Method and apparatus for monitoring models in beam management using artificial intelligence and machine learning

Abstract

The present embodiment relates to a method and an apparatus for monitoring a model in beam management by using artificial intelligence and machine learning, and provides a method comprising the steps of: receiving second reference signal resource set configuration information for monitoring a reference signal of an AI / ML model in relation to a reference signal configured for a terminal; measuring the signal strength or signal quality of the reference signal based on the second reference signal resource set configuration information; and reporting a performance result of the AI / ML model by comparing the measured value of the reference signal with a predicted value of the reference signal inferred via the AI / ML model.

Description

Technical Field

[0001] The present disclosure relates to a method and apparatus for monitoring a model in beam management using artificial intelligence and machine learning in a next-generation radio access network (hereinafter referred to as "New Radio (NR)"). Background Art

[0002] Recently, the 3rd Generation Partnership Project (3GPP) has approved the "Study on New Radio Access Technology", which is a research project for exploring next-generation / 5G radio access technology (hereinafter referred to as "New Radio" or "NR"). Based on the study on new radio access technology, the Radio Access Network Working Group 1 (RAN WG1) has been discussing frame structures, channel coding and modulation, waveforms, and multiple access methods for New Radio (NR). Designing NR not only requires providing increased data transfer rates compared to Long Term Evolution (LTE) / LTE-Advanced, but also needs to meet various QoS requirements in detailed and specific usage scenarios.

[0003] Enhanced Mobile Broadband (eMBB), Massive Machine-Type Communication (mMTC), and Ultra Reliable and Low Latency Communication (URLLC) have been proposed as representative usage scenarios for NR. To meet the requirements of each scenario, NR needs to be designed with a flexible frame structure compared to LTE / LTE-Advanced.

[0004] Since the requirements for data rate, latency, reliability, coverage, etc. are different from each other, a method for efficiently multiplexing radio resource units based on a parameter set different from other numerologies (e.g., subcarrier spacing, subframes, Transmission Time Interval (TTI), etc.) is needed as a method for efficiently meeting the requirements of each usage scenario through the frequency bands constituting any NR system.

[0005] As part of this aspect, artificial intelligence and machine learning technologies are being introduced into the field of wireless communication, so a specific architecture is needed to support the efficient management and deployment of AI / ML models. Summary of the Invention

[0006] Technical Problem

[0007] The present disclosure can provide a method and apparatus for model monitoring in beam management using artificial intelligence and machine learning in NR.

[0008] Technical solution

[0009] In one aspect, the present disclosure can provide a method for a user equipment (UE) to perform model monitoring in beam management using artificial intelligence and machine learning (AI / ML). The method may include: receiving second reference signal resource set configuration information regarding a reference signal (RS) for monitoring the AI / ML model, the second reference signal resource set configuration information being related to the reference signal (RS) configured for the UE; measuring the signal strength or signal quality of the reference signal based on the second reference signal resource configuration information; and comparing the measured value of the reference signal with the predicted value of the reference signal inferred by the AI / ML model to report the performance result of the AI / ML model.

[0010] In another aspect, the present disclosure can provide a method for a base station to perform model monitoring in beam management using artificial intelligence and machine learning. The method may include: sending second reference signal resource set configuration information regarding a reference signal (RS) for monitoring the AI / ML model, the second reference signal resource set configuration information being related to the reference signal (RS) configured for the UE; sending the reference signal based on the second reference signal resource configuration information; and receiving the performance result of the AI / ML model obtained by comparing the measured value of the reference signal with the predicted value of the reference signal inferred by the AI / ML model.

[0011] In another aspect, the present disclosure can provide a user equipment (UE) that performs model monitoring in beam management using artificial intelligence and machine learning (AI / ML). The user equipment may include a transmitter, a receiver, and a controller that controls the operations of the transmitter and the receiver, wherein the controller: receives second reference signal resource set configuration information regarding a reference signal (RS) for monitoring the AI / ML model, the second reference signal resource set configuration information being related to the reference signal (RS) configured for the UE; measures the signal strength or signal quality of the reference signal based on the second reference signal resource configuration information; and compares the measured value of the reference signal with the predicted value of the reference signal inferred by the AI / ML model to report the performance result of the AI / ML model.

[0012] In another aspect, the present disclosure may provide a base station that performs model monitoring in beam management using artificial intelligence and machine learning. The base station may include a transmitter, a receiver, and a controller configured to control the operations of the transmitter and the receiver, wherein the controller: transmits second reference signal resource set configuration information regarding a reference signal (RS) for monitoring an AI / ML model, the second reference signal resource set configuration information being related to the reference signal (RS) configured for a UE; transmits a reference signal based on the second reference signal resource configuration information, and receives a performance result of the AI / ML model obtained by comparing a measured value of the reference signal with a predicted value of the reference signal inferred by the AI / ML model.

[0013] Advantageous Effects

[0014] According to an embodiment, artificial intelligence and machine learning in NR may be used to monitor a model in beam management.

[0015] According to an embodiment, the AI / ML model may be monitored through continuous model monitoring while minimizing disconnection from the cell, and the burden of additional beam management in AI / ML model monitoring may be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a view schematically showing an NR wireless communication system according to an embodiment of the present disclosure.

[0017] Figure 2 is a view schematically showing a frame structure in an NR system according to an embodiment of the present disclosure.

[0018] Figure 3 is a view for explaining a resource grid supported by a radio access technology according to an embodiment of the present disclosure.

[0019] Figure 4 is a view for explaining a bandwidth part supported by a radio access technology according to an embodiment of the present disclosure.

[0020] Figure 5 is a view showing an example of a synchronization signal block in a radio access technology according to an embodiment of the present disclosure.

[0021] Figure 6 is a signal diagram for explaining a random access procedure in a radio access technology according to an embodiment of the present disclosure.

[0022] Figure 7 is a view for explaining a CORESET.

[0023] Figure 8It is a view showing the operation of two UEs performing initial beam management at two different positions during the beam transmission operation of a base station.

[0024] Figure 9 It is a diagram showing the initial access process of a UE and a base station.

[0025] Figure 10 And Figure 11 It is a view showing the setting of candidate beams assigned to a UE.

[0026] Figure 12 It is a view showing the process for beam management.

[0027] Figures 13 to 15 It is a signal flow diagram showing the beam reporting method according to the reporting method.

[0028] Figure 16 It is a flowchart showing the method by which a UE performs model monitoring in beam management using artificial intelligence and machine learning according to an embodiment.

[0029] Figure 17 It is a flowchart showing the method by which a base station performs model monitoring in beam management using artificial intelligence and machine learning according to an embodiment.

[0030] Figure 18 And Figure 19 It is a view showing the operations of performing beam measurement and beam prediction using AI / ML according to an embodiment.

[0031] Figure 20 It is a signal flow diagram showing the operations of a UE and a base station when the UE performs model inference according to an embodiment.

[0032] Figure 21 It is a signal flow diagram showing the operations of a UE and a base station when the base station performs model inference according to an embodiment.

[0033] Figure 22 It is a block diagram showing a UE according to an embodiment.

[0034] Figure 23 It is a block diagram showing a base station according to an embodiment. Detailed Description

[0035] In the following, some embodiments of the present disclosure will be described in detail with reference to the illustrative drawings. In the drawings, the same reference numerals are used throughout the drawings to refer to the same elements, even if they are shown in different drawings. Further, in the following description of the present disclosure, if a detailed description of known functions and configurations incorporated herein may obscure the subject matter of the present disclosure, it will be omitted. When using expressions such as "comprising", "having" or "including" as mentioned herein, any other part may be added, unless the expression "only" is used. When an element is expressed in the singular form, the element may cover the plural form, unless the element is explicitly and particularly mentioned.

[0036] In addition, terms such as first, second, A, B, (A) or (B) etc. may be used herein when describing the components of the present disclosure. Each of these terms does not define the nature, order or sequence of the corresponding component, but is only used to distinguish the corresponding component from other components.

[0037] In describing the positional relationship between components, if two or more components are described as being "connected", "combined" or "coupled" to each other, it should be understood that two or more components may be "connected", "combined" or "coupled" directly to each other, and two or more components may be "connected", "combined" or "coupled" to each other with another component "inserted" therebetween. In this case, the other component may be included in at least one of the two or more components that are "connected", "combined" or "coupled" to each other.

[0038] In the description of a series of operation methods or manufacturing methods, for example, expressions such as "after", "subsequently", "then" and "before" etc. may also cover cases where the operations or processes are not executed continuously, unless the expressions "immediately" or "directly" are used in the expression.

[0039] The numerical values mentioned herein for components or the information corresponding thereto (e.g., levels, etc.) may be interpreted as including the error ranges caused by various factors (e.g., process factors, internal or external influences, noise, etc.), even if no explicit description is provided therefor.

[0040] The wireless communication system in this specification refers to a system for providing various communication services such as voice services and data services using radio resources. The wireless communication system may include a user equipment (UE), a base station, a core network, etc.

[0041] The embodiments disclosed below can be applied to wireless communication systems using various radio access technologies. For example, the embodiments can be applied to various radio access technologies such as Code Division Multiple Access (CDMA), Frequency Division Multiple Access (FDMA), Time Division Multiple Access (TDMA), Orthogonal Frequency Division Multiple Access (OFDMA), Single Carrier Frequency Division Multiple Access (SC-FDMA), or Non-Orthogonal Multiple Access (NOMA), etc. In addition, the radio access technology can refer to generations of communication technologies established by various communication organizations such as 3GPP, 3GPP2, WiFi, Bluetooth, IEEE, or ITU, etc., and specific access technologies. For example, CDMA can be implemented as wireless technologies such as Universal Terrestrial Radio Access (UTRA) or CDMA2000. TDMA can be implemented as wireless technologies such as Global System for Mobile Communications (GSM) / General Packet Radio Service (GPRS) / Enhanced Data Rates for GSM Evolution (EDGE). OFDMA can be implemented as wireless technologies such as IEEE (Institute of Electrical and Electronics Engineers) 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802-20, Evolved UTRA (E-UTRA). IEEE 802.16m is an evolution of IEEE 802.16e, which provides backward compatibility with IEEE802.16e-based systems. UTRA is part of the Universal Mobile Telecommunications System (UMTS). 3GPP (Third Generation Partnership Project) LTE (Long Term Evolution) is part of E-UMTS (Evolved UMTS) using Evolved UMTS Terrestrial Radio Access (E-UTRA), which employs OFDMA in the downlink and SC-FDMA in the uplink. As described above, the embodiments can be applied to radio access technologies that have been enabled or commercialized, and can also be applied to radio access technologies that are under development or will be developed in the future.

[0042] The UE used in this specification must be interpreted in a broad sense, which indicates a device including a wireless communication module that communicates with a base station in a wireless communication system. For example, the UE includes user equipment (UE) in WCDMA, LTE, NR, HSPA, and IMT-2020 (5G or New Radio), mobile stations in GSM, user terminals (UT), subscriber stations (SS), and wireless devices, etc. In addition, the UE can be a portable user device such as a smart phone, or can be a vehicle in a V2X communication system, a device in the vehicle that includes a wireless communication module, etc. (depending on its usage type). In the case of a Machine Type Communication (MTC) system, the UE can refer to an MTC terminal, an M2M terminal, or a URLLC terminal, which employs a communication module capable of performing machine type communication.

[0043] A base station or cell in this specification refers to an endpoint that communicates with a UE via a network and encompasses various coverage areas, such as Node-B, evolved Node-B (eNB), gNode-B, low-power node (LPN), sector, site, various types of antennas, base transceiver system (BTS), access point, point (e.g., transmission point, reception point, or transmission / reception point), relay node, mega cell, macro cell, micro cell, pico cell, femto cell, radio remote head (RRH), radio unit (RU), and small cell, etc. Additionally, a cell can be used to mean a bandwidth part (BWP) in the frequency domain. For example, a serving cell can refer to the active BWP of a UE.

[0044] The various cells listed above are provided with a base station that controls one or more cells, and the base station can be interpreted in the following two meanings. The base station can be 1) a device for providing a connection to a wireless area, such as a mega cell, macro cell, micro cell, pico cell, femto cell, or small cell, or the base station can be 2) the wireless area itself. In the above description 1), the base station can be multiple devices controlled by the same entity and providing a predetermined wireless area, or all devices that interact with each other and cooperatively configure the wireless area. For example, depending on the configuration method of the wireless area, the base station can be a point, transmission / reception point, transmission point, and reception point, etc. In the above description 2), the base station can be a wireless area that enables a user equipment (UE) to send data to other UEs or adjacent base stations and receive data from them.

[0045] In this specification, a cell can refer to the coverage range of a signal transmitted from a transmission / reception point, a component carrier having the coverage range of a signal transmitted from a transmission / reception point (or transmission point), or the transmission / reception point itself.

[0046] The uplink (UL) refers to a scheme for sending data from a UE to a base station, and the downlink (DL) refers to a scheme for sending data from a base station to a UE. The downlink can mean communication or a communication path from multiple transmission / reception points to a UE, and the uplink can refer to communication or a communication path from a UE to multiple transmission / reception points. In the downlink, the transmitter can be part of multiple transmission / reception points, and the receiver can be part of the UE. Additionally, in the uplink, the transmitter can be part of the UE, and the receiver can be part of multiple transmission / reception points.

[0047] The uplink and downlink transmit and receive control information on control channels such as the Physical Downlink Control Channel (PDCCH) and the Physical Uplink Control Channel (PUCCH). The uplink and downlink transmit and receive data on data channels such as the Physical Downlink Shared Channel (PDSCH) and the Physical Uplink Shared Channel (PUSCH). Hereinafter, the transmission and reception of signals on channels such as PUCCH, PUSCH, PDCCH, or PDSCH, etc., may be expressed as "transmit and receive PUCCH, PUSCH, PDCCH, or PDSCH, etc.".

[0048] For clarity, the following description will focus on 3GPP LTE / LET-A / NR (New Radio) communication systems, but the technical features of the present disclosure are not limited to the corresponding communication systems.

[0049] After studying 4G (Fourth Generation) communication technologies, 3GPP has been developing 5G (Fifth Generation) communication technologies to meet the requirements of the ITU-R's next-generation radio access technologies. In particular, 3GPP is developing LTE-A pro as a 5G communication technology by improving LTE-Advanced technologies so that it complies with the ITU-R and the requirements of the new NR communication technology that is completely different from 4G communication technologies. Both LTE-A pro and NR refer to 5G communication technologies. Hereinafter, unless a specific communication technology is specified, 5G communication technologies will be described based on NR.

[0050] Considering satellites, automobiles, and new vertical fields, etc. in typical 4G LTE scenarios, various operation scenarios have been defined in NR to support enhanced mobile broadband (eMBB) scenarios in terms of services; massive machine-type communication (mMTC) scenarios, where UEs cross a vast area with a high UE density and thus require low data rates and asynchronous connections; and ultra-reliable low-latency (URLLC) scenarios, which require high responsiveness and reliability and support high-speed mobility.

[0051] To meet such scenarios, NR introduces a wireless communication system that employs new waveform and frame structure technologies, low-latency technologies, ultra-high frequency band (millimeter wave) support technologies, and forward compatibility supply technologies. Specifically, the NR system has various technical changes in terms of flexibility to provide forward compatibility. The main technical features of NR will be described below with reference to the accompanying drawings.

[0052] <Overview of the NR System>

[0053] Figure 1 is a view schematically showing an NR system to which the present embodiment can be applied.

[0054] Reference Figure 1, the NR system is divided into a 5G core network (5GC) and an NG-RAN part. The NG-RAN includes gNBs and ng-eNBs that provide the user plane (SDAP / PDCP / RLC / MAC / PHY) and the control plane (RRC) protocol terminations for user equipment (UE). Multiple gNBs or a gNB and an ng-eNB are interconnected with each other via the Xn interface. The gNBs and ng-eNBs are connected to the 5GC via the NG interface respectively. The 5GC can be configured to include an access and mobility management function (AMF) for managing the control plane, such as UE connection and mobility control functions, and a user plane function (UPF) for controlling user data. NR supports both frequency bands below 6 GHz (frequency range 1, FR1) and frequency bands equal to or greater than 6 GHz (frequency range 2, FR2).

[0055] A gNB refers to a base station that provides the NR user plane and control plane protocol terminations for a UE. An ng-eNB refers to a base station that provides the E-UTRA user plane and control plane protocol terminations for a UE. The base stations described in this specification should be understood to cover gNBs and ng-eNBs. However, as needed, a base station can also be used to refer to separate gNBs or ng-NBs.

[0056] <NR Waveform, Parameter Set, and Frame Structure>

[0057] NR uses the CP-OFDM waveform with a cyclic prefix for downlink transmission and uses CP-OFDM or DFT-s-OFDM for uplink transmission. OFDM technology is easy to combine with multiple-input multiple-output (MIMO) schemes and allows the use of a low-complexity receiver in the case of high frequency efficiency.

[0058] Since the three scenarios described above have mutually different requirements for data rate, latency rate, coverage, etc. in NR, it is necessary to efficiently meet the requirements for each scenario on the frequency bands constituting the NR system. For this purpose, a technique for efficiently multiplexing radio resources based on multiple different parameter sets has been proposed.

[0059] In particular, the NR transmission parameter set is determined based on the subcarrier spacing and the cyclic prefix (CP). As shown in Table 1 below, "μ" is used as the exponent value of 2 to change exponentially based on 15 kHz.

[0060] [Table 1]

[0061] μ Subcarrier spacing Cyclic prefix Support data Support synchronization 0 15 Conventional Yes Yes 1 30 Conventional Yes Yes 2 60 Conventional, extended Yes No 3 120 Conventional Yes Yes 4 240 Conventional No Yes

[0062] As shown in Table 1 above, NR can have five types of parameter sets according to the subcarrier spacing. This is different from LTE, which is one of the 4G communication technologies where the subcarrier spacing is fixed at 15 kHz. In particular, in NR, the subcarrier spacing for data transmission is 15 kHz, 30 kHz, 60 kHz, or 120 kHz, and the subcarrier spacing for synchronization signal transmission is 15 kHz, 30 kHz, 120 kHz, or 240 kHz. In addition, the extended CP is only applied to the subcarrier spacing of 60 kHz. In the frame structure of NR, a frame is defined, which includes 10 subframes, each subframe having the same length of 1 ms, and the frame has a length of 10 ms. A frame can be divided into two half-frames of 5 ms, and each half-frame includes 5 subframes. When the subcarrier spacing is 15 kHz, a subframe includes one time slot, and each time slot includes 14 OFDM symbols. Figure 2 is a view for explaining the frame structure in the NR system to which this embodiment can be applied. Refer to Figure 2 , a time slot includes 14 OFDM symbols, which is fixed in the case of the normal CP, but the length of the time slot in the time domain can vary depending on the subcarrier spacing. For example, in the case of the parameter set with a subcarrier spacing of 15 kHz, the time slot is configured to have the same length of 1 ms as the subframe. On the other hand, in the case of the parameter set with a subcarrier spacing of 30 kHz, the time slot includes 14 OFDM symbols, but a subframe can include two time slots, each having a length of 0.5 ms. That is, the subframe and the frame can be defined using fixed time lengths, and the time slot can be defined as the number of symbols such that its time length varies depending on the subcarrier spacing.

[0063] NR defines the basic unit of scheduling as a time slot and also introduces micro time slots (or sub time slots, or non-time-slot-based scheduling) in order to reduce the transmission delay of the radio section. If a wide subcarrier spacing is used, the length of a time slot will be shortened in inverse proportion to it, thereby reducing the transmission delay in the radio section. The micro time slot (or sub time slot) is designed to efficiently support the URLLC scenario, and the micro time slot can be scheduled in units of 2, 4, or 7 symbols.

[0064] In addition, different from LTE, NR defines the uplink and downlink resource allocation at the symbol level in a time slot. In order to reduce the HARQ delay, a time slot structure capable of directly sending HARQ ACK / NACK in the transmission time slot has been defined. Such a time slot structure is called a "self-contained structure", which will be described.

[0065] NR is designed to support a total of 256 slot formats, and 62 of these slot formats are used in 3GPP Rel-15. In addition, NR supports a general frame structure that constitutes an FDD or TDD frame through various combinations of slots. For example, NR supports: i) a slot structure in which all symbols of a slot are configured for downlink, ii) a slot structure in which all symbols are configured for uplink, and iii) a slot structure in which downlink symbols and uplink symbols are mixed. In addition, NR supports data transmission scheduled to be allocated to one or more slots. Therefore, the base station can use a Slot Format Indicator (SFI) to notify the UE whether the slot is a downlink slot, an uplink slot, or a flexible slot. The base station can notify the slot format by using the SFI to indicate the index of a table configured by UE-specific RRC signaling. In addition, the base station can dynamically indicate the slot format through Downlink Control Information (DCI), or can statically or semi-statically indicate the slot format through RRC signaling.

[0066] <Physical Resources of NR>

[0067] Regarding the physical resources in NR, antenna ports, resource grids, resource elements, resource blocks, and bandwidth parts are considered, etc.

[0068] An antenna port is defined as inferring another channel carrying another symbol on the same antenna port from a channel carrying one symbol on the antenna port. If the large-scale characteristics of the channel carrying the symbol on the antenna port can be inferred from another channel carrying the symbol on another antenna port, then these two antenna ports can have a Quasi-Co-Location or Quasi-Co-Location (QC / QCL) relationship. The large-scale characteristics include at least one of delay spread, Doppler spread, frequency shift, average received power, and reception timing.

[0069] Figure 3 A resource grid supported by a radio access technology according to an embodiment of the present disclosure is shown.

[0070] Reference Figure 3 , the resource grid can exist according to the corresponding parameter set because NR supports multiple parameter sets in the same carrier. In addition, the resource grid can depend on the antenna port, subcarrier spacing, and transmission direction.

[0071] A resource block includes 12 subcarriers and is defined only in the frequency domain. In addition, a resource element includes one OFDM symbol and one subcarrier. Therefore, as Figure 3 shown, the size of a resource block can vary according to the subcarrier spacing. In addition, "Point A" which acts as a common reference point, a common resource block, and a virtual resource block for the resource block grid is defined in NR.

[0072] Figure 4 Shows a bandwidth part supported by a radio access technology according to an embodiment of the present disclosure.

[0073] Unlike LTE where the carrier bandwidth is fixed at 20 MHz, in NR, the maximum carrier bandwidth depends on the subcarrier spacing configured to be 50 MHz to 400 MHz. Therefore, it is not assumed that all UEs use the entire carrier bandwidth. Thus, as Figure 4 shown, in NR, a bandwidth part (BWP) can be specified within the carrier bandwidth such that a UE can use that bandwidth part. Additionally, the bandwidth part can be associated with a parameter set, can include a subset of consecutive common resource blocks, and can be dynamically activated over time. A UE has up to four bandwidth parts in each of the uplink and downlink. The UE uses the activated bandwidth part to transmit and receive data during a given time period.

[0074] In the case of paired spectrum, the uplink bandwidth part and the downlink bandwidth part are configured independently. In the case of unpaired spectrum, to prevent unnecessary frequency retuning between downlink operation and uplink operation, the downlink bandwidth part and the uplink bandwidth part are configured in pairs to share the center frequency.

[0075] <Initial access in NR>

[0076] In NR, a UE performs a cell search and a random access procedure to access a base station and communicate with it.

[0077] Cell search is a process of synchronizing a UE with a cell of a corresponding base station by using a synchronization signal block (SSB) sent from the base station and acquiring the physical layer cell ID and system information.

[0078] Figure 5 Shows an example of a synchronization signal block in a radio access technology according to an embodiment of the present disclosure.

[0079] Referring to Figure 5 , the SSB includes a primary synchronization signal (PSS) and a secondary synchronization signal (SSS) that occupy one symbol and 127 subcarriers, and a PBCH that spans three OFDM symbols and 240 subcarriers.

[0080] The UE monitors the SSB in the time domain and the frequency domain, thereby receiving the SSB.

[0081] The SSB can be transmitted up to 64 times within 5 ms. Multiple SSBs are transmitted within 5 ms through different transmission beams, and the UE assumes to detect based on the SSB transmitted once every 20 ms using a specific beam for transmission. The number of beams available for SSB transmission within 5 ms can increase as the frequency band increases. For example, up to 4 SSB beams can be transmitted on a frequency band of 3 GHz or lower, while up to 8 SSB beams can be transmitted on a frequency band of 3 to 6 GHz. In addition, up to 64 different beams can be used to transmit the SSB on a frequency band of 6 GHz or higher.

[0082] One time slot includes two SSBs, and the starting symbol and the repetition times in the time slot are determined according to the following subcarrier spacing.

[0083] Different from the SS in a typical LTE system, the SSB is not transmitted at the center frequency of the carrier bandwidth. That is to say, the SSB can also be transmitted at a frequency other than the center of the system frequency band, and multiple SSBs can be transmitted in the frequency domain in the case of supporting broadband operation. Therefore, the UE uses a synchronization raster to monitor the SSB, which is the candidate frequency position for monitoring the SSB. The carrier raster and the synchronization raster are the center frequency position information of the channels newly defined in NR for initial connection, and the synchronization raster can support the UE's fast SSB search because its frequency spacing is configured wider than that of the carrier raster.

[0084] The UE can obtain the MIB on the PBCH of the SSB. The MIB (Master Information Block) includes the minimum information for the UE to receive the remaining minimum system information (RMSI) broadcast by the network. In addition, the PBCH can include information about the position of the first DM-RS symbol in the time domain, information for the UE to monitor SIB1 (such as SIB1 parameter set information, information related to SIB1 CORESET, search space information, PDCCH-related parameter information, etc.), and the offset information between the common resource block and the SSB (the absolute SSB position in the carrier is sent through SIB1), etc. The SIB1 parameter set information is also applied to some messages used in the random access process for the UE to access the base station after completing the cell search process. For example, the parameter set information of SIB1 can be applied to at least one of Messages 1 to 4 for the random access process.

[0085] The above-mentioned RMSI may mean SIB1 (System Information Block 1), and SIB1 is broadcast periodically (e.g., 160 ms) in the cell. SIB1 includes the information required for the UE to perform the initial random access procedure, and SIB1 is sent periodically on the PDSCH. To receive SIB1, the UE must receive the parameter set information for SIB1 transmission and the CORESET (Control Resource Set) information for the scheduling of SIB1 on the PBCH. The UE uses the SI-RNTI in the CORESET to identify the scheduling information for SIB1. The UE obtains SIB1 on the PDSCH according to the scheduling information. The remaining SIBs other than SIB1 can be sent periodically, or the remaining SIBs can be sent according to the request of the UE.

[0086] Figure 6 is a view for explaining the random access procedure in the radio access technology to which this embodiment can be applied.

[0087] Refer to Figure 6 , if the cell search is completed, the UE sends a random access preamble for random access to the base station. The random access preamble is sent on the PRACH. In particular, the random access preamble is sent periodically on the PRACH to the base station, and the PRACH includes continuous radio resources in specific time slots that are repeated. Generally, when the UE initially accesses the cell, a contention-based random access procedure is performed, while when the UE performs random access for beam failure recovery (BFR), a non-contention-based random access procedure is performed.

[0088] The UE receives a random access response to the sent random access preamble. The random access response may include a random access preamble identifier (ID), UL grant (uplink radio resources), temporary C-RNTI (temporary cell-radio network temporary identifier), and TAC (time alignment command). Since one random access response may include random access response information for one or more UEs, a random access preamble identifier may be included to indicate for which UE the included UL grant, temporary C-RNTI, and TAC are valid. The random access preamble identifier may be the identifier of the random access preamble received by the base station. The TAC may include information for the UE to adjust the uplink synchronization. The random access response may be indicated by a random access identifier on the PDCCH, that is, a random access-radio network temporary identifier (RA-RNTI).

[0089] Upon receiving a valid random access response, the UE processes the information included in the random access response and performs a scheduled transmission to the base station. For example, the UE applies the TAC and stores the temporary C-RNTI. In addition, the UE uses the UL grant to send the data stored in the UE's buffer or newly generated data to the base station. In this case, the information used to identify the UE must be included in the data.

[0090] Finally, the UE receives a downlink message to resolve the contention.

[0091] <NR CORESET>

[0092] The downlink control channel in NR is transmitted in a CORESET (Control Resource Set) with a length of 1 to 3 symbols, and the downlink control channel transmits uplink / downlink scheduling information, SFI (Slot Format Index), and TPC (Transmit Power Control) information, etc.

[0093] As described above, NR has introduced the concept of CORESET to ensure system flexibility. A CORESET (Control Resource Set) refers to the time-frequency resources for downlink control signals. The UE can use one or more search spaces in the CORESET time-frequency resources to decode control channel candidates. A CORESET-specific QCL (Quasi-Co-Location) assumption is configured and used to provide information about the analog beam direction, as well as characteristics such as delay spread, Doppler spread, Doppler shift, and average delay, which are all characteristics of the existing QCL assumption.

[0094] Figure 7 A CORESET is shown.

[0095] Reference Figure 7 , the CORESET can exist in various forms within the carrier bandwidth in a single slot, and the CORESET can include up to 3 OFDM symbols in the time domain. In addition, the CORESET is defined as a multiple of 6 resource blocks in the frequency domain up to the carrier bandwidth.

[0096] The first CORESET, as part of the initial bandwidth part, is specified (e.g., indicated, allocated) by the MIB to receive additional configuration information and system information from the network. After establishing a connection with the base station, the UE can receive and configure one or more CORESET information through RRC signaling.

[0097] In this specification, a frequency, a frame, a subframe, a resource, a resource block, a region, a band, a sub-band, a control channel, a data channel, a synchronization signal, various reference signals related to NR (New Radio), various signals, or various messages may be interpreted as the meaning used currently or in the past, or may be interpreted as various meanings to be used in the future.

[0098] Wider bandwidth operation

[0099] A typical LTE system supports scalable bandwidth operation for any LTE CC (Component Carrier). That is to say, according to the frequency deployment scenario, an LTE provider can configure a bandwidth from a minimum of 1.4 MHz to a maximum of 20 MHz in a single configured LTE CC, and a common LTE UE supports the transmission / reception capability for a bandwidth of 20 MHz for a single LTE CC (for example, each component carrier).

[0100] However, NR is designed to support a UE having different transmission / reception bandwidth capabilities on a single wideband NR CC. Therefore, it is necessary to configure one or more Bandwidth Parts (BWPs), including sub-divided bandwidths for an NR CC, thereby supporting more flexible and wider bandwidth operations by configuring and activating different bandwidth parts for the corresponding UE.

[0101] In particular, one or more bandwidth parts can be configured by a single serving cell configured for the UE in NR, and the UE is defined to activate one downlink (DL) bandwidth part and one uplink (UL) bandwidth part for uplink / downlink data transmission / reception in the corresponding serving cell. In addition, in the case where multiple serving cells are configured for the UE (that is, the UE to which CA is applied), the UE is also defined to activate one downlink bandwidth part and / or one uplink bandwidth part in each serving cell to perform uplink and / or downlink data transmission and reception using the radio resources of the corresponding serving cell.

[0102] In particular, an initial bandwidth part for the initial access procedure of the UE can be defined in the serving cell; one or more UE-specific bandwidth parts can be configured for each UE by dedicated RRC signaling, and a default bandwidth part for fallback operation can be defined for each UE.

[0103] It is possible to define the simultaneous activation and use of multiple downlink and / or uplink bandwidth parts according to the UE's capabilities and the configuration of the bandwidth parts in the serving cell. However, NR rel-15 defines that only one downlink (DL) bandwidth part and one uplink (UL) bandwidth part are activated and used at a time.

[0104] The present disclosure relates to determining beams for model monitoring and defining new reference signal (RS) resources associated therewith as a method for measuring the accuracy of the deployed AI / ML of a UE capable of performing beam management using AI / ML.

[0105] Beam management in 3GPP NR

[0106] The beam management method of 3GPP NR can be divided into an initial access phase and a phase after cell connection establishment. The UE performing the initial access process configures the initial Tx / Rx beam of the UE through the RACH process. To provide the gNB tx.beam (e.g., gNB transmission beam) configuration to the UE without cell connection, the base station periodically and repeatedly sends the SSB (e.g., by default, the SSB is sent once every 20 ms period within 5 ms) mapped with beams in different directions. The UE can select a qualified SSB by performing signal measurements on the periodically sent SSB and notify the base station of information about the selected tx beam by sending the PRACH preamble mapped for the SSB.

[0107] Figure 8 is a view showing the operations of two UEs performing initial beam management at two different positions during the beam transmission operation of the base station.

[0108] Reference Figure 8 , the base station can use specific time-frequency resources in a preset frame to send the synchronization signal block SSB. In this case, the base station can perform a beam scanning operation by forming various beams.

[0109] For example, beams can be sent by performing spatial division from beam index 0 to 11. If UE 1 performs measurements on the SSB, the signal intensity of beam index 3 that matches the beam direction is the maximum, while the intensities of the surrounding beams are measured to be low, as Figure 8 shown. Similarly, UE 2 can measure the signal intensity of beam index 9 through its position characteristics. Each UE can perform the initial access to the base station by performing a random access process based on the signal intensity measurement results of the SSB.

[0110] Figure 9 is an example view showing the initial access process of the UE and the base station.

[0111] Reference Figure 9 ,the initial access process performed by the UE in NR is as follows.

[0112] 1. The UE receives cell-related parameter information (e.g., PRACH information corresponding to each SSB) required for the initial connection step through the system information message.

[0113] 2. The UE measures the RSRP of the periodically transmitted SSB.

[0114] 3. The UE performs beam (SSB) selection based on the measurement results of the SSB. For example, the UE can select the beam that exhibits the highest RSRP based on the measurement results.

[0115] 4. The UE can notify the base station of the selected initial beam information by sending a preamble belonging to the PRACH resource corresponding to the selected beam to the base station.

[0116] 5. The UE can receive a random access response to the transmitted random access preamble through the selected beam. Thereafter, access to the initial cell can be performed by transmitting / receiving Msg 3 and Msg 4.

[0117] As described above, a base station that does not know the location / beam information of the first-entering UE (i.e., the UE that performs the contention-based random access procedure (CBRA)) can generally set up to 64 beams at the cell level for the beam setting of unconnected UEs. The UE performs an operation of sequentially measuring all beams to find the best beam at its location. As the number of beams in the cell increases, beam selection and cell connection may experience a time delay, and the UE may need to measure more beams, which may increase the power consumption of the UE.

[0118] To solve the above problems, the base station can estimate the approximate location / beam of the initially accessing UE by mapping a wider beam to the SSB, and after the UE accesses the cell, a narrow beam can be set through a beam refinement operation. However, although the narrow beam provides a high data rate for the UE, it is sensitive to the movement or environmental changes of the UE, which makes the possibility of disconnection greater. For this reason, the base station allocates the CSI resources (CSI-RS / SSB) to which the candidate beams are mapped in a UE-specific manner as shown Figure 10 such that the UE continuously measures the environmental beam intensity and reports the measurement results to the base station. This can be set by the base station through CSI resource configuration and CSI report configuration.

[0119] The UE configured with beam reporting performs reporting based on the configuration of the reference signal (RS) allocated to the UE by measurement. It follows the CSI framework defined by 3GPP. However, there is a problem with the UE-specific CSI configuration method that as the number of UEs in the cell increases, the RS resources allocated to each UE also increase rapidly. To alleviate the resource overhead problem, the base station can choose a method of allocating the same candidate beams (CSI resources) to UEs at similar locations, as shown Figure 11As shown (UE group-specific CSI resource configuration). However, if UEs with different mobilities share the same resources, a problem arises where new candidate beam resources should be allocated to UEs outside the resource area. If the minimum candidate beam is allocated to UEs with high / medium mobility as a way to reduce resource overhead, the UEs will experience frequent RRC reconfigurations, and the reconfiguration of candidate beams via RRC will result in a relatively large delay, which may cause beam disconnections. The base station can operate candidate beams by appropriately increasing the number of beams belonging to the CSI resource to alleviate this problem. However, a trade-off problem emerges. From the perspective of the UE, the measurement burden also increases due to the increase in the number of beams.

[0120] Currently in NR, in terms of the physical layer, beam management is generally divided into three categories (P1, P2, and P3). Refer to Figure 12 , P1 refers to the operation of discovering transmission / reception (tx / rx) beam pairing while simultaneously performing UE beam scanning and transmission / reception point (TRP) beam scanning, similar to the beam configuration performed during the initial access process. The UE that enters the connected mode identifies the beams to be scanned configured by the base station via the CSI resource set, and performs signal strength measurements on the TRP beams. If the TRP beam of the UE is selected through P2, the base station continuously transmits the selected beam through P3. The UE can select the UE beam while performing UE beam scanning. The specific beam selection at this time depends on the implementation. This operation may be applicable to DL / UL.

[0121] Beam scanning uses a method that implicitly indicates beam information by mapping beam information to RS resource information, where the base station notifies the UE of the reference signal resource information through a specific candidate beam (CSI resource set) setting. In other words, the base station can identify the beam information it maps through the RS resource indicator (RI) using the index information implicitly associated with the RS, rather than notifying the UE of the actual beam index. This is configured using the 3GPP CSI framework, and the UE implicitly reports the RSRP information about the four best beam RIs to the base station by measuring the RS intensity of the resources configured by the base station.

[0122] The method of reporting this information also follows the RRC configuration of the base station and is defined to be one of three methods: periodic reporting, aperiodic reporting, and semi-persistent reporting.

[0123] As Figure 13 shown, the periodic CSI reporting method is triggered through RRC configuration. The UE that receives the configuration of CSI-related RS resources and the reporting method through the RRC message measures the signal strength of the beam according to this configuration through the periodically received RS, and repeatedly reports the measured results.

[0124] In Figure 14 , although CSI-related RS resources and reporting methods are configured through RRC messages in the aperiodic CSI reporting method, RS-based beam measurements are not performed unless a trigger message is received at a lower layer. Only when a trigger indication is received through MAC CE or DCI, the signal strength of the measurement beam is transmitted based on the trigger through a one-shot RS transmission, and the result value is reported once.

[0125] Figure 15 is a semi-persistent reporting method, which is an intermediate method between the aperiodic CSI reporting method and the periodic reporting method, and the UE that receives the configuration of CSI-related RS resources and reporting methods through RRC messages performs CSI reporting periodically until it receives a deactivation message only when it is activated by MAC CE. If the UE receives a CSI deactivation message via MAC CE, the CSI reporting stops.

[0126] To mitigate the latency in such beam search (measurement) and UE power consumption, the application of AI / ML models is considered. The radio interface can utilize a function to enhance, which enables improved support for AI / ML-based algorithms to reduce complexity / overhead.

[0127] Key use cases of applications applying AI / ML models include the following.

[0128] o CSI band enhancement, such as reducing overhead, enhanced accuracy, and prediction.

[0129] o Beam management, such as time-domain and / or spatial-domain prediction, to reduce overhead and standby time, and enhance beam selection accuracy.

[0130] o Enhanced positioning accuracy in various scenarios, including scenarios with poor non-line-of-sight (NLOS) conditions.

[0131] In this regard, the following terms can be defined for wireless communication applying AI / ML.

[0132] Data collection refers to the process of a network node, management entity, or UE collecting data for AI / ML model training, data analysis, and inference.

[0133] An AI / ML model refers to a data-driven algorithm that produces an output from an input by applying AI / ML techniques. AI / ML model training refers to the process of learning the input-output relationship from data and generating an AI / ML model for inference. AI / ML model inference refers to the process of using a trained AI / ML model to generate an output from an input. AI / ML model validation refers to a sub-process of training that evaluates the quality of an AI / ML model using a dataset different from the one used for model training. AI / ML model testing refers to a sub-process of training that evaluates the performance of the final AI / ML model using a dataset different from the ones used for model training and validation. Different from AI / ML model validation, this testing does not involve any subsequent model adjustment.

[0134] A UE-side AI / ML model refers to an AI / ML model in which the inference is performed entirely by the UE. A network-side AI / ML model refers to an AI / ML model in which the inference is performed entirely on the network. A unilateral AI / ML model refers to a UE-side AI / ML model or a network-side AI / ML model. A bilateral AI / ML model refers to a paired AI / ML model that performs joint inference. Here, joint inference refers to AI / ML inference jointly performed by the UE and the network. In other words, the initial partial inference is performed by the UE, and the remaining inference is performed by the gNB, or vice versa.

[0135] AI / ML model transmission refers to sending an AI / ML model over a wireless interface as parameters of a known model structure at the receiving side or a new model with these parameters. The transmission may involve the complete model or a partial model. Model download refers to transmitting the model from the network to the UE. Model upload refers to transmitting the model from the UE to the network.

[0136] Federated learning (or federated training) refers to a machine learning technique in which an AI / ML model is trained across multiple distributed edge nodes (such as UEs, gNBs), and each node performs local model training using its own local data samples. This technique involves multiple rounds of model interaction without exchanging local data samples. Offline field data refers to data collected in the field and used for offline AI / ML model training. Online field data refers to data collected in the field and used for real-time or online AI / ML model training.

[0137] Model monitoring refers to the process of monitoring (e.g., evaluating) the inference performance of an AI / ML model.

[0138] Supervised learning refers to the process of training a model using inputs paired with corresponding labels. Unsupervised learning refers to the process of training a model without labeled data. Semi-supervised learning refers to the process of training a model using a combination of labeled and unlabeled data. Reinforcement learning (RL) refers to the process of training an AI / ML model in an interactive environment using inputs (i.e., states) and feedback signals (i.e., rewards) based on the output of the model (i.e., actions).

[0139] Model activation refers to activating (e.g., enabling) an AI / ML model to perform a specific function. Model deactivation refers to deactivating (e.g., disabling) an AI / ML model from performing a specific function. Model switching refers to deactivating the currently active AI / ML model and activating another AI / ML model for a specific function.

[0140] When applying an AI / ML model, the following network UE cooperation levels are considered.

[0141] 1. Level x: No cooperation.

[0142] 2. Level y: Signaling-based cooperation, no model transfer.

[0143] 3. Level z: Signaling-based cooperation with model transfer.

[0144] BM-Case1 and BM-Case2 are defined to support the feature description and basic performance evaluation of AI / ML-based beam management as follows.

[0145] BM-Case1: Based on the measurement results of beam set B, perform downlink (DL) beam prediction on the beam set (set A) in the spatial domain.

[0146] BM-Case2: Based on the past measurement results of beam set B, perform downlink DL beam prediction on beam set A in the time domain.

[0147] In this case, for both BM-Case1 and BM-Case2, beam sets A and B can be in the same frequency range.

[0148] Sets A and B can be configured differently; for example, set B can be a subset of set A, or set A can include narrower beams while set B includes wider beams. Here, set A can be defined for DL beam prediction and set B can be defined for DL beam measurement.

[0149] Hereinafter, a method of using artificial intelligence and machine learning to monitor a model in beam management is described with reference to the relevant drawings.

[0150] Figure 16FIG. 1600 is a flowchart showing a process in which a UE performs model monitoring in beam management using artificial intelligence and machine learning according to an embodiment.

[0151] Referring Figure 16 , the UE may receive, based on a reference signal RS configured for the UE, second reference signal resource set configuration information regarding a reference signal resource for monitoring an AI / ML model (e.g., configuration information of a second reference signal resource set related to the reference signal used therefor) (S1610).

[0152] For example, when a reference signal such as CSI-RS is configured for the UE, an entire reference signal resource set (e.g., set A) for the corresponding reference signal may be configured. In this case, the entire reference signal resource set (e.g., set A) may represent a set of reference signal resources configured for the UE using a conventional method. In addition, each reference signal resource may be associated with a corresponding beam. Although CSI-RS is used as an example of a reference signal, the concept is not limited to a specific signal as long as the technical spirit of the present disclosure can be similarly applied.

[0153] According to an embodiment, for a reference signal that can be separately transmitted in a reference signal resource in an entire reference signal resource set (set A), the UE may obtain a predicted value inferred by an AI / ML model instead of directly measuring signal strength or signal quality. In this case, a measured value of a reference signal measured based on first reference signal resource set configuration information regarding the reference signal may be used as an input to infer a predicted value of a reference signal that can be separately transmitted in a reference signal resource included in the entire reference signal resource set, and the first reference signal resource set configuration information is configured based on the entire reference signal resource set. (For example, in this case, a measured value obtained from a reference signal configured based on a first reference signal set (set (B)) may be used as an input to infer a predicted value of a reference signal in set A, and the set (B) is derived from the complete set A).

[0154] To this end, in addition to the second reference signal resource set configuration information, the UE may also receive the first reference signal resource set configuration information through higher layer signaling. The UE may measure the signal strength or signal quality of a reference signal transmitted through a reference signal resource in a first reference signal resource set (set B) based on the first reference signal resource set configuration information.

[0155] In this case, as Figure 18 shown, the first reference signal resource set (set B) may be configured to include a subset of some of the entire reference signal resource set (set A). Alternatively, as Figure 19As shown, the first reference signal resource set (Set B) can be configured as a separate new reference signal resource different from the entire reference signal resource set (Set A). In this case, the number of reference signal resources constituting the first reference signal resource set (Set B) is set to be less than the number of reference signal resources included in the entire reference signal resource set (Set A).

[0156] The UE can use the measured values of the reference signals measured based on the configuration information of the first reference signal resource set (Set B) as the input to the AI / ML model (e.g., by inputting the measured values from the first reference signal resource set (Set B) into the AI / ML model) to obtain the predicted values of the reference signals that can be respectively transmitted among the reference signal resources included in the entire reference signal resource set (Set A). In this case, if the predicted values of the entire reference signal resource set (Set A) can be inferred by using the measured values of the first reference signal resource set (Set B) as the input, the AI / ML model is not limited to a specific model and can include various known or later-known models.

[0157] The UE can determine (e.g., select) N reference signal resources (Top-N) in order (e.g., in descending order of the predicted values among the resources in Set A) starting from the reference signal resource where the highest predicted value is inferred among the reference signal resources included in the entire reference signal resource set (Set A). Here, N can be preset as a natural number. However, this is merely an example, and the method for determining the N reference signal resources can vary as needed.

[0158] If the UE reports N reference signal resources (Top-N), the base station can determine the best beam to be used in the subsequent communication with the UE by referring to the reported information. The base station can perform data transmission / reception with the UE using the determined beam.

[0159] In this case, the second reference signal resource set (Set C) configuration information (e.g., the configuration information of the second reference signal resource set (Set C)) can be configured to monitor the AI / ML model for inferring the predicted values of the entire reference signal resource set (Set A). According to an embodiment, the second reference signal resource set configuration information can be configured based on at least one reference signal resource selected according to the predicted values inferred by the AI / ML model for the reference signals that can be respectively transmitted among the reference signal resources included in the entire reference signal resource set.

[0160] According to an embodiment, if the above-described N reference signal resources (Top-N) are selected, the second reference signal resource set (Set C) may be composed of N reference resources. In other words, the beam corresponding to the second reference signal resource set (Set C) may be composed of the top-N beams reported to the base station among the beams belonging to the entire reference signal resource set (Set A). The base station may implicitly map the beams for top-N received from the UE to the reference signal resources previously allocated to the UE through the second reference signal resource set (Set C) configuration information.

[0161] Therefore, whenever at least one reference signal resource selected according to the predicted values of the reference signals that can be respectively transmitted in the reference signal resources included in the entire reference signal resource set is reported, the second reference signal resource set configuration information may be configured. In other words, whenever the Top-N reporting is performed, the second reference signal resource set (Set C) may be dynamically configured between the UE and the base station.

[0162] In addition, in the second reference signal resource set configuration information, the time domain resource information may be configured regarding the time when at least one reference signal resource selected according to the predicted values of the reference signals that can be respectively transmitted in the reference signal resources included in the entire reference signal resource set is reported, or the transmission time of the first reference signal resource set.

[0163] In other words, the base station may allocate at least the first reference signal resource set (Set B) and the second reference signal resource set (Set C) to each UE, and configure the second reference signal resource set (Set C) to be transmitted in time correlation with the first reference signal resource set (Set B). For example, the second reference signal resource set (Set C) may be configured to be transmitted for x time slots starting from the Top-N CSI reporting time inferred from the measurement values of the first reference signal resource set (Set B). Alternatively, the second reference signal resource set (Set C) may be configured to be transmitted for y time slots after the transmission time of the reference signal through the first reference signal resource set (Set B).

[0164] Return reference Figure 16 , the UE may measure the signal strength or signal quality of the reference signal based on the second reference signal resource set configuration information (S1620), compare the measured value of the reference signal with the predicted value of the reference signal inferred by the AI / ML model, and report the performance result of the AI / ML model (S1630).

[0165] The UE can measure the signal strength or signal quality of the reference signal transmitted through the second reference signal resource set (set C) based on the configuration information of the second reference signal resource set. In this case, the UE can expect the corresponding reference signals to be mapped and transmitted beam-by-beam / implicitly for the Top-N beams reported by the UE. In other words, the RI corresponding to the reference signal resources belonging to the second reference signal resource set (set C) can be mapped in a pre-agreed order, such as in ascending / descending order of beam id for the reported Top-N or resource indicator (RI), or in ascending / descending order for the reported predicted values.

[0166] The UE can determine the accuracy of the model by comparing the measured values of the reference signal with the predicted values of the reference signal inferred by the AI / ML model. In other words, the UE can determine the accuracy of the model by comparing the measured values of each beam mapped to the second reference signal resource set (set C) with the predicted values of each beam corresponding to the Top-N derived by the AI / ML model.

[0167] The UE can report the performance evaluation results to the base station, such as the accuracy, update, or reselection of the AI / ML model. If it is found that the model accuracy is low based on the performance evaluation results, the UE can request a new CSI resource setting (e.g., a new CSI resource configuration) by requesting / indicating the base station to deactivate, replace the AI / ML model, or fallback from the current AI / ML model.

[0168] According to an embodiment, the reporting of the performance evaluation results can be sent or omitted depending on the model monitoring results. For example, if the accuracy of the model reaches or exceeds a predetermined value, the reporting can be omitted. Alternatively, the reporting of the performance evaluation results can be configured to occur consistently regardless of the model monitoring results.

[0169] According to the embodiments described above, artificial intelligence and machine learning can be used to monitor the model in beam management. In addition, the AI / ML model monitoring method and device according to the embodiments can minimize the disconnection from the cell through continuous model monitoring and reduce the burden of additional beam management in AI / ML model monitoring.

[0170] Figure 17 is a flowchart showing a process 1700 in which a base station performs model monitoring in beam management using artificial intelligence and machine learning according to an embodiment. For the sake of brevity, the description related to Figure 16 can be omitted here, and the omitted content can be similarly applied to the base station as long as it does not conflict with the technical spirit of the present invention.

[0171] Refer to Figure 17, the base station may send second reference signal resource set configuration information (configuration information of the second reference signal resource set) regarding the reference signal for monitoring the AI / ML model, which is related to the reference signal RS configured for the UE (S1710).

[0172] For example, when a reference signal such as CSI-RS is configured for the UE, the entire reference signal resource set (Set A) for the reference signal may be configured. In this case, the entire reference signal resource set (Set A) may refer to a set of reference signal resources configured for the UE using conventional methods. In addition, each reference signal resource may be configured to correspond to an independent beam.

[0173] According to an embodiment, for the reference signals that can be respectively transmitted in the reference signal resources in the entire reference signal resource set (Set A), the UE may obtain the predicted values inferred by the AI / ML model instead of the directly measured signal strength or signal quality values. In this case, the measured values of the reference signals measured based on the first reference signal resource set (Set B) configuration information regarding the reference signal may be used as inputs to infer the predicted values of the reference signals that can be respectively transmitted in the reference signal resources included in the entire reference signal resource set, and the first reference signal resource set configuration information is configured based on the entire reference signal resource set.

[0174] For this purpose, in addition to the second reference signal resource set configuration information, the base station may also send the first reference signal resource set configuration information through higher layer signaling. The UE may measure the signal strength or signal quality of the reference signals transmitted through the reference signal resources in the first reference signal resource set (Set B) based on the first reference signal resource set configuration information.

[0175] In this case, as Figure 18 shown, the first reference signal resource set (Set B) may be configured to include a subset of some of the entire reference signal resource set (Set A). Alternatively, as Figure 19 shown, the first reference signal resource set (Set B) may be configured as a separate new reference signal resource different from the entire reference signal resource set (Set A). In this case, the number of reference signal resources constituting the first reference signal resource set (Set B) is set to be less than the number of reference signal resources included in the entire reference signal resource set (Set A).

[0176] The UE may use the measured values of the reference signals measured based on the first reference signal resource set (Set B) configuration information as inputs to the AI / ML model to obtain the predicted values for the reference signals that can be respectively transmitted in the reference signal resources included in the entire reference signal resource set (Set A).

[0177] The UE may determine N reference signal resources (Top-N) in order starting from the reference signal resource where the highest predicted value is inferred among the reference signal resources included in the entire reference signal resource set (Set A). Here, N may be preset as a natural number. However, this is merely an example, and the method for determining the N reference signal resources may be configured differently as needed.

[0178] If the UE reports N reference signal resources (Top-N), the base station may determine the best beam to be used in subsequent communication with the UE by referring to the reported information. The base station may perform data transmission / reception with the UE using the determined beam.

[0179] In this case, the second reference signal resource set (Set C) configuration information may be configured to monitor the AI / ML model for inferring the predicted value for the entire reference signal resource set (Set A). According to an embodiment, the second reference signal resource set configuration information may be configured based on at least one reference signal resource selected based on the predicted value inferred by the AI / ML model for the reference signals that can be respectively transmitted among the reference signal resources included in the entire reference signal resource set.

[0180] According to an example, if the above-described N reference signal resources (Top-N) are selected, the second reference signal resource set (Set C) may consist of N reference resources. In other words, the beam corresponding to the second reference signal resource set (Set C) may be composed of the top-N beams reported to the base station among the beams belonging to the entire reference signal resource set (Set A). The base station may implicitly map the beam for top-N received from the UE to the reference signal resources previously allocated to the UE through the second reference signal resource set (Set C) configuration information.

[0181] Therefore, the second reference signal resource set configuration information may be configured whenever at least one reference signal resource selected based on the predicted value for the reference signals that can be respectively transmitted among the reference signal resources included in the entire reference signal resource set is reported. In other words, the second reference signal resource set (Set C) may be dynamically configured between the UE and the base station whenever the Top-N report is performed.

[0182] In addition, in the second reference signal resource set configuration information, the time domain resource information may be configured regarding the time when at least one reference signal resource selected based on the predicted value for the reference signals that can be respectively transmitted among the reference signal resources included in the entire reference signal resource set is reported, or the transmission time of the first reference signal resource set.

[0183] In other words, the base station can allocate at least a first reference signal resource set (set B) and a second reference signal resource set (set C) for each UE, and configure the second reference signal resource set (set C) to be transmitted in temporal association with the first reference signal resource set (set B). For example, the second reference signal resource set (set C) can be configured to be transmitted for x time slots starting from the Top-N CSI reporting time inferred from the measurement values of the first reference signal resource set (set B). Alternatively, the second reference signal resource set (set C) can be configured to be transmitted for y time slots after the transmission time of the reference signal through the first reference signal resource set (set B).

[0184] Return reference Figure 17 , the base station can transmit a reference signal based on the second reference signal resource set configuration information (S1720), and receive the performance result of the AI / ML model obtained by comparing the measurement value of the reference signal with the predicted value of the reference signal inferred by the AI / ML model (S1730).

[0185] The UE can measure the signal strength or signal quality of the reference signal transmitted through the second reference signal resource set based on the second reference signal resource set (set C) configuration information. In this case, the UE can expect the corresponding reference signals to be mapped and transmitted beam-sequentially / implicitly for the Top-N beams reported for the UE. In other words, the base station can map the RI corresponding to the reference signal resources belonging to the second reference signal resource set (set C) in a pre-agreed order, such as in ascending / descending order of beam id for the reported Top-N or resource indicator (RI), or in ascending / descending order for the reported predicted values.

[0186] The UE can determine the accuracy of the model by comparing the measurement value of the reference signal with the predicted value of the reference signal inferred by the AI / ML model. In other words, the UE can determine the accuracy of the model by comparing the measurement values of each beam mapped to the second reference signal resource set (set C) with the predicted values of each beam corresponding to Top-N derived by the AI / ML model.

[0187] The base station can receive performance evaluation results from the UE, such as the accuracy, update, or reselection of the AI / ML model. When it is determined that the model accuracy is low based on the performance evaluation result, the base station can receive a request for a new CSI resource setting (e.g., a new CSI resource configuration) from the UE by requesting / indicating the base station to deactivate, replace, or fallback the AI / ML model.

[0188] According to an embodiment, a report of a performance evaluation result may be sent or omitted based on a model monitoring result. For example, when the accuracy of a model reaches or exceeds a predetermined value as a result of a performance evaluation, the report may be omitted. Alternatively, the report of the performance evaluation result may be configured to be executed each time, regardless of the model monitoring result.

[0189] According to the embodiments described above, an AI and machine learning may be used to monitor a model in beam management. In addition, the AI / ML model monitoring method and apparatus according to an embodiment may minimize disconnection from a cell through continuous model monitoring and reduce the burden of additional beam management in AI / ML model monitoring.

[0190] Hereinafter, each embodiment related to a model monitoring method in beam management using artificial intelligence and machine learning will be described in detail with reference to the accompanying drawings.

[0191] Typical beam management operations in NR cause problems such as increased system overhead and UE power consumption as the number of beams and UEs increases. In addition, for a UE in the initial cell connection phase, since the UE selects an initial beam only after measuring all available beams, there may be a delay in cell connection. To solve this problem, the use of an AI / ML model has been proposed, which predicts the overall beam intensity based on some beam measurements. However, the detailed process or method has not been defined. The present disclosure aims to propose a more specific operation of model monitoring as part of efficient AI / ML-based beam management.

[0192] Currently, discussions on AI / ML for beam management are ongoing, and it is agreed to explore spatial DL beam prediction (BM-Case1) and temporal DL beam prediction (BM-Case2) as sub-use cases. This enables predicting the beam intensity of set A based on measurements of the beams in set B. In the case of spatial DL beam prediction, as Figure 18 shown, set B may be configured as a subset of set A. Alternatively, as Figure 19 shown, set B may consist of wide beams, while set A may consist of narrow beams. For temporal DL beam prediction, in addition to the cases of Figure 18 and Figure 19 , a configuration where set A and set B are the same may also be considered. Temporal DL beam prediction involves predicting future beam information based on past measurements. It may also utilize spatial DL beam prediction to first predict the entire beam set and then reuse it for temporal prediction. Therefore, the spatial DL beam prediction cases shown in Figure 18 and Figure 19 may be used as a basic beam prediction method.

[0193] In AI / ML-based beam prediction, the product may vary depending on the location of the AI / ML model and the entity performing the training / inference. The terms "unilateral (AI / ML) model" and "bilateral (AI / ML) model" refer to whether the inference is performed by a single node alone or jointly by the network (NW) and the UE. For AI / ML-based beam management (BM), only unilateral models are currently considered. The collaboration level can be determined based on the location of the training, and beam management (BM) can be implemented in one of the following four forms.

[0194] 1. Perform AI / ML model training and inference on the network side.

[0195] 2. Perform AI / ML model training and inference on the UE side.

[0196] 3. Perform AI / ML model training on the network side and perform AI / ML model inference on the UE side.

[0197] 4. Perform AI / ML model training on the UE side and perform AI / ML model inference on the network side.

[0198] In the above cases 1 and 2, it is not necessary to transmit the model to the air interface because both training / inference run in one node. However, it may be necessary to signal the information required to operate the AI / ML model. This may correspond to the currently defined collaboration level y (only transmitting / receiving signaling, without model transmission). In the above cases 3 and 4, it is necessary to transmit the model to the air interface because training and inference run in different nodes. This may correspond to the currently defined collaboration level z.

[0199] In addition, the beam management method using AI / ML is expected to reduce the burden of measuring the entire beam by predicting the signal strength of all beams (set A) by measuring some of all beams or other beams except all beams, and reduce the overhead on the RS. However, if the predicted values of the beams in set A are inaccurate, the AI / ML method may cause frequent beam failures or continuous disconnections by selecting inaccurate beams. To solve these problems, it is very important to continuously monitor the accuracy of the predicted beams. It is necessary to compare the actual measurement values and the predicted values of the beams belonging to set A to monitor the accuracy of the beams. However, measuring the beams in set A may increase the measurement burden on the UE every time all the beams in set A are inferred.

[0200] The location where inference and training are performed may also affect the model monitoring method. In other words, the inference entity performs the monitoring, and the result value should be sent to the training entity. This is because the model monitoring result may allow model updates to be performed through new training. In addition, even when training and inference are performed on a single node, the result of model monitoring may trigger a new process as needed, making it possible to transmit feedback to the base station or UE according to the model performance.

[0201] Based on the above, the present disclosure proposes a specific method for effectively monitoring an AI / ML model when performing beam management using AI / ML.

[0202] The present disclosure proposes to define a set C (e.g., a set of beams for AI / ML model monitoring) as a means for evaluating the accuracy of the result value (e.g., the output of the inferred beam (set A)) of the AI / ML model when performing a beam management process between a base station (NW) and a UE using the AI / ML model. Here, the set C consists of one or more beams for model monitoring purposes, and the node performing the model monitoring can select / define it as a subset of set A.

[0203] In the present disclosure, set A can be defined for downlink (DL) beam prediction, and set B can be defined for DL beam measurement.

[0204] More specifically, the above-proposed set C can consist of N beams (e.g., the top-N beams in order starting from the beam with the highest RSRP) selected by any algorithm (e.g., high RSRP) among the beams of set A derived by the AI / ML model. Information about the beams constituting set C can be determined by the UE or the base station according to the model inference location. When the UE performs model inference, set C can consist of the top-N beams selected from among the beams of set A derived through model inference and used for reporting to the base station, and set C can be dynamically configured between the UE and the base station whenever the top-N is reported. At this time, the network (NW) can implicitly map the beams for the top-N received from the UE to the RS resources pre-allocated to the UE for set C, so that the base station and the UE can identify the configuration of set C with each other and perform resource mapping for measurement. The UE can determine the model accuracy by comparing the result value of the output of set C inferred by the AI / ML model with the measured value of set C actually measured. When the NW performs model inference, the beams of set C can consist of beams arbitrarily selected from among the beams of set A derived through model inference at the base station, which may depend on the base station implementation.

[0205] Here, the base station can set at least two different CSI resource sets through RRC messages to allow the UE to measure the signal strength of each beam that constitutes set B and set C. In other words, at least a first CSI resource set mapped to the beams belonging to set B and a second CSI resource set for set C can be allocated for each UE. Here, the second CSI resource set for set C can be configured to be transmitted in time correlation with the first CSI resource set for set B. More specifically, the second CSI resource set can be configured to be transmitted at a time (e.g., x time slots after the report) associated with the report for the first CSI resource set.

[0206] In addition, the signal strength of the beam measured through the second CSI resource set can be used to evaluate the model performance based on the comparison with the inferred predicted value.

[0207] Hereinafter, the operations of the UE for performing model inference according to an embodiment are described. In addition to the CSI resource set of set B for measuring the beams (set B) to be used as the input of the AI / ML model, the UE can additionally receive the RS resource information (CSI resource set for model monitoring / set C) of the beams (set C) to be used for evaluating the performance of the AI / ML model. In this case, after the beam measurement (set B) for model inference, the UE can perform an additional beam measurement (set C) to determine the accuracy of the corresponding inference related to the reporting time of set B. The UE can determine the model accuracy by comparing the predicted value and the measured value of set C. For example, the report on the measurement result of the beam of set C can send the performance evaluation result (e.g., accuracy, model update, or reselection) of the model to the base station or a fallback indication (e.g., fallback to the beam management process that does not use the AI / ML model).

[0208] According to an embodiment of the present disclosure, the detailed process can be broadly classified based on whether the model inference is performed at the network (gNB) or at the UE.

[0209] When the UE performs model inference, referring to Figure 20 , the operations of the UE and the base station are as follows.

[0210] (Operations of the UE)

[0211] 1. The UE can receive an RRC configuration message from the base station that includes information about at least two CSI resource sets.

[0212] The RRC configuration message can include the configuration information about the first CSI resource set and its reporting method and the second CSI resource set and its reporting method.

[0213] Here, the configuration information of the first CSI resource set can be configured in the same manner as in the prior art.

[0214] According to the present disclosure, a configuration method and a reporting method for a second CSI resource set associated with a first CSI resource set can be newly defined.

[0215] Time resource information regarding the second CSI resource set can be configured in association with the reporting time of the first CSI resource set.

[0216] Based on the number of RS resources (beams) in the second CSI resource set, the UE (e.g., after measuring the first CSI resource set) can use an AI / ML model to infer a predicted value for set A and identify N RS resources to be reported to the base station. For example, the time resource information of the second CSI resource set may include a value (e.g., x time slots) related to the transmission time of the first CSI resource set or the CSI reporting time of the Top-N inferred based on the measurement of the first CSI resource set. Here, N can mean Top-N corresponding to N measurement values in order from the highest measurement value.

[0217] Different from the typical reporting of beam measurement results, the reporting method of the second CSI resource set may include setting request information corresponding to the model monitoring result value. For example, information related to model accuracy, model deactivation / change, or fallback identification can be reported.

[0218] 2. The UE can measure the signal strength of one or more RSs transmitted in the first CSI resource set (set B).

[0219] 3. The UE can use the measured beam information set B as an input value of the AI / ML model to infer a predicted value for the entire beam set A.

[0220] 4. The UE can send a CSI report to the base station according to setting 1, and the CSI report includes information about the beam set C belonging to Top-N among the predicted values derived by inference.

[0221] 5. If necessary, the UE can receive information about new serving beams from the base station. (Beam indication)

[0222] 6. The UE measures the signal strength of one or more RSs of the second CSI resource set according to setting 1. The UE expects to map and transmit the corresponding RSs for the Top-N beams reported by the UE sequentially / implicitly.

[0223] Here, "sequentially / implicitly" can mean mapping to the RI corresponding to the RS belonging to the second CSI resource set in a pre-agreed order, such as in ascending / descending order for the resource indicator (RI) or beam id of Top-N, or in ascending / descending order for the reported predicted value (e.g., RSRP).

[0224] 7. The UE can determine the accuracy of the model by comparing the measured signal strength of the beam mapped to the second CSI resource set with the value of Top-N (the predicted value derived by inference).

[0225] 8. The UE can send the model performance evaluation result to the base station according to Setting 1.

[0226] Depending on the model monitoring result, this report may be sent or omitted. Alternatively, the result can be reported each time regardless of the monitoring result.

[0227] If the performance evaluation indicates low model accuracy, the UE can request / indicate the base station to deactivate, replace, or fallback the AI / ML model and request a new CSI resource.

[0228] (Operation of the base station)

[0229] 1. The base station can send an RRC configuration message to the UE, including information about at least two CSI resource sets.

[0230] The RRC configuration message can include configuration information about the first CSI resource set and its reporting method, and the second CSI resource set and its reporting method.

[0231] Here, the configuration information of the first CSI resource set can be the same as the configuration information in the prior art.

[0232] According to the present disclosure, a new method for configuring and reporting the second CSI resource set associated with the first CSI resource set can be newly defined.

[0233] The time resource information of the second CSI resource set can be configured relative to the reporting time of the first CSI resource set.

[0234] The second CSI resource set can be configured to include N RS resources to correspond to the number of beams reported by the UE to the base station. For example, the time resource information can indicate: relative to the transmission time of the first CSI resource set or the value of the Top-N CSI reporting time inferred based on the measurement of the beam mapped to the first CSI resource set (e.g., x time slots).

[0235] Different from the typical beam measurement result reporting, the reporting method of the second CSI resource set can include request information based on the model monitoring result. For example, the report may include information about model accuracy, model deactivation / change, or fallback identification.

[0236] 2. The base station can use the first CSI resource set to send the RS mapped to the beams of Set B according to Setting 1.

[0237] 3. The base station can receive a CSI report from the UE that includes information about N beams (set C).

[0238] 4. If necessary, the base station can select a beam and notify the UE of information about the selected beam.

[0239] (Beam indication)

[0240] 5. The base station can sequentially / implicitly map the beams included in the CSI report received at 3 to one or more RSs of the second CSI resource set pre-allocated at 1 and send them.

[0241] Here, "sequentially / implicitly" means mapping to the RI corresponding to the RS belonging to the second CSI resource set in a pre-agreed order, such as in ascending / descending order for the resource indicator (RI) or beam id of Top-N reported from the UE, or in ascending / descending order for the reported predicted value (e.g., RSRP).

[0242] 6. The base station can receive the model performance evaluation result from the UE.

[0243] Depending on the model monitoring result, this result can be received or omitted from the UE. Alternatively, the result can be reported each time regardless of the monitoring result.

[0244] If the performance evaluation indicates low model accuracy, the UE can request / indicate the base station to deactivate, replace, or fallback the AI / ML model, so that the corresponding information can be used as a request for a new CSI resource configuration.

[0245] Reference Figure 21 , when the base station / NW performs model inference, the operations of the base station and the UE are as follows.

[0246] (Operations of the base station)

[0247] 1. The base station can send an RRC configuration message to the UE that includes information about at least two CSI resource sets.

[0248] The RRC configuration message can include configuration information about the first CSI resource set and its reporting method and the second CSI resource set and its reporting method.

[0249] Here, the configuration information of the first CSI resource set can be the same as the configuration information in the prior art.

[0250] According to the present disclosure, a new method for configuring and reporting the second CSI resource set associated with the first CSI resource set can be defined.

[0251] The time resource information of the second CSI resource set can be configured relative to the reporting time of the first CSI resource set. For example, the time resource information may indicate: a value of the CSI reporting time relative to the transmission time of the first CSI resource set or the result value of the measurement of the beam mapped to the first CSI resource set (e.g., x time slots).

[0252] The reporting time information of the second CSI resource set can be configured relative to the reporting time of the first CSI resource set. For example, the time resource information may indicate: a value of the CSI reporting time relative to the transmission time of the first CSI resource set or the measurement result of the beam mapped to the first CSI resource set (e.g., y time slots).

[0253] 2. The base station can use the first CSI resource set to transmit the RS mapped to the beams of set B according to setting 1.

[0254] 3. The base station can receive a CSI report from the UE, and the CSI report includes the measurement values of the beam set B mapped to the first CSI resource set.

[0255] 4. The base station can use the beam information included in the CSI report received in 3 as the input value of the AI / ML model to infer the predicted values for the entire beam set A.

[0256] 5. If necessary, the base station can select a best beam among the predicted values for the inferred beam (set A) and notify the UE of the information about the selected beam.

[0257] 6. The base station can arbitrarily select N beams (set C) from set A that do not belong to set B, map them to the RS of the second CSI resource set, and send them to the UE according to the setting.

[0258] 7. The base station can receive the measurement results of set C from the UE at the time according to setting 1.

[0259] 8. The base station can compare the measurement results received in 7 with the predicted result values inferred in 5 and perform model monitoring.

[0260] 9. The base station can indicate a new setting to the UE according to the model performance evaluation result.

[0261] This can be sent or omitted depending on the model monitoring result.

[0262] If the performance evaluation result indicates low model accuracy, the base station can indicate a new CSI resource configuration to the UE in response to the deactivation, replacement, or fallback of the AI / ML model.

[0263] (UE's operations)

[0264] 1. The UE can receive an RRC configuration message from the base station, which includes information about at least two CSI resource sets.

[0265] It can include configuration information about the first CSI resource set and its reporting method, as well as about the second CSI resource set and its reporting method.

[0266] Here, the configuration information of the first CSI resource set can be the same as the configuration information in the prior art.

[0267] According to the present disclosure, a new method can be defined for configuring and reporting the second CSI resource set relative to the first CSI resource set.

[0268] The time resource information of the second CSI resource set can also be configured relative to the reporting time of the first CSI resource set. For example, the time resource information can indicate: a value (e.g., x time slots) of the CSI reporting time relative to the transmission time of the first CSI resource set or the result value of the measurement of the beam mapped to the first CSI resource set.

[0269] The reporting time information of the second CSI resource set can be configured relative to the reporting time of the first CSI resource set. For example, the time resource information indicates: a value (e.g., y time slots) of the CSI reporting time relative to the transmission time of the first CSI resource set or the measurement result of the beam mapped to the first CSI resource set.

[0270] 2. The UE can measure the signal strength of the RS to which the beams of set B are mapped according to setting 1 using the first CSI resource set.

[0271] 3. The UE can send / report a CSI report to the base station, which includes the measurement values of the beam set B mapped to the first CSI resource set.

[0272] 4. If necessary, the UE can receive information about the selected beam from the base station. (Beam indication)

[0273] 5. The UE can measure the signal strength of the RS of the second CSI resource set (set C) transmitted in association with the transmission time of 3 according to setting 1.

[0274] 6. The UE can report / send the measurement result of 5 to the base station at the time according to setting 1.

[0275] 7. According to the model performance evaluation result, the UE can receive new settings from the base station.

[0276] This can be sent or omitted depending on the model monitoring result at the base station.

[0277] If the performance evaluation result indicates low model accuracy, the base station may, after deactivating, replacing, or reverting the AI / ML model, indicate a new CSI resource configuration to the UE.

[0278] As described above, when performing AI / ML-based beam management, due to frequent beam switching and disconnections, low model accuracy may lead to a degradation in communication quality. To prevent frequent disconnections due to incorrect beam predictions, continuous model monitoring is required during AI / ML model inference. However, typical model monitoring compares predicted beam results derived from measurements of all candidate beams, which incurs a large amount of measurement overhead during each inference. According to the present disclosure, a new beam set (set C) is implicitly defined to minimize the configuration signaling for model monitoring, enabling the UE to evaluate model accuracy using a minimum number of beam measurements per inference. This approach reduces the burden of additional beam measurements and helps minimize disconnections from the battery through continuous model monitoring.

[0279] Hereinafter, the hardware and software configurations of the UE and the base station that can execute all or some of the embodiments described above in conjunction with Figures 1 to 21 the description will be described with reference to the accompanying drawings. For the sake of brevity, the above description may be omitted, and in such a case, as long as it does not conflict with the technical spirit of the present invention, the omitted content may be similar to the following description.

[0280] Figure 22 is a block diagram showing a UE 2200 according to an embodiment.

[0281] Referring to Figure 22 , the UE 2200 according to an embodiment includes a transmitter 2220, a receiver 2230, and a controller 2210 that controls the transmitter and the receiver.

[0282] The controller 2210 controls the overall operation of the UE 2200 according to a model monitoring method in beam management using artificial intelligence and machine learning required to execute the above-described embodiments.

[0283] The transmitter 2220 and the receiver 2230 are used to transmit or receive signals, messages, or data required to execute the above-described embodiments together with the base station.

[0284] The controller 2210 may receive second reference signal resource set configuration information regarding a reference signal (RS) for monitoring the AI / ML model, and the second reference signal resource set configuration information is related to the synchronization signal (SS) configured for the UE.

[0285] When configuring a reference signal such as CSI-RS for a UE, an entire reference signal resource set (set A) for the reference signal can be configured. In this case, the entire reference signal resource set (set A) can refer to a set of reference signal resources configured for the UE using a conventional method. In addition, each reference signal resource can correspond to an independent beam.

[0286] According to an embodiment, for a reference signal that can be separately transmitted in a reference signal resource in the entire reference signal resource set (set A), the controller 2210 can obtain a predicted value inferred by an AI / ML model, rather than a measured value of directly measured signal strength or signal quality. In this case, a measured value of the reference signal measured based on configuration information of a first reference signal resource set (set B) regarding the reference signal can be used as an input to infer a predicted value of the reference signal that can be separately transmitted in the reference signal resources included in the entire reference signal resource set, and the configuration information of the first reference signal resource set is configured based on the entire reference signal resource set.

[0287] To this end, in addition to the second reference signal resource set configuration information, the controller 2210 can also receive the first reference signal resource set configuration information through higher layer signaling. The controller 2210 can measure the signal strength or signal quality of the reference signal transmitted through the reference signal resources in the first reference signal resource set (set B) based on the first reference signal resource set configuration information.

[0288] The controller 2210 can use the measured value of the reference signal measured based on the configuration information of the first reference signal resource set (set B) as an input to the AI / ML model to obtain a predicted value for the reference signal that can be separately transmitted in the reference signal resources included in the entire reference signal resource set (set A).

[0289] The controller 2210 can sequentially determine N reference signal resources (Top-N) starting from the reference signal resource where the highest predicted value is inferred among the reference signal resources included in the entire reference signal resource set (set A). Here, N can be preset as a natural number. However, this is merely an example, and the method for determining the N reference signal resources can be configured differently as needed.

[0290] If the controller 2210 reports N reference signal resources (Top-N), the base station can determine the best beam to be used in subsequent communication with the UE by referring to the reported information. The base station can perform data transmission / reception with the UE using the determined beam.

[0291] In this case, the configuration information of the second reference signal resource set (set C) can be configured to monitor an AI / ML model for inferring predicted values for the entire reference signal resource set (set A). According to an embodiment, the second reference signal resource set configuration information can be configured based on at least one reference signal resource selected according to predicted values inferred by the AI / ML model for reference signals that can be respectively transmitted among the reference signal resources included in the entire reference signal resource set.

[0292] According to an embodiment, if the above-described N reference signal resources (Top-N) are selected, the second reference signal resource set (set C) can be composed of N reference resources. In other words, the beam corresponding to the second reference signal resource set (set C) can be composed of the top-N beams reported to the base station among the beams belonging to the entire reference signal resource set (set A). The base station can implicitly map the beams for the top-N received from the UE to the reference signal resources previously allocated to the UE through the second reference signal resource set (set C) configuration information.

[0293] Therefore, whenever at least one reference signal resource selected according to predicted values for reference signals that can be respectively transmitted among the reference signal resources included in the entire reference signal resource set is reported, the second reference signal resource set configuration information can be configured. In other words, whenever the Top-N report is performed, the second reference signal resource set (set C) can be dynamically configured between the UE and the base station.

[0294] In addition, in the second reference signal resource set configuration information, time domain resource information can be configured regarding the time of reporting at least one reference signal resource selected according to predicted values for reference signals that can be respectively transmitted among the reference signal resources included in the entire reference signal resource set, or the transmission time of the first reference signal resource set.

[0295] In other words, the base station can allocate at least the first reference signal resource set (set B) and the second reference signal resource set (set C) to each UE, and configure the second reference signal resource set (set C) to be transmitted in time association with the first reference signal resource set (set B). For example, the second reference signal resource set (set C) can be configured to be transmitted for x time slots starting from the Top-N CSI reporting time inferred from the measurement values of the first reference signal resource set (set B). Alternatively, the second reference signal resource set (set C) can be configured to be transmitted for y time slots after the transmission time of the reference signal through the first reference signal resource set (set B).

[0296] The controller 2210 may measure the signal strength or signal quality of a reference signal based on the second reference signal resource configuration information. In addition, the controller 2210 may compare the measured value of the reference signal with the predicted value of the reference signal inferred by the AI / ML model and report the performance result of the AI / ML model.

[0297] The controller 2210 may measure the signal strength or signal quality of the reference signal transmitted through the second reference signal resource set (set C) based on the second reference signal resource set (set C) configuration information. In this case, the controller 2210 may expect to map and transmit the corresponding reference signal beam-sequentially / implicitly for the Top-N reported by the UE.

[0298] The controller 2210 may determine the accuracy of the model by comparing the measured value of the reference signal with the predicted value of the reference signal inferred by the AI / ML model. In other words, the controller 2210 may determine the accuracy of the model by comparing the measured value of each beam mapped to the second reference signal resource set (set C) with the predicted value of each beam corresponding to the Top-N derived by the AI / ML model.

[0299] The controller 2210 may report performance evaluation results to the base station, such as the accuracy, update, or reselection of the AI / ML model. When the model accuracy is low according to the performance evaluation result, the controller 2210 may request a new CSI resource setting by requesting / indicating the base station to deactivate, replace, or fallback the AI / ML model.

[0300] According to an embodiment, the reporting of the performance evaluation result may be sent or omitted according to the model monitoring result. For example, when the accuracy of the model reaches or exceeds a predetermined value as a result of the performance evaluation, the reporting may be omitted. Alternatively, the reporting of the performance evaluation result may be configured to be executed each time, regardless of the model monitoring result.

[0301] According to the embodiments described above, artificial intelligence and machine learning may be used to monitor the model in beam management. In addition, the AI / ML model monitoring method and device according to an embodiment may minimize the disconnection from the cell through continuous model monitoring and reduce the burden of additional beam management in AI / ML model monitoring.

[0302] Figure 23 is a block diagram showing a base station 2300 according to an embodiment.

[0303] Refer to Figure 23 , the base station 2300 according to an embodiment includes a transmitter 2320, a receiver 2330, and a controller 2310 that controls the transmitter and the receiver.

[0304] The controller 2310 controls the overall operation of the UE 2300 and the overall operation of the repeater according to the model monitoring method in beam management using artificial intelligence and machine learning required to implement the above-described embodiments.

[0305] The transmitter 2320 and the receiver 2330 are used to transmit or receive signals, messages, or data required to implement the above-described embodiments together with the UE.

[0306] The controller 2310 may send second reference signal resource set configuration information regarding a reference signal (RS) for monitoring the AI / ML model, and the second reference signal resource set configuration information is related to the reference signal (SS) configured for the UE.

[0307] When a reference signal such as CSI-RS is configured for the UE, the entire reference signal resource set (set A) for the reference signal may be configured. In this case, the entire reference signal resource set (set A) may refer to a set of reference signal resources that can be configured for the UE using conventional methods. In addition, each reference signal resource may be configured to correspond to a corresponding beam.

[0308] According to an embodiment, for the reference signals that can be respectively transmitted in the reference signal resources in the entire reference signal resource set (set A), the UE may obtain predicted values inferred by the AI / ML model instead of directly measuring the signal strength or signal quality measurement values. In this case, the measurement values of the reference signals measured based on the first reference signal resource set (set B) configuration information regarding the reference signal may be used as inputs to infer the predicted values of the reference signals that can be respectively transmitted in the reference signal resources included in the entire reference signal resource set, and the first reference signal resource set configuration information is based on the entire reference signal resource set configuration.

[0309] To this end, in addition to the second reference signal resource set configuration information, the controller 2310 may also send the first reference signal resource set configuration information through higher layer signaling. The UE may measure the signal strength or signal quality of the reference signals transmitted through the reference signal resources in the first reference signal resource set (set B) based on the first reference signal resource set configuration information.

[0310] The UE may use the measurement values of the reference signals measured based on the first reference signal resource set (set B) configuration information as inputs to the AI / ML model to obtain predicted values for the reference signals that can be respectively transmitted in the reference signal resources included in the entire reference signal resource set (set A).

[0311] The UE may determine N reference signal resources (Top-N) in order starting from the reference signal resource where the highest predicted value is inferred among the reference signal resources included in the entire reference signal resource set (set A). Here, N may be preset as a natural number. However, this is an example, and the method for determining the N reference signal resources may be set differently as needed.

[0312] If the UE reports N reference signal resources (Top-N), the controller 2310 may determine the best beam to be used in subsequent communication with the UE by referring to the reported information. The controller 2310 may perform data transmission / reception with the UE using the determined beam.

[0313] In this case, the second reference signal resource set (set C) configuration information may be configured to monitor the AI / ML model for inferring the predicted values for the entire reference signal resource set (set A). According to an embodiment, the second reference signal resource set configuration information may be configured based on at least one reference signal resource selected according to the predicted values inferred by the AI / ML model for the reference signals that can be respectively transmitted among the reference signal resources included in the entire reference signal resource set.

[0314] According to an embodiment, if the above-described N reference signal resources (Top-N) are selected, the second reference signal resource set (set C) may be composed of N reference resources. In other words, the beam corresponding to the second reference signal resource set (set C) may be composed of the top-N beams reported to the base station among the beams belonging to the entire reference signal resource set (set A). The controller 2310 may implicitly map the beam for top-N received from the UE to the reference signal resources previously allocated to the UE through the second reference signal resource set (set C) configuration information.

[0315] Therefore, whenever at least one reference signal resource selected according to the predicted values for the reference signals that can be respectively transmitted among the reference signal resources included in the entire reference signal resource set is reported, the second reference signal resource set configuration information may be configured. In other words, whenever the Top-N report is performed, the second reference signal resource set (set C) may be dynamically configured between the UE and the base station.

[0316] In addition, the second reference signal resource set configuration information may include time domain resource information configured in relation to: the reporting time of at least one reference signal resource selected based on the predicted values of the reference signals of the reference signal resources in the entire reference signal resource set, or the transmission time of the first reference signal resource set.

[0317] In other words, the controller 2310 can allocate at least a first reference signal resource set (set B) and a second reference signal resource set (set C) for each UE, and configure the second reference signal resource set (set C) to be transmitted in temporal association with the first reference signal resource set (set B). For example, the second reference signal resource set (set C) can be configured to be transmitted for x time slots starting from the Top-N CSI reporting time inferred from the measurement values of the first reference signal resource set (set B). Alternatively, the second reference signal resource set (set C) can be configured to be transmitted for y time slots after the transmission time of the reference signal through the first reference signal resource set (set B).

[0318] The controller 2310 can transmit a reference signal based on the second reference signal resource set configuration information and receive the performance result of the AI / ML model obtained by comparing the measurement value of the reference signal with the predicted value of the reference signal inferred by the AI / ML model.

[0319] The UE can measure the signal strength or signal quality of the reference signal transmitted through the second reference signal resource set based on the second reference signal resource set (set C) configuration information. In this case, the UE can expect the corresponding reference signals to be mapped and transmitted beam-sequentially / implicitly for the Top-N beams reported for the UE. In other words, the controller 2310 can map the RI corresponding to the reference signal resources belonging to the second reference signal resource set (set C) in a predetermined order, such as in ascending / descending order of beam id for the reported Top-N or resource indicator (RI), or in ascending / descending order for the reported predicted values.

[0320] The UE can determine the accuracy of the model by comparing the measurement value of the reference signal with the predicted value of the reference signal inferred by the AI / ML model. In other words, the UE can determine the accuracy of the model by comparing the measurement values of each beam mapped to the second reference signal resource set (set C) with the predicted values of each beam corresponding to the Top-N derived by the AI / ML model.

[0321] The controller 2310 can receive performance evaluation results from the UE, such as the accuracy, update, or reselection of the AI / ML model. When the model accuracy is low according to the performance evaluation result, the controller 2310 can receive a request for a new CSI resource setting from the UE by requesting / indicating the base station to deactivate, replace, or fallback the AI / ML model.

[0322] According to an embodiment, a report of a performance evaluation result may be sent or omitted based on a model monitoring result. For example, when the accuracy of a model reaches or exceeds a predetermined value as a result of a performance evaluation, the report may be omitted. Alternatively, the report of the performance evaluation result may be configured to be executed each time, regardless of the model monitoring result.

[0323] According to the embodiment described above, an artificial intelligence and machine learning may be used to monitor a model in beam management. In addition, the AI / ML model monitoring method and apparatus according to an embodiment may minimize disconnection from a cell through continuous model monitoring and reduce the burden of additional beam management in AI / ML model monitoring.

[0324] The embodiments described above may be supported by standard documents disclosed in at least one of radio access systems such as IEEE 802, 3GPP, and 3GPP2. That is, steps, configurations, and parts not described in the present embodiments may be supported by the standard documents mentioned above for clarifying the technical concept of the present disclosure. In addition, all terms disclosed herein may be described by the standard documents set forth above.

[0325] The embodiments described above may be implemented by any one of various components. For example, the present embodiment may be implemented as hardware, firmware, software, or a combination thereof.

[0326] In the case of being implemented by hardware, the method according to the present embodiment may be implemented as at least one of an application specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a processor, a controller, a microcontroller, or a microprocessor.

[0327] In the case of being implemented by firmware or software, the method according to the present embodiment may be implemented in the form of a device, a process, or a function for performing the functions or operations described above. The software code may be stored in a memory unit and may be driven by a processor. The memory unit may be provided inside or outside the processor and may exchange data with the processor through various well-known components.

[0328] In addition, terms such as "system", "processor", "controller", "component", "module", "interface", "model", and "unit" generally may refer to computer-related entities, combinations of hardware and software, software, or software in operation. For example, the components described above may be, but are not limited to, processes driven by a processor, processors, controllers, control processors, entities, execution threads, programs, and / or computers. For example, an application running in a controller or processor and the controller or processor may both be components. One or more components may be provided in a process and / or an execution thread, and these components may be provided in a single device (e.g., a system, a computing device, etc.), or may be distributed over two or more devices.

[0329] The above-described embodiments of the present disclosure have been described for illustrative purposes only, and those of ordinary skill in the art will recognize that various modifications and changes can be made thereto without departing from the scope and spirit of the present disclosure. In addition, the embodiments of the present disclosure are not intended to be limiting, but rather to illustrate the technical idea of the present disclosure, and thus the scope of the technical idea of the present disclosure is not limited by these embodiments. The scope of the present disclosure should be construed based on the appended claims, such that all technical ideas within the scope equivalent to the claims belong to the present disclosure.

[0330] Cross-reference to related applications

[0331] This application claims priority under 35 U.S.C. 119(a) from Korean Patent Application No. 10-2022-0129855, filed on October 11, 2022, and No. 10-2023-0134641, filed on October 10, 2023, in the Korean Intellectual Property Office, the entire contents of which are incorporated herein by reference in their entirety.

Claims

1. A method for user equipment (UE) to perform model monitoring in beam management using artificial intelligence and machine learning (AI / ML), the method include: receiving second reference signal resource set configuration information about a reference signal (RS) for monitoring an AI / ML model, wherein the second reference signal resource set configuration information is related to a reference signal (RS) configured for a UE; Measuring the signal strength or signal quality of the reference signal based on the second reference signal resource configuration information; and Compare the measured value of the reference signal with the predicted value of the reference signal inferred by the AI / ML model, and report the performance results of the AI / ML model.

2. The method according to claim 1, in, The second reference signal resource set configuration information is configured based on at least one reference signal resource, and the at least one reference signal resource is selected according to a predicted value inferred by the AI / ML model for a reference signal that can be transmitted via reference signal resources included in the entire reference signal resource set of the reference signal.

3. The method according to claim 2, in, A predicted value of a reference signal that can be transmitted via reference signal resources included in the entire reference signal resource set is inferred using a measured value of the reference signal measured based on first reference signal resource set configuration information about the reference signal as input, wherein the first reference signal resource set configuration information is based on the entire reference signal resource set configuration.

4. The method according to claim 3, in, The second reference signal resource set configuration information includes time domain resource information, and the time domain resource information is configured with respect to the following items: the transmission time of the first reference signal resource set, or the time of reporting at least one reference signal resource selected based on the predicted value of the reference signal that can be respectively transmitted via the reference signal resources included in the entire reference signal resource set.

5. The method according to claim 2, in, The second reference signal resource set configuration information is configured in the following case: each time at least one reference signal resource selected according to a predicted value of a reference signal respectively transmittable via reference signal resources included in the entire reference signal resource set is reported.

6. A method for a base station to perform model monitoring in beam management using artificial intelligence and machine learning, the method include: Sending second reference signal resource set configuration information about a reference signal (RS) for monitoring an AI / ML model, wherein the second reference signal resource set configuration information is related to a reference signal (RS) configured for the UE; sending the reference signal based on the second reference signal resource configuration information; and Receive performance results of the AI / ML model obtained by comparing the measured value of the reference signal with the predicted value of the reference signal inferred by the AI / ML model.

7. The method according to claim 6, in, The second reference signal resource set configuration information is configured based on at least one reference signal resource, and the at least one reference signal resource is selected according to a predicted value inferred by the AI / ML model for reference signals that can be respectively transmitted via reference signal resources included in the entire reference signal resource set of the reference signal.

8. The method according to claim 7, in, The predicted value of a reference signal that can be transmitted respectively via the reference signal resources included in the entire reference signal resource set is inferred using the measured value of the reference signal measured based on the first reference signal resource set configuration information about the reference signal as input, and the first reference signal resource set configuration information is based on the entire reference signal resource set configuration.

9. The method according to claim 8, in, The second reference signal resource set configuration information includes time domain resource information, and the time domain resource information is configured with respect to the following items: the transmission time of the first reference signal resource set, or the time of reporting at least one reference signal resource selected based on the predicted value of the reference signal that can be respectively transmitted via the reference signal resources included in the entire reference signal resource set.

10. The method according to claim 7, in, The second reference signal resource set configuration information is configured in the following case: each time at least one reference signal resource selected according to a predicted value of a reference signal respectively transmittable via reference signal resources included in the entire reference signal resource set is reported.

11. A user equipment (UE) for performing model monitoring in beam management using artificial intelligence and machine learning (AI / ML), include: Transmitter; Receiver; as well as a controller configured to control operation of the transmitter and the receiver, wherein the controller: receiving second reference signal resource set configuration information about a reference signal (RS) for monitoring an AI / ML model, the second reference signal resource set configuration information being related to a reference signal (RS) configured for the UE; measuring the signal strength or signal quality of the reference signal based on the second reference signal resource configuration information; and The measured value of the reference signal is compared with the predicted value of the reference signal inferred by the AI / ML model to report the performance results of the AI / ML model.

12. The UE according to claim 11, in, The second reference signal resource set configuration information is configured based on at least one reference signal resource, and the at least one reference signal resource is selected according to a predicted value inferred by the AI / ML model for reference signals that can be respectively transmitted via reference signal resources included in the entire reference signal resource set of the reference signal.

13. The UE according to claim 12, in, The predicted value of a reference signal that can be transmitted respectively via the reference signal resources included in the entire reference signal resource set is inferred using the measured value of the reference signal measured based on the first reference signal resource set configuration information about the reference signal as input, and the first reference signal resource set configuration information is based on the entire reference signal resource set configuration.

14. The UE according to claim 13, in, The second reference signal resource set configuration information includes time domain resource information, and the time domain resource information is configured with respect to the following items: the transmission time of the first reference signal resource set, or the time of reporting at least one reference signal resource selected based on the predicted value of the reference signal that can be respectively transmitted via the reference signal resources included in the entire reference signal resource set.

15. The UE according to claim 12, in, The second reference signal resource set configuration information is configured in the following case, each time at least one reference signal resource selected according to a predicted value of a reference signal respectively transmittable via reference signal resources included in the entire reference signal resource set is reported.

Images