more consecutive input mismatches (1025). The synchronization indication message can have, for example, a first format. The input mismatch can be detected, for example, by a discrepancy between output data acquired by the AI ​​model at the base station (1010) and the terminal (1020). The base station (1010) can preset a time period for detecting consecutive input mismatches. The time interval for detecting the above-described consecutive input mismatches can be defined, for example, by the number of slots, the number of symbols, the number of subframes, or a time offset. That is, the period for transmitting the synchronization indication message can be set based on information that can identify the time interval (e.g., slots, symbols, subframes, time offset, etc.).

[0226] For example, when a predetermined number of consecutive input mismatches are detected, the base station (1010) may transmit a synchronization instruction message to the terminal (1020) after a preset waiting time has elapsed. The preset waiting time may be defined as, for example, a predetermined time period. The predetermined time period may be defined by, for example, the number of slots, the number of symbols, the number of subframes, or a time offset. That is, the time interval between the detection of a predetermined number of consecutive input mismatches and the transmission of the synchronization instruction message may be set based on information (e.g., slots, symbols, subframes, time offsets, etc.) that can identify the time interval.

[0227] The terminal (1020) may reset input data for the second AI model based on receiving a synchronization instruction message. Additionally, the base station (1010) may reset input data for the first AI model in response to transmitting the synchronization instruction message to the terminal (1020) (1027). As the input data for each of the first AI model and the second AI model are reset, inference values ​​based on the first AI model and the second AI model may substantially match. The input data for each of the first AI model and the second AI model may be, for example, input data accumulated in each of the first AI model and the second AI model.

[0228] According to the above-described FIG. 10a or FIG. 10b, the synchronization instruction message (e.g., AIR message) of the first format can be used for the purpose of setting the input data of the target AI model for which input mismatch is predicted as initial input data. For example, the base station (1010) or the terminal (1020) can prevent input mismatch from accumulating by periodically transmitting the AIR message (see FIG. 10a). For example, if the base station (1010) or the terminal (1020) determines that synchronization of input data due to input mismatch is necessary (e.g., occurrence of a synchronization event due to detection of an output error), the base station (1010) or the terminal (1020) can prevent input mismatch from accumulating by transmitting the AIR message (see FIG. 10b).

[0229] The above initial input data may be pre-defined. The pre-defined initial value may be set, for example, during model transfer. The model transfer may mean, for example, that the base station (1020) transfers the corresponding AI model to the terminal (1020). The pre-defined initial value may be set, for example, when the base station (1010) and the terminal (1020) are associated. The pre-defined initial value may differ, for example, by mismatch decision category that determines the input mismatch. The input mismatch decision category may be classified by factors such as, for example, a control channel error or a sudden change in the wireless communication environment.

[0230] For example, an AIR message may transmit a model ID, a time offset, or an input index. The model ID may be, for example, an identifier of an AI model to be initialized. The model ID may be set at the time of model transfer. The model ID may be set at the time of initial access. The time offset may indicate, for example, a number of slots after which accumulated input data is to be initialized from the time of transmission of the AIR message. The input index may be, for example, an identifier indicating input data to be initialized among input data for AI models. The input index may be, for example, an identifier indicating that input data of all AI models are to be initialized.

[0231] FIGS. 11A and 11B illustrate an example of a base station-driven synchronization procedure using a synchronization indication message (e.g., a synchronization input definition (AID) message) according to one embodiment of the present disclosure.

[0232] Referring to FIG. 11A, the base station (1110) can determine whether synchronization is required. For example, the base station (1110) can transmit a synchronization indication message (e.g., an AID message) to the terminal (1120) based on the arrival of a preset period (e.g., an AID period) (1130) (1111, 1115). The synchronization indication message can have, for example, a second format. The base station (1110) can transmit a synchronization indication message to the terminal (1120) whenever the preset period arrives (1111, 1115). The preset period can be defined as, for example, a predetermined time period. The predetermined time period can be defined by, for example, the number of slots, the number of symbols, the number of subframes, or a time offset. That is, the period for transmitting the synchronization indication message can be set based on information that can identify a time interval (e.g., slots, symbols, subframes, time offset, etc.).

[0233] The terminal (1120) may redefine or modify input data for the second AI model based on receiving the synchronization instruction message. Furthermore, the base station (1110) may redefine or modify input data for the first AI model in response to transmitting the synchronization instruction message to the terminal (1120) (1113, 1117). As the input data for each of the first and second AI models is redefined or modified, the inference values ​​based on the first and second AI models may substantially match.

[0234] The input data to be newly defined or modified can be determined based on the synchronization instruction message. The input data to be synchronized with the second AI model of the terminal (1120) can be newly defined or modified based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message. In addition, the base station (1110) that transmitted the synchronization instruction message can also newly define or modify the input data to be synchronized with the first AI model based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message.

[0235] Referring to FIG. 11B, the base station (1110) may determine whether synchronization of input data is required. For example, the base station (1110) may identify an input mismatch. The mismatch category illustrated in FIG. 11B may include, but is not limited to, a channel error. In one embodiment, the base station (1110) may transmit a synchronization indication message to the terminal (1120) based on the detection of two or more input mismatches (1125). The synchronization indication message may have, for example, a second format. The terminal (1120) may redefine or modify input data for the second AI model based on receiving the synchronization indication message. Additionally, the base station (1110) may redefine or modify input data for the first AI model in response to transmitting the synchronization indication message to the terminal (1120) (1127). As the input data for each of the first AI model and the second AI model is newly defined or modified, the inference values ​​based on the first AI model and the second AI model can be substantially consistent.

[0236] The input data to be newly defined or modified can be determined based on the synchronization instruction message. The input data to be synchronized of the second AI model of the terminal (1120) can be newly defined or modified based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message. In addition, the base station (1110) that transmitted the synchronization instruction message can also newly define or modify the input data of the AI ​​model to be synchronized based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message (1127).

[0237] FIG. 12 illustrates an example of a terminal-driven synchronization procedure using a synchronization instruction message (e.g., a synchronization input definition (AID) message) according to one embodiment of the present disclosure.

[0238] Referring to FIG. 12 , the terminal (1220) may determine whether synchronization is required. For example, the terminal (1220) may identify an input mismatch (1211, 1213). The mismatch category illustrated in FIG. 12 may include, but is not limited to, a channel error. In one embodiment, the terminal (1220) may transmit a synchronization indication message to the base station (1210) based on the detection of two or more input mismatches (1221).

[0239] The base station (1210) may redefine or modify input data for the first AI model based on receiving a synchronization instruction message. Furthermore, the terminal (1220) may redefine or modify input data for the second AI model in response to transmitting the synchronization instruction message to the base station (1210) (1223). As the input data for each of the second AI model and the first AI model are redefined or modified, the inference values ​​based on the second AI model and the first AI model may substantially match.

[0240] The input data to be newly defined or modified can be determined based on the synchronization instruction message. The input data to be synchronized for the first AI model of the base station (1210) can be newly defined or modified based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message. In addition, the terminal (1220) that transmitted the synchronization instruction message can also newly define or modify the input data to be synchronized for the second AI model based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message.

[0241] According to the above-described FIG. 11a, FIG. 11b or FIG. 12, the synchronization instruction message (e.g., AID message) of the second format can be used to redefine the input data of the target AI model for which input mismatch is predicted. The input data can be redefined by changing or modifying predetermined input data. The AID message can transmit an input message appropriate to the situation. For this purpose, the AID message can directly include the input data to be newly defined. For example, the base station (1010) or the terminal (1020) can prevent input mismatch from accumulating by periodically transmitting the AID message (see FIG. 11a). For example, if the base station (1010) or the terminal (1020) determines that synchronization of input data due to input mismatch is necessary (e.g., occurrence of a synchronization event due to detection of an output error), the base station (1010) or the terminal (1020) can prevent input mismatch from accumulating by transmitting the AID message (see FIG. 11b or FIG. 12). The above synchronization event may be triggered in response to the detection of a single output error or multiple output errors in succession. The AID message may be transmitted appropriately to the situation in which the output error occurred.

[0242] For example, an AIR message may transmit a model ID, a time offset, or an input index. The model ID may be, for example, an identifier of an AI model for which input data is to be newly defined. The model ID may be set at the time of model transfer. The model ID may be set at the time of initial access. The time offset may indicate, for example, a number of slots after the time of transmission of the AID message to apply new input data in place of accumulated input data. The input index may be, for example, an identifier indicating input data to be newly defined among input data for AI models. The input index may be, for example, an identifier indicating that input data of all AI models are to be newly defined.

[0243] The newly applied input data may be pre-defined. The newly applied input data may be set, for example, during model transfer. The model transfer may mean, for example, that the base station (1020) transfers the corresponding AI model to the terminal (1020). The newly applied input data may be set, for example, when the base station (1010) and the terminal (1020) are associated. The newly applied input data may differ, for example, by category that determines the input mismatch. The input mismatch category may be classified by factors such as a control channel error or a sudden change in the wireless communication environment, for example.

[0244] FIG. 13 illustrates an example of a terminal-driven synchronization procedure using a synchronization instruction message (e.g., an AID (Asynchronous Input Definition) message) according to one embodiment of the present disclosure. Specifically, FIG. 13 exemplarily illustrates a terminal-driven synchronization procedure including a synchronization authorization request message (AUR (AID Uplink Request) message) and a synchronization authorization response message (AUG (AID Uplink Grant) message).

[0245] Referring to FIG. 13, the terminal (1320) may determine whether synchronization of input data is required. For example, the terminal (1320) may identify an input mismatch (1311, 1313). The mismatch category illustrated in FIG. 13 may include, but is not limited to, a channel error. In one embodiment, the terminal (1320) may transmit a synchronization authorization request message (e.g., an AID Uplink Request) to the base station (1310) based on the detection of two or more input mismatches (1321).

[0246] The base station (1310) may transmit a synchronization grant response message (AID Uplink Grant) to the terminal (1320) in response to receiving a synchronization grant request message (1323). The synchronization grant response message may be an ACK signal for the synchronization grant request message. For example, the base station (1310) may obtain time information included in the synchronization grant request message and determine a time point for transmitting the synchronization grant response message based on the time information. The time information may be set by, for example, a time offset. This may be intended to allow the synchronization grant response message to be synchronized with the synchronization grant request message. The time information may be defined by, for example, a predetermined time offset. The predetermined time offset may be defined by, for example, the number of slots, the number of symbols, or the number of subframes. That is, the time offset may be set based on information (e.g., slots, symbols, subframes, time offset, etc.) that can identify a time interval from receiving a synchronization grant request message to transmitting a synchronization grant response message.

[0247] The terminal (1320) may transmit a synchronization instruction message of either the first format (AIR) or the second format (AID) to the base station (1310) based on receiving a synchronization authorization response message (1325). FIG. 13 illustrates an example of transmitting a synchronization instruction message of the second format (AID), but the present invention is not limited thereto and a synchronization instruction message of the first format (AIR) may also be transmitted. For convenience of explanation, the following description will assume a case of transmitting a synchronization instruction message of the second format.

[0248] The base station (1310) may redefine or modify input data for the first AI model based on receiving a synchronization instruction message. Furthermore, the terminal (1320) may redefine or modify input data for the second AI model in response to transmitting the synchronization instruction message to the base station (1310) (1327). As the input data for each of the second AI model and the first AI model are redefined or modified, the inference values ​​based on the second AI model and the first AI model may substantially match.

[0249] The input data to be newly defined or modified can be determined based on the synchronization instruction message. The input data to be synchronized for the first AI model of the base station (1310) can be newly defined or modified based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message. In addition, the terminal (1320) that transmitted the synchronization instruction message can also newly define or modify the input data to be synchronized for the second AI model based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message.

[0250] FIGS. 14A and 14B illustrate an example of a base station-driven synchronization procedure using a synchronization confirmation message (AC (AIR / AID confirmation) message) according to one embodiment of the present disclosure.

[0251] In one embodiment, the synchronization confirmation message may be transmitted aperiodically in response to receiving a synchronization instruction message (1423, 1443). In one embodiment, the synchronization confirmation message may also be transmitted periodically (1423, 1443). The synchronization confirmation message transmitted periodically may further include a field indicating whether the synchronization instruction message has been received. The field indicating whether the reception has been received may be set to at least one of reception or non-reception. A device receiving the synchronization confirmation message (e.g., base station (1410) or terminal (1420)) may reset, redefine, or modify input data for the AI ​​model based on whether the information indicating whether the reception has been received is identified as reception (1425, 1445).

[0252] Referring to FIG. 14A, the base station (1410) can determine whether synchronization is required for input data. For example, the base station (1410) can identify an input mismatch (1411, 1413). The mismatch category illustrated in FIG. 14A may include, but is not limited to, a channel error. In one embodiment, the base station (1410) can transmit a synchronization indication message of either the first format (AIR) or the second format (AID) to the terminal (1420) based on the detection of two or more input mismatches (1421). For convenience of explanation, the case of transmitting a synchronization indication message of the second format will be described below as an example.

[0253] The terminal (1420) may transmit a synchronization confirmation message (e.g., an ACM (AIR / AID confirmation message); AC message) to the base station (1410) based on receiving the synchronization instruction message (1423). In one embodiment, the terminal (1420) may transmit the synchronization confirmation message to the base station (1410) in response to receiving the synchronization instruction message (1423). For example, the terminal (1420) may obtain time information included in the synchronization instruction message and determine a time point for transmitting the synchronization confirmation message based on the time information. This may be for the purpose of allowing the synchronization confirmation message to be synchronized with the synchronization instruction message. The time information may be defined, for example, by a predetermined time offset. The predetermined time offset may be defined, for example, by the number of slots, the number of symbols, or the number of subframes. That is, it can be set based on information (e.g., slot, symbol, subframe, etc.) that can identify a time offset corresponding to the time interval from receiving a synchronization instruction message to transmitting a synchronization confirmation message.

[0254] In response to receiving the synchronization instruction message, the terminal (1420) transmits a synchronization confirmation message to the base station (1410), and based on the transmission of the synchronization confirmation message, the terminal (1420) may newly define or modify input data for the second AI model (1425). That is, in one embodiment, the terminal (1420) may newly define or modify input data for the second AI model in response to transmitting the synchronization confirmation message (1425).

[0255] The base station (1410) may redefine or modify input data for the first AI model based on receiving a synchronization confirmation message from the terminal (1420). As the input data for each of the first AI model and the second AI model are redefined or modified, the inference values ​​based on the first AI model and the second AI model may substantially match.

[0256] The input data to be newly defined or modified can be determined based on the synchronization instruction message. The input data to be synchronized with the second AI model of the terminal (1420) can be newly defined or modified based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message. In addition, the base station (1410) that transmitted the synchronization instruction message can also newly define or modify the input data to be synchronized with the first AI model based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message.

[0257] Referring to FIG. 14B, the terminal (1420) can determine whether synchronization is required for input data. For example, the terminal (1420) can identify an input mismatch (1431, 1433). The mismatch category illustrated in FIG. 14B may include, but is not limited to, a channel error. According to one embodiment, the terminal (1420) can transmit a synchronization instruction message in either the first format (AIR) or the second format (AID) to the base station (1410) based on the detection of two or more input mismatches (1441). For convenience of explanation, the case of transmitting a synchronization instruction message in the second format will be described below as an example.

[0258] In response to receiving a synchronization instruction message, the base station (1410) may transmit a synchronization confirmation message (e.g., an ACM (AIR / AID confirmation message); AC message) to the terminal (1420) (1443). For example, the base station (1410) may obtain time information included in the synchronization instruction message and determine a time point for transmitting the synchronization confirmation message based on the time information. This may be intended to allow the synchronization confirmation message to be synchronized with the synchronization instruction message. The time information may be defined by, for example, a predetermined time offset. The predetermined time offset may be defined by, for example, the number of slots, the number of symbols, or the number of subframes. That is, the time offset corresponding to the time interval from receiving the synchronization instruction message to transmitting the synchronization confirmation message may be set based on information (e.g., slots, symbols, subframes, etc.) that can identify the time offset.

[0259] The base station (1410) may, in response to receiving the synchronization instruction message, transmit a synchronization confirmation message to the terminal (1420), and may redefine or modify input data for the first AI model based on the transmission of the synchronization confirmation message (1445). That is, in one embodiment, the base station (1410) may also redefine or modify input data for the first AI model in response to transmitting the synchronization confirmation message (1445).

[0260] The terminal (1420) may redefine or modify input data for the second AI model based on receiving a synchronization confirmation message from the base station (1410). As the input data for each of the second AI model and the first AI model is redefined or modified, the inference values ​​based on the second AI model and the first AI model may substantially match.

[0261] The input data to be newly defined or modified can be determined based on the synchronization instruction message. The input data to be synchronized for the first AI model of the base station (1410) can be newly defined or modified based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message. In addition, the terminal (1420) that transmitted the synchronization instruction message can also newly define or modify the input data to be synchronized for the second AI model based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message.

[0262] According to the above-described FIG. 14a or FIG. 14b, a synchronization confirmation message (e.g., ACM (AIR / AID confirmation message); AC message) can be used as a response to a synchronization instruction message of the first format (e.g., AIR message) or a synchronization instruction message of the second format (e.g., AID message). That is, the AC message can be a response message to an AIR message or an AID message. In this case, the base station (1010) or the terminal (1020) can initialize (e.g., response to an AIR message) or newly define (e.g., response to an AID message) accumulated input data after transmitting or receiving the AC message (see FIG. 14a or FIG. 14b). The transmission of the AC message can be transmitted in the uplink in response to an AIR message or an AID message transmitted in the downlink (see FIG. 14a), or can be transmitted in the downlink in response to an AIR message or an AID message transmitted in the uplink (see FIG. 14b).

[0263] In one example, the AC message may transmit a model ID, a time offset, or a message index. The model ID may be, for example, an identifier of an AI model to be initialized or newly defined for input data. The model ID may be set during model transfer. The model ID may be set during initial access. The time offset may indicate, for example, a number of slots after the transmission time of the AC message to apply or initialize new input data in place of accumulated input data. The message index may be, for example, an identifier indicating when the AC message is a response to an AIR message or AID message transmitted. The message index may be, for example, an identifier indicating an AIR message to which the AC message responds. The message index may be, for example, an identifier indicating an AID message to which the AC message responds. The above message index may include, for example, an identifier that distinguishes the type of message being responded to (e.g., an AIR message or an AID message).

[0264] FIG. 15 illustrates an example of a terminal-driven synchronization procedure using a synchronization request message (AR (AIR / AID request) message) according to one embodiment of the present disclosure.

[0265] In one embodiment, the synchronization request message may be transmitted periodically or aperiodically based on whether synchronization is determined to be necessary. The synchronization request message may be transmitted aperiodically in response to a determination that synchronization is necessary. The synchronization request message may also be transmitted periodically.

[0266] A synchronization request message transmitted periodically may include a determination as to whether synchronization of input data is required. The determination may include at least one of: required, unnecessary, or unclear. The synchronization request message may further include a specific field (e.g., a field for identifying a synchronization requirement), and the determination may be included in the specific field. A device (e.g., a base station (1510) or a terminal (1520)) receiving the synchronization request message may reset, redefine, or modify input data for the AI ​​model based on whether the result of the determination as to whether synchronization is required is identified as required based on the synchronization request message.

[0267] Referring to FIG. 15, the terminal (1520) may determine whether synchronization of input data is required. For example, the terminal (1520) may identify an input mismatch (1511, 1513). The mismatch category illustrated in FIG. 15 may include, but is not limited to, a channel error. According to one embodiment, the terminal (1520) may transmit a synchronization request message (e.g., an ARM (AIR / AID request message); AR message) to the base station (1510) based on the detection of two or more input mismatches (1521).

[0268] The base station (1510) may transmit a synchronization instruction message in either the first format (AIR) or the second format (AID) to the terminal (1520) based on the reception of a synchronization request message (1523). For example, the base station (1510) may obtain time information included in the synchronization request message and determine a time point for transmitting the synchronization instruction message based on the time information. This may be intended to allow the synchronization instruction message to be synchronized with the AI ​​synchronization request message. The time information may be defined, for example, by a predetermined time offset. The predetermined time offset may be defined, for example, by the number of slots, the number of symbols, or the number of subframes. That is, the time offset corresponding to the time interval from receiving the synchronization request message to transmitting the synchronization instruction message may be set based on information (e.g., slots, symbols, subframes, etc.) that can identify the time offset. For convenience of explanation, the case of transmitting a synchronization instruction message in the second format will be described below as an example.

[0269] The terminal (1520) may redefine or modify input data for the second AI model based on receiving the synchronization instruction message. Furthermore, the base station (1510) may redefine or modify input data for the first AI model in response to transmitting the synchronization instruction message to the terminal (1520) (1525). As the input data for each of the first and second AI models are redefined or modified, the inference values ​​based on the first and second AI models may substantially match.

[0270] The input data to be newly defined or modified can be determined based on the synchronization instruction message. The input data to be synchronized with the second AI model of the terminal (1520) can be newly defined or modified based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message. In addition, the base station (1510) that transmitted the synchronization instruction message can also newly define or modify the input data to be synchronized with the first AI model based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message.

[0271] According to the above-described FIG. 15, a synchronization request message (e.g., ARM (AIR / AID request message); AR message) can be used for the purpose of requesting a synchronization instruction message of a first format (e.g., AIR message) or a synchronization instruction message of a second format (e.g., AID message) from a first device (e.g., terminal 1520) to a second device (e.g., base station 1510). That is, the AR message can be a request message for an AIR message or an AID message. For example, the terminal 1520 can transmit an AR message to the base station 1510 in response to detecting an input mismatch. The terminal 1520 can resolve the input mismatch by transmitting the AR message to the base station 1520 and receiving an AIR message or an AID message from the base station 1520 (see FIG. 15). For example, the terminal (1520) may use different AR messages depending on whether it is requesting an AIR message or an AID message. For example, the AR message may include a message identifier indicating the type of synchronization instruction message to be requested.

[0272] In one example, the AR message may transmit a model ID or an input mismatch decision category index. The model ID may be, for example, an identifier of an AI model to be initialized or an identifier of an AI model to be newly defined. The model ID may be set during model transfer. The model ID may be set during initial access. The input mismatch decision category index may be, for example, an index for identifying input data to be initialized or newly defined. The input data to be initialized or newly defined may differ, for example, by categories that determine input mismatch. Therefore, by checking the input mismatch decision category, the input data to be initialized or newly defined can be determined. The input mismatch decision category may be classified by factors such as a control channel error or a sudden change in the wireless communication environment, for example.

[0273] FIG. 16 illustrates an example of a base station-driven synchronization procedure using an AI input gather (AIG) message according to one embodiment of the present disclosure.

[0274] In one embodiment, the synchronization collection messages may be transmitted periodically or aperiodically.

[0275] In one embodiment, a synchronization collection message transmitted periodically may include at least one of a first gather field for identifying whether synchronization is required, or a second gather field for identifying whether a synchronization request message has been received. The first gather field may include at least one of information of required, unnecessary, or unclear. The second gather field may include at least one of information of received or not received. In one embodiment, a device (e.g., a base station 1610 or a terminal 1620) receiving the synchronization collection message may deactivate an underlying AI model based on satisfying at least one of the first gather field being identified as 'required' or the second gather field being identified as 'received'.

[0276] In one embodiment, the aperiodically transmitted synchronization collection message may be transmitted to another device (e.g., terminal 1620 or base station 1610) based on a device (e.g., base station 1610 or terminal 1620) determining that synchronization of input data is required or upon receiving a synchronization request message. The other device may be the device that transmitted the synchronization request message.

[0277] Referring to FIG. 16, the terminal (1620) may determine whether synchronization is required for input data. For example, the terminal (1620) may identify an input mismatch (1613, 1615). The mismatch category illustrated in FIG. 16 may include, but is not limited to, a channel error. According to one embodiment, the terminal (1620) may transmit a synchronization request message (e.g., an ARM (AIR / AID request message); AR message) to the base station (1610) based on the detection of two or more input mismatches (1621).

[0278] The base station (1610) may transmit a synchronization gathering message (e.g., an artificial intelligence gather (AIG) message) to the terminal (1620) based on the reception of a synchronization request message (1623). For example, the base station (1610) may obtain time information included in the synchronization request message and determine a time point for transmitting the synchronization gathering message based on the time information. This may be for the purpose of allowing the synchronization gathering message to be synchronized with the synchronization request message. The time information may be defined by, for example, a predetermined time offset. The predetermined time offset may be defined by, for example, the number of slots, the number of symbols, or the number of subframes. That is, the time interval from receiving the synchronization request message to transmitting the synchronization gathering message may be set based on information (e.g., slots, symbols, subframes, etc.) that can identify the time interval.

[0279] The base station (1610) may deactivate the first AI model for a time period (e.g., 1 TTI) to be used for data collection based on transmitting a synchronization collection message. The time period may be defined, for example, by a time offset based on the time point at which the synchronization collection message was transmitted.

[0280] The terminal (1620) may deactivate the second AI model based on receiving a synchronization collection message. The second AI model to be deactivated may be identified based on the model ID field included in the synchronization collection message. The second AI model may be deactivated during a time period for data collection. The time period during which the second AI model is deactivated may coincide with the time period during which the first AI model is deactivated. The time period may be defined as a time offset, for example, by the number of slots, the number of symbols, or the number of subframes.

[0281] The base station (1610) can acquire base data that serves as the basis for a synchronization instruction message in either the first format (AIR) or the second format (AID) during a data collection period. The base station (1610) can generate a synchronization instruction message in either the first format or the second format based on the end of the data collection period, and transmit the generated synchronization instruction message to the terminal (1620) (1627). For convenience of explanation, the case of transmitting a synchronization instruction message in the second format will be described below as an example.

[0282] The terminal (1620) may redefine or modify input data for the second AI model based on receiving the synchronization instruction message. Furthermore, the base station (1610) may redefine or modify input data for the first AI model in response to transmitting the synchronization instruction message to the terminal (1620) (1629). As the input data for each of the first and second AI models are redefined or modified, the inference values ​​based on the first and second AI models may substantially match.

[0283] The input data to be newly defined or modified can be determined based on the synchronization instruction message. The input data to be synchronized with the second AI model of the terminal (1620) can be newly defined or modified based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message. In addition, the base station (1610) that transmitted the synchronization instruction message can also newly define or modify the input data to be synchronized with the first AI model based on at least one of the model ID field, the input index field, and / or the input data field of the synchronization instruction message.

[0284] In various embodiments of the present disclosure, the synchronization instruction message in the first format may be mixed with the synchronization instruction message of [Table 4] in which the synchronization method field is set to the first type of information. In addition, the synchronization instruction message may be mixed with the synchronization instruction message of [Table 4] in which the synchronization method field is set to the second type of information.

[0285] According to the above-described FIG. 16, a synchronization gathering message (e.g., an artificial intelligence gather (AIG) message) can be used to notify the collection of data for transmitting a synchronization instruction message of a first format (e.g., an AIR message) or a synchronization instruction message of a second format (e.g., an AID message). For example, the AIG message can notify whether the data gathering is for transmitting an AIR message or an AID message. For example, the AIG message can indicate whether the target AI model is to be kept on or off while collecting data.

[0286] For example, the AR message may transmit a target AI model status indicator or a time offset. The target AI model status indicator may indicate whether the target AI model is to be kept on or off while collecting data. The target AI model status indicator may include a model ID. The model ID may be, for example, an identifier of an AI model to be initialized or an identifier of an AI model to be newly defined. The target AI model status indicator may be determined, for example, by considering an input mismatch decision category. The input mismatch decision category may be classified by factors such as a control channel error or a sudden change in the wireless communication environment. The target AI model status indicator may be determined, for example, by considering the target AI model. The time offset may indicate, for example, how many slots from the time point of transmission of the AIG message the data will be collected.

[0287] Additionally, the device according to various embodiments of the present specification can be described as follows.

[0288] In a wireless communication system according to one aspect of the present disclosure, a method of operating a first device may include an operation of determining whether synchronization is required for first input data of a first artificial intelligence (AI) model shared with a second device for one function, and an operation of transmitting, in response to a determination that synchronization is required, a first synchronization instruction message to the second device, the first synchronization instruction message instructing synchronization of second input data of a second AI model having substantially the same neural network structure as the first AI model with the first input data.

[0289] According to one embodiment of the present disclosure, the operation of determining whether synchronization is necessary may be an operation of determining that synchronization of the first input data and the second input data is necessary based on a preset cycle.

[0290] According to one embodiment of the present disclosure, the operation of determining whether synchronization is necessary may include the operation of identifying that an input mismatch has occurred between the first AI model and the second AI model, and the operation of determining that synchronization of the first input data and the second input data is necessary in response to identifying that the input mismatch has occurred.

[0291] According to one embodiment of the present disclosure, the operation of determining whether synchronization is necessary may include an operation of identifying that an input mismatch has occurred between the first AI model and the second AI model, and an operation of determining that synchronization of the first input data and the second input data is necessary in response to identifying that the input mismatch has occurred at least N times (where N is a natural number greater than or equal to 2) consecutively.

[0292] According to one embodiment of the present disclosure, the first synchronization instruction message may include at least one of a mismatch category field, a model identifier (ID) field, an input index field, or a time offset field.

[0293] According to one embodiment of the present disclosure, (a) the mismatch category field is provided to indicate at least one of a plurality of preset input mismatch categories, (b) the model ID field is provided to indicate an AI model that requires synchronization for input data, (c) the input index field is provided to indicate input data that requires synchronization, and (d) the type offset field may be provided to indicate the number of slots from the time of receiving a synchronization instruction message to the time of executing synchronization of the input data.

[0294] According to one embodiment of the present disclosure, the method may include an operation of transmitting an AI model to the second device or generating a model ID to be transmitted via the model ID field in response to an initial connection with the second device.

[0295] According to one embodiment of the present disclosure, the method may include an action of resetting the first input data to an initial value, or changing or modifying the first input data to a new value, in response to determining that the synchronization is necessary.

[0296] According to one embodiment of the present disclosure, the operation of determining whether synchronization is necessary may include an operation of identifying an output error based on output data of the first AI model and output data of the second AI model, and an operation of determining that synchronization of the first input data and the second input data is necessary by identifying the output error.

[0297] According to one embodiment of the present disclosure, the first synchronization instruction message may include the second input data to be changed or modified for synchronization with the first input data.

[0298] According to one embodiment of the present disclosure, the operation of transmitting the first synchronization instruction message may include the operation of transmitting a synchronization authorization request (AID uplink request) message to the second device and the operation of receiving a synchronization authorization response (AID uplink grant) message from the second device in response to the synchronization authorization request message.

[0299] According to one embodiment of the present disclosure, the method may include an operation of receiving a synchronization confirmation (AIR / AID confirmation) message from the second device in response to the first synchronization instruction message, and an operation of resetting input data of the first AI model to an initial value or changing or modifying it to a new value in response to receiving the synchronization confirmation message.

[0300] According to one embodiment of the present disclosure, the method may include receiving a synchronization request (AIR / AID request) message from the second device as a requirement for transmitting the first synchronization instruction message to the second device.

[0301] According to one embodiment of the present disclosure, the method may include an operation of transmitting a synchronization gathering (AIG (artificial intelligence gather)) message to the second device in response to receiving the synchronization request message, and an operation of collecting data for synchronization of the first input data and the second input data during a predetermined time offset.

[0302] According to one embodiment of the present disclosure, the method may include an operation of receiving a second synchronization instruction message from the second device, and an operation of resetting the first input data to an initial value or changing or modifying the first input data to a new value in response to receiving the second synchronization instruction message.

[0303] According to one embodiment of the present disclosure, the method may include an operation of receiving a synchronization approval request message from the second device, an operation of transmitting a synchronization approval response message to the second device in response to receiving the synchronization approval request message, an operation of receiving a second synchronization instruction message from the second device in response to the synchronization approval response message, and an operation of resetting the first input data to an initial value or changing or modifying the first input data to a new value in response to receiving the second synchronization instruction message.

[0304] According to one embodiment of the present disclosure, the method may include an operation of receiving a second synchronization instruction message from the second device, an operation of transmitting a synchronization confirmation message to the second device in response to the second synchronization instruction message, and an operation of resetting the first input data to an initial value or changing or modifying the first input data to a new value.

[0305] According to one embodiment of the present disclosure, the method may include an operation of transmitting a synchronization request message to the second device, an operation of receiving a second synchronization instruction message from the second device in response to the synchronization request message, and an operation of resetting the first input data to an initial value or changing or modifying the first input data to a new value in response to receiving the second synchronization instruction message.

[0306] According to one embodiment of the present disclosure, the method may include an operation of transmitting a synchronization request message to the second device, an operation of receiving a synchronization collection message from the second device in response to the synchronization request message, and an operation of deactivating the first AI model for a predetermined time offset in response to receiving the synchronization collection message and collecting data for synchronizing the first input data and the second input data.

[0307] According to one embodiment of the present disclosure, the method may include an operation of deactivating the first AI model during a preset time offset.

[0308] Electronic devices according to various embodiments disclosed herein may take various forms. Electronic devices may include, for example, display devices, portable communication devices (e.g., smartphones), computer devices, portable multimedia devices, portable medical devices, cameras, wearable devices, or home appliances. Electronic devices according to embodiments disclosed herein are not limited to the aforementioned devices.

[0309] The various embodiments of this specification and the terminology used therein are not intended to limit the technical features described in this specification to specific embodiments, but should be understood to include various modifications, equivalents, or substitutes of the embodiments. For example, a component expressed in the singular should be understood to include plural components unless the context clearly indicates only the singular. It should be understood that the term "and / or" used herein encompasses any and all possible combinations of one or more of the listed items. The terms "comprise," "have," "consist of," and the like used herein are intended only to specify the presence of a feature, component, part, or combination thereof described in this specification, and the use of such terms does not exclude the presence or addition of one or more other features, components, parts, or combinations thereof. In this specification, phrases such as "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 each include any one of the items listed together in that phrase, or all possible combinations thereof. Terms such as "first", "second", or "first" or "second" may be used merely to distinguish the corresponding element from other corresponding elements and do not limit the corresponding elements in any other respect (e.g., importance or order).

[0310] The term "part" or "module" used in various embodiments of the present specification may include a unit implemented by hardware, software, or firmware, and may be used interchangeably with terms such as logic, logic block, component, or circuit, for example. The "part" or "module" may be an integrally configured component or a minimum unit of the component or a part thereof that performs one or more functions. For example, according to one embodiment, the "part" or "module" may be implemented in the form of an application-specific integrated circuit (ASIC).

[0311] The term “if” as used in various embodiments of this specification may be interpreted to mean “when” or “when” or “in response to determining” or “in response to detecting,” depending on the context. Similarly, “if it is determined that” or “if this is detected” may be interpreted to mean “upon determining” or “in response to determining” or “upon detecting” or “in response to detecting,” depending on the context.

[0312] The program executed by the electronic device described herein may be implemented as hardware components, software components, and / or a combination of hardware components and software components. The program may be executed by any system capable of executing computer-readable instructions.

[0313] Software may include a computer program, code, instructions, or a combination of one or more of these, which can configure a processing device to perform a desired operation or command the processing device, either independently or collectively. Software may be implemented as a computer program including instructions stored on a computer-readable storage medium. Examples of the computer-readable storage medium include magnetic storage media (e.g., read-only memory (ROM), random-access memory (RAM), floppy disks, hard disks, etc.) and optical reading media (e.g., CD-ROMs, digital versatile discs (DVDs)). The computer-readable storage medium may be distributed across network-connected computer systems so that the computer-readable code can be stored and executed in a distributed manner. Computer programs can be distributed online (e.g., by download or upload) through 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 created on a device-readable storage medium, such as the memory of a manufacturer's server, an application store's server, or an intermediary server.

[0314] According to various embodiments, each component (e.g., a module or a program) of the above-described components may include one or more entities, and some of the entities may be separated and placed in other components. 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 the 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. In a method of operating a first device in a wireless communication system, An operation for determining whether synchronization is required for the first input data of a first artificial intelligence (AI) model shared with a second device for one function; and A method comprising: in response to a determination that said synchronization is necessary, transmitting to the second device a first synchronization instruction message instructing to synchronize second input data of a second AI model having substantially the same neural network structure as the first AI model with the first input data.

2. In paragraph 1, The action that determines whether the above synchronization is necessary is: A method for determining that synchronization of the first input data and the second input data is necessary based on a preset cycle.

3. In paragraph 1, The action that determines whether the above synchronization is necessary is: An operation for identifying that an input mismatch has occurred between the first AI model and the second AI model; and A method comprising: an operation for determining that synchronization of the first input data and the second input data is necessary in response to identifying that the input mismatch has occurred at least N times (where N is a natural number).

4. In paragraph 1, The above first synchronization instruction message is, Contains at least one of a mismatch category field, a model identifier (ID) field, an input index field, or a time offset field; A method according to claim 1, wherein (a) the mismatch category field is provided to indicate at least one of a plurality of preset input mismatch categories, (b) the model ID field is provided to indicate an AI model that requires synchronization for input data, (c) the input index field is provided to indicate input data that requires synchronization, and (d) the type offset field is provided to indicate the number of slots from the time of receiving a synchronization instruction message to the time of executing synchronization of the input data.

5. In paragraph 4, A method comprising the action of transmitting an AI model to said second device or generating a model ID to be transmitted via said model ID field in response to an initial connection with said second device.

6. In paragraph 1, A method comprising resetting the first input data to an initial value, or changing or modifying the first input data to a new value, in response to determining that the synchronization is necessary.

7. In paragraph 1, The action that determines whether the above synchronization is necessary is: An operation for identifying an output error based on the output data of the first AI model and the output data of the second AI model; and A method comprising an operation of determining that synchronization of the first input data and the second input data is necessary by identifying the output error.

8. In paragraph 1, The action of transmitting the above first synchronization instruction message is: An operation of transmitting a synchronization approval request (AID uplink request) message to the second device; and A method comprising the action of receiving a synchronization authorization response (AID uplink grant) message from the second device in response to the synchronization authorization request message.

9. In paragraph 1, An operation of receiving a synchronization confirmation (AIR / AID confirmation) message from the second device in response to the first synchronization instruction message; and A method comprising an action of resetting input data of the first AI model to an initial value or changing or modifying it to a new value in response to receiving the synchronization confirmation message.

10. In paragraph 1, An action of receiving a synchronization request (AIR / AID request) message from the second device as a requirement for transmitting the first synchronization instruction message to the second device; An operation of transmitting a synchronization gathering (AIG (artificial intelligence gather)) message to the second device in response to receiving the synchronization request message; and An operation of collecting data for synchronizing the first input data and the second input data during a predetermined time offset; A method comprising the action of deactivating the first AI model during the predetermined time offset.

11. In paragraph 1, An operation of receiving a second synchronization instruction message from the second device; and A method comprising an action of resetting the first input data to an initial value, or changing or modifying the first input data to a new value, in response to receiving the second synchronization instruction message.

12. In paragraph 1, An action of receiving a synchronization approval request message from the second device; An action of transmitting a synchronization approval response message to the second device in response to receiving the synchronization approval request message; An operation of receiving a second synchronization instruction message from the second device in response to the synchronization approval response message; and A method comprising an action of resetting the first input data to an initial value, or changing or modifying the first input data to a new value, in response to receiving the second synchronization instruction message.

13. In paragraph 1, An operation of receiving a second synchronization instruction message from the second device; An operation of transmitting a synchronization confirmation message to the second device in response to the second synchronization instruction message; and A method comprising an action of resetting the first input data to an initial value or changing or modifying the first input data to a new value.

14. In paragraph 1, An action of transmitting a synchronization request message to the second device; An operation of receiving a second synchronization instruction message from the second device in response to the synchronization request message; and A method comprising an action of resetting the first input data to an initial value, or changing or modifying the first input data to a new value, in response to receiving the second synchronization instruction message.

15. In paragraph 1, An action of transmitting a synchronization request message to the second device; An operation of receiving a synchronization collection message from the second device in response to the synchronization request message; and A method comprising: in response to receiving the synchronization collection message, deactivating the first AI model for a predetermined time offset and collecting data for synchronization of the first input data and the second input data.

Images