WO2023040700A1 - 一种人工智能ai通信方法及装置 - Google Patents

一种人工智能ai通信方法及装置 Download PDF

Info

Publication number
WO2023040700A1
WO2023040700A1 PCT/CN2022/117147 CN2022117147W WO2023040700A1 WO 2023040700 A1 WO2023040700 A1 WO 2023040700A1 CN 2022117147 W CN2022117147 W CN 2022117147W WO 2023040700 A1 WO2023040700 A1 WO 2023040700A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
information
unit
terminal device
capability information
Prior art date
Application number
PCT/CN2022/117147
Other languages
English (en)
French (fr)
Inventor
王四海
秦城
杨锐
李雪茹
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 WO2023040700A1 publication Critical patent/WO2023040700A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W8/00—Network data management
    • H04W8/22—Processing or transfer of terminal data, e.g. status or physical capabilities
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W8/00—Network data management
    • H04W8/22—Processing or transfer of terminal data, e.g. status or physical capabilities
    • H04W8/24—Transfer of terminal data

Definitions

  • the present application relates to the field of communication technology, and in particular to an artificial intelligence AI communication method and device.
  • AI Artificial intelligence
  • the wireless communication system introduces AI technology, and may gradually use AI modules to replace the functional modules in the wireless communication system.
  • a possible working mode is that the network device sends the AI model to the terminal device, and the terminal device receives the AI model from the network device and applies the AI model for wireless communication.
  • the AI model delivered by the network device may not be executed by the terminal device, which makes it impossible to apply the AI model for wireless communication.
  • the present application provides an artificial intelligence AI communication method and device, in order to better apply AI technology in wireless communication systems.
  • the present application provides an AI communication method, which can be executed by a first device, where the first device can be a terminal device, or a recipient of an AI model in a communication system.
  • the method may be implemented through the following steps: the first device obtains AI capability information, and the AI capability information includes the time and/or energy consumption of each reference AI unit in at least one reference AI unit executed by the first device; The device sends the AI capability information.
  • the first device sends its AI capability information to the second device, and the second device can evaluate the matching between the AI model and the AI capability of the first device to ensure that the first device and the second device use the AI model Availability of communication.
  • the method further includes: the first device receives configuration information sent by the second device, where the configuration information instructs the first device to start the AI communication mode , or, the configuration information indicates at least one AI model, or, the configuration information indicates configuration parameters of at least one AI model, or, the configuration information indicates an acquisition method of at least one AI model.
  • At least one AI model is determined according to AI capability information.
  • configuration information is response information to AI capability information.
  • the time for the first device to execute the first AI reference unit in at least one reference AI unit is a first time value, first time level or first time range, and the first device executes at least one reference AI unit.
  • the energy consumption of the first AI reference unit is a first energy consumption value, a first energy consumption level or a first energy consumption range.
  • the AI capability information includes the amount of input data used by each reference AI unit in the at least one reference AI unit executed by the first device.
  • the AI capability information includes one or more of the following: the first device executes the numerical accuracy of the input data used by each reference AI unit in at least one reference AI unit; each reference AI unit in at least one reference AI unit The numerical precision of the weight of the AI unit; the calculation precision when the first device executes each reference AI unit in the at least one reference AI unit.
  • the resources used by each reference AI unit in executing at least one reference AI unit on the first device are all available computing resources of the first device. Based on the above AI capability information, the second device can more easily, efficiently and accurately evaluate the matching between the AI capability of the first device and the complexity of the AI model, and then better perform AI communication.
  • the AI capability information includes at least one of an upper limit of time for the first device to execute the AI model, an upper limit of energy consumption, and resource usage.
  • the first device may send at least one of time upper limit information, energy consumption upper limit information, and resource usage information for executing the AI model to the second device, so as to inform the second device that it is executing the AI model.
  • the method before the first device sends AI capability information to the second device, the method further includes: the first device receives request information from the second device, and the request information is used to request the first device to send AI capability information to the second device Send the above AI capability information.
  • the first device sends AI capability information to the second device, including: the first device periodically sends AI capability information to the second device; or when the first device accesses the network where the second device is located, the first device The device sends AI capability information to the second device; or when the first device establishes a communication connection with the second device, the first device sends AI capability information to the second device; or when the computing resources used by the first device to execute the AI model change, The first device sends AI capability information to the second device.
  • the present application provides an AI communication method, which can be executed by a second device, where the second device can be a network device, or a sender of an AI model in a communication system.
  • the method may be implemented through the following steps: the second device receives AI capability information sent by the first device, where the AI capability information includes time and/or energy consumption for executing at least one reference AI unit by the first device.
  • the method further includes: the second device sends configuration information to the first device, where the configuration information instructs the first device to start the AI communication mode , or, the configuration information indicates at least one AI model, or, the configuration information indicates configuration parameters of at least one AI model, or, the configuration information indicates an acquisition method of at least one AI model, wherein the at least one AI model is based on the above-mentioned The AI capability information is determined.
  • At least one AI model is determined according to AI capability information.
  • configuration information is response information to AI capability information.
  • the time for the first device to execute the first AI reference unit in at least one reference AI unit is a first time value, first time level or first time range, and the first device executes at least one reference AI unit.
  • the energy consumption of the first AI reference unit is a first energy consumption value, a first energy consumption level or a first energy consumption range.
  • the AI capability information includes the amount of input data used by each reference AI unit in the at least one reference AI unit executed by the first device.
  • the AI capability information includes one or more of the following: the first device executes the numerical accuracy of the input data used by each reference AI unit in at least one reference AI unit; each reference AI unit in at least one reference AI unit The numerical precision of the weight of the AI unit; the calculation precision when the first device executes each reference AI unit in the at least one reference AI unit.
  • the AI capability information includes at least one of an upper limit of time for the first device to execute the AI model, an upper limit of energy consumption, and resource usage.
  • the method before the second device receives the AI capability information sent by the first device, the method further includes: the second device sends request information to the first device, and the request information is used to request the first device to send an AI capability information to the second device. Send the above AI capability information.
  • the method further includes: the second device determines at least one AI model according to the above AI capability information.
  • the second device determining at least one AI model according to the AI capability information includes: the second device determines at least one AI model according to the AI capability information and M sets of similarity information corresponding to the M AI models, where M Each group of similarity information in the group similarity information is the similarity information between one AI model among the M AI models and K reference AI units. Wherein, M and K are positive integers.
  • the M AI models are pre-stored in the second device or other devices; or the M AI models are acquired by the second device from other devices; or the M AI models are generated by the second device.
  • the first group of similarity information in the M groups of similarity information is the similarity information of the first AI model and K reference AI units
  • the first group of similarity information includes the first similarity information
  • the second A piece of similarity information is associated with the first ratio and/or the second ratio, wherein the first ratio is the ratio of the total number of layers in the first AI model that are the same as the first reference AI unit to the total number of layers in the first AI model
  • the second ratio is the proportion of the total calculation amount of the same layer as the first reference AI unit in the first AI model to the total calculation amount of the first AI model
  • M AI models include the first AI model
  • K reference AI units include First reference to the AI unit.
  • the first group of similarity information in the M groups of similarity information is the similarity information of the first AI model and K reference AI units, the first group of similarity information includes the first similarity information,
  • the first similarity information is associated with the first ratio and/or the second ratio, wherein the first ratio is the total number of layers of units in the first AI model that are the same as the first reference AI unit to the total number of layers of the first AI model
  • the second ratio is the ratio of the total calculation amount of the same unit as the first reference AI unit in the first AI model to the total calculation amount of the first AI model
  • M AI models include the first AI model
  • the cells include a first reference AI cell.
  • the present application provides an AI communication method, which can be executed by a first device, where the first device can be a terminal device, or a recipient of an AI model in a communication system.
  • the method can be implemented through the following steps: the first device receives the AI model information sent by the second device, the AI model information includes M groups of similarity information corresponding to M AI models, and each group of similarity information in the M groups of similarity information is the similarity information between an AI model among the M AI models and K reference AI units, where M and K are positive integers; the first device sends feedback information to the second device according to the AI model information.
  • the second device sends AI model information to the first device, and the first device can evaluate the matching between the AI model and the AI capabilities of the first device, ensuring that the first device and the second device communicate using the AI model feasibility.
  • the first group of similarity information in the M groups of similarity information is the similarity information of the first AI model and K reference AI units
  • the first group of similarity information includes the first similarity information
  • the second A piece of similarity information is associated with the first ratio and/or the second ratio, wherein the first ratio is the ratio of the total number of layers in the first AI model that are the same as the first reference AI unit to the total number of layers in the first AI model
  • the second ratio is the proportion of the total calculation amount of the same layer as the first reference AI unit in the first AI model to the total calculation amount of the first AI model
  • M AI models include the first AI model
  • K reference AI units include First reference to the AI unit.
  • the first group of similarity information in the M groups of similarity information is the similarity information of the first AI model and K reference AI units, the first group of similarity information includes the first similarity information,
  • the first similarity information is associated with the first ratio and/or the second ratio, wherein the first ratio is the total number of layers of units in the first AI model that are the same as the first reference AI unit to the total number of layers of the first AI model
  • the second ratio is the ratio of the total calculation amount of the same unit as the first reference AI unit in the first AI model to the total calculation amount of the first AI model
  • M AI models include the first AI model
  • the cells include a first reference AI cell.
  • the AI model information includes one or more of the following: the numerical accuracy of the input data used when executing each of the M AI models; the weight of each of the M AI models Numerical precision; the calculation precision when executing each of the M AI models.
  • the AI model information includes the total number of layers and/or the total calculation amount of each AI model in the M AI models.
  • the first device can more easily, efficiently and accurately evaluate the matching between the AI capability of the first device and the complexity of the AI model, and then better perform AI communication.
  • the AI model information includes an upper limit of time for executing the AI model.
  • the first device can obtain the time budget requirement of the second device for executing the AI model to be delivered to the first device, so as to better determine whether to start the AI communication mode or select an appropriate AI model.
  • the above feedback information indicates that the first device requests to start the AI mode, or, the feedback information indicates the evaluation result of at least one AI model among the M AI models, or, the feedback information requests the second device to send to the first device At least one AI model, wherein the M AI models include at least one AI model.
  • the method further includes: the first device receives configuration information sent by the second device, and the configuration information instructs the first device to start the AI mode , or, the configuration information indicates at least one AI model, or, the configuration information indicates configuration parameters of at least one AI model, or, the configuration information indicates an acquisition method of at least one model, wherein the above M AI models include the at least one AI model.
  • the method before the first device receives the AI model information sent by the second device, the method further includes: the first device sends request information to the second device, and the request information is used to request the second device to send an AI model information to the first device. Send the above AI model information.
  • the method before the first device sends feedback information to the second device based on the AI model information, the method further includes: the first device determines the feedback information according to the AI model information and the AI capability information of the first device.
  • the AI capability information of the first device indicates the execution time and/or energy consumption of at least one reference AI unit by the first device.
  • the possible design of the AI capability information reference may be made to the relevant description of the first aspect, and details are not repeated here.
  • the present application provides an AI communication method, which can be executed by a second device, where the second device can be a network device, or a sender of an AI model in a communication system.
  • the method can be implemented through the following steps: the second device sends AI model information to the first device, the AI model information includes M sets of similarity information corresponding to M AI models, and the M sets of similarity information is one of the M AI models Similarity information between the AI model and K reference AI units, wherein M and K are positive integers; the second device receives feedback information sent by the first device, and the feedback information is determined according to the AI model information.
  • the first group of similarity information in the M groups of similarity information is the similarity information of the first AI model and K reference AI units
  • the first group of similarity information includes the first similarity information
  • the second A piece of similarity information is associated with the first ratio and/or the second ratio, wherein the first ratio is the ratio of the total number of layers in the first AI model that are the same as the first reference AI unit to the total number of layers in the first AI model
  • the second ratio is the proportion of the total calculation amount of the same layer as the first reference AI unit in the first AI model to the total calculation amount of the first AI model
  • M AI models include the first AI model
  • K reference AI units include First reference to the AI unit.
  • the first group of similarity information in the M groups of similarity information is the similarity information of the first AI model and K reference AI units, the first group of similarity information includes the first similarity information,
  • the first similarity information is associated with the first ratio and/or the second ratio, wherein the first ratio is the total number of layers of units in the first AI model that are the same as the first reference AI unit to the total number of layers of the first AI model
  • the second ratio is the ratio of the total calculation amount of the same unit as the first reference AI unit in the first AI model to the total calculation amount of the first AI model
  • M AI models include the first AI model
  • the cells include a first reference AI cell.
  • the AI model information includes one or more of the following: the numerical accuracy of the input data used when executing each of the M AI models; the weight of each of the M AI models Numerical precision; the calculation precision when executing each of the M AI models.
  • the AI model information includes the total number of layers and/or the total calculation amount of each AI model in the M AI models.
  • the AI model information includes an upper limit of time for executing the AI model.
  • the above feedback information indicates that the first device requests to start the AI mode, or, the feedback information indicates the evaluation result of at least one AI model among the M AI models, or, the feedback information requests the second device to send to the first device At least one AI model, wherein the M AI models include at least one AI model.
  • the method further includes: the second device sends configuration information to the first device, and the configuration information instructs the first device to start the AI mode, or, the The configuration information indicates at least one AI model, or the configuration information indicates configuration parameters of the at least one AI model, or the configuration information indicates an acquisition method of the at least one AI model, wherein the M AI models include the at least one AI model.
  • the method before the second device sends the AI model information to the first device, the method further includes: the second device receives the request information sent by the first device, and the request information is used to request the second device to send the AI model information to the first device Send AI model information.
  • the above M AI models are pre-stored in the second device or other devices, or, the above M AI models are acquired by the second device from other devices, or, the above M AI models are the generated by the second device.
  • the present application also provides a communication device, the communication device may be a terminal device, or the communication device may be a receiving end device in a communication system, and the communication device has the function of realizing the above-mentioned first aspect or the third aspect A function of the first device of any aspect.
  • the functions described above may be implemented by hardware, or may be implemented by executing corresponding software on the hardware.
  • the hardware or software includes one or more modules corresponding to the above functions.
  • the structure of the communication device includes a transceiver unit and a processing unit, and these units can perform the corresponding functions of the first device in any one of the first aspect or the third aspect above, for details, refer to the method example The detailed description is not repeated here.
  • the structure of the communication device includes a transceiver and a processor, and optionally also includes a memory, and the transceiver is used for sending and receiving data, and for communicating and interacting with other devices in the communication system,
  • the processor is configured to support the communication device to execute the corresponding functions of the first device in any one of the first aspect or the third aspect above.
  • the memory coupled to the processor, holds program instructions and data necessary for the communication device.
  • the present application also provides a communication device, the communication device may be a network device, or the communication device may be a sending end device in a communication system, and the communication device has the function of realizing the above-mentioned second aspect or the fourth aspect
  • the functionality of the second device of any aspect may be implemented by hardware, or may be implemented by executing corresponding software on the hardware.
  • the hardware or software includes one or more modules corresponding to the above functions.
  • the structure of the communication device includes a transceiver unit and a processing unit, and these units can perform the corresponding functions of the second device in any aspect of the second aspect or the fourth aspect above, for details, refer to the method example The detailed description is not repeated here.
  • the structure of the communication device includes a transceiver and a processor, and optionally also includes a memory, and the transceiver is used for sending and receiving data, and for communicating and interacting with other devices in the communication system,
  • the processor is configured to support the communication device to perform the corresponding functions of the second device in any one of the second aspect or the fourth aspect above.
  • the memory coupled to the processor, holds program instructions and data necessary for the communication device.
  • the embodiment of the present application provides a communication system, which may include the first device and the second device mentioned above.
  • an embodiment of the present application provides a computer-readable storage medium, the computer-readable storage medium stores program instructions, and when the program instructions are run on a computer, the computer is made to perform the above-mentioned first to fourth aspects.
  • Exemplary, computer readable storage media may be any available media that can be accessed by a computer.
  • computer readable media may include non-transitory computer readable media, random-access memory (random-access memory, RAM), read-only memory (read-only memory, ROM), electrically erasable In addition to programmable read-only memory (electrically EPROM, EEPROM), CD-ROM or other optical disk storage, magnetic disk storage medium or other magnetic storage device, or can be used to carry or store the desired program code in the form of instructions or data structures and can Any other media accessed by a computer.
  • random-access memory random-access memory
  • read-only memory read-only memory
  • ROM read-only memory
  • CD-ROM or other optical disk storage magnetic disk storage medium or other magnetic storage device, or can be used to carry or store the desired program code in the form of instructions or data structures and can Any other media accessed by a computer.
  • the embodiment of the present application provides a computer program product including computer program codes or instructions, which, when run on a computer, enables the computer to realize any aspect of the above-mentioned first aspect to the fourth aspect and any possible method in the design.
  • the present application also provides a chip, including a processor, the processor is coupled to a memory, and is used to read and execute program instructions stored in the memory, so that the chip realizes the first aspect above to the method in any aspect of the fourth aspect and any possible design thereof.
  • FIG. 1 is a schematic structural diagram of a communication system provided by the present application.
  • FIG. 2 is a schematic diagram of a process of wireless communication between a network device and a terminal device using an AI model in an embodiment of the present application;
  • FIG. 3 is one of the flow charts of the AI communication method in the embodiment of the present application.
  • FIG. 4 is a schematic flow diagram of evaluating the AI capability and AI model matching degree of a network device in an embodiment of the present application
  • FIG. 5 is the second schematic flow diagram of the AI communication method in the embodiment of the present application.
  • FIG. 6 is a schematic flow diagram of evaluating the AI capability and AI model matching degree of the terminal device in the embodiment of the present application.
  • FIG. 7 is one of the structural schematic diagrams of the communication device in the embodiment of the present application.
  • FIG. 8 is the second structural diagram of the communication device in the embodiment of the present application.
  • the technical solution provided by this application can be applied to various communication systems, such as: the fifth generation (5th generation, 5G) or new radio (new radio, NR) system, long term evolution (long term evolution, LTE) system, LTE frequency division Duplex (frequency division duplex, FDD) system, LTE time division duplex (time division duplex, TDD) system, etc.
  • 5G fifth generation
  • NR new radio
  • long term evolution long term evolution
  • LTE long term evolution
  • LTE frequency division Duplex frequency division duplex
  • FDD frequency division duplex
  • TDD time division duplex
  • the technical solution provided by this application can also be applied to device to device (device to device, D2D) communication, vehicle to everything (vehicle-to-everything, V2X) communication, machine to machine (machine to machine, M2M) communication, machine type Communication (machine type communication, MTC), and Internet of things (internet of things, IoT) communication system or other communication systems.
  • D2D device to device
  • V2X vehicle-to-everything
  • M2M machine to machine
  • M2M machine type Communication
  • MTC machine type communication
  • IoT Internet of things
  • Embodiments of the present application provide an artificial intelligence AI communication method and device.
  • the method and the device described in this application are based on the same technical concept. Since the principles of the method and the device to solve the problem are similar, the implementation of the device and the method can be referred to each other, and the repetition will not be repeated.
  • At least one (species) refers to one (species) or multiple (species), and multiple (species) refers to two (species) or more than two (species).
  • Fig. 1 shows a schematic diagram of a wireless communication system applicable to the embodiment of the present application.
  • the wireless communication system may include at least one network device, such as the network device 111 and the network device 112 shown in Figure 1, and the wireless communication system may also include at least one terminal device, such as the terminal shown in Figure 1 Device 121 , terminal device 122 , terminal device 123 , terminal device 124 , terminal device 125 , terminal device 126 , terminal device 127 .
  • Both network devices and terminal devices can be configured with multiple antennas, and network devices and terminal devices can communicate using multi-antenna technology.
  • the network device when the network device communicates with the terminal device, the network device may manage one or more cells, and there may be an integer number of terminal devices in one cell. It should be noted that a cell may be understood as an area within the wireless signal coverage of the network device.
  • This application can be used in the scenario of communication between network equipment and terminal equipment, such as communication between network equipment 111 and terminal equipment 121, terminal equipment 122, and terminal equipment 123; for example, network equipment 111 and network equipment 112 can communicate with terminal equipment 124 communications.
  • the present application can also be used in a scenario of communication between terminal devices, for example, terminal device 122 can communicate with terminal device 125 .
  • the present application can also be used in a scenario where network devices communicate with each other, for example, network device 111 can communicate with network device 112 .
  • FIG. 1 is only a simplified schematic diagram for easy understanding, and the present application is not limited thereto.
  • the embodiments of the present application may be applicable to any communication scenario in which a sending-end device communicates with a receiving-end device.
  • the terminal equipment in the embodiment of the present application may also be referred to as user equipment (user equipment, UE), access terminal, subscriber unit, subscriber station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal , wireless communication device, user agent, or user device.
  • user equipment user equipment
  • UE user equipment
  • access terminal subscriber unit, subscriber station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal , wireless communication device, user agent, or user device.
  • a terminal device may be a device that provides voice/data to a user, for example, a handheld device with a wireless connection function, a vehicle-mounted device, and the like.
  • some terminals are: mobile phone (mobile phone), tablet computer, notebook computer, palmtop computer, mobile internet device (mobile internet device, MID), wearable device, virtual reality (virtual reality, VR) device, augmented reality (augmented reality, AR) equipment, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in remote medical surgery, smart grid Wireless terminals in transportation safety, wireless terminals in smart city, wireless terminals in smart home, cellular phones, cordless phones, session initiation protocols protocol, SIP) telephone, wireless local loop (wireless local loop, WLL) station, personal digital assistant (personal digital assistant, PDA), handheld device with wireless communication function, computing device or other processing device connected to a wireless modem, Wearable devices, terminal devices in a 5G network, or terminal devices in a future evolving public land mobile network (PLMN), etc., are not
  • the terminal device may also be a wearable device.
  • Wearable devices can also be called wearable smart devices, which is a general term for the application of wearable technology to intelligently design daily wear and develop wearable devices, such as glasses, gloves, watches, clothing and shoes.
  • a wearable device is a portable device that is worn directly on the body or integrated into the user's clothing or accessories. Wearable devices are not only a hardware device, but also achieve powerful functions through software support, data interaction, and cloud interaction.
  • Generalized wearable smart devices include full-featured, large-sized, complete or partial functions without relying on smart phones, such as smart watches or smart glasses, etc., and only focus on a certain type of application functions, and need to cooperate with other devices such as smart phones Use, such as various smart bracelets and smart jewelry for physical sign monitoring.
  • the device for realizing the function of the terminal device may be the terminal device, or may be a device capable of supporting the terminal device to realize the function, such as a chip system or a chip, and the device may be installed in the terminal device.
  • the system-on-a-chip may be composed of chips, or may include chips and other discrete devices.
  • the network device in this embodiment of the present application may be a device for communicating with a terminal device, and the network device may also be called an access network device or a wireless access network device, for example, the network device may be a base station.
  • the network device in this embodiment of the present application may refer to a radio access network (radio access network, RAN) node (or device) that connects a terminal device to a wireless network.
  • radio access network radio access network, RAN node (or device) that connects a terminal device to a wireless network.
  • the base station can broadly cover various names in the following, or replace with the following names, such as: Node B (NodeB), evolved base station (evolved NodeB, eNB), next generation base station (next generation NodeB, gNB), relay station, Access point, transmission point (transmitting and receiving point, TRP), transmitting point (transmitting point, TP), primary station, secondary station, multi-standard wireless (motor slide retainer, MSR) node, home base station, network controller, access point Ingress node, wireless node, access point (AP), transmission node, transceiver node, baseband unit (BBU), remote radio unit (RRU), active antenna unit (AAU), radio frequency Head (remote radio head, RRH), central unit (central unit, CU), distributed unit (distributed unit, DU), positioning node, etc.
  • NodeB Node B
  • eNB evolved base station
  • next generation NodeB next generation NodeB, gNB
  • relay station Access point
  • transmission point transmission
  • a base station may be a macro base station, a micro base station, a relay node, a donor node, or the like, or a combination thereof.
  • a base station may also refer to a communication module, a modem or a chip configured in the aforementioned equipment or device.
  • the base station can also be a mobile switching center, a device that assumes the function of a base station in D2D, V2X, and M2M communications, a network-side device in a 6G network, and a device that assumes the function of a base station in a future communication system.
  • Base stations can support networks of the same or different access technologies. The embodiment of the present application does not limit the specific technology and specific device form adopted by the network device.
  • Base stations can be fixed or mobile.
  • a helicopter or drone can be configured to act as a mobile base station, and one or more cells can move according to the location of the mobile base station.
  • a helicopter or drone may be configured to serve as a device in communication with another base station.
  • the network device mentioned in the embodiment of the present application may be a device including CU, or DU, or a device including CU and DU, or a control plane CU node (central unit-control plane, CU -CP)) and the user plane CU node (central unit-user plane (CU-UP) of the user plane) and the equipment of the DU node.
  • CU central unit-control plane
  • CU-UP central unit-user plane
  • Network equipment and terminal equipment can be deployed on land, including indoors or outdoors, hand-held or vehicle-mounted; they can also be deployed on water; they can also be deployed on aircraft, balloons and satellites in the air.
  • the scenarios where the network device and the terminal device are located are not limited.
  • AI Artificial intelligence
  • the signal detection can be the process of extracting the received signal containing interference noise from the wireless channel
  • the channel estimation is the process of estimating the model parameters of an assumed channel model from the received signal.
  • AI-based end-to-end communication link design is another example.
  • AI-based channel state information (CSI) feedback scheme which encodes CSI through a neural network and feeds it back to network devices.
  • an AI encoder and a corresponding AI decoder may be deployed in a network device.
  • the network device can send or deliver the AI coder to the terminal device, or instruct the terminal device to obtain the AI coder, so that the subsequent terminal device uses the AI coder to perform CSI coding.
  • the terminal device can use the AI encoder to encode the CSI, and send the encoded information to the network device.
  • the network device can use the AI decoder to decode the encoded information of the terminal device to obtain the restored information.
  • the AI encoder can also be understood as an AI model for information encoding
  • the AI decoder can also be understood as an AI model for information decoding.
  • AI models can be implemented based on neural network models.
  • the neural network model is a mathematical calculation model that imitates the behavioral characteristics of the human brain neural network and performs distributed parallel information processing.
  • Some complex neural network models may contain a large number of parameters or calculations, and the capabilities of terminal devices (such as computing power, storage capacity or energy) may be limited.
  • the AI capabilities of the terminal devices can support the execution (or: run, process) of the AI models sent by the network devices, for example, the storage capacity of the terminal devices can accommodate the AI
  • the computing capability of the model and the terminal device can support the AI model to complete the calculation within the required time, and the operating power consumption (or energy consumption) of the terminal device executing the AI model is within the expected and acceptable range.
  • the AI model calculation delay upper limit t i (that is, the AI model reasoning completion time limit) is known. Therefore, in one scheme, the computing capability C UE of the terminal device can be compared with the computational complexity C M of the AI model To evaluate the matching between the AI model and the AI capability of the terminal device, when C M /C UE +t th ⁇ t i , it means that the AI model can complete the calculation within the required delay, that is, the complexity of the AI model and the calculation of the terminal device Otherwise, it is considered that the complexity of the AI model does not match the computing power of the terminal device.
  • the computing capability C UE of the terminal device may be in units of floating-point operations per second (FLOPS), and the computational complexity C M of the AI model may be in units of floating-point operations (floating-point operations, FLOP) is the unit, and t th is the margin that can be configured in advance.
  • FLOPS floating-point operations per second
  • t th is the margin that can be configured in advance.
  • AI models for the same purpose may vary greatly.
  • the computing efficiencies of AI models with different structures may vary greatly.
  • different software conversion and optimization methods due to the need for a large amount of software conversion and optimization between AI models and hardware computing resources, different software conversion and optimization methods also bring different computing efficiencies.
  • different AI computing hardware implementations of terminal devices may also bring different AI model computing efficiencies. Therefore, the aforementioned AI model structure design, software environment, hardware implementation and other factors may lead to hundreds of times of deviation in computing efficiency, and it may be difficult to evaluate the actual matching between the AI model and the AI capabilities of the terminal device only by configuring the margin t th .
  • t th is too large, computing power resources may be wasted, and if t th is too small, the execution of the AI model may not be completed within the upper limit of the delay; and when the AI model structure, software environment, and hardware implementation are different, the value of t th may be different. It is also different.
  • terminal hardware information such as an emulator of a terminal device
  • the matching server can obtain an accurate evaluation result of the matching between the complexity of the AI model and the AI capability of the terminal device.
  • this solution needs to introduce a matching server and a corresponding interaction protocol, which complicates the network structure and interaction process, and increases the cost and evaluation delay.
  • this application proposes an AI communication method and device, which can realize the evaluation of AI model complexity and equipment AI matching, and ensure the feasibility of using AI models for communication services.
  • the first device may have computing hardware that executes the AI model, and is ready to obtain and use the AI model from the second device
  • the second device may have the AI model, and is ready to be sent to the second device for use.
  • the first device is the terminal device 121, the terminal device 122, or the terminal device 123 in FIG. 1
  • the second device is the network device 111 in FIG. 1; for another example, the first device is the terminal device 122 in FIG. 1,
  • the second device is the terminal device 125 in FIG. 1 ; for another example, the first device is the network device 111 in FIG. 1 , and the second device is the network device 112 in FIG. 1 .
  • the embodiment of the present application provides an AI communication method, which is applicable to the communication system shown in FIG. 1 .
  • the method may include but not limited to the following steps:
  • the terminal device acquires AI capability information, where the AI capability information includes time and/or energy consumption for the terminal device to execute each reference AI unit in at least one reference AI unit.
  • the AI capability information includes: the time and/or energy consumption of the terminal device to execute the reference AI unit P1 , the time and/or energy consumption of the terminal device executing the reference AI unit P2, and the time and/or energy consumption of the terminal device executing the reference AI unit P3.
  • the reference AI unit information may be standardized in advance or specified offline, and disclosed to the terminal device.
  • the terminal device executes the reference AI unit, which may also be understood as the terminal device performs an inference or training operation on the reference AI unit.
  • the reference AI unit may be at least one of a reference AI model, a reference AI module, and a reference AI operator.
  • the AI model may be a neural network model
  • the AI module may be a neural network module
  • the AI operator may be a neural network operator.
  • common neural network models such as ResNet series, MobileNet series, Transformer series, etc.
  • neural network models include layer-by-layer operators (neural network operators), and current neural network models (typically such as ResNet series models) are often Modular design, that is, the neural network model includes a series of modules (neural network modules) and neural network operators.
  • the AI module includes AI operators
  • the AI model includes AI modules and/or AI operators.
  • the execution time and/or energy consumption of each reference unit by the terminal device may be represented by an actual value, an approximate value, a level (for example: low/medium/high, or 1/2/3/4), or a range.
  • the above-mentioned at least one reference AI unit may include a first AI reference unit, and the time when the terminal device executes the first AI reference unit may be a first time value, a first time level, or a first time range, and the terminal device executes the first AI reference unit.
  • the energy consumption of an AI reference unit may be a first energy consumption value, a first energy consumption level or a first energy consumption range.
  • the terminal device may execute the above at least one reference AI unit by itself, and obtain its AI capability information.
  • the above-mentioned process of executing at least one AI reference unit may be completed in a device with the same or similar AI capabilities as the terminal device, or in a simulator that can truly reflect the AI capabilities of the terminal device. Specifically, the error of the calculation result obtained in a device or simulator with the same or similar AI capabilities as the terminal device may not exceed an expected value.
  • the terminal device can obtain the AI capability information from other devices or simulators.
  • the time and/or energy consumption information for executing at least one AI reference unit by the terminal device may be factory-set.
  • the terminal device sends AI capability information to the network device.
  • the network device receives the AI capability information sent by the terminal device.
  • the AI capability information sent by the terminal device may include the amount of input data used by the terminal device to execute at least one reference AI unit.
  • the quantity of input data may also be referred to as sample quantity.
  • the terminal device performs model training, it executes a batch of samples at a time, and the size of the sample batch determines the number of samples for one training.
  • end devices will get different computational efficiencies.
  • the AI capability information sent by the terminal device may also include at least one of the following: the terminal device (or another device with the same AI capability as the terminal device, or a simulator) executes at least one reference AI unit in which each reference AI unit uses The numerical accuracy of the input data; the numerical accuracy of the weight (or weight, coefficient) of each reference AI unit in the at least one reference AI unit; the terminal device (or other devices with the same AI capabilities as the terminal device, or simulators ) performs an operation accuracy of each reference AI unit in at least one reference AI unit.
  • the numerical precision of input data and weights will affect the computational efficiency of end devices. Wherein, the numerical precision may be int (integer type) 8, int16, float (floating point type) 16, etc., for example. Numerical precision can also be equivalently expressed as the precision of multiplication and addition. The precision of calculation will also affect the calculation efficiency of terminal equipment. For example, the precision of calculation can be int8, int16, float16, etc.
  • the resources used by the terminal device to execute each reference AI unit in the at least one reference AI unit are all available computing resources of the terminal device.
  • There may be many types of operating hardware for terminal devices for example, central processing unit (central processing unit, CPU), microprocessor (graphics processing unit, GPU), embedded neural network processor (neural-network processing unit, NPU) , or field programmable gate array (field programmable gate array, FPGA), etc.
  • the operating hardware may also be a combination of the above types.
  • heterogeneous computing refers to the distribution of AI models in multiple Execute on different types of computing units.
  • the AI model is distributed and executed on three types of computing units: CPU, GPU, and FPGA.
  • the computing resources available to the terminal device may include one or more of CPU resources, GPU resources, and FPGA resources.
  • the resource used by the terminal device to execute each reference AI unit in the at least one reference AI unit is a part of all available computing resources of the terminal device. It can be understood that the proportion of computing resources used by the terminal device to execute the reference AI unit will affect its AI capability information (ie, the time and/or energy consumption information of the terminal device to execute at least one AI reference unit).
  • mode 1 the terminal device may calculate the completion time or energy consumed for executing each reference AI unit in the at least one reference AI unit according to the ratio of the used computing resources to all available computing resources. Convert to obtain the final reported AI capability information.
  • Mode 2 The terminal device can report the proportion information of the used computing resources to all available computing resources to the network device, and then the network device can convert the AI capability information reported by the terminal device according to the resource proportion information.
  • the AI capability information sent by the terminal device may include (or indicate) the ratio of computing resources used by each reference AI unit in the execution of at least one reference AI unit by the terminal device to all available computing resources of the terminal device, or, The terminal device may separately send the resource ratio information to the network device in other signaling.
  • the process for the terminal device to execute the foregoing at least one reference AI unit may be completed when the terminal device leaves the factory.
  • AI capability information may be included in the same signaling, or may be included in different signaling, which may be referred to as specific capability information of the terminal device.
  • the terminal device may report the above AI capability information to the network device at one time, for example, the terminal device sends the first information to the network device, and the first information may include or Indicate the above-mentioned AI capability information, or the first information is the above-mentioned AI capability information.
  • the terminal device may report the AI capability information to the network device one or more times.
  • Example 1 The terminal device reports to the network device one or more times different capabilities such as the time and/or energy consumption of the AI unit, the amount of input data, and the accuracy of the input data.
  • the terminal device sends the first information to the network device, the first information includes or indicates the time and/or energy consumption of each reference AI unit in the at least one reference AI unit executed by the terminal device, and the terminal device sends the second information to the network device , the second information includes or indicates the amount of input data used to execute each reference AI unit in at least one reference AI unit, and the AI capability information may include the first information and the second information.
  • Example 2 The terminal device reports the above AI capability information to the network device one or more times.
  • the terminal device sends the first information to the network device, the first information includes or indicates capability information such as time and/or energy consumption for executing the AI unit, the amount of input data, the accuracy of the input data, etc., and the terminal device sends the first information to the network device
  • the second information includes or indicates updated capability information
  • the AI capability information may include the first information and the second information.
  • the method further includes:
  • the network device sends configuration information to the terminal device.
  • the terminal device receives the configuration information sent by the network device.
  • the configuration information is response information of the aforementioned AI capability information.
  • the configuration information may satisfy one or more of the following: the configuration information may instruct the terminal device to start the AI mode, and further, AI communication may be performed between the terminal device and the network device.
  • the configuration information may indicate at least one AI model, and further, the at least one AI model may be used for AI communication between the terminal device and the network device.
  • the configuration information may indicate configuration parameters of at least one AI model.
  • the configuration information may indicate an acquisition method of at least one AI model, for example, the configuration information may indicate a download address or an acquisition address of at least one AI model.
  • the at least one AI model may be determined according to the aforementioned AI capability information.
  • the network device may not respond after receiving the above AI capability information sent by the terminal device.
  • the terminal device sends the above AI capability information to the network device, and the specific triggering method may include:
  • Mode 1 The network device sends request information to the terminal device.
  • the terminal device receives the request information sent by the network device.
  • the request information is used to request the terminal device to send (or report) the aforementioned AI capability information to the network device.
  • Mode 2 The terminal device periodically sends the above AI capability information to the network device. Alternatively, the terminal device sends the aforementioned AI capability information to the network device at a predefined time interval or at a predefined specific time.
  • Mode 3 When the terminal device accesses the network where the network device is located, the terminal device sends AI capability information to the network device.
  • the terminal device accesses the network where the network device is located, it may also be described as after the terminal device accesses the network where the network device is located, or within a certain period of time after the terminal device accesses the network where the network device is located.
  • the terminal device accessing the network where the network device is located may also be understood as establishing a communication connection between the terminal device and the network device.
  • Mode 4 When the computing resources that the terminal device can use to execute the AI model change (for example: increase or decrease), or when the proportion of computing resources that the terminal device can use to execute the AI model changes, the terminal device sends a message to the network device AI capability information.
  • the computing resources that the terminal device can use to execute the AI model change it can also be described as after the computing resources that the terminal device can use to execute the AI model change, or the computing resources that the terminal device can use to execute the AI model send a change For a certain period of time, it will not be repeated here.
  • the complexity of the operations that need to be completed is different, and the power consumption/energy requirements, time requirements, and resource requirements that the terminal device can accept (or tolerate) for executing the AI model are also different. of. That is to say, in different application scenarios, the maximum time requirements, maximum energy consumption requirements, or resource usage requirements for the terminal device to execute the AI model are different.
  • the maximum time requirement can be how long the terminal device should After executing the AI model
  • the maximum energy consumption requirement can be the maximum energy consumption that the terminal device is allowed to consume after executing the AI model
  • the resource usage requirement can be the maximum ratio of the resources allowed to be used by the terminal device to execute the AI model to the available resources of the terminal device, or the terminal device
  • the terminal device when it is in a different application scenario, it may send AI capability information to the network device, where the AI capability information may include an upper limit value of time for the terminal device to execute the AI model, an upper limit value of energy consumption, and At least one of resource usage.
  • the time upper limit may be referred to as a time upper limit, maximum time, or maximum time requirement
  • the energy consumption upper limit may be referred to as an energy consumption upper limit, maximum energy consumption, or maximum energy consumption requirement
  • resource usage may be, for example, The hardware resource configuration that the terminal device can use, the upper limit of the resource ratio, etc.
  • the resource ratio upper limit value may be referred to as a resource ratio upper limit, a maximum resource ratio, or a maximum resource ratio requirement.
  • the network device may also send query information to the terminal device, wherein the query information instructs the terminal device to report the above-mentioned time upper limit value for executing the AI model, energy consumption One or more of Limits, and Resource Usage.
  • the terminal device may report to the network device at least one of the upper limit value of the time used to execute the AI model, the upper limit value of energy consumption, and the resource usage information, so as to inform At least one of time budget requirements, energy consumption constraints, and resource ratio consumption constraints for the network device to execute the AI model.
  • the time upper limit information, energy consumption upper limit information, or resource usage information and the information reported in step S302 may be carried in different signaling.
  • the terminal device reports its AI capability information, so that the network device can evaluate the matching situation between the AI capability of the terminal device and the AI model (for example: the complexity of the AI model) through the AI capability information, and then decide whether to start
  • the AI communication mode or the delivery of a suitable AI model can improve the simplicity, efficiency, and accuracy of the network device evaluation results, thereby enabling better AI communication.
  • the network device may evaluate the terminal device's AI capability and the complexity of the AI model, and determine the configuration information.
  • FIG. 4 exemplarily shows a flow chart of a method for network equipment to evaluate AI capability and AI model matching degree.
  • the terminal device and the network device may obtain reference AI unit information.
  • the reference AI unit information may be standardized in advance or specified offline, and disclosed to network devices and terminal devices.
  • the AI reference table may be standardized in advance or formulated offline, where the AI reference table may include names or numbers or indexes of N reference AI units (for example: AI models, AI modules, or AI operators), and the N The structure description parameters of the N reference AI units, wherein the structure description parameters of the N reference AI units can also be provided by citing references or links.
  • N reference AI units for example: AI models, AI modules, or AI operators
  • the specific weight coefficient value (weight value) of each AI unit may be undefined, or predefined, or be a random number.
  • the AI reference table is disclosed to terminal devices and network devices in advance. Specifically, in step S301, at least one reference AI unit executed by the terminal device may be K reference AI units among the N reference AI units, where N is a positive integer, and K is a positive integer not greater than N.
  • the network device acquires the similarity information of the M AI models.
  • M is a positive integer.
  • the M AI models may be pre-stored in the network device or other devices, or the M AI models are acquired by the network device from other devices, or the M AI models are obtained by the network device through existing Transformed or generated by AI models.
  • the network device acquires M sets of similarity information between the M AI models and the K reference AI units.
  • the M AI models include the first AI model
  • the K reference AI units include the first reference AI unit.
  • the similarity information corresponding to the first AI model is the first group of similarity information in the above M groups of similarity information
  • the first group of similarity information includes the first similarity information
  • the first similarity information and the first ratio sum /or the second ratio is associated
  • the first ratio is the ratio of the total number of layers in the first AI model that are the same as the first reference AI unit to the total number of layers in the first AI model
  • the second ratio is the ratio of the first AI model The proportion of the total calculation amount of the same layer as the first reference AI unit in the total calculation amount of the first AI model.
  • the M sets of similarity information include: M sets of similarity information corresponding to M AI models.
  • the first set of similarity information corresponding to the first AI model includes: K pieces of similarity information corresponding to the first AI model and K AI reference units.
  • Table 1 shows a common neural network model, a residual neural network (ResNet) model.
  • AI models can include AI modules, for example, the ResNet-34 model shown in Table 1 includes AI modules Further, the AI module can also include AI operators, for example, the AI module Including the AI operator [3 ⁇ 3,64] (a convolution operator).
  • the first AI model as a ResNet-50 model (that is, a 50-layer ResNet model)
  • the first reference AI unit is a model, such as the ResNet-34 model shown in Table 1
  • the second ratio can be obtained by dividing the total calculation amount (in FLOP) of these same layers by the total calculation amount of the ResNet-50 model, which is 3.8 ⁇ 10 9 FLOPs.
  • the M AI models include the first AI model
  • the K reference AI units include the first reference AI unit.
  • the similarity information corresponding to the first AI model is the first group of similarity information in the above M groups of similarity information
  • the first group of similarity information includes the first similarity information
  • the first similarity information and the first ratio sum /or the second ratio is associated
  • the first ratio is the ratio of the total number of layers of units in the first AI model that are the same as the first reference AI unit to the total number of layers of the first AI model
  • the second ratio is the ratio of the first AI The proportion of the total calculation amount of the same unit as the first reference AI unit in the model to the total calculation amount of the first AI model.
  • the first AI model as the ResNet-50 model shown in Table 1
  • the first reference AI unit is a model, such as the ResNet-34 model shown in Table 1
  • the first ratio is 0.
  • the first reference AI unit is a module, such as a module
  • the first reference AI unit is a module, such as a module
  • the number of the module included in the ResNet-50 model is 0, and at this time, the first ratio is 0.
  • the first reference AI unit is an operator, for example, the operator [3 ⁇ 3,64], and the number of the operator included in the ResNet-50 model is 3, then the first ratio is 3/50.
  • the network device may perform local calculations to obtain similarity information of the M AI models.
  • the calculation process can also be completed in other devices, and then the network device can obtain the calculation result from other devices.
  • S403 The network device evaluates the matching degree between the AI capability and the AI model.
  • the network device may evaluate the matching between the AI capability of the terminal device and the M AI models based on the AI capability information received in step S302.
  • the network device may obtain one or more of expected completion time, expected power consumption, or expected resource ratio used by the M models executed in the terminal device through analogy evaluation. For example, when the similarity is represented by the aforementioned first ratio, assuming that the total number of layers of the first AI model is L0, in step S402, assuming that the similarity between the first AI model obtained by the network device and the reference AI model K1 is S1, The similarity between the first AI model and the reference AI module K2 is S2, and the similarity between the first AI model and the reference AI operator K3 is S3.
  • the AI capability information sent by the terminal device to the network device includes t 1 , t 2 and t 3 when the terminal device executes the aforementioned reference AI model, reference AI module, and reference AI operator, respectively.
  • the network device can estimate that the expected completion time for the terminal device to execute the first AI model is
  • L0, L1, L2, and L3 can represent the total amount of calculation, and the network device can also adopt a similar scheme to obtain the above results, which will not be repeated here.
  • the method for the network device to evaluate the expected energy consumed by the terminal device to execute the first AI model and the expected resource ratio can also be obtained by analogy evaluation, which will not be repeated here.
  • the above description is an exemplary method for the network device to evaluate the matching degree between the AI capability of the terminal device and the AI model.
  • the network device can use M models and refer to the detailed structure of the AI unit. analysis to obtain more accurate evaluation results, which is not limited in this application.
  • the network device can determine whether the M models match the AI capability of the terminal device. For example, if the expected completion time of the first AI model among the M models does not exceed the time budget requirement of the terminal device, and the expected energy consumption does not exceed the energy consumption limit of the terminal device, and the proportion of resources expected to be consumed does not exceed the resource of the terminal device Ratio limitation, that is, it can be determined that the first AI model matches the AI capability of the terminal device.
  • the network device may send configuration information to the terminal device.
  • step S303 when the configuration information indicates at least one AI model, or the configuration information indicates the configuration parameters of at least one AI model, or the configuration information indicates the acquisition method of at least one AI model, before the network device sends the configuration information to the terminal device, the network device Determine at least one AI model according to the AI capability information. Specifically, in one case, the network device has only one AI model for the current application, and if the AI model matches the AI capability of the terminal device, then the AI model is the AI model determined by the network device; in another case, The network device has more than one AI model for the current application, and the network device can select at least one AI model from the AI models that match the AI capabilities of the terminal device to send or configure it to the terminal device.
  • the selection principle of the network device can be as follows: The expected completion time is the shortest, the expected energy consumption is the least, the resource ratio expected to be used is the smallest, or the random principle. Understandably, if the network device can determine at least one AI model, it is considered that the network device can start the AI mode.
  • the network device may indicate a download address or an acquisition address of at least one AI model.
  • the M AI models may be pre-stored in other devices (eg, the third device), and at this time, the network device may instruct the terminal device to acquire the at least one AI model from the third device.
  • the terminal device sends the data obtained by executing the reference AI model, AI module or AI operator to the network device as AI capability information, which can accurately represent the AI capability of the terminal device, so that the network device can obtain more accurate evaluation results , and then decide whether to enable AI communication or deliver AI models.
  • the network device and the terminal device can obtain the reference AI unit first, based on the reference AI unit, the terminal device can obtain the AI capability information as described in step S301, and the network device can obtain the AI model similarity information as described in step S402, and then, The terminal device can send the acquired AI capability information to the network device, so that the network device can evaluate the matching degree between the AI capability and the AI model. After the evaluation is completed, the network device can send a suitable AI encoder to the terminal device. After receiving the AI encoder, the terminal device can use the AI encoder to encode the CSI, and send the encoded information to the network device. Then, the network device can use the AI decoder corresponding to the AI encoder to decode the encoded information of the terminal device to obtain the restored information.
  • the embodiment of the present application also provides an AI communication method, which is applicable to the communication system shown in FIG. 1 .
  • the method may include but not limited to the following steps:
  • the network device sends AI model information to the terminal device, and accordingly, the terminal device receives the AI model information sent by the network device.
  • the AI model information may indicate the complexity information of each AI model in the M AI models.
  • the complexity information of each AI model in the M AI models can be represented by similarity data.
  • the AI model information indicates M sets of similarity information (or similarity data) corresponding to M AI models, and each set of similarity information in the M sets of similarity information is one of the M AI models and one AI model.
  • Similarity information of K reference AI units wherein the K reference AI units belong to N reference AI units, M and N are positive integers, and K is a positive integer less than or equal to N.
  • the AI model information includes 3 sets of similarity information corresponding to 3 AI models (Q1, Q2, and Q3) (the first set of similarity information, the second set of similarity information, The third group of similarity information), where the first group of similarity information is the similarity information between AI model Q1 and 5 reference AI units, and the second group of similarity information is the similarity between AI model Q2 and 5 reference AI units Information, the third group of similarity information is the similarity information between AI model Q3 and 5 reference AI units.
  • the M AI models may be pre-stored in the network device or other devices, or the M AI models are acquired by the network device from other devices, or the M AI models are obtained by the network device through existing Transformed or generated by AI models.
  • the N reference AI units may be standardized in advance or specified offline, and disclosed to network devices and terminal devices.
  • the N reference AI units may be in the AI reference table, and the specific content of the AI reference table may refer to the above description.
  • the M AI models include the first AI model
  • the K reference AI units include the first reference AI unit.
  • the similarity information corresponding to the first AI model is the first group of similarity information in the above M groups of similarity information
  • the first group of similarity information includes the first similarity information
  • the first similarity information and the first ratio sum /or the second ratio is associated
  • the first ratio is the ratio of the total number of layers in the first AI model that are the same as the first reference AI unit to the total number of layers in the first AI model
  • the second ratio is the ratio of the first AI model The proportion of the total calculation amount of the same layer as the first reference AI unit in the total calculation amount of the first AI model.
  • the M AI models include the first AI model
  • the K reference AI units include the first reference AI unit.
  • the similarity information corresponding to the first AI model is the first group of similarity information in the above M groups of similarity information
  • the first group of similarity information includes the first similarity information
  • the first similarity information and the first ratio sum /or the second ratio is associated
  • the first ratio is the ratio of the total number of layers of units in the first AI model that are the same as the first reference AI unit to the total number of layers of the first AI model
  • the second ratio is the ratio of the first AI The proportion of the total calculation amount of the same layer as the first reference AI unit in the model to the total calculation amount of the first AI model.
  • the M sets of similarity information corresponding to the M AI models may be represented by actual values, approximate values, levels, or ranges.
  • the above-mentioned first similarity information may be a first similarity value, a first similarity level, or a first similarity range.
  • the network device may locally calculate and obtain the above M sets of similarity information.
  • the process of calculating M sets of similarity information may be completed on a third device (eg, other network devices or third-party servers), and the network device may obtain the M sets of similarity information from the third device.
  • the AI model information sent by the network device may include at least one of the following: the numerical precision of the input data used when the network device or the third device executes each of the above M AI models; The numerical precision of the weight (or weight, coefficient) of each AI model; the calculation precision when the network device or the third device executes each AI model in the M AI models.
  • the AI model information sent by the network device may also include the total number of layers and the total calculation amount of each of the M AI models. It can be understood that by acquiring more detailed information on the M AI models, the terminal device can more accurately evaluate the complexity of the M AI models.
  • the time budget information for executing the AI model varies.
  • the aforementioned AI model information may also include an upper limit of time for executing the AI model, that is, time budget information for executing the AI model to be delivered, that is, how long it takes to execute the AI model.
  • the time upper limit value may be referred to as a time upper limit, a maximum time, or a maximum time requirement.
  • the information included in the above AI model information may be included in the same signaling, or may be included in different information.
  • the network device may send the above AI model information to the terminal device at one time.
  • the network device may send the above AI model information to the terminal device one or more times.
  • S502 The terminal device sends feedback information to the network device.
  • the network device receives the feedback information sent by the terminal device.
  • the feedback information may satisfy one or more of the following: the feedback information may be used to request to start the AI communication mode. Alternatively, the feedback information may include or indicate an evaluation result of at least one AI model among the above M AI models. Alternatively, the feedback information may request the network device to send at least one AI model to the terminal device, where the at least one AI model belongs to the aforementioned M AI models.
  • the method before the aforementioned step S501, that is, before the network device sends AI model information to the terminal device, the method further includes:
  • the network device acquires the AI model information.
  • the network device can locally calculate and obtain the above M sets of similarity information, so as to obtain the AI model information.
  • the network device may acquire the M sets of similarity information from the third device.
  • the method further includes:
  • the network device sends configuration information to the terminal device.
  • the terminal device receives the configuration information sent by the network device.
  • the configuration information is response information of the above feedback information.
  • the configuration information may satisfy one or more of the following: the configuration information may instruct the terminal device to start the AI mode, and further, AI communication may be performed between the terminal device and the network device.
  • the configuration information may indicate at least one AI model, and further, the at least one AI model may be used for AI communication between the terminal device and the network device.
  • the configuration information may indicate an acquisition method of at least one AI model, for example, the configuration information may indicate a download address or an acquisition address of at least one AI model.
  • the at least one AI model is determined according to the above feedback information.
  • the M AI models may be pre-stored in other devices (for example: the third device), at this time, the network device may instruct the terminal device to obtain from the first The at least one AI model is acquired in the third device.
  • the network device may not respond after receiving the above feedback information sent by the terminal device.
  • the method before the aforementioned step S501, that is, before the network device sends AI model information to the terminal device, the method further includes: the terminal device sends request information to the network device, and the request information is used to request the network device to send the AI model information to the terminal device.
  • the terminal device sends the aforementioned AI model information.
  • the terminal device can evaluate the complexity of the M AI models, and then determine the AI capabilities of the M AI models and the terminal device matches.
  • the method further includes: the terminal device determines according to the AI model information received in step S501 and the AI capability information of the terminal device Feedback.
  • the terminal device determines according to the AI model information received in step S501 and the AI capability information of the terminal device Feedback.
  • FIG. 6 exemplarily shows a flow chart of a method for evaluating AI capability and AI model matching degree of a terminal device.
  • the network device and the terminal device may obtain reference AI unit information.
  • the terminal device After obtaining the reference AI unit information, the terminal device obtains AI capability information.
  • step S601 the specific implementation of network equipment and terminal equipment obtaining reference AI unit information can refer to the expression in step S401 of the aforementioned embodiment, and the specific implementation of terminal equipment acquiring AI capability information in step S602 can refer to the description in step S301 of the aforementioned embodiment expression and will not be repeated here.
  • S603 The terminal device evaluates the matching degree between the AI capability and the model.
  • the terminal device After the aforementioned step S501, that is, after the terminal device receives the AI model information sent by the network device, the terminal device can evaluate the matching between the AI capability of the terminal device and the M AI models based on the AI model information received in step S501.
  • the terminal device may also obtain one or more of the expected completion time, expected power consumption, or expected resource ratio of the M models executed in the terminal device through analogy evaluation.
  • the AI model information sent by the network device to the terminal device includes the total number of layers of the first AI model as L0, and the similarity information: the first AI model
  • the similarity with the reference AI model K1 is S1
  • the similarity between the first AI model and the reference AI module K2 is S2
  • the similarity between the first AI model and the reference AI operator K3 is S3.
  • the terminal device obtains t 1 , t 2 and t 3 respectively for the completion time of executing the above-mentioned reference AI model, reference AI module and reference AI operator.
  • the terminal device can evaluate the expected completion time of executing the first AI model as
  • L0, L1, L2, and L3 can represent the total amount of calculation
  • the network device can also adopt a similar scheme to obtain the above results, which will not be repeated here.
  • the method for the terminal device to evaluate the expected energy consumed and the expected resource ratio for executing the first AI model can also be obtained by analogy evaluation, which will not be repeated here.
  • the terminal The device can determine whether the M models match their AI capabilities. For example, if the expected completion time of the first AI model among the M models does not exceed the time budget requirement of the terminal device, and the expected energy consumption does not exceed the energy consumption limit of the terminal device, and the proportion of resources expected to be consumed does not exceed the resource of the terminal device ratio limit, the terminal device can determine that the first AI model matches the AI capability of the terminal device.
  • the time budget requirement of the terminal device may be the time budget information sent by the aforementioned network device for the terminal device to execute the AI model, or may be the local time budget requirement of the terminal device;
  • the energy consumption limit of the terminal device may be the local terminal device The energy consumption limit;
  • the resource ratio limit of the terminal device may be a local resource ratio limit of the terminal device.
  • the terminal device may send feedback information to the network device.
  • the feedback information may indicate at least one AI model.
  • the terminal device may select at least one AI model from models matching its AI capabilities, and request the network device to send the model through the feedback information, and the terminal device
  • the selection principle of can be the shortest expected completion time, the least expected energy consumption, the smallest expected resource ratio, or a random principle.
  • steps S500 , S501 , S502 , and S503 can refer to the relevant expressions in the embodiment shown in FIG. 5 , and details are not repeated here.
  • the terminal device can evaluate the matching between the AI model to be delivered and the AI capability of the terminal device more simply, efficiently and accurately. Further determining whether to request to start the AI model or to request the delivered AI model can improve the accuracy and efficiency of the terminal device in evaluating the AI capability and the matching degree of the AI model.
  • the terminal device can send its AI capability information to the network device, so that the network device can accurately and efficiently evaluate the matching between the AI model and the AI capability of the terminal device, or, the network device can send the AI capability information to the terminal device.
  • Model information so that terminal devices can accurately and efficiently evaluate the match between AI models and their AI capabilities.
  • the method provided by the embodiment of the present application can realize the evaluation of AI capability and AI model matching degree between the terminal device and the network device without introducing additional server devices and interaction protocols.
  • the network device and the terminal device include hardware structures and/or software modules corresponding to each function.
  • the present application can be implemented in the form of hardware or a combination of hardware and computer software with reference to the units and method steps of the examples described in the embodiments disclosed in the present application. Whether a certain function is executed by hardware or computer software drives the hardware depends on the specific application scenario and design constraints of the technical solution.
  • FIG. 7 and FIG. 8 are schematic structural diagrams of a possible communication device provided by an embodiment of the present application. These communication apparatuses may be used to realize the functions of the terminal device or the network device in the foregoing method embodiments, and thus also realize the beneficial effects of the foregoing method embodiments.
  • the communication device may be a terminal device or a network device, and may also be a module (such as a chip) applied to the terminal device or the network device.
  • a communication device 700 includes a transceiver unit 701 and a processing unit 702 .
  • the processing unit 702 is configured to call the transceiver unit 701 to receive information from other communication devices or send information to other communication devices.
  • the transceiver unit 701 may further include a receiving unit and a sending unit, the receiving unit is used to receive information from other communication devices, and the sending unit is used to send information to other communication devices.
  • the communication device 700 is configured to realize the functions of the terminal device or the network device in the method embodiments shown in FIG. 3 , FIG. 4 , FIG. 5 , and FIG. 6 above.
  • the embodiment in FIG. 4 is based on the embodiment in FIG. 3
  • the embodiment in FIG. 6 is based on the embodiment in FIG.
  • the transceiver unit 701 is used to send AI capability information to the network device, the AI capability information includes the terminal device executing at least one reference AI unit The time and/or energy consumption of each reference AI unit; the processing unit 702 is configured to acquire the AI capability information.
  • the communication device 700 is used to realize the function of the network device in the method embodiment shown in FIG.
  • the transceiver unit 701 is used to receive the AI capability information sent by the terminal device, the AI capability information includes that the terminal device executes at least one reference AI unit The time and/or energy consumption of each reference AI unit; the transceiver unit 701 is also configured to send configuration information to the terminal device, where the configuration information instructs the terminal device to start the AI mode, or the configuration information indicates at least one AI model , or, the configuration information indicates configuration parameters of at least one AI model, or, the configuration information indicates an acquisition method of at least one AI model, wherein the at least one AI model is determined according to the AI capability information.
  • the transceiver unit 701 is used to receive the AI model information sent by the network device, the AI model information includes M groups corresponding to M AI models Similarity information, each set of similarity information in M groups of similarity information is similarity information between an AI model in M AI models and K reference AI units, where M and K are positive integers; the transceiver unit 701, It is also used to send feedback information to network devices based on AI model information.
  • the communication device 700 is used to implement the functions of the network device in the method embodiment shown in FIG.
  • the transceiver unit 701 is used to send AI model information to the terminal device, the AI model information includes M groups of similarity information corresponding to M AI models Each group of similarity information in M groups of similarity information is the similarity information between an AI model in M AI models and K reference AI units, wherein M and K are positive integers; the transceiver unit 701 also It is used to receive the feedback information sent by the terminal device, where the feedback information is determined according to the AI model information.
  • transceiver unit 701 and processing unit 702 can be directly obtained by referring to the related descriptions in the method embodiments shown in FIG. 3 and FIG. 5 , and will not be repeated here.
  • an embodiment of the present application further provides a communication device 800 .
  • the communication device 800 includes an interface circuit 801 and a processor 802 .
  • the interface circuit 801 and the processor 802 are coupled to each other.
  • the interface circuit 801 may be a transceiver or an input-output interface.
  • the communication device 800 may further include a memory 803 for storing instructions executed by the processor 802, or storing input data required by the processor 802 to execute the instructions, or storing data generated by the processor 802 after executing the instructions.
  • the processor 802 is used to implement the functions of the above-mentioned processing unit 702
  • the interface circuit 801 is used to implement the functions of the above-mentioned transceiver unit 701 .
  • 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 network device; or, the terminal device chip sends information to other modules in the terminal device (such as radio frequency modules or antenna) to send information, which is sent by the terminal device to the network device.
  • processor in the embodiments of the present application may be a central processing unit (central processing unit, CPU), and may also be other general processors, digital signal processors (digital signal processor, DSP), application specific integrated circuits (application specific integrated circuit, ASIC), field programmable gate array (field programmable gate array, FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof.
  • CPU central processing unit
  • DSP digital signal processor
  • ASIC application specific integrated circuit
  • FPGA field programmable gate array
  • a general-purpose processor can be a microprocessor, or any conventional processor.
  • the embodiments of the present application provide a communication system, and the communication system may include the terminal device and the network device involved in the above embodiments.
  • the embodiment of the present application also provides a computer-readable storage medium, where the computer-readable storage medium is used to store a computer program.
  • the computer program When the computer program is executed by a computer, the computer can implement the downlink scheduling method provided by the foregoing method embodiment.
  • the embodiment of the present application also provides a computer program product, the computer program product is used to store a computer program, and when the computer program is executed by a computer, the computer can realize the downlink scheduling method provided by the above method embodiment.
  • the embodiment of the present application further provides a chip, including a processor, the processor is coupled to a memory, and is configured to call a program in the memory so that the chip implements the downlink scheduling method provided by the above method embodiment.
  • the embodiment of the present application further provides a chip, the chip is coupled with a memory, and the chip is used to implement the downlink scheduling method provided in the above method embodiment.
  • the method steps in the embodiments of the present application may be implemented by means of hardware, or may be implemented by means of a processor executing software instructions.
  • Software instructions can be composed of corresponding software modules, and software modules can be stored in random access memory (random access memory, RAM), flash memory, read-only memory (read-only memory, ROM), programmable read-only memory (programmable ROM) , PROM), erasable programmable read-only memory (erasable PROM, EPROM), electrically erasable programmable read-only memory (electrically EPROM, EEPROM), register, hard disk, mobile hard disk, CD-ROM or known in the art any other form of storage medium.
  • An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium.
  • the storage medium may also be a component of the processor.
  • the processor and storage medium can be located in the ASIC.
  • the ASIC can be located in a network device or a terminal device.
  • the processor and the storage medium may also exist in the network device or the terminal device as discrete components.
  • all or part of them may be implemented by software, hardware, firmware or any combination thereof.
  • software When implemented using software, it may be implemented in whole or in part in the form of a computer program product.
  • the computer program product comprises one or more computer programs or instructions. When the computer program or instructions are loaded and executed on the computer, the processes or functions described in the embodiments of the present application are executed in whole or in part.
  • the computer may be a general purpose computer, a special purpose computer, a computer network, network equipment, user equipment, or other programmable devices.
  • the computer program or instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer program or instructions may be downloaded from a website, computer, The first device or data center transmits to another website site, computer, first device or data center in a wired or wireless manner.
  • the computer-readable storage medium may be any available medium that can be accessed by a computer, or a data storage device such as a first device or a data center integrating one or more available media. Described usable medium can be magnetic medium, for example, floppy disk, hard disk, magnetic tape; It can also be optical medium, for example, digital video disc (digital video disc, DVD); It can also be semiconductor medium, for example, solid state drive (solid state drive) , SSD).

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本申请提供一种人工智能AI通信方法，应用于通信系统中进行通信的任意两个设备，该方法包括：第一设备获取AI能力信息，其中，该AI能力信息包括第一设备执行至少一个参考AI单元中每个参考AI单元的时间和/或能耗；第一设备向第二设备发送该AI能力信息。进一步地，该方法还包括：第一设备接收第二设备发送的配置信息，其中，该配置信息指示第一设备启动AI通信模式，或，该配置信息指示至少一个AI模型，或，该配置信息指示至少一个AI模型的配置参数，或，该配置信息指示至少一个AI模型的获取方法。这样，第二设备能够通过第一设备上报的AI能力信息评估第一设备的AI能力和AI模型的匹配情况，可以提高第二设备评估结果的简便性、高效性和准确度。

Description

一种人工智能AI通信方法及装置
本申请要求在2021年9月16日提交中国国家知识产权局、申请号为202111085791.9的中国专利申请的优先权，发明名称为“一种AI能力表示方法和装置”的中国专利申请的优先权，在2021年10月26日提交中国国家知识产权局、申请号为202111250477.1的中国专利申请的优先权，发明名称为“一种人工智能AI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
本申请涉及通信技术领域，尤其涉及一种人工智能AI通信方法及装置。
背景技术
人工智能(artificial intelligence，AI)技术是计算机科学的一个分支，贯穿在计算机发展的历史过程中，是信息科技业界的一个重要的发展方向。随着通信技术的发展，越来越多的应用将通过AI实现智能化。目前，无线通信系统引入AI技术，可能会逐步使用AI模块代替无线通信系统中的功能模块。无线通信系统引入AI技术后，一种可能的工作模式是，网络设备将AI模型发送给终端设备，终端设备从网络设备接收AI模型并应用AI模型进行无线通信。
不同的终端设备的AI能力存在差异，因此，网络设备下发的AI模型可能无法被终端设备执行，这样就会导致无法应用AI模型进行无线通信。
发明内容
本申请提供一种人工智能AI通信方法及装置，以期更好地在无线通信系统中应用AI技术。
第一方面，本申请提供了一种AI通信方法，该方法可以由第一设备执行，其中，第一设备可以是终端设备，或者通信系统中AI模型的接收方。该方法可以通过以下步骤实现：第一设备获取AI能力信息，该AI能力信息包括第一设备执行至少一个参考AI单元中每个参考AI单元的时间和/或能耗；第一设备向第二设备发送该AI能力信息。
本实现方式中，通过第一设备向第二设备发送其AI能力信息，第二设备能够对AI模型和第一设备的AI能力的匹配情况进行评估，保证第一设备和第二设备使用AI模型进行通信的可行性。
在一个可能的设计中，第一设备向第二设备发送AI能力信息之后，该方法还包括：第一设备接收第二设备发送的配置信息，其中，该配置信息指示第一设备启动AI通信模式，或，该配置信息指示至少一个AI模型，或，该配置信息指示至少一个AI模型的配置参数，或，该配置信息指示至少一个AI模型的获取方法。
在一个可能的设计中，至少一个AI模型是根据AI能力信息确定的。
在一个可能的设计中，配置信息是AI能力信息的响应信息。
在一个可能的设计中，第一设备执行至少一个参考AI单元中第一AI参考单元的时间为第一时间数值、第一时间等级或第一时间范围，第一设备执行至少一个参考AI单元中第一AI参考单元的能耗为第一能耗数值、第一能耗等级或第一能耗范围。
在一个可能的设计中，AI能力信息包括第一设备执行至少一个参考AI单元中每个参考AI单元使用的输入数据的数量。
在一个可能的设计中，AI能力信息包括以下一种或多种：第一设备执行至少一个参考AI单元中每个参考AI单元使用的输入数据的数值精度；至少一个参考AI单元中每个参考AI单元的权值的数值精度；第一设备执行至少一个参考AI单元中每个参考AI单元时的运算精度。
在一个可能的设计中，第一设备执行至少一个参考AI单元中每个参考AI单元使用的资源为所述第一设备所有可用的计算资源。基于上述AI能力信息，第二设备能够更简便、高效和准确地评估第一设备的AI能力和AI模型复杂度的匹配情况，进而更好地进行AI通信。
在一个可能的设计中，AI能力信息包括第一设备用于执行AI模型的时间上限值，能耗上限值，和资源使用情况中的至少一种。这样，第一设备可以向第二设备发送其用于执行AI模型的时间上限值信息、能耗上限值信息、和资源使用情况信息中的至少一种，以告知第二设备其执行AI模型的时间预算要求、能量消耗限制、或资源使用情况。
在一个可能的设计中，第一设备向第二设备发送AI能力信息之前，该方法还包括：第一设备接收来自第二设备的请求信息，该请求信息用于请求第一设备向第二设备发送上述AI能力信息。
在一个可能的设计中，第一设备向第二设备发送AI能力信息，包括：第一设备周期性向第二设备发送AI能力信息；或第一设备接入第二设备所在的网络时，第一设备向第二设备发送AI能力信息；或第一设备与第二设备建立通信连接时，第一设备向第二设备发送AI能力信息；或第一设备用于执行AI模型的计算资源改变时，第一设备向第二设备发送AI能力信息。
第二方面，本申请提供了一种AI通信方法，该方法可以由第二设备执行，其中，第二设备可以是网络设备，或者通信系统中AI模型的发送方。该方法可以通过以下步骤实现：第二设备接收第一设备发送的AI能力信息，该AI能力信息包括第一设备执行至少一个参考AI单元的时间和/或能耗。
在一个可能的设计中，第二设备接收第一设备发送的AI能力信息之后，该方法还包括：第二设备向第一设备发送配置信息，其中，该配置信息指示第一设备启动AI通信模式，或，该配置信息指示至少一个AI模型，或，该配置信息指示至少一个AI模型的配置参数，或，该配置信息指示至少一个AI模型的获取方法，其中，该至少一个AI模型是根据上述AI能力信息确定的。
在一个可能的设计中，至少一个AI模型是根据AI能力信息确定的。
在一个可能的设计中，配置信息是AI能力信息的响应信息。
在一个可能的设计中，第一设备执行至少一个参考AI单元中第一AI参考单元的时间为第一时间数值、第一时间等级或第一时间范围，第一设备执行至少一个参考AI单元中第 一AI参考单元的能耗为第一能耗数值、第一能耗等级或第一能耗范围。
在一个可能的设计中，AI能力信息包括第一设备执行至少一个参考AI单元中每个参考AI单元使用的输入数据的数量。
在一个可能的设计中，AI能力信息包括以下一种或多种：第一设备执行至少一个参考AI单元中每个参考AI单元使用的输入数据的数值精度；至少一个参考AI单元中每个参考AI单元的权值的数值精度；第一设备执行至少一个参考AI单元中每个参考AI单元时的运算精度。
在一个可能的设计中，AI能力信息包括第一设备用于执行AI模型的时间上限值，能耗上限值，和资源使用情况中的至少一种。
在一个可能的设计中，第二设备接收第一设备发送的AI能力信息之前，该方法还包括：第二设备向第一设备发送请求信息，该请求信息用于请求第一设备向第二设备发送上述AI能力信息。
在一个可能的设计中，配置信息指示至少一个AI模型，或配置指示至少一个AI模型的配置参数，或配置信息指示至少一个AI模型的获取方法时，第二设备向第一设备发送配置信息之前，该方法还包括：第二设备根据上述AI能力信息确定至少一个AI模型。
在一个可能的设计中，第二设备根据AI能力信息确定至少一个AI模型包括：，第二设备根据AI能力信息和M个AI模型对应的M组相似度信息确定至少一个AI模型，其中，M组相似度信息中的每组相似度信息为M个AI模型中的一个AI模型与K个参考AI单元的相似度信息。其中，M和K为正整数。
在一个可能的设计中，M个AI模型为第二设备或其他设备中预先存储的；或M个AI模型为第二设备从其他设备获取的；或M个AI模型为第二设备生成的。
在一个可能的设计中，M组相似度信息中的第一组相似度信息为第一AI模型和K个参考AI单元的相似度信息，第一组相似度信息包括第一相似度信息，第一相似度信息与第一比例和/或第二比例相关联，其中，第一比例为第一AI模型中与第一参考AI单元相同的层的总数量占第一AI模型总层数的比例，第二比例为第一AI模型中与第一参考AI单元相同的层的总计算量占第一AI模型总计算量的比例，M个AI模型包括第一AI模型，K个参考AI单元包括第一参考AI单元。
在另一个可能的设计中，M组相似度信息中的第一组相似度信息为第一AI模型和K个参考AI单元的相似度信息，第一组相似度信息包括第一相似度信息，第一相似度信息与第一比例和/或第二比例相关联，其中，第一比例为第一AI模型中与第一参考AI单元相同的单元的总层数占第一AI模型总层数的比例，第二比例为第一AI模型中与第一参考AI单元相同的单元的总计算量占第一AI模型总计算量的比例，M个AI模型包括第一AI模型，K个参考AI单元包括第一参考AI单元。
第二方面及各个可能的设计的有益效果可以参考第一方面相关的描述，在此不予赘述。
第三方面，本申请提供了一种AI通信方法，该方法可以由第一设备执行，其中，第一设备可以是终端设备，或者通信系统中AI模型的接收方。该方法可以通过以下步骤实现：第一设备接收第二设备发送的AI模型信息，该AI模型信息包括M个AI模型对应的M组相似度信息，M组相似度信息中的每组相似度信息为M个AI模型中的一个AI模型与K个参考AI单元的相似度信息，其中，M和K为正整数；第一设备根据该AI模型信息向第二 设备发送反馈信息。
本实现方式中，第二设备向第一设备发送AI模型信息，第一设备能够对AI模型和第一设备的AI能力的匹配情况进行评估，保证第一设备和第二设备使用AI模型进行通信的可行性。
在一个可能的设计中，M组相似度信息中的第一组相似度信息为第一AI模型和K个参考AI单元的相似度信息，第一组相似度信息包括第一相似度信息，第一相似度信息与第一比例和/或第二比例相关联，其中，第一比例为第一AI模型中与第一参考AI单元相同的层的总数量占第一AI模型总层数的比例，第二比例为第一AI模型中与第一参考AI单元相同的层的总计算量占第一AI模型总计算量的比例，M个AI模型包括第一AI模型，K个参考AI单元包括第一参考AI单元。
在另一个可能的设计中，M组相似度信息中的第一组相似度信息为第一AI模型和K个参考AI单元的相似度信息，第一组相似度信息包括第一相似度信息，第一相似度信息与第一比例和/或第二比例相关联，其中，第一比例为第一AI模型中与第一参考AI单元相同的单元的总层数占第一AI模型总层数的比例，第二比例为第一AI模型中与第一参考AI单元相同的单元的总计算量占第一AI模型总计算量的比例，M个AI模型包括第一AI模型，K个参考AI单元包括第一参考AI单元。
在一个可能的设计中，AI模型信息包括以下一种或多种：执行M个AI模型中每个AI模型时使用的输入数据的数值精度；M个AI模型中每个AI模型的权值的数值精度；执行M个AI模型中每个AI模型时的运算精度。
在一个可能的设计中，AI模型信息包括M个AI模型中每个AI模型的总层数和/或总计算量。
基于上述AI模型信息，第一设备能够更简便、高效和准确地评估的第一设备的AI能力和AI模型复杂度的匹配情况，进而更好地进行AI通信。
在一个可能的设计中，AI模型信息包括执行AI模型的时间上限值。这样，第一设备能够获取第二设备对第一设备执行待下发AI模型的时间预算要求，以便更好地确定是否启动AI通信模式或选择合适的AI模型。
在一个可能的设计中，上述反馈信息指示第一设备请求启动AI模式，或，反馈信息指示M个AI模型中至少一个AI模型的评估结果，或，反馈信息请求第二设备向第一设备发送至少一个AI模型，其中，M个AI模型包括至少一个AI模型。
在一个可能的设计中，第一设备根据AI模型信息向第二设备发送反馈信息之后，该方法还包括：第一设备接收第二设备发送的配置信息，该配置信息指示第一设备启动AI模式，或，该配置信息指示至少一个AI模型，或，该配置信息指示至少一个AI模型的配置参数，或，该配置信息指示至少一个模型的获取方法，其中，上述M个AI模型包括该至少一个AI模型。
在一个可能的设计中，第一设备接收第二设备发送的AI模型信息之前，该方法还包括：第一设备向第二设备发送请求信息，该请求信息用于请求第二设备向第一设备发送上述AI模型信息。
在一个可能的设计中，第一设备基于AI模型信息向第二设备发送反馈信息之前，该方法还包括：第一设备根据AI模型信息和第一设备的AI能力信息确定反馈信息。
在一个可能的设计中，第一设备的AI能力信息指示所述第一设备执行至少一个参考AI单元的时间和/或能耗。关于AI能力信息的可能的设计可以参考第一方面相关的描述，在此不予赘述。
第四方面，本申请提供了一种AI通信方法，该方法可以由第二设备执行，其中，第二设备可以是网络设备，或者通信系统中AI模型的发送方。该方法可以通过以下步骤实现：第二设备向第一设备发送AI模型信息，该AI模型信息包括M个AI模型对应的M组相似度信息，M组相似度信息为M个AI模型中的一个AI模型与K个参考AI单元的相似度信息，其中，所述M和K为正整数；第二设备接收第一设备发送的反馈信息，该反馈信息是根据AI模型信息确定的。
在一个可能的设计中，M组相似度信息中的第一组相似度信息为第一AI模型和K个参考AI单元的相似度信息，第一组相似度信息包括第一相似度信息，第一相似度信息与第一比例和/或第二比例相关联，其中，第一比例为第一AI模型中与第一参考AI单元相同的层的总数量占第一AI模型总层数的比例，第二比例为第一AI模型中与第一参考AI单元相同的层的总计算量占第一AI模型总计算量的比例，M个AI模型包括第一AI模型，K个参考AI单元包括第一参考AI单元。
在另一个可能的设计中，M组相似度信息中的第一组相似度信息为第一AI模型和K个参考AI单元的相似度信息，第一组相似度信息包括第一相似度信息，第一相似度信息与第一比例和/或第二比例相关联，其中，第一比例为第一AI模型中与第一参考AI单元相同的单元的总层数占第一AI模型总层数的比例，第二比例为第一AI模型中与第一参考AI单元相同的单元的总计算量占第一AI模型总计算量的比例，M个AI模型包括第一AI模型，K个参考AI单元包括第一参考AI单元。
在一个可能的设计中，AI模型信息包括以下一种或多种：执行M个AI模型中每个AI模型时使用的输入数据的数值精度；M个AI模型中每个AI模型的权值的数值精度；执行M个AI模型中每个AI模型时的运算精度。
在一个可能的设计中，AI模型信息包括M个AI模型中每个AI模型的总层数和/或总计算量。
在一个可能的设计中，AI模型信息包括执行AI模型的时间上限值。
在一个可能的设计中，上述反馈信息指示第一设备请求启动AI模式，或，反馈信息指示M个AI模型中至少一个AI模型的评估结果，或，反馈信息请求第二设备向第一设备发送至少一个AI模型，其中，M个AI模型包括至少一个AI模型。
在一个可能的设计中，第二设备接收第一设备发送的反馈信息之后，该方法还包括：第二设备向第一设备发送配置信息，该配置信息指示第一设备启动AI模式，或，该配置信息指示至少一个AI模型，或，该配置信息指示至少一个AI模型的配置参数，或，该配置信息指示至少一个AI模型的获取方法，其中，M个AI模型包括该至少一个AI模型。
在一个可能的设计中，第二设备向第一设备发送AI模型信息之前，该方法还包括：第二设备接收第一设备发送的请求信息，该请求信息用于请求第二设备向第一设备发送AI模型信息。
在一个可能的设计，上述M个AI模型为第二设备或其他设备中预先存储的，或，上述M个AI模型为第二设备从其他设备获取的，或，上述M个AI模型为所述第二设备生成的。
第四方面及各个可能的设计的有益效果可以参考第三方面相关的描述，在此不予赘述。
第五方面，本申请还提供了一种通信装置，所述通信装置可以是终端设备，或者所述通信装置可以是通信系统中的接收端设备，该通信装置具有实现上述第一方面或第三方面中任一方面的第一设备的功能。所述功能可以通过硬件实现，也可以通过硬件执行相应的软件实现。所述硬件或软件包括一个或多个与上述功能相对应的模块。
在一个可能的设计中，所述通信装置的结构中包括收发单元和处理单元，这些单元可以执行上述第一方面或第三方面中任一方面的第一设备的相应功能，具体参见方法示例中的详细描述，此处不做赘述。
在一个可能的设计中，所述通信装置的结构中包括收发器和处理器，可选的还包括存储器，所述收发器用于收发数据，以及用于与通信系统中的其他设备进行通信交互，所述处理器被配置为支持所述通信装置执行上述第一方面或第三方面中任一方面的第一设备的相应的功能。所述存储器与所述处理器耦合，其保存所述通信装置必要的程序指令和数据。
第六方面，本申请还提供了一种通信装置，所述通信装置可以是网络设备，或者所述通信装置可以是通信系统中的发送端设备，该通信装置具有实现上述第二方面或第四方面中任一方面的第二设备的功能。所述功能可以通过硬件实现，也可以通过硬件执行相应的软件实现。所述硬件或软件包括一个或多个与上述功能相对应的模块。
在一个可能的设计中，所述通信装置的结构中包括收发单元和处理单元，这些单元可以执行上述第二方面或第四方面中任一方面的第二设备的相应功能，具体参见方法示例中的详细描述，此处不做赘述。
在一个可能的设计中，所述通信装置的结构中包括收发器和处理器，可选的还包括存储器，所述收发器用于收发数据，以及用于与通信系统中的其他设备进行通信交互，所述处理器被配置为支持所述通信装置执行上述第二方面或第四方面中任一方面的第二设备的相应的功能。所述存储器与所述处理器耦合，其保存所述通信装置必要的程序指令和数据。
第七方面，本申请实施例提供了一种通信系统，可以包括上述提及的第一设备和第二设备。
第八方面，本申请实施例提供的一种计算机可读存储介质，该计算机可读存储介质存储有程序指令，当程序指令在计算机上运行时，使得计算机执行上述第一方面至第四方面中任一方面及其任一可能的设计中的方法。示例性的，计算机可读存储介质可以是计算机能够存取的任何可用介质。以此为例但不限于：计算机可读介质可以包括非瞬态计算机可读介质、随机存取存储器(random-access memory，RAM)、只读存储器(read-only memory，ROM)、电可擦除可编程只读存储器(electrically EPROM，EEPROM)、CD-ROM或其他光盘存储、磁盘存储介质或者其他磁存储设备、或者能够用于携带或存储具有指令或数据结构形式的期望的程序代码并能够由计算机存取的任何其他介质。
第九方面，本申请实施例提供一种包括计算机程序代码或指令的计算机程序产品，当其在计算机上运行时，使得计算机实现上述第一方面至第四方面中任一方面及其任一可能的设计中的方法。
第十方面，本申请还提供了一种芯片，包括处理器，所述处理器与存储器耦合，用于读取并执行所述存储器中存储的程序指令，以使所述芯片实现上述第一方面至第四方面中任一方面及其任一可能的设计中的方法。
上述第五方面至第十方面中的各个方面以及各个方面可能达到的技术效果请参照上述针对第一方面至第四方面中的各种可能方案可以达到的技术效果说明，这里不再重复赘述。
附图说明
图1为本申请提供的一种通信系统的架构示意图；
图2为本申请实施例中网络设备与终端设备利用AI模型进行无线通信的过程示意图；
图3为本申请实施例中AI通信方法流程示意图之一；
图4为本申请实施例中网络设备进行AI能力和AI模型匹配度评估的流程示意图；
图5为本申请实施例中AI通信方法流程示意图之二；
图6为本申请实施例中终端设备进行AI能力和AI模型匹配度评估的流程示意图；
图7为本申请实施例中通信装置结构示意图之一；
图8为本申请实施例中通信装置结构示意图之二。
具体实施方式
下面将结合附图，对本申请中的技术方案作进一步地详细描述。
本申请提供的技术方案可以应用于各种通信系统，例如：第五代(5th generation，5G)或新无线(new radio，NR)系统、长期演进(long term evolution，LTE)系统、LTE频分双工(frequency division duplex，FDD)系统、LTE时分双工(time division duplex，TDD)系统等。本申请提供的技术方案还可以应用于未来的通信系统，如第六代移动通信系统。本申请提供的技术方案还可以应用于设备到设备(device to device，D2D)通信，车到万物(vehicle-to-everything，V2X)通信，机器到机器(machine to machine，M2M)通信，机器类型通信(machine type communication，MTC)，以及物联网(internet of things，IoT)通信系统或者其他通信系统。
本申请实施例提供一种人工智能AI通信方法及装置。其中，本申请所述方法和装置基于同一技术构思，由于方法及装置解决问题的原理相似，因此装置与方法的实施可以相互参见，重复之处不再赘述。
在本申请的描述中，“第一”、“第二”等词汇，仅用于区分描述的目的，而不能理解为指示或暗示相对重要性，也不能理解为指示或暗示顺序。
在本申请的描述中，“至少一个(种)”是指一个(种)或者多个(种)，多个(种)是指两个(种)或者两个(种)以上。
在本申请的描述中，除非另有说明，“/”表示或的意思，例如，A/B可以表示A或B；本文中的“和/或”仅仅是一种描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。另外，在本申请实施例的描述中，“多个”是指两个或多于两个。为了更加清晰地描述本申请实施例的技术方案，下面结合附图，对本申请实施例提供的下行调度方法及装置进行详细说明。
图1示出了适用于本申请实施例的无线通信系统的示意图。如图1所示，该无线通信系统可以包括至少一个网络设备，例如图1所示的网络设备111、网络设备112，该无线通信系统还可以包括至少一个终端设备，例如图1所示的终端设备121、终端设备122、终端设备123、终端设备124、终端设备125、终端设备126、终端设备127。网络设备和终端设 备均可配置多个天线，网络设备与终端设备可使用多天线技术通信。
其中，网络设备和终端设备通信时，网络设备可以管理一个或多个小区，一个小区中可以有整数个终端设备。需要说明的是，小区可以理解为网络设备的无线信号覆盖范围内的区域。
本申请可以用于网络设备和终端设备之间通信的场景，如网络设备111与终端设备121、终端设备122、终端设备123之间可以通信；又如网络设备111和网络设备112可以与终端设备124通信。本申请还可以用于终端设备和终端设备之间通信的场景，如终端设备122可以与终端设备125通信。本申请还可以用于网络设备与网络设备之间通信的场景，如网络设备111可以与网络设备112通信。
应理解，图1仅为便于理解而示例的简化示意图，本申请并未限定于此。本申请实施例可以适用于发送端设备和接收端设备通信的任何通信场景。
本申请实施例中的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、终端、无线通信设备、用户代理或用户装置。
终端设备可以是一种向用户提供语音/数据的设备，例如，具有无线连接功能的手持式设备、车载设备等。目前，一些终端的举例为：手机(mobile phone)、平板电脑、笔记本电脑、掌上电脑、移动互联网设备(mobile internet device，MID)、可穿戴设备，虚拟现实(virtual reality，VR)设备、增强现实(augmented reality，AR)设备、工业控制(industrial control)中的无线终端、无人驾驶(self-driving)中的无线终端、远程手术(remote medical surgery)中的无线终端、智能电网(smart grid)中的无线终端、运输安全(transportation safety)中的无线终端、智慧城市(smart city)中的无线终端、智慧家庭(smart home)中的无线终端、蜂窝电话、无绳电话、会话启动协议(session initiation protocol，SIP)电话、无线本地环路(wireless local loop，WLL)站、个人数字助理(personal digital assistant，PDA)、具有无线通信功能的手持设备、计算设备或连接到无线调制解调器的其它处理设备、可穿戴设备，5G网络中的终端设备或者未来演进的公用陆地移动通信网络(public land mobile network，PLMN)中的终端设备等，本申请实施例对此并不限定。
作为示例而非限定，在本申请实施例中，该终端设备还可以是可穿戴设备。可穿戴设备也可以称为穿戴式智能设备，是应用穿戴式技术对日常穿戴进行智能化设计、开发出可以穿戴的设备的总称，如眼镜、手套、手表、服饰及鞋等。可穿戴设备即直接穿在身上，或是整合到用户的衣服或配件的一种便携式设备。可穿戴设备不仅仅是一种硬件设备，更是通过软件支持以及数据交互、云端交互来实现强大的功能。广义穿戴式智能设备包括功能全、尺寸大、可不依赖智能手机实现完整或者部分的功能，例如：智能手表或智能眼镜等，以及只专注于某一类应用功能，需要和其它设备如智能手机配合使用，如各类进行体征监测的智能手环、智能首饰等。
本申请实施例中，用于实现终端设备的功能的装置可以是终端设备，也可以是能够支持终端设备实现该功能的装置，例如芯片系统或芯片，该装置可以被安装在终端设备中。本申请实施例中，芯片系统可以由芯片构成，也可以包括芯片和其他分立器件。
本申请实施例中的网络设备可以是用于与终端设备通信的设备，该网络设备也可以称为接入网设备或无线接入网设备，如网络设备可以是基站。本申请实施例中的网络设备可 以是指将终端设备接入到无线网络的无线接入网(radio access network，RAN)节点(或设备)。基站可以广义的覆盖如下中的各种名称，或与如下名称进行替换，比如：节点B(NodeB)、演进型基站(evolved NodeB，eNB)、下一代基站(next generation NodeB，gNB)、中继站、接入点、传输点(transmitting and receiving point，TRP)、发射点(transmitting point，TP)、主站、辅站、多制式无线(motor slide retainer，MSR)节点、家庭基站、网络控制器、接入节点、无线节点、接入点(AP)、传输节点、收发节点、基带单元(BBU)、射频拉远单元(remote radio unit，RRU)、有源天线单元(active antenna unit，AAU)、射频头(remote radio head，RRH)、中心单元(central unit，CU)、分布式单元(distributed unit，DU)、定位节点等。基站可以是宏基站、微基站、中继节点、施主节点或类似物，或其组合。基站还可以指用于设置于前述设备或装置内的通信模块、调制解调器或芯片。基站还可以是移动交换中心以及D2D、V2X、M2M通信中承担基站功能的设备、6G网络中的网络侧设备、未来的通信系统中承担基站功能的设备等。基站可以支持相同或不同接入技术的网络。本申请的实施例对网络设备所采用的具体技术和具体设备形态不做限定。
基站可以是固定的，也可以是移动的。例如，直升机或无人机可以被配置成充当移动基站，一个或多个小区可以根据该移动基站的位置移动。在其他示例中，直升机或无人机可以被配置成用作与另一基站通信的设备。
在一些部署中，本申请实施例所提及的网络设备可以为包括CU、或DU、或包括CU和DU的设备、或者控制面CU节点(控制面的中央单元(central unit-control plane，CU-CP))和用户面CU节点(用户面的中央单元(central unit-user plane，CU-UP))以及DU节点的设备。
网络设备和终端设备可以部署在陆地上，包括室内或室外、手持或车载；也可以部署在水面上；还可以部署在空中的飞机、气球和卫星上。本申请实施例中对网络设备和终端设备所处的场景不做限定。
人工智能AI技术可以与无线空口相结合以提升无线网络性能。例如，基于AI的信道估计和信号检测。其中，信号检测可以是从无线信道中把所接收到的含干扰噪声的信号提取出来的过程；信道估计是从接收到的信号中将假定的某个信道模型的模型参数估计出来的过程。又例如，基于AI的端到端的通信链路设计。再例如，基于AI的信道状态信息(channel state information，CSI)反馈方案，即通过神经网络对CSI进行编码并反馈给网络设备。
以基于AI的信道状态信息(channel state information，CSI)反馈方案为例，介绍网络设备与终端设备利用AI模型进行无线通信的过程。如图2中(a)所示，网络设备中可以部署有AI编码器和对应的AI解码器。网络设备可以向终端设备发送或下发AI编码器，或指示终端设备获取AI编码器，以便后续终端设备使用AI编码器进行CSI编码。如图2中(b)所示，终端设备可以使用该AI编码器对CSI进行编码，并将编码后的信息发送给网络设备。然后，网络设备可以使用AI解码器对终端设备编码后的信息进行解码，以得到恢复后的信息。需要说明的是，AI编码器也可以理解为用于信息编码的AI模型，AI解码器也可以理解为用于信息解码的AI模型。
AI模型多种多样，不同的应用场景中，可以采用不同的AI模型。常见的，AI模型可以基于神经网络(neural network)模型实现。神经网络模型是一种模仿人脑神经网络的行为特征，进行分布式并行信息处理的数学计算模型。一些复杂的神经网络模型 可能包含的参数量或计算量很大，而终端设备的能力(例如计算能力、存储能力或能量)可能受限。因此，网络设备和终端设备之间进行AI通信之前，需要确保终端设备的AI能力能够支持执行(或称：运行、处理)网络设备发送的AI模型，例如，终端设备的存储能力能够容纳该AI模型、终端设备的计算能力能够支持该AI模型在要求的时间内完成计算、终端设备执行该AI模型的运行功耗(或称，能量功耗)在预期的可接受范围内。
通常，AI模型计算时延上限t i(即AI模型推理完成时间上限)是已知的，因此，一种方案中，可以通过比较终端设备的计算能力C UE和AI模型的计算复杂度C M来评估AI模型和终端设备AI能力的匹配情况，当C M/C UE+t th≤t i时，表示AI模型可以在要求的时延内完成计算，即AI模型复杂度和终端设备的计算能力相匹配，否则认为AI模型复杂度和终端设备的计算能力不匹配。其中，终端设备的计算能力C UE可以以每秒浮点运算次数(floating-point operations per second，FLOPS)为单位，AI模型的计算复杂度C M可以以浮点运算数(floating-point operations，FLOP)为单位，t th为可以提前配置的裕量。
然而，相同用途的AI模型的内部结构可能差异较大，在实际计算时，因为硬件计算率、数据调度时延等不同，不同结构的AI模型的计算效率可能大相径庭。另外，由于AI模型到硬件计算资源之间，需要大量的软件转换和优化，不同的软件转换和优化方法同样带来了不同的计算效率。除此之外，终端设备不同的AI计算硬件实现也可能会带来不同的AI模型计算效率。因此，上述AI模型结构设计、软件环境、硬件实现等因素均可能导致上百倍的计算效率偏差，仅通过配置裕量t th可能难以评估AI模型和终端设备AI能力的实际匹配情况。例如，t th过大可能浪费算力资源，t th过小则可能导致无法在时延上限内完成AI模型的执行；而且AI模型结构、软件环境、硬件实现等不同时，t th取值可能也是不同的。
另一种方案中，可以在匹配服务器上配置终端硬件信息，例如终端设备的模拟器。在获取到AI模型的详细信息和计算时延之后，匹配服务器可以获得精准的AI模型复杂度和终端设备AI能力的匹配情况的评估结果。然而，该方案需要引入匹配服务器和相应的交互协议，使得网络结构和交互流程复杂，同时增加了成本和评估时延。
基于此，本申请提出了一种AI通信方法及装置，可以实现AI模型复杂度和设备AI的匹配情况的评估，保证使用AI模型进行通信业务的可行性。
需要说明的是，本申请可以应用于通信系统中任意两个设备(第一设备和第二设备)进行通信的场景。其中，第一设备中可以有执行AI模型的计算硬件、并准备从第二设备中获取AI模型并使用，第二设备中可以有AI模型、并准备发送给第二设备使用。例如，第一设备为图1中的终端设备121、终端设备122、或终端设备123，第二设备为图1中的网络设备111；又例如，第一设备为图1中的终端设备122，第二设备为图1中的终端设备125；再例如，第一设备为图1中的网络设备111，第二设备为图1中的网络设备112。
下文示出的实施例中，仅为便于理解和说明，以网络设备与终端设备之间的交互为例，详细说明本申请实施例提供的方法。
下面结合附图详细介绍本申请实施例提供的方法及相关装置。需要说明的是，本申请实施例的展示顺序仅代表实施例的先后顺序，并不代表实施例所提供的技术方案的优劣。
基于以上描述，本申请实施例提供了一种AI通信方法，适用于图1所示的通信系统。 如图3所示，该方法可以包括但不限于以下步骤：
S301：终端设备获取AI能力信息，该AI能力信息包括终端设备执行至少一个参考AI单元中每个参考AI单元的时间和/或能耗。
例如，当至少一个参考AI单元的数量为3(参考AI单元P1、参考AI单元P2、参考AI单元P3)时，该AI能力信息包括：终端设备执行参考AI单元P1的时间和/或能耗，终端设备执行参考AI单元P2的时间和/或能耗，以及终端设备执行参考AI单元P3的时间和/或能耗。
具体地，该参考AI单元信息可以是提前标准化的或者离线指定的，并向终端设备公开。
其中，终端设备执行参考AI单元，也可以理解为终端设备对参考AI单元执行推理或训练操作。示例性地，参考AI单元可以是参考AI模型、参考AI模块、和参考AI算子中的至少一种。例如，AI模型可以为神经网络模型，AI模块可以为神经网络模块，AI算子可以为神经网络算子。其中，常见的神经网络模型例如ResNet系列、MobileNet系列、Transformer系列等，神经网络模型包括一层层算子(神经网络算子)，而当前的神经网络模型(典型的如ResNet系列模型)常常是模块化设计，即神经网络模型包括一系列模块(神经网络模块)及神经网络算子。可以认为，AI模块包括AI算子，而AI模型包括AI模块和/或AI算子。
可选地，终端设备执行每个参考单元的时间和/或能耗可以使用实际数值、近似数值、等级(例如：低/中/高、或1/2/3/4)、或者范围表示。具体地，上述至少一个参考AI单元可以包括第一AI参考单元，终端设备执行该第一AI参考单元的时间可以为第一时间数值、第一时间等级或第一时间范围，终端设备执行该第一AI参考单元的能耗可以为第一能耗数值、第一能耗等级或第一能耗范围。
一种实现方式中，终端设备可以自己执行上述至少一个参考AI单元，并获取其AI能力信息。另一种实现方式中，上述执行至少一个AI参考单元的过程可以在与该终端设备AI能力相同或相近的设备、或能够真实体现该终端设备AI能力的模拟器中完成。具体地，在与终端设备AI能力相同或相近的设备或模拟器中得到的计算结果的误差可以不超过预期值。此时，终端设备可以从其他设备或模拟器中获取该AI能力信息。又一种实现方式中，终端设备执行至少一个AI参考单元的时间和/或能耗信息可以是出厂设置的。
S302：终端设备向网络设备发送AI能力信息。相应地，网络设备接收终端设备发送的AI能力信息。
可选地，终端设备发送的AI能力信息可以包括终端设备执行至少一个参考AI单元使用的输入数据的数量。其中，输入数据的数量也可以称为样本数量。通常，终端设备进行模型训练时，一次执行一批(batch)样本，样本批量大小决定一次训练的样本数目。使用不同的数量的输入数据时，终端设备会得到不同的计算效率。
可选地，终端设备发送的AI能力信息还可以包括以下至少一种：终端设备(或与终端设备AI能力相同的其他设备、或模拟器)执行至少一个参考AI单元中每个参考AI单元使用的输入数据的数值精度；该至少一个参考AI单元中每个参考AI单元的权值(或称权重、系数)的数值精度；终端设备(或与终端设备AI能力相同的其他设备、或模拟器)执行至少一个参考AI单元中每个参考AI单元时的运算精度。
神经网络可以是由神经单元组成的，神经单元可以是指以x s和截距1为输入数据的运算单元，该运算单元的输出可以为：
Figure PCTCN2022117147-appb-000001
其中，s＝1、2、……n，n为自然数，W s即为x s的权重，b为神经单元的偏置(也可以看作是一种权重)。输入数据和权值的数值精度会影响终端设备的计算效率。其中，数值精度例如可以是int(整型)8、int16、float(浮点型)16等。数值精度也可以等效地表示为其中的乘法和加法运算精度，运算精度同样会影响终端设备的计算效率，运算精度例如可以是int8、int16、float16等
一种可能的实现方式中，终端设备执行上述至少一个参考AI单元中每个参考AI单元使用的资源为该终端设备所有可用的计算资源。终端设备的运行硬件可能有多种类型，例如，中央控制单元(central processing unit，CPU)、微型处理器(graphics processing unit，GPU)、嵌入式神经网络处理器(neural-network processing unit，NPU)、或现场可编程门阵列(field programmable gate array，FPGA)等，如果终端设备支持异构计算，运行硬件还可能是上述多种类型的组合，其中，异构计算指的是AI模型分布在多种类型的计算单元上执行。例如，AI模型分布在CPU、GPU和FPGA三种的计算单元上执行，进而，终端设备可用的计算资源可以包括CPU资源、GPU资源和FPGA资源中的一种或多种。
另一种可能的实现方式中，终端设备执行上述至少一个参考AI单元中每个参考AI单元使用的资源为该终端设备所有可用的计算资源的一部分。可以理解，终端设备执行参考AI单元使用的计算资源的比例会影响其AI能力信息(即终端设备执行至少一个AI参考单元的时间和/或能耗信息)。
本实现方式中，方式1：终端设备可以根据使用的计算资源占所有可用的计算资源的比例，对计算出的执行上述至少一个参考AI单元中每个参考AI单元的完成时间或消耗的能量进行换算，得到最终上报的AI能力信息。方式2：终端设备可以将使用的计算资源占所有可用的计算资源的比例信息上报给网络设备，进而，网络设备可以根据该资源比例信息对终端设备上报的AI能力信息进行换算。可选地，终端设备发送的AI能力信息可以包括(或指示)该终端设备执行至少一个参考AI单元中每个参考AI单元使用的计算资源占该终端设备所有可用的计算资源的比例，或者，终端设备可以在其他信令中单独向网络设备发送该资源比例信息。
可选地，终端设备执行上述至少一个参考AI单元的过程可以在终端设备出厂时完成。
需要说明的是，上述AI能力信息中包括的信息可以在同一信令中，也可以包括在不同信令中，其可以称为终端设备的专有能力信息。
当上述AI能力信息中包括的信息在同一信令中时，终端设备可以一次性将上述AI能力信息上报给网络设备，例如，终端设备向网络设备发送第一信息，该第一信息可以包括或指示上述AI能力信息，或该第一信息为上述AI能力信息。
当上述AI能力信息中包括的信息在不同信令中时，终端设备可以一次或多次将AI能力信息上报给网络设备。示例一：终端设备一次或多次将执行AI单元的时间和/或能耗、输入数据的数量、输入数据的精度等不同能力上报给网络设备。例如，终端设备向网络设备发送第一信息，该第一信息包括或指示终端设备执行至少一个参考AI单元中 每个参考AI单元的时间和/或能耗，终端设备向网络设备发送第二信息，该第二信息包括或指示执行至少一个参考AI单元中每个参考AI单元使用的输入数据的数量，上述AI能力信息可以包括该第一信息和第二信息。示例二：终端设备一次或多次将上述AI能力信息上报给网络设备。例如，终端设备向网络设备发送第一信息，该第一信息包括或指示执行AI单元的时间和/或能耗、输入数据的数量、输入数据的精度等能力信息，终端设备向网络设备发送第二信息，该第二信息包括或指示更新后的能力信息，上述AI能力信息可以包括该第一信息和第二信息。
在一种实施方式中，如图3所示，在前述的步骤S302之后，也就是终端设备向网络设备发送AI能力信息之后，该方法还包括：
S303：网络设备向终端设备发送配置信息。相应地，终端设备接收网络设备发送的配置信息。在一实施例中，该配置信息是上述AI能力信息的响应信息。
具体地，该配置信息可以满足以下一项或多项：该配置信息可以指示终端设备启动AI模式，进一步地，终端设备和网络设备之间可以进行AI通信。或者，该配置信息可以指示至少一个AI模型，进一步地，终端设备和网络设备之间可以使用该至少一个AI模型进行AI通信。又或者，该配置信息可以指示至少一个AI模型的配置参数。再或者，该配置信息可以指示至少一个AI模型的获取方法，例如，该配置信息可以指示至少一个AI模型的下载地址或获取地址。其中，该至少一个AI模型可以是根据上述AI能力信息确定的。
可以理解，网络设备接收终端设备发送的上述AI能力信息后，也可以不做响应。
在一种实施方式中，终端设备向网络设备发送上述AI能力信息，具体触发方法可以包括：
方式1：网络设备向终端设备发送请求信息。相应地，终端设备接收网络设备发送的请求信息。其中，该请求信息用于请求终端设备向网络设备发送(或称：上报)上述AI能力信息。
方式2：终端设备周期性地向网络设备发送上述AI能力信息。或者，终端设备按照预定义的时间间隔或者预定义的特定时间向网络设备发送上述AI能力信息。
方式3：当终端设备接入网络设备所在的网络时，终端设备向网络设备发送AI能力信息。其中，终端设备接入网络设备所在的网络时，也可以描述为终端设备接入网络设备所在的网络后、或终端设备接入网络设备所在的网络一定时间内。此外，终端设备接入网络设备所在的网络，也可以理解为，终端设备与网络设备建立通信连接。
方式4：当终端设备可以用于执行AI模型的计算资源发生变化(例如：增加或减少)时，或者终端设备可以用于执行AI模型的计算资源的比例发生变化时，终端设备向网络设备发送AI能力信息。同上，终端设备可以用于执行AI模型的计算资源发生变化时，也可以描述为终端设备可以用于执行AI模型的计算资源发送变化后、或终端设备可以用于执行AI模型的计算资源发送变化一定时间内，此处不再赘述。
当终端设备处于不同的应用场景时，需要完成的操作的复杂度不同，终端设备对执行AI模型所能接受(或称容忍)的功耗/能量要求、时间要求、使用的资源要求等也是不同的。也就是说，在不同的应用场景下，终端设备用于执行AI模型的最大时间要求、最大能耗要求、或资源使用情况要求不同，其中，最大时间要求可以是终端设备应该在 多长时间内执行完AI模型，最大能耗要求可以是终端设备执行完AI模型允许消耗的最大能耗，资源使用情况要求可以是终端设备执行AI模型允许使用的资源占终端设备可用资源的最大比例、或者终端设备执行AI模型可以使用的硬件资源配置。需要说明的是，上述所称终端设备执行AI模型，指的是终端设备在AI通信模式下执行AI模型的过程，并不限定终端设备所执行的AI模型的种类。
进一步地，终端设备处于不同的应用场景时，可以向网络设备发送AI能力信息，其中，AI能力信息可以包括用于终端设备用于执行AI模型的时间上限值、能耗上限值、和资源使用情况中的至少一种。可以理解地，时间上限值或称为时间上限、最大时间、或最大时间要求，能耗上限值或称为能耗上限、最大能耗、或最大能耗要求，资源使用情况例如可以为终端设备可以使用的硬件资源配置、资源比例上限值等。其中，资源比例上限值或称为资源比例上限、最大资源比例、或最大资源比例要求。可选地，上述终端设备向网络设备发送AI能力之前，网络设备还可以向终端设备发送查询信息，其中，该查询信息指示终端设备上报上述用于执行AI模型的时间上限值、能耗上限值、和资源使用情况中的一个或多个。
应理解，当终端设备的应用场景改变时，终端设备可以向网络设备上报其用于执行AI模型的时间上限值、能耗上限值、和资源使用情况信息中的至少一种，以告知网络设备其执行AI模型的时间预算要求、能量消耗限制、和资源比例消耗限制中的至少一种。此时，该时间上限值信息、能耗上限值信息、或资源使用情况信息与步骤S302中上报的信息可以携带在不同信令中。
通过上述的实施方式，终端设备通过上报其AI能力信息，使网络设备能够通过该AI能力信息评估终端设备的AI能力和AI模型(例如：AI模型的复杂度)的匹配情况，进而决定是否启动AI通信模式或者下发合适的AI模型，可以提高网络设备评估结果的简便性、高效性和准确度，进而更好地进行AI通信。
上述实施例中，终端设备向网络设备发送其AI能力信息后，网络设备可以对终端设备的AI能力和AI模型的复杂度进行评估，并确定配置信息。
图4示例性给出一种网络设备进行AI能力和AI模型匹配度评估的方法流程图。如图4所示，终端设备获取AI能力信息之前，步骤S401中，终端设备和网络设备可以获取参考AI单元信息。
如前文所述，该参考AI单元信息可以是提前标准化的或者离线指定的，并向网络设备和终端设备公开。
示例性地，可以事先标准化或离线制定AI参考表格，其中，该AI参考表格可以包括N个参考AI单元(例如：AI模型、AI模块或AI算子)的名称或编号或索引，以及该N个参考AI单元的结构描述参数，其中，该N个参考AI单元的结构描述参数也可以通过引用参考文献或链接的方式提供。通过AI参考表格，网络设备和终端设备可以统一理解N个参考AI单元的网络结构。可选地，每个AI单元的具体权重系数值(权值)可以不定义、或者预定义、或者为随机数。该AI参考表格提前向终端设备和网络设备公开。具体地，步骤S301中，终端设备执行的至少一个参考AI单元，可以是该N个参考AI单元中的K个参考AI单元，其中，N为正整数，K为不大于N的正整数。
S402：网络设备获取M个AI模型的相似度信息。其中，M为正整数。
可选地，该M个AI模型可以是网络设备或其他设备中预先存储的，或者，该M个AI模型是网络设备从其他设备获取的，或者，该M个AI模型是网络设备通过已有的AI模型转化或生成的。
具体地，网络设备获取该M个AI模型与上述K个参考AI单元的M组相似度信息。
一种实现方式中，该M个AI模型包括第一AI模型，该K个参考AI单元包括第一参考AI单元。该第一AI模型对应的相似度信息为上述M组相似度信息中的第一组相似度信息，该第一组相似度信息包括第一相似度信息，第一相似度信息与第一比例和/或第二比例相关联，其中，第一比例为第一AI模型中与第一参考AI单元相同的层的总数量占第一AI模型总层数的比例，第二比例为第一AI模型中与第一参考AI单元相同的层的总计算量占第一AI模型总计算量的比例。
也就是说，该M组相似度信息包括：M个AI模型对应的M组相似度信息。该第一AI模型对应的第一组相似度信息包括：第一AI模型与K个AI参考单元对应的K个相似度信息。
表1示意出了一种常见的神经网络模型，残差神经网络(residual neural network，ResNet)模型。AI模型可以包括AI模块，例如，表1所示ResNet-34模型包括AI模块
Figure PCTCN2022117147-appb-000002
进一步地，AI模块又可以包括AI算子，例如，AI模块
Figure PCTCN2022117147-appb-000003
包括AI算子[3×3,64](一种卷积操作算子)。
表1：ResNet神经网络模型
Figure PCTCN2022117147-appb-000004
例如，以第一AI模型为ResNet-50模型(即50层的ResNet模型)为例，当第一参考AI单元为模型，例如表1所示的ResNet-34模型时，由于ResNet-50模型和ResNet-34模型中相同的层为三层[3×3,64]、四层[3×3,128]、六层[3×3,256]、三层[3×3,512]以及[7×7,64,stride2]、[3×3max pool,stride 2]，一共18层，相同的层数占ResNet-50模型总层数的比例为18/50＝9/25，即第一比例为9/25。当第一参考AI单元为模块，例如为模块
Figure PCTCN2022117147-appb-000005
时，ResNet-50模型中包含6个该模块，一共18层，占ResNet-50模型总层数的比例为18/50＝9/25，即第一比例为9/25。当第一参考AI单元为算子，例如为算子[3×3,256]时，ResNet-50模型 和该算子相同的层为[3×3,256]，一共6层，相同的层所占的比例为6/50＝3/25，即第一比例为3/25。同理，以这些相同层的总计算量(以FLOP为单位)除以ResNet-50模型的总计算量3.8×10 9FLOPs即可得到第二比例。
另一种实现方式中，该M个AI模型包括第一AI模型，该K个参考AI单元包括第一参考AI单元。该第一AI模型对应的相似度信息为上述M组相似度信息中的第一组相似度信息，该第一组相似度信息包括第一相似度信息，第一相似度信息与第一比例和/或第二比例相关联，其中，第一比例为第一AI模型中与第一参考AI单元相同的单元的总层数占第一AI模型总层数的比例，第二比例为第一AI模型中与第一参考AI单元相同的单元的总计算量占第一AI模型总计算量的比例。
例如，同样以第一AI模型为表1所示ResNet-50模型为例，当第一参考AI单元为模型，例如表1所示的ResNet-34模型时，由于ResNet-50模型与ResNet-34模型不一样，即第一AI模型中没有与第一参考AI单元相同的单元，此时，第一比例为0。当第一参考AI单元为模块，例如为模块
Figure PCTCN2022117147-appb-000006
时，ResNet-50模型中包括的该模块的数量为3，此时，第一比例为3×3/50＝9/50。当第一参考AI单元为模块，例如为模块
Figure PCTCN2022117147-appb-000007
时，ResNet-50模型中包括的该模块的数量为0，此时，第一比例为0。当第一参考AI单元为算子，例如为算子[3×3,64]，ResNet-50模型中包括的该算子的数量为3，此时，第一比例为3/50。
可选地，网络设备可以进行本地计算，进而获取M个AI模型的相似度信息。或者，该计算过程也可以在其他设备中完成，进而，网络设备可以其他设备获取该计算结果。
S403：网络设备进行AI能力与AI模型匹配度评估。
在前述的步骤S302之后，即网络设备接收终端设备发送的AI能力信息之后，网络设备可以基于步骤S302中收到的AI能力信息，评估终端设备的AI能力和M个AI模型的匹配情况。
一些实施例中，网络设备可以通过类比评估获得M个模型在该终端设备中执行的预期完成时间、预期功耗、或预期使用的资源比例中的一种或多种。例如，当相似度用前述第一比例表示时，假设第一AI模型的总层数为L0，在步骤S402中，假设网络设备得到的第一AI模型与参考AI模型K1的相似度为S1，第一AI模型与参考AI模块K2的相似度为S2，第一AI模型与参考AI算子K3的相似度为S3。步骤S401中，终端设备和网络设备可以获知：K1、K2、K3的总层数分别为L1、L2、L3(L3＝1，算子的层数为1)。步骤S302中，终端设备给网络设备的AI能力信息中，包括终端设备执行上述参考AI模型、参考AI模块、参考AI算子的完成时间分别为t 1、t 2和t 3。则网络设备可以评估终端设备执行该第一AI模型的预期完成时间为
Figure PCTCN2022117147-appb-000008
当相似度用前述第二比例表示时，L0、L1、L2、L3可以表示总计算量，网络设备同样可以采用类似方案获得上述结果，此处不再赘述。网络设备评估终端设备执行该第一AI模型所消耗的预期能量和预期使用的资源比例的方法同样可以采用类比评估获得，此处不再赘述。
应理解，上述描述是网络设备进行终端设备AI能力与AI模型匹配度评估的一种示例性方法，另一些实施例中，网络设备可以通过M个模型以及参考AI单元的详细结构，进行更精细的分析，获取更为准确的评估结果，本申请对此不予限定。
基于上述评估获得的预期完成时间、预期功耗、或预期使用的资源比例中的一种或多 种，以及前述终端设备发送的用于执行AI模型的时间上限值信息、能耗上限值信息、或资源使用情况信息中的一种或多种，网络设备可以判断该M个模型与终端设备的AI能力是否匹配。例如，如果M个模型中的第一AI模型的预期完成时间不超过终端设备的时间预算要求，且预期消耗能量不超过终端设备的能量消耗限制，且预期消耗的资源比例不超过终端设备的资源比例限制，即可确定该第一AI模型与终端设备的AI能力相匹配。
进一步地，如前文步骤S303所述，网络设备可以向终端设备发送配置信息。
步骤S303中，当配置信息指示至少一个AI模型，或配置信息指示至少一个AI模型的配置参数，或配置信息指示至少一个AI模型的获取方法时，网络设备向终端设备发送配置信息之前，网络设备根据AI能力信息确定至少一个AI模型。具体地，一种情况是，网络设备只有一个面向当前应用的AI模型，如果该AI模型与终端设备AI能力相匹配，则该AI模型即为网络设备确定的AI模型；另一种情况是，网络设备有多于一个面向当前应用的AI模型，网络设备可以从与终端设备AI能力相匹配的AI模型中选择至少一个AI模型发送或配置给该终端设备，其中，网络设备的选取原则可以为预期完成时间最短、预期消耗能量最少、预期使用的资源比例最小、或者随机原则。可以理解地，网络设备如果能够确定至少一个AI模型，则认为网络设备可以启动AI模式。
步骤S303中，当配置信息指示至少一个AI模型的获取方法时，具体地，网络设备可以指示至少一个AI模型的下载地址或获取地址。如前所述，该M个AI模型可以是其他设备(例如：第三设备)中预先存储的，此时，网络设备可以指示终端设备从第三设备中获取该至少一个AI模型。
需要说明的是，图4所示实施例中，步骤S301、S302及S303的具体实现过程可以参见图3所示实施例中的相关表述，在此不予赘述。上述实施例中，终端设备将执行参考AI模型、AI模块或AI算子获取的数据作为AI能力信息发送给网络设备，可以准确地表示终端设备的AI能力，使网络设备获得更准确的评估结果，进而决定是否开启AI通信或者下发AI模型。
下面以基于AI的CSI反馈方案为例，对本申请实施例进行简要说明。
网络设备和终端设备可以先获取参考AI单元，基于该参考AI单元，终端设备可以获取如步骤S301所述的AI能力信息，网络设备可以获取如步骤S402所述的AI模型相似度信息，接着，终端设备可以将其获取的AI能力信息发送给网络设备，以便网络设备根据进行AI能力与AI模型匹配度评估，评估完成后，网络设备可以向终端设备发送合适的AI编码器。终端设备接收该AI编码器后，可以使用该AI编码器对CSI进行编码，并将编码后的信息发送给网络设备。然后，网络设备可以使用AI编码器对应的AI解码器对终端设备编码后的信息进行解码，以得到恢复后的信息。
本申请实施例还提供了一种AI通信方法，适用于图1所示的通信系统。如图5所示，该方法可以包括但不限于以下步骤：
S501：网络设备向终端设备发送AI模型信息，相应地，终端设备接收网络设备发送的AI模型信息。其中，该AI模型信息可以指示M个AI模型中每个AI模型的复杂度信息。
示例性地，M个AI模型中每个AI模型的复杂度信息可以用相似度数据来表示。例如，该AI模型信息指示M个AI模型对应的M组相似度信息(或称相似度数据)， 该M组相似度信息中的每组相似度信息为M个AI模型中的一个AI模型与K个参考AI单元的相似度信息，其中，该K个参考AI单元属于N个参考AI单元，M和N为正整数，K为小于等于N的正整数。
例如，当M＝3、K＝5时，该AI模型信息包括3个AI模型(Q1、Q2和Q3)对应的3组相似度信息(第1组相似度信息、第2组相似度信息、第3组相似度信息)，其中，第1组相似度信息为AI模型Q1与5个参考AI单元的相似度信息，第2组相似度信息为AI模型Q2与5个参考AI单元的相似度信息，第3组相似度信息为AI模型Q3与5个参考AI单元的相似度信息。
可选地，该M个AI模型可以是网络设备或其他设备中预先存储的，或者，该M个AI模型是网络设备从其他设备获取的，或者，该M个AI模型是网络设备通过已有的AI模型转化或生成的。
示例性地，该N个参考AI单元可以是提前标准化的或者离线指定的，并向网络设备和终端设备公开。例如，该N个参考AI单元可以是AI参考表格中的，AI参考表格的具体内容可以参见上文表述。
一种实现方式中，该M个AI模型包括第一AI模型，该K个参考AI单元包括第一参考AI单元。该第一AI模型对应的相似度信息为上述M组相似度信息中的第一组相似度信息，该第一组相似度信息包括第一相似度信息，第一相似度信息与第一比例和/或第二比例相关联，其中，第一比例为第一AI模型中与第一参考AI单元相同的层的总数量占第一AI模型总层数的比例，第二比例为第一AI模型中与第一参考AI单元相同的层的总计算量占第一AI模型总计算量的比例。
另一种实现方式中，该M个AI模型包括第一AI模型，该K个参考AI单元包括第一参考AI单元。该第一AI模型对应的相似度信息为上述M组相似度信息中的第一组相似度信息，该第一组相似度信息包括第一相似度信息，第一相似度信息与第一比例和/或第二比例相关联，其中，第一比例为第一AI模型中与第一参考AI单元相同的单元的总层数占第一AI模型总层数的比例，第二比例为第一AI模型中与第一参考AI单元相同的层的总计算量占第一AI模型总计算量的比例。关于相似度信息的具体示例可以参见上文相关表述，此处不再赘述。
可选地，该M个AI模型对应的M组相似度信息可以使用实际数值、近似数值、等级、或者范围表示。例如，上述第一相似度信息可以为第一相似度数值、第一相似度等级、或第一相似度范围。
一种实现方式中，网络设备可以在本地计算得到上述M组相似度信息。另一种实现方式中，计算M组相似度信息的过程可以在第三设备(例如：其他网络设备或第三方服务器)完成，网络设备可以从第三设备中获取该M组相似度信息。
可选地，网络设备发送的AI模型信息可以包括以下至少一种：网络设备或第三设备执行上述M个AI模型中每个AI模型时使用的输入数据的数值精度；该M个AI模型中每个AI模型的权值(或称权重、系数)的数值精度；网络设备或第三设备执行M个AI模型中每个AI模型时的运算精度。
可选地，网络设备发送的AI模型信息还可以包括该M个AI模型中每个AI模型各自的总层数、总计算量。可以理解，通过获取该M个AI模型更详细的信息，终端设备可以更准确地评估该M个AI模型的复杂度。
不同的应用场景下，执行AI模型的时间预算信息有所差别。进一步地，上述AI模型信息还可以包括执行AI模型的时间上限值，也即，执行待下发AI模型的时间预算信息，即需要在多长时间内执行完AI模型。可以理解地，时间上限值或称为时间上限、最大时间、或最大时间要求。
需要说明的是，上述AI模型信息中包括的信息可以在同一信令中，可以包括在不同信息中。当上述AI模型信息中包括的信息在同一信令中时，网络设备可以一次性将上述AI模型信息发送给终端设备。当上述AI模型信息中包括的信息在不同信令中时，网络设备可以一次或多次将上述AI模型信息发送给终端设备。
S502：终端设备向网络设备发送反馈信息。相应地，网络设备接收终端设备发送的反馈信息。
示例性地，该反馈信息可以满足以下一项或多项：该反馈信息可以用于请求启动AI通信模式。或者，该反馈信息可以包括或指示上述M个AI模型中至少一个AI模型的评估结果。或者，该反馈信息可以请求网络设备向终端设备发送至少一个AI模型，其中，该至少一个AI模型属于上述M个AI模型。
在一种实施方式中，如图5所示，在前述的步骤S501之前，也就是网络设备向终端设备发送AI模型信息之前，该方法还包括：
S500：网络设备获取该AI模型信息。
如上所述，网络设备可以在本地计算得到上述M组相似度信息，从而获取该AI模型信息。或者，网络设备可以从第三设备中获取该M组相似度信息。
在一种实施方式中，如图5所示，在前述的步骤S502之后，也就是终端设备向网络设备发送反馈信息之后，该方法还包括：
S503：网络设备向终端设备发送配置信息。相应地，终端设备接收网络设备发送的配置信息。其中，该配置信息是上述反馈信息的响应信息。
具体地，该配置信息可以满足以下一项或多项：该配置信息可以指示终端设备启动AI模式，进一步地，终端设备和网络设备之间可以进行AI通信。或者，该配置信息可以指示至少一个AI模型，进一步地，终端设备和网络设备之间可以使用该至少一个AI模型进行AI通信。又或者，该配置信息可以指示至少一个AI模型的获取方法，例如，该配置信息可以指示至少一个AI模型的下载地址或获取地址。其中，该至少一个AI模型是根据上述反馈信息确定的。
当配置信息指示至少一个AI模型的获取方法时，如前所述，该M个AI模型可以是其他设备(例如：第三设备)中预先存储的，此时，网络设备可以指示终端设备从第三设备中获取该至少一个AI模型。
可以理解，网络设备接收终端设备发送的上述反馈信息后，也可以不做响应。
在一种实施方式中，在前述的步骤S501之前，也就是网络设备向终端设备发送AI模型信息之前，该方法还包括：终端设备向网络设备发送请求信息，该请求信息用于请求网络设备向终端设备发送上述AI模型信息。
上述实施例中，网络设备向终端设备发送M个AI模型的M组相似度信息后，终端设备可以对该M个AI模型进行复杂度评估，进而确定该M个AI模型与终端设备的AI能力的匹配情况。
上述实施例中，在前述的步骤S502之前，也就是终端设备向网络设备发送反馈信 息之前，该方法还包括：终端设备根据步骤S501中收到的AI模型信息和该终端设备的AI能力信息确定反馈信息。下面通过具体实施例示例性介绍详细介绍一种终端设备确定反馈信息的方法。
图6示例性给出一种终端设备进行AI能力和AI模型匹配度评估的方法流程图。如图6所示，网络设备在获取AI模型信息之前，步骤S601中，网络设备和终端设备可以获取参考AI单元信息。
S602：终端设备获取参考AI单元信息后，获取AI能力信息。
其中，步骤S601中网络设备和终端设备获取参考AI单元信息的具体实现可以参考前述实施例步骤S401中的表述，步骤S602中终端设备获取AI能力信息的具体实现可以参考前述实施例步骤S301中的表述，在此不再赘述。
S603：终端设备进行AI能力与模型匹配度评估。
在前述的步骤S501之后，即终端设备接收到网络设备发送的AI模型信息之后，终端设备可以基于步骤S501中收到的AI模型信息，评估终端设备的AI能力和M个AI模型的匹配情况。
与前述实施例类似，终端设备同样可以通过类比评估获得M个模型在该终端设备中执行的预期完成时间、预期功耗、或预期使用的资源比例中的一种或多种。例如，当相似度用前述第一比例表示时，在步骤S501中，网络设备发送给终端设备的AI模型信息中包括第一AI模型的总层数为L0，及相似度信息：第一AI模型与参考AI模型K1的相似度为S1，第一AI模型与参考AI模块K2的相似度为S2，第一AI模型与参考AI算子K3的相似度为S3。步骤S601中，终端设备和网络设备可以获知：K1、K2、K3的总层数分别为L1、L2、L3(L3＝1，算子的层数为1)。步骤S602中，终端设备得到执行上述参考AI模型、参考AI模块、参考AI算子的完成时间分别为t 1、t 2和t 3。则终端设备可以评估其执行该第一AI模型的预期完成时间为
Figure PCTCN2022117147-appb-000009
当相似度用前述第二比例表示时，L0、L1、L2、L3可以表示总计算量，网络设备同样可以采用类似方案获得上述结果，此处不再赘述。终端设备评估其执行该第一AI模型所消耗的预期能量和预期使用的资源比例的方法同样可以采用类比评估获得，此处不再赘述。
基于上述评估的预期完成时间、预期功耗、或预期使用的资源比例中的一种或多种，以及前述网络设备发送的用于执行AI模型的时间上限值(即时间预算信息)，终端设备可以判断该M个模型与其AI能力是否匹配。例如，如果M个模型中的第一AI模型的预期完成时间不超过终端设备的时间预算要求，且预期消耗能量不超过终端设备的能量消耗限制，且预期消耗的资源比例不超过终端设备的资源比例限制，终端设备即可确定该第一AI模型与终端设备的AI能力相匹配。其中，终端设备的时间预算要求可以是前述网络设备发送的终端设备用于执行AI模型的时间预算信息，也可以是终端设备本地的时间预算要求；终端设备的能量消耗限制可以是终端设备本地的能量消耗限制；终端设备的资源比例限制可以是终端设备本地的资源比例限制。
进一步地，终端设备可以向网络设备发送反馈信息。例如，步骤S502中，该反馈信息可以指示至少一个AI模型，具体地，终端设备可以从与其AI能力相匹配的模型中选择至少一个AI模型，并通过反馈信息申请网络设备发送该模型，终端设备的选取原则可以为预期完成时间最短、预期消耗能量最少、预期使用的资源比例最小、或者随 机原则。
需要说明的是，图6所示实施例中，步骤S500、S501、S502及S503的具体实现过程可以参见图5所示实施例中的相关表述，在此不予赘述。
上述实施例中，网络设备将待下发的AI模型的复杂度传递给终端设备后，终端设备可以更加简便、高效、准确地评估待下发的AI模型与终端设备的AI能力的匹配情况，进而确定是否请求启动AI模型或请求下发的AI模型，可以提高终端设备对AI能力和AI模型匹配度评估的准确度和效率。
通过本申请实施例提供的方法，终端设备可以向网络设备发送其AI能力信息，以便网络设备准确高效地评估AI模型与终端设备的AI能力的匹配情况，或者，网络设备可以向终端设备发送AI模型信息，以便终端设备准确高效地评估AI模型与其AI能力的匹配情况。此外，本申请实施例提供的方法无需引入额外的服务器设备和交互协议，即可实现终端设备和网络设备之间的AI能力和AI模型匹配度评估。
可以理解的是，为了实现上述实施例中功能，网络设备和终端设备包括了执行各个功能相应的硬件结构和/或软件模块。本领域技术人员应该很容易意识到，结合本申请中所公开的实施例描述的各示例的单元及方法步骤，本申请能够以硬件或硬件和计算机软件相结合的形式来实现。某个功能究竟以硬件还是计算机软件驱动硬件的方式来执行，取决于技术方案的特定应用场景和设计约束条件。
图7和图8为本申请的实施例提供的可能的通信装置的结构示意图。这些通信装置可以用于实现上述方法实施例中终端设备或网络设备的功能，因此也能实现上述方法实施例所具备的有益效果。在本申请的实施例中，该通信装置可以是终端设备或网络设备，还可以是应用于终端设备或网络设备的模块(如芯片)。
如图7所示，通信装置700包括收发单元701和处理单元702。处理单元702用于调用收发单元701接收其他通信装置的信息或者向其他通信装置发送信息。收发单元701还可以进一步包括接收单元和发送单元，接收单元用于接收其他通信装置的信息，发送单元用于向其他通信装置发送信息。通信装置700用于实现上述图3、图4、图5、图6所示的方法实施例中终端设备或网络设备的功能。其中图4实施例是基于图3实施例的，图6实施例是基于图5实施例的。以下以图3实施例和图5实施例进行举例，说明收发单元701和处理单元702分别执行的操作，其它实施例中两个单元执行的操作可以参考方法实施例得到。
当通信装置700用于实现图3所示的方法实施例中终端设备的功能时：收发单元701，用于向网络设备发送AI能力信息，该AI能力信息包括终端设备执行至少一个参考AI单元中每个参考AI单元的时间和/或能耗；处理单元702，用于获取该AI能力信息。当通信装置700用于实现图3所示的方法实施例中网络设备的功能时：收发单元701，用于接收终端设备发送的AI能力信息，该AI能力信息包括终端设备执行至少一个参考AI单元中每个参考AI单元的时间和/或能耗；收发单元701，还用于向终端设备发送配置信息，其中，该配置信息指示终端设备启动AI模式，或，该配置信息指示至少一个AI模型，或，该配置信息指示至少一个AI模型的配置参数，或，该配置信息指示至少一个AI模型的获取方法，其中，至少一个AI模型是根据AI能力信息确定的。
当通信装置700用于实现图5所示的方法实施例中终端设备的功能时：收发单元701，用于接收网络设备发送的AI模型信息，该AI模型信息包括M个AI模型对应的M组相似 度信息，M组相似度信息中的每组相似度信息为M个AI模型中的一个AI模型与K个参考AI单元的相似度信息，其中，M和K为正整数；收发单元701，还用于根据AI模型信息向网络设备发送反馈信息。当通信装置700用于实现图5所示的方法实施例中网络设备的功能时：收发单元701，用于向终端设备发送AI模型信息，该AI模型信息包括M个AI模型对应的M组相似度信息，M组相似度信息中的每组相似度信息为M个AI模型中的一个AI模型与K个参考AI单元的相似度信息，其中，M和K为正整数；收发单元701，还用于接收终端设备发送的反馈信息，其中，该反馈信息是根据AI模型信息确定的。
有关上述收发单元701和处理单元702更详细的描述可以直接参考图3、图5所示的方法实施例中相关描述直接得到，这里不加赘述。
基于同一技术构思，如图8所示，本申请实施例还提供一种通信装置800。通信装置800包括接口电路801和处理器802。接口电路801和处理器802之间相互耦合。可以理解的是，接口电路801可以为收发器或输入输出接口。可选的，通信装置800还可以包括存储器803，用于存储处理器802执行的指令、或存储处理器802运行指令所需要的输入数据、或存储处理器802运行指令后产生的数据。
当通信装置800用于实现图3、图4、图5、图6所示的方法时，处理器802用于实现上述处理单元702的功能，接口电路801用于实现上述收发单元701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，该信息是网络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网络设备的。
可以理解的是，本申请的实施例中的处理器可以是中央处理单元(central processing unit，CPU)，还可以是其它通用处理器、数字信号处理器(digital signal processor，DSP)、专用集成电路(application specific integrated circuit，ASIC)、现场可编程门阵列(field programmable gate array，FPGA)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。
基于以上实施例，本申请实施例提供了一种通信系统，该通信系统可以包括上述实施例涉及的终端设备和网络设备等。
本申请实施例还提供一种计算机可读存储介质，所述计算机可读存储介质用于存储计算机程序，该计算机程序被计算机执行时，所述计算机可以实现上述方法实施例提供的下行调度方法。
本申请实施例还提供一种计算机程序产品，所述计算机程序产品用于存储计算机程序，该计算机程序被计算机执行时，所述计算机可以实现上述方法实施例提供的下行调度方法。
本申请实施例还提供一种芯片，包括处理器，所述处理器与存储器耦合，用于调用所述存储器中的程序使得所述芯片实现上述方法实施例提供的下行调度方法。
本申请实施例还提供一种芯片，所述芯片与存储器耦合，所述芯片用于实现上述方法实施例提供的下行调度方法。
本申请的实施例中的方法步骤可以通过硬件的方式来实现，也可以由处理器执行软件指令的方式来实现。软件指令可以由相应的软件模块组成，软件模块可以被存放于随机存取存储器(random access memory，RAM)、闪存、只读存储器(read-only memory，ROM)、可编程只读存储器(programmable ROM，PROM)、可擦除可编程只读存储器(erasable PROM， EPROM)、电可擦除可编程只读存储器(electrically EPROM，EEPROM)、寄存器、硬盘、移动硬盘、CD-ROM或者本领域熟知的任何其它形式的存储介质中。一种示例性的存储介质耦合至处理器，从而使处理器能够从该存储介质读取信息，且可向该存储介质写入信息。当然，存储介质也可以是处理器的组成部分。处理器和存储介质可以位于ASIC中。另外，该ASIC可以位于网络设备或终端设备中。当然，处理器和存储介质也可以作为分立组件存在于网络设备或终端设备中。
在上述实施例中，可以全部或部分地通过软件、硬件、固件或者其任意组合来实现。当使用软件实现时，可以全部或部分地以计算机程序产品的形式实现。所述计算机程序产品包括一个或多个计算机程序或指令。在计算机上加载和执行所述计算机程序或指令时，全部或部分地执行本申请实施例所述的流程或功能。所述计算机可以是通用计算机、专用计算机、计算机网络、网络设备、用户设备或者其它可编程装置。所述计算机程序或指令可以存储在计算机可读存储介质中，或者从一个计算机可读存储介质向另一个计算机可读存储介质传输，例如，所述计算机程序或指令可以从一个网站站点、计算机、第一设备或数据中心通过有线或无线方式向另一个网站站点、计算机、第一设备或数据中心进行传输。所述计算机可读存储介质可以是计算机能够存取的任何可用介质或者是集成一个或多个可用介质的第一设备、数据中心等数据存储设备。所述可用介质可以是磁性介质，例如，软盘、硬盘、磁带；也可以是光介质，例如，数字视频光盘(digital video disc，DVD)；还可以是半导体介质，例如，固态硬盘(solid state drive，SSD)。
显然，本领域的技术人员可以对本申请实施例进行各种改动和变型而不脱离本申请实施例的范围。这样，倘若本申请实施例的这些修改和变型属于本申请权利要求及其等同技术的范围之内，则本申请也意图包含这些改动和变型在内。

Claims (17)

  1. 一种人工智能AI通信方法，其特征在于，包括：
    第一设备获取AI能力信息，所述AI能力信息包括所述第一设备执行至少一个参考AI单元中每个参考AI单元的时间和/或能耗；
    所述第一设备向第二设备发送所述AI能力信息。
  2. 根据权利要求1所述的方法，其特征在于，所述第一设备向第二设备发送AI能力信息之后，所述方法还包括：
    所述第一设备接收所述第二设备发送的配置信息，其中，所述配置信息指示所述第一设备启动AI通信模式，或，所述配置信息指示至少一个AI模型，或，所述配置信息指示至少一个AI模型的配置参数，或，所述配置信息指示至少一个AI模型的获取方法。
  3. 根据权利要求1或2所述的方法，其特征在于，所述第一设备向第二设备发送AI能力信息之前，所述方法还包括：
    所述第一设备接收来自所述第二设备的请求信息，所述请求信息用于请求所述第一设备向所述第二设备发送所述AI能力信息。
  4. 根据权利要求1至3中任一项所述的方法，其特征在于，所述第一设备向第二设备发送AI能力信息，包括：
    所述第一设备周期性向所述第二设备发送所述AI能力信息；或
    所述第一设备接入所述第二设备所在的网络时，所述第一设备向所述第二设备发送所述AI能力信息；或
    所述第一设备与所述第二设备建立通信连接时，所述第一设备向所述第二设备发送所述AI能力信息；或
    所述第一设备用于执行AI模型的计算资源改变时，所述第一设备向所述第二设备发送所述AI能力信息。
  5. 一种人工智能AI通信方法，其特征在于，包括：
    第二设备接收第一设备发送的AI能力信息，所述AI能力信息包括所述第一设备执行至少一个参考AI单元中每个参考AI单元的时间和/或能耗。
  6. 根据权利要求5所述的方法，其特征在于，所述第二设备接收第一设备发送的AI能力信息之后，所述方法还包括：
    所述第二设备向所述第一设备发送配置信息，其中，所述配置信息指示所述第一设备启动AI通信模式，或，所述配置信息指示至少一个AI模型，或，所述配置信息指示至少一个AI模型的配置参数，或，所述配置信息指示至少一个AI模型的获取方法。
  7. 根据权利要求6所述的方法，其特征在于，所述配置信息指示至少一个AI模型，或所述配置信息指示至少一个AI模型的配置参数，或所述配置信息指示至少一个AI模型的获取方法时，所述第二设备向所述第一设备发送配置信息之前，所述方法还包括：
    所述第二设备根据所述AI能力信息确定所述至少一个AI模型。
  8. 根据权利要求5至7中任一项所述的方法，其特征在于，所述第二设备接收第一设备发送的AI能力信息之前，所述方法还包括：
    所述第二设备向所述第一设备发送请求信息，所述请求信息用于请求所述第一设备向所述第二设备发送所述AI能力信息。
  9. 根据权利要求2或6所述的方法，其特征在于，所述配置信息是所述AI能力信息的响应信息。
  10. 根据权利要求1至9中任一项所述的方法，其特征在于，所述第一设备执行所述至少一个参考AI单元中第一AI参考单元的时间为第一时间数值、第一时间等级或第一时间范围，所述第一设备执行所述第一AI参考单元的能耗为第一能耗数值、第一能耗等级或第一能耗范围。
  11. 根据权利要求1至10中任一项所述的方法，其特征在于，所述AI能力信息包括以下一种或多种：
    所述第一设备执行所述至少一个参考AI单元中每个参考AI单元使用的输入数据的数量；
    所述第一设备执行所述至少一个参考AI单元中每个参考AI单元使用的输入数据的数值精度；
    所述至少一个参考AI单元中每个参考AI单元的权值的数值精度；
    所述第一设备执行所述至少一个参考AI单元中每个参考AI单元时的运算精度；
    所述第一设备用于执行AI模型的时间上限值；
    所述第一设备用于执行AI模型的能耗上限值；
    所述第一设备用于执行AI模型的资源使用情况。
  12. 根据权利要求1至11中任一项所述的方法，其特征在于，所述第一设备执行所述至少一个参考AI单元中每个参考AI单元使用的资源为所述第一设备所有可用的计算资源。
  13. 一种第一设备，其特征在于，包括：一个或多个处理器和一个或多个存储器；
    所述一个或多个存储器与所述一个或多个处理器耦合，所述一个或多个存储器用于存储计算机程序代码，所述计算机程序代码包括计算机指令，当所述一个或多个处理器执行所述计算机指令时，使得所述第一设备执行如权利要求1-4或权利要求9-12任一项所述的方法。
  14. 一种第二设备，其特征在于，包括：一个或多个处理器和一个或多个存储器；
    所述一个或多个存储器与所述一个或多个处理器耦合，所述一个或多个存储器用于存储计算机程序代码，所述计算机程序代码包括计算机指令，当所述一个或多个处理器执行所述计算机指令时，使得所述第二设备执行如权利要求5-12任一项所述的方法。
  15. 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中存储有计算机可执行指令，所述计算机可执行指令在被所述计算机调用时用于使所述计算机执行上述权利要求1-12中任一项所述的方法。
  16. 一种包含指令的计算机程序产品，其特征在于，当所述计算机程序产品在计算机上运行时，使得所述计算机如执行权利要求1-12中任一项所述的方法。
  17. 一种芯片，其特征在于，所述芯片与存储器耦合，用于读取并执行所述存储器中存储的程序指令，以实现如权利要求1-12中任一项所述的方法。
PCT/CN2022/117147 2021-09-16 2022-09-06 一种人工智能ai通信方法及装置 WO2023040700A1 (zh)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
CN202111085791 2021-09-16
CN202111085791.9 2021-09-16
CN202111250477.1 2021-10-26
CN202111250477.1A CN115835182A (zh) 2021-09-16 2021-10-26 一种人工智能ai通信方法及装置

Publications (1)

Publication Number Publication Date
WO2023040700A1 true WO2023040700A1 (zh) 2023-03-23

Family

ID=85516078

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/117147 WO2023040700A1 (zh) 2021-09-16 2022-09-06 一种人工智能ai通信方法及装置

Country Status (2)

Country Link
CN (1) CN115835182A (zh)
WO (1) WO2023040700A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117596619A (zh) * 2024-01-12 2024-02-23 北京小米移动软件有限公司 通信方法、设备和存储介质

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112204532A (zh) * 2018-08-30 2021-01-08 华为技术有限公司 一种终端对ai任务支持能力的评测方法及终端
WO2021128746A1 (zh) * 2019-12-25 2021-07-01 华为技术有限公司 通信方法及装置
WO2021142609A1 (zh) * 2020-01-14 2021-07-22 Oppo广东移动通信有限公司 信息上报方法、装置、设备和存储介质

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112204532A (zh) * 2018-08-30 2021-01-08 华为技术有限公司 一种终端对ai任务支持能力的评测方法及终端
WO2021128746A1 (zh) * 2019-12-25 2021-07-01 华为技术有限公司 通信方法及装置
WO2021142609A1 (zh) * 2020-01-14 2021-07-22 Oppo广东移动通信有限公司 信息上报方法、装置、设备和存储介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
OPPO, SAMSUNG: "Update to AMMT use case - Session-specific model transfer split computation decision operation", 3GPP DRAFT; S1-210378, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. SA WG1, no. Electronic Meeting; 20210222 - 20210304, 15 March 2021 (2021-03-15), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France , XP051986485 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117596619A (zh) * 2024-01-12 2024-02-23 北京小米移动软件有限公司 通信方法、设备和存储介质
CN117596619B (zh) * 2024-01-12 2024-04-23 北京小米移动软件有限公司 通信方法、设备和存储介质

Also Published As

Publication number Publication date
CN115835182A (zh) 2023-03-21

Similar Documents

Publication Publication Date Title
US11765680B2 (en) Data analytics for multi-access edge computation
EP4066542A1 (en) Performing a handover procedure
WO2022121804A1 (zh) 半异步联邦学习的方法和通信装置
CN115280338A (zh) 一种基于联邦学习的模型训练方法、电子设备及存储介质
WO2023040700A1 (zh) 一种人工智能ai通信方法及装置
EP4152803A1 (en) Channel state information processing method, electronic device, and storage medium
US10271218B2 (en) Enable access point availability prediction
US10547359B2 (en) Calibration data
WO2023088241A1 (zh) 一种人工智能ai通信方法及装置
WO2021237463A1 (en) Method and apparatus for position estimation
WO2024036453A1 (zh) 一种联邦学习方法及相关装置
CN114846822A (zh) 定位方法、装置及系统
US20240089742A1 (en) Data transmission method and related apparatus
WO2022088003A1 (zh) 一种信息传输的方法、轻量化处理的方法以及相关通信装置
WO2024040536A1 (zh) 信息传输方法、装置、芯片、芯片模组及存储介质
WO2024093739A1 (zh) 一种通信方法及装置
WO2024099175A1 (zh) 一种算法管理方法和装置
WO2022193908A1 (zh) 一种通信方法及装置
WO2024008004A1 (zh) 一种通信方法及装置
WO2024036454A1 (zh) 一种数据特征测量方法及相关装置
WO2023044904A1 (zh) 通信方法、终端设备及网络设备
WO2023208082A1 (zh) 一种通信方法、装置、芯片及模组设备
WO2024065804A1 (zh) 数据压缩传输方法、装置、设备以及存储介质
WO2023116746A1 (zh) 一种波束确定方法及其装置
WO2022077392A1 (zh) 功率控制方法、装置及系统

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: 22869071

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 2022869071

Country of ref document: EP

ENP Entry into the national phase

Ref document number: 2022869071

Country of ref document: EP

Effective date: 20240313

NENP Non-entry into the national phase

Ref country code: DE