WO2025009901A1 - 인공지능 및/또는 머신러닝 기반 통신을 위한 모델 훈련 방법 및 장치 - Google Patents
인공지능 및/또는 머신러닝 기반 통신을 위한 모델 훈련 방법 및 장치 Download PDFInfo
- Publication number
- WO2025009901A1 WO2025009901A1 PCT/KR2024/009443 KR2024009443W WO2025009901A1 WO 2025009901 A1 WO2025009901 A1 WO 2025009901A1 KR 2024009443 W KR2024009443 W KR 2024009443W WO 2025009901 A1 WO2025009901 A1 WO 2025009901A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- data
- terminal
- model
- message
- information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/098—Distributed learning, e.g. federated learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L5/00—Arrangements affording multiple use of the transmission path
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B7/00—Radio transmission systems, i.e. using radiation field
- H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
- H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
- H04B7/06—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station
- H04B7/0686—Hybrid systems, i.e. switching and simultaneous transmission
- H04B7/0695—Hybrid systems, i.e. switching and simultaneous transmission using beam selection
Definitions
- This specification relates to wireless communications applicable to 5G NR, 5G-Advanced and 6G.
- next-generation 5G system which is an improved wireless broadband communication system than the existing LTE system
- NewRAT communication scenarios are divided into Enhanced Mobile BroadBand (eMBB) / Ultra-reliability and low-latency communication (URLLC) / Massive Machine-Type Communications (mMTC).
- eMBB Enhanced Mobile BroadBand
- URLLC Ultra-reliability and low-latency communication
- mMTC Massive Machine-Type Communications
- eMBB is a next-generation mobile communication scenario with the characteristics of High Spectrum Efficiency, High User Experienced Data Rate, and High Peak Data Rate
- URLLC is a next-generation mobile communication scenario with the characteristics of Ultra Reliable, Ultra Low Latency, and Ultra High Availability (e.g., V2X, Emergency Service, Remote Control)
- mMTC is a next-generation mobile communication scenario with the characteristics of Low Cost, Low Energy, Short Packet, and Massive Connectivity (e.g., IoT).
- An object of the present specification is to provide a method and device for data collection between a terminal and a network when performing AI/ML model training on the terminal side to efficiently perform beam management using AI/ML in a wireless communication system.
- One embodiment of the present specification provides a method for a wireless communication system in which a terminal receives data for AI/ML model training from a base station and performs AI/ML model training based on the received data. After performing the AI/ML model training, the terminal transmits a first message to the base station instructing termination of data collection.
- one embodiment of the present specification provides a method in which, in a wireless communication system, a base station transmits data for AI/ML model training to a terminal. After performing AI/ML model training, the base station receives a first message from the terminal instructing termination of data collection.
- one embodiment of the present specification provides a wireless communication system, comprising at least one processor, and at least one memory storing instructions and being operably electrically connectable to the at least one processor, wherein operations performed based on the instructions being executed by the at least one processor include: receiving data for AI/ML model training from a base station, and performing AI/ML model training based on the received data. After performing the AI/ML model training, a communication device is provided that transmits a first message to the base station instructing termination of data collection.
- one embodiment of the present specification provides a wireless communication system, comprising at least one processor, and at least one memory storing instructions and being operably electrically connectable to the at least one processor, wherein the operations performed based on the instructions being executed by the at least one processor include: transmitting data for AI/ML model training to a terminal. After performing the AI/ML model training, a base station is provided that receives a first message from the terminal instructing termination of data collection.
- the base station transmits setting information for receiving the above data to the terminal, and the terminal can receive the same.
- the setting information can include event information related to the end of collection of the data, and the first message can be transmitted based on the event information.
- the above event information may include at least one of threshold information related to the performance of the AI/ML model and time information for the end of training of the AI/ML model.
- the above first message may be for disabling transmission of the data.
- the terminal transmits a second message requesting transmission of the data to the base station, and the base station can receive it. Based on the second message, the setting information is transmitted from the base station to the terminal, and the terminal can receive it.
- the second message may include at least one of time information and data amount information for the end of the AI/ML model training.
- Figure 1 is a diagram illustrating a wireless communication system.
- Figure 2 illustrates the structure of a radio frame used in NR.
- FIGS. 3A to 3C are exemplary diagrams showing exemplary architectures for wireless communication services.
- Figure 4 illustrates the slot structure of an NR frame.
- Figure 5 shows examples of subframe types in NR.
- Figure 6 illustrates the structure of a self-contained slot.
- Figure 7 shows an example of initial beam measurement and selection in NR.
- Figure 8 shows an example of an initial connection procedure between a terminal and a base station in NR.
- Figure 9 shows an example of candidate beam settings in NR.
- Figures 10a to 10c illustrate three procedures for beam management in NR.
- Figures 11a to 11c illustrate examples of beam reporting procedures in NR.
- Figures 12a and 12b show examples of beam measurement and spatial domain beam prediction using AI/ML.
- Figure 13 shows an example of temporal domain beam prediction using AI/ML.
- Figure 14 shows an operation method of a terminal according to one embodiment of the present specification.
- Figure 15 illustrates an operation method of a terminal according to another embodiment of the present specification.
- Figure 16 illustrates an operation method of a terminal according to another embodiment of the present specification.
- Figures 17a and 17b illustrate procedures of a terminal and a base station according to embodiments of the present specification.
- Figure 18 shows a procedure of a terminal and a base station for event-based model training according to one embodiment of the present specification.
- Figure 19 illustrates a procedure of a terminal and a base station for event-based model training according to another embodiment of the present specification.
- Figure 20 illustrates a procedure of a terminal and a base station for event-based model training according to another embodiment of the present specification.
- Figure 21 shows a procedure of a terminal and a base station for event-based model training according to another embodiment of the present specification.
- Figure 24 illustrates a device according to one embodiment of the present specification.
- Figure 25 is a block diagram showing the configuration of a terminal according to one embodiment of the present specification.
- FIG. 26 illustrates a block diagram of a processor in which the disclosure of the present specification is implemented.
- FIG. 27 is a block diagram showing in detail the transceiver of the first device illustrated in FIG. 24 or the transceiver unit of the device illustrated in FIG. 25.
- first, second, etc. used in this specification may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another.
- first component may be referred to as the second component
- second component may also be referred to as the first component.
- a component When it is said that a component is connected or connected to another component, it may be directly connected or connected to that other component, but there may be other components in between. On the other hand, when it is said that a component is directly connected or connected to another component, it should be understood that there are no other components in between.
- a or B can mean “only A,” “only B,” or “both A and B.” In other words, as used herein, “A or B” can be interpreted as “A and/or B.” For example, as used herein, “A, B or C” can mean “only A,” “only B,” “only C,” or “any combination of A, B and C.”
- a slash (/) or a comma can mean “and/or.”
- A/B can mean “A and/or B.”
- A/B can mean “only A,” “only B,” or “both A and B.”
- A, B, C can mean “A, B, or C.”
- At least one of A and B can mean “only A”, “only B” or “both A and B”. Additionally, as used herein, the expressions “at least one of A or B” or “at least one of A and/or B” can be interpreted identically to “at least one of A and B”.
- “at least one of A, B and C” can mean “only A,” “only B,” “only C,” or “any combination of A, B and C.” Additionally, “at least one of A, B or C” or “at least one of A, B and/or C” can mean “at least one of A, B and C.”
- control information when it is indicated as “control information (PDCCH)”, “PDCCH (Physical Downlink Control Channel)” may be suggested as an example of “control information”.
- control information in this specification is not limited to “PDCCH”, and “PDDCH” may be suggested as an example of “control information”.
- PDCCH Physical Downlink Control Channel
- PDCCH Physical Downlink Control Channel
- the attached drawing illustrates an example of a UE (User Equipment), the illustrated UE may also be referred to as a terminal, an ME (Mobile Equipment), etc.
- the UE may be a portable device such as a laptop, a mobile phone, a PDA, a smart phone, a multimedia device, etc., or a non-portable device such as a PC or a vehicle-mounted device.
- UE is used as an example of a device capable of wireless communication (e.g., a wireless communication device, a wireless device, or a wireless device).
- the operations performed by the UE can be performed by any device capable of wireless communication.
- a device capable of wireless communication may also be referred to as a wireless communication device, a wireless device, or a wireless device.
- base station generally refers to a fixed station that communicates with wireless devices, and can be used as a comprehensive term that includes eNodeB (evolved-NodeB), eNB (evolved-NodeB), BTS (Base Transceiver System), Access Point, gNB (Next generation NodeB), RRH (remote radio head), TP (transmission point), RP (reception point), relay, etc.
- eNodeB evolved-NodeB
- eNB evolved-NodeB
- BTS Base Transceiver System
- Access Point gNB (Next generation NodeB)
- RRH remote radio head
- TP transmission point
- RP reception point
- relay etc.
- LTE long term evolution
- LTE-A LTE-Advanced
- 5G 5th generation
- ITU proposes three usage scenarios: eMBB (enhanced Mobile BroadBand), mMTC (massive Machine Type Communication), and URLLC (Ultra Reliable and Low Latency Communications).
- eMBB enhanced Mobile BroadBand
- mMTC massive Machine Type Communication
- URLLC Ultra Reliable and Low Latency Communications
- URLLC is for use scenarios that require high reliability and low latency.
- services such as autonomous driving, factory automation, and augmented reality require high reliability and low latency (e.g., latency below 1ms).
- the current latency of 4G (LTE) is statistically 21-43ms (best 10%), 33-75ms (median). This is insufficient to support services requiring latency below 1ms.
- eMBB use scenarios are for use scenarios that require mobile ultra-wideband.
- the 5th generation mobile communication system can support higher capacity than the current 4G LTE, increase the density of mobile broadband users, and support D2D (Device to Device), high stability, and MTC (Machine type communication).
- 5G research and development also aims for lower standby time and lower battery consumption than the 4G mobile communication system to better implement the Internet of Things.
- a new radio access technology (New RAT or NR) can be proposed.
- the NR frequency band can be defined by two types of frequency ranges (FR1, FR2).
- the numerical values of the frequency ranges can be changed, and for example, the two types of frequency ranges (FR1, FR2) can be as shown in Table 1 below.
- FR1 can mean “sub 6GHz range”
- FR2 can mean “above 6GHz range” and can be called millimeter wave (mmW).
- mmW millimeter wave
- FR1 can include a band of 410 MHz to 7125 MHz as shown in Table 1. That is, FR1 can include a frequency band of 6 GHz (or 5850, 5900, 5925 MHz, etc.) or higher.
- the frequency band of 6 GHz (or 5850, 5900, 5925 MHz, etc.) or higher included in FR1 can include an unlicensed band.
- the unlicensed band can be used for various purposes, for example, it can be used for communication for vehicles (e.g., autonomous driving).
- 3GPP-based communication standards define downlink physical channels corresponding to resource elements carrying information originating from upper layers, and downlink physical signals corresponding to resource elements used by the physical layer but not carrying information originating from upper layers.
- a physical downlink shared channel (PDSCH), a physical broadcast channel (PBCH), a physical multicast channel (PMCH), a physical control format indicator channel (PCFICH), a physical downlink control channel (PDCCH), and a physical hybrid ARQ indicator channel (PHICH) are defined as downlink physical channels, and a reference signal and a synchronization signal are defined as downlink physical signals.
- a physical uplink shared channel (PUSCH), a physical uplink control channel (PUCCH), and a physical random access channel (PRACH) are defined as uplink physical channels
- a demodulation reference signal (DMRS) for uplink control/data signals
- a sounding reference signal (SRS) used for uplink channel measurement are defined.
- PDCCH Physical Downlink Control CHannel
- PCFICH Physical Control Format Indicator CHannel
- PHICH Physical Hybrid automatic retransmit request Indicator CHannel
- PDSCH Physical Downlink Shared CHannel
- DCI Downlink Control Information
- CFI Control Format Indicator
- Downlink ACK/NACK ACKnowlegement/Negative ACK
- PUCCH Physical Uplink Control CHannel
- PUSCH Physical Uplink Shared CHannel
- PRACH Physical Random Access CHannel
- UCI Uplink Control Information
- the wireless communication system includes at least one base station (BS).
- the BS is divided into a gNodeB (or gNB) (20a) and an eNodeB (or eNB) (20b).
- the gNB (20a) supports 5th generation mobile communication.
- the eNB (20b) supports 4th generation mobile communication, i.e., LTE (long term evolution).
- Each base station (20a and 20b) provides communication services for a specific geographic area (generally called a cell) (20-1, 20-2, 20-3).
- the cell may be further divided into a number of areas (called sectors).
- a UE usually belongs to one cell, and the cell to which the UE belongs is called a serving cell.
- a base station that provides communication services for a serving cell is called a serving BS. Since a wireless communication system is a cellular system, there are other cells adjacent to the serving cell. Other cells adjacent to a serving cell are called neighbor cells.
- a base station that provides communication services for a neighbor cell is called a neighbor BS. The serving cell and neighbor cells are determined relatively based on the UE.
- downlink means communication from a base station (20) to a UE (10)
- uplink means communication from a UE (10) to a base station (20).
- the transmitter may be part of the base station (20), and the receiver may be part of the UE (10).
- the transmitter may be part of the UE (10), and the receiver may be part of the base station (20).
- wireless communication systems can be largely divided into FDD (frequency division duplex) and TDD (time division duplex).
- FDD frequency division duplex
- TDD time division duplex
- uplink transmission and downlink transmission are performed while occupying different frequency bands.
- TDD time division duplex
- the channel response of the TDD method is substantially reciprocal. This means that the downlink channel response and the uplink channel response are almost the same in a given frequency domain. Therefore, in a wireless communication system based on TDD, the downlink channel response has the advantage of being able to be obtained from the uplink channel response.
- the entire frequency band is time-divided into uplink transmission and downlink transmission, so the downlink transmission by the base station and the uplink transmission by the UE cannot be performed simultaneously.
- uplink transmission and downlink transmission are divided into subframe units, uplink transmission and downlink transmission are performed in different subframes.
- Figure 2 illustrates the structure of a radio frame used in NR.
- a radio frame has a length of 10 ms and is defined by two 5 ms half-frames (Half-Frames, HF).
- a half-frame is defined by five 1 ms subframes (Subframes, SF).
- a subframe is divided into one or more slots, and the number of slots in a subframe depends on the Subcarrier Spacing (SCS).
- SCS Subcarrier Spacing
- Each slot contains 12 or 14 OFDM (A) symbols depending on the cyclic prefix (CP). When a normal CP is used, each slot contains 14 symbols. When an extended CP is used, each slot contains 12 symbols.
- a symbol may include an OFDM symbol (or a CP-OFDM symbol), an SC-FDMA symbol (or a DFT-s-OFDM symbol).
- multiple numerologies may be provided to a terminal as wireless communication technology advances. For example, when the SCS is 15 kHz, it supports a wide area in traditional cellular bands; when the SCS is 30 kHz/60 kHz, it supports dense-urban, lower latency, and wider carrier bandwidth; and when the SCS is 60 kHz or higher, it supports a bandwidth larger than 24.25 GHz to overcome phase noise.
- the above numerology can be defined by the CP (cycle prefix) length and the subcarrier spacing (SCS).
- One cell can provide multiple numerologies to the terminal.
- the index of the numerology is represented as ⁇
- each subcarrier spacing and the corresponding CP length can be as shown in the table below.
- N slot symb the number of OFDM symbols per slot
- N frame, ⁇ slot the number of slots per frame
- N subframe, ⁇ slot the number of slots per subframe
- ⁇ ⁇ f 2 ⁇ 15 [kHz] N slot symb N frame, ⁇ slot N subframe, ⁇ slot 0 15 14 10 1 1 30 14 20 2 2 60 14 40 4 3 120 14 80 8 4 240 14 160 16 5 480 14 320 32 6 960 14 640 64
- N slot symb the number of OFDM symbols per slot
- N frame, ⁇ slot the number of slots per frame
- N subframe, ⁇ slot the number of slots per subframe
- OFDM(A) numerology e.g., SCS, CP length, etc.
- OFDM(A) numerology e.g., SCS, CP length, etc.
- the (absolute time) section of a time resource e.g., SF, slot or TTI
- TU Time Unit
- Figures 3a to 3c are exemplary diagrams showing exemplary architectures for wireless communication services.
- the UE is connected to an LTE/LTE-A based cell and an NR based cell in a DC (dual connectivity) manner.
- DC dual connectivity
- the above NR-based cell is connected to the core network for existing 4th generation mobile communications, i.e. Evolved Packet Core (EPC).
- EPC Evolved Packet Core
- an LTE/LTE-A-based cell is connected to a core network for 5th generation mobile communications, i.e., a 5G core network.
- NSA non-standalone
- the UE is connected only to NR-based cells.
- a service method based on this architecture is called SA (standalone).
- reception from a base station uses a downlink subframe, and transmission to a base station uses an uplink subframe.
- This method can be applied to paired spectrums and non-paired spectrums.
- a pair of spectrums means that two carrier spectrums are included for downlink and uplink operations.
- one carrier can include a downlink band and an uplink band that are paired with each other.
- Figure 4 illustrates the slot structure of an NR frame.
- a slot includes multiple symbols in the time domain. For example, in the case of a normal CP, one slot includes 14 symbols, but in the case of an extended CP, one slot includes 12 symbols.
- a carrier includes multiple subcarriers in the frequency domain.
- An RB Resource Block
- a BWP Bandwidth Part
- a terminal can be configured with up to N (e.g., 4) BWPs in the downlink and uplink, respectively.
- each element is referred to as a Resource Element (RE), to which one complex symbol can be mapped.
- RE Resource Element
- Figure 5 shows examples of subframe types in NR.
- the TTI (transmission time interval) illustrated in FIG. 5 may be called a subframe or slot for NR (or new RAT).
- the subframe (or slot) of FIG. 5 may be used in a TDD system of NR (or new RAT) to minimize data transmission delay.
- the subframe (or slot) includes 14 symbols.
- the symbols in the front of the subframe (or slot) may be used for a downlink (DL) control channel, and the symbols in the back of the subframe (or slot) may be used for an uplink (UL) control channel.
- the remaining symbols may be used for DL data transmission or UL data transmission.
- downlink transmission and uplink transmission may be sequentially performed in one subframe (or slot). Therefore, downlink data may be received within a subframe (or slot), and an uplink acknowledgement (ACK/NACK) may be transmitted within the subframe (or slot).
- ACK/NACK uplink acknowledgement
- subframes or slots
- slots self-contained subframes
- the first N symbols in a slot are used to transmit a DL control channel (hereinafter, DL control region), and the last M symbols in the slot can be used to transmit a UL control channel (hereinafter, UL control region).
- N and M are each an integer greater than or equal to 0.
- a resource region (hereinafter, data region) between the DL control region and the UL control region can be used for DL data transmission or UL data transmission.
- a physical downlink control channel (PDCCH) can be transmitted in the DL control region
- a physical downlink shared channel (PDSCH) can be transmitted in the DL data region.
- a physical uplink control channel (PUCCH) can be transmitted in the UL control region, and a physical uplink shared channel (PUSCH) can be transmitted in the UL data region.
- a time gap may be required for a transition process from a transmission mode to a reception mode or from a reception mode to a transmission mode.
- some OFDM symbols when switching from DL to UL in the subframe structure can be set as a guard period (GP).
- Figure 6 illustrates the structure of a self-contained slot.
- a frame is characterized by a self-contained structure in which a DL control channel, DL or UL data, and a UL control channel can all be included in one slot.
- the first N symbols in a slot can be used to transmit a DL control channel (hereinafter, referred to as a DL control region), and the last M symbols in a slot can be used to transmit a UL control channel (hereinafter, referred to as a UL control region).
- N and M are each integers greater than or equal to 0.
- a resource region hereinafter, referred to as a data region
- a data region between the DL control region and the UL control region can be used for DL data transmission or UL data transmission.
- the following configuration can be considered. Each section is listed in chronological order.
- DL Area (i) DL Data Area, (ii) DL Control Area + DL Data Area
- UL domain (i) UL data domain, (ii) UL data domain + UL control domain.
- a PDCCH In the DL control region, a PDCCH can be transmitted, and in the DL data region, a PDSCH can be transmitted.
- a PUCCH In the UL control region, a PUCCH can be transmitted, and in the UL data region, a PUSCH can be transmitted.
- DCI Downlink Control Information
- UCI Uplink Control Information
- ACK/NACK Positive Acknowledgement/Negative Acknowledgement
- CSI Channel State Information
- SR Service Request
- GP provides a time gap during the process in which a base station and a terminal switch from a transmission mode to a reception mode or during the process in which they switch from a reception mode to a transmission mode. Some symbols at the time of switching from DL to UL within a subframe can be set to GP.
- the current beam management method of 3GPP NR can be divided into the initial access phase and the cell connection establishment phase.
- a terminal performing the initial access procedure establishes its initial transmit/receive (Tx/Rx) beam through a random access procedure, i.e., a RACH (random access channel) procedure.
- Tx/Rx transmit/receive
- RACH random access channel
- Figure 7 shows an example of initial beam measurement and selection in NR.
- the base station in order to provide base station transmission beam (gNB Tx beam) setting to terminals (UE1/UE2) without cell connection, the base station repeatedly transmits SSBs (synchronization signal blocks) to which beams in different directions are mapped periodically. And, SSBs can be transmitted at 20ms cycles within 5ms. Specifically, the default value for initial cell selection can be 20ms.
- a terminal can select a qualified SSB through signal measurement for periodically transmitted SSBs and transmit a PRACH (physical random access channel) preamble mapped to the selected SSB, thereby informing the base station of information about the selected Tx beam.
- PRACH physical random access channel
- terminals at different locations i.e., UE1, can select an SSB having an SSB index of 3 and UE2, can select an SSB having an SSB index of 9, and then UE1 and UE2 can each transmit a corresponding PRACH preamble for the selected SSB.
- UE1 and UE2 can each transmit a corresponding PRACH preamble for the selected SSB.
- each SSB is beamformed in a specific direction.
- Figure 8 shows an example of an initial connection procedure between a terminal and a base station in NR.
- the UE receives cell-related parameter information (e.g., PRACH information corresponding to each SSB) required in the initial access stage through a system information message transmitted by a base station (gNB) (S802).
- the system information message includes a master information block (MIB) and a system information block 1 (SIB1) including cell common information.
- MIB master information block
- SIB1 system information block 1
- the terminal After the terminal acquires the system information message, it receives SSBs periodically transmitted from the base station (S803). Then, the terminal measures RSRP (reference signal received power) for the received SSBs. Among the N SSBs, i.e., beams, it selects one SSB (beam) with the highest/qualified value (value).
- RSRP reference signal received power
- the terminal transmits an RA (random access) preamble belonging to the PRACH resource corresponding to the selected SSB (beam) to the base station (S805). Through this, the terminal can inform the base station of the selected initial beam information.
- RA random access
- the base station receives an RA (random access) preamble belonging to a PRACH resource corresponding to an SSB (beam) selected from a terminal, and in response transmits an RAR (random access response) to the terminal using the selected SSB (beam) (S806).
- RA random access
- RAR random access response
- a base station that does not know the location/beam information of a terminal that first enters a cell i.e., a terminal performing the CBRA (contention based random access) procedure
- a terminal performing the CBRA (contention based random access) procedure can set up to 64 beams in common (cell commonly) for the beam setting of a terminal that has no connection, and the terminal sequentially measures all beams to find the optimal beam at its location. This not only causes a time delay in beam selection and cell connection as the number of beams in the cell increases, but can also increase the power consumption of the terminal by requiring the terminal to measure a large number of beams.
- the base station can identify the approximate location/beam of the initially connected terminal by mapping a wide beam for SSB, and can set a narrow beam through a beam refinement operation after the terminal accesses the cell.
- the narrow beam provides a high data rate to the terminal
- the base station allocates a CSI resource (CSI-RS/SSB) to which a candidate beam is mapped to the terminal in a UE-specific manner, so that the terminal continuously measures the surrounding beam strength and reports the measurement result to the base station.
- CSI-RS/SSB CSI resource
- Figure 9 shows an example of candidate beam settings in NR.
- a terminal that has received a beam report performs a report based on the configuration of the base station by measuring the reference signal (RS) allocated to it.
- RS reference signal
- this UE-specific CSI configuration method has a problem in that as the number of terminals in a cell increases, the RS resources allocated to each terminal also rapidly increase.
- the base station can select a method of allocating the same candidate beam, i.e., CSI resources, to terminals in similar locations, as shown in Fig. 9. This can be called UE group-specific CSI resource configuration.
- CSI resources i.e., CSI resources
- the base station can operate candidate beams by appropriately increasing the number of beams belonging to the CSI resource set.
- the increased number of beams increases the burden on measurement.
- Figures 10a to 10c illustrate three procedures for beam management in NR.
- Beam management in NR can be defined by dividing into three procedures in terms of procedures defined in the physical layer.
- Fig. 10a shows Procedure 1 (P1)
- Fig. 10b shows Procedure 2 (P2)
- Fig. 10c shows Procedure 3 (P3), respectively.
- P1 is an operation to find a transmission reception point (TRP) beam sweeping and UE beam sweeping simultaneously while performing beam setting of a terminal performing the initial access procedure described above.
- a terminal entering the connected mode recognizes that beams set by the base station through candidate beam (i.e., CSI resource set) setting will be swept, and first performs signal strength measurement for the TRP beam.
- the base station When the TRP beam of the terminal is selected through P2, the base station repeatedly transmits the selected one beam through P3.
- the terminal can select a UE beam while performing UE beam sweeping. It is up to the terminal implementation which beam the UE selects in this operation.
- the above-described operation can be applied to both downlink (DL) and uplink (UL).
- Figures 11a to 11c illustrate examples of beam reporting procedures in NR.
- Beam sweeping uses a method in which the base station notifies the terminal of reference signal (RS) resource information by setting a specific candidate beam, i.e., a CSI resource set, so that information about the beam is implicitly notified by mapping it with the RS resource information. That is, rather than notifying the terminal of the actual beam index, the base station recognizes the information about the mapped beam through the index information implicitly mapped to the RS information using the RS resource indicator (RI). This is set using the 3GPP CSI framework, and the terminal implicitly reports RSRP information about the best four beams (RI) to the base station by measuring the RS strength for the resources set by the base station.
- the method for reporting the measurement results also depends on the RRC setting of the base station, and 3GPP defines it to be set in one of the following three ways.
- FIG. 11a shows a periodic CSI reporting method, which is triggered through RRC configuration. That is, the terminal receives an RRC configuration message from the base station, and the RRC configuration message includes settings for CSI-related RS resources and reporting methods, i.e., CSI resource set information, and information that CSI reporting is periodic (S1101a). Thereafter, the terminal receives RSs periodically transmitted based on the received RRC configuration message (S1102a and S1105a), and measures signal strength for a beam based on the received RSs (S1103a and S1106a). Then, the terminal periodically reports the measured result (value) to the base station (S1104a and S1107a).
- the terminal receives an RRC configuration message from the base station, and the RRC configuration message includes settings for CSI-related RS resources and reporting methods, i.e., CSI resource set information, and information that CSI reporting is periodic (S1101a). Thereafter, the terminal receives RSs periodically transmitted based on the received RRC configuration
- FIG. 11b shows an aperiodic CSI reporting method. Even if CSI-related RS resources and a reporting method are configured through an RRC configuration message, beam measurement through RS is not performed without a trigger message (or information) from a lower layer. That is, the terminal receives an RRC configuration message including configuration of CSI-related RS resources and a reporting method, that is, CSI resource set information and information that CSI reporting is aperiodic, from the base station (S1101b), and the CSI report trigger is performed through a medium access control (MAC) control element (CE) or downlink control information (DCI).
- MAC medium access control
- CE control element
- DCI downlink control information
- the terminal receives CSI report trigger information including a trigger indication from the base station through the MAC CE or DCI (S1102b), and receives RSs transmitted once based on the received trigger indication (S1103b).
- the transmission of RSs for the CSI resource set can be transmitted after a specific time (e.g., X slots) at which the CSI report trigger information is transmitted.
- the terminal measures the signal strength for the beam based on the received RSs (S1104b).
- the terminal reports the measured result (value) to the base station once (S1105b).
- the CSI report can be transmitted after a specific time (e.g., Y slots) at which the CSI report trigger information is received.
- Fig. 11c shows a semi-persistent reporting method, which is an intermediate method between the periodic reporting method and the aperiodic reporting method.
- the terminal Upon receiving a configuration for CSI-related RS resources and a reporting method through an RRC configuration message, the terminal performs CSI reporting periodically until it receives a deactivation message (or information) only when activated by MAC CE. That is, the terminal receives an RRC configuration message including a configuration for CSI-related RS resources and a reporting method, that is, CSI resource set information and information that CSI reporting is semi-persistent, from the base station (S1101c), and CSI report activation is performed through MAC CE.
- a terminal receives CSI report activation information including an activation indication from a base station via MAC CE (S1102c and S1110c), receives RSs periodically transmitted based on the received activation indication (S1103c, S1106c, S1111c and S1114c), and measures signal strength for a beam based on the received RSs (S1104c, S1107c, S1112c and S1115c). Then, the terminal periodically reports the measured result (value) to the base station (S1105c, S1108c, S1113c and S1116c). After CSI reporting is activated, if CSI report deactivation information including a deactivation indication is received from the base station via MAC CE (S1109c), the terminal stops CSI reporting.
- 3GPP is considering applying AI/ML models to improve the delay and terminal power consumption of such beam search/measurement, and has started a study to discuss the feasibility and potential spec impact of this.
- Terminology Description Data collection The process of collecting data by network nodes, management entities, or UEs for the purpose of AI/ML model training, data analysis, and inference.
- AI/ML Model Data-driven algorithms that apply AI/ML techniques to generate output sets based on input sets.
- AI/ML model training The process of training an AI/ML model [by learning input/output relationships] in a data-driven manner and obtaining a trained AI/ML model for inference.
- AI/ML model testing A sub-process of training to evaluate the performance of the final AI/ML model using a different dataset than that used for model training and validation. Unlike AI/ML model validation, testing does not assume any subsequent tuning of the model. (A subprocess of training, to evaluate the performance of a final AI/ML model using a dataset different from one used for model training and validation.
- Joint inference is AI/ML inference where inference is performed jointly by the UE and the network, i.e., the first part of the inference is performed by the UE first and the remaining part by the gNB, or vice versa.
- Unsupervised learning The process of training a model without labeled data.
- Semi-supervised learning The process of training a model using a mixture of labeled and unlabeled data.
- Reinforcement Learning (RL)Reinforcement Learning (RL) The process of training an AI/ML model from feedback signals (reward) based on inputs (states) and outputs (actions) of the model in an environment where the model interacts.
- 3GPP has decided to conduct a study on the specification impact on beam result reporting methods in relation to data collection for AI model training on the UE side in relation to beam management procedures.
- the beam management operation in the conventional NR causes the problem of increasing the system overhead and the power consumption of the terminal as the number of beams and the number of terminals increase.
- the terminal since the terminal goes through the process of selecting the initial beam after measuring all beams, it may cause a delay in cell access.
- Figures 12a and 12b show examples of beam measurement and spatial domain beam prediction using AI/ML.
- FIGS. 12a and 12b The case of spatial DL beam prediction is illustrated in FIGS. 12a and 12b.
- FIG. 12a shows a case where Set B is a subset of Set A
- FIG. 12b considers a set where Set B consists of wide beams and Set A consists of narrow beams, that is, sets composed of different beams.
- temporal DL beam prediction in addition to the cases where i) set B is a subset of set A, ii) sets A and B are different sets, we consider the case where iii) sets A and B consist of the same set.
- Temporal DL beam prediction predicts future beam information based on past beam measurement information, so we may consider a method where spatial DL beam prediction is used as the basis for predicting the entire beam and then applying it to the case iii) where sets A and B consist of the same set. For this reason, it is expected that the cases where i) set B is a subset of set A, and ii) sets A and B are different sets, for spatial DL beam prediction, will be used as the basic beam prediction method.
- Figure 13 shows an example of temporal domain beam prediction using AI/ML.
- Temporal beam prediction of BM Case 2 is defined as an operation of predicting a beam result (i.e., output) at a specific point in the near future based on past beam measurement result information (i.e., input), as illustrated in Fig. 13.
- the set of beams to be used as input and the set of beams derived as output can consider, in addition to the cases described above, i) when set B is a subset of set A, ii) when sets A and B are different sets, and iii) when sets A and B are composed of the same set.
- a model training process must be prerequisite.
- the terminal In order to perform model training for beam management on the terminal (UE) side, the terminal must be set up with a DL RS (downlink reference signal) for training from the base station.
- DL RS downlink reference signal
- the terminal can start training, but there is no specific discussion on how the base station will provide data for training or how long the terminal should collect data.
- model training can be terminated when the model performance reaches the target value.
- the terminal or NW that determines this needs to terminate the model training or release the related settings, but there has been no discussion on this so far.
- the present invention proposes a method for determining a setup procedure for data transmission and termination of data transmission for effective UE-side model training when performing beam management using an AI/ML model based on the aforementioned contents.
- the present invention proposes a method for a data collection procedure for model training, including a step of requesting a base station to transmit data for data collection when a terminal (UE) needs to collect data related to model training in order to perform communication using an AI/ML model, or receiving a message notifying the start of data transmission for model training from the base station, the terminal performs model training through data collection by receiving data transmitted using resources set by the base station, and transmitting a data termination message to the base station to terminate data collection when a specific event is satisfied. More specifically, the present invention defines a step of determining a termination point in time for data collection for model training, a message definition related thereto, and terminal/base station operations.
- Figure 14 shows an operation method of a terminal according to one embodiment of the present specification.
- the terminal receives data for AI/ML model training from the base station (S1401) and performs AI/ML model training based on the received data (S1402). After performing the AI/ML model training, the terminal transmits a first message instructing the termination of data collection to the base station (S1403).
- the first message may be for deactivating transmission of data for AI/ML model training.
- Figure 15 illustrates an operation method of a terminal according to another embodiment of the present specification.
- a request for transmission of data for AI/ML model training transmitted by a terminal to a base station is shown.
- a terminal that has decided to start collecting data for AI/ML model training transmits a request message (second message) to request the base station to transmit data required for data collection (S1500).
- the second message may include at least one of the following information i) to iii).
- the base station that receives the second message from the terminal must start transmitting data to the terminal based on the received information. This may be because resource information for transmitting the data requested by the terminal is preset for the terminal during the functionality/model identification process, or a configuration message may be additionally transmitted from the base station to the terminal through reception of the second message.
- the transmission of the second message from the terminal to the base station may mean that the actual terminal requests that resources for data transmission for model training be set, or if the resources have already been set, requests that the corresponding resources be activated.
- the terminal may transmit the request message, i.e., the second message, to the base station and start model training corresponding to it.
- the terminal after transmitting the second message requesting transmission of data for AI/ML model training, the terminal receives data for AI/ML model training from the base station (S1501), and performs AI/ML model training based on the received data (S1502). After performing the AI/ML model training, the terminal transmits a first message instructing the termination of data collection to the base station (S1503).
- the base station transmits a response message to the second message and the terminal expects to receive it.
- Figure 16 illustrates an operation method of a terminal according to another embodiment of the present specification.
- the terminal receives a notification from the base station to start transmission of data for AI/ML model training.
- the terminal receives a notification of start of transmission of data for AI/ML model training from the base station (S1600). Thereafter, data for AI/ML model training is received from the base station (S1601), and AI/ML model training is performed based on the received data (S1602). After performing AI/ML model training, the terminal transmits a first message instructing the end of data collection to the base station (S1603).
- the notification of the start of transmission of data for AI/ML model training from the base station can be transmitted and received without transmission of the second message requesting transmission of data for AI/ML model training from the terminal described in FIG. 15 above.
- the notification of the start of data transmission for AI/ML model training can be made through a configuration message related to data for specific model training.
- a configuration message related to data for specific model training For example, if the activation of data transmission resources for specific model training is made through the configuration message, it can mean the start of data transmission for model training of the terminal.
- Figures 17a and 17b illustrate procedures of a terminal and a base station according to embodiments of the present specification.
- FIGS. 17A and 17B illustrate procedures according to embodiments of data collection for model training, wherein FIG. 17A illustrates a procedure for starting data collection by a terminal request, and FIG. 17B illustrates a procedure for starting data collection by base station settings.
- the terminal transmits AI/ML related capability information to a base station, i.e., NW (network) (S1701a), and receives AI/ML related setting(s) from the NW through this (S1702a).
- NW network
- the terminal transmits including capability information for function IDs #1 and #2, and the base station that receives this transmits AI/ML related setting(s) to the terminal based on the information received from the terminal.
- the above-described process may be referred to as a functionality identification process or a model identification process.
- the terminal has the capability to perform AI/ML based beam management, and AI/ML related capability information may be transmitted to the base station including a functionality/model ID and related additional information.
- the base station transmits beam information (e.g., CSI-RS resource configuration) for Set A/B required for model training to the terminal based on the AI/ML related capability information received from the terminal, and through this, the terminal may receive at least one AI/ML related configuration.
- FIG. 17a illustrates the following AI/ML related configurations.
- the data request message may instruct, i.e., request, activation/start of one configuration for the configured functionality/model.
- the terminal transmits a data request for model training including information about a configuration of one of the configured functionality/models (S1703a).
- the data request message for model training may be transmitted including information about 'functionality ID #1: CSI-RS resource set #1'.
- the base station may be defined to transmit a response message to the data request message, and the terminal may expect to receive the same.
- the base station transmits a confirmation message in response to the data request message transmitted by the terminal (S1704a).
- a configuration message notifying resource information for data transmission may be transmitted from the base station to the terminal in response to a data request message from the terminal.
- the data request message transmitted by the terminal may mean that the terminal requests that transmission resources for data for model training be configured.
- the terminal transmits a data request message for model training including AI/ML related capability information to the base station, i.e., the NW (network) (S1703b), and receives a configuration message including AI/ML related configuration from the NW through this (S1704b).
- the terminal may transmit including capability information for function ID #1, and the base station that receives this may transmit a configuration message including information on CSI-RS resource set #1 to the terminal based on the information received from the terminal.
- the base station activates resources for data transmission for model training. That is, it activates CSI-RS resource set #1 (S1705a, S1705b) and performs corresponding data transmission (S1706a, S1706b).
- the corresponding data transmission means that RSs for CSI-RS resource set #1 are repeatedly transmitted.
- the terminal performs model training by collecting data transmitted from the base station, and when the model training is terminated, it instructs the base station to terminate the model training (S1707a, S1707b).
- the message instructing the termination of the model training may include corresponding function ID information, i.e., information on function ID #1.
- the present invention additionally proposes that a terminal performing model training decides to terminate data collection for a model when the following events (1 to 4) are satisfied.
- the threshold for performance and the value for the timer applied to the time interval/period can be set by the base station, and it is preferable that it is set in units of functionality/model received from the terminal, i.e. instructed.
- the threshold may be applied with the same or different parameters to events 1 and 2. If it is defined with different parameters, it is preferable that the threshold defined for event 2 has a smaller value than the threshold defined for event 1.
- the value corresponding to the timer may replace the value included in the data request message of the terminal.
- the above-mentioned event parameters may be set by the base station to the terminal during the functionality/model identification process between the terminal and the base station, or may be set through a confirm/configuration message received from the base station in response to a data request message transmitted by the terminal.
- the base station transmits a configuration message for data for model training of the terminal without transmitting a data request message from the terminal, they may be included in the configuration message.
- timers defined in events 2 and 4 can be defined with the same or different parameters.
- the timer(s) proposed in the above-mentioned event may be defined differently based on various factors that can measure the total amount of data, such as data volume/size/time instance, etc. Application of other factors that are used in the same meaning as the data volume/time required for the entire model training proposed above may be judged to be the same technology as the technology proposed in the present invention.
- a terminal that decides to terminate model training based on event 1 or event 2 may transmit a message to the base station notifying that model training at the terminal is terminated to notify that data transmission being provided from the base station to the terminal is no longer necessary, and the message may include at least one of the following information:
- the reason for termination of model training may be model training success information in the case of event 1, or model training failure information in the case of event 2.
- model training termination message may be defined to be transmitted from the base station to the terminal according to event 3 or event 4.
- the termination of model training may be implicitly recognized without message transmission and data collection and provision operations may be deactivated, and the transmission of the termination message may be performed from the base station to the terminal or from the terminal to the base station as needed.
- Figure 18 shows a procedure of a terminal and a base station for event-based model training according to one embodiment of the present specification.
- Figure 18 shows the procedure of a terminal and a base station when at least event 1 is set.
- Event 1 When the model's performance is judged to have exceeded a certain threshold while performing model training.
- the terminal receives a message including a transmission start notification of data for model training from the base station or data-related setting information (S1802).
- the message may include an event parameter (e.g., event 1) for model training termination and/or a related threshold parameter (e.g., threshold value).
- event parameter e.g., event 1
- threshold value e.g., threshold value
- a setting/confirmation message may be received including information about function ID #1, event 1 information, and threshold information for event 1.
- the terminal Before the terminal receives a message including a notification of start of transmission of data for model training or data-related configuration information, the terminal optionally transmits a message requesting activation of data transmission for model training to the base station (S1801). For example, a data request message for model training may be transmitted including information about function ID #1.
- the terminal receives data transmitted from the base station (S1804) and performs model training based on the received data.
- the reception of the data means that RSs for the activated CSI-RS resource set #1 are repeatedly received.
- the terminal can periodically evaluate the performance of the model. When the performance value of the model reaches the threshold received from the base station, i.e., satisfies event 1, the terminal decides to terminate the model training.
- the terminal transmits a message to the base station notifying that model training is ended (S1805).
- the model training end message may be transmitted with information about function ID #1 included.
- the base station transmits a message including a start notification of data transmission for model training to the terminal or data-related setting information (S1802).
- the message may include an event parameter (e.g., event 1) for model training termination and/or a related threshold parameter (e.g., threshold value).
- event parameter e.g., event 1
- threshold value e.g., threshold value
- a setting/confirmation message may be transmitted including information about function ID #1, information about event 1, and threshold information about event 1.
- the base station Before the base station transmits a message including a start notification of data transmission for model training or data-related configuration information, the base station optionally receives a message requesting activation of data transmission for model training from the terminal (S1801). For example, a data request message for model training may be received including information about function ID #1.
- the base station activates the transmission resources of data for model training. That is, it activates the CSI-RS resource set #1 (S1803) and performs the corresponding data transmission (S1804).
- the corresponding data transmission means that RSs for the activated CSI-RS resource set #1 are repeatedly transmitted.
- the beam can be periodically transmitted through the configured CSI-RS.
- a message notifying that model training is terminated is received from the terminal (S1805).
- the model training termination message may be received including information about function ID #1.
- the base station deactivates transmission resources for data for model training (S1806). That is, the base station deactivates CSI-RS resource set #1 for model training and stops transmission of the corresponding data.
- Figure 19 illustrates a procedure of a terminal and a base station for event-based model training according to another embodiment of the present specification.
- Figure 19 shows the procedure of the terminal and the base station when at least event 2 is set.
- Event 2 If the model performance does not exceed a certain threshold within a certain time interval (Timer) after model training begins.
- Event 2 can operate in the same way as event 1 operates with a specific timer. If event 1 is set to operate with a timer, it can check whether the performance satisfies the threshold while the timer is operating, and if it satisfies the threshold, the model training can be terminated even before the timer expires, and if the timer expires, the terminal can notify the base station of the end of the model training.
- the terminal receives a message including a transmission start notification of data for model training from the base station or data-related setting information (S1902).
- the message may include an event parameter for model training termination (e.g., event 2), a related timer parameter (e.g., timer value), and a threshold parameter (e.g., threshold value).
- event parameter for model training termination e.g., event 2
- timer parameter e.g., timer value
- threshold parameter e.g., threshold value
- a setting/confirmation message may be received including information on function ID #1, information on event 2, timer information for event 2, and threshold information.
- the terminal Before the terminal receives a message including a notification of start of transmission of data for model training or data-related configuration information, the terminal optionally transmits a message requesting activation of data transmission for model training to the base station (S1901). For example, a data request message for model training may be transmitted including information about function ID #1.
- the terminal receives data transmitted from the base station (S1904) and performs model training based on the received data.
- the reception of the data means that RSs for the activated CSI-RS resource set #1 are repeatedly received.
- the terminal starts the timer based on the timer information previously received.
- the timer can be started i) when a message including a notification of the start of transmission of data for model training or data-related configuration information is received from the base station, or ii) when the first data resource configured/transmitted from the base station is received.
- the terminal periodically evaluates the performance of the model while the timer is running. If the performance value of the model reaches the previously received threshold, i.e., satisfies Event 1, the terminal decides to terminate the model training. In addition, if the threshold received from the base station is not reached while the timer is running and the timer expires, i.e., satisfies Event 2, the terminal also decides to terminate the model training.
- the terminal transmits a message to the base station notifying that model training is ended (S1905).
- the model training end message may be transmitted including information about function ID #1.
- the base station transmits a message including a start notification of data transmission for model training to the terminal or data-related setting information (S1902).
- the message may include an event parameter for model training termination (e.g., event 2), a related timer parameter (e.g., timer value), and a threshold parameter (e.g., threshold value).
- an event parameter for model training termination e.g., event 2
- a related timer parameter e.g., timer value
- a threshold parameter e.g., threshold value
- a setting/confirmation message may be transmitted including information on function ID #1, information on event 2, timer information on event 2, and threshold information.
- the base station Before the base station transmits a message including a start notification of data transmission for model training or data-related configuration information, the base station optionally receives a message requesting activation of data transmission for model training from the terminal (S1901). For example, a data request message for model training may be received including information about function ID #1.
- the base station activates the transmission resources of data for model training. That is, it activates the CSI-RS resource set #1 (S1903) and performs the corresponding data transmission (S1904).
- the corresponding data transmission means that RSs for the activated CSI-RS resource set #1 are repeatedly transmitted.
- the beam can be transmitted periodically through the configured CSI-RS.
- a message notifying that model training is terminated is received from the terminal (S1905).
- the model training termination message may be received including information about function ID #1.
- the base station deactivates transmission resources for data for model training (S1906). That is, the base station deactivates CSI-RS resource set #1 for model training and stops transmission of the corresponding data.
- Figure 20 illustrates a procedure of a terminal and a base station for event-based model training according to another embodiment of the present specification.
- Figure 20 shows the procedures of a terminal and a base station when at least event 3 is set.
- the terminal receives a message including a transmission start notification of data for model training from a base station or data-related setting information (S2002).
- the message may include event parameters for model training termination (e.g., event 1, event 2, and/or event 3), related timer parameters (e.g., timer value), and threshold parameters (e.g., threshold value).
- event parameters for model training termination e.g., event 1, event 2, and/or event 3
- related timer parameters e.g., timer value
- threshold parameters e.g., threshold value
- a setting/confirmation message may be received including information on function ID #1, information on event 1/2/3, timer information for event 2, and threshold information.
- the terminal Before the terminal receives a message including a notification of start of data transmission for model training or data-related configuration information, the terminal optionally transmits a message requesting activation of data transmission for model training to the base station (S2001). For example, a data request message for model training may be transmitted including information about function ID #1.
- the terminal receives data transmitted from the base station (S2004) and performs model training based on the received data.
- the reception of the data means that RSs for the activated CSI-RS resource set #1 are repeatedly received.
- the terminal starts the timer based on the timer information previously received.
- the timer can be started i) when a message including a notification of the start of transmission of data for model training or data-related configuration information is received from the base station, or ii) when the first data resource configured/transmitted from the base station is received.
- the terminal periodically evaluates the performance of the model while the timer is running. If the performance value of the model reaches a previously received threshold, i.e., satisfies Event 1, the terminal decides to terminate the model training. If, while the timer is running, the terminal satisfies Event 3 by receiving a message from the base station notifying release/deactivation of resource settings for data transmitted for data collection of the terminal, the terminal also decides to terminate the model training. For example, the terminal decides to terminate the model training by receiving a data release message including function ID #1 information from the base station (S2006).
- the terminal decides to terminate model training.
- the base station transmits a message including a start notification of data transmission for model training to the terminal or data-related setting information (S2002).
- the message may include event parameters for model training termination (e.g., event 1, event 2, and/or event 3), related timer parameters (e.g., timer value), and threshold parameters (e.g., threshold value).
- event parameters for model training termination e.g., event 1, event 2, and/or event 3
- related timer parameters e.g., timer value
- threshold parameters e.g., threshold value
- a setting/confirmation message may be transmitted including information on function ID #1, information on event 1/2/3, timer information for event 2, and threshold information.
- the base station Before the base station transmits a message including a start notification of data transmission for model training or data-related configuration information, the base station optionally receives a message requesting activation of data transmission for model training from the terminal (S2001). For example, a data request message for model training may be received including information about function ID #1.
- the base station activates transmission resources for data for model training. That is, it activates CSI-RS resource set #1 (S2003) and performs corresponding data transmission (S2004).
- the corresponding data transmission means that RSs for the activated CSI-RS resource set #1 are repeatedly transmitted.
- the beam can be periodically transmitted through the configured CSI-RS.
- the base station decides to release/deactivate the CSI-RS resources set for model training of the terminal, it transmits a message indicating termination of model training (or a message notifying release/deactivation of data-related resources) to the terminal and stops the requested data transmission. For example, the base station decides to deactivate CSI-RS resource set #1 (S2005) and transmits a data release message including function ID #1 information to the terminal (S2006).
- Figure 21 shows a procedure of a terminal and a base station for event-based model training according to another embodiment of the present specification.
- Figure 21 shows the procedures of the terminal and base station when event 4 is set.
- Event 4 When the data collection period (timer) requested/set by the terminal is reached
- Event 4 can be applied when it operates independently without events 1 to 3. That is, the terminal or base station sets a time duration and performs model training only during that time.
- the terminal transmits a message requesting activation of data transmission for model training to the base station (S2101).
- the data request message for model training may be transmitted including information about function ID #1 and corresponding time information (e.g., timer) required for model training.
- the terminal receives a message including a transmission start notification of data for model training from the base station or data-related setting information (S2102).
- the message may include time information (e.g., timer) for continuing data transmission.
- time information e.g., timer
- a setting/confirmation message may be received including information about function ID #1 and timer information about event 4.
- the timer value may be included in a message transmitted from a terminal to a base station or from a base station to a terminal, and if the timer value is included in both messages, it is preferable to preferentially apply the timer value included in the message transmitted from the base station to the terminal.
- the terminal receives data transmitted from the base station (S2104) and performs model training based on the received data.
- the reception of the data means that RSs for the activated CSI-RS resource set #1 are repeatedly received.
- the terminal starts the timer described above.
- the timer can be started either i) upon receiving a message from the base station indicating the start of transmission of data for model training or including data-related configuration information, or ii) upon receiving the first data resource configured/transmitted from the base station.
- the terminal performs training of the corresponding model while the timer is running, and when the timer expires, i.e., event 4 is satisfied, the terminal decides to terminate model training (S2105).
- the base station receives a message from the terminal requesting to activate data transmission for model training (S2101). For example, a data request message for model training may be received including information about function ID #1 and corresponding time information (e.g., timer) required for model training.
- a data request message for model training may be received including information about function ID #1 and corresponding time information (e.g., timer) required for model training.
- the base station transmits a message including a notification of the start of transmission of data for model training to the terminal or data-related setting information (S2102).
- the message may include time information (e.g., timer) for continuing data transmission.
- time information e.g., timer
- information about function ID #1 and timer information about event 4 may be included and transmitted in the setting/confirmation message.
- the timer value may be included in a message transmitted from a terminal to a base station or from a base station to a terminal, and if the timer value is included in both messages, it is preferable to preferentially apply the timer value included in the message transmitted from the base station to the terminal.
- the base station activates the transmission resources of data for model training. That is, it activates the CSI-RS resource set #1 (S2103) and performs the corresponding data transmission (S2104).
- the corresponding data transmission means that RSs for the activated CSI-RS resource set #1 are repeatedly transmitted.
- the beam can be periodically transmitted through the configured CSI-RS.
- the base station starts the timer described above.
- the timer can be started by i) transmitting a message containing a start notification of data transmission for model training to the terminal or data-related configuration information, or ii) transmitting the first data resource set by the base station.
- the base station repeatedly (or periodically) transmits the data according to the setting while the timer is running.
- the base station decides to terminate the data transmission. That is, the base station performs deactivation of CSI-RS resource set #1 (S2106).
- Figures 22 and 23 illustrate procedures of a terminal and a base station for explaining termination of model training according to embodiments of the present specification.
- Figure 22 shows a procedure for terminating model training by a terminal and a procedure for terminating model training by a base station.
- the terminal transmits a message requesting activation of data transmission for model training to the base station (S2201, S2301).
- the data request message for model training may be transmitted including information on function ID #1 and corresponding time interval information (e.g., timer).
- the base station transmits a message including a start notification of data transmission for model training or data-related setting information to the terminal (S2202, S2302).
- a message including a start notification of data transmission for model training or data-related setting information For example, information about function ID #1 may be included in the setting/confirmation message and transmitted.
- the base station activates transmission resources for data for model training, that is, activates CSI-RS resource set #1 (S2203, S2303), and performs corresponding data transmission (S2204, S2304).
- the corresponding data transmission means that RSs for the activated CSI-RS resource set #1 are repeatedly transmitted.
- the beam can be transmitted periodically through the configured CSI-RS.
- the terminal proposed in this specification can transmit to the base station, when transmitting a data request message to the base station to collect data for model training, the amount/size of data that the terminal wants to collect or the minimum time information required to collect the same.
- the model training can be defined to be terminated based on the time or data amount included in the data request message that the terminal transmits to the base station (or by including the time or data amount in the setting message that notifies that the base station starts transmitting data to the terminal).
- the base station recognizes the information received from the terminal as one of the assistance information of the terminal, and can continue to provide data until the time or amount of data received from the terminal is satisfied. However, this means that data transmission from the base station can be terminated at any time even if the information is not satisfied, depending on the implementation of the base station. That is, as shown in FIGS. 22 and 23, the terminal that started model training or the base station that started data transmission can determine to terminate the model training based on their own internal decisions (S2205, S2305), and thereafter, the terminal transmits a model training termination message to the base station (S2206), or the base station can notify the terminal that it will stop data transmission (S2306).
- the base station that receives the model training termination message from the terminal deactivates the corresponding data transmission resource (S2207), and the terminal that receives the data release/deactivation message from the base station terminates the model training (S2308).
- the base station that has transmitted a data release/deactivation message to the terminal then deactivates the corresponding data transmission resource (S2307).
- the base station must provide data to the terminal at least until the time or data amount received from the terminal (or transmitted to the terminal) is satisfied in order to provide the data requested by the terminal.
- the base station and the terminal must perform data transmission/collection until the corresponding time or data amount is satisfied, and after the corresponding time or data amount is satisfied, the model training can be terminated through a model training termination message or a data release/deactivation message depending on the base station implementation or the terminal implementation.
- the base station and the terminal may recognize the time or data amount information set by the terminal as the end time for data collection and provision, and may mean that the model training should be terminated when the set time or data amount is reached.
- time duration/instance or data size/volume transmitted by the terminal or received from the base station may be used as a value to determine the corresponding time, and may be used in the same way as the timer defined in the operation according to the set event proposed in this specification.
- the time instance can be defined by an arbitrary COUNT parameter.
- the terminal or base station can include the maximum count (MAX_CNT) value in the data request or data-related configuration message. This has the same role as the timer described above. First, the case where MAX_CNT is used instead of the timer in the previously proposed event.
- the terminal performs model training while periodically receiving a data set from the base station. Whenever the terminal receives a data set of one period, the COUNT value is increased by 1, and if the COUNT value has not reached the MAX_CNT value, the terminal can evaluate the performance of the model. If the performance exceeds the threshold, the model training is terminated by transmitting a model training termination message to the base station.
- the same operation can be performed by setting the MAX_CNT value instead of the timer value in the case where the terminal transmits only a timer to the base station or the base station transmits only a timer to the terminal without an event.
- the data request/termination message can be defined using a PHY (physical), MAC (medium access control) CE (control element) or RRC (radio resource control) message.
- PHY physical
- MAC medium access control
- CE control element
- RRC radio resource control
- the data request PUCCH can be defined according to the functionality/model of the terminal, and the same PUCCH can be used even when the terminal transmits termination signaling. This is preferably applied in cases where training occurs frequently, such as online learning, and if defined as PUCCH, it is preferable that the parameter information proposed in this specification be set in a prior process, such as functionality/model identification.
- Figure 24 illustrates a device according to one embodiment of the present specification.
- the wireless communication system may include a first device (100a) and a second device (100b).
- the above first device (100a) may be a base station, a network node, a transmitting terminal, a receiving terminal, a wireless device, a wireless communication device, a vehicle, a vehicle equipped with an autonomous driving function, a connected car, a drone (Unmanned Aerial Vehicle, UAV), an AI (Artificial Intelligence) module, a robot, an AR (Augmented Reality) device, a VR (Virtual Reality) device, an MR (Mixed Reality) device, a hologram device, a public safety device, an MTC device, an IoT device, a medical device, a fintech device (or a financial device), a security device, a climate/environmental device, a device related to 5G services, or any other device related to the 4th industrial revolution field.
- UAV Unmanned Aerial Vehicle
- AI Artificial Intelligence
- a robot an AR (Augmented Reality) device, a VR (Virtual Reality) device, an MR (Mixed
- the second device (100b) may be a base station, a network node, a transmitting terminal, a receiving terminal, a wireless device, a wireless communication device, a vehicle, a vehicle equipped with an autonomous driving function, a connected car, a drone (Unmanned Aerial Vehicle, UAV), an AI (Artificial Intelligence) module, a robot, an AR (Augmented Reality) device, a VR (Virtual Reality) device, an MR (Mixed Reality) device, a hologram device, a public safety device, an MTC device, an IoT device, a medical device, a fintech device (or a financial device), a security device, a climate/environmental device, a device related to 5G services, or any other device related to the 4th industrial revolution field.
- UAV Unmanned Aerial Vehicle
- AI Artificial Intelligence
- a robot an AR (Augmented Reality) device, a VR (Virtual Reality) device, an MR (Mixed Reality
- the first device (100a) may include at least one processor, such as a processor (1020a), at least one memory, such as a memory (1010a), and at least one transceiver, such as a transceiver (1031a).
- the processor (1020a) may perform the functions, procedures, and/or methods described above.
- the processor (1020a) may perform one or more protocols.
- the processor (1020a) may perform one or more layers of a wireless interface protocol.
- the memory (1010a) may be connected to the processor (1020a) and may store various forms of information and/or commands.
- the transceiver (1031a) may be connected to the processor (1020a) and may be controlled to transmit and receive wireless signals.
- the second device (100b) may include at least one processor, such as a processor (1020b), at least one memory device, such as a memory (1010b), and at least one transceiver, such as a transceiver (1031b).
- the processor (1020b) may perform the functions, procedures, and/or methods described above.
- the processor (1020b) may implement one or more protocols.
- the processor (1020b) may implement one or more layers of a wireless interface protocol.
- the memory (1010b) may be connected to the processor (1020b) and may store various forms of information and/or commands.
- the transceiver (1031b) may be connected to the processor (1020b) and may be controlled to transmit and receive wireless signals.
- the above memory (1010a) and/or the above memory (1010b) may be connected internally or externally to the processor (1020a) and/or the processor (1020b), respectively, and may be connected to another processor via various technologies such as a wired or wireless connection.
- the first device (100a) and/or the second device (100b) may have one or more antennas.
- the antenna (1036a) and/or the antenna (1036b) may be configured to transmit and receive wireless signals.
- Figure 25 is a block diagram showing the configuration of a terminal according to one embodiment of the present specification.
- FIG. 25 is a drawing illustrating the device of FIG. 24 in more detail.
- the device includes a memory (1010), a processor (1020), a transceiver (1031), a power management module (1091), a battery (1092), a display (1041), an input unit (1053), a speaker (1042), a microphone (1052), a subscriber identification module (SIM) card, and one or more antennas.
- the processor (1020) may be configured to implement the proposed functions, procedures and/or methods described herein. Layers of a radio interface protocol may be implemented in the processor (1020).
- the processor (1020) may include an application-specific integrated circuit (ASIC), another chipset, logic circuitry and/or data processing devices.
- the processor (1020) may be an application processor (AP).
- the processor (1020) may include at least one of a digital signal processor (DSP), a central processing unit (CPU), a graphics processing unit (GPU), and a modem (modulator and demodulator).
- DSP digital signal processor
- CPU central processing unit
- GPU graphics processing unit
- modem modulator and demodulator
- Examples of the processor (1020) may be a SNAPDRAGONTM series processor manufactured by Qualcomm®, an EXYNOSTM series processor manufactured by Samsung®, an A series processor manufactured by Apple®, a HELIOTM series processor manufactured by MediaTek®, an ATOMTM series processor manufactured by INTEL®, a KIRINTM series processor manufactured by HiSilicon®, or a corresponding next-generation processor.
- the power management module (1091) manages power to the processor (1020) and/or the transceiver (1031).
- the battery (1092) supplies power to the power management module (1091).
- the display (1041) outputs the results processed by the processor (1020).
- the input unit (1053) receives input to be used by the processor (1020).
- the input unit (1053) can be displayed on the display (1041).
- a SIM card is an integrated circuit used to securely store an international mobile subscriber identity (IMSI) and its associated keys, which are used to identify and authenticate subscribers in mobile devices such as mobile phones and computers. Contact information can also be stored on many SIM cards.
- IMSI international mobile subscriber identity
- the memory (1010) is operably coupled with the processor (1020) and stores various information for operating the processor (610).
- the memory (1010) may include a read-only memory (ROM), a random access memory (RAM), a flash memory, a memory card, a storage medium, and/or other storage devices.
- ROM read-only memory
- RAM random access memory
- flash memory a non-transitory computer-readable medium
- the modules may be stored in the memory (1010) and executed by the processor (1020).
- the memory (1010) may be implemented within the processor (1020). Alternatively, the memory (1010) may be implemented outside the processor (1020) and may be communicatively connected to the processor (1020) via various means known in the art.
- the transceiver (1031) is operably coupled to the processor (1020) and transmits and/or receives a radio signal.
- the transceiver (1031) includes a transmitter and a receiver.
- the transceiver (1031) may include a baseband circuit for processing a radio frequency signal.
- the transceiver controls one or more antennas to transmit and/or receive a radio signal.
- the processor (1020) transmits command information to the transceiver (1031) to initiate communication, for example, to transmit a radio signal constituting voice communication data.
- the antenna functions to transmit and receive radio signals.
- the transceiver (1031) may transmit the signal for processing by the processor (1020) and convert the signal to a baseband.
- the processed signal may be converted into audible or readable information output through the speaker (1042).
- the speaker (1042) outputs sound-related results processed by the processor (1020).
- the microphone (1052) receives sound-related input to be used by the processor (1020).
- a user inputs command information, such as a telephone number, for example, by pressing (or touching) a button on an input unit (1053) or by voice activation using a microphone (1052).
- the processor (1020) receives the command information and processes it to perform an appropriate function, such as making a call to the telephone number.
- Operational data may be extracted from a SIM card or memory (1010).
- the processor (1020) may display command information or operational information on a display (1041) for the user's recognition and convenience.
- FIG. 26 illustrates a block diagram of a processor in which the disclosure of the present specification is implemented.
- the processor (1020) implementing the disclosure of the present specification may include a plurality of circuits to implement the proposed functions, procedures and/or methods described herein.
- the processor (1020) may include a first circuit (1020-1), a second circuit (1020-2) and a third circuit (1020-3).
- the processor (1020) may include more circuits.
- Each circuit may include a plurality of transistors.
- the above processor (1020) may be called an ASIC (application-specific integrated circuit) or AP (application processor) and may include at least one of a DSP (digital signal processor), a CPU (central processing unit), and a GPU (graphics processing unit).
- ASIC application-specific integrated circuit
- AP application processor
- DSP digital signal processor
- CPU central processing unit
- GPU graphics processing unit
- FIG. 27 is a block diagram showing in detail the transceiver of the first device illustrated in FIG. 24 or the transceiver unit of the device illustrated in FIG. 25.
- the transceiver (1031) includes a transmitter (1031-1) and a receiver (1031-2).
- the transmitter (1031-1) includes a DFT (Discrete Fourier Transform) unit (1031-11), a subcarrier mapper (1031-12), an IFFT unit (1031-13), a CP insertion unit (1031-14), and a wireless transmitter (1031-15).
- the transmitter (1031-1) may further include a modulator.
- the transmitter may further include a scramble unit (not shown), a modulation mapper (not shown), a layer mapper (not shown), and a layer permutator (not shown), which may be arranged before the DFT unit (1031-11).
- the transmitter (1031-1) first causes information to pass through a DFT (1031-11) before mapping the signal to a subcarrier.
- the signal spread (or precoded in the same sense) by the DFT unit (1031-11) is mapped to a subcarrier through a subcarrier mapper (1031-12) and then passes through an IFFT (Inverse Fast Fourier Transform) unit (1031-13) to be converted into a signal on the time axis.
- IFFT Inverse Fast Fourier Transform
- the DFT unit (1031-11) performs DFT on the input symbols and outputs complex-valued symbols. For example, if Ntx symbols are input (where Ntx is a natural number), the DFT size is Ntx.
- the DFT unit (1031-11) may be called a transform precoder.
- the subcarrier mapper (1031-12) maps the complex symbols to each subcarrier in the frequency domain. The complex symbols may be mapped to resource elements corresponding to resource blocks allocated for data transmission.
- the subcarrier mapper (1031-12) may be called a resource element mapper.
- the IFFT unit (1031-13) performs IFFT on the input symbols and outputs a baseband signal for data, which is a time-domain signal.
- the CP insertion unit (1031-14) copies a portion of the rear part of the base band signal for data and inserts it into the front part of the base band signal for data.
- CP insertion ISI (Inter-Symbol Interference) and ICI (Inter-Carrier Interference) are prevented, so that orthogonality can be maintained even in a multipath channel.
- the receiver (1031-2) includes a wireless receiving unit (1031-21), a CP removing unit (1031-22), an FFT unit (1031-23), and an equalizer unit (1031-24).
- the wireless receiving unit (1031-21), the CP removing unit (1031-22), and the FFT unit (1031-23) of the receiver (1031-2) perform the inverse functions of the wireless transmitting unit (1031-15), the CP inserting unit (1031-14), and the IFF unit (1031-13) of the transmitting terminal (1031-1).
- the receiver (1031-2) may further include a demodulator.
- the methods are described based on the flow chart as a series of steps or blocks, but the order of the steps described is not limited, and some steps may occur in a different order or simultaneously with other steps described above. Furthermore, those skilled in the art will understand that the steps depicted in the flow chart are not exclusive, and other steps may be included or one or more of the steps in the flow chart may be deleted without affecting the scope of the rights.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Signal Processing (AREA)
- Neurology (AREA)
- Probability & Statistics with Applications (AREA)
- Computer Networks & Wireless Communication (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
Description
| Frequency Range designation | Corresponding frequency range | Subcarrier Spacing |
| FR1 | 410MHz - 7125MHz | 15, 30, 60kHz |
| FR2 | 24250MHz - 52600MHz | 60, 120, 240kHz |
| μ | △f=2μ15 [kHz] | CP |
| 0 | 15 | 일반 |
| 1 | 30 | 일반 |
| 2 | 60 | 일반, 확장 |
| 3 | 120 | 일반 |
| 4 | 240 | 일반 |
| 5 | 480 | 일반 |
| 6 | 960 | 일반 |
| μ | △f=2μ15 [kHz] | Nslot symb | Nframe,μ slot | Nsubframe,μ slot |
| 0 | 15 | 14 | 10 | 1 |
| 1 | 30 | 14 | 20 | 2 |
| 2 | 60 | 14 | 40 | 4 |
| 3 | 120 | 14 | 80 | 8 |
| 4 | 240 | 14 | 160 | 16 |
| 5 | 480 | 14 | 320 | 32 |
| 6 | 960 | 14 | 640 | 64 |
| μ | SCS (15*2u) | Nslot symb | Nframe,μ slot | Nsubframe,μ slot |
| 2 | 60KHz (u=2) | 12 | 40 | 4 |
| 용어(Terminology) | 기술(Description) |
| 데이터 수집(Data collection) | AI/ML 모델 훈련, 데이터 분석 및 추론을 목적으로 네트워크 노드들, 관리 엔티티 또는 UE에 의해 데이터를 수집하는 프로세스 (A process of collecting data by the network nodes, management entity, or UE for the purpose of AI/ML model training, data analytics and inference) |
| AI/ML 모델(AI/ML Model) | 입력 세트를 기반으로 출력 세트를 생성하기 위한 AI/ML 기법을 적용하는 데이터 기반 알고리즘 (A data driven algorithm that applies AI/ML techniques to generate a set of outputs based on a set of inputs.) |
| AI/ML 모델 훈련(AI/ML model training) | 데이터 기반 방식으로 [입력/출력 관계를 학습함으로써] AI/ML 모델을 훈련하고 추론을 위해 훈련된 AI/ML 모델을 획득하는 과정 (A process to train an AI/ML Model [by learning the input/output relationship] in a data driven manner and obtain the trained AI/ML Model for inference) |
| AI/ML 모델 추론(AI/ML model Inference) | 입력 세트에 기초한 출력 세트를 생성하기 위해 훈련된 AI/ML 모델을 사용하는 프로세스 (A process of using a trained AI/ML model to produce a set of outputs based on a set of inputs) |
| AI/ML 모델 검증(AI/ML model validation) | 모델 훈련에 사용되는 것과 다른 데이터 세트를 사용하여 AI/ML 모델의 품질을 평가하는 훈련의 하위 프로세스로, 모델 훈련에 사용되는 데이터 세트 이상으로 일반화되는 모델 파라미터를 선택하는 데 도움이 됨 (A subprocess of training, to evaluate the quality of an AI/ML model using a dataset different from one used for model training, that helps selecting model parameters that generalize beyond the dataset used for model training.) |
| AI/ML 모델 테스트(AI/ML model testing) | 모델 훈련 및 검증에 사용된 것과 다른 데이터 세트를 사용하여 최종 AI/ML 모델의 성능을 평가하기 위한 훈련의 하위 프로세스. AI/ML 모델 검증과 달리 테스트에서는 모델의 후속 조정을 가정하지 않음. (A subprocess of training, to evaluate the performance of a final AI/ML model using a dataset different from one used for model training and validation. Differently from AI/ML model validation, testing does not assume subsequent tuning of the model.) |
| UE측 (AI/ML) 모델(UE-side (AI/ML) model) | 추론이 전적으로 UE에서 수행되는 AI/ML 모델 (An AI/ML Model whose inference is performed entirely at the UE) |
| 네트워크측 (AI/ML) 모델(Network-side (AI/ML) model) | 추론이 전적으로 네트워크에서 수행되는 AI/ML 모델 (An AI/ML Model whose inference is performed entirely at the network) |
| 단측 (AI/ML) 모델(One-sided (AI/ML) model) | UE 측 (AI/ML) 모델 또는 네트워크 측 (AI/ML) 모델 (A UE-side (AI/ML) model or a Network-side (AI/ML) model) |
| 양측 (AI/ML) 모델(Two-sided (AI/ML) model) | 공동 추론이 수행되는 쌍을 이루는 AI/ML 모델(들). 여기서, 공동 추론은 추론이 UE와 네트워크를 통해 공동으로 수행되는 AI/ML 추론으로, 즉, 추론의 첫번째 부분은 UE에 의해 먼저 수행되고 나머지 부분은 gNB에 의해 수행되거나 또는 그 반대의 경우도 해당됨. (A paired AI/ML Model(s) over which joint inference is performed, where joint inference comprises AI/ML Inference whose inference is performed jointly across the UE and the network, i.e, the first part of inference is firstly performed by UE and then the remaining part is performed by gNB, or vice versa.) |
| AI/ML 모델 전송(AI/ML model transfer) | 수신측에 알려진 모델 구조의 파라미터 또는 파라미터를 가지는 새 모델을 무선 인터페이스를 통해 AI/ML 모델의 전달. 전달에는 전체 모델 또는 부분 모델이 포함될 수 있음. (Delivery of an AI/ML model over the air interface, either parameters of a model structure known at the receiving end or a new model with parameters. Delivery may contain a full model or a partial model.) |
| 모델 다운로드(Model download) | 네트워크에서 UE로 모델 전송 (Model transfer from the network to UE) |
| 모델 업로드(Model upload) | UE에서 네트워크로 모델 전송 (Model transfer from UE to the network) |
| 연합 학습 / 연합 훈련(Federated learning / federated training) | 로컬 데이터 샘플을 사용하여 각각 로컬 모델 훈련을 수행하는 여러 분산형 에지 노드 (예로, UE, gNB)에서 AI/ML 모델을 훈련하는 머신 러닝 기술. 이 기술에서는 모델의 여러 상호 작용이 필요하지만 로컬 데이터 샘플의 교환은 필요하지 않음. (A machine learning technique that trains an AI/ML model across multiple decentralized edge nodes (e.g., UEs, gNBs) each performing local model training using local data samples. The technique requires multiple interactions of the model, but no exchange of local data samples.) |
| 오프라인 필드 데이터(Offline field data) | 필드에서 수집되어 AI/ML 모델의 오프라인 교육에 사용되는 데이터 (The data collected from field and used for offline training of the AI/ML model) |
| 온라인 필드 데이터(Online field data) | 필드에서 수집되어 AI/ML 모델의 온라인 교육에 사용되는 데이터 (The data collected from field and used for online training of the AI/ML model) |
| 모델 모니터링(Model monitoring) | AI/ML 모델의 추론 성능을 모니터링하는 절차 (A procedure that monitors the inference performance of the AI/ML model) |
| 지도 학습(Supervised learning) | 입력 및 해당 레이블로부터 모델을 훈련하는 프로세스 (A process of training a model from input and its corresponding labels.) |
| 비지도 학습(Unsupervised learning) | 레이블이 지정된 데이터가 없이 모델을 훈련하는 프로세스 (A process of training a model without labelled data.) |
| 준지도 학습(Semi-supervised learning ) | 레이블이 지정된 데이터와 레이블이 지정되지 않은 데이터가 혼합된 모델을 훈련하는 프로세스 (A process of training a model with a mix of labelled data and unlabelled data) |
| 강화 학습 (RL)Reinforcement Learning (RL) | 모델이 상호 작용하는 환경에서 입력(상태)과 모델의 출력(작업)에 따른 피드백 신호(보상)로부터 AI/ML 모델을 훈련하는 프로세스 (A process of training an AI/ML model from input (a.k.a. state) and a feedback signal (a.k.a. reward) resulting from the model's output (a.k.a. action) in an environment the model is interacting with.) |
| 모델 활성화 (Model activation) |
특정 기능에 대한 AI/ML 모델 활성화 (enable an AI/ML model for a specific function) |
| 모델 비활성화(Model deactivation) | 특정 기능에 대한 AI/ML 모델 비활성화 (disable an AI/ML model for a specific function) |
| 모델 스위칭(Model switching) | 특정 기능에 대해, 현재 활성화된 AI/ML 모델을 비활성화하고 다른 AI/ML 모델을 활성화 Deactivating a currently active AI/ML model and activating a different AI/ML model for a specific function |
Claims (18)
- 무선 통신 시스템에서 단말의 동작 방법에 있어서,AI/ML(artificial intelligence/machine learning) 모델 훈련을 위한 데이터를 수신하는 단계;상기 데이터를 기초로, 상기 AI/ML 모델 훈련을 수행하는 단계; 및상기 AI/ML 모델 훈련의 수행 후, 상기 데이터의 수집 종료를 지시하는 제1 메시지를 전송하는 단계를 포함하는, 방법.
- 제1항에 있어서,상기 데이터의 수신을 위한 설정 정보를 수신하는 단계를 더 포함하고,상기 설정 정보는 상기 데이터의 수집 종료와 연관된 이벤트 정보를 포함하며, 상기 이벤트 정보를 기초로 상기 제1 메시지가 전송되는, 방법.
- 제1항에 있어서,상기 이벤트 정보는, AI/ML 모델의 성능과 연관된 임계치(threshold) 정보 및 상기 AI/ML 모델 훈련의 종료를 위한 시간 정보 중 적어도 하나를 포함하는, 방법.
- 제1항에 있어서,상기 제1 메시지는, 상기 데이터의 전송 비활성화를 위한 것인, 방법.
- 제2항에 있어서,상기 데이터의 전송을 요청하는 제2 메시지를 전송하는 단계를 더 포함하고,상기 설정 정보는, 상기 제2 메시지를 기초로 수신되는, 방법.
- 제5항에 있어서,상기 제2 메시지는, 상기 AI/ML 모델 훈련의 종료를 위한 시간 정보 및 데이터 양 정보 중 적어도 하나를 포함하는, 방법.
- 무선 통신 시스템에서 기지국의 동작 방법에 있어서,AI/ML(artificial intelligence/machine learning) 모델 훈련을 위한 데이터를 전송하는 단계; 및상기 AI/ML 모델 훈련의 수행 후, 상기 데이터의 수집 종료를 지시하는 제1 메시지를 수신하는 단계를 포함하는, 방법.
- 제7항에 있어서,상기 데이터의 전송을 위한 설정 정보를 전송하는 단계를 더 포함하고,상기 설정 정보는 상기 데이터의 수집 종료와 연관된 이벤트 정보를 포함하며, 상기 이벤트 정보를 기초로 상기 제1 메시지가 수신되는, 방법.
- 제7항에 있어서,상기 이벤트 정보는, AI/ML 모델의 성능과 연관된 임계치(threshold) 정보 및 상기 AI/ML 모델 훈련의 종료를 위한 시간 정보 중 적어도 하나를 포함하는, 방법.
- 제7항에 있어서,상기 제1 메시지는, 상기 데이터의 전송 비활성화를 위한 것인, 방법.
- 제8항에 있어서,상기 데이터의 전송을 요청하는 제2 메시지를 수신하는 단계를 더 포함하고,상기 설정 정보는, 상기 제2 메시지를 기초로 전송되는, 방법.
- 제11항에 있어서,상기 제2 메시지는, 상기 AI/ML 모델 훈련의 종료를 위한 시간 정보 및 데이터 양 정보 중 적어도 하나를 포함하는, 방법.
- 무선 통신 시스템에서의 통신 기기로서,적어도 하나의 프로세서; 및명령어(instructions)를 저장하고, 상기 적어도 하나의 프로세서와 동작 가능하게(operably) 전기적으로 연결가능한, 적어도 하나의 메모리를 포함하고, 상기 명령어가 상기 적어도 하나의 프로세서에 의해서 실행되는 것에 기초하여, 수행되는 동작은:AI/ML(artificial intelligence/machine learning) 모델 훈련을 위한 데이터를 수신하는 단계와,상기 데이터를 기초로, 상기 AI/ML 모델 훈련을 수행하는 단계와, 그리고상기 AI/ML 모델 훈련의 수행 후, 상기 데이터의 수집 종료를 지시하는 제1 메시지를 전송하는 단계를 포함하는, 통신 기기.
- 제13항에 있어서,상기 명령어가 상기 적어도 하나의 프로세서에 의해서 실행되는 것에 기초하여, 수행되는 동작은:상기 데이터의 수신을 위한 설정 정보를 수신하는 단계를 더 포함하고,상기 설정 정보는 상기 데이터의 수집 종료와 연관된 이벤트 정보를 포함하며, 상기 이벤트 정보를 기초로 상기 제1 메시지가 전송되는, 통신 기기.
- 제13항에 있어서,상기 이벤트 정보는, AI/ML 모델의 성능과 연관된 임계치(threshold) 정보 및 상기 AI/ML 모델 훈련의 종료를 위한 시간 정보 중 적어도 하나를 포함하는, 통신 기기.
- 제13항에 있어서,상기 제1 메시지는, 상기 데이터의 전송 비활성화를 위한 것인, 통신 기기.
- 제14항에 있어서,상기 명령어가 상기 적어도 하나의 프로세서에 의해서 실행되는 것에 기초하여, 수행되는 동작은:상기 데이터의 전송을 요청하는 제2 메시지를 전송하는 단계를 더 포함하고,상기 설정 정보는, 상기 제2 메시지를 기초로 수신되는, 통신 기기.
- 제17항에 있어서,상기 제2 메시지는, 상기 AI/ML 모델 훈련의 종료를 위한 시간 정보 및 데이터 양 정보 중 적어도 하나를 포함하는, 통신 기기.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| EP24836348.3A EP4723506A1 (en) | 2023-07-06 | 2024-07-04 | Method and apparatus for training model for artificial intelligence/machine learning-based communication |
| CN202480045560.0A CN121464585A (zh) | 2023-07-06 | 2024-07-04 | 用于基于人工智能和/或机器学习的通信的模型训练方法及装置 |
Applications Claiming Priority (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| KR20230087741 | 2023-07-06 | ||
| KR10-2023-0087741 | 2023-07-06 | ||
| KR1020240085445A KR20250007985A (ko) | 2023-07-06 | 2024-06-28 | 인공지능 및/또는 머신러닝 기반 통신을 위한 모델 훈련 방법 및 장치 |
| KR10-2024-0085445 | 2024-06-28 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2025009901A1 true WO2025009901A1 (ko) | 2025-01-09 |
Family
ID=94172348
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/KR2024/009443 Ceased WO2025009901A1 (ko) | 2023-07-06 | 2024-07-04 | 인공지능 및/또는 머신러닝 기반 통신을 위한 모델 훈련 방법 및 장치 |
Country Status (3)
| Country | Link |
|---|---|
| EP (1) | EP4723506A1 (ko) |
| CN (1) | CN121464585A (ko) |
| WO (1) | WO2025009901A1 (ko) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220377844A1 (en) * | 2021-05-18 | 2022-11-24 | Qualcomm Incorporated | Ml model training procedure |
| CN116264684A (zh) * | 2021-12-10 | 2023-06-16 | 华为技术有限公司 | 一种无线网络中的人工智能ai模型训练方法及装置 |
| US20230209368A1 (en) * | 2020-05-21 | 2023-06-29 | Lg Electronics Inc. | Wireless communication method using on-device learning-based machine learning network |
-
2024
- 2024-07-04 WO PCT/KR2024/009443 patent/WO2025009901A1/ko not_active Ceased
- 2024-07-04 EP EP24836348.3A patent/EP4723506A1/en active Pending
- 2024-07-04 CN CN202480045560.0A patent/CN121464585A/zh active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20230209368A1 (en) * | 2020-05-21 | 2023-06-29 | Lg Electronics Inc. | Wireless communication method using on-device learning-based machine learning network |
| US20220377844A1 (en) * | 2021-05-18 | 2022-11-24 | Qualcomm Incorporated | Ml model training procedure |
| CN116264684A (zh) * | 2021-12-10 | 2023-06-16 | 华为技术有限公司 | 一种无线网络中的人工智能ai模型训练方法及装置 |
Non-Patent Citations (2)
| Title |
|---|
| MINGJU LI, XIAOMI: "Potential specification impact on AI/ML for beam management", 3GPP DRAFT; R1-2302978; TYPE DISCUSSION; FS_NR_AIML_AIR, vol. RAN WG1, 7 April 2023 (2023-04-07), pages 1 - 17, XP052293548 * |
| ZHENGXUAN LIU, XIAOMI: "Further discussion on specification impact for CSI feedback based on AI/ML", 3GPP DRAFT; R1-2304894; TYPE DISCUSSION; FS_NR_AIML_AIR, vol. RAN WG1, 15 May 2023 (2023-05-15), pages 1 - 17, XP052310349 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN121464585A (zh) | 2026-02-03 |
| EP4723506A1 (en) | 2026-04-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2024035018A1 (ko) | 하향링크 신호를 수신하는 방법, 사용자기기, 프로세싱 장치, 저장 매체 및 컴퓨터 프로그램, 그리고 하향링크 신호를 전송하는 방법 및 기지국 | |
| WO2021157998A1 (en) | Method and apparatus for performing communication in wireless communication system | |
| WO2022270964A1 (en) | Method and apparatus for handling ul timing alignment upon receiving tci state update signal in a wireless communication system | |
| WO2018174586A1 (ko) | 빔 회복 과정 수행 방법과 사용자기기, 및 빔 회복 과정 지원 방법 및 기지국 | |
| WO2023048541A1 (en) | System and method of transmitting and receiving system information by reduced capability ues | |
| WO2016013901A1 (ko) | 단말 간 통신을 지원하는 무선 통신 시스템에서 파워 제어 방법 및 이를 위한 장치 | |
| WO2020141963A1 (en) | Method and apparatus for signaling multiple resources for message 3 (msg3) transmissions | |
| WO2022025613A1 (ko) | 무선 통신 시스템에서 사이드링크의 비연속적 수신을 지원하기 위한 방법 및 장치 | |
| WO2024035027A1 (ko) | 하향링크 신호를 수신하는 방법, 사용자기기, 프로세싱 장치 및 저장 매체, 그리고 하향링크 신호를 전송하는 방법 및 기지국 | |
| WO2024029911A1 (ko) | 하향링크 채널을 수신하는 방법, 사용자기기, 프로세싱 장치, 저장 매체 및 컴퓨터 프로그램, 그리고 하향링크 채널을 전송하는 방법 및 기지국 | |
| WO2024035058A1 (ko) | 하향링크 신호를 수신하는 방법, 사용자기기, 프로세싱 장치 및 저장 매체, 그리고 하향링크 신호를 전송하는 방법 및 기지국 | |
| WO2023163471A1 (ko) | 무선 통신 시스템에서 상향링크 송수신을 수행하는 방법 및 장치 | |
| WO2023014143A1 (ko) | 무선 통신 시스템에서 상향링크 송수신을 수행하는 방법 및 장치 | |
| WO2021141389A1 (ko) | 사이드링크를 위한 자원할당 방법 | |
| WO2025009901A1 (ko) | 인공지능 및/또는 머신러닝 기반 통신을 위한 모델 훈련 방법 및 장치 | |
| WO2024072182A1 (ko) | 무선 통신 시스템에서 통신을 수행하기 위한 방법, 네트워크 노드, 프로세싱 장치 및 저장 매체 | |
| WO2024253486A1 (ko) | 인공지능 및/또는 머신러닝 모델 추론을 위한 빔 측정과 결과 보고 방법 및 장치 | |
| WO2024242527A1 (ko) | 인공지능 및/또는 머신러닝 모델 추론을 위한 빔 설정과 측정 방법 및 장치 | |
| WO2024242398A1 (ko) | 인공지능 및/또는 머신러닝 모델 추론을 위한 빔 측정과 결과 보고 방법 및 장치 | |
| WO2022139394A1 (ko) | 무선 통신 시스템에서 빔 실패 복구 방법 및 장치 | |
| WO2025116500A1 (ko) | 인공지능 및/또는 머신러닝 모델의 성능 모니터링 주기를 변경하는 방법 및 장치 | |
| WO2024162826A1 (en) | Method and apparatus for application layer measurement reporting in unlicensed spectrum | |
| WO2024237680A1 (ko) | 인공지능 및/또는 머신러닝을 이용한 빔 관리에서 측정 결과 보고 방법 및 장치 | |
| WO2025071238A1 (ko) | 인공지능 및 머신러닝 모델의 기능을 인식하는 방법 및 장치 | |
| WO2024167159A1 (ko) | 인공지능 및/또는 머신러닝을 이용한 빔 관리 방법 및 장치 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 24836348 Country of ref document: EP Kind code of ref document: A1 |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 2024836348 Country of ref document: EP |
|
| ENP | Entry into the national phase |
Ref document number: 2024836348 Country of ref document: EP Effective date: 20251231 |
|
| ENP | Entry into the national phase |
Ref document number: 2024836348 Country of ref document: EP Effective date: 20251231 |
|
| ENP | Entry into the national phase |
Ref document number: 2024836348 Country of ref document: EP Effective date: 20251231 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| ENP | Entry into the national phase |
Ref document number: 2024836348 Country of ref document: EP Effective date: 20251231 |
|
| WWP | Wipo information: published in national office |
Ref document number: 2024836348 Country of ref document: EP |