WO2023066346A1 - 一种通信方法及装置 - Google Patents

一种通信方法及装置 Download PDF

Info

Publication number
WO2023066346A1
WO2023066346A1 PCT/CN2022/126467 CN2022126467W WO2023066346A1 WO 2023066346 A1 WO2023066346 A1 WO 2023066346A1 CN 2022126467 W CN2022126467 W CN 2022126467W WO 2023066346 A1 WO2023066346 A1 WO 2023066346A1
Authority
WO
WIPO (PCT)
Prior art keywords
training data
artificial intelligence
training
intelligence model
model
Prior art date
Application number
PCT/CN2022/126467
Other languages
English (en)
French (fr)
Inventor
柴晓萌
吴艺群
Original Assignee
华为技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Publication of WO2023066346A1 publication Critical patent/WO2023066346A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L5/00—Arrangements affording multiple use of the transmission path
    • H04L5/003—Arrangements for allocating sub-channels of the transmission path
    • H04L5/0058—Allocation criteria
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B17/00—Monitoring; Testing
    • H04B17/30—Monitoring; Testing of propagation channels
    • H04B17/309—Measuring or estimating channel quality parameters
    • H04B17/318—Received signal strength
    • H04B17/327—Received signal code power [RSCP]
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L5/00—Arrangements affording multiple use of the transmission path
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L5/00—Arrangements affording multiple use of the transmission path
    • H04L5/003—Arrangements for allocating sub-channels of the transmission path
    • H04L5/0048—Allocation of pilot signals, i.e. of signals known to the receiver
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L5/00—Arrangements affording multiple use of the transmission path
    • H04L5/003—Arrangements for allocating sub-channels of the transmission path
    • H04L5/0053—Allocation of signaling, i.e. of overhead other than pilot signals
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W72/00—Local resource management
    • H04W72/04—Wireless resource allocation
    • H04W72/044—Wireless resource allocation based on the type of the allocated resource
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W72/00—Local resource management
    • H04W72/04—Wireless resource allocation
    • H04W72/044—Wireless resource allocation based on the type of the allocated resource
    • H04W72/0446—Resources in time domain, e.g. slots or frames
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W72/00—Local resource management
    • H04W72/04—Wireless resource allocation
    • H04W72/044—Wireless resource allocation based on the type of the allocated resource
    • H04W72/0453—Resources in frequency domain, e.g. a carrier in FDMA
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W72/00—Local resource management
    • H04W72/04—Wireless resource allocation
    • H04W72/044—Wireless resource allocation based on the type of the allocated resource
    • H04W72/046—Wireless resource allocation based on the type of the allocated resource the resource being in the space domain, e.g. beams
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Definitions

  • the present disclosure relates to the field of communication technologies, and in particular, to a communication method and device.
  • a wireless communication network such as a mobile communication network
  • services supported by the network are becoming more and more diverse, and therefore requirements to be met are becoming more and more diverse.
  • the network needs to be able to support ultra-high speed, ultra-low latency, and/or very large connections.
  • This feature makes network planning, network configuration, and/or resource scheduling increasingly complex.
  • functions of the network become more and more powerful, such as supporting higher and higher spectrum, supporting high-order multiple input multiple output (MIMO) technology, supporting beamforming, and/or supporting new technologies such as beam management, etc. technology, making network energy saving a hot research topic.
  • MIMO multiple input multiple output
  • These new requirements, new scenarios, and new features have brought unprecedented challenges to network planning, O&M, and efficient operations.
  • artificial intelligence technology can be introduced into the wireless communication network to realize network intelligence. Based on this, how to effectively implement artificial intelligence in the network is a problem worth studying.
  • the present disclosure provides a communication method and device for realizing how to construct a training data set, so as to train and obtain a better neural network.
  • the present disclosure provides a communication method, which can be executed on a network device side, such as an access network device side or an AI entity independent of the access network device.
  • the method can be implemented by software, hardware, or a combination of software and hardware.
  • the method is performed by an access network device, or by a larger device including the access network device, or by a circuit system capable of implementing the functions of the access network device, or by a device independent of the access network device
  • the execution of the AI entity is not limited.
  • the access network device is a base station.
  • the method includes: obtaining N training data, for one training data in the N training data, the training data corresponds to a tag information, the tag information is used to indicate the attributes of the training data, N is an integer greater than 0;
  • the device sends indication information, and the indication information is used to indicate the information of M artificial intelligence models.
  • the artificial intelligence model is trained according to X training data among N training data.
  • the X pieces of training data are determined according to the label information, and M and X are integers greater than 0.
  • training data corresponding to different label information can be constructed from the N training data, so as to train M or more artificial intelligence models.
  • the training data for training an artificial intelligence model satisfies the tag information corresponding to the artificial intelligence model, and scene-based artificial intelligence model training can be realized.
  • the reasoning performance of the scene-based artificial intelligence model in the corresponding scene is more accurate.
  • the artificial intelligence model corresponds to a training condition
  • the label information corresponding to the X training data used to train the artificial intelligence model satisfies the artificial intelligence
  • the training conditions corresponding to the model, or the attributes corresponding to the X training data meet the training conditions corresponding to the artificial intelligence model.
  • the training data sets corresponding to different training conditions are constructed, so that the wireless channel distribution corresponding to the training data of each artificial intelligence model is relatively concentrated or similar.
  • the training data for training an artificial intelligence model satisfies the training conditions corresponding to the artificial intelligence model, and scene-based artificial intelligence model training can be realized.
  • the tag information is used to indicate at least one of the following attributes: the index of the synchronization signal block associated with the training data; the reference signal received power of the synchronization signal block associated with the training data; the type of the wireless channel associated with the training data , the types include line-of-sight and non-line-of-sight; the timing advance associated with the training data; and the cell ID of the cell associated with the training data.
  • the above attributes are related to the wireless channel. Through the above attributes, different training data can be bound to the wireless channel, so that when training the artificial intelligence model, an appropriate scene-based model can be adaptively selected according to the above attributes.
  • the training condition includes at least one of the following: an index of the synchronization signal block; a value range of the received power of the reference signal; a type of the wireless channel; a value range of the timing advance;
  • One or more of the index of the synchronization signal block, the value range of the reference signal received power, the type of the wireless channel, and the value range of the timing advance are related to the wireless channel.
  • the training conditions include the above information, the training conditions can be passed. Screen out the required wireless channel-related training data to build a scenario-based training data set and enable scenario-based model training.
  • each artificial intelligence model can be trained with the training data of multiple cells, enabling the scene model of the multi-cell environment, so that multiple cells can share the same artificial intelligence model.
  • the training conditions corresponding to the artificial intelligence model are preset.
  • the training conditions corresponding to the artificial intelligence model are determined according to the X training data used to train the artificial intelligence model, and the X training data used to train the artificial intelligence model have the same
  • the training data of the clustering feature, or, the X training data used for training the artificial intelligence model is the training data of the clustering feature in the same range among the N training data.
  • the method further includes: sending M training conditions corresponding to the M artificial intelligence models to the terminal device, wherein the M artificial intelligence models correspond to the M training conditions one-to-one.
  • sending indication information to the terminal device includes: receiving model indication information from the terminal device, where the model indication information is used to indicate the artificial intelligence model requested by the terminal device; sending indication information to the terminal device, where the indication information indicates Information about the artificial intelligence model requested by the terminal device.
  • the model indication information is used to indicate: the index of the requested artificial intelligence model, the training condition of the requested artificial intelligence model, the attribute or label information of the training data of the requested artificial intelligence model.
  • the terminal device requests the required artificial intelligence model, which can realize the precise indication of the artificial intelligence model, avoid sending the unnecessary artificial intelligence model to the terminal device, and reduce signaling overhead.
  • a communication method is provided, which can be executed on a terminal device side.
  • the method can be implemented by software, hardware, or a combination of software and hardware.
  • the method is performed by a terminal device, or by a circuit system, or by a larger device including the terminal device, and the circuit system can realize the functions of the terminal device.
  • the method includes:
  • the instruction information is used to indicate the information of M artificial intelligence models, for one artificial intelligence model in the M artificial intelligence models, the artificial intelligence model is trained according to X training data in N training data, M, N and X are integers greater than 0.
  • the training data corresponds to a piece of tag information
  • the tag information is used to indicate the attribute of the training data.
  • the artificial intelligence model corresponds to a training condition
  • the label information corresponding to the X training data satisfies the training condition corresponding to the artificial intelligence model.
  • the method further includes: receiving M training conditions corresponding to M artificial intelligence models, wherein the M artificial intelligence models correspond to the M training conditions one-to-one.
  • receiving indication information includes: sending model indication information, where the model indication information is used to indicate the requested artificial intelligence model;
  • Receive indication information where the indication information indicates the information of the requested artificial intelligence model.
  • the present application further provides a communication device, where the communication device implements any method or any implementation manner provided in the foregoing first aspect.
  • the communication device may be realized by hardware, may be realized by software, or may be realized by executing corresponding software by hardware.
  • the hardware or software includes one or more units or modules corresponding to the above functions.
  • the communication device includes: a processor, where the processor is configured to support the communication device to execute the method described in the above first aspect.
  • the communication device may also include a memory, which may be coupled to the processor, which holds program instructions and data necessary for the communication device.
  • the communication device further includes an interface circuit for supporting the communication device to communicate with other communication devices.
  • the structure of the communication device includes a processing unit and a communication unit, and these units can perform corresponding functions in the foregoing method examples.
  • these units can perform corresponding functions in the foregoing method examples.
  • the present application further provides a communication device, where the communication device implements any method or any implementation manner provided in the above second aspect.
  • the communication device may be realized by hardware, may be realized by software, or may be realized by executing corresponding software by hardware.
  • the hardware or software includes one or more units or modules corresponding to the above functions.
  • the communication device includes: a processor, where the processor is configured to support the communication device to execute the method shown in the second aspect above.
  • the communication device may also include a memory, which may be coupled to the processor, which holds program instructions and data necessary for the communication device.
  • the communication device further includes an interface circuit for supporting the communication device to communicate with other communication devices.
  • the structure of the communication device includes a processing unit and a communication unit, and these units can perform corresponding functions in the above method examples.
  • these units can perform corresponding functions in the above method examples.
  • a communication device including a processor and an interface circuit, and the interface circuit is used to receive signals from other communication devices other than the communication device and transmit them to the processor or send signals from the processor
  • the processor is configured to execute the computer program or instruction stored in the memory to implement the method in any possible implementation manner of the foregoing first aspect.
  • the apparatus further includes a memory in which computer programs or instructions are stored.
  • a communication device including a processor and an interface circuit, and the interface circuit is used to receive signals from other communication devices other than the communication device and transmit them to the processor or send signals from the processor
  • the processor is configured to execute the computer program or instruction stored in the memory to implement the method in any possible implementation manner of the aforementioned second aspect.
  • the apparatus further includes a memory in which computer programs or instructions are stored.
  • a computer-readable storage medium in which a computer program or an instruction is stored, and when the computer program or instruction is run on a computer, the computer is enabled to implement the aforementioned first aspect A method in any possible implementation of .
  • a computer-readable storage medium in which a computer program or instruction is stored, and when the computer program or instruction is run on a computer, the computer realizes the aforementioned second aspect A method in any possible implementation of .
  • a computer program product including computer readable instructions is provided, and when the computer readable instructions are run on a computer, the computer is made to implement the method in any possible implementation manner in the aforementioned first aspect.
  • a computer program product including computer-readable instructions is provided, and when the computer-readable instructions are run on a computer, the computer is made to implement the method in any possible implementation manner in the aforementioned second aspect.
  • a chip in an eleventh aspect, includes a processor, and may also include a memory, the processor is coupled to the memory, and is used to execute computer programs or instructions stored in the memory, so that the chip implements the aforementioned first aspect A method in any possible implementation of .
  • a chip in a twelfth aspect, includes a processor, and may also include a memory, the processor is coupled to the memory, and is used to execute computer programs or instructions stored in the memory, so that the chip implements the aforementioned second aspect A method in any possible implementation of .
  • a communication system in a thirteenth aspect, includes the device (such as an access network device) for realizing the first aspect and the device (such as a terminal device) for realizing the second aspect.
  • the device such as an access network device
  • the device such as a terminal device
  • FIG. 1 is a schematic structural diagram of a communication system to which the present disclosure can be applied;
  • FIG. 2 is a schematic structural diagram of another communication system to which the present disclosure can be applied;
  • FIG. 3 is a schematic diagram of a layer relationship of a neural network provided by the present disclosure.
  • FIG. 4(a) to FIG. 4(e) are exemplary diagrams of a network architecture to which the method provided by the present disclosure can be applied;
  • FIG. 5 is a schematic flowchart of a configuration method provided by the present disclosure.
  • FIG. 6 is a schematic diagram of synchronous signal transmission provided by the present disclosure.
  • FIG. 7 is a schematic structural diagram of a communication device provided by the present disclosure.
  • Fig. 8 is a schematic structural diagram of a communication device provided by the present disclosure.
  • the technical solution of the present disclosure can be applied to various communication systems, such as: long term evolution (long term evolution, LTE) system, fifth generation (5th generation, 5G) mobile communication system, or next generation mobile communication system, etc. Do limit.
  • long term evolution long term evolution
  • 5G fifth generation
  • next generation mobile communication system etc. Do limit.
  • the 5G system may also be called a new radio (NR) system.
  • NR new radio
  • the interaction between the terminal device and the access network device is used as an example for description. It should be noted that the method provided in the present disclosure can be applied not only to the interaction between the terminal device and the In the interaction between any two devices, the present disclosure does not limit this.
  • Fig. 1 is a schematic structural diagram of a communication system to which the present disclosure can be applied, and the communication system includes an access network device and a terminal device.
  • the terminal device can establish a connection with the access network device and communicate with the access network device.
  • Fig. 1 is only a schematic diagram, and the present disclosure does not limit the number of access network devices and terminal devices included in the communication system.
  • a terminal device may be referred to as a terminal for short.
  • the terminal device can communicate with one or more core networks via a radio access network (radio access network, RAN).
  • radio access network radio access network
  • a terminal device may be a device with a wireless transceiver function or a chip that may be set in the device.
  • a terminal device may also be referred to as user equipment (UE), access terminal, subscriber unit, subscriber station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, wireless communication device, user agent, or user device.
  • the terminal device in the present disclosure may be a mobile phone, a tablet computer (Pad), a computer with a wireless transceiver function, a virtual reality (virtual reality, VR) terminal, an augmented reality (augmented reality, AR) terminal, a wearable device , vehicles, drones, helicopters, airplanes, ships, robots, robotic arms, or smart home appliances, etc.
  • VR virtual reality
  • AR augmented reality
  • the terminal device in the present disclosure can be widely applied to communication in various scenarios, including but not limited to at least one of the following scenarios: enhanced mobile broadband (enhanced mobile broadband, eMBB), ultra-reliable low-latency communication (ultra-reliable low- latency communication (URLLC), device-to-device (D2D), vehicle to everything (V2X), machine-type communication (MTC), massive machine-type communication (massive machine- type communication, mMTC), Internet of Things (IOT), virtual reality, augmented reality, industrial control, automatic driving, telemedicine, smart grid, smart furniture, smart office, smart wear, smart transportation, or smart city, etc. .
  • the present disclosure does not limit the specific technology and specific equipment form adopted by the terminal.
  • the device for realizing the function of the terminal may be a terminal; it may also be a device capable of supporting the terminal to realize the function, such as a chip system, a hardware circuit, a software module, or a hardware circuit plus a software module, and the device may be Installs in the terminal or can be used with the terminal.
  • the technical solution provided by the present disclosure will be described below by taking the terminal as an example of the device for realizing the function of the terminal.
  • Access network equipment can be base station (base station), node B (NodeB), evolved node B (evolved NodeB, eNodeB), transmission reception point (transmission reception point, TRP), fifth generation (5th generation, 5G) mobile
  • the access network equipment may be
  • the protocol layer structure may include a control plane protocol layer structure and a user plane protocol layer structure.
  • the control plane protocol layer structure may include a radio resource control (radio resource control, RRC) layer, a packet data convergence protocol (packet data convergence protocol, PDCP) layer, a radio link control (radio link control, RLC) layer, a media The access control (media access control, MAC) layer and the function of the protocol layer such as the physical layer.
  • the user plane protocol layer structure may include the functions of the PDCP layer, the RLC layer, the MAC layer, and the physical layer.
  • the service data adaptation protocol (service data adaptation protocol) may also be included above the PDCP layer. protocol, SDAP) layer.
  • the protocol layer structure between the access network device and the terminal may also include an artificial intelligence (AI) layer, which is used to transmit data related to the AI function.
  • AI artificial intelligence
  • Access network devices may include CUs and DUs. Multiple DUs can be centrally controlled by one CU.
  • the interface between the CU and the DU may be referred to as an F1 interface.
  • the control plane (control panel, CP) interface may be F1-C
  • the user plane (user panel, UP) interface may be F1-U.
  • the present disclosure does not limit the specific names of the interfaces.
  • CU and DU can be divided according to the protocol layer of the wireless network: for example, the functions of the PDCP layer and above protocol layers are set in the CU, and the functions of the protocol layers below the PDCP layer (such as RLC layer and MAC layer, etc.) are set in the DU; another example, PDCP The functions of the protocol layer above the layer are set in the CU, and the functions of the PDCP layer and the protocol layer below are set in the DU, without restriction.
  • the CU or DU may be divided into functions having more protocol layers, and for example, the CU or DU may also be divided into part processing functions having protocol layers.
  • part of the functions of the RLC layer and the functions of the protocol layers above the RLC layer are set in the CU, and the rest of the functions of the RLC layer and the functions of the protocol layers below the RLC layer are set in the DU.
  • the functions of the CU or DU can also be divided according to the business type or other system requirements, for example, according to the delay, and the functions whose processing time needs to meet the delay requirement are set in the DU, which does not need to meet the delay
  • the required feature set is in the CU.
  • the CU may also have one or more functions of the core network.
  • the CU can be set on the network side to facilitate centralized management.
  • the wireless unit (radio unit, RU) of the DU is set remotely.
  • the RU may have a radio frequency function.
  • DUs and RUs can be divided in a physical layer (physical layer, PHY).
  • the DU can implement high-level functions in the PHY layer
  • the RU can implement low-level functions in the PHY layer.
  • the functions of the PHY layer may include at least one of the following: adding a cyclic redundancy check (cyclic redundancy check, CRC) bit, channel coding, rate matching, scrambling, modulation, layer mapping, precoding, Resource mapping, physical antenna mapping, or radio frequency transmission functions.
  • CRC cyclic redundancy check
  • the functions of the PHY layer may include at least one of the following: CRC check, channel decoding, de-rate matching, descrambling, demodulation, de-layer mapping, channel detection, resource de-mapping, physical antenna de-mapping, or RF receiving function.
  • the high-level functions in the PHY layer may include a part of the functions of the PHY layer, for example, this part of the functions is closer to the MAC layer, and the lower-level functions in the PHY layer may include another part of the functions of the PHY layer, for example, this part of the functions is closer to the radio frequency function.
  • high-level functions in the PHY layer may include adding CRC codes, channel coding, rate matching, scrambling, modulation, and layer mapping
  • low-level functions in the PHY layer may include precoding, resource mapping, physical antenna mapping, and radio transmission functions
  • high-level functions in the PHY layer may include adding CRC codes, channel coding, rate matching, scrambling, modulation, layer mapping, and precoding
  • low-level functions in the PHY layer may include resource mapping, physical antenna mapping, and radio frequency send function.
  • the high-level functions in the PHY layer may include CRC check, channel decoding, de-rate matching, decoding, demodulation, and de-layer mapping
  • the low-level functions in the PHY layer may include channel detection, resource de-mapping, physical antenna de-mapping, and RF receiving functions
  • the high-level functions in the PHY layer may include CRC check, channel decoding, de-rate matching, decoding, demodulation, de-layer mapping, and channel detection
  • the low-level functions in the PHY layer may include resource de-mapping , physical antenna demapping, and RF receiving functions.
  • the function of the CU may be implemented by one entity, or may also be implemented by different entities.
  • the functions of the CU can be further divided, that is, the control plane and the user plane are separated and realized by different entities, namely, the control plane CU entity (ie, the CU-CP entity) and the user plane CU entity (ie, the CU-UP entity) .
  • the CU-CP entity and the CU-UP entity can be coupled with the DU to jointly complete the functions of the access network equipment.
  • any one of the foregoing DU, CU, CU-CP, CU-UP, and RU may be a software module, a hardware structure, or a software module+hardware structure, without limitation.
  • the existence forms of different entities may be different, which is not limited.
  • DU, CU, CU-CP, and CU-UP are software modules
  • RU is a hardware structure.
  • the access network device includes CU-CP, CU-UP, DU and RU.
  • the execution subject of the present disclosure includes DU, or includes DU and RU, or includes CU-CP, DU and RU, or includes CU-UP, DU and RU, without limitation.
  • the methods performed by each module are also within the protection scope of the present disclosure.
  • the device for realizing the function of the access network device may be the access network device; it may also be a device capable of supporting the access network device to realize the function, such as a chip system.
  • the device can be installed in the access network equipment or matched with the access network equipment.
  • a dedicated AI entity (or called an AI module) may also be introduced into the network.
  • the AI entity can correspond to an independent network element; or it can be located inside a certain network element, which can be a core network device, an access network device, or a network management (operation, administration and maintenance, OAM), etc.
  • the AI entity is located outside the access network device and can communicate with the access network device.
  • the access network device can forward the data related to the AI model reported by the terminal device to the AI entity, and the AI entity performs operations such as training data set construction and model training, and forwards the trained artificial intelligence model to the AI entity through the access network device.
  • Each terminal device can forward the data related to the AI model reported by the terminal device to the AI entity, and the AI entity performs operations such as training data set construction and model training, and forwards the trained artificial intelligence model to the AI entity through the access network device.
  • OAM is used to operate, manage and/or maintain core network equipment (network management of core network equipment), and/or is used to operate, manage and/or maintain access network equipment (network management of access network equipment) .
  • the present disclosure includes a first OAM and a second OAM, the first OAM is the network management of the core network equipment, and the second OAM is the network management of the access network equipment.
  • the first OAM and/or the second OAM may include an AI entity.
  • the present disclosure includes a third OAM, and the third OAM is the network manager of the core network device and the access network device at the same time.
  • an AI entity may be integrated in a terminal or a terminal chip.
  • the AI entity may also be called by other names, which are mainly used to implement AI functions (or called AI-related operations), and this disclosure does not limit its specific names.
  • the AI model is a specific method for realizing the AI function, and the AI model represents the mapping relationship between the input and output of the model.
  • AI models can be neural networks or other machine learning models.
  • the AI model may be referred to as a model for short.
  • AI-related operations may include at least one of the following: data collection, model training, model information release, model inference (or called model inference, prediction), or reasoning result release, etc.
  • the present disclosure may be applicable to a scenario where a network device uses training data to train an AI model.
  • the training data may be collected by the terminal device and reported to the network device, or may be collected by the network device itself, or part of the data may be collected by the terminal device and reported to the network device and part of the data may be collected by the network device itself.
  • the training data may be any data, such as downlink channel state information (channel state information, CSI) acquired by the terminal device, downlink received signal, etc., without limitation.
  • the access network device may determine the CSI of the wireless channel between them, so as to determine the resource and modulation and coding scheme for scheduling the downlink data channel of the terminal device according to the CSI ( modulation and coding scheme, MCS), precoding matrix and other configuration parameters.
  • MCS modulation and coding scheme
  • the terminal device may measure the downlink CSI and feed back the obtained CSI to the access network device.
  • the terminal device may perform operations such as compression and/or quantization on the CSI, and feed back the compressed and/or quantized CSI to the terminal device, which will cause loss of the accuracy of the fed back CSI.
  • an AI-based wireless network signal processing method can be used, that is, by training a pair of AI models, the terminal device uses the AI model A in the pair of AI models to compress the downlink CSI, and the AI model A The output compressed CSI is fed back to the access network device.
  • the access network device recovers the compressed CSI by using the AI model B in the pair of AI models to recover the downlink CSI.
  • the method may be applicable to a frequency division duplex (frequency division duplex, FDD) system or a time division duplex (time division duplex, TDD) system. This method is more commonly used in FDD systems.
  • the access network device can measure the uplink CSI and use the uplink CSI as the CSI of the downlink channel.
  • the method can be applied to an FDD system or a time division duplex TDD system. This method is more commonly used in TDD systems.
  • the training data required for training these wireless channel-related AI models can be reported by the terminal device to the access network device or measured by the access network device, and then obtained by the access network device or other network devices with training functions on the network side (such as core network equipment, OAM, server or AI node, etc.) are trained according to the training data.
  • the access network device may send the training data to the other device.
  • the training data may also be referred to as training samples.
  • the AI model may also be called a machine learning model.
  • Training data is one of the important contents of machine learning.
  • the training process of the model is essentially to learn features from the training data, so that the output of the AI model is as close as possible to the target output, such as the difference between the output of the AI model and the target output. Small. Wherein, the target output may also be called a label.
  • machine learning can include supervised learning, unsupervised learning, and reinforcement learning.
  • Supervised learning uses machine learning algorithms to learn the mapping relationship between training data and data labels based on training data and data labels, and uses AI models to express the learned mapping relationship.
  • the process of training the AI model is the process of learning this mapping relationship.
  • Unsupervised learning uses only the training data and uses the algorithm to discover the internal patterns of the samples on its own.
  • model training is usually performed by network-side devices (such as access network devices, core network devices, OAM, servers, or AI nodes, etc.), and optional
  • the network side device sends the trained model information to the access network device, and then the access network device sends the trained model information to the terminal device.
  • network-side devices such as access network devices, core network devices, OAM, servers, or AI nodes, etc.
  • the network side device sends the trained model information to the access network device, and then the access network device sends the trained model information to the terminal device.
  • the method described below takes the model training performed by the access network device as an example for description.
  • the access network device needs to collect a large amount of training data of one or more terminal devices. After the access network device collects the training data of the one or more terminal devices, it can construct a training data set according to the training data.
  • the access network device forms all training data of all terminal devices into a mixed data set, uses the mixed data set to train a general model, and then broadcasts the general model to all terminal devices, and all terminal devices Devices all use this common model.
  • the general model trained in this way leads to suboptimal performance under each channel distribution.
  • this disclosure proposes a method for collecting training data and constructing training data sets in wireless networks. Aiming at training data sets related to wireless channels, a scenario-based training data set is constructed, that is, the scene is similar, that is, the channel distribution Similar training data is divided into a data set, so as to divide and construct a scene-based training data set, and then associate a scene-based artificial intelligence model.
  • a data set may also be called a sample set
  • a training data set may also be called a training sample set.
  • Machine learning includes neural networks.
  • Neural network is a specific implementation of machine learning. According to the general approximation theorem, the neural network can theoretically approximate any continuous function, so that the neural network has the ability to learn any mapping. Therefore, neural networks can accurately and abstractly model complex high-dimensional problems.
  • the idea of a neural network is derived from the neuronal structure of brain tissue.
  • Each neuron performs a weighted sum operation on its input values, and passes the weighted sum result through an activation function to generate an output.
  • the bias of the weighted summation is b.
  • the form of the activation function can be varied.
  • the output of the neuron is: w i , xi , and b can be various possible values such as decimals, integers (including 0, positive integers or negative integers, etc.), or complex numbers.
  • the activation functions of different neurons in a neural network can be the same or different.
  • a neural network generally includes a multi-layer structure, and each layer may include one or more neurons. Increasing the depth and/or width of a neural network can improve the expressive power of the neural network, providing more powerful information extraction and abstract modeling capabilities for complex systems.
  • the depth of the neural network may refer to the number of layers included in the neural network, and the number of neurons included in each layer may be referred to as the width of the layer.
  • Figure 3 it is a schematic diagram of the layer relationship of the neural network.
  • a neural network includes an input layer and an output layer. The input layer of the neural network processes the input received by neurons, and then passes the result to the output layer, and the output layer obtains the output result of the neural network.
  • a neural network in another implementation, includes an input layer, a hidden layer, and an output layer.
  • the input layer of the neural network processes the input received by neurons, and then passes the result to the hidden layer in the middle, and the hidden layer then passes the calculation result to the output layer or the adjacent hidden layer, and finally the output layer obtains the result of the neural network. Output the result.
  • a neural network may include one or more sequentially connected hidden layers without limitation.
  • a loss function can be defined. The loss function describes the gap or difference between the output value of the neural network and the ideal target value, and the disclosure does not limit the specific form of the loss function.
  • the training process of the neural network is to adjust the parameters of the neural network, such as the number of layers of the neural network, the width, the weight of the neuron, and/or the parameters in the activation function of the neuron, etc., so that the value of the loss function is less than the threshold threshold value Or the process of satisfying the requirement of the goal, that is, the minimum difference between the output of the neural network and the ideal target value.
  • Figure 4(a) is an example diagram of the first application framework of AI in the communication system.
  • the data source is used to store training data and inference data.
  • the model training host (model training host) obtains the AI model by analyzing or training the training data provided by the data source, and deploys the AI model in the model inference host (model inference host).
  • the AI model represents the mapping relationship between the input and output of the model. Learning the AI model through the model training node is equivalent to using the training data to learn the mapping relationship between the input and output of the model.
  • the model inference node uses the AI model to perform inference based on the inference data provided by the data source, and obtains the inference result.
  • the model inference node inputs the inference data into the AI model, and obtains an output through the AI model, and the output is the inference result.
  • the inference result may indicate: configuration parameters used (executed) by the execution object, and/or operations performed by the execution object.
  • the reasoning result can be uniformly planned by the execution (actor) entity, and sent to one or more execution objects (for example, network elements) for execution.
  • the reasoning data is the data used for reasoning by the AI model, which is equivalent to the input data when the AI model makes predictions.
  • the application framework shown in Figure 4(a) can be deployed on the network element shown in Figure 1 or Figure 2, for example, the application framework shown in Figure 4(a) can be deployed on the access network device in Figure 1 Or the AI entity in Figure 2.
  • a model training node may analyze or train training data provided by a data source to obtain a model.
  • the model inference node can use the model and the inference data provided by the data source to perform inference and obtain the output of the model. That is, the input of the model includes inference data, and the output of the model is the inference result corresponding to the model.
  • the access network device can send the inference data and/or inference results corresponding to the model to the terminal device, and the terminal device can perform corresponding actions based on the inference data and/or inference results. operate.
  • the access network device includes a near real-time access network intelligent controller (RAN intelligent controller, RIC) module for model learning and/or reasoning.
  • RAN intelligent controller RIC
  • the near real-time RIC can obtain network-side and/or terminal-side information from at least one of the CU, DU, and RU, and the information can be used as training data or inference data.
  • the near real-time RIC may submit the reasoning result to at least one of the CU, the DU and the RU.
  • the inference results can be exchanged between the CU and the DU.
  • the inference results can be exchanged between the DU and the RU, for example, the near real-time RIC submits the inference results to the DU, and the DU submits the inference results to the RU.
  • the near real-time RIC can be used to train an AI model and use that AI model for inference.
  • a non-real-time RIC is included outside the access network (optionally, the non-real-time RIC can be located in the OAM or in the core network device) for model learning and reasoning.
  • the non-real-time RIC can obtain network-side and/or terminal-side information from at least one of CU, DU, and RU, which can be used as training data or inference data, and the inference results can be submitted to CU, DU, and RU at least one of the .
  • the inference results can be exchanged between the CU and the DU.
  • the reasoning result can be exchanged between the DU and the RU, for example, the non-real-time RIC submits the reasoning result to the DU, and then the DU submits it to the RU.
  • non-real-time RIC is used to train an AI model and use that model for inference.
  • the access network equipment includes a near-real-time RIC, and the access network equipment includes a non-real-time RIC (optionally, the non-real-time RIC can be located in the OAM or core network equipment).
  • the non-real-time RIC can be used for model learning and/or reasoning; and/or, like the first possible implementation above, the near real-time RIC can be used for model learning and/or reasoning; And/or, the near-real-time RIC can obtain AI model information from the non-real-time RIC, and obtain network-side and/or terminal-side information from at least one of CU, DU, and RU, and use the information and the AI model information to obtain inference results , optional, the near real-time RIC can submit the reasoning result to at least one of CU, DU and RU, optional, the reasoning result can be interactive between CU and DU, optional, the reasoning can be interactive between DU and RU As a result, for example, the near real-time RIC submits the reasoning results to the DU, and the DU submits it to the RU.
  • near real-time RIC is used to train model A and use model A for inference.
  • non-real-time RIC is used to train Model B and utilize Model B for inference.
  • the non-real-time RIC is used to train the model C, and the model C is submitted to the near-real-time RIC, and the near-real-time RIC uses the model C for inference.
  • FIG. 4(c) is an example diagram of a network architecture to which the method provided by the present disclosure can be applied. Compared with FIG. 4( b ), in FIG. 4( b ), the CU is separated into CU-CP and CU-UP.
  • FIG. 4( d ) is an example diagram of a network architecture to which the method provided by the present disclosure can be applied.
  • the access network device includes one or more AI entities, and the functions of the AI entities are similar to the above-mentioned near real-time RIC.
  • the OAM includes one or more AI entities, and the functions of the AI entities are similar to the non-real-time RIC described above.
  • the core network device includes one or more AI entities, and the functions of the AI entities are similar to the above-mentioned non-real-time RIC.
  • both the OAM and the core network equipment include AI entities, the models trained by their respective AI entities are different, and/or the models used for reasoning are different.
  • the different models include at least one of the following differences: structural parameters of the model (such as the number of layers and/or weights of the model), input parameters of the model, or output parameters of the model.
  • FIG. 4(e) is an example diagram of a network architecture to which the method provided by the present disclosure can be applied.
  • the access network devices in Fig. 4(e) are separated into CU and DU.
  • the CU may include an AI entity, and the function of the AI entity is similar to the above-mentioned near real-time RIC.
  • the DU may include an AI entity, and the function of the AI entity is similar to the above-mentioned near real-time RIC.
  • both the CU and the DU include AI entities, the models trained by their respective AI entities are different, and/or the models used for reasoning are different.
  • the CU may be further split into CU-CP and CU-UP.
  • one or more AI models may be deployed in the CU-CP.
  • one or more AI models can be deployed in CU-UP.
  • the OAM of the access network device and the OAM of the core network device may be deployed separately and independently.
  • a model can infer one parameter, or infer multiple parameters.
  • the learning process of different models can be deployed in different devices or nodes, and can also be deployed in the same device or node.
  • the inference process of different models can be deployed in different devices or nodes, or can be deployed in the same device or node.
  • the AI entity or the access network device, etc. may perform some or all of the steps in the present disclosure, and these steps or operations are only examples, and the present disclosure may also perform other operations or variations of various operations.
  • various steps may be performed in a different order than presented in the disclosure, and it may not be necessary to perform all operations in the disclosure.
  • FIG. 5 it is a schematic flow diagram of a communication method provided by the present disclosure.
  • the flow takes the interaction between a terminal device and an access network device as an example.
  • the method includes:
  • S501 The access network device acquires N pieces of training data, where N is an integer greater than 0.
  • the training data may include but not limited to compressed downlink CSI, uncompressed CSI, reference signal receiving power (reference signal receiving power, RSRP) and other data.
  • the training data may be any data related to wireless communication that can be obtained by the access network device or the terminal device.
  • the specific type of training data can be related to the functions that the artificial intelligence model needs to implement. For example, if the artificial intelligence model that needs to be trained is used to predict CSI, then the training data can be CSI; the artificial intelligence model that needs to be trained is For predicting the direction of data transmission (ie beam), the training data can be RSRP.
  • the present disclosure does not limit the specific type of training data.
  • the training data may be reported by the terminal device to the access network device, or may be obtained by measurement by the access network device, which will be described respectively below.
  • the training data is reported by the terminal device to the access network device.
  • the terminal device may send training data to the access network device, and the specific amount of training data sent by the terminal device is not limited.
  • the access network device can acquire N pieces of training data through one or more terminal devices.
  • the training data may be wireless channel data, for example, the training data may be compressed CSI, etc., without limitation.
  • the terminal device can measure the reference signal from the access network device, so as to obtain training data.
  • the terminal device may also obtain training data in other ways, which is not limited in the present disclosure.
  • the terminal device may also send label information corresponding to the training data to the access network device, and correspondingly, the access network device may also obtain N pieces of label information corresponding to the N pieces of training data.
  • the method can also be described as: the access network device can pass one or more terminal devices To acquire N pieces of training data, the access network device can also simultaneously acquire L pieces of tag information corresponding to the N pieces of training data. Wherein, L is smaller than N.
  • the terminal device when it acquires the training data, it may record the attribute or the value of the attribute corresponding to the training data.
  • the attribute or the value of the attribute corresponding to the training data can be collectively referred to as the tag information corresponding to the training data, that is, the tag information can be used to indicate the attribute of the training data corresponding to the tag information.
  • the terminal device may send the tag information corresponding to the training data to the access network device.
  • the tag information of different training data may be the same or different, without limitation. When the tag information of multiple training data is the same, one tag information may be recorded and/or reported for the multiple training data.
  • a cell of an access network device includes 6 synchronizing signal/physical broadcast channel blocks (synchronizing signal/physical broadcast channel block, SSB), the indexes are SSB1 to SSB6, each SSB Corresponding transmission beams have different coverage areas, as shown in FIG. 6 for details.
  • the terminal device obtains training data (for example, measuring downlink CSI through channel estimation), it can obtain the reference signal receiving power (reference signal receiving power, RSRP) of each SSB in multiple SSBs, and the terminal device can use the measured value
  • the multiple SSBs corresponding to the larger multiple RSRPs are used as the SSBs associated with the training data.
  • the SSB corresponding to the RSRP with the largest value may be used as the SSB associated with the training data
  • the SSB corresponding to the RSRP with the second largest value may be used as the SSB associated with the training data.
  • the label information can be used to indicate at least one of the following attributes:
  • the training data can be associated with one or more SSBs.
  • the SSB associated with the training data may include the SSB corresponding to the measured RSRP with the largest value, and/or the SSB corresponding to the measured RSRP with the second largest value, etc.; the SSB associated with the training data may refer to the SSB obtained by the terminal device When receiving the training data, the terminal device may establish an association relationship between the training data and the multiple SSBs corresponding to the multiple RSRPs with larger measured measurement values.
  • the training data may be associated with one or more RSRPs, and the RSRP may be the RSRP of the SSB associated with the training data. If there are multiple SSBs associated with the training data, the RSRPs are RSRPs corresponding to the multiple SSBs.
  • the type includes line of sight (line of sight, LOS) and non-line of sight (non line of sight, NLOS), indicating that the wireless channel is a LOS channel or an NLOS channel.
  • LOS and NLOS can be determined by traditional non-AI methods, or by using a trained AI model.
  • the type may further include uplink, downlink, and sidelink (sidelink), respectively indicating that the CSI acquired in the wireless channel is uplink CSI, downlink CSI, and sidelink CSI.
  • the timing advance (TA) associated with the training data may obtain the TA through a random access process, or determine the TA according to the TA information sent by the access network device. After the terminal device obtains the TA, the TA is valid within the preset TA valid time, or before the next time the TA is obtained, that is, the TA associated with the training data is the valid TA at the time of obtaining the training data, and the specific process will not be repeated.
  • the cell identity of the cell associated with the training data is used to indicate which cell the training data is obtained from.
  • Dimensional information of the training data includes one or more items of time domain dimension, frequency domain dimension, and antenna domain (air domain) dimension.
  • the dimension information may also include the granularity of each dimension, for example, the granularity of the time domain dimension includes orthogonal frequency division multiplexing (orthogonal frequency division multiplexing, OFDM) symbols or time slots, and the granularity of the frequency domain dimension includes subcarriers Or resource block (resource block, RB).
  • tag information may also include other attributes, which will not be described one by one here.
  • the training data is obtained through measurement by the access network device.
  • the access network device may indicate to the terminal device the time unit (such as radio frame, subframe, time slot, or symbol, etc.) for obtaining training data; or, the time unit for obtaining training data is the protocol Agreed, without limitation. If the access network device acquires the training data periodically, the access network device may also indicate the period for acquiring the training data and/or the offset (offset) of the time unit for acquiring the training data in the period; or, the period or at least one of the offsets may be agreed upon. For example, the period is indicated by the access network equipment, and the offset is agreed upon in the protocol. For another example, both the period and the offset are indicated by the access network device.
  • the time unit such as radio frame, subframe, time slot, or symbol, etc.
  • the terminal device may send the time unit or the attribute or attribute value acquired near the time unit to the access network device.
  • the access network device may use the attribute or the value of the attribute as the tag information corresponding to the training data.
  • the time unit indicated by the access network device to the terminal device is a time unit T
  • the terminal device can send at least one of the following information obtained in the time unit T to the access network device: one or more measured RSRP; one or more SSBs corresponding to one or more measured RSRPs; a cell identity of a cell to which the measured RSRPs belong; a type of a wireless channel corresponding to the measured RSRPs.
  • the access network device uses the above at least one item of information from the terminal device as the tag information of the training data acquired by the time unit T.
  • tag information please refer to the previous article, and will not repeat it here.
  • the SSB reported by the terminal device may include the SSB corresponding to the measured RSRP with the largest value, and/or the SSB corresponding to the measured RSRP with the second largest value, and so on.
  • the training data acquired by the access network device can be used to train the artificial intelligence model.
  • the access network device may train one or more artificial intelligence models according to N training data.
  • the function of the artificial intelligence model is not limited, for example, it may include but not limited to an artificial intelligence model for CSI compression feedback, an artificial intelligence model for constellation modulation and demodulation, an artificial intelligence model for channel encoding and decoding, or Artificial intelligence models for uplink precoding, etc.
  • the N training data can be divided to construct a scenario-based training data set, and the training data in each training data set includes the same features.
  • the same feature means that the above-mentioned attributes associated with the training data are the same, or the values of the attributes are in the same range.
  • Each training data set can be used to train an artificial intelligence model, thereby realizing scene-based artificial intelligence model training.
  • scenario-based artificial intelligence model training that is, training artificial intelligence models for different training data sets, rather than emphasizing that one training data set is used to train a fixed artificial intelligence model.
  • scenario-based artificial intelligence model training that is, training artificial intelligence models for different training data sets, rather than emphasizing that one training data set is used to train a fixed artificial intelligence model.
  • scenario-based artificial intelligence model training that is, training artificial intelligence models for different training data sets, rather than emphasizing that one training data set is used to train a fixed artificial intelligence model.
  • different machine learning algorithms or different model parameters such as the number of neural network layers, etc.
  • there may be performance differences among the multiple different artificial intelligence models there may be performance differences among the multiple different artificial intelligence models.
  • the same machine learning algorithm can be used for model learning, or different algorithms can be used for model learning, without limitation.
  • an artificial intelligence model may correspond to a training condition, and the training condition may be preset or agreed upon in an agreement, or may be determined by an access network device.
  • the training condition can be used to filter out training data for training the artificial intelligence model.
  • the training data used to train the artificial intelligence model is determined according to the label information, specifically, one or more data used to train the artificial intelligence model (in this disclosure, used to train the artificial intelligence).
  • One or more training data of the intelligent model can also be recorded as X training data, wherein X is a positive integer, or described as X is an integer greater than 0) training data, which is the corresponding label information in N training data ( or attributes) that satisfy the training data corresponding to the training conditions of the artificial intelligence model.
  • the training conditions can also be described as conditions.
  • this condition can be regarded as a training condition
  • this condition can be regarded as an application condition.
  • the label information includes SSB indexes
  • the training condition corresponding to the artificial intelligence model 1 includes at least one SSB index, for example, the included indexes are SSB1 and SSB2.
  • the tag information includes the training data whose SSB index is SSB1 and/or SSB2 can be divided into a training data set, and the artificial intelligence model 1 can be trained by using the above training data set.
  • the training condition is determined first, and then one or more training data satisfying the training condition are determined, and the model is trained using the one or more training data.
  • N training data can be clustered, and the training data with the same clustering features or clustering features in the same range can be divided into a training data set, where the clustering features can be variance, Features such as covariance matrix or multipath delay distribution.
  • the training data whose variance is in the same range is divided into a training data set; for another example, the training data whose covariance matrix is in the same range is divided into a training data set.
  • the attributes corresponding to the training data with the same clustering feature or in the same range can be determined, and the corresponding attributes can be used as training conditions, which are the same as those using the clustering feature Correspondence to the artificial intelligence model trained on the training data. For example, if the training data with the variance in the same range is divided into a training data set, and it is determined that the SSB indexes associated with the training data included in the training data set are all SSB1, then SSB1 can be used as an attribute included in the training condition.
  • one or more training data for training the artificial intelligence model is determined first, and then the training conditions of the artificial intelligence model are determined according to the one or more training data.
  • the specific content of the training conditions is not limited.
  • the training conditions may include at least one of the following:
  • the index of the synchronization signal block is one or more; the value range of the reference signal received power; the type of the wireless channel; the value range of the timing advance; the cell identifier, the identifier can be the identifier of a cell or multiple cells logo.
  • training conditions may also include other contents, which will not be described one by one here.
  • the access network device may also send the training conditions corresponding to the artificial intelligence model to the terminal device, and the specific process will not be described again.
  • the access network device sends indication information to the terminal device; correspondingly, the terminal device receives the indication information from the access network device.
  • the indication information is used to indicate information of M artificial intelligence models, and M is an integer greater than 0.
  • the M artificial intelligence models may be all the artificial intelligence models trained according to the N training data, or may be part of the artificial intelligence models, which are determined according to the actual situation.
  • the access network device may actively send instruction information to the terminal device.
  • the access network device may also send M training conditions corresponding to the M artificial intelligence models to the terminal device.
  • the M artificial intelligence models are in one-to-one correspondence with the M training conditions, that is, each artificial intelligence model corresponds to a training condition.
  • the terminal device after the terminal device obtains the information of the M artificial intelligence models, it can determine the artificial intelligence model to be used according to the actual situation. For example, the terminal device may obtain the attribute of the current wireless channel data and the training condition satisfied by the attribute when using the artificial intelligence model or before using the artificial intelligence model, and use the artificial intelligence model corresponding to the training condition as the artificial intelligence model to be used.
  • the attributes of the current wireless channel data acquired by the terminal device include: the associated SSB is SSB1. If the training condition of one artificial intelligence model among the M artificial intelligence models is: the training data associated with SSB1 is used for training, then the terminal device can use the artificial intelligence model.
  • the training condition can be regarded as an application condition on the terminal device side.
  • the access network device sends instruction information to the terminal device after receiving a request message from the terminal device after training the artificial intelligence model.
  • the number of artificial intelligence models indicated by the access network device to the terminal device may be less than or equal to the number of actually trained artificial intelligence models.
  • the terminal device may request a corresponding artificial intelligence model from the access network device according to the actual required artificial intelligence model.
  • the terminal device may send model indication information to the access network device, and the model indication information may be used to indicate the artificial intelligence model requested by the terminal device.
  • the access network device may determine the artificial intelligence model requested by the terminal device according to the model indication information, and indicate the information of the artificial intelligence model requested by the terminal device through the indication information.
  • the model indication information may be an index of an artificial intelligence model.
  • different indexes may be configured for the artificial intelligence models in different scenarios, or the indexes of the artificial intelligence models in different scenarios may be preset.
  • the access network device may indicate the index of the artificial intelligence model of each scene to the terminal device.
  • the terminal device needs to use the artificial intelligence model, it can send the index corresponding to the artificial intelligence model to the access network device, and then obtain the information of the artificial intelligence model from the access network device.
  • the model indication information may be the training conditions corresponding to the artificial intelligence model.
  • the access network device may send the training condition corresponding to the artificial intelligence model to the terminal device in advance, and if the training condition is preset, the access network device may not send the training condition.
  • the terminal device needs to use the artificial intelligence model, it can send the training conditions corresponding to the artificial intelligence model to the access network device, and then obtain the information of the artificial intelligence model from the access network device.
  • the model indication information may be label information corresponding to the training data for training the artificial intelligence model.
  • the terminal device can obtain relevant attributes of the current wireless channel through measurement, such as obtaining multiple RSRPs, multiple SSBs corresponding to multiple RSRPs, cell identities of the cells to which multiple RSRPs belong, and the type of wireless channel corresponding to RSRPs.
  • tag information One or more items of , as mentioned above, these information can be collectively referred to as tag information.
  • the terminal device may send the tag information including at least one of the above items to the access network device, and the access network device may indicate to the terminal device the artificial intelligence model corresponding to the training condition that the tag information satisfies.
  • model indication information may also be in other implementation forms, which is not limited in the present disclosure.
  • the terminal device After the terminal device obtains the artificial intelligence model trained by the access network device, it can use the corresponding artificial intelligence model according to actual needs.
  • the access network device when it obtains N training data, it can divide the N training data according to different training conditions, and construct training data sets corresponding to different training conditions.
  • the attributes included in the training condition can represent the scene corresponding to the training data, which is equivalent to dividing the N training data into a training data set of multiple scenes.
  • the training data for training an artificial intelligence model satisfies the training conditions corresponding to the artificial intelligence model, and scene-based artificial intelligence model training can be realized. Therefore, the obtained model is more in line with the channel condition, the correctness of the model is higher, and the communication efficiency is higher.
  • this method can also maintain low signaling overhead when obtaining a scene-based artificial intelligence model.
  • the access network device forms all training data of each terminal device into a terminal-device-specific (specific) data set, and uses each terminal-device-specific data set to train a terminal-device-specific model , and then configure the model corresponding to the terminal device for each terminal device, and each terminal device uses its dedicated model.
  • training a dedicated model for each terminal device leads to a large signaling overhead on the network.
  • the distribution of wireless channel samples collected by the terminal device at different times may also be quite different.
  • the access network device can transfer N training data Forward to the AI entity, and the AI entity will train the artificial intelligence model; the AI entity can indicate the information of the trained artificial intelligence model to the access network device, and then the access network device will indicate the information of the trained artificial intelligence model to the terminal device , and other processes are similar to the process shown in FIG. 5 , and will not be repeated here.
  • the access network device is taken as an example to describe.
  • the actual model training can be performed by modules (or units, network elements, etc.) in the access network device, such as near real-time RIC, access network
  • the AI entity in the device, the AI entity in the CU, or the AI entity in the DU is not limited.
  • the modules used to train different models can be the same or different without limitation.
  • the model information can be sent to the terminal device after being submitted by the model training module to other modules in the access network. For example, after the AI entity training in the CU obtains a model, the information of the model passes through the DU and RU in turn, and then is sent to the terminal device through the air interface. No more examples.
  • the access network device constructs training data sets corresponding to different training conditions when the tag information includes different attributes and under different training conditions.
  • An SSB is associated with a transmission beam of an access network device.
  • the coverage of each transmission beam is usually relatively concentrated, and will not change significantly for a long time, that is, the wireless channel environment and distribution within the coverage of each transmission beam are different. It is similar, so the training data associated with an SSB can be used as a training data set.
  • the terminal device when the terminal device obtains training data through measurement, it can measure and obtain the RSRP of at least one SSB, and the terminal device can consider that it is within the coverage of the SSB associated with the largest RSRP, so it can use the SSB associated with the largest RSRP
  • the index of the training data is marked, that is, the marking information of the training data may include the index of the SSB with the largest RSRP among the at least one SSB.
  • one cell of the access network device includes 6 SSBs, and the indexes are SSB1 to SSB6 respectively.
  • Each SSB corresponds to a different coverage of a transmission beam.
  • the terminal device measures the training data, if the measured index of the largest RSRP-associated SSB is SSB1, then the tag information corresponding to the training data may include SSB1.
  • the access network device or the AI entity when it is training the artificial intelligence model, it can determine one or more training data for training the artificial intelligence model according to the index of the SSB, that is, construct a training data set for training the artificial intelligence model. Specifically, when an access network device or an AI entity constructs a training data set, it may divide corresponding training data with the same index of the SSB into a training data set. At this time, the training condition of the artificial intelligence model may include the index of the SSB. For example, for an artificial intelligence model, its corresponding training condition includes the first index of the first SSB; correspondingly, among the N training data, the corresponding label information includes one or more of the first index of the first SSB Training data, used to train the artificial intelligence model.
  • the cell includes terminal equipment 1 and terminal equipment 2, the training data reported by terminal equipment 1 are training data 1 to training data 6, respectively, and the training data reported by terminal equipment 2 are training data 7 to training data data12.
  • the SSBs associated with training data 1 to training data 12 can refer to Table 1.
  • the access network device or AI entity trains three artificial intelligence models through the above training data, where the index included in the training condition corresponding to artificial intelligence model 1 is SSB1, and the index included in the training condition corresponding to artificial intelligence model 2 is SSB2.
  • the index included in the training condition corresponding to intelligent model 3 is SSB3.
  • one or more training data used to train each artificial intelligence model can be as shown in Table 2.
  • the access network device or AI entity can use the training data in training data set 1 to train artificial intelligence model 1, use the training data in training data set 2 to train artificial intelligence model 2, and use the training data in training data set 3 to train artificial intelligence model 2.
  • the access network device may indicate the trained artificial intelligence models and training conditions associated with each artificial intelligence model to the terminal device.
  • the terminal device determines which artificial intelligence model should be used according to the training conditions corresponding to the current environment. For example, the access network device indicates to the terminal device the information of the artificial intelligence model 1, the information of the artificial intelligence model 2, and the information of the artificial intelligence model 3, and indicates the training conditions corresponding to each artificial intelligence model.
  • Terminal devices can use artificial intelligence models and reasoning data for reasoning. If the terminal device measures each SSB when measuring the reasoning data, and the measured RSRP of SSB2 is the largest, then the terminal device may use the artificial intelligence model 2 .
  • the terminal device when it needs to use the artificial intelligence model, it requests the corresponding artificial intelligence model from the access network device.
  • the terminal device can request the artificial intelligence model used by sending the index of the artificial intelligence model; the terminal device can also send the tag information corresponding to the current wireless channel data to the access network device, and the access network device can indicate the tag information to the terminal device
  • the tag information sent by the terminal device includes the index of SSB3, and the access network device determines that the training condition corresponding to the artificial intelligence model 3 includes SSB3, thereby indicating the information of the artificial intelligence model 3 to the terminal device.
  • the training condition includes one SSB index as an example.
  • the training condition may include multiple SSB indexes.
  • the access network device may group SSBs, group one or more SSBs with adjacent beam directions or similar beam-pointing environments, and divide training data associated with the same group of SSBs into a training data set.
  • the access network device may group SSBs, group one or more SSBs with adjacent beam directions or similar beam-pointing environments, and divide training data associated with the same group of SSBs into a training data set.
  • the access network device may group SSBs, group one or more SSBs with adjacent beam directions or similar beam-pointing environments, and divide training data associated with the same group of SSBs into a training data set.
  • the beam directions corresponding to the SSBs with indexes SSB1, SSB2 and SSB3 are close to each other, group the SSBs with indexes SSB1, SSB2 and SSB3 into one group;
  • the N pieces of training data acquired by the access network device are shown in Table 3.
  • training data 10 SSB4 training data 11 SSB5 training data 12 SSB6
  • training artificial intelligence model 1 use training data associated with at least one of SSB1, SSB2 or SSB3 for training; when training artificial intelligence model 2, use at least one of SSB4, SSB5 or SSB6 Item-associated training data for training.
  • AI model 1 is to be used here; if the RSRP of SSB4 or SSB5 or SSB6 is measured to be the largest, it is determined that the current wireless channel data is associated with SSB4 or SSB5 or SSB6, so that it can be associated with SSB4 or SSB5 or SSB6
  • the artificial intelligence model trained by the training data is used as the artificial intelligence model to be used, and the artificial intelligence model 2 is to be used here.
  • these training data can be divided by the index of its associated SSB to construct a scene-based training data set, so that each training data set
  • the distribution of wireless channels is relatively concentrated or similar, enabling scene-based model training.
  • the training data may also be divided according to other attributes, so that the wireless channel environments and distributions corresponding to the divided training data are similar.
  • the first RSRP associated with the training data where the first RSRP is the maximum RSRP measured by the terminal device when acquiring the training data;
  • the second RSRP associated with the training data where the second RSRP is the second largest RSRP measured by the terminal device when acquiring the training data;
  • An index of a second SSB associated with the training data where the second SSB is the SSB corresponding to the second RSRP;
  • the type of the wireless channel associated with the training data the type including line-of-sight and non-line-of-sight;
  • the tag information includes more attributes.
  • the terminal device when the terminal device obtains the training data, it records the tag information of the training data, and when the terminal device reports the training data to the access network device, it also reports the tag information corresponding to the training data to the access network device at the same time. network access equipment.
  • one or more training data whose corresponding tag information meets the training conditions corresponding to the artificial intelligence model can be used to train the personal intelligent model, that is, to build and train the artificial intelligence model
  • the training data set, the training data set includes one or more training data.
  • the access network device or the AI entity trains the artificial intelligence model according to each constructed training data set, and one training data set can train one artificial intelligence model.
  • association relationship between the training data set, artificial intelligence model and training conditions is shown in Table 5.
  • the access network device When the access network device indicates the artificial intelligence model to the terminal device, it may also indicate to the terminal device the training conditions corresponding to the artificial intelligence model. When using the artificial intelligence model, the terminal device determines which artificial intelligence model should be used according to the training conditions corresponding to the artificial intelligence model.
  • the terminal device when the terminal device obtains the current wireless channel data, it measures the attributes of the current wireless channel data as follows: the largest RSRP is located in [a3, a4], and the RSRP corresponds to SSB1, and the type of the wireless channel is determined to be non-line-of-sight, and If it is determined that TA is located within [b3, b4], it is determined that the attributes of the current wireless channel data meet the training conditions corresponding to the artificial intelligence model 2, and then it can be determined to use the artificial intelligence model 2.
  • the terminal device determines that the current wireless channel data obtained by measurement does not meet the training conditions corresponding to any artificial intelligence model, the terminal device does not use any artificial intelligence model, or selects an artificial intelligence model arbitrarily, or selects a training condition from it.
  • the access network device or AI entity when the access network device or AI entity collects multiple training data, it can divide these training data through its associated SSB index, RSRP, TA, LOS/NLOS and other information to build a scenario-based Training data sets, so that the distribution of wireless channels in each training data set is concentrated or similar, to achieve scene-based artificial intelligence model training.
  • the artificial intelligence model is mainly trained on the training data collected in the same community.
  • it can also be considered to use training data from different communities to train the artificial intelligence model, that is, different
  • the training data of the cells can be divided into a training data set.
  • the tag information of the training data may further include: a cell identifier of a cell associated with the training data.
  • the training data of multiple cells can be divided into multiple different training data sets directly according to the training conditions of the artificial intelligence model, so the training data set corresponding to each artificial intelligence model Contains training data for one or more cells.
  • the training data for each cell may be divided into multiple training data sets, where one training data set includes one or more training data.
  • one training data set includes one or more training data.
  • the training data set it can be divided according to the label information, for example, the training data including the index of the same SSB in the label information can be divided into one training data set.
  • each subdistrict has divided the training data set
  • data analysis can be performed on each training data set of each subdistrict, and the main features of each training data set can be extracted, or the statistical characteristics of each training data set can be counted, For example, variance, covariance matrix, multipath delay distribution, etc., if the training data sets with similar characteristics in multiple cells are further fused into one training data set, the training data sets of multiple cells can be further fused.
  • the training conditions corresponding to the artificial intelligence model 1 to be trained include label information 1, label information 2, and label information 4, the training conditions corresponding to the artificial intelligence model 2 include label information 3 and label information 5, and the training conditions corresponding to the artificial intelligence model 3 Including label information 6, then the relationship between the fused training data set, training conditions and artificial intelligence model is shown in Table 7.
  • the training data set 7 used to train the artificial intelligence model 1 is obtained by merging the training data set 1 in the community 1, the training data set 2 and the training data set 4 in the community 2. In other cases, the And so on, no more details.
  • the access network device can combine the artificial intelligence model associated with the tag information of the cell and the associated artificial intelligence model One or more tag information of the cell is sent to the terminal device in the cell, and the terminal device locally saves the artificial intelligence model that may be used in the cell.
  • the access network device may indicate AI model 1 and AI model 2 to the terminal device in cell 1, but not AI model 3;
  • the terminal device indicates the training condition corresponding to the artificial intelligence model 1, it may only indicate the label information 1 and the label information 2, and not indicate the label information 4.
  • the access network device may also send the artificial intelligence model associated with the tag information of the cell and other cells, and one or more tag information of the cell and the other cells associated with the artificial intelligence model, to each terminal device in the cell.
  • each artificial intelligence model associated with the tag information of the current cell and several adjacent cells, and one or more tag information of the current cell and several adjacent cells associated with each artificial intelligence model are sent to the terminal device of the current cell.
  • the artificial intelligence model associated with the tag information of the access network notification area (RAN-based notification area, RNA) where the cell is located and one or more tag information of the RNA associated with the artificial intelligence model are sent to the cell's Terminal Equipment.
  • the terminal device can save the artificial intelligence models that may be used in the current cell and several adjacent cells, so as to quickly determine the applicable artificial intelligence model after cell handover.
  • the access network device can indicate artificial intelligence model 1, artificial intelligence model 2, and artificial intelligence model 3 to the terminal device in cell 1;
  • the access network device can indicate tag information 1, the tag information 2, and the tag information 4; similarly, when the access network device indicates the training conditions corresponding to the artificial intelligence model 3 to the terminal device in the cell 1 , can indicate tag information6.
  • the training data in multiple communities will be fused, so that the training data in multiple communities can jointly construct a training data set for training artificial intelligence models, so that multiple communities can share the same artificial intelligence model to improve the training efficiency of artificial intelligence models.
  • the training data set used for training the artificial intelligence model is determined depending on the tag information of the training data reported by the terminal device, that is, the training data is divided according to the preset training conditions.
  • the training data it is also possible to divide the training data first, and then determine the training conditions of the artificial intelligence model according to the divided training data, that is, analyze from the training data itself which attributes in the tag information are related to the distribution of the training data.
  • the training conditions corresponding to the artificial intelligence model are no longer preset or preconfigured, and the training conditions corresponding to the artificial intelligence model are determined according to one or more training data for training the artificial intelligence model.
  • the access network device or AI entity After the access network device or AI entity obtains N training data, it clusters the N training data, and divides the training data with the same clustering characteristics or within the same range into a training data set and a training data set.
  • the data set includes one or more training data, where the clustering feature can be the variance, covariance matrix, or multipath delay distribution of the training data. Since each training data is associated with a label information, the intersection or union of the attributes corresponding to the training data in each training data set can be determined according to the divided training data sets, so as to determine the training conditions corresponding to the artificial intelligence model.
  • the SSB associated with the training data in the training data set is relatively concentrated, one part is associated with SSB1, and the other part is associated with SSB2, but other information, such as RSRP, TA, LOS/NLOS and other information If there is no obvious rule, it can be considered that the distribution of the wireless channel is related to the SSB, and it can be determined that the training condition of the artificial intelligence model trained by the training data includes the index of the SSB, for example, SSB1 and SSB2.
  • the wireless channel types associated with the training data in a training data set are all line-of-sight, but other information, such as RSRP, TA, etc., have no obvious rules, it can be considered that the distribution of wireless channels is related to the line-of-sight. It may be determined that the training condition of the artificial intelligence model trained by the training data includes that the wireless channel type is line-of-sight.
  • the access network device or AI entity performs artificial intelligence model training according to the constructed training data sets.
  • Each training data set can be trained to obtain an artificial intelligence model, and each artificial intelligence model obtained through training is also associated with a training condition.
  • the access network device sends each artificial intelligence model and its associated training conditions to the terminal device. For details, please refer to the previous description, which will not be repeated here.
  • the accuracy of data division can be improved, and the training of artificial intelligence models can be improved. accuracy.
  • the access network device or the terminal device may include a hardware structure and/or a software module, and realize the above-mentioned in the form of a hardware structure, a software module, or a hardware structure plus a software module.
  • various functions Whether one of the above-mentioned functions is executed in the form of a hardware structure, a software module, or a hardware structure plus a software module depends on the specific application and design constraints of the technical solution.
  • each functional module in each embodiment of the present application may be integrated into one processor, or physically exist separately, or two or more modules may be integrated into one module.
  • the above-mentioned integrated modules can be implemented in the form of hardware or in the form of software function modules.
  • the embodiment of the present application further provides a communication device 700 for realizing the functions of the access network device or the terminal device in the above method.
  • the form of the communication device is not limited, and may be a hardware structure, a software module, or a hardware structure plus a software module.
  • the device may be a software module or a system on a chip.
  • the system-on-a-chip may be composed of chips, or may include chips and other discrete devices.
  • the apparatus 700 may include: a processing unit 701 and a communication unit 702 .
  • the communication unit may also be referred to as a transceiver unit, and may include a sending unit and/or a receiving unit, which are respectively used to perform the sending and receiving performed by the access network device, the AI entity or the terminal device in the method embodiments above. A step of.
  • a communication unit may also be referred to as a transceiver.
  • a processing unit may also be referred to as a processing module, or a processing device, or the like.
  • the device used to realize the receiving function in the communication unit 702 can be regarded as a receiving unit, and the device used to realize the sending function in the communication unit 702 can be regarded as a sending unit, that is, the communication unit 702 includes a receiving unit and a sending unit.
  • a communication unit may sometimes be implemented as a pin, a transceiver, a transceiver, or a transceiver circuit, and the like.
  • the processing unit may sometimes be implemented as a processor, or a processing board, or the like.
  • the receiving unit can sometimes be realized as a pin, a receiver, a receiver or a receiving circuit, etc.
  • the sending unit can sometimes be realized as a pin, a transmitter, a transmitter, or a transmitting circuit, and the like.
  • the processing unit is used to obtain N training data through the communication unit.
  • the training data corresponds to a tag information
  • the tag information is used to indicate the attribute of the training data, and N is greater than 0 integer;
  • the processing unit is configured to send instruction information to the terminal device through the communication unit, the instruction information is used to indicate information of M artificial intelligence models, and for one artificial intelligence model in the M artificial intelligence models, the artificial intelligence model is trained according to N
  • the X training data in the data are trained, and the X training data are determined according to the label information, and M and X are integers greater than 0.
  • the artificial intelligence model corresponds to a training condition, and the label information corresponding to the X training data satisfies the artificial intelligence The training conditions corresponding to the model.
  • the tag information is used to indicate at least one of the following attributes:
  • the reference signal received power of the synchronization signal block associated with the training data
  • the type of wireless channel associated with the training data the type including line-of-sight and non-line-of-sight;
  • a cell identifier of a cell associated with the training data is a cell identifier of a cell associated with the training data.
  • the training condition includes at least one of the following: an index of a synchronization signal block; a value range of received power of a reference signal; a type of wireless channel; a value range of timing advance;
  • the training conditions corresponding to the artificial intelligence model are preset.
  • the training conditions corresponding to the artificial intelligence model are determined according to the X training data used to train the artificial intelligence model, and the X training data used to train the artificial intelligence model
  • the training data is the training data having the same clustering feature in the N training data, or the X training data used to train the artificial intelligence model is that the clustering features in the N training data are in the same range in the training data.
  • the communication unit is further configured to: send M training conditions corresponding to the M artificial intelligence models to the terminal device, where the M artificial intelligence models and the M training conditions One to one correspondence.
  • the communication unit is further configured to: receive model indication information from the terminal device, where the model indication information is used to indicate the artificial intelligence model requested by the terminal device;
  • the indication information indicates the information of the artificial intelligence model requested by the terminal device.
  • a communication unit configured to receive indication information, the indication information is used to indicate information of M artificial intelligence models, and for one artificial intelligence model in the M artificial intelligence models, the artificial intelligence model is based on N training data In the X training data training, M, N and X are integers greater than 0.
  • the training data corresponds to a piece of tag information
  • the tag information is used to indicate an attribute of the training data
  • the artificial intelligence model corresponds to a training condition, and the label information corresponding to the X training data satisfies the requirement of the artificial intelligence model corresponding training conditions.
  • the tag information is used to indicate at least one of the following attributes:
  • the reference signal received power of the synchronization signal block associated with the training data
  • the type of wireless channel associated with the training data the type including line-of-sight and non-line-of-sight;
  • a cell identifier of a cell associated with the training data is a cell identifier of a cell associated with the training data.
  • the training condition includes at least one of the following: an index of a synchronization signal block; a value range of a reference signal received power; a type of a wireless channel; a value range of a timing advance;
  • the training conditions corresponding to the artificial intelligence model are preset.
  • the training condition corresponding to the artificial intelligence model is determined according to X training data used for training the artificial intelligence model, and the X training data used for training the artificial intelligence model It is the training data with the same clustering feature in the N training data, or the X training data used to train the artificial intelligence model is the clustering feature in the N training data within the same range training data.
  • the communication unit is further configured to: receive M training conditions corresponding to the M artificial intelligence models, where the M artificial intelligence models correspond to the M training conditions one-to-one.
  • the communication unit is further configured to: send model indication information, where the model indication information is used to indicate the artificial intelligence model requested by the terminal device;
  • the indication information indicates the information of the artificial intelligence model requested by the terminal device.
  • processing unit 701 and the communication unit 702 may also perform other functions.
  • processing unit 701 and the communication unit 702 may also perform other functions.
  • FIG. 8 shows an apparatus 800 provided in the embodiment of the present application.
  • the apparatus shown in FIG. 8 may be a hardware circuit implementation manner of the apparatus shown in FIG. 7 .
  • the communication device may be applicable to the flow chart shown above, and execute the functions of the terminal device or the access network device in the above method embodiments. For ease of illustration, FIG. 8 only shows the main components of the communication device.
  • the communication device 800 includes a processor 810 and an interface circuit 820 .
  • the processor 810 and the interface circuit 820 are coupled to each other.
  • the interface circuit 820 may be a transceiver, a pin, an interface circuit or an input-output interface.
  • the communication device 800 may further include a memory 830 for storing instructions executed by the processor 810 or storing input data required by the processor 810 to execute the instructions or storing data generated after the processor 810 executes the instructions.
  • part or all of the memory 830 may reside in the processor 810 .
  • the processor 810 is used to implement the functions of the above-mentioned processing unit 701
  • the interface circuit 820 is used to implement the functions of the above-mentioned communication unit 702 .
  • the terminal device chip implements the functions of the terminal device in the above method embodiment.
  • the terminal device chip receives information from other modules in the terminal device (such as radio frequency modules or antennas), and the information is sent to the terminal device by the access network device; or, the terminal device chip sends information to other modules in the terminal device (such as radio frequency modules) module or antenna) to send information, which is sent by the terminal device to the access network device.
  • the access network equipment chip implements the functions of the access network equipment in the above method embodiments.
  • the access network device chip receives information from other modules (such as radio frequency modules or antennas) in the access network device, and the information is sent by the terminal device to the access network device; or, the access network device chip sends information to the access network device Other modules in the device (such as radio frequency modules or antennas) send information, which is sent by the access network device to the terminal device.
  • the method in Figure 5 takes the interaction between the access network device and the terminal device as an example.
  • the access network device can The N training data are forwarded to the AI entity, and the AI entity performs artificial intelligence model training; the AI entity can indicate the information of the trained artificial intelligence model to the access network device, and then the access network device can indicate the trained artificial intelligence model to the terminal device.
  • Information about the smart model, and other processes are similar to those in Figure 5. Therefore, the above introduction about access network equipment can be applied to AI entities.
  • processor in the present disclosure may be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof .
  • a general-purpose processor can be a microprocessor, or any conventional processor.
  • the memory in the present disclosure may be random access memory, flash memory, read-only memory, programmable read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, register, hard disk, mobile hard disk or Any other form of storage medium known in the art.
  • the present disclosure may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, optical storage, etc.) having computer-usable program code embodied therein.
  • These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions
  • the device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

Landscapes

  • Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Quality & Reliability (AREA)
  • Electromagnetism (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本公开提供一种通信方法及装置，其中方法包括：获取N个训练数据，对于该N个训练数据中的一个训练数据，该训练数据对应一个标记信息，该标记信息用于指示该训练数据的属性；向终端设备发送指示信息，指示信息用于指示M个人工智能模型的信息，人工智能模型是根据所述N个训练数据中的X个训练数据训练的该，X个训练数据是根据所述标记信息确定的。通过上面的方法，在获取到N个训练数据时，可以将N个训练数据构建不同标记信息对应的训练数据，从而训练M个人工智能模型或更多个人工智能模型。当对人工智能模型进行训练时，训练一个人工智能模型的训练数据满足该人工智能模型对应的标记信息，可以实现场景化的人工智能模型训练。

Description

一种通信方法及装置
相关申请的交叉引用
本申请要求在2021年10月21日提交中国专利局、申请号为202111227090.4、申请名称为“一种通信方法及装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本公开涉及通信技术领域，尤其涉及一种通信方法及装置。
背景技术
在无线通信网络中，例如在移动通信网络中，网络支持的业务越来越多样，因此需要满足的需求越来越多样。例如，网络需要能够支持超高速率、超低时延、和/或超大连接。该特点使得网络规划、网络配置、和/或资源调度越来越复杂。此外，由于网络的功能越来越强大，例如支持的频谱越来越高、支持高阶多入多出(multiple input multiple output，MIMO)技术、支持波束赋形、和/或支持波束管理等新技术，使得网络节能成为了热门研究课题。这些新需求、新场景和新特性给网络规划、运维和高效运营带来了前所未有的挑战。为了迎接该挑战，可以将人工智能技术引入无线通信网络中，从而实现网络智能化。基于此，如何在网络中有效地实现人工智能是一个值得研究的问题。
发明内容
本公开提供一种通信方法及装置，用以实现如何构建训练数据集，从而训练获得较优的神经网络。
第一方面，本公开提供一种通信方法，该方法可在网络设备侧，例如接入网设备侧或独立于接入网设备的AI实体，执行。该方法可通过软件、硬件、或软硬件结合的方式执行。例如，该方法由接入网设备执行，或由包括接入网设备的较大设备执行，或由电路系统执行，该电路系统能够实现接入网设备的功能，或者由独立于接入网设备的AI实体执行，不予限制。示例性地，所述接入网设备为基站。该方法包括：获取N个训练数据，对于N个训练数据中的一个训练数据，该训练数据对应一个标记信息，该标记信息用于指示该训练数据的属性，N为大于0的整数；向终端设备发送指示信息，指示信息用于指示M个人工智能模型的信息，对于M个人工智能模型中的一个人工智能模型，该人工智能模型是根据N个训练数据中的X个训练数据训练的，该X个训练数据是根据所述标记信息确定的，M和X为大于0的整数。
通过上面的方法，在获取到N个训练数据时，可以将N个训练数据构建不同标记信息对应的训练数据，从而训练M个或更多个人工智能模型。当对人工智能模型进行训练时，训练一个人工智能模型的训练数据满足该人工智能模型对应的标记信息，可以实现场景化的人工智能模型训练。场景化的人工智能模型在相应的场景中的推理性能更加准确。
一种可能的实现方式中，对于M个人工智能模型中的一个人工智能模型，该人工智能 模型对应一个训练条件，用于训练该人工智能模型的X个训练数据对应的标记信息满足该人工智能模型对应的训练条件，或者所述X个训练数据对应的属性满足所述人工智能模型对应的训练条件。
通过将N个训练数据中标记信息满足训练条件的训练数据对人工智能模型进行训练，实现构建不同训练条件对应的训练数据集，使得训练每个人工智能模型的训练数据对应的无线信道分布比较集中或者相似。当对人工智能模型进行训练时，训练一个人工智能模型的训练数据满足该人工智能模型对应的训练条件，可以实现场景化的人工智能模型训练。
一种可能的实现方式中，标记信息用于指示以下至少一项属性：训练数据关联的同步信号块的索引；训练数据关联的同步信号块的参考信号接收功率；训练数据关联的无线信道的类型，类型包括视距和非视距；训练数据关联的定时提前；训练数据关联的小区的小区标识。
上述属性与无线信道相关，通过上述属性，实现将不同训练数据与无线信道绑定，使得在训练人工智能模型时，可以根据上述属性自适应的选择适当的场景化模型。
一种可能的实现方式中，训练条件包括以下至少一项：同步信号块的索引；参考信号接收功率的取值范围；无线信道的类型；定时提前的取值范围；小区标识。
同步信号块的索引、参考信号接收功率的取值范围、无线信道的类型以及定时提前的取值范围中的一项或多项与无线信道相关，训练条件包括上述信息时，可以实现通过训练条件筛选出所需要的无线信道相关的训练数据，从而构建场景化的训练数据集，使能场景化的模型训练。
另外，当训练条件包括小区标识时，每个人工智能模型可以通过多个小区的训练数据进行训练，使能多小区环境的场景化模型，从而使多个小区可以共享相同的人工智能模型。
一种可能的实现方式中，人工智能模型对应的训练条件为预设的。
一种可能的实现方式中，人工智能模型对应的训练条件是根据用于训练人工智能模型的X个训练数据确定的，用于训练人工智能模型的X个训练数据为N个训练数据中具有相同聚类特征的训练数据，或者，用于训练人工智能模型的X个训练数据为N个训练数据中聚类特征在同一范围内的训练数据。
一种可能的实现方式中，该方法还包括：向终端设备发送M个人工智能模型对应的M个训练条件，其中，M个人工智能模型和M个训练条件一一对应。
一种可能的实现方式中，向终端设备发送指示信息，包括：接收来自终端设备的模型指示信息，模型指示信息用于指示终端设备请求的人工智能模型；向终端设备发送指示信息，指示信息指示终端设备请求的人工智能模型的信息。可选的，模型指示信息用于指示：所请求的人工智能模型的索引，所请求的人工智能模型的训练条件，所请求的人工智能模型的训练数据的属性或标记信息。
通过上面的方法，终端设备请求所需的人工智能模型，可以实现人工智能模型的精确指示，避免向终端设备发送其不需要的人工智能模型，降低信令开销。
第二方面，提供一种通信方法，该方法可在终端设备侧执行。该方法可通过软件、硬件、或软硬件结合的方式执行。例如，该方法由终端设备执行，或由电路系统执行，或者由包括终端设备的较大设备执行，该电路系统能够实现终端设备的功能。该方法包括：
接收指示信息，指示信息用于指示M个人工智能模型的信息，对于M个人工智能模型中的一个人工智能模型，人工智能模型是根据N个训练数据中的X个训练数据训练的， M、N和X为大于0的整数。
一种可能的实现方式中，对于N个训练数据中的一个训练数据，训练数据对应一个标记信息，标记信息用于指示训练数据的属性。
一种可能的实现方式中对于M个人工智能模型中的一个人工智能模型，人工智能模型对应一个训练条件，X个训练数据对应的标记信息满足人工智能模型对应的训练条件。
关于标记信息、训练条件等的介绍请参考第一方面，此处不再赘述。
一种可能的实现方式中，方法还包括：接收M个人工智能模型对应的M个训练条件，其中，M个人工智能模型和M个训练条件一一对应。
一种可能的实现方式中，接收指示信息，包括：发送模型指示信息，模型指示信息用于指示所请求的人工智能模型；
接收指示信息，指示信息指示所请求的人工智能模型的信息。
关于指示信息的介绍请参考第一方面，此处不再赘述。
第三方面，本申请还提供一种通信装置，该通信装置具有实现上述第一方面提供的任一方法或任一实现方式。该通信装置可以通过硬件实现，可以通过软件实现，或者可以通过硬件执行相应的软件实现。该硬件或软件包括一个或多个与上述功能相对应的单元或模块。
在一种可能的实现方式中，该通信装置包括：处理器，该处理器被配置为支持该通信装置执行以上第一方面所示方法。该通信装置还可以包括存储器，该存储可以与处理器耦合，其保存该通信装置必要的程序指令和数据。可选地，该通信装置还包括接口电路，该接口电路用于支持该通信装置与其他通信装置进行通信。
在一种可能的实施方式中，通信装置的结构中包括处理单元和通信单元，这些单元可以执行上述方法示例中相应功能，具体参见第一方面中的描述，此处不做赘述。
第四方面，本申请还提供一种通信装置，该通信装置具有实现上述第二方面中提供的任一方法或任一实现方式。该通信装置可以通过硬件实现，可以通过软件实现，或者可以通过硬件执行相应的软件实现。该硬件或软件包括一个或多个与上述功能相对应的单元或模块。
在一种可能的实现方式中，该通信装置包括：处理器，该处理器被配置为支持该通信装置执行以上第二方面所示方法。该通信装置还可以包括存储器，该存储可以与处理器耦合，其保存该通信装置必要的程序指令和数据。可选地，该通信装置还包括接口电路，该接口电路用于支持该通信装置与其他通信装置进行通信。
在一种可能的实施方式中，通信装置的结构中包括处理单元和通信单元，这些单元可以执行上述方法示例中相应功能，具体参见第二方面提供的方法中的描述，此处不做赘述。
第五方面，提供了一种通信装置，包括处理器和接口电路，接口电路用于接收来自该通信装置之外的其它通信装置的信号并传输至该处理器或将来自该处理器的信号发送给该通信装置之外的其它通信装置，该处理器用于执行所述存储器中存储的计算机程序或指令，实现前述第一方面中任意可能的实现方式中的方法。可选地，该装置还包括存储器，所述存储器中存储计算机程序或指令。
第六方面，提供了一种通信装置，包括处理器和接口电路，接口电路用于接收来自该通信装置之外的其它通信装置的信号并传输至该处理器或将来自该处理器的信号发送给该通信装置之外的其它通信装置，该处理器用于执行所述存储器中存储的计算机程序或指 令，实现前述第二方面中任意可能的实现方式中的方法。可选地，该装置还包括存储器，所述存储器中存储计算机程序或指令。
第七方面，提供了一种计算机可读存储介质，该计算机可读存储介质中存储有计算机程序或指令，当所述计算机程序或指令在计算机上运行时，使得所述计算机实现前述第一方面中任意可能的实现方式中的方法。
第八方面，提供了一种计算机可读存储介质，该计算机可读存储介质中存储有计算机程序或指令，当所述计算机程序或指令在计算机上运行时，使得所述计算机实现前述第二方面中任意可能的实现方式中的方法。
第九方面，提供了一种包括计算机可读指令的计算机程序产品，当所述计算机可读指令在计算机上运行时，使得所述计算机实现前述第一方面中任意可能的实现方式中的方法。
第十方面，提供了一种包括计算机可读指令的计算机程序产品，当所述计算机可读指令在计算机上运行时，使得所述计算机实现前述第二方面中任意可能的实现方式中的方法。
第十一方面，提供一种芯片，该芯片包括处理器，还可以包括存储器，所述处理器与存储器耦合，用于执行所述存储器中存储的计算机程序或指令，使得芯片实现前述第一方面中任意可能的实现方式中的方法。
第十二方面，提供一种芯片，该芯片包括处理器，还可以包括存储器，所述处理器与存储器耦合，用于执行所述存储器中存储的计算机程序或指令，使得芯片实现前述第二方面中任意可能的实现方式中的方法。
第十三方面，提供一种通信系统，所述系统包括实现第一方面所述的装置(如接入网设备)以及实现第二方面所述的装置(如终端设备)。
附图说明
图1为本公开可以应用的一种通信系统的架构示意图；
图2为本公开可以应用的另一种通信系统的架构示意图；
图3为本公开提供的一种神经网络的层关系示意图；
图4(a)至图4(e)为本公开提供的方法能够应用的一种网络架构的示例图；
图5为本公开提供的一种配置方法流程示意图；
图6为本公开提供的一种同步信号传输示意图；
图7为本公开提供的一种通信装置结构示意图；
图8为本公开提供的一种通信装置结构示意图。
具体实施方式
下面结合说明书附图对本公开做详细描述。
本公开的技术方案可以应用于各种通信系统，例如：长期演进(long term evolution，LTE)系统、第五代(5th generation，5G)移动通信系统、或下一代移动通信系统等，在此不做限制。其中，5G系统还可以称为新无线(new radio，NR)系统。
本公开中，以终端设备与接入网设备之间的交互为例进行描述，需要说明的是，本公开提供的方法，不仅可以应用于终端设备与网络侧之间的交互，还可以应用于任意两个设备之间的交互中，本公开对此并不限定。
为便于理解本公开，首先以图1中示出的通信系统为例详细说明适用于本公开的通信系统。图1是本公开可以应用的通信系统的架构示意图，该通信系统中包括接入网设备和终端设备。终端设备可以与接入网设备建立连接，并和接入网设备进行通信。图1只是示意图，本公开对该通信系统中包括的接入网设备和终端设备的数量不做限定。
本公开中，终端设备可以简称为终端。终端设备可以经无线接入网(radio access network，RAN)与一个或多个核心网进行通信。
本公开中，终端设备，可以为具有无线收发功能的设备或可设置于该设备中的芯片。终端设备也可以称为用户设备(user equipment，UE)、接入终端、用户单元、用户站、移动站、移动台、远方站、远程终端、移动设备、用户终端、无线通信设备、用户代理或用户装置。本公开中的终端设备可以是手机(mobile phone)、平板电脑(Pad)、带无线收发功能的电脑、虚拟现实(virtual reality，VR)终端、增强现实(augmented reality，AR)终端、可穿戴设备、车辆、无人机、直升机、飞机、轮船、机器人、机械臂、或智能家居设等等。本公开中的终端设备可以广泛应用于各种场景中的通信，例如包括但不限于以下至少一个场景：增强移动宽带(enhanced mobile broadband，eMBB)、超可靠低时延通信(ultra-reliable low-latency communication，URLLC)、设备到设备(device-to-device，D2D)、车物(vehicle to everything，V2X)、机器类通信(machine-type communication，MTC)、大规模机器类通信(massive machine-type communication，mMTC)、物联网(internet of things，IOT)、虚拟现实、增强现实、工业控制、自动驾驶、远程医疗、智能电网、智能家具、智能办公、智能穿戴、智能交通、或智慧城市等。本公开对终端所采用的具体技术和具体设备形态不做限定。
在本公开中，用于实现终端的功能的装置可以是终端；也可以是能够支持终端实现该功能的装置，例如芯片系统、硬件电路、软件模块、或硬件电路加软件模块，该装置可以被安装在终端中或可以与终端匹配使用。为了便于描述，下文以用于实现终端的功能的装置是终端为例，描述本公开提供的技术方案。
接入网设备可以是基站(base station)、节点B(NodeB)、演进型节点B(evolved NodeB，eNodeB)、发送接收点(transmission reception point，TRP)、第五代(5th generation，5G)移动通信系统中的下一代节点B(next generation NodeB，gNB)、开放无线接入网(open radio access network，O-RAN或open RAN)中的接入网设备、第六代(6th generation，6G)移动通信系统中的下一代基站、未来移动通信系统中的基站或无线保真(wireless fidelity，WiFi)系统中的接入节点等；或者可以是完成基站部分功能的模块或单元，例如，可以是集中式单元(central unit，CU)、分布式单元(distributed unit，DU)、集中单元控制面(CU control plane，CU-CP)模块、或集中单元用户面(CU user plane，CU-UP)模块。接入网设备可以是宏基站，也可以是微基站或室内站，还可以是中继节点或施主节点等。本公开中对接入网设备所采用的具体技术和具体设备形态不做限定。
(1)协议层结构。
接入网设备和终端之间的通信遵循一定的协议层结构。该协议层结构可以包括控制面协议层结构和用户面协议层结构。例如，控制面协议层结构可以包括无线资源控制(radio resource control，RRC)层、分组数据汇聚层协议(packet data convergence protocol，PDCP)层、无线链路控制(radio link control，RLC)层、媒体接入控制(media access control，MAC)层和物理层等协议层的功能。例如，用户面协议层结构可以包括PDCP层、RLC 层、MAC层和物理层等协议层的功能，在一种可能的实现中，PDCP层之上还可以包括业务数据适配协议(service data adaptation protocol，SDAP)层。
可选的，接入网设备和终端之间的协议层结构还可以包括人工智能(artificial intelligence，AI)层，用于传输AI功能相关的数据。
(2)集中单元(central unit，CU)和分布单元(distributed unit，DU)。
接入网设备可以包括CU和DU。多个DU可以由一个CU集中控制。作为示例，CU和DU之间的接口可以称为F1接口。其中，控制面(control panel，CP)接口可以为F1-C，用户面(user panel，UP)接口可以为F1-U。本公开不限制各接口的具体名称。CU和DU可以根据无线网络的协议层划分：比如，PDCP层及以上协议层的功能设置在CU，PDCP层以下协议层(例如RLC层和MAC层等)的功能设置在DU；又比如，PDCP层以上协议层的功能设置在CU，PDCP层及以下协议层的功能设置在DU，不予限制。
上述对CU和DU的处理功能按照协议层的划分仅仅是一种举例，也可以按照其他的方式进行划分。例如可以将CU或者DU划分为具有更多协议层的功能，又例如将CU或DU还可以划分为具有协议层的部分处理功能。在一种设计中，将RLC层的部分功能和RLC层以上的协议层的功能设置在CU，将RLC层的剩余功能和RLC层以下的协议层的功能设置在DU。在另一种设计中，还可以按照业务类型或者其他系统需求对CU或者DU的功能进行划分，例如按时延划分，将处理时间需要满足时延要求的功能设置在DU，不需要满足该时延要求的功能设置在CU。在另一种设计中，CU也可以具有核心网的一个或多个功能。示例性的，CU可以设置在网络侧方便集中管理。在另一种设计中，将DU的无线单元(radio unit，RU)拉远设置。可选的，RU可以具有射频功能。
可选的，DU和RU可以在物理层(physical layer，PHY)进行划分。例如，DU可以实现PHY层中的高层功能，RU可以实现PHY层中的低层功能。其中，用于发送时，PHY层的功能可以包括以下至少一项：添加循环冗余校验(cyclic redundancy check，CRC)位、信道编码、速率匹配、加扰、调制、层映射、预编码、资源映射、物理天线映射、或射频发送功能。用于接收时，PHY层的功能可以包括以下至少一项：CRC校验、信道解码、解速率匹配、解扰、解调、解层映射、信道检测、资源解映射、物理天线解映射、或射频接收功能。其中，PHY层中的高层功能可以包括PHY层的一部分功能，例如该部分功能更加靠近MAC层，PHY层中的低层功能可以包括PHY层的另一部分功能，例如该部分功能更加靠近射频功能。例如，PHY层中的高层功能可以包括添加CRC码、信道编码、速率匹配、加扰、调制、和层映射，PHY层中的低层功能可以包括预编码、资源映射、物理天线映射、和射频发送功能；或者，PHY层中的高层功能可以包括添加CRC码、信道编码、速率匹配、加扰、调制、层映射和预编码，PHY层中的低层功能可以包括资源映射、物理天线映射、和射频发送功能。例如，PHY层中的高层功能可以包括CRC校验、信道解码、解速率匹配、解码、解调、和解层映射，PHY层中的低层功能可以包括信道检测、资源解映射、物理天线解映射、和射频接收功能；或者，PHY层中的高层功能可以包括CRC校验、信道解码、解速率匹配、解码、解调、解层映射、和信道检测，PHY层中的低层功能可以包括资源解映射、物理天线解映射、和射频接收功能。
示例性的，CU的功能可以由一个实体来实现，或者也可以由不同的实体来实现。例如，可以对CU的功能进行进一步划分，即将控制面和用户面分离并通过不同实体来实现，分别为控制面CU实体(即CU-CP实体)和用户面CU实体(即CU-UP实体)。该CU-CP 实体和CU-UP实体可以与DU相耦合，共同完成接入网设备的功能。
可选的，上述DU、CU、CU-CP、CU-UP和RU中的任一个可以是软件模块、硬件结构、或者软件模块+硬件结构，不予限制。其中，不同实体的存在形式可以是不同的，不予限制。例如DU、CU、CU-CP、CU-UP是软件模块，RU是硬件结构。这些模块及其执行的方法也在本公开的保护范围内。
一种可能的实现中，接入网设备包括CU-CP、CU-UP、DU和RU。例如，本公开的执行主体包括DU，或者包括DU和RU，或者包括CU-CP、DU和RU，或者包括CU-UP、DU和RU，不予限制。各模块所执行的方法也在本公开的保护范围内。
本公开中，用于实现接入网设备的功能的装置可以是接入网设备；也可以是能够支持接入网设备实现该功能的装置，例如芯片系统。该装置可以被安装在接入网设备中或者和接入网设备匹配使用。
为了在无线网络中支持AI，网络中还可能引入专门的AI实体(或称为AI模块)。AI实体可以对应一个独立的网元；或者可以位于某个网元内部，该网元可以是核心网设备、接入网设备、或网管(operation,administration and maintenance，OAM)等。例如，如图2所示，AI实体位于接入网设备的外部，可以和接入网设备进行通信。接入网设备可以将终端设备上报的与AI模型相关的数据转发给AI实体，由AI实体执行训练数据集构建与模型训练等操作，并将训练好的人工智能模型通过接入网设备转发给各终端设备。本公开中，OAM用于操作、管理和/或维护核心网设备(核心网设备的网管)，和/或，用于操作、管理和/或维护接入网设备(接入网设备的网管)。例如，本公开中包括第一OAM和第二OAM，第一OAM是核心网设备的网管，第二OAM是接入网设备的网管。第一OAM和/或第二OAM中可以包括AI实体。再例如，本公开中包括第三OAM，第三OAM同时是核心网设备和接入网设备的网管。
可选的，为了匹配支持AI，终端或终端芯片中可以集成AI实体。
可选的，本公开中，AI实体还可以称为其他名称，主要用于实现AI功能(或称为AI相关的操作)，本公开不限制其具体名称。
本公开中，AI模型是实现AI功能的具体方法，AI模型表征了模型的输入和输出之间的映射关系。AI模型可以是神经网络或者其他机器学习模型。其中，AI模型可以简称为模型。AI相关的操作可以包括以下至少一项：数据收集、模型训练、模型信息发布、模型推断(或称为模型推理、预测)、或推理结果发布等。
本公开可以适用于网络设备采用训练数据训练AI模型的场景。训练数据可以是终端设备收集并上报给网络设备的，可以是网络设备自己采集的，或者可以是部分数据由终端设备收集并上报给网络设备且部分数据由网络设备自己采集。训练数据可以是任何数据，比如终端设备获取到的下行信道状态信息(channel state information，CSI)，下行接收信号等，不予限制。举例来说，接入网设备与终端设备进行通信之前，接入网设备可以确定他们二者之间的无线信道的CSI，以便根据CSI确定调度终端设备的下行数据信道的资源、调制编码方案(modulation and coding scheme，MCS)、预编码矩阵等配置参数。例如，对于下行通信，接入网设备获取到的下行CSI越精确，所调度的下行传输就越能匹配当前的下行信道，下行传输的性能就越好。
一种可能的实现中，为了获取到相对准确的下行CSI，可以由终端设备测量得到下行CSI，并将所得到的CSI反馈给接入网设备。而终端设备为了降低CSI反馈的开销，可以 对CSI进行压缩、和/或量化等操作，将压缩和/或量化后的CSI反馈给终端设备，这就会对所反馈的CSI的精度造成损失。为了提高下行CSI反馈的精度，可以采用基于AI的无线网络信号处理方法，即通过训练一对AI模型，终端设备利用该一对AI模型中的AI模型A进行下行CSI的压缩，将AI模型A输出的压缩CSI反馈给接入网设备，相应的，接入网设备利用该一对AI模型中的AI模型B对压缩CSI进行恢复，恢复出下行CSI。可选的，该方法可以适用于频分双工(frequency division duplex，FDD)系统或时分双工(time division duplex，TDD)系统。该方法较常用于FDD系统。
一种可能的实现中，利用上下行互易性，可以由接入网设备测量上行CSI，并将上行CSI用作下行信道的CSI。可选的，该方法可以适用FDD系统或时分双工TDD系统。该方法较常用于TDD系统。
训练这些无线信道相关的AI模型所需要的训练数据可以由终端设备上报至接入网设备或由接入网设备测量得到，再由接入网设备或者网络侧的具有训练功能的其他网络设备(如核心网设备、OAM、服务器或AI节点等)根据训练数据进行训练。其中，当训练是由该其他网络设备实现时，接入网设备可以将训练数据发送至该其他设备。可选的，训练数据还可以称为训练样本。可选的，AI模型还可以称为机器学习模型。
训练数据是机器学习的重要内容之一，模型的训练过程本质上就是从训练数据中学习特征，使得AI模型的输出尽可能接近目标输出，如AI模型的输出与目标输出之间的差异尽可能小。其中，目标输出也可以被称为标签。例如，机器学习可以包括监督学习、非监督学习、和强化学习。监督学习根据训练数据和数据标签，利用机器学习算法学习训练数据到数据标签的映射关系，并用AI模型来表达学到的映射关系。训练AI模型的过程就是学习这种映射关系的过程。无监督学习仅依据训练数据，利用算法自行发掘样本的内在模式。例如，无监督学习中有一类算法将训练数据自身作为监督信号，即学习从训练数据到训练数据的映射关系，称为自监督学习。训练时，通过计算模型的预测值与样本本身之间的误差来优化模型参数。强化学习是一种通过与环境进行交互来学习解决问题的策略的算法。与监督、无监督学习不同，强化学习问题并没有明确的“正确的”动作标签数据，算法需要与环境进行交互，获取环境反馈的奖励信号，进而调整决策动作以获得更大的奖励信号数值。
在无线网络中，由于模型训练对算力的要求较高，通常都是由网络侧设备(如接入网设备、核心网设备、OAM、服务器或AI节点等)进行模型训练，并可选的由该网络侧设备将训练得到的模型的信息发送给接入网设备，再由接入网设备将训练好的模型信息下发给终端设备。为了描述简洁，如无特别说明，下文描述的方法以接入网设备执行模型训练为例进行描述。
为了提高模型的准确性，需要接入网设备收集一个或多个终端设备的大量的训练数据。接入网设备收集到该一个或多个终端设备的训练数据后，可以根据这些训练数据构建训练数据集。
一种可能的实现方式中，接入网设备将所有终端设备的所有训练数据组成为一个混合数据集，用该混合数据集训练一个通用模型，然后将该通用模型广播给所有终端设备，所有终端设备都使用该通用模型。通过这种方式训练得到的通用模型为了兼顾各种信道分布，导致在每个信道分布下性能都不是最优的。
为了解决上面的问题，本公开提出一种无线网络中训练数据收集与训练数据集构建的 方法，针对无线信道有关的训练数据集，构建场景化的训练数据集，即将场景相似，也就是信道分布相近的训练数据划分为一个数据集，从而划分和构建场景化的训练数据集，进而关联场景化的人工智能模型。可选的，本公开中，数据集还可以称为样本集，训练数据集还可以称为训练样本集。
在描述本公开的方法之前，先简单介绍一些关于人工智能的相关知识。人工智能，可以让机器具有人类的智能，例如可以让机器应用计算机的软硬件来模拟人类某些智能行为。为了实现人工智能，可以采取机器学习方法或很多其他方法。例如，机器学习包括神经网络。神经网络(neural network,NN)是机器学习的一种具体实现方式。根据通用近似定理，神经网络理论上可以逼近任意连续函数，从而使得神经网络具备学习任意映射的能力。因此神经网络可以对复杂的高维度问题进行准确地抽象建模。神经网络的思想来源于大脑组织的神经元结构。每个神经元都对其输入值做加权求和运算，将加权求和结果通过一个激活函数产生输出。假设神经元的输入为x＝[x 0,…,x n],与输入对应的权值为w＝[w 0,…,w n],加权求和的偏置为b。激活函数的形式可以多样化。假设一个神经元的激活函数为：y＝f(z)＝max(0,z)，则该神经元的输出为：
Figure PCTCN2022126467-appb-000001
再例如一个神经元的激活函数为：y＝f(z)＝z，则该神经元的输出为：
Figure PCTCN2022126467-appb-000002
w i、x i、和b可以为小数、整数(包括0、正整数或负整数等)、或复数等各种可能的取值。神经网络中不同神经元的激活函数可以相同或不同。
神经网络一般包括多层结构，每层可包括一个或多个神经元。增加神经网络的深度和/或宽度可以提高该神经网络的表达能力，为复杂系统提供更强大的信息提取和抽象建模能力。其中，神经网络的深度可以指神经网络包括的层数，每层包括的神经元个数可以称为该层的宽度。如图3所示，为神经网络的层关系示意图。一种实现中，神经网络包括输入层和输出层。神经网络的输入层将接收到的输入经过神经元处理后，将结果传递给输出层，由输出层得到神经网络的输出结果。另一种实现中，神经网络包括输入层、隐藏层和输出层。神经网络的输入层将接收到的输入经过神经元处理后，将结果传递给中间的隐藏层，隐藏层再将计算结果传递给输出层或者相邻的隐藏层，最后由输出层得到神经网络的输出结果。一个神经网络可以包括一层或多层依次连接的隐藏层，不予限制。神经网络的训练过程中，可以定义损失函数。损失函数描述了神经网络的输出值和理想目标值之间的差距或差异，本公开不限制损失函数的具体形式。神经网络的训练过程就是通过调整神经网络参数，如神经网络的层数、宽度、神经元的权值、和/或神经元的激活函数中的参数等，使得损失函数的值小于阈值门限值或者满足目标需求的过程，即使得神经网络的输出与理想目标值之间的差异最小。
如图4(a)所示为AI在通信系统中的第一种应用框架的示例图。在图4(a)中，数据源(data source)用于存储训练数据和推理数据。模型训练节点(model trainning host)通过对数据源提供的训练数据(training data)进行分析或训练，得到AI模型，且将AI模型部署在模型推理节点(model inference host)中。其中，AI模型表征了模型的输入和输出之间的映射关系。通过模型训练节点学习得到AI模型，相当于由模型训练节点利用训练数据学习得到模型的输入和输出之间的映射关系。模型推理节点使用AI模型，基于数据源提供的推理数据进行推理，得到推理结果。该方法还可以描述为：模型推理节点将推理数据输入到AI模型，通过AI模型得到输出，该输出即为推理结果。该推理结果可以指 示：由执行对象使用(执行)的配置参数、和/或由执行对象执行的操作。推理结果可以由执行(actor)实体统一规划，并发送给一个或多个执行对象(例如，网元)去执行。其中，推理数据是用于AI模型进行推理的数据，相当于AI模型进行预测时的输入数据。
在本公开中，图4(a)所示的应用框架可以部署在图1或者图2中所示的网元，例如，图4(a)的应用框架可以部署在图1的接入网设备或者图2的AI实体。例如在接入网设备中，模型训练节点可对数据源提供的训练数据(training data)进行分析或训练，得到一个模型。模型推理节点可以使用该模型和数据源提供的推理数据进行推理，得到模型的输出。即该模型的输入包括推理数据，该模型的输出即为该模型所对应的推理结果。将终端设备视为图4(a)中的执行对象，接入网设备可以将模型对应的推理数据和/或推理结果发送给终端设备，终端设备可以根据推理数据和/或推理结果进行相应的操作。
下面结合图4(b)～图4(e)对本公开提供的通信方案能够应用的网络架构进行介绍。
如图4(b)所示，第一种可能的实现中，接入网设备中包括近实时接入网智能控制(RAN intelligent controller，RIC)模块，用于进行模型学习和/或推理。例如，近实时RIC可以从CU、DU和RU中的至少一个获得网络侧和/或终端侧的信息，该信息可以作为训练数据或者推理数据。可选的，近实时RIC可以将推理结果递交至CU、DU和RU中的至少一个。可选的，CU和DU之间可以交互推理结果。可选的，DU和RU之间可以交互推理结果，例如近实时RIC将推理结果递交至DU，由DU递交给RU。例如，近实时RIC可以用于训练AI模型，利用该AI模型进行推理。
如图4(b)所示，第二种可能的实现中，接入网之外包括非实时RIC(可选的，非实时RIC可以位于OAM中或者核心网设备中)，用于进行模型学习和推理。例如，非实时RIC可以从CU、DU和RU中的至少一个获得网络侧和/或终端侧的信息，该信息可以作为训练数据或者推理数据，该推理结果可以被递交至CU、DU和RU中的至少一个。可选的，CU和DU之间可以交互推理结果。可选的，DU和RU之间可以交互推理结果，例如非实时RIC将推理结果递交至DU，由DU递交给RU。例如，非实时RIC用于训练AI模型，利用该模型进行推理。
如图4(b)所示，第三种可能的实现中，接入网设备中包括近实时RIC，接入网设备之外包括非实时RIC(可选的，非实时RIC可以位于OAM中或者核心网设备中)。同上述第二种可能的实现，非实时RIC可以用于进行模型学习和/或推理；和/或，同上述第一种可能的实现，近实时RIC可以用于进行模型学习和/或推理；和/或，近实时RIC可以从非实时RIC获得AI模型信息，并从CU、DU和RU中的至少一个获得网络侧和/或终端侧的信息，利用该信息和该AI模型信息得到推理结果，可选的，近实时RIC可以将推理结果递交至CU、DU和RU中的至少一个，可选的，CU和DU之间可以交互推理结果，可选的，DU和RU之间可以交互推理结果，例如近实时RIC将推理结果递交至DU，由DU递交给RU。例如，近实时RIC用于训练模型A，利用模型A进行推理。例如，非实时RIC用于训练模型B，利用模型B进行推理。例如，非实时RIC用于训练模型C，将模型C递交给近实时RIC，近实时RIC利用模型C进行推理。
图4(c)所示为本公开提供的方法能够应用的一种网络架构的示例图。相对图4(b)，图4(b)中将CU分离为了CU-CP和CU-UP。
图4(d)所示为本公开提供的方法能够应用的一种网络架构的示例图。如图4(d)所示，可选的，接入网设备中包括一个或多个AI实体，该AI实体的功能类似上述近实时 RIC。可选的，OAM中包括一个或多个AI实体，该AI实体的功能类似上述非实时RIC。可选的，核心网设备中包括一个或多个AI实体，该AI实体的功能类似上述非实时RIC。当OAM和核心网设备中都包括AI实体时，他们各自的AI实体所训练得到的模型不同，和/或用于进行推理的模型不同。
本公开中，模型不同包括以下至少一项不同：模型的结构参数(例如模型的层数、和/或权值等)、模型的输入参数、或模型的输出参数。
图4(e)所示为本公开提供的方法能够应用的一种网络架构的示例图。相对图4(d)，图4(e)中的接入网设备分离为CU和DU。可选的，CU中可以包括AI实体，该AI实体的功能类似上述近实时RIC。可选的，DU中可以包括AI实体，该AI实体的功能类似上述近实时RIC。当CU和DU中都包括AI实体时，他们各自的AI实体所训练得到的模型不同，和/或，用于进行推理的模型不同。可选的，还可以进一步将CU拆分为CU-CP和CU-UP。可选的，CU-CP中可以部署有一个或多个AI模型。可选的，CU-UP中可以部署有一个或多个AI模型。可选的，图4(d)或图4(e)中，接入网设备的OAM和核心网设备的OAM可以分开独立部署。
本公开中，一个模型可以推理得到一个参数，或者推理得到多个参数。不同模型的学习过程可以部署在不同的设备或节点中，也可以部署在相同的设备或节点中。不同模型的推理过程可以部署在不同的设备或节点中，也可以部署在相同的设备或节点中。
可以理解的是，本公开中，AI实体或接入网设备等可以执行本公开中的部分或全部步骤，这些步骤或操作仅是示例，本公开还可以执行其它操作或者各种操作的变形。此外，各个步骤可以按照本公开呈现的不同的顺序来执行，并且有可能并非要执行本公开中的全部操作。
在本公开的各个实施例中，如果没有特殊说明以及逻辑冲突，不同的示例之间的术语和/或描述具有一致性、且可以相互引用，不同的示例中的技术特征根据其内在的逻辑关系可以组合形成新的示例。
可以理解的是，在本公开的示例中涉及的各种编号仅为描述方便进行的区分，并不用来限制本公开的示例的范围。上述各过程的序号的大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定。
本公开描述的网络架构以及业务场景是为了更加清楚的说明本公开的技术方案，并不构成对于本公开提供的技术方案的限定，本领域普通技术人员可知，随着网络架构的演变和新业务场景的出现，本公开提供的技术方案对于类似的技术问题，同样适用。
参见图5，为本公开提供的一种通信方法流程示意图，该流程以终端设备与接入网设备之间交互为例，该方法包括：
S501：接入网设备获取N个训练数据，N为大于0的整数。
本公开中，训练数据可以包括但不限于压缩后的下行CSI，未压缩的CSI，参考信号接收功率(reference signal receiving power，RSRP)等数据。训练数据可以是接入网设备或终端设备能够获得到的任何一种和无线通信相关的数据。一种实现方式中，训练数据的具体类型可以和人工智能模型所需实现的功能有关，例如需要训练的人工智能模型为用于预测CSI，那么训练数据可以为CSI；需要训练的人工智能模型为用于预测数据传输方向(即波束)，那么训练数据可以为RSRP。本公开对训练数据的具体类型并不限定。
本公开中，训练数据可以是由终端设备上报给接入网设备，也可以由接入网设备测量获得的，下面分别进行描述。
第一种可能的实现方式，训练数据是由终端设备上报给接入网设备的。
在该实现方式中，终端设备可以向接入网设备发送训练数据，终端设备发送的训练数据的具体数量并不限定。相应的，接入网设备可以通过一个或多个终端设备获取N个训练数据。该训练数据可以为无线信道数据，例如训练数据可以为压缩后的CSI等，不予限制。终端设备可以对来自接入网设备的参考信号进行测量，从而获得训练数据。终端设备也可以通过其他方式获得训练数据，本公开对此并不限定。
在该实现方式中，终端设备还可以向接入网设备发送训练数据对应的标记信息，相应的，接入网设备还可以获取到N个训练数据对应的N个标记信息。可选地，如果该N个训练数据中存在一组或多组训练数据，每组训练数据各自对应相同的标记信息，该方法还可以描述为：接入网设备可以通过一个或多个终端设备获取N个训练数据，接入网设备还可以同时获取到N个训练数据对应的L个标记信息。其中，L小于N。
本公开中，终端设备在获取训练数据时，可以记录与该训练数据对应的属性或属性的取值。为了描述方便，可以将与该训练数据对应的属性或属性的取值统称为该训练数据对应的标记信息，也就是说，标记信息可以用于指示与该标记信息对应的训练数据的属性。终端设备可以将训练数据对应的标记信息发送至接入网设备。可选的，不同的训练数据的标记信息可以相同或不同，不予限制。当多个训练数据的标记信息相同时，可以针对该多个训练数据记录和/或上报一个标记信息。
举例来说，如图6所示，接入网设备的一个小区内包括6个同步信号/物理广播信道块(synchronizing signal/physical broadcast channel block，SSB)，索引分别为SSB1至SSB6，每个SSB对应的发送波束的覆盖范围不同，具体可以参考图6所示。终端设备在获取训练数据(例如通过信道估计测量下行CSI)时，可以获取到多个SSB中每个SSB的参考信号接收功率(reference signal receiving power，RSRP)，终端设备可以将测量到的测量值较大的多个RSRP对应的多个SSB，作为训练数据关联的SSB。例如可以将取值最大的RSRP对应的SSB作为训练数据关联的SSB，将取值第二大的RSRP对应的SSB作为训练数据关联的SSB等。
结合前面的描述，一种可能的实现方式中，对于一个训练数据或者一组具有相同标记信息的多个训练数据，标记信息可以用于指示以下至少一项属性：
·训练数据关联的SSB的索引。其中，训练数据可以关联一个或多个SSB。例如，训练数据关联的SSB可以包括测量到的取值最大的RSRP对应的SSB，和/或测量到的取值第二大的RSRP对应的SSB等；训练数据关联的SSB可以是指终端设备获取到该训练数据时，测量到的测量值较大的多个RSRP对应的多个SSB，终端设备可以建立训练数据和该多个SSB的关联关系。
·训练数据关联的SSB的RSRP。其中，训练数据可以关联一个或多个RSRP，该RSRP可以为训练数据关联的SSB的RSRP。如果训练数据关联的SSB为多个，则该RSRP为该多个SSB分别对应的RSRP。
·训练数据关联的无线信道的类型。一种实现方式，所述类型包括视距(line of sight，LOS)和非视距(non line of sight，NLOS)，表示该无线信道为LOS信道或NLOS信道。LOS和NLOS可以通过传统非AI的方式确定，也可以使用已经训练好的 AI模型进行确定。另一种实现方式，所述类型还可以包括上行、下行和侧行(sidelink)，分别表示在该无线信道中获取到的CSI为上行CSI、下行CSI以及侧行CSI。
·训练数据关联的定时提前(time advance，TA)。其中，终端设备可以通过随机接入过程获得TA，或者，根据接入网设备发送的TA信息确定TA。终端设备获取TA后，该TA在预设的TA有效时间内，或者下次获取TA之前有效，即训练数据关联的TA为获取该训练数据时有效的TA，具体过程不再赘述。
·训练数据关联的小区的小区标识。换句话说，该小区标识用于指示该训练数据是在哪个小区获取的。
·训练数据的维度信息。例如，包括时域维度、频域维度、天线域(空域)维度中的一项或多项。可选地，该维度信息还可以包括每个维度的粒度，例如时域维度的粒度包括正交频分复用(orthogonal frequency division multiplexing，OFDM)符号或时隙，频域维度的粒度包括子载波或资源块(resource block，RB)。
以上只是示例，标记信息还可以包括其他属性，在此不再逐一举例说明。
第二种可能的实现方式，训练数据是由接入网设备测量获得的。
在该实现方式中，接入网设备可以将其获取训练数据的时间单元(例如无线帧、子帧、时隙、或符号等)指示给终端设备；或者，该获取训练数据的时间单元是协议约定的，不予限制。如果接入网设备是周期性的获取训练数据，接入网设备还可以指示其获取训练数据的周期和/或其获取训练数据的时间单元在周期中的偏置(offset)；或者，该周期或偏置中的至少一个可以是协议约定的。例如，周期是由接入网设备指示的，偏置是协议约定的。再例如，周期和偏置都是由接入网设备指示的。
终端设备接收到接入网设备指示的时间单元，则可以将该时间单元或者该时间单元附近获取到的属性或属性的取值发送给接入网设备。接入网设备可以将该属性或属性的取值作为该训练数据对应的标记信息。
举例来说，接入网设备向终端设备指示的时间单元为时间单元T，终端设备可以将在时间单元T获取到的以下至少一项信息发送至接入网设备：测量到的一个或多个RSRP；测量到的一个或多个RSRP对应的一个或多个SSB；测量到的RSRP所属的小区的小区标识；测量到的RSRP所对应的无线信道的类型。接入网设备将来自终端设备的以上至少一项信息作为时间单元T获取到的训练数据的标记信息。关于标记信息的更加详细的介绍请参加前文，此处不再赘述。例如，终端设备上报的SSB可以包括测量到的取值最大的RSRP对应的SSB，和/或测量到的取值第二大的RSRP对应的SSB等。
本公开中，接入网设备获取到的训练数据可以用于训练人工智能模型。具体的，接入网设备可以根据N个训练数据训练一个或多个人工智能模型。其中，人工智能模型的功能并不限定，例如可以包括但不限于用于CSI压缩反馈的人工智能模型，用于星座调制与解调的人工智能模型，用于信道编解码的人工智能模型，或用于上行预编码的人工智能模型等。本公开中，可以对N个训练数据进行划分，构建场景化的训练数据集，每个训练数据集内的训练数据包括相同的特征。相同的特征指的是训练数据关联的上述属性相同，或者属性的取值在同一范围内，每个训练数据集可以用于训练一个人工智能模型，从而实现场景化的人工智能模型训练。本公开中，强调的是场景化的人工智能模型训练，即针对不同的训练数据集分别训练人工智能模型，而不是强调一个训练数据集用于训练一个固定的人 工智能模型。例如，利用一个训练数据集，采用不用的机器学习算法或者不同的模型参数(例如神经网络层数等)，可以学习出多个不同的人工智能模型。可选的，这多个不同的人工智能模型之间可能存在性能差异。针对不同的训练数据集，可以采用相同的机器学习算法进行模型学习，也可以采用不同的算法进行模型学习，不予限制。
举例来说，一种可能的实现方式中，一个人工智能模型可以对应一个训练条件，该训练条件可以为预设的或协议约定的，也可以为接入网设备确定的。该训练条件可以用于筛选出训练该人工智能模型的训练数据。在该实现方式中，用于训练该人工智能模型的训练数据是根据所述标记信息确定的，具体的，用于训练该人工智能模型的一个或多个(本公开中，用于训练该人工智能模型的一个或多个训练数据，还可以记为X个训练数据，其中，X为正整数，或者描述为X为大于0的整数)训练数据，为N个训练数据中对应的标记信息(或属性)满足该人工智能模型对应的训练条件的训练数据。可选的，训练条件还可以描述为条件。例如，对于接入网设备或者其他进行模型训练的网元，该条件可以看做训练条件，对于终端设备或者其他应用该模型进行推理的网元，该条件可以看做应用条件。
例如，标记信息包括SSB的索引，人工智能模型1对应的训练条件包括至少一个SSB的索引，例如包括的索引为SSB1和SSB2。此时可以将N个训练数据中，标记信息包括SSB的索引为SSB1和/或SSB2的训练数据，划分为一个训练数据集，可以采用上述训练数据集对人工智能模型1进行训练。
通过上面的描述可知，该实现方式中，先确定训练条件，再确定满足该训练条件的一个或多个训练数据，并利用该一个或多个训练数据训练模型。
另一种可能的实现方式中，可以将N个训练数据进行聚类，将聚类特征相同或者聚类特征在同一范围内的训练数据划分为一个训练数据集，其中聚类特征可以为方差、协方差矩阵或多径时延分布等特征。例如，将方差在同一范围内的训练数据划分为一个训练数据集；再例如，将协方差矩阵在同一范围内的训练数据划分为一个训练数据集。
由于训练数据关联第一标记信息，可以确定聚类特征相同或者在同一范围内的训练数据共同对应的属性，并将所述共同对应的属性作为训练条件，该训练条件与采用该聚类特征的训练数据训练得到的人工智能模型对应。例如，将方差在同一范围内的训练数据划分为一个训练数据集，确定该训练数据集包括的训练数据关联的SSB的索引均为SSB1，则可以将SSB1作为训练条件包括的属性。
通过上面的描述可知，该实现方式中，先确定训练人工智能模型的一个或多个训练数据，再根据该一个或多个训练数据确定该人工智能模型的训练条件。
本公开中，对训练条件的具体内容并不限定，举例来说，训练条件可以包括以下至少一项：
同步信号块的索引，该索引为一个或多个；参考信号接收功率的取值范围；无线信道的类型；定时提前的取值范围；小区标识，该标识可以为一个小区的标识或者多个小区的标识。
以上只是示例，训练条件还可以包括其他内容，在此不再逐一举例说明。
本公开中，接入网设备还可以向终端设备发送人工智能模型对应的训练条件，具体过程不再赘述。
S502：接入网设备向终端设备发送指示信息；相应的，终端设备接收来自接入网设备的指示信息。
其中，指示信息用于指示M个人工智能模型的信息，M为大于0的整数。M个人工智能模型可以是根据N个训练数据训练的全部人工智能模型，也可以是其中的一部分人工智能模型，具体根据实际情况确定。
本公开中，一种实现方式中，接入网设备训练完成人工智能模型之后，接入网设备可以主动向终端设备发送指示信息。接入网设备还可以向终端设备发送M个人工智能模型对应的M个训练条件。该M个人工智能模型和该M个训练条件一一对应，即每个人工智能模型对应一个训练条件。
在该方式中，终端设备获取到M个人工智能模型的信息后，可以根据实际情况确定所使用的人工智能模型。例如，终端设备可以在使用人工智能模型时或者使用人工智能模型之前，获取当前无线信道数据的属性和该属性满足的训练条件，将该训练条件对应的人工智能模型作为待使用的人工智能模型。例如，终端设备获取的当前无线信道数据的属性包括：关联的SSB为SSB1。如果M个人工智能模型中，有一个人工智能模型的训练条件为：采用关联SSB1的训练数据进行训练，那么终端设备可以使用该人工智能模型。该训练条件在终端设备侧可以看做应用条件。
另一种实现方式中，接入网设备训练完成人工智能模型之后，在接收到来自终端设备的请求消息后，再向终端设备发送指示信息。在该实现方式中，接入网设备向终端设备指示的人工智能模型的数量，可以小于或等于实际训练的人工智能模型的数量。例如，接入网设备通过指示信息指示1个人工智能模型的信息，即M＝1。M的具体取值，可以根据终端设备实际请求的数量确定。
在该方式中，终端设备可以根据实际需求的人工智能模型，向接入网设备请求相应的人工智能模型。例如，终端设备可以向接入网设备发送模型指示信息，该模型指示信息可以用于指示终端设备请求的人工智能模型。接入网设备可以根据该模型指示信息确定终端设备请求的人工智能模型，并通过指示信息指示终端设备请求的该人工智能模型的信息。
举例来说，模型指示信息可以为人工智能模型的索引。接入网设备在训练人工智能模型时，可以为不同场景的人工智能模型配置不同的索引，或者不同场景的人工智能模型的索引为预设的。接入网设备可以向终端设备指示每个场景的人工智能模型的索引。终端设备需要使用人工智能模型时，可以向接入网设备发送该人工智能模型对应的索引，就可以从接入网设备获得该人工智能模型的信息。
举例来说，模型指示信息可以为人工智能模型对应的训练条件。接入网设备可以预先将人工智能模型对应的训练条件发送至终端设备，如果训练条件为预设的，接入网设备也可以不发送训练条件。终端设备需要使用人工智能模型时，可以向接入网设备发送该人工智能模型对应的训练条件，就可以从接入网设备获得该人工智能模型的信息。
举例来说，模型指示信息可以为训练该人工智能模型的训练数据对应的标记信息。终端设备需要使用人工智能模型时，可能并不清楚具体需要哪一个人工智能模型。此时终端设备可以通过测量，获得当前无线信道的相关属性，例如获得多个RSRP、多个RSRP对应的多个SSB、多个RSRP所属的小区的小区标识以及RSRP所对应的无线信道的类型中的一项或多项，如前所述，这些信息可以统称为标记信息。终端设备可以向接入网设备发送包括上述至少一项的标记信息，接入网设备可以将该标记信息满足的训练条件对应的人工智能模型，指示给终端设备。
以上只是示例，模型指示信息还可能为其他实现形式，本公开对此并不限定。
终端设备获取到接入网设备训练的人工智能模型之后，可以根据实际需求使用相应的人工智能模型。
通过上面的过程，接入网设备在获取到N个训练数据时，可以将N个训练数据按照不同训练条件进行划分，构建不同训练条件对应的训练数据集。而训练条件包括的属性可以表征训练数据对应的场景，也就是相当于将N个训练数据划分为多个场景的训练数据集。当对人工智能模型进行训练时，训练一个人工智能模型的训练数据满足该人工智能模型对应的训练条件，可以实现场景化的人工智能模型训练。从而使得所得到的模型更加符合信道条件，模型的正确性更高，通信效率更高。此外，该方法在获得场景化人工智能模型时还能够保持较低的信令开销。例如，一种可能的实现方式中，接入网设备将每个终端设备的所有训练数据构成一个终端设备专用(specific)的数据集，使用每个终端设备专用的数据集训练终端设备专用的模型，然后给每个终端设备配置该终端设备对应的模型，每个终端设备都使用其专用的模型。这种方式中，针对每个终端设备训练专用的模型导致网络的信令开销较大，此外，即使同一个终端设备，终端设备移动也会使其所处的位置、环境发生变化，导致同一个终端设备不同时刻收集到的无线信道样本的分布也可能有较大差异。该实现方式虽然能获得终端设备特定的模型，但是却无法利用数据的相关性，且信令开销较高。
上面的过程中，以接入网设备与终端设备之间交互为例，当AI实体位于接入网设备之外，是一个独立的模块或网元时，接入网设备可以将N个训练数据转发至AI实体，由AI实体进行人工智能模型训练；AI实体可以向接入网设备指示训练好的人工智能模型的信息，再由接入网设备向终端设备指示训练好的人工智能模型的信息，其他过程都和图5的流程类似，在此不再赘述。
上面的过程中，以接入网设备为例进行描述，针对一个模型，实际执行模型训练的可以是接入网设备中的模块(或单元、网元等)，如近实时RIC、接入网设备中的AI实体、CU中AI实体、或DU中的AI实体，不予限制。用于训练不同模型的模块可以相同，也可以不同，不予限制。模型的信息可以由训练模型的模块递交给接入网中的其他模块后，发送给终端设备。例如，CU中的AI实体训练得到一个模型后，该模型的信息依次经过DU和RU后，通过空口发送给终端设备。不再一一举例。
结合前面的描述，下面给出标记信息包括不同属性时，在不同训练条件下，接入网设备如何构建不同训练条件对应的训练数据集。
实现方式一，假设训练数据对应的标记信息包括该训练数据关联的SSB的索引。
一个SSB关联一个接入网设备的发送波束，每个发送波束的覆盖范围通常比较集中，且较长时间不会发生明显的变化，即每个发送波束的覆盖范围内的无线信道环境、分布都比较相似，因此可以将一个SSB关联的训练数据作为一个训练数据集。
在该实现方式中，终端设备通过测量获得训练数据时，可以测量获得至少一个SSB的RSRP，终端设备可以认为其处于最大的RSRP关联的SSB的覆盖范围内，因此可以用最大的RSRP关联的SSB的索引对该训练数据进行标记，即该训练数据的标记信息可以包括所述至少一个SSB中RSRP最大的SSB的索引。
例如，如图6所示，接入网设备的一个小区内包括6个SSB，索引分别为SSB1至SSB6，每个SSB对应的发送波束的覆盖范围不同，具体可以参考图6所示。终端设备在测量训练数据时，如果测量到的最大的RSRP关联的SSB的索引为SSB1，那么该训练数据对应的 标记信息可以包括SSB1。
在该实现方式中，接入网设备或者AI实体在训练人工智能模型时，可以根据SSB的索引确定训练人工智能模型的一个或多个训练数据，即构建训练人工智能模型的训练数据集。具体的，接入网设备或者AI实体在构建训练数据集时，可以将对应的SSB的索引相同的训练数据划分为一个训练数据集，此时人工智能模型的训练条件可以包括SSB的索引。例如，针对一个人工智能模型，其对应的训练条件包括第一SSB的第一索引；相应的，可以将N个训练数据中，对应的标记信息包括第一SSB的第一索引的一个或多个训练数据，用于训练人工智能模型。
举例来说，结合图6，小区中包括终端设备1和终端设备2，终端设备1上报的训练数据分别为训练数据1至训练数据6，终端设备2上报的训练数据分别为训练数据7至训练数据12。训练数据1至训练数据12分别关联的SSB可以参考表1所示。
表1
训练数据 SSB的索引
训练数据1 SSB1
训练数据2 SSB2
训练数据3 SSB3
训练数据4 SSB1
训练数据5 SSB2
训练数据6 SSB3
训练数据7 SSB1
训练数据8 SSB2
训练数据9 SSB3
训练数据10 SSB1
训练数据11 SSB2
训练数据12 SSB3
假设接入网设备或者AI实体通过上面的训练数据训练3个人工智能模型，其中人工智能模型1对应的训练条件包括的索引为SSB1，人工智能模型2对应的训练条件包括的索引为SSB2，人工智能模型3对应的训练条件包括的索引为SSB3。那么，用于训练每个人工智能模型的一个或多个训练数据可以如表2所示。
表2
Figure PCTCN2022126467-appb-000003
Figure PCTCN2022126467-appb-000004
结合表2，接入网设备或AI实体可以采用训练数据集1中的训练数据训练人工智能模型1，采用训练数据集2中的训练数据训练人工智能模型2，采用训练数据集3中的训练数据训练人工智能模型3。
一种实现方式中，接入网设备可以将训练得到的各个人工智能模型以及每个人工智能模型关联的训练条件指示给终端设备。终端设备在使用人工智能模型时，根据当前环境对应的训练条件确定应使用哪一个人工智能模型。例如，接入网设备向终端设备指示人工智能模型1的信息、人工智能模型2的信息以及人工智能模型3的信息，以及指示每个人工智能模型对应的训练条件。终端设备可以使用人工智能模型和推理数据进行推理。如果终端设备测量该推理数据时测量各个SSB，且测量到的SSB2的RSRP最大，那么终端设备可以使用人工智能模型2。
另一种实现方式中，终端设备在需要使用人工智能模型时，向接入网设备请求相应的人工智能模型。终端设备可以通过发送人工智能模型的索引，请求所使用的人工智能模型；终端设备也可以向接入网设备发送当前无线信道数据对应的标记信息，接入网设备可以向终端设备指示该标记信息满足的训练条件对应的人工智能模型。例如，终端设备发送的标记信息包括SSB3的索引，接入网设备确定人工智能模型3对应的训练条件包括SSB3，从而向终端设备指示人工智能模型3的信息。
上面的举例中，以训练条件包括一个SSB的索引为例，实际应用中，训练条件中可以包括多个SSB的索引。例如，接入网设备可以对SSB进行分组，将波束方向相邻或者波束指向的环境相似的一个或者多个SSB分为一组，将关联到同一组SSB的训练数据划分为一个训练数据集。如图6所示，如果索引为SSB1、SSB2以及SSB3的SSB对应的波束方向覆盖范围接近，则将索引为SSB1、SSB2以及SSB3的SSB分为一组；同理，也可以将索引为SSB4、SSB5以及SSB6的SSB分为一组。此时，可以将关联到SSB1、SSB2以及SSB3的训练数据划分为一个训练数据集，将关联到SSB4、SSB5以及SSB6的训练数据划分为另一个训练数据集。
例如，接入网设备获取到的N个训练数据如表3所示。
表3
训练数据 SSB的索引
训练数据1 SSB1
训练数据2 SSB2
训练数据3 SSB3
训练数据4 SSB4
训练数据5 SSB5
训练数据6 SSB6
训练数据7 SSB1
训练数据8 SSB2
训练数据9 SSB3
训练数据10 SSB4
训练数据11 SSB5
训练数据12 SSB6
结合前面的例子，根据表3，接入网设备确定的训练数据集、人工智能模型以及SSB的关联关系可以如表4所示。
表4
Figure PCTCN2022126467-appb-000005
结合表4，在训练人工智能模型1时，采用与SSB1、SSB2或SSB3中的至少一项关联的训练数据进行训练；在训练人工智能模型2时，采用与SSB4、SSB5或SSB6中的至少一项关联的训练数据进行训练。
终端设备如果测量到SSB1或SSB2或SSB3的RSRP最大，则确定当前无线信道数据与SSB1或SSB2或SSB3关联，从而可以将通过SSB1或SSB2或SSB3关联的训练数据训练的人工智能模型，作为待使用的人工智能模型，此处待使用人工智能模型1；如果测量到SSB4或SSB5或SSB6的RSRP最大，则确定当前无线信道数据与SSB4或SSB5或SSB6关联，从而可以将通过SSB4或SSB5或SSB6关联的训练数据训练的人工智能模型，作为待使用的人工智能模型，此处待使用人工智能模型2。
通过上面的方法，接入网设备或AI实体收集到多个训练数据时，可以通过其关联的SSB的索引对这些训练数据进行划分，构建场景化的训练数据集，使得每个训练数据集内的无线信道分布比较集中或者相似，实现场景化的模型训练。
上面描述的实现方式中，多个训练数据关联到同一个SSB或同一组SSB，就可以认为这些训练数据对应的无线信道环境、分布相似。本公开中，还可以按照其他属性对训练数据进行划分，使得划分后的训练数据对应的无线信道环境、分布相似。
实现方式二，假设训练数据对应的标记信息包括以下至少一项属性：
该训练数据关联的第一RSRP，第一RSRP为终端设备在获取训练数据时，测量到的最大RSRP；
该训练数据关联的第一SSB的索引，第一SSB为第一RSRP对应的SSB；
该训练数据关联的第二RSRP，第二RSRP为终端设备在获取训练数据时，测量到的第二大RSRP；
该训练数据关联的第二SSB的索引，第二SSB为第二RSRP对应的SSB；
训练数据关联的无线信道的类型，所述类型包括视距和非视距；
训练数据关联的定时提前。
通过上面的描述可知，相比于实现方式一，实现方式二中，标记信息包括的属性较多，例如，由于视距类型的无线信道与非视距类型的无线信道的无线信道环境不同，因此通过增加无线信道的类型，可以避免将不同类型的无线信道的训练数据划分为一个训练数据集。
和实现方式一类似，实现方式二中，终端设备获取到训练数据时，记录训练数据的标记信息，终端设备向接入网设备上报训练数据时，将训练数据对应的标记信息也同时上报给接入网设备。
接入网设备或者AI实体在训练人工智能模型时，可以将对应的标记信息满足人工智能模型对应的训练条件的一个或多个训练数据用于训练该个人智能模型，即构建训练该人工智能模型的训练数据集，训练数据集包括一个或多个训练数据。训练数据集构建完成后，接入网设备或者AI实体根据构建好的各个训练数据集对人工智能模型进行训练，一个训练数据集可以训练一个人工智能模型。
例如，训练数据集、人工智能模型以及训练条件之间的关联关系如表5所示。
表5
Figure PCTCN2022126467-appb-000006
接入网设备向终端设备指示人工智能模型时，还可以向终端设备指示该人工智能模型对应的训练条件。终端设备在使用人工智能模型时，根据人工智能模型对应的训练条件确定应使用哪一个人工智能模型。例如，终端设备在获取当前无线信道数据时，测量到当前无线信道数据属性为：最大的RSRP位于[a3,a4]内，且该RSRP对应SSB1，且确定无线信 道的类型为非视距，且确定TA位于[b3,b4]内，则确定当前无线信道数据的属性满足人工智能模型2对应的训练条件，则可以确定使用人工智能模型2。
另外，如果终端设备确定测量获取到的当前无线信道数据不满足任何一个人工智能模型对应的训练条件，则该终端设备不使用任何一个人工智能模型，或者任意选择一个人工智能模型，或者从中选择训练条件较为接近的人工智能模型，或者使用默认人工智能模型，默认人工智能模型为预定义或者通过接入网设备配置的，不予限制。
通过上面的方法，接入网设备或AI实体收集到多个训练数据时，可以通过其关联的SSB的索引、RSRP、TA、LOS/NLOS等信息，对这些训练数据进行划分，构建场景化的训练数据集，使得每个训练数据集内的无线信道分布比较集中或者相似，实现场景化的人工智能模型训练。
前两个实现方式中，主要针对在同一个小区内收集到的训练数据对人工智能模型进行训练，在本实施例中，还可以考虑使用不同小区的训练数据对人工智能模型进行训练，即不同小区的训练数据可以划分为一个训练数据集。
实现方式三，在实现方式二的基础上，训练数据的标记信息还可以包括：训练数据关联的小区的小区标识。
在实现方式三中，一种实现方式中，可以直接根据人工智能模型的训练条件，将多个小区的训练数据划分为多个不同的训练数据集，因此每个人工智能模型对应的训练数据集中包括一个或多个小区的训练数据。
另一种实现方式中，可以针对每个小区的训练数据，划分为多个训练数据集，其中一个训练数据集包括一个或多个训练数据。在划分训练数据集时，可以按照标记信息进行划分，例如将标记信息中包括相同SSB的索引的训练数据划分为一个训练数据集。
在每个小区已经划分好训练数据集的情况下，可以对每个小区的每个训练数据集进行数据分析，提取每个训练数据集的主要特征，或者统计每个训练数据集的统计特征，例如方差、协方差矩阵、多径时延分布等，将多个小区中特征相似的训练数据集进一步融合成一个训练数据集，则可将多个小区的训练数据集进一步融合。
例如，存在小区1和小区2，小区1内收集到的训练数据划分为训练数据集1至训练数据集3，每个小区的训练数据集和标记信息的关联关系如表6所示。
表6
小区 标记信息 训练数据集
小区1 标记信息1 训练数据集1
小区1 标记信息2 训练数据集2
小区1 标记信息3 训练数据集3
小区2 标记信息4 训练数据集4
小区2 标记信息5 训练数据集5
小区2 标记信息6 训练数据集6
假设需要训练的人工智能模型1对应的训练条件包括标记信息1、标记信息2以及标记信息4，人工智能模型2对应的训练条件包括标记信息3以及标记信息5，人工智能模型3对应的训练条件包括标记信息6，那么得到融合后的训练数据集、训练条件和人工智能模型的关联关系如表7所示。
表7
Figure PCTCN2022126467-appb-000007
从表7可知，用于训练人工智能模型1的训练数据集7，是由小区1中的训练数据集1、训练数据集2以及小区2中的训练数据集4合并后得到的，其他情况以此类推，不再赘述。
对于一个人工智能模型，如果用于训练该人工智能模型的训练数据来自多个小区，在其中一个小区，接入网设备可以将关联到该小区的标记信息的人工智能模型以及人工智能模型关联的该小区的一个或者多个标记信息发送给该小区的终端设备，终端设备在本地保存在本小区内可能用到的人工智能模型。例如，结合表7，对于小区1，接入网设备可以向小区1中的终端设备指示人工智能模型1和人工智能模型2，不指示人工智能模型3；当接入网设备向小区1中的终端设备指示人工智能模型1对应的训练条件时，可以只指示标记信息1和标记信息2，不指示标记信息4。
接入网设备也可以将关联到该小区和其他小区的标记信息的人工智能模型以及人工智能模型关联的本小区和上述其他小区的一个或者多个标记信息发送给本小区的各个终端设备。例如将关联到本小区和若干相邻小区的标记信息的各个人工智能模型以及每个人工智能模型关联的本小区和若干相邻小区一个或者多个标记信息发送给本小区的终端设备。又例如，关联到本小区所在的接入网通知区(RAN-based notification area，RNA)的标记信息的人工智能模型以及人工智能模型关联的该RNA的一个或者多个标记信息发送给本小区的终端设备。相应的，终端设备可以保存在本小区以及若干相邻小区可能用到的人工智能模型，便于在小区切换后快速确定适用的人工智能模型。
例如，结合表7，对于小区1，接入网设备可以向小区1中的终端设备指示人工智能模型1，人工智能模型2以及人工智能模型3；当接入网设备向小区1中的终端设备指示人工智能模型1对应的训练条件时，可以指示标记信息1、标记信息2以及标记信息4；同样的，当接入网设备向小区1中的终端设备指示人工智能模型3对应的训练条件时，可以指示标记信息6。
通过上面的方法，将对多个小区内的训练数据进行融合，使得多个小区内的训练数据可以共同构建用于训练人工智能模型的训练数据集，从而使多个小区可以共享相同的人工智能模型，提高人工智能模型的训练效率。
实现方式一至实现方式三中，都是依赖于终端设备上报的训练数据的标记信息确定用于训练人工智能模型的训练数据集，即是根据预先设定的训练条件对训练数据进行划分。本公开中，还可以先对训练数据进行划分，再根据划分后的训练数据，确定人工智能模型的训练条件，即从训练数据本身分析出训练数据的分布与标记信息中的哪些属性相关。
实现方式四中，人工智能模型对应的训练条件不再是预设的或预配置的，人工智能模型对应的训练条件根据训练该人工智能模型的一个或多个训练数据确定。
具体的，接入网设备或者AI实体获取到N个训练数据后，对这N个训练数据进行聚 类，将聚类特征相同或者在同一范围内的训练数据划分为一个训练数据集，一个训练数据集中包括一个或多个训练数据，其中聚类特征可以为训练数据的方差、协方差矩阵或多径时延分布等特征。由于每个训练数据关联一个标记信息，可以根据划分好的训练数据集，确定每个训练数据集中训练数据对应的属性的交集或并集，从而确定人工智能模型对应的训练条件。
例如，根据划分好的训练数据集确定，该训练数据集内的训练数据关联的SSB比较集中，一部分与SSB1关联，另一部分与SSB2关联，但其他信息，比如RSRP、TA、LOS/NLOS等信息无明显规律，则可以认为无线信道的分布与SSB相关，可以确定该训练数据训练的人工智能模型的训练条件包括SSB的索引，例如包括SSB1和SSB2。
再例如，一个训练数据集内的训练数据关联的无线信道类型均为视距，但其他信息，比如RSRP、TA等信息无明显规律，则可以认为无线信道的分布与视距存在关联，此时可以确定该训练数据训练的人工智能模型的训练条件包括无线信道类型为视距。
接入网设备或者AI实体根据构建好的各个训练数据集进行人工智能模型训练，每个训练数据集可以训练得到一个人工智能模型，训练得到的每个人工智能模型也同样关联一个训练条件。接入网设备将每个人工智能模型与其关联的训练条件发送给终端设备，具体可以参考前面的描述，在此不再赘述。
上面的方法中，通过对训练数据进行聚类，确定无线信道的分布与哪些信息相关，而不是根据预设的训练条件划分训练数据集，可以提高数据划分的准确性，提高对人工智能模型训练的准确性。
上述本申请提供的实施例中，分别从各个设备之间交互的角度对本申请实施例提供的方法进行了介绍。为了实现上述本申请实施例提供的方法中的各功能，接入网设备或终端设备可以包括硬件结构和/或软件模块，以硬件结构、软件模块、或硬件结构加软件模块的形式来实现上述各功能。上述各功能中的某个功能以硬件结构、软件模块、还是硬件结构加软件模块的方式来执行，取决于技术方案的特定应用和设计约束条件。
本申请实施例中对模块的划分是示意性的，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式。另外，在本申请各个实施例中的各功能模块可以集成在一个处理器中，也可以是单独物理存在，也可以两个或两个以上模块集成在一个模块中。上述集成的模块既可以采用硬件的形式实现，也可以采用软件功能模块的形式实现。
与上述构思相同，如图7所示，本申请实施例还提供一种通信装置700用于实现上述方法中接入网设备或终端设备的功能。该通信装置的形式不受限，可以是硬件结构、软件模块、或硬件结构加软件模块。例如，该装置可以为软件模块或者芯片系统。本申请实施例中，芯片系统可以由芯片构成，也可以包含芯片和其他分立器件。该装置700可以包括：处理单元701和通信单元702。
本申请实施例中，通信单元也可以称为收发单元，可以包括发送单元和/或接收单元，分别用于执行上文方法实施例中接入网设备、AI实体或终端设备执行的发送和接收的步骤。
以下，结合图7至图8详细说明本申请实施例提供的通信装置。应理解，装置实施例的描述与方法实施例的描述相互对应，因此，未详细描述的内容可以参见上文方法实施例，为了简洁，这里不再赘述。
通信单元也可以称为收发装置。处理单元也可以称为处理模块、或处理装置等。可选的，可以将通信单元702中用于实现接收功能的器件视为接收单元，将通信单元702中用 于实现发送功能的器件视为发送单元，即通信单元702包括接收单元和发送单元。通信单元有时可以实现为管脚、收发机、收发器、或收发电路等。处理单元有时可以实现为处理器、或处理单板等。接收单元有时可以实现为管脚、接收机、接收器或接收电路等。发送单元有时可以实现为管脚、发射机、发射器或者发射电路等。
通信装置700执行上面实施例中图5所示的流程中接入网设备的功能时：
处理单元，用于通过通信单元获取N个训练数据，对于N个训练数据中的一个训练数据，该训练数据对应一个标记信息，该标记信息用于指示该训练数据的属性，N为大于0的整数；
处理单元，用于通过通信单元向终端设备发送指示信息，指示信息用于指示M个人工智能模型的信息，对于M个人工智能模型中的一个人工智能模型，该人工智能模型是根据N个训练数据中的X个训练数据训练的，该X个训练数据是根据所述标记信息确定的，M和X为大于0的整数。
一种可能的实现方式中，对于所述M个人工智能模型中的所述一个人工智能模型，所述人工智能模型对应一个训练条件，所述X个训练数据对应的标记信息满足所述人工智能模型对应的训练条件。
一种可能的实现方式中，所述标记信息用于指示以下至少一项属性：
所述训练数据关联的同步信号块的索引；
所述训练数据关联的同步信号块的参考信号接收功率；
所述训练数据关联的无线信道的类型，所述类型包括视距和非视距；
所述训练数据关联的定时提前；
所述训练数据关联的小区的小区标识。
一种可能的实现方式中，所述训练条件包括以下至少一项：同步信号块的索引；参考信号接收功率的取值范围；无线信道的类型；定时提前的取值范围；小区标识。
一种可能的实现方式中，所述人工智能模型对应的训练条件为预设的。
一种可能的实现方式中，所述人工智能模型对应的训练条件是根据用于训练所述人工智能模型的所述X个训练数据确定的，用于训练所述人工智能模型的所述X个训练数据为所述N个训练数据中具有相同聚类特征的训练数据，或者，用于训练所述人工智能模型的所述X个训练数据为所述N个训练数据中聚类特征在同一范围内的训练数据。
一种可能的实现方式中，通信单元还用于：向所述终端设备发送所述M个人工智能模型对应的M个训练条件，其中，所述M个人工智能模型和所述M个训练条件一一对应。
一种可能的实现方式中，通信单元还用于：接收来自所述终端设备的模型指示信息，所述模型指示信息用于指示所述终端设备请求的人工智能模型；向所述终端设备发送所述指示信息，所述指示信息指示所述终端设备请求的人工智能模型的信息。
通信装置700执行上面实施例中图5所示的流程中终端设备的功能时：
通信单元，用于接收指示信息，所述指示信息用于指示M个人工智能模型的信息，对于所述M个人工智能模型中的一个人工智能模型，所述人工智能模型是根据N个训练数据中的X个训练数据训练的，M、N和X为大于0的整数。
一种可能的实现方式中，对于所述N个训练数据中的一个训练数据，所述训练数据对应一个标记信息，所述标记信息用于指示所述训练数据的属性。
一种可能的实现方式中对于所述M个人工智能模型中的所述一个人工智能模型，所述 人工智能模型对应一个训练条件，所述X个训练数据对应的标记信息满足所述人工智能模型对应的训练条件。
一种可能的实现方式中，标记信息用于指示以下至少一项属性：
所述训练数据关联的同步信号块的索引；
所述训练数据关联的同步信号块的参考信号接收功率；
所述训练数据关联的无线信道的类型，所述类型包括视距和非视距；
所述训练数据关联的定时提前；
所述训练数据关联的小区的小区标识。
一种可能的实现方式中，所述训练条件包括以下至少一项：同步信号块的索引；参考信号接收功率的取值范围；无线信道的类型；定时提前的取值范围；小区标识。
一种可能的实现方式中，所述人工智能模型对应的训练条件为预设的。
一种可能的实现方式中，所述人工智能模型对应的训练条件是根据用于训练所述人工智能模型的X个训练数据确定的，用于训练所述人工智能模型的所述X个训练数据为所述N个训练数据中具有相同聚类特征的训练数据，或者，用于训练所述人工智能模型的所述X个训练数据为所述N个训练数据中聚类特征在同一范围内的训练数据。
一种可能的实现方式中，通信单元还用于：接收所述M个人工智能模型对应的M个训练条件，其中，所述M个人工智能模型和所述M个训练条件一一对应。
一种可能的实现方式中，通信单元还用于：发送模型指示信息，所述模型指示信息用于指示所述终端设备请求的人工智能模型；
接收所述指示信息，所述指示信息指示所述终端设备请求的人工智能模型的信息。
以上只是示例，处理单元701和通信单元702还可以执行其他功能，更详细的描述可以参考图5所示的方法实施例中相关描述，这里不加赘述。
如图8所示为本申请实施例提供的装置800，图8所示的装置可以为图7所示的装置的一种硬件电路的实现方式。该通信装置可适用于前面所示出的流程图中，执行上述方法实施例中终端设备或者接入网设备的功能。为了便于说明，图8仅示出了该通信装置的主要部件。
如图8所示，通信装置800包括处理器810和接口电路820。处理器810和接口电路820之间相互耦合。可以理解的是，接口电路820可以为收发器、管脚、接口电路或输入输出接口。可选的，通信装置800还可以包括存储器830，用于存储处理器810执行的指令或存储处理器810运行指令所需要的输入数据或存储处理器810运行指令后产生的数据。可选地，存储器830的部分或全部可以位于处理器810中。
当通信装置800用于实现图5所示的方法时，处理器810用于实现上述处理单元701的功能，接口电路820用于实现上述通信单元702的功能。
当上述通信装置为应用于终端设备的芯片时，该终端设备芯片实现上述方法实施例中终端设备的功能。该终端设备芯片从终端设备中的其它模块(如射频模块或天线)接收信息，该信息是接入网设备发送给终端设备的；或者，该终端设备芯片向终端设备中的其它模块(如射频模块或天线)发送信息，该信息是终端设备发送给接入网设备的。
当上述通信装置为应用于接入网设备的芯片时，该接入网设备芯片实现上述方法实施例中接入网设备的功能。该接入网设备芯片从接入网设备中的其它模块(如射频模块或天线)接收信息，该信息是终端设备发送给接入网设备的；或者，该接入网设备芯片向接入 网设备中的其它模块(如射频模块或天线)发送信息，该信息是接入网设备发送给终端设备的。
如前文所述，图5的方法以接入网设备与终端设备之间交互为例，当AI实体位于接入网设备之外，是一个独立的模块或网元时，接入网设备可以将N个训练数据转发至AI实体，由AI实体进行人工智能模型训练；AI实体可以向接入网设备指示训练好的人工智能模型的信息，再由接入网设备向终端设备指示训练好的人工智能模型的信息，其他过程都和图5的流程类似。因此，上述关于接入网设备的介绍可以应用于AI实体。
可以理解的是，本公开中的处理器可以是中央处理单元，还可以是其它通用处理器、数字信号处理器、专用集成电路或者其它可编程逻辑器件、晶体管逻辑器件，硬件部件或者其任意组合。通用处理器可以是微处理器，也可以是任何常规的处理器。
本公开中的存储器可以是随机存取存储器、闪存、只读存储器、可编程只读存储器、可擦除可编程只读存储器、电可擦除可编程只读存储器、寄存器、硬盘、移动硬盘或者本领域熟知的任何其它形式的存储介质中。
本领域内的技术人员可以理解，本公开可提供为方法、系统、或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、光学存储器等)上实施的计算机程序产品的形式。
本申请是参照根据本申请的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
显然，本领域的技术人员可以对本公开进行各种改动和变型而不脱离本公开的范围。这样，倘若本公开的这些修改和变型属于本公开权利要求及其等同技术的范围之内，则本公开也意图包含这些改动和变型在内。

Claims (20)

  1. 一种通信方法，其特征在于，包括：
    获取N个训练数据，对于所述N个训练数据中的一个训练数据，所述训练数据对应一个标记信息，所述标记信息用于指示所述训练数据的属性，N为大于0的整数；
    向终端设备发送指示信息，所述指示信息用于指示M个人工智能模型的信息，对于所述M个人工智能模型中的一个人工智能模型，所述人工智能模型是根据所述N个训练数据中的X个训练数据训练的，所述X个训练数据是根据所述标记信息确定的，M和X为大于0的整数。
  2. 根据权利要求1所述的方法，其特征在于，对于所述M个人工智能模型中的所述一个人工智能模型，所述人工智能模型对应一个训练条件，所述X个训练数据对应的标记信息满足所述人工智能模型对应的训练条件。
  3. 根据权利要求1或2所述的方法，其特征在于，所述标记信息用于指示以下至少一项属性：
    所述训练数据关联的同步信号块的索引；
    所述训练数据关联的同步信号块的参考信号接收功率；
    所述训练数据关联的无线信道的类型，所述类型包括视距和非视距；
    所述训练数据关联的定时提前；
    所述训练数据关联的小区的小区标识。
  4. 根据权利要求2至3任一项所述的方法，其特征在于，所述训练条件包括以下至少一项：
    同步信号块的索引；
    参考信号接收功率的取值范围；
    无线信道的类型；
    定时提前的取值范围；
    小区标识。
  5. 根据权利要求2至4任一所述的方法，其特征在于，所述人工智能模型对应的训练条件为预设的。
  6. 根据权利要求2至4任一所述的方法，其特征在于，所述人工智能模型对应的训练条件是根据用于训练所述人工智能模型的所述X个训练数据确定的，用于训练所述人工智能模型的所述X个训练数据为所述N个训练数据中具有相同聚类特征的训练数据，或者，用于训练所述人工智能模型的所述X个训练数据为所述N个训练数据中聚类特征在同一范围内的训练数据。
  7. 根据权利要求2至6任一所述的方法，其特征在于，所述方法还包括：
    向所述终端设备发送所述M个人工智能模型对应的M个训练条件，其中，所述M个人工智能模型和所述M个训练条件一一对应。
  8. 根据权利要求1至7任一所述的方法，其特征在于，所述向终端设备发送指示信息，包括：
    接收来自所述终端设备的模型指示信息，所述模型指示信息用于指示所述终端设备请求的人工智能模型；
    向所述终端设备发送所述指示信息，所述指示信息指示所述终端设备请求的人工智能模型的信息。
  9. 一种通信方法，其特征在于，包括：
    接收指示信息，所述指示信息用于指示M个人工智能模型的信息，对于所述M个人工智能模型中的一个人工智能模型，所述人工智能模型是根据N个训练数据中的X个训练数据训练的，M、N和X为大于0的整数。
  10. 根据权利要求9所述的方法，其特征在于，对于所述N个训练数据中的一个训练数据，所述训练数据对应一个标记信息，所述标记信息用于指示所述训练数据的属性。
  11. 根据权利要求9或10所述的方法，其特征在于，
    对于所述M个人工智能模型中的所述一个人工智能模型，所述人工智能模型对应一个训练条件，所述X个训练数据对应的标记信息满足所述人工智能模型对应的训练条件。
  12. 根据权利要求10或11所述的方法，其特征在于，所述标记信息用于指示以下至少一项属性：
    所述训练数据关联的同步信号块的索引；
    所述训练数据关联的同步信号块的参考信号接收功率；
    所述训练数据关联的无线信道的类型，所述类型包括视距和非视距；
    所述训练数据关联的定时提前；
    所述训练数据关联的小区的小区标识。
  13. 根据权利要求11或12所述的方法，其特征在于，所述训练条件包括以下至少一项：
    同步信号块的索引；
    参考信号接收功率的取值范围；
    无线信道的类型；
    定时提前的取值范围；
    小区标识。
  14. 根据权利要求11至13任一所述的方法，其特征在于，所述人工智能模型对应的训练条件为预设的。
  15. 根据权利要求11至14任一所述的方法，其特征在于，所述人工智能模型对应的训练条件是根据用于训练所述人工智能模型的X个训练数据确定的，用于训练所述人工智能模型的所述X个训练数据为所述N个训练数据中具有相同聚类特征的训练数据，或者，用于训练所述人工智能模型的所述X个训练数据为所述N个训练数据中聚类特征在同一范围内的训练数据。
  16. 根据权利要求11至15任一所述的方法，其特征在于，所述方法还包括：
    接收所述M个人工智能模型对应的M个训练条件，其中，所述M个人工智能模型和所述M个训练条件一一对应。
  17. 根据权利要求9至16任一所述的方法，其特征在于，所述接收指示信息，包括：
    发送模型指示信息，所述模型指示信息用于指示所请求的人工智能模型；
    接收所述指示信息，所述指示信息指示所请求的人工智能模型的信息。
  18. 一种通信装置，其特征在于，用于实现如权利要求1～8任一项所述的方法，或用于实现如权利要求9～17任一项所述的方法。
  19. 一种通信装置，其特征在于，包括处理器和存储器，所述存储器和所述处理器耦合， 所述处理器用于执行如权利要求1～8任一项所述的方法，或用于执行如权利要求9～17任一项所述的方法。
  20. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质用于存储计算机程序，当所述计算机程序在计算机上运行时，使得所述计算机执行如权利要求1～8中任一项所述的方法，或使得所述计算机执行如权利要求9～17中任一项所述的方法。
PCT/CN2022/126467 2021-10-21 2022-10-20 一种通信方法及装置 WO2023066346A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111227090.4 2021-10-21
CN202111227090.4A CN116055012A (zh) 2021-10-21 2021-10-21 一种通信方法及装置

Publications (1)

Publication Number Publication Date
WO2023066346A1 true WO2023066346A1 (zh) 2023-04-27

Family

ID=86057937

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/126467 WO2023066346A1 (zh) 2021-10-21 2022-10-20 一种通信方法及装置

Country Status (2)

Country Link
CN (1) CN116055012A (zh)
WO (1) WO2023066346A1 (zh)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020214168A1 (en) * 2019-04-17 2020-10-22 Nokia Technologies Oy Beam prediction for wireless networks
US20210084601A1 (en) * 2019-09-16 2021-03-18 Movandi Corporation 5g signals detection using neural network
CN113169887A (zh) * 2018-09-28 2021-07-23 诺基亚技术有限公司 基于来自无线电网络和时空传感器的数据的无线电网络自优化
WO2021177867A1 (en) * 2020-03-03 2021-09-10 Telefonaktiebolaget Lm Ericsson (Publ) Determining location information about a drone

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113169887A (zh) * 2018-09-28 2021-07-23 诺基亚技术有限公司 基于来自无线电网络和时空传感器的数据的无线电网络自优化
WO2020214168A1 (en) * 2019-04-17 2020-10-22 Nokia Technologies Oy Beam prediction for wireless networks
US20210084601A1 (en) * 2019-09-16 2021-03-18 Movandi Corporation 5g signals detection using neural network
WO2021177867A1 (en) * 2020-03-03 2021-09-10 Telefonaktiebolaget Lm Ericsson (Publ) Determining location information about a drone

Also Published As

Publication number Publication date
CN116055012A (zh) 2023-05-02

Similar Documents

Publication Publication Date Title
US20220116799A1 (en) Method and device for o-ran-based performance optimization and configuration
JP6858852B2 (ja) 非同期マルチポイント送信方式
US20220052925A1 (en) Predicting Network Communication Performance using Federated Learning
US10652880B2 (en) Facilitating physical downlink shared channel resource element mapping indicator
WO2023036268A1 (zh) 一种通信方法及装置
US11296854B2 (en) Facilitating a mobile device specific physical downlink shared channel resource element mapping indicator
JP7038194B2 (ja) 無線通信システムにおいて前方互換受信機を容易にすること
KR20190132898A (ko) 무선 통신 시스템에서 데이터 수집 및 분석 기능을 활용하는 방법 및 장치
WO2023143267A1 (zh) 一种模型配置方法及装置
WO2023155468A1 (zh) 一种确定根因故障的方法及装置
WO2022172198A1 (en) Deep generative models for downlink channel estimation in fdd massive mimo systems
CN116848828A (zh) 机器学习模型分布
WO2023066346A1 (zh) 一种通信方法及装置
WO2023098860A1 (zh) 通信方法和通信装置
WO2023040971A1 (zh) 一种人工智能模型下载方法、装置及系统
WO2023093725A1 (zh) 一种校准方法及装置
WO2023036280A1 (zh) 一种模型测试方法及装置
WO2024051789A1 (zh) 一种波束管理方法
WO2023093777A1 (zh) 一种通信方法及装置
US20230336302A1 (en) Method for configuring reference signal
WO2023030221A1 (zh) 一种通信方法及装置
WO2024027424A1 (zh) 一种数据质量度量方法及装置
WO2024109682A1 (zh) 一种用于定位的方法及装置
WO2023125985A1 (zh) 模型的数据处理方法及装置
WO2023125996A1 (zh) 一种上行预编码方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22882951

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 2022882951

Country of ref document: EP

ENP Entry into the national phase

Ref document number: 2022882951

Country of ref document: EP

Effective date: 20240423