Method and apparatus for transmitting and receiving signal in wireless communication system

Abstract

A terminal according to at least one embodiment disclosed in the present specification may receive artificial intelligence / machine learning (AI / ML)-related configuration information, configure an AI / ML model for at least one transmit-receive point (TRP) based on the AI / ML-related configuration information, and receive information for managing the configured AI / ML model, in which a plurality of TRPs are supported on the basis of the terminal. The information for managing the AI / ML model individually indicates the AI / ML model to be enabled, disabled, updated, or switched for each of the plurality of TRPs.

Description

Technical Field

[0001] The present disclosure relates to a wireless communication system, and more particularly, to a method and apparatus for transmitting / receiving uplink / downlink wireless signals in a wireless communication system. Background Art

[0002] In general, wireless communication systems are developing to cover a wide range in various ways to provide communication services such as audio communication services, data communication services, etc. Wireless communication is a multiple access system that can support communication with multiple users by sharing available system resources (e.g., bandwidth, transmission power, etc.). For example, the multiple access system can be any of a code division multiple access (CDMA) system, a frequency division multiple access (FDMA) system, a time division multiple access (TDMA) system, an orthogonal frequency division multiple access (OFDMA) system, and a single carrier frequency division multiple access (SC-FDMA) system. Summary of the invention

[0003] Technical issues

[0004] An object of the present disclosure is to provide a method and apparatus for efficiently performing a wireless signal transmission / reception process.

[0005] Those skilled in the art will understand that the objectives that can be achieved using the present disclosure are not limited to those specifically described above, and the above and other objectives that can be achieved by the present disclosure will be more clearly understood from the following detailed description.

[0006] Technical Solution

[0007] In one aspect of the present disclosure, a method for operating a user equipment (UE) in a wireless communication system is provided herein. The method may include the following steps: receiving artificial intelligence / machine learning (AI / ML) related configuration information; configuring an AI / ML model for at least one transmit receive point (TRP) based on the received AI / ML related configuration information; and receiving information for managing the configured AI / ML model. Based on the UE supporting multiple TRPs, the information for managing the AI / ML model may separately indicate the activation, deactivation, update, or switching of the AI / ML model for each of the multiple TRPs.

[0008] Information used to manage AI / ML models can be received via a media access control (MAC) control element (CE).

[0009] Information for managing AI / ML models may be received via downlink control information (DCI).

[0010] AI / ML related configuration information can be received via high-level signaling.

[0011] Timers related to activation, deactivation, updating, or switching of AI / ML models may be configured individually for each of multiple TRPs.

[0012] The timer can be configured via network signaling.

[0013] The timer may be restarted based on information received for managing the AI / ML model.

[0014] The information used to manage the AI / ML model may include a related AI / ML model identifier (ID) or a related AI / ML function ID.

[0015] The activation, deactivation, update, or switching of the indicated AI / ML model may be performed based on the reception timing of the information for managing the AI / ML model or the transmission timing of the acknowledgment (ACK) of the information for managing the AI / ML model.

[0016] In another aspect of the present disclosure, there is provided a computer-readable recording medium having recorded thereon a program for executing the above-mentioned method.

[0017] In another aspect of the present disclosure, a UE configured to perform the above method is provided herein.

[0018] In another aspect of the present disclosure, a device configured to control a UE to execute the above method is provided herein.

[0019] In another aspect of the present disclosure, a method for operating a base station (BS) in a wireless communication system is provided herein. The method may include the following steps: sending AI / ML related configuration information to a UE to configure an AI / ML model for at least one TRP; and sending information for managing the AI / ML model configured to the UE. Based on the UE supporting multiple TRPs, the information for managing the AI / ML model may separately indicate the activation, deactivation, update, or switching of the AI / ML model for each of the multiple TRPs.

[0020] In another aspect of the present disclosure, a BS configured to perform the above method is provided herein.

[0021] Beneficial Effects

[0022] According to the present disclosure, wireless signal transmission and reception can be efficiently performed in a wireless communication system.

[0023] Those skilled in the art will appreciate that the effects that can be achieved using the present disclosure are not limited to those specifically described above, and other advantages of the present disclosure will be more clearly understood from the following detailed description in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1Physical channels used in a 3rd Generation Partnership Project (3GPP) system as an exemplary wireless communication system and a general signal transmission method using the same are shown.

[0025] Figure 2 Shows the radio frame structure.

[0026] Figure 3 A resource grid showing time slots.

[0027] Figure 4 An exemplary mapping of physical channels in time slots is shown.

[0028] Figure 5 An exemplary physical downlink shared channel (PDSCH) and ACK / NACK transmission and reception process is shown.

[0029] Figure 6 An exemplary physical uplink shared channel (PUSCH) transmission process is shown.

[0030] Figure 7 An example of a channel state information (CSI) related process is shown.

[0031] Figure 8 A diagram for explaining the concept of artificial intelligence / machine learning (AI / ML) / deep learning.

[0032] Figures 9 to 12 Various AI / ML models for deep learning are shown.

[0033] Fig.13 is a diagram illustrating split AI reasoning.

[0034] Fig.14 is a diagram showing the framework of 3GPP Radio Access Network (RAN) intelligence.

[0035] Figures 15 to 17 Shows the AI ​​model training and reasoning environment.

[0036] Fig.18 is a diagram for explaining operations of a user equipment (UE) according to an embodiment.

[0037] Fig.19 is a diagram for explaining the operation of a base station (BS) according to an embodiment.

[0038] Figure 20 to Figure 23 An example of the communication system 1 and the wireless device applied to the present disclosure is shown.

[0039] Fig.24 An exemplary discontinuous reception (DRX) operation suitable for use with the present disclosure is shown. DETAILED DESCRIPTION

[0040] Embodiments of the present disclosure are applicable to various wireless 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), and single carrier frequency division multiple access (SC-FDMA). CDMA can be implemented as a radio technology such as Universal Terrestrial Radio Access (UTRA) or CDMA2000. TDMA can be implemented as a radio technology 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 a radio technology such as Institute of Electrical and Electronics Engineers (IEEE) 802.11 (Wireless Fidelity (Wi-Fi)), IEEE 802.16 (Worldwide Interoperability for Microwave Access (WiMAX)), IEEE 802.20, and Evolved UTRA (E-UTRA). UTRA is part of Universal Mobile Telecommunications System (UMTS). 3rd Generation Partnership Project (3GPP) Long Term Evolution (LTE) is a part of Evolved UMTS (E-UMTS) using E-UTRA, and LTE-Advanced (A) is an evolved version of 3GPP LTE. 3GPP NR (New Radio or New Radio Access Technology) is an evolved version of 3GPP LTE / LTE-A.

[0041] As more and more communication devices require greater communication capacity, enhanced mobile broadband communications relative to traditional radio access technologies (RATs) are needed. In addition, large-scale machine-type communications (MTC), which can provide various services anytime and anywhere by connecting multiple devices and objects, is another important issue to be considered for next-generation communications. Discussions are also underway to design communication systems that take into account services / UEs that are sensitive to reliability and latency. Therefore, discussions are underway to introduce new radio access technologies that take into account enhanced mobile broadband communications (eMBB), massive MTC, and ultra-reliable low-latency communications (URLLC). In this disclosure, for simplicity, this technology will be referred to as NR (new radio or new RAT).

[0042] For the sake of brevity, 3GPP NR is mainly described, but the technical concept of the present disclosure is not limited thereto.

[0043] In the present disclosure, the term "setting" may be replaced by "configuration", and the two may be used interchangeably. In addition, conditional expressions (e.g., "if", "in the case of..." or "when") may be replaced by "based on..." or "under the state of...". In addition, the operation or software / hardware (SW / HW) configuration of the user equipment (UE) / base station (BS) may be derived / understood based on satisfying the corresponding conditions. When the processing of the receiving (or transmitting) side can be derived / understood from the processing of the transmitting (or receiving) side in the signal transmission / reception between the wireless communication device (e.g., BS and UE), its description may be omitted. For example, the signal determination / generation / encoding / transmission of the transmitting side may be understood as the signal monitoring reception / decoding / determination of the receiving side. In addition, when it is said that the UE performs (or does not perform) a specific operation, this may also be interpreted as the BS expecting / assuming (or not expecting / assuming) that the UE performs the specific operation. When it is said that the BS performs (or does not perform) a specific operation, this may also be interpreted as the UE expecting / assuming (or not expecting / assuming) that the BS performs the specific operation. In the following description, for the convenience of description, sections, implementations, examples, options, methods, schemes, etc. are distinguished and indexed from each other, which does not mean that each of them necessarily constitutes an independent invention or that each of them should only be implemented separately. Unless explicitly contradicted, it can be deduced / understood that at least some sections, implementations, examples, options, methods, schemes, etc. can be implemented in combination or can be omitted.

[0044] In a wireless communication system, a user equipment (UE) receives information from a base station (BS) through a downlink (DL) and transmits information to the BS through an uplink (UL). The information transmitted and received by the BS and the UE includes data and various control information, and includes various physical channels according to the type / purpose of the information transmitted and received by the UE and the BS.

[0045] Figure 1 Physical channels used in the 3GPP NR system and a general signal transmission method using the same are shown.

[0046] When the UE is powered on again from a power-off state or enters a new cell, in step S101, the UE performs an initial cell search process (e.g., establishes synchronization with the BS). To this end, the UE receives a synchronization signal block (SSB) from the BS. The SSB includes a primary synchronization signal (PSS), a secondary synchronization signal (SSS), and a physical broadcast channel (PBCH). The UE establishes synchronization with the BS based on the PSS / SSS and obtains information such as a cell identity (ID). The UE can obtain broadcast information in the cell based on the PBCH. The UE can receive a DL reference signal (RS) during the initial cell search process to monitor the DL channel status.

[0047] After the initial cell search, the UE may acquire more specific system information by receiving a physical downlink control channel (PDCCH) and receiving a physical downlink shared channel (PDSCH) based on information of the PDCCH in step S102.

[0048] The UE may perform a random access procedure to access the BS in steps S103 to S106. For random access, the UE may send a preamble to the BS on a physical random access channel (PRACH) (S103) and receive a response message to the preamble on a PDCCH and a PDSCH corresponding to the PDCCH (S104). In the case of contention-based random access, the UE may perform a contention resolution procedure by further sending a PRACH (S105) and receiving a PDCCH and a PDSCH corresponding to the PDCCH (S106).

[0049] After the foregoing process, the UE may receive PDCCH / PDSCH (S107) and send a physical uplink shared channel (PUSCH) / physical uplink control channel (PUCCH) (S108) as a general downlink / uplink signal transmission process. The control information sent from the UE to the BS is called uplink control information (UCI). UCI includes hybrid automatic repeat and request confirmation / negative determination (HARQ-ACK / NACK), scheduling request (SR), channel state information (CSI), etc. CSI includes channel quality indicator (CQI), precoding matrix indicator (PMI), rank indicator (RI), etc. Although UCI is usually sent on PUCCH, UCI may be sent on PUSCH when control information and business data need to be sent simultaneously. In addition, UCI may be sent aperiodically via PUSCH according to the request / command of the network.

[0050] Figure 2 The radio frame structure is shown. In NR, uplink transmission and downlink transmission are configured in frames. Each radio frame has a length of 10ms and is divided into two 5ms half frames (HF). Each half frame is divided into five 1ms subframes (SF). A subframe is divided into one or more time slots, and the number of time slots in a subframe depends on the subcarrier spacing (SCS). Depending on the cyclic prefix (CP), each time slot includes 12 or 14 orthogonal frequency division multiplexing (OFDM) symbols. When a normal CP is used, each time slot includes 14 OFDM symbols. When an extended CP is used, each time slot includes 12 OFDM symbols.

[0051] Table 1 exemplarily shows that the number of symbols per slot, the number of slots per frame, and the number of slots per subframe vary according to the SCS when a normal CP is used.

[0052] [Table 1]

[0053] <![CDATA[SCS(15*2 u )]]> <![CDATA[N slot symb ]]> <![CDATA[N frame,u slot ]]> <![CDATA[N subframe,u slot ]]> 15KHz(u=0) 14 10 1 30KHz(u=1) 14 20 2 60KHz(u=2) 14 40 4 120KHz(u=3) 14 80 8 240KHz(u=4) 14 160 16

[0054] *N slot symb : The number of symbols in a time slot

[0055] *N frame,u slot : Number of time slots in a frame

[0056] *N subframe,u slot : Number of time slots in a subframe

[0057] Table 2 shows that the number of symbols per slot, the number of slots per frame, and the number of slots per subframe vary according to the SCS when the extended CP is used.

[0058] [Table 2]

[0059] <![CDATA[SCS(15*2 u )]]> <![CDATA[N slot symb ]]> <![CDATA[N frame,u slot ]]> <![CDATA[N subframe,u slot ]]> 60KHz(u=2) 12 40 4

[0060] The structure of the frame is only an example. The number of subframes, the number of time slots, and the number of symbols in a frame may vary.

[0061] In the NR system, OFDM parameter sets (e.g., SCS) may be configured differently for multiple cells aggregated for one UE. Therefore, the (absolute time) duration of a time resource (e.g., SF, time slot, or TTI) (referred to as a time unit (TU) for simplicity) consisting of the same number of symbols may be configured differently between aggregated cells. Here, the symbol may include an OFDM symbol (or CP-OFDM symbol) and an SC-FDMA symbol (or a discrete Fourier transform-spread-OFDM (DFT-s-OFDM) symbol).

[0062] Figure 3 A resource grid showing a time slot. A time slot includes multiple symbols in the time domain. For example, when a normal CP is used, a time slot includes 14 symbols. However, when an extended CP is used, a time slot includes 12 symbols. A carrier includes multiple subcarriers in the frequency domain. A resource block (RB) is defined as a plurality of consecutive subcarriers (e.g., 12 consecutive subcarriers) in the frequency domain. A bandwidth part (BWP) may be defined as a plurality of consecutive physical RBs (PRBs) in the frequency domain and corresponds to a single parameter set (e.g., SCS, CP length, etc.). A carrier may include up to N (e.g., five) BWPs. Data communication may be performed via enabled BWPs, and only one BWP may be enabled for a UE. In a resource grid, each element is referred to as a resource element (RE), and a complex symbol may be mapped to each RE.

[0063] Figure 4An exemplary mapping of physical channels in a time slot is shown. PDCCH may be sent in the DL control region, and PDSCH may be sent in the DL data region. PUCCH may be sent in the UL control region, and PUSCH may be sent in the UL data region. A guard period (GP) provides a time gap for a transmission mode to a reception mode switch or a reception mode to a transmission mode switch at the BS and the UE. Some symbols in a subframe at the time of a DL to UL switch may be configured as a GP.

[0064] The individual physical channels are described in more detail below.

[0065] PDCCH transmits DCI. For example, PDCCH (i.e., DCI) may carry information about the transmission format and resource allocation of the DL shared channel (DL-SCH), resource allocation information of the uplink shared channel (UL-SCH), paging information about the paging channel (PCH), system information about DL-SCH, information about resource allocation of high-level control messages (e.g., RAR sent on PDSCH), transmit power control commands, information about the activation / release of the configured scheduling, etc. DCI includes a cyclic redundancy check (CRC). The CRC is masked using various identifiers (IDs) (e.g., radio network temporary identifier (RNTI)) according to the owner or purpose of the PDCCH. For example, if the PDCCH is used for a specific UE, the CRC is masked by the UE ID (e.g., cell-RNTI (C-RNTI)). If the PDCCH is used for a paging message, the CRC is masked by the paging-RNTI (P-RNTI). If the PDCCH is for system information (eg, system information block (SIB)), the CRC is masked by a system information RNTI (SI-RNTI). When the PDCCH is for RAR, the CRC is masked by a random access-RNTI (RA-RNTI).

[0066] The PDCCH includes 1, 2, 4, 8 or 16 control channel elements (CCEs) depending on its aggregation level (AL). CCE is a logical allocation unit for providing a specific code rate to the PDCCH according to the radio channel state. CCE includes 6 resource element groups (REGs), each REG being defined by one OFDM symbol × one (P)RB. The PDCCH is transmitted in a control resource set (CORESET). A CORESET is defined as a set of REGs with a given set of parameters (e.g., SCS, CP length, etc.). Multiple CORESETs for one UE may overlap with each other in the time / frequency domain. The CORESET may be configured by system information (e.g., master information block (MIB)) or UE-specific high-level signaling (e.g., radio resource control (RRC) signaling). Specifically, the number of RBs and the number of symbols (up to 3) in a CORESET may be configured by high-level signaling.

[0067] For PDCCH reception / detection, the UE monitors PDCCH candidates. PDCCH candidates are CCEs that the UE should monitor to detect PDCCH. Each PDCCH candidate is defined as 1, 2, 4, 8 or 16 CCEs according to the AL. Monitoring includes (blind) decoding of the PDCCH candidates. The set of PDCCH candidates decoded by the UE is defined as the PDCCH search space (SS). The SS can be a common search space (CSS) or a UE-specific search space (USS). The UE can obtain DCI by monitoring PDCCH candidates in one or more SSs configured by MIB or high-level signaling. Each CORESET is associated with one or more SSs, and each SS is associated with a CORESET. The SS can be defined based on the following parameters.

[0068] -controlResourceSetId: CORESET related to SS.

[0069] -monitoringSlotPeriodicityAndOffset: PDCCH monitoring periodicity (time slot) and PDCCH monitoring offset (time slot).

[0070] - monitoringSymbolsWithinSlot: PDCCH monitoring symbols in a slot (eg, the first symbol of a CORESET).

[0071] -nrofCandidates: the number of PDCCH candidates for each AL={1,2,4,8,16} (one of 0, 1, 2, 3, 4, 5, 6 and 8).

[0072] *The timing (eg, time / frequency resources) at which the UE is to monitor a PDCCH candidate is defined as a PDCCH (monitoring) timing. One or more PDCCH (monitoring) timings may be configured in a time slot.

[0073] Table 3 shows the characteristics of each SS.

[0074] [Table 3]

[0075]

[0076]

[0077] Table 4 shows the DCI format transmitted on the PDCCH.

[0078] [Table 4]

[0079]

[0080] DCI format 0_0 can be used to schedule TB-based (or TB-level) PUSCH, and DCI format 0_1 ​​can be used to schedule TB-based (or TB-level) PUSCH or code block group (CBG)-based (or CBG-level) PUSCH. DCI format 1_0 can be used to schedule TB-based (or TB-level) PDSCH, and DCI format 1_1 can be used to schedule TB-based (or TB-level) PDSCH or CBG-based (or CBG-level) PDSCH (or DL ​​grant DCI). DCI format 0_0 / 0_1 can be called UL grant DCI or UL scheduling information, and DCI format 1_0 / 1_1 can be called DL grant DCI or DL ​​scheduling information. DCI format 2_0 is used to transmit dynamic slot format information (e.g., dynamic slot format indicator (SFI)) to the UE, and DCI format 2_1 is used to transmit DL preemption information to the UE. DCI format 2_0 and / or DCI format 2_1 can be transmitted to a corresponding group of UEs on a group common PDCCH (a PDCCH pointing to a group of UEs).

[0081] DCI format 0_0 and DCI format 1_0 may be referred to as fallback DCI formats, while DCI format 0_1 ​​and DCI format 1_1 may be referred to as non-fallback DCI formats. Under the fallback DCI format, the DCI size / field configuration remains the same regardless of the UE configuration. In contrast, under the non-fallback DCI format, the DCI size / field configuration varies according to the UE configuration.

[0082] PDSCH transmits DL data (e.g., DL shared channel transport block (DL-SCH TB)) and uses modulation schemes such as quadrature phase shift keying (QPSK), 16-ary quadrature amplitude modulation (16QAM), 64QAM, or 256QAM. TB is encoded as a codeword. PDSCH can transmit up to two codewords. Scrambling and modulation mapping can be performed on the basis of the codeword, and the modulation symbols generated from each codeword can be mapped to one or more layers. Each layer is mapped to a resource together with a demodulation reference signal (DMRS), and an OFDM symbol signal is generated from the layer mapped with the DMRS and sent through the corresponding antenna port.

[0083] PUCCH transmits uplink control information (UCI). UCI includes the following information.

[0084] -SR (Scheduling Request): Information used to request UL-SCH resources.

[0085] -HARQ (Hybrid Automatic Repeat Request) -ACK (Acknowledgement): A response to a DL data packet (e.g., a codeword) on the PDSCH. HARQ-ACK indicates whether the DL data packet was successfully received. In response to a single codeword, a 1-bit HARQ-ACK may be sent. In response to two codewords, a 2-bit HARQ-ACK may be sent. HARQ-ACK responses include positive ACK (abbreviated as ACK), negative ACK (NACK), discontinuous transmission (DTX), or NACK / DTX. The term HARQ-ACK may be used interchangeably with HARQ ACK / NACK and ACK / NACK.

[0086] -CSI (Channel State Information): Feedback information of DL channels. Multiple Input Multiple Output (MIMO) related feedback information includes RI and PMI.

[0087] Table 5 shows an exemplary PUCCH format. Based on PUCCH transmission duration, the PUCCH format may be divided into short PUCCH (formats 0 and 2) and long PUCCH (formats 1, 3, and 4).

[0088] [Table 5]

[0089]

[0090] PUCCH format 0 transmits up to 2 bits of UCI and is mapped in a sequence-based manner for easy transmission. Specifically, the UE sends a specific UCI to the BS by sending one of multiple sequences on the PUCCH of PUCCH format 0. The UE sends the PUCCH of PUCCH format 0 in the PUCCH resources configured for the corresponding SR only when the UE sends a positive SR.

[0091] PUCCH format 1 transmits up to 2 bits of UCI, and the modulation symbol of UCI is spread in the time domain with an orthogonal cover code (OCC) (configured differently depending on whether frequency hopping is performed). DMRS is transmitted in symbols where modulation symbols are not transmitted (ie, transmitted in time division multiplexing (TDM)).

[0092] PUCCH format 2 transmits more than 2 bits of UCI, and the modulation symbols of DCI are sent with frequency division multiplexing (FDM) using DMRS. DMRS are located in symbols #1, #4, #7, and #10 of a given RB at a density of 1 / 3. Pseudo-noise (PN) sequences are used for DMRS sequences. For 2-symbol PUCCH format 2, frequency hopping can be enabled.

[0093] PUCCH format 3 does not support UE multiplexing in the same PRBS and transmits UCI of more than 2 bits. In other words, the PUCCH resource of PUCCH format 3 does not include OCC. The modulation symbol is transmitted in TDM using DMRS.

[0094] PUCCH format 4 supports multiplexing of up to 4 UEs in the same PRBS and transmits UCI of more than 2 bits. In other words, the PUCCH resource of PUCCH format 3 includes OCC. The modulation symbol is transmitted in TDM using DMRS.

[0095] At least one of the one or two or more cells configured in the UE may be configured for PUCCH transmission. At least the primary cell may be configured as a cell for PUCCH transmission. Based on at least one cell configured for PUCCH transmission, at least one PUCCH cell group may be configured in the UE, and each PUCCH cell group includes one or more cells. The PUCCH cell group may be referred to as a PUCCH group. PUCCH transmission may be configured for the SCell as well as the primary cell. The primary cell belongs to the primary PUCCH group, and the PUCCH-SCell configured for PUCCH transmission belongs to the secondary PUCCH group. The PUCCH on the primary cell can be used for cells belonging to the primary PUCCH group, and the PUCCH on the PUCCH-SCell can be used for cells belonging to the secondary PUCCH group.

[0096] The PUSCH transmits UL data (e.g., UL shared channel transport block (UL-SCH TB)) and / or UCI based on a CP-OFDM waveform or a DFT-s-OFDM waveform. When the PUSCH is sent with a DFT-s-OFDM waveform, the UE sends the PUSCH with transform precoding. For example, when transform precoding is not possible (e.g., disabled), the UE may send the PUSCH with a CP-OFDM waveform, and when transform precoding is possible (e.g., enabled), the UE may send the PUSCH with a CP-OFDM or DFT-s-OFDM waveform. PUSCH transmission may be dynamically scheduled by a UL grant in the DCI, or semi-statically scheduled by higher layer (e.g., RRC) signaling (and / or layer 1 (L1) signaling such as PDCCH) (configured scheduling or configured grant). PUSCH transmission may be performed in a codebook-based or non-codebook-based manner.

[0097] Figure 5 An exemplary ACK / NACK transmission process is shown. Figure 5 , the UE may detect the PDCCH in slot #n. The PDCCH includes DL scheduling information (e.g., DCI format 1_0 or DCI format 1_1). The PDCCH indicates DL assignment with PDSCH offset K0 and PDSCH with HARQ-ACK report offset K1. For example, DCI format 1_0 and DCI format 1_1 may include the following information.

[0098] - Frequency domain resource assignment: indicates the RB set assigned to PDSCH.

[0099] - Time domain resource assignment: indicates K0 and the starting position (eg, OFDM symbol index) and length (eg, number of OFDM symbols) of the PDSCH in the slot.

[0100] -PDSCH-to-HARQ_feedback timing indicator: indicates K1.

[0101] -HARQ process number (4 bits): indicates the HARQ process ID of the data (eg, PDSCH or TB).

[0102] - PUCCH resource indicator (PRI): indicates a PUCCH resource to be used for UCI transmission among multiple PUCCH resources in a PUCCH resource set.

[0103] After receiving PDSCH in slot #(n+K0) according to the scheduling information of slot #n, the UE may send UCI on PUCCH in slot #(n+K1). UCI may include a HARQ-ACK response to PDSCH. For convenience, Figure 5Based on the assumption that the SCS of PDSCH is equal to the SCS of PUCCH and slot #n1=slot #(n+K0), this should not be construed as limiting the present disclosure. When the SCSs are different, K1 may be indicated / interpreted based on the SCS of PUCCH.

[0104] In the case where the PDSCH is configured to carry up to one TB, the HARQ-ACK response may be configured in one bit. In the case where the PDSCH is configured to carry up to two TBs, if spatial bundling is not configured, the HARQ-ACK response may be configured in 2 bits, and if spatial bundling is configured, the HARQ-ACK response may be configured in 1 bit. When slot #(n+K1) is designated as the HARQ-ACK transmission timing of multiple PDSCHs, the UCI transmitted in slot #(n+K1) includes the HARQ-ACK responses to the multiple PDSCHs.

[0105] Whether the UE should perform spatial bundling for HARQ-ACK responses may be configured for each cell group (e.g., via RRC / high-layer signaling). For example, spatial bundling may be configured for each individual HARQ-ACK response sent on the PUCCH and / or a HARQ-ACK response sent on the PUSCH.

[0106] Spatial bundling can be supported when up to two (or two or more) TBs (or codewords) can be received at a time (can be or be scheduled by one DCI) in the corresponding serving cell (for example, when the high-level parameter maxNrofCodeWordsScheduledByDCI indicates 2 TBs). More than four layers can be used for 2TB transmission, and up to four layers can be used for 1TB transmission. As a result, when spatial bundling is configured for the corresponding cell group, spatial bundling can be performed for serving cells in the cell group where more than four layers can be scheduled. A UE that wants to send a HARQ-ACK response through spatial bundling can generate a HARQ-ACK response by performing a (bit-by-bit) logical AND operation on the A / N bits of multiple TBs.

[0107] For example, assuming that a UE receives a DCI that schedules two TBs and receives the two TBs on a PDSCH based on the DCI, the UE that performs spatial bundling can generate a single A / N bit by a logical AND operation between a first A / N bit of a first TB and a second A / N bit of a second TB. As a result, when both the first TB and the second TB are ACK, the UE reports an ACK bit value to the BS, and when at least one TB is NACK, the UE reports a NACK bit value to the BS.

[0108] For example, when only one TB is actually scheduled in a serving cell configured to receive two TBs, the UE may generate a single A / N bit by performing a logical AND operation on the A / N bits of the one TB and a bit value 1. As a result, the UE reports the A / N bits of the one TB to the BS.

[0109] There are multiple parallel DL HARQ processes at the BS / UE for DL ​​transmissions. Multiple parallel HARQ processes allow continuous DL transmissions while the BS waits for HARQ feedback indicating the success or failure of reception of the previous DL transmission. Each HARQ process is associated with a HARQ buffer in the media access control (MAC) layer. Each DL HARQ process manages state variables such as the number of MAC physical data unit (PDU) transmissions, HARQ feedback for the MAC PDUs in the buffer, and the current redundancy version. Each HARQ process is identified by a HARQ process ID.

[0110] Figure 6 An exemplary PUSCH transmission process is shown. Figure 6 , the UE may detect the PDCCH in slot #n. The PDCCH includes DL scheduling information (eg, DCI format 1_0 or 1_1). The DCI format 1_0 or 1_1 may include the following information.

[0111] - Frequency domain resource assignment: indicates the RB set assigned to PUSCH.

[0112] -Time domain resource assignment: Indicates the slot offset K2 and the starting position (e.g., OFDM symbol index) and duration (e.g., number of OFDM symbols) of the PUSCH in the slot. The starting symbol and length of the PUSCH may be indicated by a start and length indicator value (SLIV) or separately.

[0113] Then, the UE may transmit the PUSCH in slot #(n+K2) according to the scheduling information in slot #n. The PUSCH includes the UL-SCH TB.

[0114] CSI related operations

[0115] Figure 7 An example of a CSI-related process is shown.

[0116] The UE receives CSI-related configuration information from the BS via RRC signaling (710). The CSI-related configuration information may include at least one of channel state information-interference measurement (CSI-IM) related information, CSI measurement related information, CSI resource configuration related information, CSI-RS resource related information, or CSI report configuration related information.

[0117] -CSI-IM resources may be configured for interference measurement (IM) of the UE. In the time domain, the CSI-IM resource set may be configured as periodic, semi-persistent, or aperiodic. The CSI-IM resources may be configured as zero power (ZP)-CSI-RS of the UE. The ZP-CSI-RS may be configured to be distinguished from the non-zero power (NZP)-CSI-RS.

[0118] - The UE may assume that the CSI-RS resources for channel measurement and the CSI-IM / NZP CSI-RS resources for interference measurement configured for one CSI report have a QCL relationship with respect to the "QCL-TypeD" of each resource (when the NZP CSI-RS resources are used for interference measurement).

[0119] - The CSI resource configuration may include at least one of a CSI-IM resource for interference measurement, an NZP CSI-RS resource for interference measurement, and an NZP CSI-RS resource for channel measurement. The channel measurement resource (CMR) may be an NZP CSI-RS for CSI acquisition, and the interference measurement resource (IMR) may be an NZP CSI-RS for CSI-IM and IM.

[0120] -CSI-RS can be configured for one or more UEs. Different CSI-RS configurations can be provided for each UE, or the same CSI-RS configuration can be provided to multiple UEs. CSI-RS can support up to 32 antenna ports. CSI-RS corresponding to N (N is 1 or greater) antenna ports can be mapped to N RE positions within a time-frequency unit corresponding to one time slot and one RB. When N is 2 or greater, N-port CSI-RS can be multiplexed by CDM, FDM and / or TDM methods. CSI-RS can be mapped to the remaining REs except for REs mapped with CORESET, DMRS and SSB. In the frequency domain, CSI-RS can be configured for the entire bandwidth, a partial bandwidth part (BWP) or a partial bandwidth. CSI-RS can be sent in each RB within the bandwidth in which CSI-RS is configured (ie, density = 1), or CSI-RS can be sent in every two RBs (eg, even or odd RBs) (ie, density = 1 / 2). When CSI-RS is used as a tracking reference signal (TRS), a single-port CSI-RS may be mapped on three subcarriers in each resource block (i.e., density = 3). One or more CSI-RS resource sets may be configured for a UE in the time domain. Each CSI-RS resource set may include one or more CSI-RS configurations. Each CSI-RS resource set may be configured as periodic, semi-persistent, or aperiodic.

[0121] -CSI report configuration may include configuration of feedback type, measurement resources, report type, etc. NZP-CSI-RS resource set may be used for CSI report configuration of corresponding UE. NZP-CSI-RS resource set may be associated with CSI-RS or SSB. Multiple periodic NZP-CSI-RS resource sets may be configured as TRS resource sets. (i) Feedback type includes channel quality indicator (CQI), precoding matrix indicator (PMI), CSI-RS resource indicator (CRI), SSB resource block indicator (SSBRI), layer indicator (LI), rank indicator (RI), first layer (L1)-reference signal received strength (RSRP), etc. (ii) Measurement resources may include configuration of downlink signals and / or downlink resources on which UE performs measurements to determine feedback information. Measurement resources may be configured as ZP and / or NZP CSI-RS resource sets associated with CSI report configuration. NZP CSI-RS resource sets may include CSI-RS sets or SSB sets. For example, L1-RSRP may be measured for a CSI-RS set or an SSB set. (iii) The report type may include the time at which the UE performs the report and the configuration of the uplink channel. The reporting time may be configured as periodic, semi-persistent, or aperiodic. Periodic CSI reports may be sent on the PUCCH. Semi-persistent CSI reports may be sent on the PUCCH or PUSCH based on a MAC CE indicating enable / disable. Aperiodic CSI reports may be indicated by DCI signaling. For example, the CSI request field of an uplink grant may indicate one of various report trigger sizes. Aperiodic CSI reports may be sent on the PUSCH.

[0122] The UE measures the CSI based on the configuration information related to the CSI. The CSI measurement may include receiving a CSI-RS (720) and acquiring the CSI by calculating the received CSI-RS (730).

[0123] The UE may send a CSI report to the BS (740). For the CSI report, time resources and frequency resources available to the UE are controlled by the BS. The channel state information (CSI) includes at least one of a channel quality indicator (CQI), a precoding matrix indicator (PMI), a CSI-RS resource indicator (CRI), a SS / PBCH block resource indicator (SSBRI), a layer indicator (LI), a rank indicator (RI), L1-RSRP, and / or L-SINR.

[0124] The time domain behavior of CSI reporting supports periodic, semi-persistent, and aperiodic. i) Periodic CSI reporting is performed in short PUCCH and long PUCCH. The periodicity and slot offset of periodic CSI reporting can be configured by RRC, and refer to CSI-ReportConfig IE. ii) SP (semi-periodic) CSI reporting is performed in short PUCCH, long PUCCH or PUSCH. For SP CSI in short / long PUCCH, the periodicity and slot offset are configured by RRC, and CSI reporting is enabled / disabled through a separate MAC CE / DCI. For SP CSI in PUSCH, the periodicity of SP CSI reporting is configured by RRC, but the slot offset is not configured by RRC and SP CSI reporting is enabled / disabled by DCI (format 0_1). For SP CSI reporting in PUSCH, a separate RNTI (SP-CSI C-RNTI) is used. The initial CSI report timing follows the PUSCH time domain allocation value indicated by DCI, and subsequent CSI report timing follows the periodicity configured by RRC. DCI format 0_1 ​​may include a CSI request field and enable / disable a specific configured SP-CSI triggering state. SP CSI reporting has the same or similar enable / disable mechanism as with data transmission in SPS PUSCH. iii) Aperiodic CSI reporting is performed in PUSCH and triggered by DCI. In this case, information related to the triggering of aperiodic CSI reporting may be transmitted / indicated / configured via MAC-CE. For AP CSI with AP CSI-RS, the AP CSI-RS timing is configured by RRC and the timing of the AP CSI report is dynamically controlled by DCI.

[0125] The CSI codebooks (e.g., PMI codebooks) defined in the NR specification can be roughly classified into type I codebooks and type II codebooks. Type I codebooks are mainly for single-user MIMO (SU-MIMO) supporting both high-order and low-order. Type II codebooks mainly support multi-user MIMO (MU-MIMO) that can handle up to two layers. Although type II codebooks can provide more accurate CSI compared to type I, type II codebooks may also increase signaling overhead. On the other hand, an enhanced type II codebook is introduced to solve the CSI overhead problem associated with the existing type II codebook. The enhanced type II codebook can reduce the payload of the codebook by considering the correlation in the frequency domain.

[0126] The CSI report on PUSCH can be configured as part 1 and part 2. Part 1 has a fixed payload size that identifies the number of information bits in part 2. Part 1 is completely transmitted before part 2.

[0127] - For Type I CSI feedback, Part 1 includes RI (if reported), CRI (if reported) and CQI for the first codeword. Part 2 includes PMI, and when RI>4, Part 2 also includes CQI.

[0128] - For Type II CSI feedback, Part 1 includes the RI (if reported), CQI, and an indication of the number of non-zero WB amplitude coefficients for each layer of Type IICSI. Part 2 includes the PMI of Type IICSI.

[0129] - For enhancement type IICSI feedback, part 1 includes the RI (if reported), CQI and an indication of the total number of non-zero WB amplitude coefficients of all layers of enhancement type IICSI. Part 2 includes the PMI of enhancement type IICSI.

[0130] For PUSCH, the CSI report consists of two parts. If the CSI payload to be reported is smaller than the payload size provided by the PUSCH resources allocated for CSI reporting, the UE may drop part of the Part 2 CSI.

[0131] Semi-persistent CSI reporting performed in PUCCH format 3 or 4 supports Type IICSI feedback, but only supports part 1 of Type IICSI feedback.

[0132] Quasi-isotope (QCL)

[0133] Two antenna ports are quasi-colocated when the channel properties of an antenna port are to be inferred from the channel of another antenna port. The channel properties may include one or more of delay spread, Doppler spread, frequency / Doppler shift, average received power, receive timing / average delay, and spatial RX parameters.

[0134] A list of multiple TCI state configurations can be configured in the UE through the high-level parameter PDSCH-Config, and each TCI state is linked to the QCL configuration parameters between one or two DL reference signals and the DM-RS port of the PDSCH. The QCL may include qcl-Type1 for the first DLRS and qcl-Type2 for the second DLRS. The QCL type may correspond to one of the following.

[0135] - "QCL-TypeA": {Doppler shift, Doppler spread, average delay, delay spread}

[0136] - "QCL-TypeB": {Doppler shift, Doppler spread}

[0137] - "QCL-TypeC": {Doppler shift, average delay}

[0138] - "QCL-TypeD": {spatial Rx parameters}

[0139] Beam Management (BM)

[0140] BM refers to a series of processes for acquiring and maintaining a BS beam set (transmit receive point (TRP) beam) and / or a UE beam set that can be used for DL ​​and UL transmission / reception. BM may include the following processes and terms.

[0141] - Beam measurement: An operation in which a BS or UE measures characteristics of a received beamforming signal.

[0142] -Beam determination: The operation by which a BS or UE selects its Tx / Rx beam.

[0143] -Beam sweeping: An operation of covering the spatial domain using Tx and / or Rx beams within a prescribed time interval according to a predetermined method.

[0144] -Beam reporting: An operation in which the UE reports information about a beamformed signal based on beam measurements.

[0145] The BM process may be divided into (1) a DLBM process using SSB or CSI-RS and (2) a ULBM process using SRS. In addition, each BM process may include a Tx beam sweep for determining a Tx beam and an Rx beam sweep for determining an Rx beam.

[0146] The DLBM process may include (1) transmission of a beamformed DL RS (eg, CSI-RS or SSB) from a BS and (2) beam reporting from a UE.

[0147] The beam report may include a preferred DL RS ID and a reference signal received power (RSRP) corresponding to the preferred DL RS ID. The DL RS ID may be a SSB resource indicator (SSBRI) or a CSI-RS resource indicator (CRI).

[0148] Artificial Intelligence / Machine Learning (AI / ML)

[0149] With the development of AI / ML technology, the nodes and UEs constituting the wireless communication network are becoming more and more intelligent / advanced, especially due to the intelligence of the network / BS, it is expected that various network / BS-determined parameter values ​​(e.g., the transmission and reception power of each BS, the transmission power of each UE, the precoder / beam of the BS / UE, the time / frequency resource allocation of each UE, or the duplex method of the BS) will be quickly optimized and derived / applied according to various environmental parameters (e.g., the distribution / location of the BS, the distribution / location / material of the buildings / furniture, the location / moving direction / speed of the UE, and climate information). In line with this trend, many standardization organizations (e.g., 3GPP or O-RAN) are considering the introduction of network / BS-determined parameter values, and research on this is also actively underway.

[0150] In a narrow sense, AI / ML can be easily referred to as artificial intelligence based on deep learning, but the concept is shown in Figure 8 middle.

[0151] - Artificial Intelligence: This can correspond to all automation where machines replace human work.

[0152] -Machine Learning: Machines learn decision-making patterns from data without explicit programming rules.

[0153] - Deep learning: This is an AI / ML model based on artificial neural networks, where the machine performs everything from unstructured data to feature extraction and determination at once, and the algorithm relies on biological neural systems, i.e., a multi-layer interconnected node network for feature extraction and transformation inspired by neural networks. Common deep learning network architectures may include deep neural networks (DNNs), recurrent neural networks (RNNs), and convolutional neural networks (CNNs).

[0154] Classification of AI / ML types according to various references

[0155] 1. Offline vs. online

[0156] (1) Offline learning: This follows a sequential process of database collection, learning, and prediction. In other words, collection and learning are performed offline, and the completed program can be installed on-site and used for prediction. In most cases, this offline learning method is used. In offline learning, the system does not learn incrementally. Instead, all available collected data is used for learning, and the results are applied without further learning. If new data needs to be learned, learning can be started again using the new complete data.

[0157] (2) Online learning: Online learning is a method of improving performance little by little by incrementally learning with additionally generated data, based on the fact that data to be used for learning is continuously generated recently via the Internet. Real-time learning is performed on specific data units (batches) collected online, allowing the system to quickly adapt to changing data.

[0158] In order to build an AI system, learning can be performed only through online learning using only real-time generated data. Alternatively, after offline learning using a specific data set, additional learning (online + offline learning) can be performed using subsequently generated real-time data.

[0159] 2. Classification based on AI / ML framework concepts

[0160] (1) Centralized learning: When training data collected from multiple different nodes are reported to a centralized node, all data resources / storage / learning (e.g., supervised, unsupervised, and reinforcement learning) are performed by one central node.

[0161] (2) Federated Learning: A collective AI / ML model is configured based on data across decentralized data owners. Instead of using the data in the AI / ML model, local nodes / devices collect data and train their copies of the AI / ML model, so there is no need to report the source data to the central node. In federated learning, the parameters / weights of the AI / ML model can be sent back to the centralized node to support general AI / ML model training. The advantages of federated learning include increased computing speed and advantages in information security. That is, there is no need to upload personal data to a central server for processing, which can prevent the leakage and abuse of personal information.

[0162] (3) Distributed learning: Machine learning processing represents the concept of scaling and deployment across a cluster of nodes. Splitting and sharing AI / ML model training across multiple nodes operating simultaneously accelerates AI / ML model training.

[0163] 3. Classification based on learning methods

[0164] (1) Supervised learning: Supervised learning is a machine learning task that aims to learn a mapping function from input to output given a labeled data set. The input data is called training data and has a known label or outcome. Examples of supervised learning may include: (i) regression: linear regression, logistic regression; (ii) instance-based algorithms: k-nearest neighbor (KNN); (iii) decision tree algorithms: CART; (iv) support vector machines: SVM; (v) Bayesian algorithms: Naive Bayes; and (vi) ensemble algorithms: extreme gradient boosting, packing: random forest. Due to regression and classification problems, supervised learning can be further grouped, and classification predicts labels and regression predicts quantities.

[0165] (2) Unsupervised learning: This is a machine learning task that aims to learn a function that describes the hidden structure in unlabeled data. The input data is unlabeled and there is no known outcome. Some examples of unsupervised learning include K-means clustering, principal component analysis (PCA), nonlinear independent component analysis (ICA), and LSTM.

[0166] (3) Reinforcement Learning: In reinforcement learning (RL), an agent aims to optimize a long-term goal by interacting with the environment based on a trial-and-error process, which is goal-oriented learning based on interaction with the environment. Examples of RL algorithms may include (i) Q-learning, (ii) multi-armed bandit learning, (iii) deep Q-networks, state-action-reward-state-action (SARSA), (iv) temporal difference learning, (v) actuator-evaluator reinforcement learning, (vi) deep deterministic policy gradient, and (vii) Monte Carlo tree search. RL can be further grouped into AI / ML model-based RL and AI / ML model-free RL. Model-based RL is an RL algorithm that uses a predictive AI / ML model to obtain the transition probability between states using various dynamic states of the environment and an AI / ML model of the rewards caused by these states. Model-free RL is a value-based or policy-based RL algorithm that achieves maximum future rewards, which has low computational complexity in terms of multi-agent environments / states and does not require an accurate representation of the environment. RL algorithms can also be classified into value-based RL versus policy-based RL, policy-based RL versus non-policy RL, etc.

[0167] AI / ML Models

[0168] Fig. 9 An example of a feed-forward neural network (FFNN) AI / ML model is shown. Fig. 9 ,FFNN AI / ML model includes input layer, hidden layer and output layer.

[0169] Fig.10 An example of a recurrent neural network (RNN) AI / ML model is shown. Fig.10 , the RNN AI / ML model is an artificial neural network in which hidden nodes are connected to directed edges to form a directed cycle. It is an AI / ML model suitable for processing data that appears sequentially (for example, speech or text). One type of RNN is the long short-term memory (LSTM), which is a structure that adds the cell state to the hidden state of the RNN. In detail, in LSTM, an input gate, a forget gate, and an output gate are added to the RNN cell, and the cell state is added. Fig.10 In this example, A represents a neural network, and x t Indicates the input value, h t Indicates the output value. Here, h t It can represent the current state value based on time, and h t-1 Can represent the previous state value.

[0170] Fig.11A convolutional neural network (CNN) AI / ML model is shown. CNN is used for two purposes, including reducing the complexity of the AI / ML model and extracting good features by applying convolution calculations commonly used in the fields of video processing or image processing. Fig.11 , a kernel or filter means a unit / structure that applies weights to an input within a specific range / unit. A kernel (or filter) can be modified by learning. The stride is the range of movement of the kernel within the input. A feature map is the result of applying a kernel to an input. Padding refers to a value added to adjust the size of a feature map. To increase robustness to distortion and variation, multiple feature maps can be extracted. Pooling refers to the calculation of reducing the size of a feature map by downsampling it (e.g., max pooling or average pooling).

[0171] Fig.12 Figure 2 shows an autoencoder AI / ML model. Fig.12 , an autoencoder is a neural network that receives a feature vector x and outputs the same or similar vector x' and an unsupervised learning where the input node and the output node have the same features. The autoencoder reconstructs the input, so the output can be called a reconstruction. The loss function can be expressed according to the following formula 1.

[0172] [Formula 1]

[0173] Where h = f(x) = Wx, x′ = g(h) = Vh

[0174] In Equation 12, the loss function of the autoencoder is calculated based on the difference between the input and the output. Based on the loss function of the autoencoder, the degree of loss in the input is evaluated, and an optimization process is applied to the autoencoder to minimize the loss.

[0175] Fig.13 is a diagram illustrating split AI reasoning.

[0176] Fig.13 It shows a case where the model inference function is performed collaboratively between a terminal device such as a UE and a network AI / ML endpoint during split AI operation.

[0177] In addition to the model inference function, the model training function, executor, and data collection function can each be split into multiple parts based on the current task and environment. These functions can be performed through the collaboration of multiple entities.

[0178] For example, the computationally intensive part and the energy intensive part can be executed at the network endpoint, while the privacy-sensitive part and the delay-sensitive part can be executed on the terminal device. In this case, the terminal device executes the task / model based on the input data up to a specific part / layer, and then sends the intermediate data to the network endpoint. The network endpoint executes the remaining parts / layers and provides the reasoning output to one or more devices that perform the operation / task.

[0179] The following describes the functional framework for AI operations.

[0180] Herein, the following terms may be defined to explain AI (or AI / ML) in more detail.

[0181] -Data Collection: Data collected from network nodes, management entities or UEs, which is used as the basis for AI model training, data analysis and inference.

[0182] -AI Model: A data-driven algorithm in which AI techniques are applied to generate a set of outputs based on a set of inputs, including prediction information and / or decision parameters.

[0183] -AI / ML training: The online or offline process of training an AI model by learning features and patterns to best represent the data and obtaining an AI / ML model trained for inference.

[0184] -AI / ML Inference: The process of using trained AI models to make predictions or draw decisions based on collected data and AI models.

[0185] Reference Fig.14 , the data collection function 10 collects input data and provides the processed input data to the model training function 20 and the model reasoning function 30.

[0186] For example, input data may include measurements from UEs or other network entities, feedback from actuators, and outputs from AI models.

[0187] The data collection function 10 performs data preparation based on the input data and provides the input data processed by the data preparation. Here, the data collection function 10 does not perform specific data preparation (e.g., data preprocessing and cleaning, forming and transforming) for each AI algorithm, but the data collection function 10 can perform data preparation common to the AI ​​algorithm.

[0188] After completing the data preparation process, the data collection function 10 provides the training data 11 to the model training function 20 and provides the inference data 12 to the model inference function 30. Here, the training data 11 is the input data required by the AI ​​model training function 20. The inference data 12 is the input data required by the AI ​​model inference function 30.

[0189] The data collection function 10 may be performed by a single entity (eg, UE, RAN node, network node, etc.) or by multiple entities. In this case, the training data 11 and the inference data 12 may be provided to the model training function 20 and the model inference function 30, respectively, from the multiple entities.

[0190] As part of the AI ​​model testing process, the model training function 20 is responsible for performing the AI ​​model training, validation, and testing required to generate model performance metrics. If necessary, the model training function 20 can handle data preparation (e.g., data preprocessing and cleaning, forming, and transformation) based on the training data 11 provided by the data collection function 10.

[0191] Here, model deployment / update 13 is used to initially deploy the trained, verified and tested AI model to the model reasoning function 30 or to provide an updated model to the model reasoning function 30.

[0192] The model reasoning function 30 is responsible for providing AI model reasoning output 16 (e.g., prediction or decision). If applicable, the model reasoning function 30 may also provide model performance feedback 14 to the model training function 20. In addition, if necessary, the model reasoning function 30 may handle data preparation (e.g., data preprocessing and cleaning, forming and transformation) based on the reasoning data 12 provided by the data collection function 10.

[0193] Here, output 16 refers to the inference output of the AI ​​model generated by the model inference function 30, and the details of the inference output may vary depending on the use case.

[0194] When available, model performance feedback 14 can be used to monitor the performance of the AI ​​model. However, the feedback can be omitted.

[0195] The executor function 40 receives the output 16 from the model reasoning function 30 and triggers or performs relevant tasks / operations. The executor function 40 may trigger tasks / operations of other entities (e.g., one or more UEs, one or more RAN nodes, one or more network nodes, etc.) or itself.

[0196] Feedback 15 can be used to derive training data 11 and inference data 12 or monitor the performance of the AI ​​model, its impact on the network, etc.

[0197] The definitions of training, validation, and testing in datasets used in AI / ML can be distinguished as follows:

[0198] -Training data: Training data refers to the dataset used to train the model.

[0199] - Validation data: Validation data refers to a dataset used to validate an already trained model. In other words, validation data is a dataset used to prevent overfitting of a typical training dataset.

[0200] In addition, validation data refers to a data set used to select the best model among multiple trained models during the training process. Therefore, validation data can also be considered as a kind of learning.

[0201] -Test data: Test data refers to the dataset used for final evaluation. Test data is independent of training.

[0202] In the case of the above datasets, it is common to split the training set so that the training data and validation data are divided in a ratio of 8:2 or 7:3. When test data is included, the training data, validation data, and test data can be divided in a ratio of 6:2:2 (training: validation: test).

[0203] According to the capabilities of the AI / ML functions between the BS and the UE, the cooperation level can be defined as follows. It is also possible to combine multiple levels or modify the level by separating any one level.

[0204] Cat 0a) No collaborative framework: AI / ML algorithms are purely implementation-based and do not require changes to the wireless interface.

[0205] Cat 0b) This level involves wireless interfaces suitable for modification of AI / ML algorithms based on efficient implementations, but this level corresponds to a framework without collaboration.

[0206] Cat 1) provides support between nodes to enhance the AI / ML algorithms of individual nodes. This level is applied when the UE receives support (for training, adaptation, etc.) from the gNB and vice versa. At this level, model exchange between network nodes is not required.

[0207] Cat 2) can perform collaborative ML tasks between UE and gNB. This level requires the exchange of AI / ML model commands between network nodes.

[0208] Fig.14 The functions shown may be implemented in a RAN node (e.g., BS, TRP, Central Unit (CU) of a BS), a network node, Operation Administration and Maintenance (OAM) of a network operator, or a UE.

[0209] Alternatively, two or more entities among the RAN node, the network node, the network operator's OAM or the UE may collaborate to implement Fig.14 For example, an entity can perform Fig.14 Some of the functions in the same entity can be performed by another entity, while the remaining functions can be performed by another entity. Fig.14If some of the functions shown are performed by a single entity (e.g., UE, RAN node, or network node), the transmission / provision of data / information between the functions may be omitted. For example, if the model training function 20 and the model inference function 30 are performed by the same entity, the transmission / provision of the model deployment / update 13 and the model performance feedback 14 may be omitted.

[0210] Alternatively, Fig.14 Any of the functions shown may be performed by collaboration between two or more entities in a RAN node, a network node, a network operator's OAM, or a UE. This may be referred to as split AI operation.

[0211] Fig.15 A scenario is shown in which the AI ​​model training function is performed by a network node (e.g., a core network node, a network operator's OAM, etc.), while the AI ​​model reasoning function is performed by a RAN node (e.g., a BS, a TRP, or a CU of a BS).

[0212] Step 1: RAN node 1 and RAN node 2 send input data (i.e., training data) for AI model training to the network node. Here, RAN node 1 and RAN node 2 may send data collected from UEs (e.g., UE measurements related to RSRP, RSRQ or SINR of serving cells and neighboring cells, UE location, speed, etc.) to the network node.

[0213] Step 2: The network nodes train the AI ​​model using the received training data.

[0214] Step 3: The network node deploys / updates the AI ​​model to RAN node 1 and / or RAN node 2. RAN node 1 (and / or RAN node 2) may continue to perform model training based on the received AI model.

[0215] For ease of explanation, it is assumed that the AI ​​model is only deployed / updated to RAN node 1.

[0216] Step 4: RAN node 1 receives input data for AI model inference (i.e., inference data) from UE and RAN node 2.

[0217] Step 5: RAN node 1 uses the received inference data to perform AI model inference to generate output data (e.g., prediction or decision).

[0218] Step 6: If applicable, the RAN node 1 may send model performance feedback to the network node.

[0219] Step 7: RAN node 1, RAN node 2 and UE (or RAN node 1 and UE or RAN node 1 and RAN node 2) perform actions based on the output data. For example, in the case of a load balancing operation, the UE may move from RAN node 1 to RAN node 2.

[0220] Step 8: RAN node 1 and RAN node 2 send feedback information to the network node.

[0221] Fig.16 A scenario is shown in which both the AI ​​model training function and the AI ​​model inference function are performed by a RAN node (e.g., a BS, a TRP, or a CU of a BS).

[0222] Step 1: UE and RAN node 2 send input data (i.e., training data) for AI model training to RAN node 1.

[0223] Step 2: RAN node 1 uses the received training data to train the AI ​​model.

[0224] Step 3: RAN node 1 receives input data for AI model inference (i.e., inference data) from UE and RAN node 2.

[0225] Step 4: RAN node 1 uses the received inference data to perform AI model inference to generate output data (e.g., prediction or decision).

[0226] Step 5: RAN node 1, RAN node 2 and UE (or RAN node 1 and UE or RAN node 1 and RAN node 2) perform actions based on the output data. For example, in the case of a load balancing operation, the UE may move from RAN node 1 to RAN node 2.

[0227] Step 6: RAN node 2 sends feedback information to RAN node 1.

[0228] Fig.17 A scenario is shown in which the AI ​​model training function is performed by a RAN node (e.g., a BS, a TRP, or a CU of a BS), while the AI ​​model inference function is performed by a UE.

[0229] Step 1: The UE sends input data (i.e., training data) for AI model training to the RAN node. Here, the RAN node may collect data (e.g., UE measurements related to RSRP, RSRQ or SINR of the serving cell and neighboring cells, UE location, speed, etc.) from various UEs and / or other RAN nodes.

[0230] Step 2: The RAN node trains the AI ​​model using the received training data.

[0231] Step 3: The RAN node deploys / updates the AI ​​model to the UE. The UE may continue to perform model training based on the received AI model.

[0232] Step 4: The UE receives input data for AI model inference (ie, inference data) from the RAN node (and / or from other UEs).

[0233] Step 5: The UE performs AI model inference using the received inference data to generate output data (e.g., prediction or decision).

[0234] Step 6: If applicable, the UE may send model performance feedback to the RAN node.

[0235] Step 7: The UE and RAN nodes perform actions based on the output data.

[0236] Step 8: The UE sends feedback information to the RAN node.

[0237] AI / ML model management

[0238] With the advancement of computing technology and AI / ML technology, nodes and UEs included in wireless communication networks are becoming more and more intelligent and complex. In particular, due to the intelligence of the network, it is expected that various network decision parameter values ​​(e.g., the transmission and reception power of each BS, the transmission power of each UE, the precoder / beam of each BS / UE, the time / frequency resource allocation of each UE, the duplex mode of each BS, etc.) can be quickly optimized and applied based on various network environment parameters (e.g., the distribution / location of BS, the distribution / location / material of buildings / furniture, the location / moving direction / speed of UE, climate information, etc.).

[0239] As described above, when a network or BS to which AI technology is applied is implemented or a network / BS to which advanced performance is implemented by other means, it is expected that integrated control and / or real-time collaboration between BSs will allow dynamic optimization of interference and traffic. The present disclosure proposes a method for efficiently managing AI / ML when multiple networks / BSs perform cooperative communications based on AI / ML in the context of an evolved network / BS.

[0240] In the following embodiments, each TRP / RRH (Remote Radio Head) may be clearly distinguished by a TRP ID, or may be identified by the following index.

[0241] Each TRP can be distinguished by the CSI-RS resource ID (and / or sequence ID), cell ID, CORESETPoolIndex or (SSB) beam index (or TCI state index) in CSI-ResourceConfig. For each TRP, the AI / ML model (pair) ID and / or function ID can also be distinguished. The ID can be configured, indicated or distinguished with respect to the CSI-RS resource ID (and / or sequence ID), cell ID, CORESETPoolIndex or (SSB) beam index (or TCI state index).

[0242] In particular, for multi-TRP (MTRP) CSI feedback, one or more AI / ML model (pair) IDs and / or one or more function IDs are configured in a specific set (e.g., CSI-RS resource set or AI / ML function set). The BS may sort the AI / ML model ID and / or function ID of each TRP based on a specific index and configure the application order of the corresponding model / function. Alternatively, implicit sorting may be performed based on the above-mentioned CSI-RS ID, cell ID, CORESETPoolIndex, or TCI state index.

[0243] In order to effectively control interference and increase yield, cooperative transmission through multiple TRPs has been introduced in Rel-16 / 17. In the MTRP environment, non-coherent joint transmission (JT) is supported, which does not guarantee the coherence of each TRP, and technology for coherent transmission between TRPs is currently being discussed in Rel-18. The present disclosure proposes a method for effectively managing AI / ML (e.g., model management or lifecycle management) when introducing AI / ML in the MTRP environment.

[0244] Typically, AI / ML models are managed by repeatedly performing some or all of the following processes: data collection, model training, model deployment, model inference, model monitoring, and model updates, which is referred to as AI / ML lifecycle management.

[0245] Table 7 shows terms related to AI / ML models used in standardization.

[0246] [Table 7]

[0247]

[0248]

[0249] Table 8 shows standardized protocols related to AI / ML models.

[0250] [Table 8]

[0251]

[0252] Proposal 1: In MTRP transmission / reception, if the network and / or UE has AI / ML training and / or inference capabilities, The AI / ML model management is performed for each TRP, and the AI / ML model management (update, enable, signaling such as activation, deactivation, switching, etc.

[0253] In Proposal 1, when AI / ML is applied to MTRP transmission / reception (Tx / Rx), AI / ML model management may vary depending on the AI / ML algorithm applied. For example, there may be a representative node or server (separate from the node). The representative node or server may perform training of AI / ML models for all nodes on behalf of the node, and handle the transmission of AI / ML model reasoning / updates to each node. In contrast, in the case of distributed algorithms / federated learning, each node may perform training and / or reasoning of its own algorithm. In addition, each node may exchange auxiliary information (other than input data) for training and / or model fine-tuning. The above auxiliary information may vary depending on the situation in which the AI / ML model is applied. Typically, the auxiliary information may represent the value of the loss function and / or utility function of each node (e.g., mean square error (MSE), spectral efficiency (SE)). In order to facilitate AI / ML model management, an AI / ML model ID may be introduced. For each AI / ML model ID, the AI / ML algorithm, model parameterization, input / output data, collaboration level (e.g., single-sided AI / ML or dual-sided or more-sided AI / ML), loss function, and AI / ML model purpose (e.g., CSI feedback, positioning, beam prediction) can be distinguished and indicated.

[0254] Examples of AI / ML model uses include CSI reporting, beam management, resource allocation, scheduling, and positioning. Based on these examples, the configuration of the AI / ML model (e.g., algorithm, model size / parameters, cost function, input / output data) may vary. Fig.12 The autoencoder shown measures the CSI-RS and encodes the CSI-RS based on AI / ML to report the CSI to the BS. The BS can decode the CSI-RS and retrieve the CSI. Alternatively, the BS may set fewer ports but predict the complete CSI through AI / ML (for example, using 8-port CSI-RS to measure 16-port CSI-RS). The BS can predict future CSI based on multiple past CSI-RS measurement samples. The above examples mainly focus on single TRP (STRP) scenarios. However, for MTRP scenarios, there may be situations where AI / ML is applied separately for each TRP by extending the STRP, and where a central node such as a centralized system trains and manages AI / ML models for all nodes participating in collaborative transmission.

[0255] Proposal 2: Regarding the activation, deactivation, switching and / or updating of AI / ML models mentioned in Proposal 1, the following information is proposed: make.

[0256] -Alternative 1: MAC Control Element (CE) based AI / ML model enabling, deactivation, switching and / or updating.

[0257] -Alternative 2: DCI-based AI / ML model activation, deactivation, switching and / or updating.

[0258] -Alternative 3: Timer-based AI / ML model enabling, deactivation, switching and / or updating.

[0259] In Proposal 2, it is assumed that the UE has AI / ML capabilities (e.g., training and / or reasoning). For example, it is assumed that the UE receives configuration information about an AI / ML model related to CSI feedback from the BS, and receives signaling / download for model deployment signaling or training model parameters from the BS or AI / ML server. In this case, the AI / ML model can be considered to be activated when the signaling reception is completed or after X milliseconds (or time slots or symbols) from the completion of reception. However, since it may be more efficient to apply the AI / ML model according to the scheduling target of the BS, it is proposed that the BS provides (dynamic) activation / deactivation signaling. In Alternative Scheme 1: In the case of MAC-CE-based activation / deactivation, the AI / ML model indicated by the MAC-CE is enabled by a MAC-CE activation command. Here, the meaning of AI / ML model activation is that the action based on the existing traditional operation of the corresponding use is disabled, while the action based on AI / ML is enabled. The example of use may follow the content of Proposal 1. One or more AI / ML models may be enabled or disabled based on MAC-CE. Alternatively, specific functions may be defined for specific configurations, and only the corresponding parts may be enabled or disabled. For example, the function of determining line-of-sight / non-line-of-sight (LoS / NLoS) of a specific signal, which can be used in positioning technologies such as TDOA (time difference of arrival) and multi-RTT (round trip time), can be configured based on AI / ML, and the function can be enabled or disabled. In the case of deactivation, a fast fallback operation can be indicated by a command to deactivate all enabled AI / ML models at once. In the case of alternative scheme 1, the time of applying the AI / ML model activation / deactivation can be assumed to be X milliseconds (or time slots / symbols) (where X≥0) after the time of receiving the MAC-CE command. Alternatively, it is assumed that the waveform is applied after the time slot m Y milliseconds (or time slots / symbols) (where Y≥0) of the ACK transmission for the received MAC-CE (activate / deactivate / switch) command.

[0260] Alternative solution 2 is a DCI-based activation / deactivation / switching method. In the case of DCI-based activation / deactivation / switching, similar to BWP switching, a specific field is defined within the DCI, and the defined field indicates the activation, deactivation or switching of a specific AI / ML model. For example, a 1-bit field may be newly introduced. This field may indicate the activation of a specific AI / ML model or the deactivation of an enabled AI / ML model. Alternatively, this field may indicate the switching of an AI / ML model. Another example is that a specific AI / ML model may be associated (in advance) with a specific BWP, or such an association may be configured by the BS using RRC or other methods. Therefore, the activation / deactivation or switching of the AI / ML model may be performed based on the existing BWP switching. For example, if AI / ML model 1 is configured for BWP1 and AI / ML model 2 is set for BWP2, and if the active BWP changes from 1 to 2, the UE may understand that AI / ML model 1 is deactivated (or suspended) and AI / ML model 2 is enabled. Alternatively, it may be interpreted as the AI / ML model switching from 1 to 2. More specifically, this field becomes valid when a non-fallback DCI (e.g., a DCI format other than DCI format 0_0 or 1_0) is scheduled, so that activation / deactivation or switching can be performed. When a fallback DCI is scheduled, all AI / ML-based operations are considered disabled. During the activation / deactivation / switching of the AI / ML model based on the above-mentioned BWP switching, if the training / update of the specific AI / ML model associated with the dynamically switched BWP is not completed, the activation / deactivation / switching of the specific AI / ML model may not operate correctly. For example, assume that the BS indicates a BWP switch, but the UE is currently updating the AI / ML model or determines that the current AI / ML model is outdated. In this case, a switching gap may be introduced during the BWP switching time. Alternatively, immediately after the BWP switch, the UE may be instructed / configured to report whether a switch to the AI / ML model of the corresponding BWP is performed. If not, the UE may be instructed / configured to report when the switch is completed. Time information may be reported based on UE capabilities or for each specific AI / ML model number or ID. Alternatively, the maximum value among multiple models may be reported as time information. In addition, before the BWP command, the BS may indicate / configure the UE to pre-train / update / deploy the AI / ML model associated with the target BWP instead of the currently active BWP. The BS may consider the above factors to indicate / configure the BWP switching to the UE for the purpose of fast BWP switching and fast model switching or activation.

[0261] In alternative 3, the activation time of a specific AI / ML model may be predefined or configured by the BS. As another example, it may operate based on a timer. For example, upon receiving an activation command (DCI or MAC-CE) for a specific AI / ML model, the timer may be (re)started. If the UE receives separate signaling before the timer expires, the timer may be restarted again. If the UE does not receive any separate signaling before the timer expires, the UE may perform a fallback operation (e.g., a traditional or non-AI / ML-based operation), or the AI / ML model may be considered to be disabled.

[0262] In alternative 2 and / or alternative 3, when activation / deactivation is applied based on DCI, the UE may send an ACK for the change indicator to the BS so that the BS can check whether the UE correctly applies the AI / ML model and / or indicator. To facilitate this operation, an AI / ML-specific BWP may be defined separately. The difference between AI / ML-specific BWP switching and existing BWP switching may include an example in which the frequency position and bandwidth are the same, but the BWP switching gap is defined differently.

[0263] The AI / ML model can be operated in a data-driven manner. This means that if the data statistics change, new operations such as readjustment or retraining may be required. Therefore, when the statistics of the input data used for AI / ML model training / readjustment fluctuate, operations such as model switching or deactivation may be required. When the UE measures, applies and / or reports input data related to a specific AI / ML model, the UE may request the BS to enable, deactivate, switch or update the specific AI / ML model. For example, if the statistics of CSI (e.g., the mean or variance of the channel matrix) changes in the AI / ML model related to CSI feedback, the UE may report the measured value or the change in the measurement to the BS so that the BS can indicate the activation, deactivation, switching or update of the AI / ML model. Alternatively, as described above, the UE may send an operation request to the BS. In the case of AI / ML model update or fine-tuning, the BS may perform operations irregularly. Alternatively, for periodic / quasi-periodic management, the BS may configure the period / offset and perform AI / ML model updates or fine-tuning.

[0264] When the UE requests activation, deactivation, switching or updating, the report can be determined in an event-triggered manner. That is, the report is determined based on the value configured according to a specific threshold defined by the BS. Alternatively, for network-triggered operations, the UE may periodically or quasi-periodically report statistics related to measurements / observations of a specific AI / ML model (e.g., the mean or variance of the channel matrix). Alternatively, the UE may directly determine whether to activate, deactivate, switch or update the AI / ML model and notify the BS of the determination. In this case, the BS may discard the value before the activation, deactivation, switching or updating of the AI / ML model and only use subsequent values ​​for scheduling.

[0265] As another example, a method of associating the activation / deactivation / switching period of the AI / ML model with a specific (periodic / semi-persistent) beam / CSI reporting configuration (e.g., CSI-ReportConfig). Alternatively, the period may be configured / indicated as an integer multiple, allowing more flexible control. For example, if AI / ML-based beam reporting / CSI reporting is performed and the number of reports (e.g., AI / ML-specific reporting number) is defined, the BS (or management server) and the UE may understand each other whether to switch, update or deactivate the AI / ML model based on a value (or a specific threshold). Alternatively, similar to the above method, the BS (or management server) and the UE may understand each other whether to switch, update or deactivate the AI / ML model based on a specific CSI reporting number (e.g., RI / CQI / PMI) or a threshold. Alternatively, for a specific periodic / semi-persistent beam / CSI reporting configuration corresponding to an update, switching, or activation / deactivation period of the AI / ML model, the BS may indicate the update, switching, or activation / deactivation through RRC, MAC-CE, or DCI based on changes in channel statistics (e.g., variance or mean) of the wireless channel in the time, frequency, and / or spatial domains.

[0266] In this case, if the BS fails to detect or receive a report from the UE, ambiguity may occur between the UE and the BS (or management server). When periodic / semi-persistent reporting is performed via PUSCH, PUSCH may be scheduled in a manner similar to the beam failure request response in beam failure recovery, where a new data indicator (NDI) switching is performed with the same ID as the HARQ process ID related to the reporting PUSCH, thereby sending an ACK for the beam / CSI report.

[0267] During the process of performing activation, deactivation, or switching of an AI / ML model based on BWP switching, if the model training for the dynamically switched BWP is not fully completed, it is necessary to define an operation method. For example, if the BS instructs BWP switching, but if the UE is currently updating the AI / ML model or determines that the current AI / ML model is outdated, the BWP switching time may need to be relaxed. Alternatively, immediately after the BWP switching, the UE may report whether a switch to the AI / ML model of the corresponding BWP is performed. If not, the UE may report when the switch is completed.

[0268] Proposal 3: For smooth handover or radio resource management (RRM), it is possible to configure and indicate information about the target and / or neighbor Information about the AI / ML model / function ID of the community.

[0269] When the UE operates according to a specific AI / ML-based purpose, information (e.g., list, model number, ID, parameterization) about the AI / ML model supported / applied by the target cell may be configured and indicated to facilitate fast handover (HO) or RRM. This is to reduce delays during HO or ensure smooth execution of AI / ML model training / inference during HO. To this end, information about additional model data of the target / neighboring cell may be pre-downloaded or signaled. In addition, the source cell and the target cell may exchange AI / ML model information applicable to a specific UE via the X2 interface. Since it may be redundant to provide detailed information about various AI / ML models, AI / ML information may be exchanged by defining model numbers and / or versions (e.g., types, parameters). Proposal 3 may be implemented by providing model numbers. Since the AI / ML models used or supported may vary depending on the communication operator or supplier, the UE may report whether a specific model is supported as a UE capability. Alternatively, information about whether a specific AI / ML model can be used by different operators / suppliers (e.g., compatibility) may be indicated together.

[0270] As described above, when a network / BS to which AI technology is applied or a network / BS to which advanced performance is achieved (by other means), it is expected that integrated control and / or real-time collaboration between BSs will allow dynamic optimization for interference and services. The present disclosure proposes a resource control and CSI-related signaling method suitable for the above purpose. Although the present disclosure is described based on a 3GPP NR system within the current system architecture, it is apparent that the present disclosure is also applicable to other communication systems.

[0271] Proposal 1: In MTRP transmission / reception, if the network and / or UE has AI / ML training and / or inference capabilities, The AI / ML model management is performed for each TRP, and the AI / ML model management (update, enable, signaling such as activation, deactivation, switching, etc.

[0272] In Proposal 1, when AI / ML is applied to MTRP transmission / reception (Tx / Rx), AI / ML model management may vary depending on the AI / ML algorithm applied. For example, there may be a representative node or server (separate from the node). The representative node or server may perform training of AI / ML models for all nodes on behalf of the node, and handle the transmission of AI / ML model reasoning / updates to each node. In contrast, in the case of distributed algorithms / federated learning, each node may perform training and / or reasoning of its own algorithm. In addition, each node may exchange auxiliary information (other than input data) for training and / or model fine-tuning. The above auxiliary information may vary depending on the situation in which the AI / ML model is applied. Typically, the auxiliary information may represent the value of the loss function and / or utility function of each node (e.g., mean square error (MSE), spectral efficiency (SE)). In order to facilitate AI / ML model management, an AI / ML model ID may be introduced. For each AI / ML model ID, the AI / ML algorithm, model parameterization, input / output data, collaboration level (e.g., single-sided AI / ML or dual-sided or more-sided AI / ML), loss function, and AI / ML model purpose (e.g., CSI feedback, positioning, beam prediction) can be distinguished and indicated.

[0273] Examples of AI / ML model uses include CSI reporting, beam management, resource allocation, scheduling, and positioning. Based on these examples, the configuration of the AI / ML model (e.g., algorithm, model size / parameters, cost function, input / output data) may vary. Fig.12 The autoencoder shown measures the CSI-RS and encodes the CSI-RS based on AI / ML to report the CSI to the BS. The BS can decode the CSI-RS and retrieve the CSI. Alternatively, the BS may set fewer ports but predict the complete CSI through AI / ML (for example, using 8-port CSI-RS to measure 16-port CSI-RS). The BS can predict future CSI based on multiple past CSI-RS measurement samples. The above examples mainly focus on single TRP (STRP) scenarios. However, for MTRP scenarios, there may be situations where AI / ML is applied separately for each TRP by extending the STRP, and where a central node such as a centralized system trains and manages AI / ML models for all nodes participating in collaborative transmission.

[0274] Proposal 2: Regarding the activation, deactivation, switching and / or updating of AI / ML models mentioned in Proposal 1, the following information is proposed: make.

[0275] -Alternative 1: MAC Control Element (CE) based AI / ML model enabling, deactivation, switching and / or updating.

[0276] -Alternative 2: DCI-based AI / ML model activation, deactivation, switching and / or updating.

[0277] -Alternative 3: Timer-based AI / ML model enabling, deactivation, switching and / or updating.

[0278] In Proposal 2, it is assumed that the UE has AI / ML capabilities (e.g., training and / or reasoning). For example, it is assumed that the UE receives configuration information about an AI / ML model related to CSI feedback from the BS, and receives signaling / download for model deployment signaling or training model parameters from the BS or AI / ML server. In this case, the AI / ML model can be considered to be activated when the signaling reception is completed or after X milliseconds (or time slots or symbols) from the completion of reception. However, since it may be more efficient to apply the AI / ML model according to the scheduling target of the BS, it is proposed that the BS provides (dynamic) activation / deactivation signaling. In Alternative Scheme 1: In the case of MAC-CE-based activation / deactivation, the AI / ML model indicated by the MAC-CE is enabled by a MAC-CE activation command. Here, the meaning of AI / ML model activation is that the action based on the existing traditional operation of the corresponding use is disabled, while the action based on AI / ML is enabled. The example of use may follow the content of Proposal 1. One or more AI / ML models may be enabled or disabled based on MAC-CE. Alternatively, specific functions may be defined for specific configurations, and only the corresponding parts may be enabled or disabled. For example, the function of determining line-of-sight / non-line-of-sight (LoS / NLoS) of a specific signal, which can be used in positioning technologies such as TDOA (time difference of arrival) and multi-RTT (round trip time), can be configured based on AI / ML, and the function can be enabled or disabled. In the case of deactivation, a fast fallback operation can be indicated by a command to deactivate all enabled AI / ML models at once. In the case of alternative scheme 1, the time of applying the AI / ML model activation / deactivation can be assumed to be X milliseconds (or time slots / symbols) (where X≥0) after the time of receiving the MAC-CE command. Alternatively, it is assumed that the waveform is applied after the time slot m Y milliseconds (or time slots / symbols) (where Y≥0) of the ACK transmission for the received MAC-CE (activate / deactivate / switch) command.

[0279] Alternative solution 2 is a DCI-based activation / deactivation / switching method. In the case of DCI-based activation / deactivation / switching, similar to BWP switching, a specific field is defined within the DCI, and the defined field indicates the activation, deactivation or switching of a specific AI / ML model. For example, a 1-bit field may be newly introduced. This field may indicate the activation of a specific AI / ML model or the deactivation of an enabled AI / ML model. Alternatively, this field may indicate the switching of an AI / ML model. Another example is that a specific AI / ML model may be associated (in advance) with a specific BWP, or such an association may be configured by the BS using RRC or other methods. Therefore, the activation / deactivation or switching of the AI / ML model may be performed based on the existing BWP switching. For example, if AI / ML model 1 is configured for BWP1 and AI / ML model 2 is set for BWP2, and if the active BWP changes from 1 to 2, the UE may understand that AI / ML model 1 is deactivated (or suspended) and AI / ML model 2 is enabled. Alternatively, it may be interpreted as the AI / ML model switching from 1 to 2. More specifically, this field becomes valid when a non-fallback DCI (e.g., a DCI format other than DCI format 0_0 or 1_0) is scheduled, so that activation / deactivation or switching can be performed. When a fallback DCI is scheduled, all AI / ML-based operations are considered disabled. During the activation / deactivation / switching of the AI / ML model based on the above-mentioned BWP switching, if the training / update of the specific AI / ML model associated with the dynamically switched BWP is not completed, the activation / deactivation / switching of the specific AI / ML model may not operate correctly. For example, assume that the BS indicates a BWP switch, but the UE is currently updating the AI / ML model or determines that the current AI / ML model is outdated. In this case, a switching gap may be introduced during the BWP switching time. Alternatively, immediately after the BWP switch, the UE may be instructed / configured to report whether a switch to the AI / ML model of the corresponding BWP is performed. If not, the UE may be instructed / configured to report when the switch is completed. Time information may be reported based on UE capabilities or for each specific AI / ML model number or ID. Alternatively, the maximum value among multiple models may be reported as time information. In addition, before the BWP command, the BS may indicate / configure the UE to pre-train / update / deploy the AI / ML model associated with the target BWP instead of the currently active BWP. The BS may consider the above factors to indicate / configure the BWP switching to the UE for the purpose of fast BWP switching and fast model switching or activation.

[0280] In alternative 3, the activation time of a specific AI / ML model may be predefined or configured by the BS. As another example, it may operate based on a timer. For example, upon receiving an activation command (DCI or MAC-CE) for a specific AI / ML model, the timer may be (re)started. If the UE receives separate signaling before the timer expires, the timer may be restarted again. If the UE does not receive any separate signaling before the timer expires, the UE may perform a fallback operation (e.g., a traditional or non-AI / ML-based operation), or the AI / ML model may be considered to be disabled.

[0281] In alternative 2 and / or alternative 3, when activation / deactivation is applied based on DCI, the UE may send an ACK for the change indicator to the BS so that the BS can check whether the UE correctly applies the AI / ML model and / or indicator. To facilitate this operation, an AI / ML-specific BWP may be defined separately. The difference between AI / ML-specific BWP switching and existing BWP switching may include an example in which the frequency position and bandwidth are the same, but the BWP switching gap is defined differently.

[0282] The AI / ML model can be operated in a data-driven manner. This means that if the data statistics change, new operations such as readjustment or retraining may be required. Therefore, when the statistics of the input data used for AI / ML model training / readjustment fluctuate, operations such as model switching or deactivation may be required. When the UE measures, applies and / or reports input data related to a specific AI / ML model, the UE may request the BS to enable, deactivate, switch or update the specific AI / ML model. For example, if the statistics of CSI (e.g., the mean or variance of the channel matrix) changes in the AI / ML model related to CSI feedback, the UE may report the measured value or the change in the measurement to the BS so that the BS can indicate the activation, deactivation, switching or update of the AI / ML model. Alternatively, as described above, the UE may send an operation request to the BS. In the case of AI / ML model update or fine-tuning, the BS may perform operations irregularly. Alternatively, for periodic / quasi-periodic management, the BS may configure the period / offset and perform AI / ML model updates or fine-tuning.

[0283] When the UE requests activation, deactivation, switching or updating, the report can be determined in an event-triggered manner. That is, the report is determined based on the value configured according to a specific threshold defined by the BS. Alternatively, for network-triggered operations, the UE may periodically or quasi-periodically report statistics related to measurements / observations of a specific AI / ML model (e.g., the mean or variance of the channel matrix). Alternatively, the UE may directly determine whether to activate, deactivate, switch or update the AI / ML model and notify the BS of the determination. In this case, the BS may discard the value before the activation, deactivation, switching or updating of the AI / ML model and only use subsequent values ​​for scheduling.

[0284] As another example, a method of associating the activation / deactivation / switching period of the AI / ML model with a specific (periodic / semi-persistent) beam / CSI reporting configuration (e.g., CSI-ReportConfig). Alternatively, the period may be configured / indicated as an integer multiple, allowing more flexible control. For example, if AI / ML-based beam reporting / CSI reporting is performed and the number of reports (e.g., AI / ML-specific reporting number) is defined, the BS (or management server) and the UE may understand each other whether to switch, update or deactivate the AI / ML model based on a value (or a specific threshold). Alternatively, similar to the above method, the BS (or management server) and the UE may understand each other whether to switch, update or deactivate the AI / ML model based on a specific CSI reporting number (e.g., RI / CQI / PMI) or a threshold. Alternatively, for a specific periodic / semi-persistent beam / CSI reporting configuration corresponding to an update, switching, or activation / deactivation period of the AI / ML model, the BS may indicate the update, switching, or activation / deactivation through RRC, MAC-CE, or DCI based on changes in channel statistics (e.g., variance or mean) of the wireless channel in the time, frequency, and / or spatial domains.

[0285] In this case, if the BS fails to detect or receive a report from the UE, ambiguity may occur between the UE and the BS (or management server). When periodic / semi-persistent reporting is performed via PUSCH, PUSCH may be scheduled in a manner similar to the beam failure request response in beam failure recovery, where a new data indicator (NDI) switching is performed with the same ID as the HARQ process ID related to the reporting PUSCH, thereby sending an ACK for the beam / CSI report.

[0286] During the process of performing activation, deactivation, or switching of an AI / ML model based on BWP switching, if the model training for the dynamically switched BWP is not fully completed, it is necessary to define an operation method. For example, if the BS instructs BWP switching, but if the UE is currently updating the AI / ML model or determines that the current AI / ML model is outdated, the BWP switching time may need to be relaxed. Alternatively, immediately after the BWP switching, the UE may report whether a switch to the AI / ML model of the corresponding BWP is performed. If not, the UE may report when the switch is completed.

[0287] Proposal 3: For smooth handover or radio resource management (RRM), it is possible to configure and indicate information about the target and / or neighbor Information about the AI / ML model / function ID of the community.

[0288] When the UE operates according to a specific AI / ML-based purpose, information (e.g., list, model number, ID, parameterization) about the AI / ML model supported / applied by the target cell may be configured and indicated to facilitate fast handover (HO) or RRM. This is to reduce delays during HO or ensure smooth execution of AI / ML model training / inference during HO. To this end, information about additional model data of the target / neighboring cell may be pre-downloaded or signaled. In addition, the source cell and the target cell may exchange AI / ML model information applicable to a specific UE via the X2 interface. Since it may be redundant to provide detailed information about various AI / ML models, AI / ML information may be exchanged by defining model numbers and / or versions (e.g., types, parameters). Proposal 3 may be implemented by providing model numbers. Since the AI / ML models used or supported may vary depending on the communication operator or supplier, the UE may report whether a specific model is supported as a UE capability. Alternatively, information about whether a specific AI / ML model can be used by different operators / suppliers (e.g., compatibility) may be indicated together.

[0289] As described above, when a network / BS to which AI technology is applied or a network / BS to which advanced performance is achieved (by other means), it is expected that integrated control and / or real-time collaboration between BSs will allow dynamic optimization for interference and services. The present disclosure proposes a resource control and CSI-related signaling method suitable for the above purpose. Although the present disclosure is described based on a 3GPP NR system within the current system architecture, it is apparent that the present disclosure is also applicable to other communication systems.

[0290] exist Fig. 22In the wireless devices 100 and 200, various elements, components, units / parts and / or modules in the wireless devices 100 and 200 may all be connected to each other through a wired interface, or at least a part thereof may be wirelessly connected through the communication unit 110. For example, in each of the wireless devices 100 and 200, the control unit 120 and the communication unit 110 may be wired, and the control unit 120 and the first unit (e.g., 130 and 140) may be wirelessly connected through the communication unit 110. The various elements, components, units / parts and / or modules within the wireless devices 100 and 200 may also include one or more elements. For example, the control unit 120 may be configured by a collection of one or more processors. As an example, the control unit 120 may be configured by a collection of a communication control processor, an application processor, an electronic control unit (ECU), a graphics processing unit and a memory control processor. As another example, the memory unit 130 may be configured by a random access memory (RAM), a dynamic RAM (DRAM), a read-only memory (ROM)), a flash memory, a volatile memory, a non-volatile memory and / or a combination thereof.

[0291] Fig.23 A vehicle or an autonomous vehicle applied to the present disclosure is shown. The vehicle or the autonomous vehicle can be implemented by a mobile robot, a car, a train, a manned / unmanned aerial vehicle (AV), a ship, etc.

[0292] Reference Fig.23 , the vehicle or autonomous driving vehicle 100 may include an antenna unit 108, a communication unit 110, a control unit 120, a drive unit 140a, a power supply unit 140b, a sensor unit 140c, and an autonomous driving unit 140d. The antenna unit 108 may be configured as a part of the communication unit 110. Blocks 110 / 130 / 140a to 140d correspond to Fig. 22 Block 110 / 130 / 140.

[0293] The communication unit 110 may send and receive signals (e.g., data and control signals) to and from external devices such as other vehicles, BSs (e.g., gNBs and roadside units), and servers. The control unit 120 may perform various operations by controlling elements of the vehicle or autonomous vehicle 100. The control unit 120 may include an electronic control unit (ECU). The drive unit 140a may enable the vehicle or autonomous vehicle 100 to travel on a road. The drive unit 140a may include an engine, a motor, a power system, wheels, brakes, a steering device, etc. The power supply unit 140b may supply power to the vehicle or autonomous vehicle 100, and include a wired / wireless charging circuit, a battery, etc. The sensor unit 140c may obtain vehicle status, surrounding environment information, user information, etc. The sensor unit 140c may include an inertial measurement unit (IMU) sensor, a collision sensor, a wheel sensor, a speed sensor, a slope sensor, a weight sensor, a heading sensor, a position module, a vehicle forward / backward sensor, a battery sensor, a fuel sensor, a tire sensor, a steering sensor, a temperature sensor, a depth sensor, an ultrasonic sensor, a lighting sensor, a pedal position sensor, etc. The autonomous driving unit 140d may implement a technology for maintaining a lane in which the vehicle is traveling, a technology for automatically adjusting a speed (e.g., adaptive cruise control), a technology for autonomously traveling along a determined path, a technology for traveling by automatically setting a path if a destination is set, and the like.

[0294] For example, the communication unit 110 may receive map data, traffic information data, etc. from an external server. The autonomous driving unit 140d may generate an autonomous driving path and a driving plan from the obtained data. The control unit 120 may control the drive unit 140a so that the vehicle or the autonomous driving vehicle 100 may move along the autonomous driving path according to the driving plan (e.g., speed / direction control). In the middle of autonomous driving, the communication unit 110 may aperiodically / periodically obtain the latest traffic information data from the external server and obtain surrounding traffic information data from neighboring vehicles. In the middle of autonomous driving, the sensor unit 140c may obtain vehicle status and / or surrounding environment information. The autonomous driving unit 140d may update the autonomous driving path and driving plan based on the newly obtained data / information. The communication unit 110 may transmit information about the vehicle position, autonomous driving path, and / or driving plan to an external server. The external server may predict traffic information data based on information collected from the vehicle or autonomous driving vehicle using AI technology, etc., and provide the predicted traffic information data to the vehicle or autonomous driving vehicle.

[0295] Fig.24 is a diagram illustrating a DRX operation of a UE according to an embodiment of the present disclosure.

[0296] The UE may perform DRX operation in the process and / or method described / proposed above. A UE configured with DRX may reduce power consumption by discontinuously receiving DL signals. DRX may be performed in the RRC_IDLE state, the RRC_INACTIVE state, and the RRC_CONNECTED state. The UE performs DRX in the RRC_IDLE state and the RRC_INACTIVE state to discontinuously receive paging signals. DRX in the RRC_CONNECTED state (RRC_CONNECTED DRX) will be described below.

[0297] Reference Fig.24 , the DRX cycle includes an on-duration and a DRX opportunity. The DRX cycle defines the time interval between periodic repetitions of the on-duration. The on-duration is the time period during which the UE monitors the PDCCH. When the UE is configured with DRX, the UE performs PDCCH monitoring during the on-duration. When the UE successfully detects the PDCCH during the PDCCH monitoring, the UE starts the inactivity timer and remains awake. On the contrary, when the UE fails to detect any PDCCH during the PDCCH monitoring, the UE transitions to a sleep state after the on-duration. Therefore, when DRX is configured, PDCCH monitoring / reception may be performed discontinuously in the time domain in the process and / or method described / proposed above. For example, when DRX is configured, the PDCCH reception timing (e.g., a time slot with a PDCCH SS) may be configured discontinuously according to the DRX configuration in the present disclosure. On the contrary, when DRX is not configured, PDCCH monitoring / reception may be performed continuously in the time domain. For example, when DRX is not configured, the PDCCH reception timing (e.g., a time slot with a PDCCH SS) may be configured continuously in the present disclosure. Regardless of whether DRX is configured, PDCCH monitoring may be restricted during time periods configured as measurement gaps.

[0298] Table 9 describes the DRX operation of the UE (in the RRC_CONNECTED state). Referring to Table 9, DRX configuration information is received through high-layer signaling (e.g., RRC signaling), and DRX on / off is controlled by a DRX command from the MAC layer. Once DRX is configured, the UE may perform PDCCH monitoring discontinuously while executing the above-described / proposed procedures and / or methods.

[0299] [Table 9]

[0300]

[0301] MAC-CellGroupConfig includes configuration information required to configure MAC parameters for a cell group. MAC-CellGroupConfig may also include DRX configuration information. For example, when defining DRX, MAC-CellGroupConfig may include the following information.

[0302] -drx-OnDurationTimer value: defines the duration of the start period of the DRX cycle.

[0303] - Value of drx-InactivityTimer: defines the duration of the time period that the UE wakes up after detecting a PDCCH opportunity indicating a PDCCH for initial UL or DL ​​data.

[0304] - Value of drx-HARQ-RTT-TimerDL: defines the duration of the maximum time period after receiving a DL initial transmission until a DL retransmission is received.

[0305] - Value of drx-HARQ-RTT-TimerDL: defines the duration of the maximum time period after receiving a UL initial transmission grant until receiving a UL retransmission grant.

[0306] -drx-LongCycleStartOffset: defines the duration and start time of the DRX cycle.

[0307] -drx-ShortCycle (optional): defines the duration of the short DRX cycle.

[0308] When any one of drx-OnDurationTimer, drx-InactivityTimer, drx-HARQ-RTT-TimerDL and drx-HARQ-RTT-TimerDL is running, the UE performs PDCCH monitoring at each PDCCH opportunity and remains in the awake state.

[0309] The above-mentioned embodiments correspond to the combination of elements and features of the present disclosure in a prescribed form. And, unless explicitly mentioned, each element or feature may be regarded as selective. Each element or feature may be implemented in a form that is not combined with other elements or features. In addition, it is possible to implement the embodiments of the present disclosure by partially combining elements and / or features together. The order of operations described for each embodiment of the present disclosure may be modified. Some configurations or features of an embodiment may be included in another embodiment, or may replace the corresponding configuration or features of another embodiment. And, it is obvious to understand that the embodiment is configured by combining claims that do not have a clear reference relationship in the attached claims, or may be included as a new claim by modification after submitting the application.

[0310] Those skilled in the art will appreciate that the present disclosure may be implemented in other specific forms other than those described herein without departing from the spirit and essential characteristics of the present disclosure. Therefore, the above-described embodiments should be interpreted as being illustrative in all respects, rather than restrictive. The scope of the present disclosure should be determined by the appended claims and their legal equivalents, rather than by the above description, and all changes falling within the meaning and equivalent scope of the appended claims are intended to be covered therein.

[0311] Industrial Applicability

[0312] The present disclosure is applicable to UE, BS or other devices in a wireless mobile communication system.

Claims

1. A method for operating a user equipment UE in a wireless communication system, the method comprising the following steps: Receive AI / ML related configuration information; configuring an AI / ML model for at least one transmission / reception point TRP based on the received AI / ML related configuration information; and Receive information used to manage configured AI / ML models, Wherein, based on the UE supporting multiple TRPs, the information for managing the AI / ML model separately indicates activation, deactivation, update or switching of the AI / ML model for each of the multiple TRPs.

2. The method according to claim 1, wherein: The information for managing the AI / ML model is received via a medium access control MAC control element CE.

3. The method according to claim 1, wherein: The information for managing the AI / ML model is received through downlink control information DCI.

4. The method according to claim 1, wherein: The AI / ML related configuration information is received through high-level signaling.

5. The method according to claim 1, wherein: A timer related to activation, deactivation, updating, or switching of the AI / ML model is configured separately for each of the plurality of TRPs.

6. The method according to claim 5, wherein: The timer is configured through network signaling.

7. The method according to claim 5, wherein: The timer is restarted based on receiving the information for managing the AI / ML model.

8. The method according to claim 1, wherein: The information for managing the AI / ML model includes a related AI / ML model identifier ID or a related AI / ML function ID.

9. The method according to claim 1, wherein: The indicated activation, deactivation, updating, or switching of the AI / ML model is performed based on a reception timing of the information for managing the AI / ML model or a transmission timing of an acknowledgment ACK for the information for managing the AI / ML model. 10 . A computer-readable recording medium having recorded thereon a program for executing the method according to claim 1 .

11. An apparatus for wireless communication, the apparatus comprising: a memory configured to store instructions; as well as a processor configured to perform operations by executing the instructions, The operation of the processor includes: Receive AI / ML related configuration information; configuring an AI / ML model for at least one transmission / reception point TRP based on the received AI / ML related configuration information; and Receive information used to manage configured AI / ML models, Wherein, based on the device supporting multiple TRPs, the information for managing the AI / ML model separately indicates the activation, deactivation, update or switching of the AI / ML model for each of the multiple TRPs.

12. The device according to claim 11, further comprising: Transceiver, The device is a user equipment UE.

13. A method for operating a base station BS in a wireless communication system, the method comprising the following steps: Sending AI / ML related configuration information for configuring an artificial intelligence / machine learning AI / ML model for at least one transmitting / receiving point TRP to a user equipment UE; and sending information for managing the AI / ML model configured in the UE, Wherein, based on the UE supporting multiple TRPs, the information for managing the AI / ML model separately indicates activation, deactivation, update or switching of the AI / ML model for each of the multiple TRPs.

14. A base station BS for wireless communication, the BS comprising: Transceiver; as well as A processor configured to control the transceiver: Sending AI / ML related configuration information for configuring an artificial intelligence / machine learning AI / ML model for at least one transmission reception point TRP to a user equipment UE; and sending information for managing the AI / ML model configured in the UE, Wherein, based on the UE supporting multiple TRPs, the information for managing the AI / ML model separately indicates activation, deactivation, update or switching of the AI / ML model for each of the multiple TRPs.

Images