KR20230153293A - Method and apparatus for configuring artificial neural network for wireless communication in mobile communication system - Google Patents

Abstract

A method and device for configuring an artificial neural network for wireless communication in a mobile communication system are disclosed. The method of the base station includes receiving a capability information reporting message related to an artificial neural network from each of the terminals in the base station; generating artificial neural network configuration information based on the capabilities of the base station and the capability report messages received from each of the terminals; and transmitting the generated artificial neural network configuration information to terminals within the base station.

Description

Method and device for configuring artificial neural network for wireless communication in mobile communication system {METHOD AND APPARATUS FOR CONFIGURING ARTIFICIAL NEURAL NETWORK FOR WIRELESS COMMUNICATION IN MOBILE COMMUNICATION SYSTEM}

This disclosure relates to technology for configuring an artificial neural network, and more specifically, to technology for configuring an artificial neural network for wireless communication in a mobile communication system.

3GPP, an international standardization organization, has selected the Artificial Intelligence (AI)/Machine Learning (ML) application plan for the NR Air Interface as a future Release 18 Study Item (SI). . The purpose of SI in Release 18 is to establish use cases that can utilize AI/ML in the NR wireless interface and confirm performance gains according to AI/ML application for each use case.

Specifically, in mobile communication systems, Channel State Information (CSI) Feedback Enhancement, Beam Management, and Positioning Accuracy Enhancement were selected as representative use cases.

CSI feedback means that the terminal reports channel status information to support the base station in applying transmission techniques such as Multiple Input Multiple Output (MIMO) or precoding in a mobile communication system. . The 5G NR standard defined by 3GPP supports feedback information such as Channel Quality Indicator (CQI), Precoding Matrix Indicator (PMI), and Rank Indicator (RI) in relation to the CSI feedback method. do. In NR systems, discussions are ongoing to improve the CSI feedback technique to effectively support transmission techniques such as multi-user MIMO (MU-MIMO).

The purpose of the present disclosure to solve the above needs is to categorize the configuration and learning methods of artificial neural networks that can exist in various ways in a mobile communication system consisting of a base station and one or more terminals, so that the base station and the terminal each have To facilitate management and implementation.

A method of a base station according to a first embodiment of the disclosure for achieving the above object includes receiving a capability information reporting message related to an artificial neural network from each of the terminals in the base station; generating artificial neural network configuration information based on the capabilities of the base station and the capability report messages received from each of the terminals; and transmitting the generated artificial neural network setting information to the terminals within the base station, wherein the artificial neural network setting information is a type in which the artificial neural network is shared on a cell basis and the artificial neural network on a terminal group basis. It may include information indicating one of the shared types, the type in which the artificial neural network is shared on a terminal basis, or the type in which the artificial neural network is shared on a cell group basis.

The capability information reporting message related to the artificial neural network includes a type of the artificial neural network shared on a cell basis supported by each of the terminals, a type of the artificial neural network shared on a terminal group basis, and the artificial neural network on a terminal basis. It may include information indicating one of the shared type or the type in which the artificial neural network is shared on a cell group basis.

When the artificial neural network setting information indicates a type in which the artificial neural network is shared for each terminal group, the terminal group may include terminals with the same classification indicator.

When the artificial neural network setting information indicates a type in which the artificial neural network is shared on a cell group basis, the cell group may include cells having the same classification indicator.

The artificial neural network setting information may be a type that does not support online learning, or a type that only the terminals support the online learning, or a type that supports the online learning only in the base station, or both the base station and the terminals support the online learning. It may further include information indicating one of the supported types.

The artificial neural network setting information may further include at least one of time interval information set to report online learning information from each of the terminals or time point information for collecting the online learning from each of the terminals.

The method of the base station may further include receiving an online learning information report message from the terminals based on the artificial neural network setting information,

The online learning information report message may include the entire model of the artificial neural network or a difference value between the artificial neural network before online learning and the artificial neural network after online learning.

The method of the base station includes determining an update to the weight vector of the artificial neural network shared with the terminals based on the received online learning report message; and transmitting weight update information to the terminals based on the decision to update the weight vector of the artificial neural network.

A method of a terminal according to a second embodiment of the disclosure for achieving the above object includes receiving a request message for reporting capability information for an artificial neural network from a base station of a cell to which the terminal belongs; transmitting a capability information report message related to the artificial neural network to the base station in response to receiving the capability information report request message; Receiving artificial neural network configuration information from the base station; and configuring an artificial neural network in the terminal based on the received artificial neural network setting information, wherein the artificial neural network setting information includes a type of the artificial neural network being shared on a cell basis and a terminal group unit. It may include information indicating one of the following: a type in which the artificial neural network is shared, a type in which the artificial neural network is shared on a terminal basis, or a type in which the artificial neural network is shared on a cell group basis.

The capability information reporting message related to the artificial neural network includes a type of the artificial neural network shared on a cell basis supported by the terminal, a type of the artificial neural network shared on a terminal group basis, and a type of the artificial neural network shared on a terminal basis. It may include information indicating one of the types, or the type in which the artificial neural network is shared on a cell group basis.

The artificial neural network setting information is a type that does not support online learning, or a type that supports the online learning only in the terminal, or a type that supports the online learning only in the base station, or a type that supports the online learning in both the base station and the terminal. It may further contain information indicating one of the supported types.

The artificial neural network setting information may further include at least one of time interval information set to report online learning information from the terminal or time point information for collecting online learning from the terminal.

The method of the terminal includes online learning about the artificial neural network set in the terminal; generating an online learning information report message based on the online learning; and transmitting the online learning information reporting message to the base station based on time interval information set to report the online learning information or time point information for collecting the online learning.

A base station according to a third embodiment of the disclosure for achieving the above object includes a processor, wherein the base station: receives a capability information reporting message related to an artificial neural network from each of the terminals in the base station; Generate artificial neural network configuration information based on the capabilities of the base station and the capability report message received from each of the terminals; And may cause the generated artificial neural network setting information to be transmitted to the terminals in the base station, and the artificial neural network setting information may include a type in which the artificial neural network is shared on a cell basis and a type in which the artificial neural network is shared on a terminal group basis. It may include information indicating one of the types, a type in which the artificial neural network is shared on a terminal basis, or a type in which the artificial neural network is shared in a cell group unit.

The capability information reporting message related to the artificial neural network includes a type of the artificial neural network shared on a cell basis supported by each of the terminals, a type of the artificial neural network shared on a terminal group basis, and the artificial neural network on a terminal basis. It may include information indicating one of the shared type or the type in which the artificial neural network is shared on a cell group basis.

If the artificial neural network setting information indicates a type in which the artificial neural network is shared on a per-cell group basis, the terminal group includes terminals with the same classification indicator, and the artificial neural network setting information indicates a type of the artificial neural network being shared on a per-cell group basis. When the neural network indicates a shared type, the cell group may include cells with the same classification indicator.

The artificial neural network setting information may be a type that does not support online learning, or a type that only the terminals support the online learning, or a type that supports the online learning only in the base station, or both the base station and the terminals support the online learning. It may further include information indicating one of the supported types.

The artificial neural network setting information may further include at least one of time interval information set to report online learning information from each of the terminals or time point information for collecting the online learning from each of the terminals.

The processor may further cause the base station to: receive an online learning information report message from the terminals based on the artificial neural network setting information, and the online learning information report message may include the entire model of the artificial neural network or the online learning information. It may include a difference value between the artificial neural network before and the artificial neural network after the online learning.

The processor may cause the base station to: determine an update to the weight vector of the artificial neural network shared with the terminals based on the received online learning report message; and may further cause weight update information to be transmitted to the terminals based on the update decision for the weight vector of the artificial neural network.

According to an embodiment of the present disclosure, an artificial neural network can be set in each of the base station and the terminal, and there is an advantage in that communication using the artificial neural network can be performed more smoothly through this setting.

1 is a conceptual diagram illustrating an embodiment of a communication system.
Figure 2 is a block diagram showing an embodiment of a communication node constituting a communication system.
FIG. 3A is a conceptual diagram illustrating sharing (or creating) an artificial neural network according to Type 0 of the first embodiment of the present disclosure.
FIG. 3B is a conceptual diagram illustrating sharing (or creating) an artificial neural network according to Type 1 of the first embodiment of the present disclosure.
FIG. 3C is a conceptual diagram illustrating sharing (or creating) an artificial neural network according to Type 2 of the first embodiment of the present disclosure.
FIG. 3D is a conceptual diagram illustrating sharing (or creating) an artificial neural network according to Type 3 of the first embodiment of the present disclosure.
FIG. 4A is a conceptual diagram for explaining online learning of an artificial neural network according to Type B of the third embodiment of the present disclosure.
Figure 4b is a conceptual diagram for explaining online learning of an artificial neural network according to Type C of the third embodiment of the present disclosure.
FIG. 4C is a conceptual diagram for explaining online learning of an artificial neural network according to Type D of the third embodiment of the present disclosure.
Figure 5 is a timing diagram for explaining artificial neural network settings and learning information reporting according to the sixth embodiment of the present disclosure.
Figure 6 is a signal flow diagram when learning an artificial neural network is performed between a terminal and a base station according to the initiation.

Since the present disclosure can make various changes and have various embodiments, specific embodiments will be illustrated in the drawings and described in detail. However, this is not intended to limit the present disclosure to specific embodiments, and should be understood to include all changes, equivalents, and substitutes included in the spirit and technical scope of the present disclosure.

Terms such as first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another. For example, a first component may be referred to as a second component, and similarly, the second component may be referred to as a first component without departing from the scope of the present disclosure. The term and/or includes any of a plurality of related stated items or a combination of a plurality of related stated items.

When a component is said to be "connected" or "connected" to another component, it is understood that it may be directly connected to or connected to the other component, but that other components may exist in between. It should be. On the other hand, when it is mentioned that a component is “directly connected” or “directly connected” to another component, it should be understood that there are no other components in between.

The terms used in this disclosure are only used to describe specific embodiments and are not intended to limit the disclosure. Singular expressions include plural expressions unless the context clearly dictates otherwise. In the present disclosure, terms such as “comprise” or “have” are intended to designate the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, but are not intended to indicate the presence of one or more other features. It should be understood that this does not exclude in advance the possibility of the existence or addition of elements, numbers, steps, operations, components, parts, or combinations thereof.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the technical field to which this disclosure pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related technology, and unless clearly defined in the present disclosure, should not be interpreted in an idealized or excessively formal sense. No.

A communication system to which embodiments according to the present disclosure are applied will be described. Communication systems to which embodiments according to the present disclosure are applied are not limited to those described below, and embodiments according to the present disclosure can be applied to various communication systems. Here, communication system may be used in the same sense as communication network.

Throughout the specification, network refers to, for example, wireless Internet such as WiFi (wireless fidelity), mobile Internet such as WiBro (wireless broadband internet) or WiMax (world interoperability for microwave access), and GSM (global system for mobile communication). ) or 2G mobile communication networks such as CDMA (code division multiple access), 3G mobile communication networks such as WCDMA (wideband code division multiple access) or CDMA2000, HSDPA (high speed downlink packet access) or HSUPA (high speed uplink packet access) It may include a 3.5G mobile communication network, a 4G mobile communication network such as an LTE (long term evolution) network or an LTE-Advanced network, and a 5G mobile communication network.

Throughout the specification, terminal refers to a mobile station, mobile terminal, subscriber station, portable subscriber station, user equipment, and access terminal. It may refer to the like, and may include all or part of the functions of a terminal, a mobile station, a mobile terminal, a subscriber station, a portable subscriber station, a user device, an access terminal, etc.

Here, a desktop computer, laptop computer, tablet PC, wireless phone, mobile phone, smart phone, and smart watch that can communicate with terminals. (smart watch), smart glass, e-book reader, PMP (portable multimedia player), portable game console, navigation device, digital camera, DMB (digital multimedia broadcasting) player, digital voice digital audio recorder, digital audio player, digital picture recorder, digital picture player, digital video recorder, digital video player ), etc. can be used.

Throughout the specification, base station refers to an access point, radio access station, node B, evolved node B, base transceiver station, and MMR ( It may refer to a mobile multihop relay)-BS, etc., and may include all or part of the functions of a base station, access point, wireless access station, Node B, eNodeB, transmitting and receiving base station, and MMR-BS.

Hereinafter, preferred embodiments of the present disclosure will be described in more detail with reference to the attached drawings. In order to facilitate overall understanding in explaining the present disclosure, the same reference numerals are used for the same components in the drawings, and duplicate descriptions of the same components are omitted.

1 is a conceptual diagram illustrating an embodiment of a communication system.

Referring to FIG. 1, the communication system 100 includes a plurality of communication nodes 110-1, 110-2, 110-3, 120-1, 120-2, 130-1, 130-2, 130-3, 130-4, 130-5, 130-6). A plurality of communication nodes are 4G communication (e.g., long term evolution (LTE), LTE-A (advanced)), 5G communication (e.g., new radio (NR)) specified in the 3rd generation partnership project (3GPP) standard. ), etc. can be supported. 4G communications can be performed in frequency bands below 6 GHz, and 5G communications can be performed in frequency bands above 6 GHz as well as below 6 GHz.

For example, for 4G communication and 5G communication, a plurality of communication nodes may use a communication protocol based on code division multiple access (CDMA), a communication protocol based on wideband CDMA (WCDMA), a communication protocol based on time division multiple access (TDMA), Communication protocol based on FDMA (frequency division multiple access), communication protocol based on OFDM (orthogonal frequency division multiplexing), communication protocol based on Filtered OFDM, communication protocol based on CP (cyclic prefix)-OFDM, DFT-s-OFDM (discrete Fourier transform-spread-OFDM)-based communication protocol, OFDMA (orthogonal frequency division multiple access)-based communication protocol, SC (single carrier)-FDMA-based communication protocol, NOMA (Non-orthogonal Multiple Access), GFDM (generalized frequency) division multiplexing)-based communication protocols, FBMC (filter bank multi-carrier)-based communication protocols, UFMC (universal filtered multi-carrier)-based communication protocols, and SDMA (Space Division Multiple Access)-based communication protocols. .

Additionally, the communication system 100 may further include a core network. If the communication system 100 supports 4G communication, the core network may include a serving-gateway (S-GW), a packet data network (PDN)-gateway (P-GW), a mobility management entity (MME), etc. there is. When the communication system 100 supports 5G communication, the core network may include a user plane function (UPF), a session management function (SMF), and an access and mobility management function (AMF).

Meanwhile, a plurality of communication nodes (110-1, 110-2, 110-3, 120-1, 120-2, 130-1, 130-2, 130-3, 130-) constituting the communication system 100 4, 130-5, 130-6) Each can have the following structure.

Figure 2 is a block diagram showing an embodiment of a communication node constituting a communication system.

Referring to FIG. 2, the communication node 200 may include at least one processor 210, a memory 220, and a transmitting and receiving device 230 that is connected to a network and performs communication. Additionally, the communication node 200 may further include an input interface device 240, an output interface device 250, a storage device 260, etc. Each component included in the communication node 200 is connected by a bus 270 and can communicate with each other.

However, each component included in the communication node 200 may be connected through an individual interface or individual bus centered on the processor 210, rather than the common bus 270. For example, the processor 210 may be connected to at least one of the memory 220, the transmission/reception device 230, the input interface device 240, the output interface device 250, and the storage device 260 through a dedicated interface. .

The processor 210 may execute a program command stored in at least one of the memory 220 and the storage device 260. The processor 210 may refer to a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to embodiments of the present disclosure are performed. Each of the memory 220 and the storage device 260 may be comprised of at least one of a volatile storage medium and a non-volatile storage medium. For example, the memory 220 may be comprised of at least one of read only memory (ROM) and random access memory (RAM).

Referring again to FIG. 1, the communication system 100 includes a plurality of base stations (110-1, 110-2, 110-3, 120-1, 120-2) and a plurality of terminals (130- 1, 130-2, 130-3, 130-4, 130-5, 130-6). Base stations (110-1, 110-2, 110-3, 120-1, 120-2) and terminals (130-1, 130-2, 130-3, 130-4, 130-5, 130-6) The communication system 100 that includes may be referred to as an “access network.” Each of the first base station 110-1, the second base station 110-2, and the third base station 110-3 may form a macro cell. Each of the fourth base station 120-1 and the fifth base station 120-2 may form a small cell. The fourth base station 120-1, the third terminal 130-3, and the fourth terminal 130-4 may belong to the cell coverage of the first base station 110-1. The second terminal 130-2, the fourth terminal 130-4, and the fifth terminal 130-5 may belong to the cell coverage of the second base station 110-2. The fifth base station 120-2, the fourth terminal 130-4, the fifth terminal 130-5, and the sixth terminal 130-6 may belong to the cell coverage of the third base station 110-3. there is. The first terminal 130-1 may belong to the cell coverage of the fourth base station 120-1. The sixth terminal 130-6 may belong to the cell coverage of the fifth base station 120-2.

Here, each of the plurality of base stations 110-1, 110-2, 110-3, 120-1, and 120-2 includes a NodeB, an evolved NodeB, a base transceiver station (BTS), Radio base station, radio transceiver, access point, access node, road side unit (RSU), radio remote head (RRH), transmission point (TP), TRP ( transmission and reception point), eNB, gNB, etc.

A plurality of terminals (130-1, 130-2, 130-3, 130-4, 130-5, and 130-6) each include a user equipment (UE), a terminal, an access terminal, and a mobile device. Terminal, station, subscriber station, mobile station, portable subscriber station, node, device, IoT (Internet of Thing) It may be referred to as a device, a mounted device (mounted module/device/terminal or on board device/terminal, etc.), etc.

Meanwhile, each of the plurality of base stations 110-1, 110-2, 110-3, 120-1, and 120-2 may operate in different frequency bands or may operate in the same frequency band. Each of the plurality of base stations 110-1, 110-2, 110-3, 120-1, and 120-2 may be connected to each other through an ideal backhaul link or a non-ideal backhaul link. , information can be exchanged with each other through an ideal backhaul link or a non-ideal backhaul link. Each of the plurality of base stations 110-1, 110-2, 110-3, 120-1, and 120-2 may be connected to the core network through an ideal backhaul link or a non-ideal backhaul link. Each of the plurality of base stations (110-1, 110-2, 110-3, 120-1, 120-2) transmits the signal received from the core network to the corresponding terminal (130-1, 130-2, 130-3, 130). -4, 130-5, 130-6), and the signal received from the corresponding terminal (130-1, 130-2, 130-3, 130-4, 130-5, 130-6) is sent to the core network. can be transmitted to.

In addition, each of the plurality of base stations 110-1, 110-2, 110-3, 120-1, and 120-2 performs MIMO transmission (e.g., single user (SU)-MIMO, multi user (MU)- MIMO, massive MIMO, etc.), coordinated multipoint (CoMP) transmission, carrier aggregation (CA) transmission, transmission in an unlicensed band, direct device to device communication (D2D) (or ProSe ( proximity services)), etc. can be supported. Here, each of the plurality of terminals 130-1, 130-2, 130-3, 130-4, 130-5, and 130-6 is connected to a base station 110-1, 110-2, 110-3, and 120-1. , 120-2) and operations corresponding to those supported by the base stations 110-1, 110-2, 110-3, 120-1, and 120-2. For example, the second base station 110-2 may transmit a signal to the fourth terminal 130-4 based on the SU-MIMO method, and the fourth terminal 130-4 may transmit a signal to the fourth terminal 130-4 based on the SU-MIMO method. A signal can be received from the second base station 110-2. Alternatively, the second base station 110-2 may transmit a signal to the fourth terminal 130-4 and the fifth terminal 130-5 based on the MU-MIMO method, and the fourth terminal 130-4 and the fifth terminal 130-5 can each receive a signal from the second base station 110-2 by the MU-MIMO method.

Each of the first base station 110-1, the second base station 110-2, and the third base station 110-3 may transmit a signal to the fourth terminal 130-4 based on the CoMP method, and the fourth terminal 130-4 may transmit a signal to the fourth terminal 130-4. The terminal 130-4 can receive signals from the first base station 110-1, the second base station 110-2, and the third base station 110-3 using the CoMP method. Each of a plurality of base stations (110-1, 110-2, 110-3, 120-1, 120-2) has a terminal (130-1, 130-2, 130-3, 130-4) within its cell coverage. , 130-5, 130-6), and signals can be transmitted and received based on the CA method. The first base station 110-1, the second base station 110-2, and the third base station 110-3 each control D2D between the fourth terminal 130-4 and the fifth terminal 130-5. and each of the fourth terminal 130-4 and the fifth terminal 130-5 can perform D2D under the control of each of the second base station 110-2 and the third base station 110-3. .

Next, methods for setting and managing a wireless interface in a communication system will be described. Even when a method (e.g., transmission or reception of a signal) performed in a first communication node among communication nodes is described, the corresponding second communication node is described as a method (e.g., transmitting or receiving a signal) corresponding to the method performed in the first communication node. For example, reception or transmission of a signal) can be performed. That is, when the operation of the terminal is described, the corresponding base station can perform the operation corresponding to the operation of the terminal. Conversely, when the operation of the base station is described, the corresponding terminal can perform the operation corresponding to the operation of the base station.

Meanwhile, in a communication system, a base station can perform all functions of a communication protocol (eg, remote wireless transmission and reception functions, baseband processing functions). Alternatively, among all the functions of the communication protocol, the remote wireless transmission and reception function may be performed by a transmission reception point (TRP) (e.g., f(flexible)-TRP), and the baseband processing function among all the functions of the communication protocol Can be performed by a BBU (baseband unit) block. The TRP may be a remote radio head (RRH), radio unit (RU), transmission point (TP), etc. A BBU block may include at least one BBU or at least one digital unit (DU). A BBU block may be referred to as a “BBU pool,” “centralized BBU,” etc. The TRP can be connected to the BBU block via a wired fronthaul link or a wireless fronthaul link. A communication system consisting of a backhaul link and a fronthaul link may be as follows. When the function split method of the communication protocol is applied, the TRP can selectively perform some functions of the BBU or some functions of medium access control (MAC)/radio link control (RLC).

As previously observed, Channel State Information (CSI) Feedback Enhancement, Beam Management, and Positioning Accuracy Enhancement were selected as representative use cases in mobile communication systems.

Among these use cases, CSI feedback allows the terminal to provide channel state information to support the base station in applying transmission techniques such as Multiple Input Multiple Output (MIMO) or precoding in mobile communication systems. It means reporting. The 5G NR standard defined by 3GPP supports feedback information such as Channel Quality Indicator (CQI), Precoding Matrix Indicator (PMI), and Rank Indicator (RI) in relation to the CSI feedback method. do. In NR systems, discussions are ongoing to improve the CSI feedback technique to effectively support transmission techniques such as multi-user MIMO (MU-MIMO).

Specifically, the 3GPP NR system supports two types of code books to transmit PMI information, and are named TYPE 1 code book and TYPE 2 code book, respectively. The TYPE 1 codebook represents a beam group with an oversampled discrete Fourier transform (DFT) matrix, and has a structure in which one beam is selected and transmitted. On the other hand, the TYPE 2 code book has a structure that selects a plurality of beams and transmits information in the form of a linear combination of the selected beams. The TYPE 2 code book is evaluated as a more suitable structure to support transmission techniques such as MU-MIMO than the TYPE 1 code book, but it has the disadvantage that the CSI feedback load increases significantly due to the complex code book structure. In relation to the above problem, research is currently being conducted on how to obtain a compressed latent representation for a MIMO channel using auto-encoder, one of the deep learning techniques.

Beam management refers to the process in which a base station and a terminal allocate transmission beam and/or reception beam resources in a mobile communication system, respectively, when transmitting and receiving analog beams using spatial filters. Transmit beam and/or receive beam resource allocation may be applied. In the 5G NR standard defined by 3GPP, reference signals such as Synchronization Signal Block (SSB) and/or CSI Reference Signal (CSI-RS) are transmitted in multiple analog beam directions so that the base station and/or terminal can use optimal beam resources. Supports exploration.

However, the method in which the terminal searches for multiple analog beam directions and reports the optimal beam direction to the transmitter each time has the limitation of causing time delay and signal transmission load. In relation to this problem, reinforcement learning, one of the AI/ML techniques, is used to predict the next beam information using the previous beam information, or supervised learning is used to predict high-resolution beam information using low-resolution beam information. Research is underway to make inferences possible.

Positioning refers to a technique for measuring the location of a specific terminal in a mobile communication system. In the 5G NR standard defined by 3GPP, a Positioning Reference Signal (PRS) is transmitted to allow the terminal to report the Reference Signal Time Difference (RSTD) and then the Observed Time Difference Of Arrival , OTDOA), etc. are supported. Recently, requirements for positioning accuracy have been increasing, and from the perspective described above, research to improve the accuracy of measurement values for positioning by applying AI/ML techniques is being discussed.

However, to date, AI/ML application technology has been mainly discussed from the perspective of the link between the base station and a single terminal, and when multiple terminals within a cell managed by a base station can coexist in a mobile communication system, artificial intelligence for wireless communication There was no discussion on how to construct and train the neural network model.

Therefore, the present disclosure described below proposes a method of configuring and learning an artificial neural network model for wireless communication in a mobile communication system consisting of a base station and one or more terminals.

Specifically, in the present disclosure described below, in a mobile communication system consisting of a base station and one or more terminals, types of artificial neural networks are defined according to a method of sharing (or creating) an artificial neural network and/or a method of learning an artificial neural network. In addition, the present disclosure described below provides a method for including the artificial neural network type information when transmitting setting information for an artificial neural network or transmitting capability information between a base station and a terminal. In addition, the present disclosure described below provides a method for enabling a base station and a terminal to recognize the sharing (or creation) method and learning method of an artificial neural network through the type information.

In the following description, for convenience of explanation, the artificial neural network configuration and learning method proposed in this disclosure will be mainly explained from the downlink perspective of a mobile communication system composed of a base station and a terminal. However, the method proposed in this disclosure is mainly explained in terms of the downlink of a mobile communication system composed of a base station and a terminal. It can be expanded and applied to wireless communication systems.

[First Example]

According to the first embodiment of the present disclosure, when applying an artificial neural network for wireless communication in a mobile communication system consisting of a base station and one or more terminals, one of the following regarding the artificial neural network depends on the way the artificial neural network is shared (or created): You can define one or more shared (or created) types.

(One) Type 0: A type in which artificial neural networks are shared (or created) on a cell basis.

(2) Type 1: A type in which artificial neural networks are shared (or created) on a per-device group basis.

(3) Type 2: A type in which artificial neural networks are shared (or created) on a per-device basis.

(4) Type 3: A type in which artificial neural networks are shared (or created) on a cell group basis.

In the first embodiment of the present disclosure, when a base station configures an artificial neural network for a terminal, a method of setting the artificial neural network including sharing (or creation) type information is proposed.

In configuring shared (or generated) type information for an artificial neural network according to the first embodiment of the present disclosure, a terminal group may be defined as terminals having the same classification indicator.

Additionally, when configuring shared (or generated) type information for an artificial neural network according to the first embodiment of the present disclosure, a cell group may be defined as cells having the same classification indicator.

In addition, in configuring sharing (or creation) type information for an artificial neural network according to the first embodiment of the present disclosure, sharing (or creation) of an artificial neural network means that the structure and/or model parameters of the artificial neural network are the same for the application unit. This may mean that (or a weight vector) is applied.

In this disclosure, for convenience of explanation, the artificial neural network according to the first embodiment of the present disclosure will be described assuming a mobile communication system, for example, a 5G NR system according to the 3GPP standard. However, it should be noted that the assumptions of the 5G NR system are only for convenience of understanding and can be applied to any system that includes a transmitting node and two or more receiving nodes in a wireless communication system.

In addition, the following will describe a case where the artificial neural network according to the first embodiment of the present disclosure is applied to encode/decode the channel state information (CSI) between the station and the terminal mentioned in the 3GPP standard above. will be. CSI encoding/decoding described in the first embodiment of the present disclosure is only an example, and the present disclosure is not limited thereto.

Then, in the following, we will assume that an artificial neural network (hereinafter referred to as “artificial neural network for CSI feedback”) is applied to encode/decode CSI information between a base station and a terminal in a 5G NR system according to 3GPP standards.

When multiple terminals in a cell want to use the same code for the same channel state information, such as the codebook-based CSI feedback method defined in the current standard, it can be defined in various forms.

First, it may be desirable for the artificial neural network for CSI feedback to be defined on a cell basis. In other words, all terminals in a cell can perform CSI encoding/decoding by applying the same artificial neural network structure and weight vector.

Second, depending on the channel environment, the base station may determine that it is more effective to configure the artificial neural network for CSI feedback differently for each specific beam direction. In this case, it may be desirable to define the artificial neural network for CSI feedback on a terminal group basis or a terminal basis. Let's look at this with a specific example.

In the NR system according to an embodiment of the present disclosure, terminals having the same Transmission Configuration Indicator (TCI) on the Physical Downlink Shared Channel (PDSCH) configuration information can be regarded as one terminal group. there is. In this way, terminals with the same TCI can be set as one terminal group, and the terminal group can be set to perform CSI encoding/decoding by applying the same artificial neural network structure and weight vector. In the example of the artificial neural network sharing (or creation) type for CSI feedback, if the artificial neural network is shared (or created) for each terminal group or terminal, the base station must manage multiple different artificial neural networks, so what type of sharing (or creation) is required? Can be set depending on the capabilities of the base station.

Let's take a closer look at the forms described above with reference to the attached drawings.

FIG. 3A is a conceptual diagram illustrating sharing (or creating) an artificial neural network according to Type 0 of the first embodiment of the present disclosure.

Referring to FIG. 3A, the base station 310 may correspond to a transmitting node and may have a communication area 310a based on the signal transmission distance. The base station 310 may communicate based on a communication scheme based on the 3GPP NR standard and/or may communicate based on a 6G communication standard to be supported in the future and/or may communicate based on another wireless communication scheme. In FIG. 3A, it can be assumed that the communication area 310a of the base station 310 corresponds to one cell. The following description will be made assuming that the communication area 310a of the base station 310 is one cell.

Terminals 321, 322, 323, 324, and 325 located within the communication area of the base station 310 may correspond to the receiving node. The plurality of terminals 321 to 325 may all be terminals capable of applying an artificial neural network according to the present disclosure. In addition, Figure 3a is a conceptual diagram to explain Type 0, that is, the case of sharing an artificial neural network on a cell basis, and the terminals 321 to 325 illustrated in Figure 3a are all located within the communication area 310a of the base station 310. Therefore, the same artificial neural network 331 can be shared. Accordingly, in Figure 3a, the same artificial neural network 331 is shown at the top of each terminal 321-325. The example in FIG. 3A is only for ease of understanding, and in reality, each artificial neural network 331 is stored in a memory (not shown in FIG. 3A) inside each terminal 321-325 and driven by a processor, etc. You can.

In addition, in the example of FIG. 3A, an artificial neural network is not illustrated for the base station 310, but the same artificial neural network 331 as that of the terminals 321 to 325 can be used. In other words, the base station 310 may be storing the same artificial neural network 331 as the terminals 321-325 within the communication area 310a in its internal memory (not shown in FIG. 3A). The artificial neural network 331 stored in the base station 310 may be driven by a processor or the like and may be used when communicating with terminals 321-325. Here, communication may mean transmission and/or reception of wireless (Radio Frequency, RF) signals.

The artificial neural network 331 illustrated in FIG. 3A may be the artificial neural network for CSI feedback described above. As another example, the artificial neural network 331 illustrated in FIG. 3A may be an artificial neural network related to beam management discussed in 3GPP. As another example, the artificial neural network 331 illustrated in FIG. 3A may be an artificial neural network for improving positioning accuracy that is being discussed in 3GPP. In addition, the artificial neural network 331 illustrated in FIG. 3A may be an artificial neural network based on various methods in which AI/ML is applied other than the method discussed in 3GPP.

Although the present disclosure described above describes the base station 310 and the terminals 321-325 in a communication system based on the 3GPP NR standard, it can also be understood as a transmitting node and a receiving node in other wireless communication systems. In other wireless communication systems, when there is only one receiving node, it may include only one specific terminal in the example of FIG. 3A. In addition, although only five terminals are illustrated in FIG. 3A, those skilled in the art will know that more terminals can be included in the base station 310.

FIG. 3B is a conceptual diagram illustrating sharing (or creating) an artificial neural network according to Type 1 of the first embodiment of the present disclosure.

Referring to FIG. 3B, the base station 310 may correspond to a transmitting node and may have a communication area 310a based on the signal transmission distance. The base station 310 may communicate based on a communication scheme based on the 3GPP NR standard and/or may communicate based on a 6G communication standard to be supported in the future and/or may communicate based on another wireless communication scheme. In FIG. 3B, it can be assumed that the communication area 310a of the base station 310 corresponds to one cell. The following description will be made assuming that the communication area 310a of the base station 310 is one cell.

Terminals 341a, 341b, 342a, 342b, and 342c located within the communication area of the base station 310 may correspond to the receiving node. The plurality of terminals 341a, 341b, 342a, 342b, and 342c may all be terminals capable of applying an artificial neural network according to the present disclosure.

Unlike FIG. 3A, FIG. 3B illustrates a case where terminals are composed of two different groups. The terminals 341a and 341b may be terminals of the first group, and the terminals 342a, 342b, and 342c may be terminals of the second group. The terminals 341a and 341b of the first group may have the same classification indicator. As an example of a classification indicator, there may be the same TCI in the PDSCH configuration information as described above. If the same classification indicator is the TCI, the terminals 342a-342c of the second group may also have the same TCI in the PDSCH configuration information. At this time, if the TCI indicating the first group of terminals (341a, 341b) is called TCI #1, and the TCI indicating the second group of terminals (342a-342c) is TCI #2, TCI #1 and TCI #2 can be set to different values.

In addition, the first group of terminals 341a and 341b share the same first artificial neural network 332, and the second group of terminals 342a, 342b and 342c share a different second artificial neural network 333. ) can be shared. At this time, the first artificial neural network 332 and the second artificial neural network 333 are used for the same purpose, but may be artificial neural networks of different types. Here, the same use may mean, for example, that both the first artificial neural network 332 and the second artificial neural network 333 are used as artificial neural networks for CSI feedback. In addition, different types of artificial neural networks may mean that the first artificial neural network 332 and the second artificial neural network 333 have different processing methods, so even if they have the same input value, they can have different outputs.

In the example of FIG. 3B, an artificial neural network is not included in the base station 310. In a specific case, the base station 310 uses a first artificial neural network 332 used by the first group of terminals 341a and 341b and a second artificial neural network used by the second group of terminals 342a, 342b and 342c ( 333) can all be used. In other words, the base station 310 may store the first artificial neural network 332 and the second artificial neural network 333 in an internal memory (not shown in FIG. 3A). The first artificial neural network 332 and the second artificial neural network 333 stored in the base station 310 may be driven by a processor, etc., and the first group of terminals 341a and 341b and/or the second group of terminals It can be used when communicating with (342a, 342b, 342c). When the base station 310 communicates with terminals using an artificial neural network, the base station 310 uses the first artificial neural network 332 when communicating with the terminals 341a and 341b of the first group and the terminals 341a and 341b of the second group. The second artificial neural network 333 can be used when communicating with the terminals 342a, 342b, and 342c. Here, communication may mean transmission and/or reception of radio (RF) signals.

In FIG. 3B, the first artificial neural network 332 is illustrated at the top of each first group of terminals 341a-341b in the same manner as previously described in FIG. 3a, and the first artificial neural network 332 is illustrated at the top of each second group of terminals 342a-342c. The second artificial neural network 333 is illustrated. The example in FIG. 3B is also to aid understanding. In fact, each artificial neural network 332, 333 is stored in memory (not shown in FIG. 3A) inside each terminal 321-325 and can be driven by a processor, etc. .

In the present disclosure, both the first artificial neural network 332 and the second artificial neural network 333 may be the artificial neural network for CSI feedback described above. As another example, the first artificial neural network 332 and the second artificial neural network 333 illustrated in FIG. 3B may both be artificial neural networks related to beam management discussed in 3GPP. As another example, the first artificial neural network 332 and the second artificial neural network 333 illustrated in FIG. 3B may both be artificial neural networks for improving positioning accuracy that are being discussed in 3GPP. In addition, the artificial neural network illustrated in Figure 3b can be an artificial neural network based on various methods in which AI/ML is applied other than the method discussed in 3GPP.

Although the present disclosure described above describes the base station 310 and the terminals 341a, 341b, 342a, 342b, and 342c in a communication system based on the 3GPP NR standard, they can be understood as transmitting nodes and receiving nodes in other wireless communication systems. It can be. In addition, although only five terminals are illustrated in FIG. 3b, those skilled in the art will know that more terminals can be included in the base station 310. Also, although only two groups are mentioned in FIG. 3B, the present disclosure can be applied even if three or more groups are included.

As described above, the artificial neural networks 332 and 333 illustrated in FIG. 3B can all be the artificial neural networks for CSI feedback described above. However, each of the artificial neural networks 332 and 333 may be artificial neural networks that generate CSI feedback in different ways. As another example, the artificial neural networks 332 and 333 illustrated in FIG. 3B may all be artificial neural networks related to beam management discussed in 3GPP. At this time, each of the artificial neural networks 332 and 333 may be artificial neural networks that generate information related to beam management in different ways. As another example, the artificial neural networks 332 and 333 illustrated in FIG. 3B may all be artificial neural networks for improving positioning accuracy that are being discussed in 3GPP. Even in this case, each of the artificial neural networks 332 and 333 may be artificial neural networks for generating positioning information in different ways. In addition, the artificial neural network illustrated in Figure 3b can be an artificial neural network based on various methods in which AI/ML is applied other than the method discussed in 3GPP.

FIG. 3C is a conceptual diagram illustrating sharing (or creating) an artificial neural network according to Type 2 of the first embodiment of the present disclosure.

Referring to FIG. 3C, the base station 310 may correspond to a transmitting node and may have a communication area 310a based on the signal transmission distance. The base station 310 may communicate based on a communication scheme based on the 3GPP NR standard and/or may communicate based on a 6G communication standard to be supported in the future and/or may communicate based on another wireless communication scheme. In FIG. 3C, it can be assumed that the communication area 310a of the base station 310 corresponds to one cell. The following description will be made assuming that the communication area 310a of the base station 310 is one cell.

Terminals 321, 322, 323, 324, and 325 located within the communication area of the base station 310 may correspond to the receiving node. The plurality of terminals 321 to 325 may all be terminals capable of applying an artificial neural network according to the present disclosure. In addition, the terminals 321 to 325 illustrated in FIG. 3C are all located within the communication area of the base station 310, and each terminal 321 to 325 uses different artificial neural networks 331, 332, 333, 334, 335). At this time, the different artificial neural networks 331, 332, 333, 334, and 335 are used for the same purpose, but may be of different types. In addition, although the artificial neural network is not illustrated in the base station 310 in the example of FIG. 3C, the base station 310 can use the same artificial neural networks as the artificial neural networks 331-335 provided by each of the terminals 321-325. there is. In other words, the base station 310 may have a plurality of artificial neural networks 331-335 that are the same as those of the terminals 321-325 within the communication area 310a.

Also, as described above, FIG. 3C also illustrates different artificial neural networks (331-335) on top of each terminal (321-325). Figure 3c is a conceptual diagram corresponding to Type 2 in which artificial neural networks are shared (or created) on a per-terminal basis, so different artificial neural networks 331-335 are stored in the memory inside the corresponding terminals 321-325 (Figure 3a). (not shown) may be stored. Artificial nerves 331-335 may be driven by a processor provided in the corresponding terminal.

In the example of FIG. 3C, an artificial neural network is not illustrated for the base station 310, but the base station 310 can use all artificial neural networks 331-335 used in each of the terminals 321-325. In other words, the base station 310 can store all artificial neural networks 331-335 used in each of the terminals 321-325 in its internal memory. And when the base station 310 communicates with a specific terminal, it can communicate using an artificial neural network corresponding to the specific terminal. Here, communication may mean transmission and/or reception of radio (RF) signals. Also in FIG. 3C, the artificial neural networks 331-335 may be driven by a processor or the like inside the base station 310.

The artificial neural networks 331-335 illustrated in FIG. 3C may all be the artificial neural networks for CSI feedback described above. As another example, the artificial neural networks 331-335 illustrated in FIG. 3C may all be artificial neural networks related to beam management discussed in 3GPP. As another example, the artificial neural networks 331-335 illustrated in FIG. 3C may all be artificial neural networks for improving positioning accuracy that are being discussed in 3GPP. In addition, the artificial neural networks 331-335 illustrated in FIG. 3C can all be artificial neural networks based on various methods in which AI/ML is applied in addition to the method discussed in 3GPP.

Although the present disclosure described above describes the base station 310 and the terminals 321-325 in a communication system based on the 3GPP NR standard, it can also be understood as a transmitting node and a receiving node in other wireless communication systems. In other wireless communication systems, when there is only one receiving node, it may include only one specific terminal in the example of FIG. 3C. In addition, although five terminals are illustrated in FIG. 3C, those skilled in the art will know that more terminals can be included in the base station 310.

FIG. 3D is a conceptual diagram illustrating sharing (or creating) an artificial neural network according to Type 3 of the first embodiment of the present disclosure.

Referring to FIG. 3D, a plurality of base stations 311, 312, 313, 314, and 315 are illustrated, of which base stations 311, 312, 313, and 314 are set to the first group 301, A case in which one base station 315 is set to the second group 302 is exemplified. In Figure 3d, only two groups are illustrated due to drawing constraints. However, more groups can be formed according to the method described below. It should be noted that the composition of groups is also not limited to the number of base stations and/or terminals illustrated in FIG. 3D. Additionally, in FIG. 3D, it can be assumed that each of the communication areas 311a, 312a, 313a, 314a, and 315a of the base stations 311, 312, 313, 314, and 315 corresponds to one cell. In the following description, it will be assumed that each of the communication areas 311a, 312a, 313a, 314a, and 315a of the base stations 311, 312, 313, 314, and 315 is one cell.

Each of the base stations 311, 312, 313, 314, and 315 illustrated in FIG. 3D may correspond to a transmitting node, and the base stations 311, 312, 313, 314, and 315 transmit signals based on their respective signal transmission distances. It may have communication areas 311a, 312a, 313a, 314a, and 315a. Each of the base stations 311, 312, 313, 314, and 315 all communicates based on a communication method based on the 3GPP NR standard and/or communicates based on a 6G communication standard to be supported in the future and/or other wireless communication methods. You can communicate based on

The terminals 321a, 321b, 321c, 321c, 321d, and 321e located in the first group 301 and the terminal 322a located in the second group 302 may all correspond to the receiving node. Additionally, the terminals 321a-321e and 322a illustrated in FIG. 3D may all be terminals capable of applying an artificial neural network according to the present disclosure. Additionally, the terminals 321a-321e and 322a illustrated in FIG. 3D may communicate within the communication area of a specific base station. Since the terminals 321a-321e and 322a illustrated in FIG. 3D communicate within a specific base station, they may have artificial neural networks 331 and 332 corresponding to the group to which each base station belongs. For example, terminals 321a-321e located within the first group 301 may use the first artificial neural network 331, and terminals 322a located within the second group may use the second artificial neural network 332. You can use it. Like the terminals, the base stations 311, 312, 313, and 314 belonging to the first group 301 can all use the first artificial neural network 331, and the base station 315 belonging to the second group 302 The second artificial neural network 332 may be used.

In the first embodiment of the present disclosure, the first artificial neural network 332 and the second artificial neural network 333 are used for the same purpose, but may be artificial neural networks of different types. Here, the same use may mean, for example, that both the first artificial neural network 332 and the second artificial neural network 333 are used as artificial neural networks for CSI feedback. In addition, different types of artificial neural networks may mean that the first artificial neural network 332 and the second artificial neural network 333 have different processing methods, so even if they have the same input value, they can have different outputs.

Therefore, in the present disclosure, all base stations 311, 312, 313, and 314 operating as transmitting nodes in the first group 301 and all terminals 321a-321e operating as receiving nodes use the same first artificial neural network 331. ) can be used.

Also, as described above, artificial neural networks 331-332 are illustrated at the top of each group 301 and 302 in FIG. 3D. The example in FIG. 3D is to aid understanding as described above, and in fact, each artificial neural network (331-332) has memory inside the base stations (311-315) and terminals (321a-321e, 322a) (in FIG. 3D). (not shown) and may be driven by a processor, etc.

The artificial neural networks 331-332 illustrated in FIG. 3D may all be the artificial neural networks for CSI feedback described above. As another example, the artificial neural networks 331-332 illustrated in FIG. 3D may all be artificial neural networks related to beam management discussed in 3GPP. As another example, the artificial neural networks 331-332 illustrated in FIG. 3D may all be artificial neural networks for improving positioning accuracy that are being discussed in 3GPP. In addition, the artificial neural networks 331-335 illustrated in Figure 3d can all be artificial neural networks based on various methods in which AI/ML is applied other than the method discussed in 3GPP.

Although the present disclosure described above describes base stations 311-315 and terminals 321a-321e and 322a in a communication system based on the 3GPP NR standard, they can be understood as transmitting nodes and receiving nodes in other wireless communication systems. You can.

Let's take a closer look at the settings of the artificial neural network for the first embodiment of the present disclosure with reference to FIGS. 3A to 3D described above.

When a base station or base station group sets up an artificial neural network for a terminal, it can set one of the four types as seen above depending on the type it wants to operate. In other words, when setting up an artificial neural network for a terminal, the base station divides the It can be set to include neural network sharing (or creation) type information. If the base station wishes to share (or create) an artificial neural network to each of the terminals, to some of the terminals, or to all terminals according to the first embodiment of the present disclosure, the base station may need to manage the artificial neural networks used by the terminal. . Therefore, it is necessary to consider the capabilities of the base station when sharing (or creating) the artificial neural network according to the present disclosure with the terminals.

Meanwhile, the terminal can manage artificial neural network information based on the received artificial neural network sharing (or creation) type information. For example, in the case of a cell-level artificial neural network, each terminal can manage artificial neural network information in a storage space divided by cell ID. As another example, in the case of a terminal group unit artificial neural network (designated as a terminal group for the same TCI), each terminal can manage artificial neural network information in a storage space divided by an indicator representing the group.

Additionally, the type in which the artificial neural network is shared (or created) on a cell group basis can also be defined. For example, there may be an artificial neural network model shared by several adjacent cells, reflecting regional characteristics. The advantage of supporting an artificial neural network model shared across multiple cells by reflecting regional characteristics is that there is no need to initialize or change the artificial neural network model when the terminal performs handover between cells. As an example of this case, when configuring an artificial neural network for CSI feedback and/or an artificial neural network that finds the location of a terminal within a specific area for positioning, the artificial neural network may be an artificial neural network that can be commonly used for a plurality of cells.

Meanwhile, the type of artificial neural network described above can be set to terminals using various formats. Various types of methods can be used for this, and we will look at several methods as examples.

First, the base station can indicate Type 0 to Type 3 using a specific message broadcast to all terminals communicating within its cell. Accordingly, each of the terminals can share (or create) an artificial neural network from the base station and/or through a separate server that provides the artificial neural network based on the Type indicated in a specific message broadcast by the base station.

Second, the base station can inform terminals of the types supported by the base station in different ways based on each type provided by the base station.

For example, in the case of Type 0, which shares (or creates) an artificial neural network on a cell basis, or in the case of Type 3, which shares (or creates) an artificial neural network on a cell group basis, the base station broadcasts a specific message. You can use it to instruct the sharing (or creation) of an artificial neural network.

As another example, in the case of Type 1, which shares (or creates) an artificial neural network on a terminal basis for each terminal, and/or in the case of Type 2, which shares (or creates) an artificial neural network on a terminal group basis, a predetermined configuration that can be set for each terminal A configuration message can be used to instruct the sharing (or creation) of an artificial neural network.

As another example, in the case of Type 2, which shares (or creates) an artificial neural network by terminal group, sharing (or creation) of the artificial neural network can be instructed using a predetermined configuration message set for each terminal group. .

The operation of the base station described above is an example and is intended to aid understanding, and the types according to the first embodiment described in this disclosure are not limited to the above description. Additionally, the features of the first embodiment described above may be applied together with other embodiments described below as long as they do not conflict with each other.

[Second Embodiment]

According to the second embodiment of the present disclosure, when applying an artificial neural network for wireless communication in a mobile communication system consisting of a base station and one or more terminals in the same manner as in the first embodiment described above, the artificial neural network is shared (or generated). Depending on the method, one or more sharing (or creation) types among the four types described in the first embodiment can be defined for the artificial neural network.

In the second embodiment according to the present disclosure, when the terminal reports capability information about the artificial neural network to the base station, a method of reporting including information on the sharing (or creation) type supported by the terminal for the corresponding artificial neural network is proposed. In the first embodiment described above, the terminal did not consider reporting information about the artificial neural network to the base station. In other words, the first embodiment assumes that an artificial neural network is set up at the base station without considering the terminal's capability information, for example, the terminal's capabilities. However, it will be apparent to those skilled in the art that the second embodiment and the first embodiment of the present disclosure can be considered and used together.

According to the second embodiment according to the present disclosure, the terminal can report capability information about the artificial neural network to the base station. When the terminal reports capability information to the base station, the report may include sharing (or creation) type information about the corresponding artificial neural network supported by the terminal. For example, the terminal may report to the base station including type information of Type 0 to Type 3 described in the first embodiment. At this time, in the case of Type 1, the terminal group can be defined as terminals with the same classification indicator. Additionally, sharing (or creating) an artificial neural network may mean that the same artificial neural network structure and/or model parameters (or weight vectors) are applied to the application unit.

As an example of the present disclosure, it is assumed that an artificial neural network for CSI feedback is applied in a mobile communication system consisting of a base station and a terminal, such as a 5G NR system according to the 3GPP standard. Like the existing codebook-based CSI feedback method, when multiple terminals in a cell want to use the same code for the same channel state information, it is desirable for the artificial neural network for CSI feedback to be defined on a cell basis. You can. In other words, all terminals in a cell can perform CSI encoding/decoding by applying the same artificial neural network structure and/or weight vector.

In order for all terminals in a cell to apply the same artificial neural network structure and weight vector, the terminals in the cell must be able to cooperate to learn a single artificial neural network. If the terminals in a cell cooperate to support the operation of learning a single artificial neural network, each of the terminals must support federated learning. However, if each terminal is to support federated learning, there may be terminals that cannot support federated learning, considering the complexity of the terminal and the resulting costs.

Therefore, in the second embodiment of the present disclosure, when the terminal reports capability information about an artificial neural network to the base station, the terminal may report including sharing (or creation) type information about a specific artificial neural network supported by the terminal. .

Additionally, the four types according to the second embodiment of the present disclosure may be the same as those shown in FIGS. 3A to 3D previously illustrated while explaining the first embodiment. Therefore, the four types according to the second embodiment of the present disclosure can apply the previously described FIGS. 3A to 3D.

The features of the second embodiment described above can be applied together with the first embodiment described above as well as the methods of other embodiments described below as long as they do not conflict with the method.

[Third Embodiment]

According to the third embodiment of the present disclosure, when applying an artificial neural network for wireless communication in a mobile communication system consisting of a base station and one or more terminals, one or more of the following learning types for the artificial neural network depending on the method of learning the artificial neural network can be defined.

(One) Type A: A type that does not support online learning (or a type that only supports offline learning)

(2) Type B: A type that supports online learning only on terminals

(3) Type C: A type that supports online learning only at the base station

(4) Type D: A type that supports online learning in both base stations and terminals

When a base station sets up an artificial neural network for a terminal, we propose a method of setting it including learning type information about the artificial neural network.

In setting the learning type for the artificial neural network according to the third embodiment of the present disclosure, in the case of Type A, it may mean that an artificial neural network pre-trained offline or by another network node can be utilized in the inference process. .

In the present disclosure, in the same manner as the first and second embodiments described above, when an artificial neural network for CSI feedback is applied to encode/decode CSI information in a 5G NR mobile communication system including a base station and a terminal according to the 3GPP standard. Assume.

In the first embodiment of the present disclosure, a method was proposed where the base station transmits information including sharing (or creation) type information for the artificial neural network when setting up the artificial neural network. However, the learning characteristics of an artificial neural network may not be fully described only through shared (or generated) type information.

For example, assume that the artificial neural network for CSI feedback is configured on a cell basis, and multiple terminals within the cell apply the same artificial neural network structure and weight vector. If the artificial neural network for CSI feedback is configured on a cell basis, learning of the artificial neural network for CSI feedback may be performed by the base station, the terminal, or both the base station and the terminal.

Let's look at this using an example. First, it is assumed that channel reciprocity between downlink and uplink channels is established in the 3GPP NR TDD system. If this assumption is established, the base station can learn an artificial neural network for CSI feedback using uplink channel information. Afterwards, the base station can deliver model information learned about the artificial neural network for CSI feedback to the terminal through a broadcast channel, etc.

When the above assumption holds, the terminal may perform learning. When a terminal performs learning, each terminal can calculate the amount of change in the weight vector of the artificial neural network for CSI feedback using the joint learning method. Then, the terminal can report calculation information about the amount of change in the vector to the base station. Then, the base station can determine the final weight vector update direction received from each terminal.

Then, in the following, we will look at how the third embodiment of the present disclosure is applied using an artificial neural network for CSI feedback, which is one of the various types of artificial neural networks discussed in the 3GPP standard. The description of a specific method described below, for example, using an artificial neural network for CSI feedback, is only intended to aid understanding and should not be construed as limiting the present disclosure.

First, in the third embodiment, Type A is a case where online learning is not supported for artificial neural networks. Therefore, in the case of Type A, online learning of the artificial neural network is not performed for both the base station and the terminal. However, Type A can support offline learning. In cases where online learning is not supported, the artificial neural network can be applied to the base station and terminal as initially set. In other words, if online learning is not supported, the artificial neural network is stored in the base station and/or terminal, and online learning does not proceed thereafter.

If online learning is not supported and offline learning is performed for the artificial neural network, the base station and/or terminal may store the offline learned artificial neural network. Afterwards, the base station and/or terminal does not proceed with learning of the offline learned artificial neural network.

FIG. 4A is a conceptual diagram for explaining online learning of an artificial neural network according to Type B of the third embodiment of the present disclosure.

Referring to FIG. 4A, the base station 410 may correspond to a transmitting node and may have a communication area 410a based on the signal transmission distance. The base station 410 may communicate based on a communication scheme based on the 3GPP NR standard and/or may communicate based on a 6G communication standard to be supported in the future and/or may communicate based on another wireless communication scheme. Additionally, in FIG. 4A, it can be assumed that the communication area 410a of the base station 410 corresponds to one cell. In the following description, it will be assumed that the communication area 410a of the base station 410 is one cell.

In Figure 4a, for convenience of explanation, only one terminal 421 is illustrated within the communication area 410a of the base station 410. The terminal 421 may be a terminal capable of applying the artificial neural network 430 according to the present disclosure. Additionally, the terminal 421 may be a terminal capable of supporting online learning of the artificial neural network 430. For example, let's assume that the artificial neural network 430 is an artificial neural network for CSI feedback.

The base station 410 may transmit a reference signal to measure CSI. Then, the terminal 421 can measure the channel state based on the reference signal transmitted from the base station 410. When measuring the channel state, the terminal 421 can measure the channel state using the artificial neural network 430 for CSI feedback. For example, the input 431 of the artificial neural network 430 for CSI feedback may be a reference signal received from the terminal 421, and the output 432 of the artificial neural network 430 for CSI feedback may be CSI feedback information. It can be.

The artificial neural network 430 in FIG. 4A may be an artificial neural network capable of online learning according to the present disclosure. The artificial neural network 430 may include an input node that receives the input 431 and an output node that outputs the final result. The output 432, which is the final result output from the output node, can be called a label or ground truth. It also includes hidden nodes between the input node and the output node, and each hidden node can have weights. The input of the artificial neural network 430 may be calculated based on the weights of each hidden node located between the input node and the output node. In this way, the calculation process from the input node to the output node is called forward propagation (441).

The forward propagation described above simply uses an artificial neural network 430. However, learning of the artificial neural network 430 is accomplished by verifying the weights by backward propagating the final result through hidden nodes (442). When verifying the weight, the terminal 421 can support joint learning by lowering the weight of a hidden node with a high error rate during forward propagation through backward propagation or by reporting information about the weight of the hidden node to the base station 410. In the following description, information about the weight change of the hidden node will be referred to as “weight vector change information.”

Additionally, this disclosure assumes that such learning is online learning. Online learning according to the present disclosure can be a process in which the artificial neural network model used for inference is trained in (near) real-time with the arrival of new training samples. These real-time processes can generally be trained continuously.

Online learning described in this disclosure may mean training the artificial neural network 430 in real-time or near real-time. In the present disclosure, the distinction between (near) real-time and non real-time varies depending on the situation and may be relative to the time-scale.

In Figure 4a, only one terminal 421 is illustrated for convenience of explanation. However, it is obvious to those skilled in the art that multiple terminals can communicate within the communication area 410a of the base station 410. Therefore, each terminal located within the communication area 410a of the base station 410 can report weight vector change information for the artificial neural network 430 that it is running to the base station 410 for joint learning.

As another example, when the terminal is alone or has an individual artificial neural network for each terminal, the terminal can perform online learning by immediately lowering the weight of a hidden node with a high error rate during forward propagation.

Meanwhile, the terminal 421 may separately update (or store) the weight vector change information of the CSI feedback artificial neural network 430. As another example, the terminal 421 may provide weight vector change information of the artificial neural network 430 to the base station 410 for joint learning. In the embodiment of the present disclosure, the description will be made assuming a method in which the base station 410 leads the update of the artificial neural network during joint learning. However, a separate server for online learning other than the base station 410 may be used. When a separate online learning server is installed for online learning of the artificial neural network of the terminal 421, the online learning server may be located in the upper network of the base station. In the following description, for convenience of explanation, it will be explained that the base station 410 hosts online learning of the artificial neural network 430 for CSI feedback of the terminal 421 during joint learning.

As described above, although only one terminal 421 is illustrated in FIG. 4a, if Type 0 or Type 2 or Type 3 of the first embodiment described above is applied and joint learning is required, the base station 410 has the same purpose and the same Weight vector change information can be received from a plurality of terminals for an artificial neural network according to the method. For example, the weight vector change information received by the base station 410 may collect information about a CSI feedback artificial neural network having the same vector weights among artificial neural networks for the same purpose, such as a CSI feedback artificial neural network. For example, as described in Type 1 of the first embodiment, when a first artificial neural network 332 for CSI feedback and a second artificial neural network 333 for CSI feedback exist, information about the first artificial neural network 332 for CSI feedback Only can be collected. In the case of Type 2 described in the first embodiment, it may be the case that weight vector change information for the artificial neural network for CSI feedback for each terminal is received.

Therefore, when joint learning is performed, the base station 410 can receive weight vector change information from a plurality of terminals or from one specific terminal for a specific artificial neural network. The base station 410, which has received weight vector change information from a plurality of terminals, may determine the update direction of the weight vector based on the received weight vector change information. Once the base station 410 determines the update direction of the weight vector, it can provide weight vector update information to the corresponding terminal(s). Here, the weight vector update information may be weight vector information for updating the weight for the hidden node of the artificial neural network based on the update direction of the weight vector determined by the base station. Each terminal that has received the weight vector update information can update the weights of the artificial neural network based on the weight vector update information received from the base station. In other words, weight vectors in the artificial neural network 430 of the terminal 421 illustrated in FIG. 4A can be updated based on the weight update information.

Figure 4b is a conceptual diagram for explaining online learning of an artificial neural network according to Type C of the third embodiment of the present disclosure.

Referring to FIG. 4B, the base station 411 may correspond to a transmitting node of a wireless communication system and may have a communication area 411a based on the signal transmission distance. The base station 411 may communicate based on a communication scheme based on the 3GPP NR standard and/or may communicate based on a 6G communication standard to be supported in the future and/or may communicate based on another wireless communication scheme. In FIG. 4B, it can be assumed that the communication area 411a of the base station 411 corresponds to one cell. In the following description, it will be assumed that the communication area 411a of the base station 411 is one cell. And the base station 411 in FIG. 4B is a base station capable of performing online learning according to an embodiment of the present disclosure, and uses different reference numerals from those in FIG. 4A.

The terminal 422 may correspond to a receiving node of a wireless communication system and may be a terminal capable of applying the artificial neural network 430 according to the present disclosure. However, unlike in FIG. 4A, the terminal 422 is a terminal that cannot perform online learning for the artificial neural network 430. In other words, Figure 4b corresponds to Type C in which only the base station 411 supports online learning, so the terminal 422 illustrated in Figure 4b cannot perform online learning for the artificial neural network. Therefore, the terminal 422 illustrated in FIG. 4B uses a different reference number from the terminal 421 previously described in FIG. 4A.

The embodiment of FIG. 4B will also be examined assuming that the artificial neural network 430 is an artificial neural network for CSI feedback, as previously described in FIG. 4A. Additionally, it is assumed that reciprocity is established between the downlink channels from the base station 411 to the terminal 422 and the uplink channels from the terminal 422 to the base station 411. If this assumption holds, the base station 411 can perform training of the artificial neural network 430 for CSI feedback by utilizing the uplink channel information received from the terminal 422. For example, the CSI feedback artificial neural network 430 can obtain CSI feedback information as an output 432 by using the reference signal received from the terminal 422 as an input 431 and forward propagating 443. In addition, the artificial neural network for CSI feedback 430 can detect and store the amount of change in the weight vector for the artificial neural network for CSI feedback 430 by backward propagating 444 the CSI feedback information, which is the output 432. In FIG. 4B, since forward propagation 443 and reverse propagation 444 are performed at the base station 411, different reference numerals are used than those in FIG. 4A.

The base station 411 may obtain and store the amount of change in the weight vector for the artificial neural network 430 for CSI feedback obtained through forward propagation 443 and backward propagation 444, that is, weight vector change information. And the base station 411 can update the weights of the artificial neural network 430 for CSI feedback based on the weight vector information. In other words, the base station 411 can perform online learning of the artificial neural network 430 for CSI feedback based on weight vector information. Since the online learning described in this disclosure is the same as that described in FIG. 4A, the same description will be omitted.

Meanwhile, based on the computing capability of the base station 411, etc., if necessary, an online learning server (not shown in FIG. 4B) to update the weights of the CSI feedback artificial neural network 430 can perform calculation of the weight vector. If a separate online learning server is installed, the online learning server can be connected to the base station's upper network. However, in the present disclosure described below, for convenience of explanation, it is assumed that the weight vector calculation of the CSI feedback artificial neural network 430 is performed at the base station 411, that is, online learning is performed.

The base station 411 generates weight vector update information when the weight vector of the artificial neural network 430 for CSI feedback is needed based on the weight vector operation of the artificial neural network 430 for CSI feedback, and the artificial neural network 430 of the base station 411 The neural network 430 can be updated.

Then, if necessary, the base station 411 can transmit the generated weight vector update information to the terminal 422. Here, the weight vector update information may be information for changing the weight vector applied to at least one node among the nodes within the CSI feedback artificial neural network 430. As another example, the weight vector update information may be the updated artificial neural network for CSI feedback 430 itself.

For example, when the base station 411 applies Type 0 to Type 2 described in the first embodiment, update information about the artificial neural network can be provided to terminals within the base station 411. Then, the terminal that needs to update the corresponding artificial neural network within the base station 411 can receive the artificial neural network update information and update the corresponding artificial neural network.

Meanwhile, in the case of Type 3, the base station 411 must change the artificial neural network for the cell group, so it may update the artificial neural network through collaboration with other base stations (not shown in Figure 4b). If the artificial neural network is updated through collaboration with other base stations, the base station 411 can provide the artificial neural network update information to terminals within its cell. Therefore, the terminal that receives the artificial neural network update information from the base station 411 can update the corresponding artificial neural network 430.

FIG. 4C is a conceptual diagram for explaining online learning of an artificial neural network according to Type D of the third embodiment of the present disclosure.

Referring to FIG. 4C, the base station 412 may correspond to a transmitting node of a wireless communication system and may have a communication area 412a based on the signal transmission distance. The base station 412 may communicate based on a communication scheme based on the 3GPP NR standard and/or may communicate based on a 6G communication standard to be supported in the future and/or may communicate based on another wireless communication scheme. In FIG. 4C, it can be assumed that the communication area 412a of the base station 412 corresponds to one cell. In the following description, it will be assumed that the communication area 412a of the base station 412 is one cell. And the base station 412 in FIG. 4C is a base station capable of performing online learning according to an embodiment of the present disclosure, and the terminal 423 may also be a terminal capable of performing online learning according to an embodiment of the present disclosure. .

The terminal 423 may perform online learning for the artificial neural network 350 in the same manner as previously described in FIG. 4A. For example, if the artificial neural network 350 is an artificial neural network for CSI feedback, the reference signal transmitted by the base station may be the input 433 or an input sample. The output 434 can be obtained through forward propagation 411 using the reference signal. Under the above assumption, output 434 may be CSI feedback information. Additionally, as described above, verification of the artificial neural network 450 can be performed through backward propagation 442 and weight vector change information can be obtained. When the terminal 423 performs joint learning, weight vector information may be provided to the base station 412. If the terminal does not perform joint learning, the terminal itself may update the artificial neural network 450 based on the online learning results.

Additionally, the base station 412 may perform online learning for the artificial neural network 450 as previously described in FIG. 4B. Specifically, the base station 412 may use a specific signal, such as a reference signal, received from the terminal 423 as an input sample. In other words, the input 435 can be a reference signal transmitted by the terminal 423 through the uplink. At this time, as previously described in FIG. 4B, reciprocity may be established between the downlink channels from the base station 412 to the terminal 423 and the uplink channels from the terminal 423 to the base station 412. Then, the base station 412 can obtain the output 436 of the artificial neural network 450 through forward propagation 443. Additionally, the output 436 can be verified through backward propagation of the artificial neural network 450. Based on these verification results, the base station 412 can obtain weight vector change information. The base station 412 may update the artificial neural network 450 based on the weight vector change information. Additionally, if necessary, the base station 412 can provide update information of the artificial neural network 450 to the terminal 423. This provision method may be provided to specific terminals, may be provided to all terminals through broadcasting, or may be provided for each terminal group. In other words, update information of the artificial neural network 450 can be provided to the terminal in response to Type 0 to Type 2 described in the first embodiment above. Additionally, if the base station 412 supports Type 3 described in the first embodiment of the present disclosure and updates the artificial neural network through joint learning with other base stations belonging to the same cell group, the update information of the artificial neural network 450 It can be provided to all terminals within the base station.

The types of artificial neural networks according to the third embodiment described above, specifically Type A to Type D, can be notified to terminals by the base station. Let's look at a few examples of how a base station notifies terminals of the online learning type of an artificial neural network.

First, the base station can indicate Type A to Type D using a specific message broadcast to all terminals communicating within its cell. Accordingly, each terminal can decide whether to do online learning of the artificial neural network based on the Type indicated in a specific message broadcast by the base station.

Second, the base station can inform terminals of the types supported by the base station in different ways based on each type set at the base station.

For example, in the case of Type A, which does not support online learning, and/or when only the base station supports online learning, the base station can instruct online learning of the artificial neural network using a specific message broadcasted by the base station.

As another example, in the case of Type B, which supports online learning only in the terminal, and/or in the case of Type D, which supports online learning both in the base station and the terminal, a predetermined configuration message that can be set for each terminal capable of online learning is used. It can direct online learning of artificial neural networks.

In the case of Type B and/or Type D, online learning of the artificial neural network can be instructed using a different method, using a predetermined configuration message set for each terminal group of terminals capable of online learning.

The third embodiment of the present disclosure described above can also be applied together to the extent that it does not conflict with the previously described embodiments or other embodiments to be described later.

[Fourth Embodiment]

According to the fourth embodiment of the present disclosure, when applying an artificial neural network for wireless communication in a mobile communication system consisting of a base station and one or more terminals, the learning type is changed in the same manner as Type A to Type D described in the third embodiment. It can be defined.

In the fourth embodiment of the present disclosure, Type A may mean that an artificial neural network pre-trained offline or by another network node can be utilized in the inference process.

In the fourth embodiment of the present disclosure, as previously assumed, it is assumed that an artificial neural network is applied to encode/decode CSI information in a mobile communication system consisting of a base station and a terminal, such as a 5G NR system according to the 3GPP standard. At this time, it is assumed that the artificial neural network for encoding/decoding CSI information is an artificial neural network for CSI feedback.

In the second embodiment of the present disclosure, a method for a terminal to report capability information about an artificial neural network has been described. Specifically, the terminal can report capability information to the base station, including sharing (or creation) type information for the artificial neural network.

However, the learning characteristics of the artificial neural network may not be fully described only with the shared (or created) type information of the artificial neural network described in the second embodiment. For example, if a specific terminal is designed and implemented to have low complexity to reduce costs, operations that utilize offline or pre-trained artificial intelligence models for inference can be supported, but operations that support online learning by the terminal itself are not supported. Support may not be available.

Therefore, in the fourth embodiment of the present disclosure, when the terminal reports capability information about the artificial neural network to the base station, the report may include information on the learning type that the terminal can support for the corresponding artificial neural network. In other words, in the fourth embodiment of the present disclosure, when the terminal reports capability information about the artificial neural network to the base station and/or the network, the terminal can report learning type information that the terminal can support for the corresponding artificial neural network.

For example, if the terminal does not support online learning, when the terminal reports the terminal's capability information to the base station, it can report by setting that only Type A and/or Type C is supported. As another example, if the terminal can support online learning, it can also report whether it can support federated learning. At this time, when reporting the terminal's capability information to the base station, the terminal may report the capability information by indicating that it can support Type B and/or Type D, or by indicating that it can support all types.

The fourth embodiment of the present disclosure described above can be applied together with the first to third embodiments described above and/or other embodiments described below to the extent that they do not conflict with each other.

[Fifth Embodiment]

The fifth embodiment of the present disclosure is as described in the first embodiment of the artificial neural network according to the unit in which the artificial neural network is composed when applying an artificial neural network for wireless communication in a mobile communication system consisting of a base station and one or more terminals. The same types can be defined. For example, Type 0, where the artificial neural network is shared (or created) on a cell basis, Type 1, where the artificial neural network is shared (or created) on a terminal group basis, and artificial neural networks are shared (or created) on a terminal basis. One or more of Type 2, which is the type where the artificial neural network is shared (or created) on a cell group basis, and Type 3, which is the type where the artificial neural network is shared (or created) can be defined.

Additionally, the fifth embodiment of the present disclosure can define the types of methods for learning an artificial neural network as described in the third embodiment described above. For example, Type A, which supports only offline learning (no online learning), Type B, which supports online learning only on the terminal, Type C, which supports online learning only at the base station, and Type C, which supports online learning only at the base station and both the base station and the terminal. You can define Type D, the supported type.

In the fifth embodiment of the present disclosure, the learning type of the artificial neural network that can be supported as follows can be specified for each type of sharing (or creation) of the artificial neural network based on the unit in which the artificial neural network is composed and the presence or absence of online learning support. The learning types of artificial neural networks that can be supported for each type of sharing (or creation) of artificial neural networks can be specified as shown in <Table 1> below.

Type 0 Type 1 Type 2 Type 3 Type A O O O O Type B O O O - Type C O - - - Type D O O - O

In <Table 1> above, the notation “O” indicates support.

As an example of the present disclosure, it is assumed that an artificial neural network is applied to encode/decode CSI information in a mobile communication system consisting of a base station and a terminal, such as a 5G NR system according to the 3GPP standard. Additionally, it is assumed that the artificial neural network for encoding/decoding CSI information is an artificial neural network for CSI feedback. Assume that the base station determines that the channel characteristics for each terminal are different and sets the artificial neural network for CSI feedback to be defined for each terminal (e.g., Type 2). In this case, performing learning for the corresponding artificial neural network at the base station may be a learning method with low feasibility. This is because the computational load in the process of the base station learning a different artificial neural network for each terminal for all terminals will be very high. Therefore, in this disclosure, we propose a method of specifying the learning type supported for each sharing (or creation) type of artificial neural network, as shown in <Table 1>.

As in the example in <Table 1>, when an artificial neural network is defined for each terminal, an agreement can be made in advance between the base station and the terminal to support only offline or online learning at the terminal level. When specifying a combination of a supportable sharing (or creation) type of artificial neural network and a learning type of artificial neural network, as in the method presented in this disclosure, there is an advantage in that it is possible to prevent the terminal from implementing unnecessary functions.

The fifth embodiment of the present disclosure described above can also be applied together with the first to fourth embodiments described above and/or other embodiments described below to the extent that they do not conflict with each other.

[Example 6]

The sixth embodiment of the present disclosure provides a learned neural network including one or more of the following information when applying an artificial neural network for wireless communication in a mobile communication system consisting of a base station and one or more terminals, especially when the base station sets up an artificial neural network for the terminal. You can set the reporting method of information.

(One) (maximum) time interval information for reporting learned information online

(2) Information on when (periodically) to compile online learning

Additionally, the sixth embodiment of the present disclosure proposes a method of collecting learned information using one or more of the following methods.

(One) The terminal reports new learned information within the (maximum) time interval from the previous report on the learned information, and the base station collects only the learned information reported within the (maximum) time interval from the previous report as valid information.

(2) The base station and the terminal mutually promise (periodic) time point information (eg, T ₁ , T ₂ , …, T _N-1 , T _N ) to collect online learning, and collect specific online learning ( Periodic) At time T _K , a method of collecting online learning information reported in the time interval from T _K-1 to T _K

At this time, the learned information may be the entire model (or model parameters) at that point in time or the difference between the set (before learning) model and the learned model.

Model parameters according to the present disclosure may be the structure and weight vector of the artificial neural network described above. Additionally, the difference value between models according to mon initiation may be the amount of change in the weight vector described above. Therefore, in each of the embodiments described above, the weight vector and/or the structure of the weight vector and the artificial neural network can be understood as model parameters. Additionally, in each of the embodiments described above, the amount of change in the weight vector can be understood as the difference value between models.

As an example of the present disclosure, it is assumed that an artificial neural network is applied to encode/decode CSI information in a mobile communication system consisting of a base station and a terminal, such as a 5G NR system according to the 3GPP standard. Hereinafter, it is assumed that the artificial neural network for encoding/decoding CSI information is an artificial neural network for CSI feedback.

When multiple UEs in a cell want to use the same code for the same channel state information, such as the codebook-based CSI feedback method defined in the 5G NR system according to 3GPP standards, an artificial code for CSI feedback is used. It may be desirable for a neural network to be defined on a cell basis. In other words, all terminals in a cell can perform CSI encoding/decoding by applying the same artificial neural network structure and weight vector.

However, in order for all terminals in a cell to apply the same artificial neural network structure and weight vector, the terminals in the cell must be able to cooperate to learn a single artificial neural network. In order to cooperate to learn a single artificial neural network, each terminal must be able to support federated learning. When two or more terminals within a cell perform joint learning, the base station must be able to collect learned information (for example, the entire model or differences compared to the previous model) from a plurality of terminals. Additionally, the timing of collecting learned information between the base station and the terminal must be mutually agreed upon.

In the present disclosure, based on these points, when the base station sets up an artificial neural network for the terminal, it is configured to transmit including (maximum) time interval information for reporting online learned information. Additionally, the terminal reports newly learned information within a (maximum) time interval from the previous report on the learned information. Based on the terminal's report, the base station can only collect learned information reported within a (maximum) time interval from the previous report as valid information.

As an example, the base station fixes the decoder part of the artificial neural network at the time of transmitting the artificial neural network, fixes the model actually applied in each terminal, and transmits the model to be updated to the base station before the maximum point at which the update is performed. can be considered.

In another method, when the base station sets up an artificial neural network for the terminal, it transmits it to the terminal, including information on the (periodic) time point at which online learning is collected, and at the (periodic) time point at which specific online learning is collected, T _K , the previous collection point A method of collecting online learning information reported from terminals within a time interval from T _K-1 to T _K may also be considered.

Figure 5 is a timing diagram for explaining artificial neural network settings and learning information reporting according to the sixth embodiment of the present disclosure.

Referring to FIG. 5, a base station 510 and a plurality of terminals 501, 502, ..., 50n are illustrated. The plurality of terminals 501, 502, ..., 50n illustrated in FIG. 5 are within the communication area (not shown) of the base station 510, and may all be terminals capable of using an artificial neural network. The base station 510 is also a base station that can use an artificial neural network.

The base station 510 may transmit a reporting method setting message to the terminals 501, 502, ..., 51n in step S520. As described above, the reporting method setting message may include (maximum) time interval information for reporting learned information or (periodic) time point information for collecting learning. Additionally, the reporting method setting message may be broadcast to all terminals using the same method or the same artificial neural network, and may be sent to each terminal 501, 502, 50n as illustrated in FIG. 5. It may also be transmitted individually.

In the following description, assume that the base station 510 applies an artificial neural network to encode/decode CSI information. At this time, it is assumed that the artificial neural network for encoding/decoding CSI information is an artificial neural network for CSI feedback. Then, the base station 510 sends a reporting method setting message to the terminals 501, including (maximum) time interval information for reporting learned information for the artificial neural network for CSI feedback or (periodic) time point information for collecting learning. 502, …, 50n) or can be transmitted individually.

Each of the terminals 501, 502, and 50n may receive a reporting method setting message for the artificial neural network for CSI feedback transmitted by the base station in step S520. Therefore, each of the terminals 501, 502, ..., 50n can determine the reporting time for the artificial neural network for CSI feedback based on the received reporting method setting message.

Meanwhile, the base station 510 can set a (maximum) time interval at a certain point after transmitting the reporting method setting message to each terminal. In Figure 5, the (maximum) time interval is illustrated at the top of the base station. It should be noted that the (maximum) time interval illustrated in FIG. 5 is only an example to facilitate understanding, and the present disclosure is not limited to the example in FIG. 5 .

Thereafter, the terminals 501, 502, ..., 50n may communicate with the base station 510 and perform learning of the artificial neural network for CSI feedback according to the present disclosure. It should be noted that Figure 5 does not illustrate the operation of each of the terminals 501, 502, ..., 50n learning an artificial neural network for CSI feedback.

When each of the terminals 501, 502, ..., 50n learns the artificial neural network for CSI feedback, it can report the learning results of the artificial neural network for CSI feedback to the base station. Since this reporting time is based on the learning of an artificial neural network, the reporting time for each terminal (501, 502, ..., 50n) may vary. For example, terminal 1 (501) is the closest terminal in distance from the base station 510, terminal 2 (502) is the second closest terminal in distance from the base station 510, and terminal n (50n) is the base station (50n). 510) Assume that the terminal is the furthest from the terminal in terms of distance.

If the base station 510 broadcasts a reporting method setting message, the reporting method setting message may be received based on the distance between the base station and each terminal. In other words, the broadcasted reporting method setting message is sent to terminal 1 (501), terminal 2 (502), … , the reporting method setting message may be delivered in that order to terminal n (50n).

However, the learning report message of the artificial neural network for CSI feedback reported by each of the terminals 501, 502, ..., 50n may not be transmitted based on the distance between the terminal and the base station. The learning report message for the artificial neural network for CSI feedback is learned at each of the terminals 501, 502, ..., 50n, the reporting time set inside the terminal based on the reporting method setting message, and the base station 510 in the uplink. This may be determined based on various factors, such as the bookstore allowing the transfer. In addition, the timing of online learning for the artificial neural network for CSI feedback in the terminal may be determined based on the computing ability and/or memory capacity of each terminal.

In Figure 5, the timing at which the base station 510 receives a learning report message for the artificial neural network for CSI feedback is terminal 2 (502), terminal 1 (501),... , Terminal n (50n) is exemplified in that order.

In this way, each of the terminals 501, 502, ..., 50n can report the information it has learned and transmit it to the base station 510 using a report message at a set time based on the policy setting message. The learning information for each of the terminals 501, 502, ..., 50n may be the entire model (or model parameters) at the time of configuring the report message, or may include the difference between the set (before learning) model and the model after learning. .

The base station 510 may receive a report message including the online learning results for the artificial neural network from each of the terminals 501, 502, and 50n that use the artificial neural network for CSI feedback in step S530. At this time, the base station 510 can use only the learning information of each terminal received in the set (maximum) time interval as valid information. For example, a report message received from a specific terminal before the maximum time interval may be considered invalid. To this end, the base station 9510 may set all of the memories (not illustrated in FIG. 5) that receive and store the report message to initial values when the (maximum) time interval begins. That is, if the section of the (maximum) time interval illustrated in FIG. 5 is T _k , and the section of the (maximum) time interval including step S520 is T _k-1 (not illustrated in FIG. 5), the base station ( 510) may consider the report message received in the T _k interval as valid, and determine that the report message received in the T _k-1 interval is invalid.

In describing the embodiment above, the artificial neural network for CSI feedback was explained as an example. However, the present disclosure is not limited to this and can be applied to artificial neural networks for beam management, improving positioning accuracy, as well as other purposes. In addition, the sixth embodiment of the present disclosure described above can be applied together with the previously described embodiments or the embodiments described later to the extent that they do not conflict with each other.

[Embodiment 7]

The seventh embodiment of the present disclosure is when applying an artificial neural network for wireless communication in a mobile communication system consisting of a base station and one or more terminals, especially when the terminal reports capability information related to the artificial neural network to the base station, one or more of the following: You can report it including information.

(One) Artificial neural network performance capabilities

A. Number of floating point operations per second (FLOPS)

B. Number of parameters that can be stored (memory size)

(2) Artificial neural network learning ability

A. Number of operations per unit time (FLOPS)

B. Number of parameters that can be stored (memory)

Here, performing an artificial neural network may mean an inference operation using an artificial neural network.

As an example of the present disclosure, assume that the terminal supports an artificial neural network to improve wireless communication performance in a mobile communication system consisting of a base station and a terminal, such as a 5G NR system according to the 3GPP standard. In this case, the base station can set up various artificial neural networks with a high degree of freedom for the terminal. However, depending on the implementation method of the terminal, it may be difficult to perform inference or learning on the complex artificial neural network required by the base station. Therefore, in the seventh embodiment of the present disclosure, the terminal reports the capability for artificial neural network performance and/or artificial neural network learning ability expressed in terms of the number of operations per unit time (FLOPS) and/or the number of storable parameters (memory). can do. Accordingly, the base station can configure an artificial neural network based on the capability information reported by the terminal.

The seventh embodiment of the present disclosure described above can also be applied together to the extent that it does not conflict with the previously described embodiments.

In the above, each embodiment was divided and described for each embodiment. Hereinafter, a case in which the embodiments according to the present disclosure are performed in combination will be examined using a single embodiment.

Figure 6 is a signal flow diagram when learning an artificial neural network is performed between a terminal and a base station according to the initiation.

In Figure 6, the description will be based on one terminal 601 and one base station 602. However, it will be apparent to those skilled in the art that a plurality of terminals may be included in the base station 602, and that the operation described in FIG. 6 may also be applied to other terminals.

In step S610, the base station 602 may request a calorie information report from the terminal 601. Here, the competency information can be competency information for the artificial neural network. For example, information about whether the terminal 601 can use an artificial neural network, types for sharing (or creating) artificial neural networks, online learning types of artificial neural networks, performance capabilities of artificial neural networks, learning capabilities of artificial neural networks, etc. If acquired in advance, the base station 602 may not perform the capability information reporting request.

For example, looking at cases where the base station 602 does not request capability information reporting, when the terminal 601 initially registers with the base station 602, capability information related to artificial intelligence may be reported to the base station in advance. It may be possible. In this way, if the terminal 601 reports capability information related to artificial intelligence in the initial registration step with the base station 602, step S610 may not be performed.

As another example, if the terminal 601 moves by handover while communicating at another base station, capability information related to artificial intelligence of the terminal 601 may be transmitted from the other base station. In this way, when information about the handover terminal is received from another base station, a capability information report request may not be made.

In step S612, the terminal 601 may transmit a capability information report message to the base station 602 in response to a capability information report request or on its own. Even if there is no capability report request, the terminal 601 can independently report a capability information reporting message related to the artificial neural network to the base station 602. In this case, as described above, if necessary during the initial access procedure to the base station 602, a capability information reporting message related to the artificial neural network can be reported to the base station 602. In another case, the terminal 601 may be configured to transmit a capability information report message related to the artificial neural network to the base station 602 when it is determined that use of a specific artificial neural network is necessary.

The capability information reporting message related to the artificial neural network may include the information described in the above embodiments. For example, the terminal 601 may report the capability information report message by including at least one of the types based on the sharing (or creation) method of the artificial neural network. The terminal 601 may report the capability information report message by including at least one of the learning type information of the artificial neural network. The terminal 601 sends information (for example, the information shown in <Table 1>) that combines a type based on the sharing (or creation) method of the artificial neural network and the corresponding learning type information of the artificial neural network in the capability information reporting message. It can be collected and reported as one piece of information. The terminal 601 may report the capability information report message by including artificial neural network performance capability and/or artificial neural network learning capability information.

In step S612, the base station 602 may receive a capability information report message from the terminal 601, and generate artificial neural network setting information based on the capability information report message received in step S614.

Artificial neural network setting information may include at least one of the following information. For example, the artificial neural network setting information may include the sharing (or creation) type of the artificial neural network and the learning type of the artificial neural network. In another way, as previously exemplified in <Table 1>, the mapping information between the sharing (or creation) type of the artificial neural network and the learning type of the artificial neural network is used to identify the sharing (or creation) type of a specific artificial neural network and the learning information of the artificial neural network. It may also contain one piece of information that indicates. Additionally, the artificial neural network setting information may include (maximum) time interval information or collection point information described in the sixth embodiment above.

In step S616, the base station 602 may transmit the artificial neural network setting information generated in step S614 to the terminal 601. This artificial neural network setting information may be transmitted individually to terminals, in units of terminal groups, or to terminals in the entire cell, depending on the type of sharing (or creation) of the artificial neural network. When transmitting to terminals in an entire cell, it may also be transmitted individually to each terminal. Additionally, if a message capable of broadcasting information related to the artificial neural network is defined within the communication system, it can be transmitted using the defined broadcast message.

The terminal 601, which has received the artificial neural network setting information in step S616, may store the artificial neural network setting information in its internal memory.

In step S618, the terminal 601 can learn an artificial neural network based on the received artificial neural network setting information. Additionally, the terminal 601 may update the artificial neural network included in the terminal 601 based on the learning results of the artificial neural network or store information to be updated in the internal memory.

Additionally, the terminal 601 can generate an online learning information report message. The online learning information reporting message can be generated as an online learning information reporting message for the entire updated online learning model (the updated model itself) or the difference value between the artificial neural network model before the update and the artificial neural network model after the update. Here, the difference value between artificial neural network models may be the amount of change in the weight vector described above.

In step S620, the terminal 601 may transmit an online learning information report message to the base station 602. At this time, the timing of transmitting the online learning information reporting message is the reporting timing or reporting interval set in the terminal 601 based on the (maximum) time interval information or collection timing information of the reporting message included in the artificial neural network setting information received in step S616. It can be done in

Since the structure of the online learning information reporting message reported by the terminal 601 to the base station 602 has been described above, redundant explanation will be omitted.

In step S622, the base station 602 may collect the online learning information report message received in step S620. Figure 6 illustrates how one terminal reports to the base station 602. However, as previously illustrated in FIG. 5, the report message may be received from multiple terminals. Therefore, the base station 602 can collect online learning information reporting messages received from each terminal.

Additionally, in step S622, the base station 602 can determine the weight vector update direction for the terminals. The base station 602 determines the weight vector update direction when joint learning of terminals is performed. In this disclosure, how to determine the update direction of the weight vector is beyond the topic discussed in this disclosure, so detailed description will be omitted.

The base station 602 may determine the weight vector update direction and generate this as weight vector update information. If the terminals perform joint learning and the base station 602 generates weight update information in step S622, the base station 602 may transmit the weight vector update information to each terminal in step S624.

Meanwhile, FIG. 6 described above is an example that can be implemented using the first to seventh embodiments. Accordingly, the overall features of the present disclosure are not limited to the one example described in FIG. 6, and some of them may not be performed or may be performed additionally although they are described in each embodiment, but not mentioned in the description of FIG. 6. It may be possible.

The operation of the method according to the embodiment of the present disclosure can be implemented as a computer-readable program or code on a computer-readable recording medium. Computer-readable recording media include all types of recording devices that store information that can be read by a computer system. Additionally, computer-readable recording media can be distributed across networked computer systems so that computer-readable programs or codes can be stored and executed in a distributed manner.

Additionally, computer-readable recording media may include hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. Program instructions may include not only machine language code such as that created by a compiler, but also high-level language code that can be executed by a computer using an interpreter, etc.

Although some aspects of the disclosure have been described in the context of an apparatus, it may also refer to a corresponding method description, where a block or device corresponds to a method step or feature of a method step. Similarly, aspects described in the context of a method may also be represented by corresponding blocks or items or features of a corresponding device. Some or all of the method steps may be performed by (or using) a hardware device, such as a microprocessor, programmable computer, or electronic circuit, for example. In some embodiments, at least one or more of the most important method steps may be performed by such an apparatus.

In embodiments, a programmable logic device (e.g., a field programmable gate array) may be used to perform some or all of the functionality of the methods described herein. In embodiments, a field-programmable gate array may operate in conjunction with a microprocessor to perform one of the methods described herein. In general, the methods are preferably performed by some hardware device.

Although the present disclosure has been described above with reference to preferred embodiments, those skilled in the art may modify and change the present disclosure in various ways without departing from the spirit and scope of the present disclosure as set forth in the claims below. You will understand that it is possible.

Claims (20)

In the base station method,
Receiving a capability information reporting message related to an artificial neural network from each of the terminals in the base station;
generating artificial neural network configuration information based on the capabilities of the base station and the capability report message received from each of the terminals; and
Transmitting the generated artificial neural network configuration information to the terminals in the base station,
The artificial neural network setting information may include a type in which the artificial neural network is shared on a cell basis, a type in which the artificial neural network is shared on a terminal group basis, a type in which the artificial neural network is shared on a terminal basis, or a type in which the artificial neural network is shared on a cell group basis. Containing information indicating one of the types of being,
Base station method. In claim 1,
The capability information reporting message related to the artificial neural network is:
A type in which the artificial neural network is shared on a per cell basis supported by each of the terminals, a type on which the artificial neural network is shared on a per terminal group basis, a type in which the artificial neural network is shared on a per terminal basis, or a type in which the artificial neural network is shared on a per cell group basis. Containing information indicating one of the types with which the artificial neural network is shared,
Base station method. In claim 1,
When the artificial neural network setting information indicates a type in which the artificial neural network is shared on a per terminal group basis, the terminal group includes terminals having the same classification indicator,
Base station method. In claim 1,
When the artificial neural network setting information indicates a type in which the artificial neural network is shared on a cell group basis, the cell group includes cells having the same classification indicator,
Base station method. In claim 1,
The artificial neural network setting information is,
A type that does not support online learning, or a type that supports the online learning only in the terminals, a type that supports the online learning only in the base station, or a type that supports the online learning in both the base station and the terminals. Containing further indicative information,
Base station method. In claim 5,
The artificial neural network setting information is,
Further comprising at least one of time interval information set to report online learning information from each of the terminals or time point information for collecting the online learning from each of the terminals,
Base station method. In claim 6,
The method of the base station is,
Further comprising: receiving an online learning information report message from the terminals based on the artificial neural network setting information,
The online learning information report message includes the entire model of the artificial neural network or a difference value between the artificial neural network before online learning and the artificial neural network after online learning,
Base station method. In claim 7,
The method of the base station is,
determining an update to the weight vector of the artificial neural network shared with the terminals based on the received online learning report message; and
Further comprising: transmitting weight update information to the terminals based on an update decision for the weight vector of the artificial neural network,
Base station method. In the terminal method,
Receiving a capability information report request message for an artificial neural network from a base station of a cell to which the terminal belongs;
transmitting a capability information report message related to the artificial neural network to the base station in response to receiving the capability information report request message;
Receiving artificial neural network configuration information from the base station; and
It includes setting up an artificial neural network in the terminal based on the received artificial neural network setting information,
The artificial neural network setting information may include a type of the artificial neural network being shared on a cell basis, a type of the artificial neural network being shared on a terminal group basis, a type of the artificial neural network being shared on a terminal basis, or a type of the artificial neural network being shared on a cell group basis. containing information indicating one of the types being shared;
Terminal method. In claim 9,
The capability information reporting message related to the artificial neural network is:
A type in which the artificial neural network is shared on a per cell basis supported by the terminal, a type on which the artificial neural network is shared on a per terminal group basis, a type in which the artificial neural network is shared on a per terminal basis, or a type in which the artificial neural network is shared on a per cell group basis. Containing information indicating one of the types of neural networks being shared,
Terminal method. In claim 9,
The artificial neural network setting information is,
Indicates one of a type that does not support online learning, or a type that supports the online learning only in the terminal, or a type that supports the online learning only in the base station, or a type that supports the online learning in both the base station and the terminal. Containing further information,
Terminal method. In claim 9,
The artificial neural network setting information is,
Further comprising at least one of time interval information set to report online learning information from the terminal or time point information for collecting online learning from the terminal,
Terminal method. In claim 9,
The method of the terminal is,
Online learning about the artificial neural network set in the terminal;
generating an online learning information report message based on the online learning; and
Transmitting the online learning information reporting message to the base station based on time interval information set to report the online learning information or time point information for collecting the online learning; further comprising,
Terminal method. At the base station,
comprising a processor, wherein the processor allows the base station to:
Receiving a capability information reporting message related to an artificial neural network from each of the terminals in the base station;
Generate artificial neural network configuration information based on the capabilities of the base station and the capability report message received from each of the terminals; and
causing the generated artificial neural network configuration information to be transmitted to the terminals in the base station,
The artificial neural network setting information may include a type in which the artificial neural network is shared on a cell basis, a type in which the artificial neural network is shared on a terminal group basis, a type in which the artificial neural network is shared on a terminal basis, or a type in which the artificial neural network is shared on a cell group basis. Containing information indicating one of the types of being,
Base station. In claim 14,
The capability information reporting message related to the artificial neural network is:
A type in which the artificial neural network is shared on a per cell basis supported by each of the terminals, a type on which the artificial neural network is shared on a per terminal group basis, a type in which the artificial neural network is shared on a per terminal basis, or a type in which the artificial neural network is shared on a per cell group basis. Containing information indicating one of the types with which the artificial neural network is shared,
Base station. In claim 14,
If the artificial neural network setting information indicates a type in which the artificial neural network is shared on a per terminal group basis, the terminal group includes terminals with the same classification indicator, and
When the artificial neural network setting information indicates a type in which the artificial neural network is shared on a cell group basis, the cell group includes cells having the same classification indicator,
Base station. In claim 14,
The artificial neural network setting information is,
A type that does not support online learning, or a type that supports the online learning only in the terminals, a type that supports the online learning only in the base station, or a type that supports the online learning in both the base station and the terminals. Containing further indicative information,
Base station. In claim 17,
The artificial neural network setting information is,
Further comprising at least one of time interval information set to report online learning information from each of the terminals or time point information for collecting the online learning from each of the terminals,
Base station. In claim 18,
The processor may cause the base station to:
further cause to receive an online learning information report message from the terminals based on the artificial neural network setting information,
The online learning information report message includes the entire model of the artificial neural network or a difference value between the artificial neural network before online learning and the artificial neural network after online learning,
Base station. In claim 19,
The processor may cause the base station to:
determine an update to the weight vector of the artificial neural network shared with the terminals based on the received online learning report message; and
Further causing weight update information to be transmitted to the terminals based on the update decision for the weight vector of the artificial neural network.
Base station.

Images