Method and device for registering identifier for artificial intelligence model

Abstract

The present disclosure relates to a technique for identifying and managing an artificial intelligence model between a network and a terminal in a system in which wireless communication is performed, and can provide a method and a device, the method comprising the steps of: allocating a model identifier for an artificial intelligence model stored in a terminal; transmitting, to a base station, a registration request message including application information and / or model identifier information about the artificial intelligence model; and receiving, from the base station, a response message corresponding to a registration request message.

Description

Method and device for registering identifiers for artificial intelligence models

[0001] The present disclosure relates to a technology for identifying and managing an artificial intelligence model between a network and a terminal in a system performing wireless communication.

[0002] 3GPP recently approved the "Study on New Radio Access Technology," a study item for research on the next-generation radio access technology (i.e., 5G radio access technology). Based on this, RAN WG1 is currently designing frame structures, channel coding & modulation, waveforms, and multiple access schemes for NR (New Radio). NR is required to be designed to satisfy not only improved data rates compared to LTE, but also various QoS requirements required for each segmented and specific usage scenario.

[0003] Representative usage scenarios of NR include eMBB (enhancement Mobile BroadBand), mMTC (massive Machine Type Communication), and URLLC (Ultra Reliable and Low Latency Communications), and a flexible frame structure design compared to LTE is required to satisfy the needs of each usage scenario.

[0004] Since each usage scenario has different requirements for data rates, latency, reliability, coverage, etc., there is a need for a method to efficiently multiplex radio resource units based on different numerologies (e.g., subcarrier spacing, subframe, TTI (Transmission Time Interval), etc.) to efficiently satisfy the requirements of each usage scenario through the frequency band that constitutes an arbitrary NR system.

[0005] As part of this effort, artificial intelligence and machine learning technologies are being introduced into the wireless communications sector, necessitating a specific design that can manage the AI / ML models being applied.

[0006] The present disclosure provides a technology for registering and efficiently managing artificial intelligence and machine learning in a wireless communication system.

[0007] In one aspect, the present embodiments may provide a method for registering an artificial intelligence model by a terminal, the method including: assigning a model identifier for an artificial intelligence model stored in the terminal; transmitting a registration request message including at least one of application information and model identifier information for the artificial intelligence model to a base station; and receiving a response message corresponding to the registration request message from the base station.

[0008] In another aspect, the present embodiments may provide a method for a base station to register an artificial intelligence model, the method including: receiving a registration request message from a terminal, the registration request message including at least one of model identifier information assigned to an artificial intelligence model stored in the terminal and application information for the artificial intelligence model; registering a model identifier for the artificial intelligence model; and transmitting a response message corresponding to the registration request message to the terminal.

[0009] In another aspect, the present embodiments may provide a terminal device including a control unit that allocates a model identifier for an artificial intelligence model stored in the terminal, a transmitter that transmits a registration request message including at least one of application information and model identifier information for the artificial intelligence model to a base station, and a receiver that receives a response message corresponding to the registration request message from the base station, in a terminal for registering an artificial intelligence model.

[0010] In another aspect, the present embodiments may provide a base station device including a receiving unit that receives a registration request message from a terminal, which includes at least one of model identifier information assigned to an artificial intelligence model stored in the terminal and application information for the artificial intelligence model, a control unit that registers a model identifier for the artificial intelligence model, and a transmitting unit that transmits a response message corresponding to the registration request message to the terminal, in a base station that registers an artificial intelligence model.

[0011] The present disclosure can provide the effect of enabling the application of artificial intelligence technology to a wireless communication system by registering and efficiently managing artificial intelligence and machine learning.

[0012] FIG. 1 is a schematic diagram illustrating the structure of an NR wireless communication system to which the present embodiment can be applied.

[0013] FIG. 2 is a drawing for explaining a frame structure in an NR system to which the present embodiment can be applied.

[0014] FIG. 3 is a diagram for explaining a resource grid supported by a wireless access technology to which the present embodiment can be applied.

[0015] FIG. 4 is a diagram for explaining a bandwidth part supported by a wireless access technology to which the present embodiment can be applied.

[0016] FIG. 5 is a diagram illustrating an example of a synchronization signal block in a wireless access technology to which the present embodiment can be applied.

[0017] FIG. 6 is a diagram for explaining a random access procedure in a wireless access technology to which the present embodiment can be applied.

[0018] Figure 7 is a drawing for explaining CORESET.

[0019] Figure 8 is a flowchart for explaining terminal operation according to one embodiment.

[0020] Figure 9 is a flowchart for explaining base station operation according to one embodiment.

[0021] Fig. 10 is a diagram for exemplarily explaining the configuration of an artificial intelligence model identifier according to one embodiment.

[0022] FIG. 11 is a signal diagram illustrating an artificial intelligence model identifier registration procedure according to one embodiment.

[0023] FIG. 12 is a signal diagram illustrating an artificial intelligence model identifier registration procedure according to another embodiment.

[0024] FIG. 13 is a diagram for explaining the configuration of a model activation instruction signal using an artificial intelligence model identifier according to one embodiment.

[0025] FIG. 14 is a diagram for explaining the configuration of a model activation instruction signal using an artificial intelligence model identifier according to another embodiment.

[0026] FIG. 15 is a diagram for explaining a model activation procedure using an artificial intelligence model identifier according to one embodiment.

[0027] FIG. 16 is a diagram for explaining the configuration of a model deactivation instruction signal using an artificial intelligence model identifier according to one embodiment.

[0028] FIG. 17 is a diagram for explaining the configuration of a model deactivation instruction signal using an artificial intelligence model identifier according to another embodiment.

[0029] FIG. 18 is a diagram for explaining the configuration of a model deactivation instruction signal using an artificial intelligence model identifier according to another embodiment.

[0030] FIG. 19 is a diagram for explaining the configuration of a model deactivation instruction signal using an artificial intelligence model identifier according to another embodiment.

[0031] FIG. 20 is a diagram illustrating a model deactivation procedure using an artificial intelligence model identifier according to one embodiment.

[0032] FIG. 21 is a diagram for explaining the configuration of a model switching instruction signal using an artificial intelligence model identifier according to one embodiment.

[0033] Fig. 22 is a diagram for explaining the configuration of a model switching instruction signal using an artificial intelligence model identifier according to another embodiment.

[0034] FIG. 23 is a diagram for explaining a model switching procedure using an artificial intelligence model identifier according to one embodiment.

[0035] Figure 24 is a drawing for explaining the configuration of a terminal according to one embodiment.

[0036] Fig. 25 is a drawing for explaining the configuration of a base station according to one embodiment.

[0037] Hereinafter, some embodiments of the present disclosure will be described in detail with reference to exemplary drawings. When adding reference numerals to components in each drawing, identical components may have the same numerals as much as possible even if they are shown in different drawings. In addition, when describing the present embodiments, if it is determined that a detailed description of a related known configuration or function may obscure the gist of the technical idea of ​​the present invention, the detailed description may be omitted. When "includes," "has," "consists of," etc. are used in this specification, other parts may be added unless "only" is used. When a component is expressed in the singular, it may include a case in which the plural is included unless specifically stated otherwise.

[0038] Additionally, terms such as first, second, A, B, (a), (b), etc. may be used to describe components of the present disclosure. These terms are only intended to distinguish the components from other components, and the nature, order, sequence, or number of the components are not limited by the terms.

[0039] In a description of the positional relationship of components, when it is described that two or more components are "connected," "combined," or "connected," it should be understood that the two or more components may be directly "connected," "combined," or "connected," but that the two or more components may also be further "interposed" with another component to be "connected," "combined," or "connected." Here, the other component may be included in one or more of the two or more components that are "connected," "combined," or "connected" to each other.

[0040] In the description of the temporal flow relationship related to components, operation methods, or manufacturing methods, for example, when the temporal or flow relationship is described as “after”, “following”, “next to”, “before”, etc., it may also include cases where it is not continuous, unless “immediately” or “directly” is used.

[0041] Meanwhile, when numerical values ​​or corresponding information (e.g., levels, etc.) for components are mentioned, even without separate explicit description, the numerical values ​​or corresponding information may be interpreted as including an error range that may occur due to various factors (e.g., process factors, internal or external impact, noise, etc.).

[0042] The wireless communication system in this specification refers to a system for providing various communication services such as voice, data packets, etc. using wireless resources, and may include a terminal, a base station, or a core network.

[0043] The embodiments disclosed below can be applied to wireless communication systems using various wireless access technologies. For example, the embodiments can be applied to various wireless access technologies such as CDMA (code division multiple access), FDMA (frequency division multiple access), TDMA (time division multiple access), OFDMA (orthogonal frequency division multiple access), SC-FDMA (single carrier frequency division multiple access), or NOMA (non-orthogonal multiple access). In addition, the wireless access technology may not only refer to a specific access technology, but also to each generation of communication technologies established by various communication agreement organizations such as 3GPP, 3GPP2, WiFi, Bluetooth, IEEE, and ITU. For example, CDMA can be implemented with wireless technologies such as UTRA (universal terrestrial radio access) or CDMA2000. TDMA can be implemented with wireless technologies such as GSM (global system for mobile communications) / GPRS (general packet radio service) / EDGE (enhanced data rates for GSM evolution). OFDMA can be implemented in wireless technologies such as IEEE (Institute of Electrical and Electronics Engineers) 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802-20, and E-UTRA (evolved UTRA). IEEE 802.16m is an evolution of IEEE 802.16e and provides backward compatibility with systems based on IEEE 802.16e.UTRA is part of UMTS (universal mobile telecommunications system). 3GPP (3rd generation partnership project) LTE (long term evolution) is part of E-UMTS (evolved UMTS) that uses E-UTRA (evolved-UMTSterrestrial radio access), employing OFDMA in the downlink and SC-FDMA in the uplink. Thus, the present embodiments can be applied to currently disclosed or commercialized wireless access technologies, as well as wireless access technologies currently under development or to be developed in the future.

[0044] Meanwhile, the term "terminal" in this specification is a comprehensive concept that refers to a device that includes a wireless communication module that performs communication with a base station in a wireless communication system, and should be interpreted as a concept that includes not only UE (User Equipment) in WCDMA, LTE, NR, HSPA, and IMT-2020 (5G or New Radio), but also MS (Mobile Station), UT (User Terminal), SS (Subscriber Station), and wireless device in GSM. In addition, the terminal may be a user portable device such as a smartphone depending on the usage type, and in a V2X communication system, it may mean a vehicle, a device including a wireless communication module in the vehicle, etc. In addition, in the case of a Machine Type Communication system, it may mean an MTC terminal, M2M terminal, URLLC terminal, etc. that is equipped with a communication module to perform machine type communication.

[0045] The base station or cell in this specification refers to an end that communicates with a terminal in terms of a network, and includes various coverage areas such as Node-B, eNB (evolved Node-B), gNB (gNode-B), LPN (Low Power Node), Sector, Site, various types of antennas, BTS (Base Transceiver System), Access Point, Point (e.g., Transmission Point, Reception Point, Transmission / Reception Point), Relay Node, Mega Cell, Macro Cell, Micro Cell, Pico Cell, Femto Cell, RRH (Remote Radio Head), RU (Radio Unit), and Small Cell. In addition, a cell may mean including a BWP (Bandwidth Part) in the frequency domain. For example, a serving cell may mean an Activation BWP of a terminal.

[0046] Since the various cells listed above have a base station that controls one or more cells, the base station can be interpreted in two meanings. 1) It can be a device itself that provides a mega cell, macro cell, micro cell, pico cell, femto cell, or small cell in relation to a wireless area, or 2) it can indicate the wireless area itself. In 1), all devices that provide a given wireless area are controlled by the same entity or that interact to cooperatively configure the wireless area are all indicated as a base station. Depending on how the wireless area is configured, a point, a transceiver point, a transmission point, a reception point, etc. can be an embodiment of a base station. In 2), the wireless area itself that receives or transmits a signal from the perspective of a user terminal or a neighboring base station can also be indicated as a base station.

[0047] In this specification, a cell may mean a component carrier having coverage of a signal transmitted from a transmission / reception point or a transmission / reception point itself.

[0048] Uplink (UL, or uplink) refers to a method of transmitting and receiving data from a terminal to a base station, and downlink (DL, or downlink) refers to a method of transmitting and receiving data from a base station to a terminal. Downlink may refer to communication or a communication path from multiple transmission / reception points to a terminal, and uplink may refer to communication or a communication path from a terminal to multiple transmission / reception points. In this case, in the downlink, the transmitter may be part of the multiple transmission / reception points, and the receiver may be part of the terminal. In addition, in the uplink, the transmitter may be part of the terminal, and the receiver may be part of the multiple transmission / reception points.

[0049] Uplink and downlink transmit and receive control information through control channels such as PDCCH (Physical Downlink Control CHannel) and PUCCH (Physical Uplink Control CHannel), and transmit and receive data by configuring data channels such as PDSCH (Physical Downlink Shared CHannel) and PUSCH (Physical Uplink Shared CHannel). Hereinafter, the situation in which signals are transmitted and received through channels such as PUCCH, PUSCH, PDCCH, and PDSCH is also expressed in the form of 'transmitting and receiving PUCCH, PUSCH, PDCCH, and PDSCH.'

[0050] For clarity of explanation, the technical idea of ​​this invention is described below mainly with reference to the 3GPP LTE / LTE-A / NR (New RAT) communication system, but the technical features of this invention are not limited to the communication system.

[0051] After researching 4G (4th-Generation) communication technology, 3GPP develops 5G (5th-Generation) communication technology to meet the requirements of the next-generation wireless access technology of the ITU-R. Specifically, 3GPP develops LTE-A pro, which is an enhancement of LTE-Advanced technology to meet the requirements of the ITU-R, as a 5G communication technology, and NR, a new communication technology separate from 4G communication technology. Both LTE-A pro and NR refer to 5G communication technology, and in the following, 5G communication technology will be explained with NR as the center, unless a specific communication technology is specified.

[0052] The operating scenario in NR defines various operation scenarios by adding considerations for satellites, automobiles, and new verticals to the existing 4G LTE scenario, and in terms of service, it supports the eMBB (Enhanced Mobile Broadband) scenario, the mMTC (Massive Machine Communication) scenario that has high terminal density but is deployed over a wide area and requires low data rate and asynchronous access, and the URLLC (Ultra Reliability and Low Latency) scenario that requires high responsiveness and reliability and can support high-speed mobility.

[0053] To meet these scenarios, NR introduces a wireless communication system that incorporates new waveform and frame structure technologies, low latency technologies, support for ultra-high frequency bands (mmWave), and forward compatibility technologies. In particular, NR systems offer various technological changes in terms of flexibility to ensure forward compatibility. The key technical features of NR are described below with reference to the drawings.

[0054] <NR 시스템 일반>

[0055] Figure 1 is a schematic diagram illustrating the structure of an NR system to which the present embodiment can be applied.

[0056] Referring to Fig. 1, the NR system is divided into 5GC (5G Core Network) and NR-RAN parts, and the NG-RAN is composed of gNBs and ng-eNBs that provide user plane (SDAP / PDCP / RLC / MAC / PHY) and control plane (RRC) protocol termination for UE (User Equipment). gNBs or gNBs and ng-eNBs are interconnected via the Xn interface. gNBs and ng-eNBs are each connected to the 5GC via the NG interface. The 5GC can be configured to include an AMF (Access and Mobility Management Function) that is responsible for the control plane such as terminal access and mobility control functions, and an UPF (User Plane Function) that is responsible for the control function for user data. NR includes support for both frequency bands below 6 GHz (FR1, Frequency Range 1) and frequency bands above 6 GHz (FR2, Frequency Range 2).

[0057] gNB refers to a base station that provides NR user plane and control plane protocol termination to terminals, and ng-eNB refers to a base station that provides E-UTRA user plane and control plane protocol termination to terminals. The base station described in this specification should be understood to encompass both gNB and ng-eNB, and may also be used to refer to gNB or ng-eNB separately as needed.

[0058] <NR 웨이브 폼,뉴머롤러지 및 프레임 구조>

[0059] NR uses the CP-OFDM waveform with a cyclic prefix for downlink transmission, and CP-OFDM or DFT-s-OFDM for uplink transmission. OFDM technology is easily combined with MIMO (Multiple Input Multiple Output) and offers the advantages of high spectral efficiency and low-complexity receivers.

[0060] Meanwhile, in NR, the requirements for data rates, latency, and coverage differ across the three scenarios mentioned above. Therefore, it is necessary to efficiently satisfy these requirements across the frequency bands that comprise any NR system. To this end, technologies have been proposed to efficiently multiplex radio resources based on multiple different numerologies.

[0061] Specifically, the NR transmission numerator is determined based on the sub-carrier spacing and the cyclic prefix (CP), and is changed exponentially using the μ value as an exponent value of 2 based on 15 kHz, as shown in Table 1 below.

[0062] μsubcarrier intervalCyclic prefixSupported for dataSupported for synch015NormalYesYes130NormalYesYes260Normal, ExtendedYesNo3120NormalYesYes4240NormalNoYes

[0063] As shown in Table 1 above, the numerology of NR can be divided into five types according to the subcarrier spacing. This is different from the fixed 15 kHz subcarrier spacing of LTE, one of the 4G communication technologies. Specifically, the subcarrier spacing used for data transmission in NR is 15, 30, 60, and 120 kHz, and the subcarrier spacing used for synchronization signal transmission is 15, 30, 120, and 240 kHz. In addition, the extended CP is applied only to the 60 kHz subcarrier spacing. Meanwhile, the frame structure in NR is defined as a 10 ms frame consisting of 10 subframes of the same length of 1 ms. One frame can be divided into 5 ms half frames, and each half frame contains 5 subframes. For a 15 kHz subcarrier spacing, one subframe consists of one slot, and each slot consists of 14 OFDM symbols.

[0064] FIG. 2 is a drawing for explaining a frame structure in an NR system to which the present embodiment can be applied.

[0065] Referring to FIG. 2, a slot is fixedly composed of 14 OFDM symbols in the case of a normal CP, but the length of the slot in the time domain may vary depending on the subcarrier spacing. For example, in the case of a numerology with a 15 kHz subcarrier spacing, a slot is composed of 1 ms in length, the same length as a subframe. In contrast, in the case of a numerology with a 30 kHz subcarrier spacing, a slot is composed of 14 OFDM symbols, but two slots may be included in one subframe with a length of 0.5 ms. In other words, subframes and frames are defined with fixed time lengths, while slots are defined by the number of symbols, and their time lengths may vary depending on the subcarrier spacing.

[0066] Meanwhile, NR defines slots as the basic scheduling unit and also introduces mini-slots (or sub-slots, or non-slot-based scheduling) to reduce transmission delay in the wireless section. Using wider subcarrier spacing reduces transmission delay in the wireless section by shortening the length of each slot inversely. Mini-slots (or sub-slots) are designed to efficiently support URLLC scenarios and allow scheduling in units of 2, 4, or 7 symbols.

[0067] Furthermore, unlike LTE, NR defines uplink and downlink resource allocation at the symbol level within a single slot. To reduce HARQ delay, a slot structure was defined that allows HARQ ACK / NACKs to be transmitted directly within the transmission slot. This slot structure is referred to as a self-contained structure and will be described in detail.

[0068] NR is designed to support a total of 256 slot formats, of which 62 are used in 3GPP Rel-15. It also supports a common frame structure that configures FDD or TDD frames through various combinations of slots. For example, it supports a slot structure in which all symbols in a slot are set to downlink, a slot structure in which all symbols are set to uplink, and a slot structure in which downlink and uplink symbols are combined. NR also supports data transmission being distributed and scheduled across one or more slots. Therefore, a base station can use a slot format indicator (SFI) to inform a UE whether a slot is a downlink slot, an uplink slot, or a flexible slot. The base station can indicate the slot format by indicating an index of a table configured through UE-specific RRC signaling using the SFI, and can also indicate it dynamically through DCI (Downlink Control Information) or statically or semi-statically through RRC.

[0069] <NR 물리 자원 >

[0070] In relation to physical resources in NR, antenna ports, resource grids, resource elements, resource blocks, and bandwidth parts are considered.

[0071] Antenna ports are defined such that the channel through which a symbol on an antenna port is carried can be inferred from the channel through which another symbol on the same antenna port is carried. Two antenna ports are said to be quasi co-located (QC / QCL) if the large-scale properties of the channel through which a symbol on one antenna port is carried can be inferred from the channel through which a symbol on the other antenna port is carried. Here, the large-scale properties include one or more of delay spread, Doppler spread, frequency shift, average received power, and received timing.

[0072] FIG. 3 is a diagram for explaining a resource grid supported by a wireless access technology to which the present embodiment can be applied.

[0073] Referring to Figure 3, a resource grid may exist for each numeral, as NR supports multiple numerals on the same carrier. Furthermore, resource grids may exist based on antenna ports, subcarrier spacing, and transmission direction.

[0074] A resource block (RB) consists of 12 subcarriers and is defined solely in the frequency domain. Furthermore, a resource element (RE) consists of one OFDM symbol and one subcarrier. Therefore, as shown in Figure 3, the size of a single RB can vary depending on the subcarrier spacing. NR also defines "Point A," which serves as a common reference point for the RB grid, as well as common RBs and virtual RBs.

[0075] FIG. 4 is a diagram for explaining a bandwidth part supported by a wireless access technology to which the present embodiment can be applied.

[0076] Unlike LTE, where the carrier bandwidth is fixed at 20 MHz, NR sets the maximum carrier bandwidth from 50 MHz to 400 MHz for each subcarrier interval. Therefore, it is not assumed that all terminals will use the entire carrier bandwidth. Accordingly, NR allows terminals to designate bandwidth parts (BWPs) within the carrier bandwidth, as illustrated in Figure 4. Furthermore, bandwidth parts are associated with a single numerology, consist of a subset of consecutive common resource blocks, and can be dynamically activated over time. Each terminal is configured with up to four bandwidth parts for both the uplink and downlink, and data is transmitted and received using the bandwidth parts activated at a given time.

[0077] In the case of a paired spectrum, the uplink and downlink bandwidth parts are set independently, and in the case of an unpaired spectrum, the downlink and uplink bandwidth parts are set in pairs so that they can share a center frequency to prevent unnecessary frequency re-tuning between downlink and uplink operations.

[0078] <NR 초기 접속>

[0079] In NR, a terminal performs cell search and random access procedures to connect to a base station and perform communication.

[0080] Cell search is a procedure in which a terminal synchronizes to the cell of a corresponding base station, obtains a physical layer cell ID, and obtains system information using the synchronization signal block (SSB) transmitted by the base station.

[0081] FIG. 5 is a diagram illustrating an example of a synchronization signal block in a wireless access technology to which the present embodiment can be applied.

[0082] Referring to FIG. 5, SSB is composed of a primary synchronization signal (PSS) and a secondary synchronization signal (SSS), each occupying 1 symbol and 127 subcarriers, and a PBCH spanning 3 OFDM symbols and 240 subcarriers.

[0083] The terminal receives SSB by monitoring SSB in the time and frequency domain.

[0084] SSB can be transmitted up to 64 times in 5ms. Multiple SSBs are transmitted in different transmission beams within 5ms, and the terminal performs detection assuming that SSBs are transmitted every 20ms based on a specific beam used for transmission. The number of beams that can be used for SSB transmission within 5ms can increase as the frequency band increases. For example, up to 4 SSB beams can be transmitted below 3GHz, up to 8 in the frequency band between 3GHz and 6GHz, and up to 64 different beams can be used for SSB transmission in the frequency band above 6GHz.

[0085] SSB contains two symbols in one slot, and the starting symbol and number of repetitions within the slot are determined as follows depending on the subcarrier spacing.

[0086] Meanwhile, unlike SS in conventional LTE, SSB is not transmitted at the center frequency of the carrier bandwidth. This means that SSB can be transmitted even in locations other than the center of the system bandwidth, and when supporting wideband operation, multiple SSBs can be transmitted in the frequency domain. Accordingly, the terminal monitors SSB using the synchronization raster, which is a candidate frequency location for monitoring SSB. The carrier raster, which is the center frequency location information of the channel for initial access, and the synchronization raster are newly defined in NR. The synchronization raster has a wider frequency interval than the carrier raster, which can support the terminal's fast SSB search.

[0087] A UE can obtain the MIB through the PBCH of the SSB. The MIB (Master Information Block) includes the minimum information required for the UE to receive the remaining system information (RMSI, Remaining Minimum System Information) broadcast by the network. In addition, the PBCH may include information on the position of the first DM-RS symbol in the time domain, information for the UE to monitor SIB1 (e.g., SIB1 numerology information, information related to SIB1 CORESET, search space information, PDCCH-related parameter information, etc.), offset information between the common resource block and the SSB (the absolute position of the SSB within the carrier is transmitted through SIB1), etc. Here, the SIB1 numerology information is also applied equally to some messages used in the random access procedure for the UE to access the base station after completing the cell search procedure. For example, the numerology information of SIB1 may be applied to at least one of messages 1 to 4 for the random access procedure.

[0088] The aforementioned RMSI may refer to SIB1 (System Information Block 1), and SIB1 is broadcast periodically (e.g., every 160 ms) in the cell. SIB1 contains information necessary for the UE to perform the initial random access procedure and is periodically transmitted via PDSCH. In order for the UE to receive SIB1, it must receive numerology information used for SIB1 transmission and CORESET (Control Resource Set) information used for SIB1 scheduling via PBCH. The UE checks scheduling information for SIB1 using SI-RNTI in CORESET and acquires SIB1 on PDSCH according to the scheduling information. The remaining SIBs, excluding SIB1, may be transmitted periodically or upon request of the UE.

[0089] FIG. 6 is a diagram for explaining a random access procedure in a wireless access technology to which the present embodiment can be applied.

[0090] Referring to FIG. 6, once cell search is complete, the terminal transmits a random access preamble for random access to the base station. The random access preamble is transmitted via the PRACH. Specifically, the random access preamble is transmitted to the base station via the PRACH, which consists of consecutive radio resources in a specific slot that is periodically repeated. Generally, when a terminal initially accesses a cell, a contention-based random access procedure is performed, and when performing random access for beam failure recovery (BFR), a non-contention-based random access procedure is performed.

[0091] The terminal receives a random access response to the transmitted random access preamble. The random access response may include a random access preamble identifier (ID), an UL Grant (uplink radio resource), a temporary C-RNTI (Temporary Cell - Radio Network Temporary Identifier), and a TAC (Time Alignment Command). Since one random access response may include random access response information for one or more terminals, the random access preamble identifier may be included to indicate which terminal the included UL Grant, temporary C-RNTI, and TAC are valid for. The random access preamble identifier may be an identifier for the random access preamble received by the base station. The TAC may be included as information for the terminal to adjust uplink synchronization. The random access response may be indicated by a random access identifier on the PDCCH, i.e., an RA-RNTI (Random Access - Radio Network Temporary Identifier).

[0092] Upon receiving a valid random access response, the terminal processes the information contained in the random access response and performs scheduled transmission to the base station. For example, the terminal applies TAC and stores a temporary C-RNTI. Furthermore, using the UL Grant, the terminal transmits data stored in its buffer or newly generated data to the base station. In this case, information that identifies the terminal must be included.

[0093] Finally, the terminal receives a downlink message for contention resolution.

[0094] <NR CORESET>

[0095] The downlink control channel in NR is transmitted in a CORESET (Control Resource Set) with a length of 1 to 3 symbols, and transmits uplink / downlink scheduling information, SFI (Slot format Index), and TPC (Transmit Power Control) information.

[0096] To ensure system flexibility, NR introduced the CORESET concept. CORESET (Control Resource Set) refers to time-frequency resources for downlink control signals. A terminal can decode control channel candidates using one or more search spaces within the CORESET time-frequency resources. A QCL (Quasi CoLocation) assumption is established for each CORESET, which is used to inform the characteristics of analog beam direction in addition to the delay spread, Doppler spread, Doppler shift, and average delay assumed by the conventional QCL.

[0097] Figure 7 is a drawing for explaining CORESET.

[0098] Referring to Figure 7, a CORESET can exist in various forms within the carrier bandwidth within a single slot, and in the time domain, a CORESET can consist of up to three OFDM symbols. In addition, a CORESET is defined as a multiple of six resource blocks up to the carrier bandwidth in the frequency domain.

[0099] The first CORESET is indicated via the MIB as part of the initial bandwidth part configuration, allowing the terminal to receive additional configuration and system information from the network. After establishing a connection with the base station, the terminal can receive and configure one or more CORESET information via RRC signaling.

[0100] In this specification, the terms frequency, frame, subframe, resource, resource block, region, band, subband, control channel, data channel, synchronization signal, various reference signals, various signals or various messages related to NR (New Radio) may be interpreted in the past or present meaning or in various meanings used in the future.

[0101] In this specification, the term "artificial intelligence model" encompasses machine learning and deep learning. However, the terms "AI / ML" may also be used where necessary. The term "artificial intelligence model" in this specification is not limited to this.

[0102] 3GPP began research on improving the performance of wireless communication systems by applying AI / ML starting with Rel-18.

[0103] Use cases where AI / ML is used can be considered as follows.

[0104] - Initial set of use cases includes:

[0105] o CSI feedback enhancement, eg, overhead reduction, improved accuracy, prediction

[0106] o Beam management, eg, beam prediction in time, and / or spatial domain for overhead and latency reduction, beam selection accuracy improvement

[0107] o Positioning accuracy enhancements for different scenarios including, eg, those with heavy NLOS conditions

[0108] AI / ML 모델을 무선통신 시스템에 적용함에 있어서, 아래와 같이 용어를 정의할 수 있습니다.

[0109] AI / ML model, terminology and description to identify common and specific characteristics for framework investigations:

[0110] - Characterize the defining stages of AI / ML related algorithms and associated complexity:

[0111] o Model generation, e.g., model training (including input / output, pre- / post-process, online / offline as applicable), model validation, model testing, as applicable

[0112] o Inference operation, e.g., input / output, pre- / post-process, as applicable

[0113] - Identify various levels of collaboration between UE and gNB pertinent to the selected use cases, e.g.,

[0114] o No collaboration: implementation-based only AI / ML algorithms without information exchange [for comparison purposes]

[0115] o Various levels of UE / gNB collaboration targeting at separate or joint ML operation.

[0116] - Characterize lifecycle management of AI / ML model: eg, model training, model deployment, model inference, model monitoring, model updating

[0117] - Dataset(s) for training, validation, testing, and inference

[0118] - Identify common notation and terminology for AI / ML related functions, procedures and interfaces

[0119] To apply AI / ML models to wireless communication systems, various collaborative operations between terminals and networks (base stations) must be defined. Furthermore, definitions for various situations must be established, as shown in Table 2 below.

[0120] List of terminologiesTerminologyDescriptionData collectionA process of collecting data by the network nodes, management entity, or UE for the purpose of AI / ML model training, data analytics and inferenceAI / ML ModelA data driven algorithm that applies AI / ML techniques to generate a set of outputs based on a set of inputs.AI / ML model trainingA process to train an AI / ML Model [by learning the input / output relationship] in a data driven manner and obtain the trained AI / ML Model for inferenceAI / ML model InferenceA process of using a trained AI / ML model to produce a set of outputs based on a set of inputsAI / ML model validationA subprocess of training, to evaluate the quality of an AI / ML model using a dataset different from one used for model training, that helps selecting model parameters that generalize beyond the dataset used for model training.AI / ML model testingA subprocess of training, to evaluate the performance of a final AI / ML model using a dataset different from one used for model training and validation. Differently from AI / ML model validation, testing does not assume subsequent tuning of the model.UE-side (AI / ML) modelAn AI / ML Model whose inference is performed entirely at the UENetwork-side (AI / ML) modelAn AI / ML Model whose inference is performed entirely at the networkOne-sided (AI / ML) modelA UE-side (AI / ML) model or a Network-side (AI / ML) modelTwo-sided (AI / ML) modelA paired AI / ML Model(s) over which joint inference is performed, where joint inference comprises AI / ML Inference whose inference is performed jointly across the UE and the network, i.e, the first part of inference is firstly performed by UE and then the remaining part is performed by gNB, or vice versa.AI / ML model transferDelivery of an AI / ML model over the air interface, either parameters of a model structure known at the receiving end or a new model with parameters. Delivery may contain a full model or a partial model.Model downloadModel transfer from the network to UEModel uploadModel transfer from UE to the networkFederated learning / federated trainingA machine learning technique that trains an AI / ML model across multiple decentralized edge nodes (e.g., UEs, gNBs) each performing local model training using local data samples. The technique requires multiple interactions of the model, but no exchange of local data samples.Offline field dataThe data collected from field and used for offline training of the AI / ML modelOnline field dataThe data collected from field and used for online training of the AI / ML modelModel monitoringA procedure that monitors the inference performance of the AI / ML modelSupervised learningA process of training a model from input and its correspondinglabels.Unsupervised learningA process of training a model without labelled data.Semi-supervised learning A process of training a model with a mix of labelled data and unlabelled dataReinforcement Learning (RL)A process of training an AI / ML model from input (a.k.a. state) and a feedback signal (a.k.a. reward) resulting from the model's output (a.k.a. action) in an environment the model is interacting with.Model activationenable an AI / ML model for a specific functionModel deactivationdisable an AI / ML model for a specific functionModel switchingDeactivating a currently active AI / ML model and activating a different AI / ML model for a specific function.

[0121] 협업 수준을 정의하기 위한 한 측면으로 다음과 같은 네트워크-UE 협업 수준을 고려된다. (Take the following network-UE collaboration levels as one aspect for defining collaboration levels)

[0122] 1. Level x: No collaboration

[0123] 2. Level y: Signaling-based collaboration without model transfer

[0124] 3. Level z: Signaling-based collaboration with model transfer

[0125] Study the following aspects, including the definition of components (if needed) and necessity, in Life Cycle Management:

[0126] * Data collection

[0127] o Note: This also includes associated assistance information, if applicable.

[0128] * Model training

[0129] * [Model registration]

[0130] * Model deployment

[0131] o Note: Terminology is to be defined.

[0132] * [Model configuration]

[0133] * Model inference operation

[0134] * Model selection, activation, deactivation, switching, and fallback operation

[0135] * Model monitoring

[0136] * Model update

[0137] o Note: Terminology is to be defined. This includes model finetuning, retraining, and re-development via online / offline training.

[0138] * Model transfer

[0139] * UE capability

[0140] Note: Some aspects in the list may not have specification impact.

[0141] Note: Aspects with square brackets are tentative.

[0142] Note: More aspects may be added as study progresses.

[0143] Additionally, AI / ML models can be considered for online and offline training separately, as shown in Table 3 below.

[0144] TerminologyDescriptionOnline trainingAn AI / ML training process where the model being used for inference) is (typically continuously) trained in (near) real-time with the arrival of new training samples. Note: the notion of (near) real-time vs. non real-time is context-dependent and is relative to the inference time-scale.Note: This definition only serves as a guidance. There may be cases that may not exactly conform to this definition but could still be categorized as online training by commonly accepted conventions.Note: Fine-tuning / re-training may be done via online or offline training. (This note could be removed when we define the term fine-tuning.)Offline trainingAn AI / ML training process where the model is trained based on collected dataset, and where the trained model is later used or delivered for inference.Note: This definition only serves as a guidance.There may be cases that may not exactly conform to this definition but could still be categorized as offline training by commonly accepted conventions.

[0145] Additionally, in relation to the transmission of AI / ML, it can be defined as in Table 4.

[0146] TerminologyDescriptionAI / ML model deliveryA generic term referring to delivery of an AI / ML model from one entity to another entity in any manner.Note: An entity could mean a network node / function (eg, gNB, LMF, etc.), UE, proprietary server, etc.

[0147] In order for the terminal and base station to perform the aforementioned online or offline training and apply the AI / ML model to the wireless communication system, the life cycle management (LCM) process may require identification information related to the AI / ML model.

[0148] Study LCM procedure on the basis that an AI / ML model has a model ID with associated information and / or model functionality at least for some AI / ML operations

[0149] For this, the following details need to be defined.

[0150] A definition of model ID with associated information and / or model functionality is required (Detailed discussion of model ID with associated information and / or model functionality.)

[0151] Usage of model ID with associated information and / or model functionality based LCM procedure

[0152] Whether support of model ID is provided

[0153] The detailed applicable AI / ML operations

[0154] Additionally, the following mechanisms need to be defined for model selection, activation, deactivation, switching and replacement, at least for the UE-side model and the double-sided model:

[0155] For model selection, activation, deactivation, switching, and fallback at least for UE sided models and two-sided models, study the following mechanisms:)

[0156] - Decision by the network

[0157] o Network-initiated

[0158] o UE-initiated, requested to the network

[0159] - Decision by the UE

[0160] o Event-triggered as configured by the network, UE's decision is reported to network

[0161] o UE-autonomous, UE's decision is reported to the network

[0162] o UE-autonomous, UE's decision is not reported to the network

[0163] Data collection can be performed for various purposes in LCM, including model training, model inference, model monitoring, model selection, and model updating. Each of these can be performed with different requirements and potential specification implications.

[0164] For UE-part / UE-side models, study the following mechanisms for LCM procedures:

[0165] * For functionality-based LCM procedure: indication of activation / deactivation / switching / fallback based on individual AI / ML functionality

[0166] o Note: UE may have one AI / ML model for the functionality, or UE may have multiple AI / ML models for the functionality.

[0167] * For model-ID-based LCM procedure, indication of model selection / activation / deactivation / switching / fallback based on individual model IDs

[0168] For AI / ML models, two model types can be defined as shown in Table 5.

[0169] Proprietary-format modelsML models of vendor- / device-specific proprietary format, from 3GPP perspectiveNOTE: An example is a device-specific binary executable formatOpen-format modelsML models of specified format that are mutually recognizable across vendors and allow interoperability, from 3GPP perspective

[0170] Proprietary-format models are not mutually recognizable across vendors and hide model design information from other vendors when shared. Open-format models are mutually recognizable between vendors and do not hide model design information from other vendors when shared.

[0171] To identify an AI / ML model, a model identifier can be considered, as in Table 6, or a feature-based functional identifier can be considered, as in Table 7.

[0172] TerminologyDescriptionModel identificationA process / method of identifying an AI / ML model for the common understanding between the NW and the UENote: The process / method of model identification may or may not be applicable.Note: Information regarding the AI / ML model may be shared during model identification.

[0173] TerminologyDescriptionFunctionality identificationA process / method of identifying an AI / ML functionality for the common understanding between the NW and the UENote: Information regarding the AI / ML functionality may be shared during functionality identification.FFS: granularity of functionality

[0174] Additionally, AI / ML models can be updated for various reasons, and model updates and model parameter updates can be defined separately, as shown in Table 8.

[0175] TerminologyDescriptionModel updateProcess of updating the model parameters and / or model structure of a modelModel parameter updateProcess of updating the model parameters of a model

[0176] Meanwhile, at least the following cases for model delivery / transfer to UE, training location, and model delivery / transfer format combinations for UE-side models and UE-part of two-sided models can be considered.

[0177] CaseModel delivery / transferModel storage locationTraining locationymodel delivery (if needed) over-the-topOutside 3gpp NetworkUE-side / NW-side / neutral sitez1model transfer in proprietary format3GPP NetworkUE-side / neutral sitez2model transfer in proprietary format3GPP NetworkNW-sidez3model transfer in open format3GPP NetworkUE-side / neutral sitez4model transfer in open format of a known model structure at UE3GPP NetworkNW-sidez5model transfer in open format of an unknown model structure at UE3GPP NetworkNW-side

[0178] As mentioned above, various use cases and terminology have been defined for the use of AI / ML models in wireless communication systems. However, specific procedures for identifier allocation, registration, transmission and activation, and switching operations for using AI / ML models in terminals and networks based on these use cases and definitions have not been specified.

[0179] The present disclosure proposes specific operating procedures and terminal / base station operations in such situations.

[0180] Specifically, research is needed on two main approaches to model management: functionality-based and model-based. However, the relationship between Functionality ID and Model ID, as well as the definitions of AI / ML models and functionality mentioned herein, remain ambiguous. The definition of models that can be handled in functionality-based LCM also needs to be clarified.

[0181] AI / ML-based air interface technology aims to improve system performance by applying AI / ML to several key use cases. While Rel-18 has begun discussions on three use cases—CSI enhancement, beam management, and positioning enhancement—AI / ML is expected to be applied to a wider range of features in the future. This means applying AI / ML models, selected based on the AI / ML capabilities of the terminal and network, to existing features to enable more efficient communication.

[0182] Current research suggests that using a single generalized model across different configurations, scenarios, and sites, even for each use case and sub-use case, can degrade system performance in certain situations. Therefore, it's preferable to utilize specialized models tailored to specific scenarios and settings. For this reason, even within the same sub-use case, more than one model can be defined for different conditions. Furthermore, consideration can be given to applying the appropriate model based on terminal status or assistance information.

[0183] In this way, it may be desirable for the NW to decide on model selection and model management such as activation / deactivation / switching. If the node that actually uses the model is the UE (i.e., UE-side model, UE-part of a two-sided model), an identifier for recognizing the model indicated between the NW and the UE needs to be defined. Furthermore, discussion is needed on how to identify the model ID. For example, it has not been decided whether the model ID will be a globally unique ID, and the model actually used by the terminal may be selective depending on the terminal's capabilities and applicable conditions.

[0184] Using globally unique values ​​for terminal model management may require very large model IDs, making it unsuitable for signaling for model management. Since the model / functionality ID used for model management between terminals and networks will primarily be used to identify models through signaling, a method is needed to allocate IDs that are suitable for this purpose and efficient for signaling.

[0185] In this context, the present disclosure proposes a model identifier configuration method to enable the network to efficiently manage AI / ML models applicable to terminals. Furthermore, the present disclosure proposes a model identifier registration procedure and a model management procedure using the proposed model identifier. Furthermore, the present disclosure proposes activation / deactivation / switching operations for AI / ML models.

[0186] Below, various embodiments are described to propose the aforementioned procedures and operations. Each embodiment can be applied individually to a wireless communication system, or selectively as needed. Alternatively, each embodiment can be applied to a wireless communication system in any combination.

[0187] Meanwhile, the network (NW) described below may refer to a base station. Alternatively, NW may be used to encompass a base station and some objects of the upper core network. Furthermore, each term used for information described below may be modified and applied to various terms that include relevant information for ease of understanding. In other words, there are no restrictions on the terminology.

[0188] Figure 8 is a flowchart for explaining terminal operation according to one embodiment.

[0189] Referring to FIG. 8, a method for registering an artificial intelligence model by a terminal may include a step of assigning a model identifier to an artificial intelligence model stored in the terminal (S810).

[0190] For example, a terminal can store an artificial intelligence (AI / ML) model. The AI ​​model may be trained on the terminal itself, trained externally, and stored on the terminal. Alternatively, the AI ​​model may be transmitted by a base station.

[0191] The terminal may assign a model identifier to an AI model stored in the terminal. For example, the model identifier may be configured using at least one of an identifier assigned by the terminal and a globally unique identifier. For example, the model identifier may be randomly assigned by the terminal. In another example, the model identifier may be assigned as a globally unique identifier. In yet another example, the model identifier may be configured with a globally unique identifier portion and an identifier portion randomly assigned by the terminal.

[0192] Meanwhile, model identifiers can be configured in various forms. For example, when a terminal assigns a model identifier, the model identifier may be configured in two parts.

[0193] For example, a model identifier may be composed of a first part identifier for distinguishing preset categories or functions of an AI model, and a second part identifier arbitrarily assigned by the terminal. The first part identifier may be assigned based on which category the AI ​​model will be classified into or which function it will provide. The second part identifier may be composed of an identifier for distinguishing the AI ​​model within the first part identifier in an arbitrary manner by the terminal.

[0194] Here, preset categories or functions can be distinguished through prior agreement between the terminal and base station. Furthermore, identifiers used to distinguish preset categories or functions can be shared in advance between the terminal and base station. Alternatively, the second part identifier can be arbitrarily distinguished by the terminal.

[0195] Through this, a specific function or category can be distinguished through a first part identifier, and a second part identifier can be assigned to distinguish each AI model within the function or category. The first part identifier and the second part identifier may be composed of the same number of bits or may be composed of different numbers of bits.

[0196] A method for a terminal to register an artificial intelligence model may include a step of transmitting a registration request message including at least one of application information and model identifier information for the artificial intelligence model to a base station (S820).

[0197] A terminal can register an AI model stored in the terminal with the base station by sending a registration request message to the base station. For example, the application information for the AI ​​model may include at least one of the following: capability information for the AI ​​model and condition information regarding the applicable conditions for the AI ​​model.

[0198] The capability information of an AI model may include information that enables the base station to recognize the usability and functionality of the AI ​​model, such as the AI ​​model's supported functions and usability. The applicability condition information of the AI ​​model may include information indicating the conditions under which the AI ​​model can be applied. The registration request message may be transmitted as a higher-layer message. Alternatively, the registration request message may be transmitted from the terminal to the base station via an arbitrary message, such as an L1 / L2 message.

[0199] Meanwhile, the registration request message may further include the AI ​​model if it is in an open format. For example, the registration request message may include information about the AI ​​model. For another example, the registration request message may include information about parameters applied to the AI ​​model. For another example, the registration request message may include data set information for training the AI ​​model. Furthermore, transmitting the AI ​​model involves transmitting various information necessary for the base station to identify or register the AI ​​model stored in the terminal.

[0200] If the terminal needs to transmit an AI model to the base station, the AI ​​model can be included in the registration request message or transmitted to the base station via a separate signal. For example, the AI ​​model can be transmitted to the base station via separate signaling before or after the registration request message.

[0201] If the AI ​​model is in a fixed format rather than an open format, a model identifier is transmitted to indicate the AI ​​model, so that the base station can recognize the AI ​​model, and thus information about a separate AI model may not be included.

[0202] The base station receives a registration request message for an AI model from a terminal and stores and registers the corresponding model identifier. This allows the terminal and base station to distinguish and identify specific AI models using the same model identifier.

[0203] The method for a terminal to register an artificial intelligence model may include a step of receiving a response message corresponding to a registration request message from a base station (S830).

[0204] For example, the terminal may receive a response message to a registration request message via upper layer signaling. Alternatively, the terminal may receive the response message via L1 / L2 signaling.

[0205] The response message may include indication information indicating whether registration for the corresponding model identifier at the base station was successful or failed. If registration for the corresponding model identifier at the base station fails for any reason, the response message may further include information regarding the reason for the failure, if necessary.

[0206] If the response message includes failure reason information, the terminal may send a model identifier re-registration request message depending on the type of failure reason information, or may change the model identifier and re-transmit the registration request message.

[0207] Through the above operations, the terminal can assign a model identifier to an artificial intelligence model stored in the terminal and register it with the base station.

[0208] Figure 9 is a flowchart for explaining base station operation according to one embodiment.

[0209] Referring to FIG. 9, a method for a base station to register an artificial intelligence model may include a step of receiving a registration request message from a terminal that includes at least one of model identifier information assigned to an artificial intelligence model stored in the terminal and application information for the artificial intelligence model (S910).

[0210] For example, a terminal can store an artificial intelligence (AI / ML) model. The AI ​​model may be trained on the terminal itself, trained externally, and stored on the terminal. Alternatively, the AI ​​model may be transmitted by a base station.

[0211] The terminal may assign a model identifier to an AI model stored in the terminal. For example, the model identifier may be configured using at least one of an identifier assigned by the terminal and a globally unique identifier. For example, the model identifier may be randomly assigned by the terminal. In another example, the model identifier may be assigned as a globally unique identifier. In yet another example, the model identifier may be configured with a globally unique identifier portion and an identifier portion randomly assigned by the terminal.

[0212] Meanwhile, model identifiers can be configured in various forms. For example, when a terminal assigns a model identifier, the model identifier may be configured in two parts.

[0213] For example, a model identifier may be composed of a first part identifier for distinguishing preset categories or functions of an AI model, and a second part identifier arbitrarily assigned by the terminal. The first part identifier may be assigned based on which category the AI ​​model will be classified into or which function it will provide. The second part identifier may be composed of an identifier for distinguishing the AI ​​model within the first part identifier in an arbitrary manner by the terminal.

[0214] Here, preset categories or functions can be distinguished through prior agreement between the terminal and base station. Furthermore, identifiers used to distinguish preset categories or functions can be shared in advance between the terminal and base station. Alternatively, the second part identifier can be arbitrarily distinguished by the terminal.

[0215] Through this, a specific function or category can be distinguished through a first part identifier, and a second part identifier can be assigned to distinguish each AI model within the function or category. The first part identifier and the second part identifier may be composed of the same number of bits or may be composed of different numbers of bits.

[0216] For example, application information for an artificial intelligence model may include at least one of capability information for the artificial intelligence model and condition information for applicable conditions for the artificial intelligence model.

[0217] The capability information of an AI model may include information that enables the base station to recognize the usability and functionality of the AI ​​model, such as the AI ​​model's support functions and usability. The applicability condition information of the AI ​​model may include information indicating the conditions under which the AI ​​model can be applied. The registration request message may be received as a higher-layer message. Alternatively, the registration request message may be received by the base station through an arbitrary message, such as an L1 / L2 message.

[0218] Meanwhile, the registration request message may further include the AI ​​model if it is in an open format. For example, the registration request message may include information about the AI ​​model. For another example, the registration request message may include information about parameters applied to the AI ​​model. For another example, the registration request message may include data set information for training the AI ​​model. Furthermore, transmitting the AI ​​model involves transmitting various information necessary for the base station to identify or register the AI ​​model stored in the terminal.

[0219] If the terminal needs to transmit an AI model to the base station, the AI ​​model can be included in the registration request message or received by the base station through a separate signal. For example, the AI ​​model can be received by the base station through separate signaling before or after the registration request message is transmitted.

[0220] If the AI ​​model is in a fixed format rather than an open format, a model identifier is transmitted to indicate the AI ​​model, so that the base station can recognize the AI ​​model, and thus information about a separate AI model may not be included.

[0221] The method for registering an artificial intelligence model by a base station may include a step of registering a model identifier for the artificial intelligence model (S920).

[0222] For example, a base station receives a registration request message for an AI model from a terminal and stores and registers the corresponding model identifier with the base station. This allows the terminal and base station to distinguish and identify specific AI models using the same model identifier.

[0223] The base station can register the AI ​​model stored in the terminal using the model identifier information included in the registration request message. If a core network object exists that registers and manages AI models, the base station can also request registration of the terminal's AI model to that core network object.

[0224] Additionally, base stations can register application information for AI models by mapping it to model identifiers. This allows base stations to efficiently identify and manage AI models on terminals.

[0225] The method for a base station to register an artificial intelligence model may include a step of transmitting a response message corresponding to a registration request message to a terminal (S930).

[0226] For example, the base station may transmit a response message to a registration request message via higher layer signaling. Alternatively, the base station may transmit the response message via L1 / L2 signaling.

[0227] The response message may include indication information indicating whether registration for the corresponding model identifier at the base station was successful or failed. If registration for the corresponding model identifier at the base station fails for any reason, the response message may further include information regarding the reason for the failure, if necessary.

[0228] If the response message includes failure reason information, the terminal may send a model identifier re-registration request message depending on the type of failure reason information, or may change the model identifier and re-transmit the registration request message.

[0229] Through the above operations, the terminal can assign a model identifier to an artificial intelligence model stored in the terminal and register it with the base station.

[0230] Below, we describe various embodiments that can be performed by the aforementioned terminals and base stations. Each embodiment described below can be performed by the terminals and base stations in any combination. Furthermore, the aforementioned base stations are described below as NWs (networks).

[0231] First, this embodiment describes various embodiments of a method for allocating an identifier to be used by NW to efficiently manage a terminal model, and a model identifier registration and model management procedure utilizing the same.

[0232] More specifically, the model identifier of the terminal proposed in this embodiment may be composed of a combination of two IDs defined as a first part ID and a second part ID. For example, the model identifier may be used solely as the first part ID or as a combination of the first part ID and the second part ID, depending on the purpose of model management.

[0233] For example, as shown in Table 10, the first part ID is an ID (0 to n-1) that is mapped one-to-one to n categories classified based on AI / ML related capabilities shared by NW (or described in the specification), and can be used as an ID to identify a feature to which AI / ML is applied or a functionality within a feature.

[0234] Functionality IDFeatureFunctionality0000CSI enhancementSub-use case 1(CSI feedback compression)0001Beam ManagementSub-use case 1(spatial beam prediction)0010Sub-use case 2(temporal beam prediction)0011AI / ML positioningSub-use case 1(Direct AI / ML positioning)0100Sub-use case 2(AI / ML assisted positioning)0101~1111Reserved

[0235] The second part ID may be an ID that can be temporally assigned by the NW or terminal and may be an identifier for distinguishing one or more models belonging to the feature or functionality indicated by the first part ID. Fig. 10 is a diagram for exemplarily explaining the configuration of an artificial intelligence model identifier according to one embodiment.

[0236] Referring to FIG. 10, the second part ID may follow the first part ID, and the full ID (first part ID + second part ID, hereinafter referred to as logical model ID) may be used to uniquely identify different physical AI / ML models within the terminal.

[0237] For example, a model identifier may be defined in a UE-specific manner and have a one-to-one mapping relationship with a globally uniquely identified physical model and a physical model stored within the terminal.

[0238] If the physical model is updated, logical model ID assignment to the updated model can be performed in the following two embodiments.

[0239] 1. An example of using the same logical model ID assigned to the model before updating.

[0240] a) Assign the logical model ID assigned to the previous version of the model to the updated model,

[0241] b) Delete the previous version of the model before updating; or

[0242] a) Perform an update of the model mapped to the logical model ID assigned to the previous version of the model to be updated.

[0243] 2. An embodiment of storing the logical model ID for the previous model as is and additionally assigning a new logical model ID for the updated model.

[0244] a) Assign different logical model IDs to models x and x' before the update.

[0245] The choice of which of the two methods described above to assign model identifiers may depend on the choice of the node (UE or NW) updating the model. However, this means that a specific logical model ID is mapped one-to-one to a single physical model.

[0246] Example 1

[0247] If communication is performed using a model stored in the terminal, the terminal directly assigns a logical model ID to the stored model. The terminal can transmit a message including capability / applicable conditions information about this to the NW. Through this, the terminal registers information about the conditions under which the model can be applied along with the corresponding model identifier to the NW. For example, if a physical model algorithm and parameters exist in the terminal, such as the logical model ID “00100010” in Fig. 10, but the NW does not know the existence of the corresponding model (i.e., if there is no globally unique ID (i.e., physical model ID)), the terminal can assign applicable conditions / capability information that can apply the model and the corresponding logical model ID. The terminal performs the procedure of registering the corresponding model to the NW by transmitting a message including two pieces of information (i.e., the logical model ID and the corresponding applicable condition information) to the NW.

[0248] If the model is defined in an open format, the UE can perform a registration procedure (a globally unique ID may be assigned by the NW) while uploading the physical model to the NW (model transfer to the NW from the UE). Alternatively, if the model is defined in a proprietary format, the UE may only include relevant capability / condition information and not perform a model transfer.

[0249] The aforementioned identifier registration process and physical model transfer process can be performed simultaneously or sequentially. If physical model transfer is performed sequentially, the corresponding logical model ID can be transmitted together.

[0250] Example 2

[0251] This paper describes a case where models stored in the NW are transmitted to a terminal for communication. In this case, it is assumed that the NW can globally identify each model by assigning a globally unique ID to each model when initially storing the models in the NW storage. Based on the AI / ML related capability information received from the terminal, the NW supports inference on the terminal side with a model trained appropriately for the relevant functionality. To this end, the NW may wish to transfer the physical model stored in the NW to the terminal (model transfer). In this case, the NW selects an appropriate model(s) among the models stored in the NW based on the detailed AI / ML related capability / applicable conditions for the supported functionality (i.e., indicated by the first part ID) received from the terminal. The NW may randomly / sequentially assign the selected model(s) among the currently unregistered second part IDs to distinguish them within the corresponding functionality (first part ID).

[0252] NW can complete the model registration process by sending a message to the terminal including a logical model ID consisting of a first part ID and a second part ID and a physical model mapped to the ID.

[0253] The identifier assignment process and physical model transfer process can be performed simultaneously or sequentially. If physical model transfer is performed sequentially, the logical model ID of the terminal mapped to the corresponding model must also be transmitted.

[0254] Meanwhile, when a terminal or NW allocates a second part ID, there are two methods available: reusing an identifier that is already allocated and in use, or allocating a new ID using an identifier that is not in use.

[0255] This means that both the terminal and the network will be able to detect whether a reallocation has occurred, since they both know the logical model ID assigned to the terminal. If an unassigned second-part ID is selected and newly assigned, this means that an additional physical model must be stored on the terminal. Registering an identifier identical to an identifier currently in use means deleting (or discarding / deregistrating) the existing physical model and replacing / storing (updating) it with the new model.

[0256] The first part ID of the model identifier described above can indicate that more than one functionality can be activated simultaneously (for example, by specifying first part IDs #2, 3, and 4, all functionality can be activated or deactivated at once). Even if there are more than one second part IDs identified by any activated first part ID, only the model corresponding to one second part ID can be activated. In other words, the first part ID is an identifier for distinguishing functionalities that can be performed simultaneously, and the corresponding functionality can perform communication by applying one activated physical model.

[0257] A model identifier can also be defined as consisting of multiple part IDs, each with two or more parts, depending on the purpose. For example, the first part ID may be mapped to an ID that recognizes the feature / functionality to which the model is applied, and the second part ID may be assigned to distinguish different models for different applicable conditions (e.g., configurations / scenarios / sites) to which each functionality can be applied. The third part ID may also be assigned to identify version information for each model. Even when composed of multiple parts, each part ID can have a hierarchy with the previous part ID.

[0258] Hereinafter, the process of registering an identifier mapped to a model using a model identifier is defined as a model identifier registration procedure. This procedure may include a model transfer, or may be defined so that the model transfer procedure is performed sequentially after the registration procedure. A terminal and a network that have performed the model identifier registration and model transfer procedures have mutually exchanged information about one or more physical models available for a specific functionality, and this information can be used to perform a model management procedure.

[0259] Terminal and NW operations for the model identifier registration procedure according to an example can be performed as follows.

[0260] FIG. 11 is a signal diagram illustrating an artificial intelligence model identifier registration procedure according to one embodiment.

[0261] Referring to Figure 11, when performing AI / ML enabled communication using a model stored in the NW as in the second embodiment, the procedure for assigning a proposed model identifier from the NW is described. The detailed terminal and NW operations are as follows.

[0262] (Terminal operation)

[0263] 0. The terminal receives a list of first part IDs defined by NW and corresponding functionality mapping information.

[0264] - Alternatively, the first part ID(s) and corresponding functionality mapping information may be described in the specification and stored in advance in the terminal.

[0265] 1. The terminal transmits a message to the network containing the first part ID(s) that are mapped to the functionality(s) supported by the terminal based on the information acquired from 0, and additional capability / applicable condition information(s) corresponding to each ID. (AI / ML related capability transmission)

[0266] 2. Receive one or more newly allocated second part ID(s) (i.e., logical model IDs) for the first part ID(s) transmitted by the terminal from the NW.

[0267] - Receives a physical model (e.g., model structure, parameters, etc.) identified by logical model ID(s) assigned from NW within the above message (or through a message transmitted sequentially thereafter). (Model transfer)

[0268] 3. If the terminal successfully receives 2, it stores the logical model ID and the corresponding physical model in its storage and transmits a message to the base station notifying that registration has been successfully completed.

[0269] - This may mean that the downloaded models have been successfully stored in the terminal's storage, compiled (if compiling required), and trained (if required).

[0270] - If there is a model that failed to save / compile / train, the logical model ID for that model and the cause information may be transmitted in the message (or through the registration failure message).

[0271] (NW motion)

[0272] 0. NW broadcasts a list of defined first part IDs and corresponding functionality mapping information. Alternatively, it can be transmitted unicast upon terminal request.

[0273] - Alternatively, the first part ID(s) and corresponding functionality mapping information are described in the specification.

[0274] 1. Receive a message from the terminal containing one or more first part ID(s) that map to functionality supported by the terminal and additional capability / applicable condition information(s) corresponding to each ID (AI / ML related capability reception).

[0275] 2. Based on the AI / ML capabilities transmitted by the terminal, the NW selects the appropriate model(s) from among the models stored in its model repository and assigns a logical model ID to the selected model. Here, the logical model ID means assigning a second part ID that can uniquely identify the model within the functionality (first part ID) to which it can be applied. At this time, the NW stores the mapping information between the physical model ID (i.e., globally unique IDs) for each physical model and the terminal's logical model ID as the terminal's model information.

[0276] 3. The NW transmits a message to the terminal containing one or more second part ID(s) (i.e., logical model IDs of the terminal) corresponding to each first part ID received from the terminal.

[0277] - Transmits a physical model (e.g., model structure, parameters) identified by the logical model ID(s) assigned by NW within the above message (or through a message transmitted sequentially thereafter). (Model transfer)

[0278] 4. Receive a message from the terminal notifying that registration has been successfully completed.

[0279] - This means that the downloaded models have been successfully stored in the terminal's storage, compiled (if compiling required), and trained (if required).

[0280] - If there is a model that failed to save / compile / train, the logical model ID for that model and the cause information can be transmitted through the message (or registration failure message).

[0281] FIG. 12 is a signal diagram illustrating an artificial intelligence model identifier registration procedure according to another embodiment.

[0282] Referring to Figure 12, the procedure for registering the proposed model identifier with the NW when performing AI / ML-based communication using a model stored in the UE, as in the first embodiment, is described. The detailed terminal and NW operations are as follows.

[0283] (Terminal operation)

[0284] 0. The terminal receives a list of first part IDs defined by NW and corresponding functionality mapping information.

[0285] - Alternatively, the first part ID(s) and corresponding functionality mapping information are described in the specification.

[0286] 1. If there is a model stored in the terminal, assign a first part ID(s) that maps to the functionality for the model stored in the terminal based on the information obtained from step 0, and assign a second part ID to identify the models within the functionality for the model(s) belonging to each functionality.

[0287] 2. The terminal transmits a message to the network that includes applicable condition information (i.e., logical model ID and applicable condition information corresponding to the ID) to which the model identified by the combination of the first and second part IDs (i.e., logical model ID) assigned by the terminal in step 1 can be applied, and additional capability / applicable condition information(s) corresponding to each first part ID. (AI / ML related capability transmission)

[0288] - If necessary, the physical model of the terminal can be uploaded to the NW (Model transfer).

[0289] 3. The terminal receives a message from the NW indicating that the logical model ID(s) of the terminal have been successfully registered.

[0290] (NW motion)

[0291] 0. NW broadcasts a list of defined first part IDs and corresponding functionality mapping information (may be transmitted unicast upon terminal request).

[0292] - Alternatively, the first part ID(s) and corresponding functionality mapping information are described in the specification.

[0293] 1. The NW receives a message from the terminal, including logical model ID(s) assigned by the terminal and their corresponding applicable condition(s), along with first part ID(s) that map to the functionality supported by the terminal and additional capability / applicable condition(s) corresponding to each ID (AI / ML related capability reception).

[0294] 2. NW stores the logical model ID transmitted by the terminal and the corresponding applicable condition(s) information in the terminal's AI / ML-related profile information.

[0295] 3. NW sends a message to the terminal indicating that the model identifier transmitted by the terminal has been successfully registered in NW.

[0296] The present disclosure can also be applied to a case where the aforementioned procedures of FIGS. 11 and 12 occur simultaneously. That is, while performing a procedure for registering logical model ID(s) for a model stored in a terminal to a network, an additional logical model ID assignment process for the model stored in the network can be performed based on AI / ML-related capability information received from the terminal. In this case, the message transmitted from the network can additionally include the logical model ID(s) assigned by the network along with an indicator indicating that the registration of the logical model ID(s) received from the terminal has been successfully completed.

[0297] Below, an embodiment related to a model management procedure for managing a model using a model identifier is described.

[0298] For example, it defines model management procedures (i.e., model activation, model update, model switching, model deactivation / fallback) using registered model identifiers. The NW can instruct the UE to activate model(s) suitable for the UE's condition using the registered model identifier (i.e., logical model ID). The NW determines the UE's model based on assistance information received from the UE or the cell environment information to which the UE belongs. Since the NW basically collects more information than the UE, the model management procedures can be performed according to the NW's decision.

[0299] model activation

[0300] For example, at any given point in time, only one model (second part ID) can be active for a particular functionality (i.e., first part ID) of a terminal.

[0301] Model activation can be instructed from the NW to the UE when the UE performs inference using the model (i.e., UE-sided model or UE-part of two-sided models). This assumes that all pre-steps for model activation on the UE side (e.g., model training, compiling, storing) have been completed. That is, the NW can instruct activation for a model after confirming that both model registration and model transfer procedures have been successfully completed. Here, the model transfer procedure can be performed only when downloading and uploading are required.

[0302] FIG. 13 is a diagram for explaining the configuration of a model activation instruction signal using an artificial intelligence model identifier according to one embodiment. FIG. 14 is a diagram for explaining the configuration of a model activation instruction signal using an artificial intelligence model identifier according to another embodiment.

[0303] Referring to FIGS. 13 and 14, signaling for model activation may use DCI, MAC CE, RRC, NAS, or application layer messages, depending on the model management entity. Regardless of which signaling is used, the message may include at least one of the following information.

[0304] - Activation indicator;

[0305] - First part ID indicator;

[0306] o First part ID information for the functionality that directs Activation.

[0307] * (Fig. 13) 2 n- A directive indicating the first part ID (functionality ID) from 0 to n-1 in bitmap format with a length of -bit; or

[0308] * (Fig. 14) First part ID of n-bit length corresponding to the functionality ID indicating activation.

[0309] - A second part ID associated with the indicated first part ID.

[0310] o (Fig. 13) Constructs the associated second part ID sequentially as many times as the number of indicated / set first part IDs.

[0311] o (Fig. 14) The entire logical model ID is indicated by configuring the second part ID following the first part ID. If more than one logical model is activated, additional second part IDs can be included by setting the Extension bit.

[0312] Figures 13 and 14 illustrate examples of model activation MAC CE formats when configuring MAC CE. The lengths of the first part ID and second part ID can be determined by considering the size of the defined ID. The activation indicator can be indicated by a logical channel identifier (LCID).

[0313] FIG. 15 is a diagram for explaining a model activation procedure using an artificial intelligence model identifier according to one embodiment.

[0314] Referring to Figure 15, the following terminal and base station operations can be performed.

[0315] (Terminal operation)

[0316] 1. The terminal receives a message that includes a first part ID and one second part ID associated with it, and instructs to activate an AI / ML model.

[0317] 2. The terminal activates the functionality and model mapped to the indicated logical model ID (first part ID + second part ID).

[0318] 3. The terminal transmits a message (or signaling / HARQ ACK) indicating that the model has been successfully activated (the message indicating activation has been successfully received).

[0319] (Base station operation)

[0320] 1. The base station decides to activate any AI / ML functionality and model for the terminal.

[0321] 2. The base station transmits a message to the terminal that includes the first part ID for the determined functionality and the second part ID of the model to be activated, and instructs the activation of the AI / ML model.

[0322] 3. The base station receives a message (or signaling / HARQ ACK) indicating that the model has been successfully activated (the message indicating activation has been successfully received).

[0323] Model deactivation and / or fallback

[0324] As another example, at any given point in time, only one model (second part ID) can be active for a particular functionality (i.e., first part ID) of the terminal.

[0325] Model deactivation can be indicated from the network to the terminal when the terminal performs inference using a model (i.e., a UE-sided model or a UE-part of two-sided models). This can be done by instructing the terminal to deactivate AI / ML functionality for AI / ML-based communication using a specific model (i.e., by indicating the first part ID), thereby implicitly indicating the deactivation of the model that was activated for that functionality. This can also imply a return to the fallback procedure with the deactivation.

[0326] FIG. 16 is a diagram illustrating the configuration of a model deactivation instruction signal using an artificial intelligence model identifier according to one embodiment. FIG. 17 is a diagram illustrating the configuration of a model deactivation instruction signal using an artificial intelligence model identifier according to another embodiment.

[0327] Referring to FIGS. 16 and 17, signaling for model deactivation / fallback can use NAS or application layer messages in addition to DCI, MAC CE, and RRC depending on the model management entity, and regardless of which signaling is used, the message must include at least one of the following information.

[0328] - Deactivation / fallback indicator;

[0329] - First part ID indicator;

[0330] o First part ID information for functionality that directs deactivation / fallback

[0331] * (Fig. 16) 2 n - A directive indicating the first part ID (functionality ID) from 0 to n-1 in bitmap format with a length of -bit; or

[0332] * (Fig. 17) An n-bit long first part ID corresponding to the functionality ID indicating deactivation / fallback and an Extension bit indicating the presence or absence of an additional first part ID.

[0333] Since the aforementioned ID configuration ensures that only one model belonging to a first-part ID can be activated, simply specifying the first-part ID can indicate the deactivation of not only the corresponding functionality but also the corresponding model. As described above, utilizing a two-part ID (model identifier) ​​reduces signaling overhead because models can be managed using only the first-part ID, depending on the intended purpose.

[0334] When using a logical model ID, it is desirable to signal model deactivation with only the first part ID, but it can also be defined to signal including the second part ID.

[0335] FIG. 18 is a diagram illustrating the configuration of a model deactivation instruction signal using an artificial intelligence model identifier according to another embodiment. FIG. 19 is a diagram illustrating the configuration of a model deactivation instruction signal using an artificial intelligence model identifier according to another embodiment.

[0336] Referring to FIGS. 18 and 19, a second part ID can be included to indicate deactivation.

[0337] Figures 16 to 19 illustrate examples of model deactivation MAC CE formats when configuring MAC CE. The lengths of the first part ID and second part ID can be determined by considering the size of the defined ID. The deactivation indicator can be indicated by the LCID.

[0338] FIG. 20 is a diagram illustrating a model deactivation procedure using an artificial intelligence model identifier according to one embodiment.

[0339] Referring to Figure 20, the operations of the terminal and base station are as follows.

[0340] (Terminal operation)

[0341] 1. The terminal receives a message that includes the First part ID and instructs AI / ML model deactivation / fallback.

[0342] 2. The terminal disables the functionality / model mapped to the indicated first part ID.

[0343] 3. The terminal transmits a message (or signaling / HARQ ACK) indicating that the model has been successfully deactivated (the message indicating deactivation has been successfully received).

[0344] (Base station operation)

[0345] 1. The base station decides to disable AI / ML functionality / model for any terminal.

[0346] 2. The base station sends a message to the terminal that includes the first part ID for the determined functionality and instructs the deactivation of the AI / ML model.

[0347] 3. The base station receives a message (or signaling / HARQ ACK) indicating that the model has been successfully deactivated (the message indicating deactivation has been successfully received).

[0348] model switching

[0349] As another example, at any given point in time, only one model (second part ID) can be active for a particular functionality (i.e., first part ID) of the terminal.

[0350] Model switching can be instructed from the NW to the UE when the UE performs inference using a model (i.e., UE-sided model or UE-part of two-sided models). This means that the NW instructs the UE to apply a different model based on a state change of the UE for the AI / ML function of the UE performing AI / ML-based communication by applying a specific model, thereby instructing the UE to deactivate the currently activated and used model and to activate the newly instructed model. The model switching proposed in this embodiment can only occur between models associated with the same functionality (first part ID).

[0351] Model switching can take two forms: one uses the previously described model activation format, and the other defines a new format. While retaining the model activation format and procedure is more efficient from an overall signaling overhead perspective, defining a new format to clearly indicate the models being deactivated and activated, indicating each model ID, may be preferable.

[0352] FIG. 21 is a diagram illustrating the configuration of a model switching instruction signal using an artificial intelligence model identifier according to one embodiment. FIG. 22 is a diagram illustrating the configuration of a model switching instruction signal using an artificial intelligence model identifier according to another embodiment.

[0353] First, when using the above model activation format as is, if there is a model already activated for the indicated first part ID, switching to the indicated model is performed by deactivating that model and activating the model indicated by the newly signaled second part ID. This may include cases where a switching directive is included instead of an activation directive.

[0354] Second, when defining a new format of signaling for model switching, model switching can also use NAS or application layer messages in addition to DCI, MAC CE, and RRC, depending on the model management entity. Regardless of the signaling used, the message must include at least one of the following information. Referring to FIGS. 21 and 22, the following indices may be included.

[0355] - Switching indicator;

[0356] - First part ID indicator;

[0357] o First part ID information for the functionality that directs switching.

[0358] * (Fig. 21) 2 n - A directive indicating the first part ID (functionality ID) from 0 to n-1 in bitmap format with a length of -bit; or

[0359] * (Fig. 22) An n-bit long first part ID corresponding to the functionality ID indicating switching and an Extension bit indicating the presence or absence of an additional first part ID.

[0360] - Second part ID associated with the first part ID for deactivation;

[0361] o This means that the second part ID for the model that was activated before receiving the switching instruction will be deactivated.

[0362] - Second part ID associated with the first part ID for activation.

[0363] o This means that the second part ID for the model that needs to be newly activated by the switching directive must be activated.

[0364] o Deactivated second part ID and activated second part ID must have different values ​​as IDs associated with the same first part ID.

[0365] The length of the first part ID and second part ID can be determined by considering the size of the defined ID. The switching indicator can be indicated by the LCID.

[0366] Since the ID configuration in the present disclosure allows only one model for one second part ID belonging to a first part ID to be in an activated state, it can be recognized that the deactivation of the activated model and the activation of the newly received model are indicated by signaling only one second part ID for the first part ID for which switching is required, which has the effect of reducing signaling overhead.

[0367] FIG. 23 is a diagram for explaining a model switching procedure using an artificial intelligence model identifier according to one embodiment.

[0368] Referring to Fig. 23, the model switching operation of the terminal and base station using the same format as the activation message is described.

[0369] (Terminal operation 1)

[0370] 1. The terminal receives a message that includes a first part ID and an associated second part ID and instructs AI / ML model switching.

[0371] 2. The terminal deactivates the model that was activated in association with the indicated first part ID and activates the second part ID indicated by the message.

[0372] 3. The terminal transmits a message (or signaling / HARQ ACK) indicating that the model has been successfully switched (the message indicating switching has been successfully received).

[0373] (Base station operation 1)

[0374] 1. The base station decides to switch models for AI / ML functionality for the terminal.

[0375] 2. The base station transmits a message to the terminal indicating AI / ML model switching, including the first part ID for the functionality that has decided to be switched and two second part IDs associated with the functionality.

[0376] 3. The base station receives a message (or signaling / HARQ ACK) indicating that the model has been successfully switched (the message indicating switching has been successfully received).

[0377] The following is a model switching operation of terminals and base stations using a new format of switching messages.

[0378] (Terminal action 2)

[0379] 1. The terminal receives a message that includes a first part ID and two second part IDs associated with it and instructs AI / ML model switching.

[0380] 2. The terminal deactivates the model mapped to the second part ID that was activated in association with the indicated first part ID, and activates the model for the second part ID newly indicated by the message.

[0381] 3. The terminal transmits a message (or signaling / HARQ ACK) indicating that the model has been successfully switched (the message indicating switching has been successfully received).

[0382] (Base station operation 2)

[0383] 1. The base station decides to switch models for AI / ML functionality for the terminal.

[0384] 2. The base station transmits a message to the terminal that includes the first part ID for the functionality that has decided to be switched, the second part ID for the model that is activated in association with the functionality, and the second part ID for the model to be newly activated, and instructs the switching of the AI / ML model.

[0385] 3. The base station receives a message (or signaling / HARQ ACK) indicating that the model has been successfully switched (the message indicating switching has been successfully received).

[0386] As described above, according to the present embodiments, when performing wireless communication using AI / ML, by identifying the AI / ML model of the terminal through the UE-specific model ID, signaling for the model ID used for managing the NW model can be minimized. In addition, according to the present embodiments, by identifying the UE-specific model based on a combination of two IDs, there is an effect of reducing the overall signaling overhead and facilitating model management through purpose-specific ID application and signaling.

[0387] Below, the configuration of a terminal and a base station capable of performing some or all of the above-described embodiments is described again with reference to the drawings.

[0388] Figure 24 is a drawing for explaining the configuration of a terminal according to one embodiment.

[0389] Referring to FIG. 24, a terminal (2400) for registering an artificial intelligence model may include a control unit (2410) for assigning a model identifier for an artificial intelligence model stored in the terminal, a transmission unit (2420) for transmitting a registration request message including at least one of application information and model identifier information for the artificial intelligence model to a base station, and a reception unit (2430) for receiving a response message corresponding to the registration request message from the base station.

[0390] For example, the control unit (2410) may store an artificial intelligence (AI / ML) model in the terminal. The AI ​​model may have been trained by the terminal or trained externally and stored in the terminal. Alternatively, the AI ​​model may have been transmitted by a base station.

[0391] The control unit (2410) may assign a model identifier to an artificial intelligence model stored in the terminal. For example, the model identifier may be configured using at least one of an identifier assigned by the terminal and a globally unique identifier. For example, the model identifier may be randomly assigned by the terminal. In another example, the model identifier may be designated and assigned as a globally unique identifier. In yet another example, the model identifier may be configured with a globally unique identifier portion and an identifier portion randomly assigned by the terminal.

[0392] Meanwhile, model identifiers can be configured in various forms. For example, when a terminal assigns a model identifier, the model identifier may be configured in two parts.

[0393] For example, a model identifier may be composed of a first part identifier for distinguishing preset categories or functions of an AI model, and a second part identifier arbitrarily assigned by the terminal. The first part identifier may be assigned based on which category the AI ​​model will be classified into or which function it will provide. The second part identifier may be composed of an identifier for distinguishing the AI ​​model within the first part identifier in an arbitrary manner by the terminal.

[0394] Here, preset categories or functions can be distinguished through prior agreement between the terminal and base station. Furthermore, identifiers used to distinguish preset categories or functions can be shared in advance between the terminal and base station. Alternatively, the second part identifier can be arbitrarily distinguished by the terminal.

[0395] Through this, a specific function or category can be distinguished through a first part identifier, and a second part identifier can be assigned to distinguish each AI model within the function or category. The first part identifier and the second part identifier may be composed of the same number of bits or may be composed of different numbers of bits.

[0396] The transmitter (2420) can register an AI model stored in a terminal with the base station by transmitting a registration request message to the base station. For example, the application information for the AI ​​model may include at least one of the capability information for the AI ​​model and the condition information regarding the applicable conditions for the AI ​​model.

[0397] The capability information of an AI model may include information that enables the base station to recognize the usability and functionality of the AI ​​model, such as the AI ​​model's supported functions and usability. The applicability condition information of the AI ​​model may include information indicating the conditions under which the AI ​​model can be applied. The registration request message may be transmitted as a higher-layer message. Alternatively, the registration request message may be transmitted from the terminal to the base station via an arbitrary message, such as an L1 / L2 message.

[0398] Meanwhile, the registration request message may further include the AI ​​model if it is in an open format. For example, the registration request message may include information about the AI ​​model. For another example, the registration request message may include information about parameters applied to the AI ​​model. For another example, the registration request message may include data set information for training the AI ​​model. Furthermore, transmitting the AI ​​model involves transmitting various information necessary for the base station to identify or register the AI ​​model stored in the terminal.

[0399] If the terminal needs to transmit an AI model to the base station, the AI ​​model can be included in the registration request message or transmitted to the base station via a separate signal. For example, the AI ​​model can be transmitted to the base station via separate signaling before or after the registration request message.

[0400] If the AI ​​model is in a fixed format rather than an open format, a model identifier is transmitted to indicate the AI ​​model, so that the base station can recognize the AI ​​model, and thus information about a separate AI model may not be included.

[0401] The base station receives a registration request message for an AI model from a terminal and stores and registers the corresponding model identifier. This allows the terminal and base station to distinguish and identify specific AI models using the same model identifier.

[0402] For example, the receiver (2430) may receive a response message to a registration request message through upper layer signaling. Alternatively, the receiver (2430) may receive the response message through L1 / L2 signaling.

[0403] The response message may include indication information indicating whether registration for the corresponding model identifier at the base station was successful or failed. If registration for the corresponding model identifier at the base station fails for any reason, the response message may further include information regarding the reason for the failure, if necessary.

[0404] When the response message includes failure reason information, the transmitter (2420) may transmit a model identifier re-registration request message depending on the type of failure reason information, or may change the model identifier and re-transmit the registration request message.

[0405] In addition, the control unit (2410) controls the overall operation of the terminal (2400) according to the AI / ML model identification and management method required to perform the aforementioned disclosure.

[0406] The transmitter (2420) and receiver (2430) are used to transmit and receive signals, messages, and data necessary to perform the aforementioned embodiment with the base station.

[0407] Fig. 25 is a drawing for explaining the configuration of a base station according to one embodiment.

[0408] Referring to FIG. 25, a base station (2500) that registers an artificial intelligence model may include a receiving unit (2530) that receives a registration request message from a terminal, which includes at least one of model identifier information assigned to an artificial intelligence model stored in the terminal and application information for the artificial intelligence model, a control unit (2510) that registers a model identifier for the artificial intelligence model, and a transmitting unit (2520) that transmits a response message corresponding to the registration request message to the terminal.

[0409] For example, a terminal can store an artificial intelligence (AI / ML) model. The AI ​​model may be trained on the terminal itself, trained externally, and stored on the terminal. Alternatively, the AI ​​model may be transmitted by a base station.

[0410] The terminal may assign a model identifier to an AI model stored in the terminal. For example, the model identifier may be configured using at least one of an identifier assigned by the terminal and a globally unique identifier. For example, the model identifier may be randomly assigned by the terminal. In another example, the model identifier may be assigned as a globally unique identifier. In yet another example, the model identifier may be configured with a globally unique identifier portion and an identifier portion randomly assigned by the terminal.

[0411] Meanwhile, model identifiers can be configured in various forms. For example, when a terminal assigns a model identifier, the model identifier may be configured in two parts.

[0412] For example, a model identifier may be composed of a first part identifier for distinguishing preset categories or functions of an AI model, and a second part identifier arbitrarily assigned by the terminal. The first part identifier may be assigned based on which category the AI ​​model will be classified into or which function it will provide. The second part identifier may be composed of an identifier for distinguishing the AI ​​model within the first part identifier in an arbitrary manner by the terminal.

[0413] Here, preset categories or functions can be distinguished through prior agreement between the terminal and base station. Furthermore, identifiers used to distinguish preset categories or functions can be shared in advance between the terminal and base station. Alternatively, the second part identifier can be arbitrarily distinguished by the terminal.

[0414] Through this, a specific function or category can be distinguished through a first part identifier, and a second part identifier can be assigned to distinguish each AI model within the function or category. The first part identifier and the second part identifier may be composed of the same number of bits or may be composed of different numbers of bits.

[0415] For example, application information for an artificial intelligence model may include at least one of capability information for the artificial intelligence model and condition information for applicable conditions for the artificial intelligence model.

[0416] The capability information of an AI model may include information that enables the base station to recognize the usability and functionality of the AI ​​model, such as the AI ​​model's support functions and usability. The applicability condition information of the AI ​​model may include information indicating the conditions under which the AI ​​model can be applied. The registration request message may be received as a higher-layer message. Alternatively, the registration request message may be received by the base station through an arbitrary message, such as an L1 / L2 message.

[0417] Meanwhile, the registration request message may further include the AI ​​model if it is in an open format. For example, the registration request message may include information about the AI ​​model. For another example, the registration request message may include information about parameters applied to the AI ​​model. For another example, the registration request message may include data set information for training the AI ​​model. Furthermore, transmitting the AI ​​model involves transmitting various information necessary for the base station to identify or register the AI ​​model stored in the terminal.

[0418] If the terminal needs to transmit an AI model to the base station, the AI ​​model can be included in the registration request message or received by the base station through a separate signal. For example, the AI ​​model can be received by the base station through separate signaling before or after the registration request message is transmitted.

[0419] If the AI ​​model is in a fixed format rather than an open format, a model identifier is transmitted to indicate the AI ​​model, so that the base station can recognize the AI ​​model, and thus information about a separate AI model may not be included.

[0420] The receiver (2530) receives a registration request message for an AI model from a terminal and stores and registers the corresponding model identifier in the base station. This allows the terminal and base station to distinguish and identify specific AI models using the same model identifier.

[0421] The control unit (2510) can register the AI ​​model stored in the terminal with the base station using the model identifier information included in the registration request message. If a core network object for registering and managing the AI ​​model exists, the transmitter unit (2520) can also request registration of the terminal's AI model to the core network object.

[0422] Additionally, the control unit (2510) can register application information for an artificial intelligence model by mapping it to a model identifier. Through this, the control unit (2510) can efficiently identify and manage the artificial intelligence model of the terminal.

[0423] The transmitter (2520) may transmit a response message to the registration request message via upper layer signaling. Alternatively, the transmitter (2520) may transmit the response message via L1 / L2 signaling.

[0424] The response message may include indication information indicating whether registration for the corresponding model identifier at the base station was successful or failed. If registration for the corresponding model identifier at the base station fails for any reason, the response message may further include information regarding the reason for the failure, if necessary.

[0425] If the response message includes failure reason information, the terminal may send a model identifier re-registration request message depending on the type of failure reason information, or may change the model identifier and re-transmit the registration request message.

[0426] In addition, the control unit (2510) controls the overall operation of the base station (2500) according to the AI / ML model identification and management method required to perform the aforementioned disclosure.

[0427] The transmitter (2520) and receiver (2530) are used to transmit and receive signals, messages, and data necessary for performing the aforementioned embodiment to and from the terminal.

[0428] The above-described embodiments may be supported by standard documents disclosed in at least one of the wireless access systems, IEEE 802, 3GPP, and 3GPP2. That is, steps, components, and parts not described in the present embodiments to clearly illustrate the technical concepts herein may be supported by the above-described standard documents. Furthermore, all terms disclosed in this specification may be explained by the above-described standard documents.

[0429] The embodiments described above may be implemented through various means. For example, the embodiments may be implemented through hardware, firmware, software, or a combination thereof.

[0430] In the case of hardware implementation, the method according to the present embodiments may be implemented by one or more ASICs (Application Specific Integrated Circuits), DSPs (Digital Signal Processors), DSPDs (Digital Signal Processing Devices), PLDs (Programmable Logic Devices), FPGAs (Field Programmable Gate Arrays), processors, controllers, microcontrollers, or microprocessors.

[0431] When implemented using firmware or software, the methods according to the present embodiments may be implemented in the form of devices, procedures, or functions that perform the functions or operations described above. The software code may be stored in a memory unit and executed by a processor. The memory unit may be located within or outside the processor and may exchange data with the processor using various known means.

[0432] Additionally, terms such as "system," "processor," "controller," "component," "module," "interface," "model," or "unit" as described above may generally refer to a computer-related entity, such as hardware, a combination of hardware and software, software, or software in execution. For example, the aforementioned components may be, but are not limited to, a process driven by a processor, a processor, a controller, a control processor, an object, a thread of execution, a program, and / or a computer. For example, both an application running on a controller or a processor and the controller or the processor may be components. One or more components may be within a process and / or thread of execution, and the components may be located on a single device (e.g., a system, a computing device, etc.) or distributed across two or more devices.

[0433] The above description is merely an illustrative example of the technical idea of ​​the present disclosure, and those skilled in the art to which the present disclosure pertains will appreciate that various modifications and variations can be made without departing from the essential characteristics of the technical idea of ​​the present disclosure. In addition, the present embodiments are not intended to limit the technical idea of ​​the present disclosure but rather to explain it, and therefore the scope of the technical idea of ​​the present disclosure is not limited by these embodiments. The scope of protection of the present disclosure should be interpreted by the claims below, and all technical ideas within a scope equivalent thereto should be interpreted as being included within the scope of the rights of the present disclosure.

[0434]

[0435] CROSS-REFERENCE TO RELATED APPLICATION

[0436] This patent application claims priority under 35 USC §119(a) to Korean Patent Application No. 10-2023-0046096, filed in Korea on April 7, 2023, and Korean Patent Application No. 10-2024-0043443, filed in Korea on March 29, 2024, the entire contents of which are incorporated herein by reference. This patent application also claims priority in countries other than the United States for the same reasons, the entire contents of which are incorporated herein by reference.

Claims

1. In the method of registering an artificial intelligence model by a terminal, A step of assigning a model identifier to an artificial intelligence model stored in a terminal; A step of transmitting a registration request message including at least one of application information for the artificial intelligence model and model identifier information to a base station; and A method comprising the step of receiving a response message corresponding to the registration request message from the base station.

2. In paragraph 1, The above model identifier is, A method characterized in that it is configured using at least one of an identifier assigned by the terminal and a global unique identifier.

3. In paragraph 1, The above model identifier is, A method characterized in that it comprises a first part identifier for distinguishing a preset category or function for the artificial intelligence model and a second part identifier randomly assigned by the terminal.

4. In paragraph 1, Application information for the above artificial intelligence model is as follows: A method including at least one of capability information of the artificial intelligence model and condition information on applicable conditions of the artificial intelligence model.

5. In paragraph 1, The above registration request message is, A method further including the artificial intelligence model when the artificial intelligence model is in an open format.

6. In the method of registering an artificial intelligence model by a base station, A step of receiving a registration request message from the terminal, the registration request message including at least one of model identifier information assigned to an artificial intelligence model stored in the terminal and application information for the artificial intelligence model; A step of registering the model identifier for the artificial intelligence model; and A method comprising the step of transmitting a response message corresponding to the registration request message to the terminal.

7. In paragraph 6, The above model identifier is, A method characterized in that it is configured using at least one of an identifier assigned by the terminal and a global unique identifier.

8. In paragraph 6, The above model identifier is, A method characterized in that it comprises a first part identifier for distinguishing a preset category or function for the artificial intelligence model and a second part identifier randomly assigned by the terminal.

9. In paragraph 6, Application information for the above artificial intelligence model is as follows: A method including at least one of capability information of the artificial intelligence model and condition information on applicable conditions of the artificial intelligence model.

10. In paragraph 6, The above registration request message is, A method further including the artificial intelligence model when the artificial intelligence model is in an open format.

11. In the terminal registering the artificial intelligence model, A control unit that assigns a model identifier to an artificial intelligence model stored in a terminal; A transmitter that transmits a registration request message including at least one of application information for the artificial intelligence model and model identifier information to a base station; and A terminal including a receiving unit that receives a response message corresponding to the registration request message from the base station.

12. In paragraph 11, The above model identifier is, A terminal characterized in that it is configured using at least one of an identifier assigned by the terminal and a global unique identifier.

13. In paragraph 11, The above model identifier is, A terminal characterized by comprising a first part identifier for distinguishing a preset category or function for the artificial intelligence model and a second part identifier randomly assigned by the terminal.

14. In paragraph 11, Application information for the above artificial intelligence model is as follows: A terminal including at least one of capability information of the artificial intelligence model and condition information on applicable conditions of the artificial intelligence model.

15. In paragraph 11, The above registration request message is, A terminal further including the artificial intelligence model when the artificial intelligence model is in an open format.

Images