Communication method and device

Abstract

A communication method and apparatus are provided. The method includes: A terminal device (or an access network device) acquires channel features through inference by using a channel feature extraction model, and the channel features correspond to a channel between the terminal device and the access network device. The terminal device (or the access network device) sends the channel features to a location management function LMF. The LMF acquires positioning information of the terminal device through inference by using a positioning information acquisition model based on the channel features. The LMF may determine location information of the terminal device based on the positioning information. According to the method, an artificial intelligence model may be used to implement or assist in implementing positioning of the terminal device.

Description

[Technical Field]

[0001] This application relates to the field of communication technology, and more specifically to positioning methods and apparatus. [Background technology]

[0002] This application claims priority to Chinese Patent Application No. 202111367563.0, titled "COMMUNICATION METHOD AND APPARATUS," filed with the China National Intellectual Property Administration on 18 November 2021, which is incorporated herein by reference in its entirety.

[0003] In communication systems, positioning technology can be used to obtain location information of mobile devices. For example, positioning technology may be applied to fields such as navigation and aviation, plotting and disaster relief, vehicle navigation, logistics information queries, or traffic management. With the development of mobile communications, positioning technology based on wireless cellular networks has been widely applied. In wireless cellular networks, such as fourth-generation (4G) or fifth-generation (5G) mobile communication networks, positioning technology for mobile devices has been extensively investigated. How to improve the positioning efficiency of mobile devices is a technical problem worth researching. [Overview of the Initiative]

[0004] This application provides a communication method and apparatus, more particularly a terminal device positioning method. In this method, a terminal device (or access network device) and a location management function (LMF) use a channel feature extraction model and a positioning information acquisition model in a consistent manner, thereby enabling intelligent positioning of the terminal device.

[0005] According to the first embodiment, a communication method is provided which can be implemented on the terminal device side. This method includes sending X channel features to an LMF, where the X channel features correspond to Y channels of the terminal device, the X channel features are obtained by using a channel feature extraction model, the input to the channel feature extraction model is determined based on the Y channel responses, the Y channel responses are in a one-to-one correspondence with the Y channels, and X and Y are integers greater than or equal to 1.

[0006] Optionally, Y channels are the Y channels between a terminal device and P access network devices. Optionally, there are E channels between each of the P access network devices and the terminal device, where E is an integer greater than or equal to 1. Optionally, when E is greater than 1, the different channels among the E channels correspond to one or a combination of different cells, different transmit / receive points (TRPs), different antenna panels, different antennas, different beams, or different frequency bands.

[0007] In this method, channel features sent to the LMF can be used to implement or assist in the implementation of terminal device positioning. In this case, an artificial intelligence model may be used to solve the terminal device positioning problem, thereby improving positioning efficiency. For example, compared to conventional deterministic algorithms for calculating terminal device location information, the artificial intelligence model is acquired through training using actual channel data. In the above method, when artificial intelligence is used to implement or assist in the implementation of terminal device positioning, this positioning is closer to the actual channel environment. Therefore, terminal device positioning using the model is more accurate.

[0008] In possible implementations, X is less than or equal to Y, and / or the total dimension of X channel features is less than the total dimension of Y channel responses.

[0009] This method makes it possible to reduce the signaling overhead between the LMF and the terminal device.

[0010] In a possible implementation, X channel features are used to determine the inputs to the positioning information acquisition model, and the output of the positioning information acquisition model contains the positioning information of the terminal device.

[0011] According to this method, the LMF can acquire positioning information of a terminal device by using X channel features and a positioning information acquisition model, thereby enabling the implementation of intelligent positioning of the terminal device and improving positioning efficiency.

[0012] In possible implementations, positioning information indicates the location information of the terminal device.

[0013] In possible implementations, terminal device location information includes at least one of the following: the longitude of the terminal device's location, the latitude of the terminal device's location, the altitude of the terminal device's location, and the offset of the terminal device's location relative to a reference location (e.g., the location of a reference access network device or a reference building).

[0014] According to this method, the location information of the terminal device is obtained directly through model inference, and therefore, the positioning of the terminal device is simpler and more convenient.

[0015] In a possible implementation, the positioning information indicates at least one of the following for Y channels: channel type, first route location, or relative azimuth angle of the terminal device.

[0016] Optionally, the channel type may include line of sight (LOS) or non-line of sight (NLOS).

[0017] Optionally, the first path location includes the time-domain location of the first path within a channel in one OFDM symbol.

[0018] Optionally, for each of the Y channels, the relative azimuth angle of the terminal device indicates the relative location between the terminal device corresponding to the channel and the access network device. Optionally, the relative azimuth angle includes at least one of the following: the angle of arrival (AoA) or the angle of departure (AOD) of the reference signal on the channel. Optionally, the reference signal is a reference signal that is on the channel and sent by the terminal device to the access network device.

[0019] Positioning information is used to determine the location information of a terminal device. According to this method, LMF can acquire terminal device positioning information by using a simpler and more convenient model, thereby supporting the implementation of terminal device positioning.

[0020] In possible implementations, this method includes receiving information about the channel feature extraction model. Optionally, the channel feature extraction model information may be received from the LMF or from an artificial intelligence (AI) feature network element.

[0021] According to this method, online training of channel feature extraction models can be implemented, thereby enabling terminal devices to acquire channel feature extraction models that better match the current channel state, and thereby improve positioning accuracy.

[0022] In possible implementations, the method includes determining a channel feature extraction model based on a first reference model, the first reference model including a first reference channel feature extraction model and a first reference positioning information acquisition model.

[0023] In possible implementations, when the channel feature extraction model and the first reference positioning information acquisition model are used in a matched manner, the value of the loss function is less than or equal to the first threshold.

[0024] According to this method, a terminal device can acquire a channel feature extraction model through training using a first reference model. In this way, the number of training iterations can be reduced, and the channel feature extraction model acquired through training can better match the current channel state, thereby enabling the implementation of fast, simple, and convenient intelligent positioning.

[0025] According to a second embodiment, a communication method is provided which can be implemented on the access network device side. This method includes sending E channel features to a location management function LMF, where E is an integer of 1 or more, and the E channel features correspond to E channels of a terminal device, each of the E channel features is obtained by using a channel feature extraction model, the input to the channel feature extraction model is determined based on the channel response, and the channel response corresponds to the channel corresponding to the channel feature.

[0026] In possible implementations, the dimension of the channel features is smaller than the dimension of the channel response.

[0027] In a possible implementation, E channel features are used to determine the inputs to the positioning information acquisition model, and the output of the positioning information acquisition model contains the positioning information of the terminal device.

[0028] For an explanation of positioning information, please refer to the first aspect. Further details will not be provided again in this specification.

[0029] In possible implementations, this method includes receiving information about the channel feature extraction model. Optionally, the channel feature extraction model information may be received from the LMF or from the AI ​​feature network elements.

[0030] In possible implementations, the method includes determining a channel feature extraction model based on a first reference model, the first reference model including a first reference channel feature extraction model and a first reference positioning information acquisition model.

[0031] In possible implementations, when the channel feature extraction model and the first reference positioning information acquisition model are used in a matched manner, the value of the loss function is less than or equal to the first threshold.

[0032] According to a third embodiment, a communication method is provided which can be implemented on the AI ​​functional network element side. This method includes receiving X channel features, where the X channel features correspond to Y channels of a terminal device, and where X and Y are positive integers of 1 or more, and acquiring positioning information of the terminal device based on the X channel features and a positioning information acquisition model.

[0033] Optionally, X channel features are received from terminal devices. Optionally, X channel features are received from P access network devices. For a further description of channel features, see the first or second embodiment. Details are not described again herein.

[0034] For an explanation of positioning information, please refer to the first aspect. Further details will not be provided again in this specification.

[0035] In possible implementations, X is less than or equal to Y, and / or the total dimension of X channel features is less than the total dimension of Y channel responses.

[0036] In possible implementations, the positioning information acquisition model is determined based on a first reference model, which includes a first reference channel feature extraction model and a first reference positioning information acquisition model.

[0037] In possible implementations, when the positioning information acquisition model and the first reference channel feature extraction model are used in a matched manner, the value of the loss function is less than or equal to the first threshold.

[0038] According to a fourth aspect, a communication method is provided which can be implemented on the AI ​​functional network element side. This method includes receiving a training dataset, each of which training data in the training dataset represents Y channel responses and Y positioning information for Y channels of a terminal device, wherein the Y channel responses are in a one-to-one correspondence with the Y positioning information, and obtaining a channel feature extraction model and a positioning information acquisition model through training based on the training dataset, wherein the input to the channel feature extraction model is determined based on at least one of the Y channel responses, and the input to the positioning information acquisition model is determined based on the output of the channel feature extraction model.

[0039] In possible implementations, the loss function between the output of the positioning information acquisition model and at least one positioning information corresponding to at least one channel response is less than or equal to a threshold.

[0040] In possible implementations, the channel feature extraction model is sent to the terminal device, access network device, or LMF.

[0041] In possible implementations, the positioning information acquisition model is sent to the LMF.

[0042] According to a fifth aspect, an apparatus is provided. The apparatus is configured to implement the method according to the first aspect. The apparatus may be a terminal device, or an apparatus disposed within a terminal device, or an apparatus that can be used in conjunction with a terminal device in a corresponding manner. In design, the apparatus includes a unit configured to have a one-to-one correspondence with the methods / operations / steps / actions described in the first aspect and to implement them. The unit may be implemented by hardware circuitry, software, or hardware circuitry combined with software.

[0043] For example, the device may include a processing unit and a communication unit, and the processing unit and the communication unit may perform the corresponding functions in the first embodiment. For example, as follows:

[0044] The communication unit is configured to send X channel features to the LMF, where X channel features correspond to Y channels of the terminal device. The X channel features are obtained using a channel feature extraction model, the input to which the channel feature extraction model is determined based on Y channel responses, where Y channel responses have a one-to-one correspondence with Y channels, and X and Y are integers greater than or equal to 1. The X channel features are obtained by the processing unit using the channel feature extraction model.

[0045] For explanations of the Y channels, X channel features, etc., please refer to the first aspect. Further details will not be explained again.

[0046] In a possible implementation, X channel features are used to determine the inputs to the positioning information acquisition model, and the output of the positioning information acquisition model contains the positioning information of the terminal device.

[0047] For an explanation of positioning information, please refer to the first aspect. Further details will not be provided again in this specification.

[0048] In possible implementations, the communication unit is configured to receive information about the channel feature extraction model. Optionally, the channel feature extraction model information is received from the LMF or from the AI ​​feature network element.

[0049] In a possible implementation, the processing unit is configured to determine a channel feature extraction model based on a first reference model, the first reference model including a first reference channel feature extraction model and a first reference positioning information acquisition model.

[0050] For example, the device includes a processor configured to implement the method described in the first embodiment. The device may further include memory configured to store instructions and / or data. The memory is coupled to the processor. When executing program instructions stored in memory, the processor can implement the method described in the first embodiment. The device may further include a communication interface, which is used by the device to communicate with another device. For example, the communication interface may be a transceiver, circuit, bus, module, pin, or another type of communication interface. In a possible design, the device may include A memory configured to store program instructions, Communication interface, The system includes a processor configured to send X channel features to an LMF via a communication interface, where the X channel features correspond to Y channels of a terminal device, the X channel features are obtained by using a channel feature extraction model, the input to the channel feature extraction model is determined based on the Y channel responses, the Y channel responses are in a one-to-one correspondence with the Y channels, and X and Y are integers greater than or equal to 1.

[0051] For explanations of the Y channels, X channel features, etc., please refer to the first aspect. Further details will not be explained again.

[0052] In possible implementations, the processor receives information about the channel feature extraction model through a communication interface. Optionally, the channel feature extraction model information is received from the LMF or from the AI ​​feature network elements.

[0053] In a possible implementation, the processor is configured to determine a channel feature extraction model based on a first reference model, the first reference model including a first reference channel feature extraction model and a first reference positioning information acquisition model.

[0054] According to the sixth aspect, an apparatus is provided. The apparatus is configured to implement the method according to the second aspect. The apparatus may be an access network device, or an apparatus placed within an access network device, or an apparatus that can be used in conjunction with an access network device in a consistent manner. In the design, the apparatus includes a unit configured to have a one-to-one correspondence with the methods / operations / steps / actions described in the second aspect and to implement them. The unit may be implemented by hardware circuitry, software, or hardware circuitry combined with software.

[0055] For example, the device may include a processing unit and a communication unit, and the processing unit and the communication unit may perform the corresponding functions in the second embodiment. For example, as follows:

[0056] The communication unit is configured to send E channel features to the location management function LMF, where E is an integer greater than or equal to 1. The E channel features correspond to the E channels of the terminal device, and each of the E channel features is obtained by using a channel feature extraction model, the input to which the channel feature extraction model is determined is based on the channel response, and the channel response corresponds to the channel corresponding to the channel feature. The E channel features are obtained by the processing unit by using the channel feature extraction model.

[0057] For a description of the channel characteristics, please refer to the second aspect. Further details will not be provided again.

[0058] In possible implementations, the communication unit is configured to receive information about the channel feature extraction model. Optionally, the channel feature extraction model information is received from the LMF or from the AI ​​feature network element.

[0059] In a possible implementation, the processing unit is configured to determine a channel feature extraction model based on a first reference model, the first reference model including a first reference channel feature extraction model and a first reference positioning information acquisition model.

[0060] For example, the device includes a processor configured to implement the method described in the second embodiment. The device may further include memory configured to store instructions and / or data. The memory is coupled to the processor. When executing program instructions stored in memory, the processor can implement the method described in the second embodiment. The device may further include a communication interface, which is used by the device to communicate with another device. For example, the communication interface may be a transceiver, circuit, bus, module, pin, or another type of communication interface. In possible designs, the device may include A memory configured to store program instructions, Communication interface, The system includes a processor configured to send E channel features to a location management function LMF via a communication interface, where E is an integer greater than or equal to 1, and the E channel features correspond to E channels of a terminal device. Each of the E channel features is obtained by using a channel feature extraction model, the input to which the channel feature extraction model is determined based on the channel response, and the channel response corresponds to the channel corresponding to the channel feature.

[0061] For a description of the channel characteristics, please refer to the second aspect. Further details will not be provided again.

[0062] In possible implementations, the processor receives information about the channel feature extraction model through a communication interface. Optionally, the channel feature extraction model information is received from the LMF or from the AI ​​feature network elements.

[0063] In a possible implementation, the processor is configured to determine a channel feature extraction model based on a first reference model, the first reference model including a first reference channel feature extraction model and a first reference positioning information acquisition model.

[0064] According to the seventh aspect, an apparatus is provided for implementing the method according to the third aspect. The apparatus may be an LMF, or an apparatus placed in an LMF, or an apparatus that can be used in conjunction with an LMF in a corresponding manner. In design, the apparatus includes a unit configured to perform the methods / operations / steps / actions described in the third aspect, in one-to-one correspondence with them. The unit may be implemented by hardware circuitry, software, or hardware circuitry combined with software.

[0065] For example, the device may include a processing unit and a communication unit, and the processing unit and the communication unit may perform the corresponding functions in the third embodiment. For example, as follows:

[0066] The communication unit is configured to receive X channel features, where X channel features correspond to Y channels on a terminal device, and X and Y are positive integers greater than or equal to 1.

[0067] The communication unit is configured to acquire positioning information from a terminal device based on X channel features and a positioning information acquisition model.

[0068] For explanations of channel characteristics, positioning information, etc., please refer to the third aspect. Further details will not be explained again.

[0069] In possible implementations, the processing unit is configured to determine a positioning information acquisition model based on a first reference model, the first reference model including a first reference channel feature extraction model and a first reference positioning information acquisition model.

[0070] For example, the device includes a processor configured to implement the method described in the third embodiment. The device may further include memory configured to store instructions and / or data. The memory is coupled to the processor. When executing program instructions stored in memory, the processor can implement the method described in the third embodiment. The device may further include a communication interface, which is used by the device to communicate with another device. For example, the communication interface may be a transceiver, circuit, bus, module, pin, or another type of communication interface. In possible designs, the device may include A memory configured to store program instructions, Communication interface, The system includes a processor configured to receive X channel features through a communication interface, where the X channel features correspond to Y channels on a terminal device, and X and Y are positive integers greater than or equal to 1.

[0071] The processor is configured to acquire positioning information for the terminal device based on X channel features and a positioning information acquisition model.

[0072] For explanations of channel characteristics, positioning information, etc., please refer to the third aspect. Further details will not be explained again.

[0073] In possible implementations, the processor is configured to determine a positioning information acquisition model based on a first reference model, the first reference model including a first reference channel feature extraction model and a first reference positioning information acquisition model.

[0074] According to the eighth aspect, an apparatus is provided for implementing the method according to the fourth aspect. The apparatus may be an AI function network element, or an apparatus placed within an AI function network element, or an apparatus that can be used in conjunction with an AI function network element in a consistent manner. In the design, the apparatus includes a unit configured to perform the methods / operations / steps / actions described in the fourth aspect, in one-to-one correspondence with them. The unit may be implemented by hardware circuitry, software, or hardware circuitry combined with software.

[0075] For example, the device may include a processing unit and a communication unit, and the processing unit and the communication unit may perform the corresponding functions in the fourth embodiment. For example, as follows:

[0076] The communication unit is configured to receive a training dataset, where each piece of training data in the training dataset represents Y channel responses and Y positioning information for Y channels of a terminal device, and the Y channel responses are in a one-to-one correspondence with the Y positioning information.

[0077] The processing unit is configured to acquire a channel feature extraction model and a positioning information acquisition model through training based on a training dataset. The input to the channel feature extraction model is determined based on at least one of Y channel responses, and the input to the positioning information acquisition model is determined based on the output of the channel feature extraction model.

[0078] In possible implementations, the communication unit is configured to send a channel feature extraction model to a terminal device, access network device, or LMF.

[0079] In possible implementations, the communication unit is configured to send a positioning information acquisition model to the LMF.

[0080] For example, the device includes a processor configured to implement the method described in the fourth embodiment. The device may further include memory configured to store instructions and / or data. The memory is coupled to the processor. When executing program instructions stored in memory, the processor can implement the method described in the fourth embodiment. The device may further include a communication interface, which is used by the device to communicate with another device. For example, the communication interface may be a transceiver, circuit, bus, module, pin, or another type of communication interface. In possible designs, the device may include A memory configured to store program instructions, Communication interface, The system includes a processor configured to receive a training dataset through a communication interface, where each piece of training data in the training dataset represents Y channel responses and Y positioning information for Y channels of a terminal device, and the Y channel responses are in a one-to-one correspondence with the Y positioning information.

[0081] The processor is configured to acquire a channel feature extraction model and a positioning information acquisition model through training based on a training dataset. The input to the channel feature extraction model is determined based on at least one of Y channel responses, and the input to the positioning information acquisition model is determined based on the output of the channel feature extraction model.

[0082] In possible implementations, the processor sends the channel feature extraction model to a terminal device, access network device, or LMF via a communication interface.

[0083] In possible implementations, the processor sends a positioning information acquisition model to the LMF via a communication interface.

[0084] According to the ninth aspect, a communication system is provided, Apparatus according to the fifth aspect and apparatus according to the seventh aspect, Apparatus according to the sixth aspect and apparatus according to the seventh aspect, Apparatus according to the fifth aspect and apparatus according to the eighth aspect, Apparatus according to the sixth aspect and apparatus according to the eighth aspect, Apparatus according to the fifth embodiment, apparatus according to the seventh embodiment, and apparatus according to the eighth embodiment, Apparatus according to the sixth aspect, apparatus according to the seventh aspect, and apparatus according to the eighth aspect, or Apparatus according to the fifth embodiment, apparatus according to the sixth embodiment, apparatus according to the seventh embodiment, and apparatus according to the eighth embodiment Includes.

[0085] According to the tenth aspect, a computer-readable storage medium containing instructions is provided. When the instructions are executed on a computer, the computer is enabled to carry out the method according to the first, second, third, or fourth aspect.

[0086] According to the eleventh aspect, a computer program product including instructions is provided. When the instructions are executed on a computer, the computer is enabled to carry out the method according to the first, second, third, or fourth aspect.

[0087] According to the twelfth aspect, a chip system is provided. The chip system includes a processor and may further include memory. The chip system is configured to implement the method according to the first, second, third, or fourth aspect. The chip system may include a chip, or a chip and other separate components. [Brief explanation of the drawing]

[0088] [Figure 1] This is a schematic diagram of the communication system architecture as disclosed in this document. [Figure 2] This is a schematic diagram of the layer relationships of the neural network as disclosed in this disclosure. [Figure 3A] This is a schematic diagram of the network architecture as disclosed. [Figure 3B] This is a schematic diagram of the network architecture as disclosed. [Figure 3C] This is a schematic diagram of the network architecture as disclosed. [Figure 3D] This is a schematic diagram of the network architecture as disclosed. [Figure 3E] This is a schematic diagram of the network architecture as disclosed. [Figure 4] This is a schematic flowchart of the positioning method described in this disclosure. [Figure 5A] This is a schematic diagram of the system architecture as disclosed. [Figure 5B] This is a schematic diagram of the system architecture as disclosed. [Figure 6] This is a schematic flowchart of the positioning method described in this disclosure. [Figure 7A] This is a schematic diagram of the reference signal transmission between the UE and the base station according to this disclosure. [Figure 7B] This is a schematic diagram of the reference signal transmission between the UE and the base station according to this disclosure. [Figure 8A] This is a schematic diagram of the model training structure described herein. [Figure 8B] This is a schematic diagram of the model training structure described herein. [Figure 8C] This is a schematic diagram of the model training structure described herein. [Figure 9A] This is a schematic diagram of the model application structure according to this disclosure. [Figure 9B] This is a schematic diagram of the model application structure according to this disclosure. [Figure 9C] This is a schematic diagram of the model application structure according to this disclosure. [Figure 10] This is an illustrative diagram illustrating how to estimate the location of a UE according to this disclosure. [Figure 11] This is a schematic diagram of the channel types as disclosed. [Figure 12] This is a schematic flowchart of the positioning method described in this disclosure. [Figure 13] This is a schematic flowchart of the positioning method described in this disclosure. [Figure 14A] This is a schematic diagram of the model training structure described herein. [Figure 14B] This is a schematic diagram of the model training structure described herein. [Figure 15] This is a schematic flowchart of the positioning method described in this disclosure. [Figure 16] This is a schematic diagram of the structure of the device according to this disclosure. [Figure 17] This is a schematic diagram of the structure of the device according to this disclosure. [Modes for carrying out the invention]

[0089] Figure 1 is a schematic diagram of the architecture of a communication system 1000 to which this disclosure may apply. As shown in Figure 1, the communication system includes a radio access network (RAN) 100 and a core network (CN) 200. Optionally, the communication system 1000 may further include an internet 300. The radio access network 100 includes at least one access network device (Sometimes called RAN devices, for example, 110a and 110b in Figure 1) It may include the law of nature, The system may further include at least one terminal (for example, 120a to 120j in Figure 1). The terminal is connected wirelessly to the access network device. The access network device is connected wirelessly or wired to the core network. The core network device and the access network device may be separate physical devices independent of each other, or they may be the same physical device incorporating the functions of both the core network device and the access network device, or other possible cases. For example, one physical device may incorporate the functions of the access network device and a portion of the functions of the core network device, while the other physical device implements the remaining functions of the core network device. The physical forms of the core network device and the access network device are not limited in this disclosure. The terminal may be connected wired or wirelessly to another terminal. The access network device may be connected wired or wirelessly to another access network device. Figure 1 is a schematic diagram only and is not intended to limit this disclosure. For example, the communication system may further include other network devices, such as wireless relay devices, wireless backhaul devices, and so on.

[0090] Access network devices may include base stations, Node B, evolved Node B (eNode B, or eNB), transmission reception points (TRPs), next-generation Node B (gNB) in fifth-generation (5G) mobile communication systems, access network devices in open radio access networks (O-RAN, or open RAN), next-generation base stations in sixth-generation (6G) mobile communication systems, base stations in future mobile communication systems, and access nodes in wireless fidelity (Wi-Fi) systems. Alternatively, an access network device may be a module or unit that completes a functional part of the access network device, such as a central unit (CU), distributed unit (DU), central unit control plane (CU-CP) module, or central unit user plane (CU-UP) module. Access network devices may be macro base stations (e.g., 110a in Figure 1), micro base stations or indoor base stations (e.g., 110b in Figure 1), or relay nodes, donor nodes, etc. Specific technologies and device configurations used by access network devices are not limited to this disclosure. 5G systems are sometimes referred to as new radio (NR) systems.

[0091] In this disclosure, the device configured to implement the functionality of an access network device may be an access network device, or a device capable of supporting an access network device in implementing its functionality, such as a chip system, hardware circuitry, a software module, or hardware circuitry with a software module. The device may be installed within an access network device, or used in conjunction with an access network device in a consistent manner. In this disclosure, a chip system may include a chip, or include a chip and other separate components. For ease of explanation, the technical solutions provided will be described below using an example where the device configured to implement the functionality of an access network device is an access network device, and the access network device is a base station.

[0092] (1) Protocol layer structure.

[0093] Communication between access network devices and terminals conforms to a protocol layer structure. For example, the protocol layer structure may include a control plane protocol layer structure and a user plane protocol layer structure. For example, the control plane protocol layer structure may include a radio resource control (RRC) layer, a packet data convergence protocol (PDCP) layer, a radio link control (RLC) layer, a media access control (MAC) layer, and a physical (PHY) layer. For example, the user plane protocol layer structure may include a PDCP layer, an RLC layer, a MAC layer, and a physical layer. In possible implementations, a service data adaptation protocol (SDAP) layer may further be added on top of the PDCP layer.

[0094] Optionally, the protocol layer structure between the access network device and the terminal may further include an artificial intelligence (AI) layer used for transmitting data related to AI functionality.

[0095] The protocol layer structure between an access network device and a terminal may be considered as an access stratum (AS) structure. Optionally, a non-access stratum (NAS) layer may exist above the AS, which is used by the access network device to transfer information from the core network device to the terminal, or from the terminal to the core network device. In this case, a logical interface may be considered between the terminal and the core network device. Optionally, the access network device may transfer information between the terminal and the core network device using a transparent transmission method. For example, NAS signaling / NAS messages may be mapped to RRC signaling or included in RRC signaling as an element of RRC signaling.

[0096] (2) Central unit (CU) and distributed unit (DU)

[0097] Access network devices may include CUs and DUs. This design is sometimes called a CU-DU split. Multiple DUs may be centrally controlled by a single CU. For example, the interface between a CU and a DU is called an F1 interface. plane The CP interface may be F1-C, and the user plane (user planeThe interface may be F1-U. Specific names of interfaces are not limited in this disclosure. CUs and DUs may be obtained through segmentation based on the protocol layers of the wireless network. For example, the functions of the PDCP layer and the protocol layers above the PDCP layer (e.g., the RRC layer and the SDAP layer) may be assigned to the CU, and the functions of the protocol layers below the PDCP layer (e.g., the RLC layer, MAC layer, and PHY layer) may be assigned to the DU. In another example, the functions of the protocol layers above the PDCP layer may be assigned to the CU, and the functions of the protocol layers below the PDCP layer may be assigned to the DU. This is not limited.

[0098] The above allocation of CU and DU functionality based on the protocol layer is merely an example, and other allocation methods may be used as alternatives. For example, a CU or DU may be obtained through partitioning so that it has more functionality than the protocol layer. In another example, a CU or DU may be obtained through partitioning so that it has a portion of the processing functionality of the protocol layer. For example, the functionality of the RLC layer and the portion of the protocol layer functionality above the RLC layer may be allocated to the CU, and the functionality of the RLC layer and the remaining portion of the protocol layer functionality below the RLC layer may be allocated to the DU. In yet another example, the functionality of a CU or DU may be allocated based on the service type or other system requirements, for example, based on delay. Functionality whose processing time must meet delay requirements may be allocated to the DU, and functionality whose processing time does not need to meet delay requirements may be allocated to the CU.

[0099] Optionally, the CU may have one or more functions of the core network. For example, the CU may be located on the network side to facilitate centralized management. Optionally, the DU's radio unit (RU) may be located remotely, and the RU may have radio frequency functionality.

[0100] For example, the functions of the PHY layer may be assigned to the DU and RU. For example, the DU may implement the upper-layer functions of the PHY layer, and the RU may implement the lower-layer functions of the PHY layer. When the PHY layer is used for transmission, the functions of the PHY layer may include at least one of the following functions: cyclic redundancy check (CRC) bit summing, channel coding, rate matching, scrambling, modulation, layer mapping, precoding, resource mapping, physical antenna mapping, or radio frequency transmission. When the PHY layer is used for reception, the functions of the PHY layer may include at least one of the following functions: CRC check, channel decoding, derate matching, descrambling, demodulation, delayer mapping, channel detection, resource demapping, physical antenna demapping, or radio frequency reception. The upper-layer functions of the PHY layer may include a portion of the functions of the PHY layer, which is closer to the MAC layer. The lower-layer functions of the PHY layer may include another portion of the functions of the PHY layer, which is closer to the radio frequency functions. For example, the upper layer functions of a PHY layer may include CRC bit summing, channel coding, rate matching, scrambling, modulation, and layer mapping, while the lower layer functions of a PHY layer may include precoding, resource mapping, physical antenna mapping, and radio frequency transmission. Alternatively, the upper layer functions of a PHY layer may include CRC bit summing, channel coding, rate matching, scrambling, modulation, layer mapping, and precoding, while the lower layer functions of a PHY layer may include resource mapping, physical antenna mapping, and radio frequency transmission. For example, the upper layer functions of a PHY layer may include CRC checking, channel decoding, de-rate matching, decoding, demodulation, and de-layer mapping, while the lower layer functions of a PHY layer may include channel detection, resource demapping, physical antenna demapping, and radio frequency reception.Alternatively, the upper layer functions of the PHY layer may include CRC checking, channel decoding, delayer matching, decoding, demodulation, delayer mapping, and channel detection, while the lower layer functions of the PHY layer may include resource demapping, physical antenna demapping, and radio frequency reception.

[0101] Optionally, the functionality of a CU may be implemented by one entity or by different entities. For example, the functionality of a CU may be further divided so that the control plane and user plane are split and implemented by different entities. The split entities are a control plane CU entity (i.e., a CU-CP entity) and a user plane CU entity (i.e., a CU-UP entity), respectively. The CU-CP entity and the CU-UP entity may be combined into a DU to complete the functionality of an access network device together. In this disclosure, entities may be understood as modules or units, and the form of existence of an entity may be a hardware structure, a software module, or a hardware structure with a software module, but is not limited to this.

[0102] Optionally, one of the DU, CU, CU-CP, CU-UP, and RU may be a software module, a hardware structure, or a software module with a hardware structure added. This is not limited to these. Different entities may have the same or different forms of existence. For example, the DU, CU, CU-CP, and CU-UP are software modules, and the RU is a hardware structure. For the sake of brevity of explanation, other possible combinations are not enumerated herein. These modules and the methods performed by these modules also fall within the scope of protection of this disclosure. For example, when the methods according to this disclosure are performed by an access network device, the methods may be performed in particular by a CU, DU, or a near real-time RIC as described below.

[0103] Terminals are sometimes also called terminal devices, user equipment (UE), mobile stations, or mobile terminals. Terminals can be widely used in a variety of communication scenarios. For example, scenarios include, but are not limited to, at least one of the following: enhanced mobile broadband (eMBB), ultra-reliable low-latency communication (URLLC), massive machine-type communication ( mMTC), device-to-device (D2D), vehicle-to-everything (V2X), machine-type communication (MTC), Internet of Things (IoT), virtual reality, augmented reality, industrial control, autonomous driving, telemedicine, smart grids, smart furniture, smart offices, smart wearables, intelligent transportation, and smart cities. The terminals may include mobile phones, tablet computers, computers with wireless transceiver capabilities, wearable devices, vehicles, unmanned aerial vehicles, helicopters, airplanes, ships, robots, robotic arms, and smart home devices. The specific technologies and device forms used by the terminals are not limited to those described herein.

[0104] In this disclosure, a device configured to implement terminal functionality may be a terminal, or a device capable of supporting a terminal device in implementing functionality, such as a chip system, hardware circuitry, software module, or hardware circuitry with a software module. This device may be installed in a terminal or used in conjunction with a terminal in a corresponding manner. For ease of explanation, the technical solutions provided in this disclosure will be described below by using examples where the device configured to implement terminal functionality is a terminal, and optionally by using examples where the terminal is a UE.

[0105] Base stations and / or terminals may be in a fixed location or may be mobile. Base stations and / or terminals may be deployed on the ground, indoors or outdoors, or may be handheld or vehicle-mounted, or may be deployed on water, or may be deployed in the air on an aircraft, balloon, or satellite. The environments / scenarios for base stations and terminals are not limited in this application. Base stations and terminals may be deployed in the same environment / scenarios or in different environments / scenarios. For example, both base stations and terminals may be deployed on the ground. Alternatively, a base station may be deployed on the ground and a terminal on water. Examples are not provided one by one.

[0106] The roles of base stations and terminals can be relative. For example, the helicopter or unmanned aerial vehicle 120i in Figure 1 may be configured as a mobile base station. For terminal 120j, which accesses the radio access network 100 by using 120i, terminal 120i is a base station. For base station 110a, 120i can be a terminal; in other words, 110a and 120i can communicate with each other according to the Wireless Air Interface Protocol. Alternatively, 110a and 120i communicate with each other according to the Inter-Base Station Interface Protocol. In this case, 120i is also a base station for 110a. Therefore, base stations and terminals are sometimes collectively referred to as communication equipment (or communication devices). 110a and 110b in Figure 1 may be referred to as communication equipment with base station functionality, and 120a through 120j in Figure 1 may be referred to as communication equipment with terminal functionality.

[0107] Communication may be conducted between a base station and a terminal, between base stations, or between terminals using either an authorized spectrum or an unauthorized spectrum, or using both an authorized and an unauthorized spectrum. Communication may be conducted using a spectrum below 6 gigahertz (GHz) or a spectrum above 6 GHz, or using both a spectrum below 6 GHz and a spectrum above 6 GHz. The spectrum resources used for wireless communication are not limited in this disclosure.

[0108] The core network 200 may include one or more core network elements. Using 5G as an example, the core network may include at least one of the following network elements: an access and mobility management function (AMF) network element, a session management function (SMF) network element, a user plane function (UPF) network element, a policy control function (PCF) network element, a unified data management (UDM) network element, an application function (AF) network element, a location management function (LMF) network element, and so on. These core network elements may be hardware structures, software modules, or hardware structures with added software modules. The implementation forms of different network elements may be the same or different, but are not limited to this. Different core network elements may be different physical devices (or sometimes referred to as core network devices), or multiple different core network elements may be integrated into a single physical device, in other words, this physical device may have the functions of multiple core network elements.

[0109] In this disclosure, the device configured to implement the functionality of a core network device may be a core network device, or a device capable of supporting a core network device in implementing its functionality, such as a chip system, hardware circuitry, software module, or hardware circuitry with a software module. This device may be installed within a core network device or used in conjunction with a core network device in a consistent manner. This disclosure uses an example where the device configured to implement the functionality of a core network device is a core network device to illustrate the technical solutions provided herein.

[0110] In communication systems, positioning technologies for mobile devices are sometimes implemented. Positioning technologies are used to obtain location information for mobile devices. Positioning technologies may be applied in fields such as navigation and aviation, plotting and disaster relief, vehicle navigation, logistics information queries, or traffic management. With the development of mobile communications, positioning technologies based on wireless cellular networks are widely applied. In wireless cellular networks, such as fourth-generation (4G) or 5G mobile communication networks, positioning technologies for mobile devices are being extensively investigated. 4G systems include long-term evolution (LTE) systems.

[0111] To improve positioning efficiency and enable the system to intelligently implement or assist in the implementation of positioning functions, this disclosure introduces artificial intelligence (AI) technology for positioning functions.

[0112] Artificial intelligence can enable machines to possess human intelligence. For example, a machine can simulate certain intelligent human behaviors by using computer software and hardware. To implement artificial intelligence, machine learning methods or other methods may be used, but are not limited to these. In machine learning methods, a machine acquires a model (or rule) through learning or training using training data, and uses the model to perform inference or prediction. The inference or prediction results can be used to solve practical problems. Machine learning methods include, but are not limited to, at least one of the following: neural networks (NN), decision trees, random forests, linear models, Bayesian classifiers, probabilistic graphical models, and support vector machines (SVMs).

[0113] A neural network is used as an example. According to the universal approximation theorem, a neural network can theoretically approximate any continuous function, thereby giving it the ability to learn any mapping. Therefore, neural networks can accurately perform abstract modeling for complex high-dimensional problems. The idea of ​​a neural network comes from the neuronal structure of brain tissue. Each neuron performs a weighted sum operation on the input values ​​of the neuron and outputs the result of the weighted sum through an activation function. The input of a neuron is x=[x0,…,x n ] and the weights corresponding to the input are w=[w0,…,w n ] and it is assumed that the offset of the weighted sum is b. The form of the activation function can be diverse. If the activation function of a neuron is y=f(z)=max(0,z), then the output of the neuron is

number

number

[0114] Neural networks generally involve a multi-layer structure, where each layer may contain one or more neurons. Increasing the depth and / or width of a neural network can improve its expressive power, providing stronger information extraction and abstract modeling capabilities for complex systems. The depth of a neural network indicates the number of layers it contains, and the number of neurons contained in each layer is sometimes called the layer width. Figure 2 is an illustrative diagram of the layer relationships in a neural network. In one implementation, the neural network includes an input layer and an output layer. After performing neuronal processing on the received input, the input layer of the neural network forwards the results to the output layer, which then retrieves the output result of the neural network. In another implementation, the neural network includes an input layer, a hidden layer, and an output layer. The input layer of the neural network performs neuronal processing on the received input and then forwards the results to an intermediate hidden layer. The hidden layer then forwards the computation results to the output layer or an adjacent hidden layer. Finally, the output layer retrieves the output result of the neural network. A neural network may include one or more sequentially connected hidden layers, but is not limited to this.

[0115] In machine learning model training, a loss function may be defined. The loss function describes the difference between the model's output value and an ideal target value. Specific forms of the loss function are not limited to this disclosure. Model training is the process of adjusting some or all of the model's parameters so that the value of the loss function falls below a threshold or satisfies a target requirement. For example, in training a neural network, one or more of the following parameters may be adjusted, namely the number of layers in the neural network, the width of the neural network, the connectivity between layers, the weights of some or all of the neurons, the activation function of some or all of the neurons, and the offset in the activation function, so that the difference between the neural network's output and an ideal target value is as small as possible or below a threshold. The neural network may be considered a model.

[0116] In this disclosure, the model may also be referred to as the AI ​​model. The AI ​​model may be considered a specific method for implementing AI functionality. The AI ​​model represents a mapping relationship or functional model between the inputs and outputs of a model. AI functionality includes, namely, data collection, model training (or model learning), model information exposure, model testing, model inference (and is recommended AI functionality may include at least one of the following: theory, prediction, model monitoring or checking, or publishing inference results. AI functionality is sometimes called AI (relational) computation.

[0117] This disclosure describes how a separate network element (for example, called an AI network element, AI node, or AI device) may be introduced into the communication system shown in Figure 1 to implement some or all of the AI-related calculations or functions. The AI ​​network element may be directly connected to a base station or indirectly connected to a base station through a third-party network element. Optionally, the third-party network element may be a core network element, such as an AMF network element or a UPF network element. Alternatively, an AI entity may be located in another network element in the communication system to implement the AI-related calculations or functions. The AI ​​entity may also be called an AI module, AI unit, or by another name, and is primarily configured to implement some or all of the AI ​​functions. The specific names of the AI ​​entities are not limited in this disclosure. Optionally, this other network element may be a base station, a core network element, an operation, administration, and maintenance (OAM) device, etc. In this case, the network element performing the AI-related calculations is a network element with built-in AI functionality. Both AI network elements and AI entities implement AI-related functions. For ease of explanation, AI network elements and network elements with built-in AI functions will be described together as AI function network elements below.

[0118] In this disclosure, an OAM is configured for the operation, management, and / or maintenance of core network devices (core network device operation, administration, and maintenance system) and / or for the operation, management, and / or maintenance of access network devices (access network device operation, administration, and maintenance system). For example, this disclosure includes a first OAM and a second OAM, the first OAM being a core network device operation, administration, and maintenance system and the second OAM being an access network device operation, administration, and maintenance system. Optionally, the first OAM and / or the second OAM include AI entities. In another example, this disclosure includes a third OAM, the third OAM being an operation, administration, and maintenance system for both core network devices and access network devices. Optionally, the third OAM includes AI entities.

[0119] Optionally, AI entities may be incorporated into a terminal or terminal chip to match and support AI functionality.

[0120] Figure 3A is an illustrative diagram of an AI application framework in a communication system. In Figure 3A, the data source is used to store training data and inference data. The model training host acquires an AI model by performing training or update training based on the training data provided by the data source, and deploys the AI ​​model on the model inference host. The AI ​​model represents the mapping relationship between the model's inputs and outputs. Acquiring the AI ​​model through learning by the model training host is equivalent to acquiring the mapping relationship between the model's inputs and outputs through learning by the model training host using the training data. The model inference host uses the AI ​​model to perform inference based on the inference data provided by the data source and acquires the inference result. This method can also be described as follows: The model inference host inputs the inference data into the AI ​​model and acquires an output by using the AI ​​model, the output being the inference result. The inference result may represent the configuration parameters used (acted upon) by the execution object, and / or the actions performed by the execution object. The inference results may be uniformly planned by an actor entity and sent to one or more execution objects (e.g., core network elements, base stations, or UEs) for execution. Optionally, the model inference host may feed back its inference results to the model training host. This process is sometimes called model feedback. The fed-back inference results are used by the model training host to update the AI ​​model and deploy the updated AI model on the model inference host. Optionally, execution objects may feed back network parameters collected by the execution objects to a data source. This process is sometimes called performance feedback. The fed-back network parameters may be used as training data or inference data.

[0121] In this disclosure, the application framework shown in Figure 3A may be extended to the network elements shown in Figure 1. For example, the application framework in Figure 3A may be extended to at least one of the terminal devices, access network devices, core network devices, or independently deployed AI network elements (not shown) in Figure 1. For example, an AI network element (which may be considered a model training host) may perform analysis or training on training data provided by the terminal device and / or access network device to obtain a model. At least one of the terminal device, access network device, or core network device (which may be considered a model inference host) may perform inference using the model and inference data to obtain the output of the model. The inference data may be provided by the terminal device and / or access network device. The input to the model includes the inference data, and the output to the model is the inference result corresponding to the model. At least one of the terminal device, access network device, or core network device (which may be considered an execution object) may perform a corresponding action based on the inference result. The model inference host and the execution object may be the same or different, but are not limited to this.

[0122] Referring to Figures 3B through 3E, the following describes, by example, a network architecture to which the methods provided in this disclosure may be applied.

[0123] As shown in Figure 3B, in the first possible implementation, the access network device includes a near real-time access network intelligent controller (RAN intelligent controller, RIC) configured to perform model training and inference. For example, the near real-time RIC may be configured to train an AI model and use the AI ​​model for inference. For example, the near real-time RIC may obtain information about the network side and / or terminal side from at least one of the CU, DU, or RU. This information may be used as training data or inference data. Optionally, the near real-time RIC may submit the inference results to at least one of the CU, DU, RU, or terminal device. Optionally, the CU and DU may exchange inference results. Optionally, the DU and RU may exchange inference results. For example, the near real-time RIC submits the inference results to the DU, and the DU forwards the inference results to the RU.

[0124] As shown in Figure 3B, in a second possible implementation, there is a non-real-time RIC located outside the access network, configured to perform model training and inference (optionally, the non-real-time RIC may be located within the OAM or core network device). For example, the non-real-time RIC is configured to train an AI model and use this model for inference. For example, the non-real-time RIC may obtain network-side and / or terminal-side information from at least one of the CU, DU, or RU. This information may be used as training data or inference data. The inference results may be submitted to at least one of the CU, DU, RU, or terminal device. Optionally, the CU and DU may exchange inference results. Optionally, the DU and RU may exchange inference results. For example, the non-real-time RIC submits the inference results to the DU, and the DU forwards the inference results to the RU.

[0125] As shown in Figure 3B, in the third possible implementation, the access network device includes a near real-time RIC, and outside the access network there is a non-real-time RIC (optionally, the non-real-time RIC may be located in the OAM or core network device). Similar to the second possible implementation, the non-real-time RIC may be configured to perform model training and inference. As an addition / alternative, similar to the first possible implementation, the near real-time RIC may be configured to perform model training and inference. As an addition / alternative, the non-real-time RIC may perform model training, and the near real-time RIC may obtain AI model information from the non-real-time RIC, obtain network-side and / or terminal-side information from at least one of the CU, DU, or RU, and obtain inference results by using this information and the AI ​​model information. Optionally, the near real-time RIC may submit the inference results to at least one of the CU, DU, RU, or terminal device. Optionally, the CU and DU may exchange inference results. Optionally, the DU and RU may exchange inference results. For example, the near real-time RIC submits the inference results to the DU, and the DU forwards the inference results to the RU. For example, a near real-time RIC is configured to train model A and use model A for inference. For example, a non-real-time RIC is configured to train model B and use model B for inference. For example, a non-real-time RIC is configured to train model C and send information about model C to a near real-time RIC, which then uses model C for inference.

[0126] Figure 3C is an exemplary diagram of a network architecture to which the methods provided in this disclosure may be applied. Compared with Figure 3B, in Figure 3C the CU is split into CU-CP and CU-UP.

[0127] Figure 3D is an illustrative diagram of a network architecture to which the methods provided in this disclosure may be applied. As shown in Figure 3D, optionally, an access network device includes one or more AI entities, the functionality of which is similar to that of the near real-time RIC described above. Optionally, an OAM includes one or more AI entities, the functionality of which is similar to that of the non-real-time RIC described above. Optionally, a core network device includes one or more AI entities, the functionality of which is similar to that of the non-real-time RIC described above. When the OAM and the core network device each include AI entities, the models obtained through training by the AI ​​entities of the OAM and the core network device are different, and / or the models used for inference are different.

[0128] In this disclosure, the difference in a model includes at least one of the following differences: the structural parameters of the model (e.g., the number of neural network layers, the neural network width, the connectivity between layers, the neuron weights, the neuron activation function, or the offset in the activation function), the input parameters of the model (e.g., the type and / or dimensions of the input parameters), or the output parameters of the model (e.g., the type and / or dimensions of the output parameters).

[0129] Figure 3E is an illustrative diagram of a network architecture to which the methods provided in this disclosure may be applied. Compared to Figure 3D, the access network device in Figure 3E is split into CU and DU. Optionally, the CU may include AI entities, the functionality of which is similar to that of the near real-time RIC described above. Optionally, the DU may include AI entities, the functionality of which is similar to that of the near real-time RIC described above. When the CU and DU each include AI entities, the models trained by the AI ​​entities in the CU and DU are different, and / or the models used for inference are different. Optionally, the CU in Figure 3E may be further split into CU-CP and CU-UP. Optionally, one or more AI models may be deployed in CU-CP. Optionally, one or more AI models may be deployed in CU-UP.

[0130] In Figure 3D or Figure 3E, the OAM for the access network device and the OAM for the core network device are deployed uniformly. Alternatively, as described above, the OAM for the access network device and the OAM for the core network device may be deployed separately in Figure 3D or Figure 3E.

[0131] In this disclosure, a single model may obtain a single output through inference, the output including one or more parameters. The learning or training processes of different models may be deployed on different devices or nodes, or on the same device or node. The inference processes of different models may be deployed on different devices or nodes, or on the same device or node.

[0132] In this disclosure, network elements may perform some or all of the steps or actions related to the network element. These steps or actions are merely examples. In this disclosure, other actions or variations of various actions may be performed. In addition, the steps may be performed in a sequence different from the sequence presented in this disclosure, and not all actions in this disclosure are required to be performed.

[0133] In the examples of this disclosure, unless otherwise stated or unless there is a logical inconsistency, terms and / or descriptions in different examples may be cross-referenced, and technical features in different examples may be combined on the basis of their internal logical relationships to form new examples.

[0134] In this disclosure, 「 at least one 」 This may be described as one or more, 「 multiple 」There may be two, three, four, or more. This is not limited to them. " / " may indicate an "or" relationship between the related objects. For example, A / B may indicate A or B. "And / or" may be used to describe that there are three relationships between the related objects. For example, A and / or B may indicate the following three cases: only A exists, both A and B exist, and only B exists, where A and B may be singular or plural. In this disclosure, terms such as "first," "second," "A," or "B" may be used to distinguish technical features that have the same or similar functionality, for the sake of ease of describing the technical solutions. Terms such as "first," "second," "A," or "B" do not limit the quantity or execution sequence. In addition, terms such as "first," "second," "A," or "B" are not limited to clearly different ones. Terms such as "example" or "for example" are used to indicate an example, illustration, or explanation. Any design solution described as an "example" or "for example" should not be presented as preferable or advantageous to another design solution. The use of terms such as "example" or "for example" is intended to present the concepts involved in a particular way that is easy to understand.

[0135] The network architectures and service scenarios described in this disclosure are intended to provide a clearer explanation of the technical solutions in this disclosure and do not constitute a limitation on the technical solutions provided herein. Those skilled in the art will know that, as network architectures evolve and new service scenarios emerge, the technical solutions provided herein may also be applicable to similar technical problems.

[0136] Figure 4 is a schematic flowchart of the positioning method according to this disclosure. As shown in Figure 4, in this method, the channel feature extraction model and positioning information acquisition model used in the matched scheme are used to acquire location information of a UE in order to implement or assist in the implementation of UE positioning. In this method, the channel feature extraction model is deployed on the RAN side, for example, to a base station or UE, and is used to map channel information, such as channel responses, to channel features. The base station or UE sends the channel features to the core network. The positioning information acquisition model is deployed on the core network side, for example to an LMF, and is used to map the channel features to positioning information of the UE. The positioning information may be used to determine the location information of the UE.

[0137] In this method, the AI ​​model can be used to intelligently locate the UE, thereby improving positioning efficiency. For example, compared to conventional deterministic algorithms for calculating UE location information, in the method according to this disclosure, the model is acquired through channel information training, and the UE positioning information acquired based on the model is closer to the actual channel environment. Therefore, the positioning of UEs implemented or in which the model is used to assist in implementation is more accurate.

[0138] A channel feature extraction model is an AI model, and the name and / or model content of a channel feature extraction model (including, for example, the model structure and model parameters) are not limited. For example, a channel feature extraction model may also be called a first model, a feature extraction model, or another name, and is used to map channel information, such as channel responses, to channel features.

[0139] The positioning information acquisition model is an AI model, and the name and / or model content of the positioning information acquisition model (including, for example, the model structure and model parameters) is not limited. For example, the positioning information acquisition model may be called the second model, the positioning model, or another name, and is used to map channel features to the positioning information of the UE.

[0140] In possible implementations, UE positioning information indicates UE location information. For example, UE location information includes at least one of the following: the longitude of the UE location, the latitude of the UE location, the altitude of the UE location, and the offset of the UE location relative to a reference location (e.g., the location of a reference base station or a reference building). In this implementation, a positioning information acquisition model can be used to determine the UE location information.

[0141] In possible implementations, the positioning information of a UE indicates at least one of the following: the type of channel between the UE and the base station, the location of the first path of the channel between the UE and the base station, or the relative azimuth angle between the UE and the base station (e.g., the angle of arrival (AoA) and / or angle of departure (AOD) of the signal transmitted between the UE and the base station). In this disclosure, the channel type includes line of sight (LOS) or non-line of sight (NLOS). The LMF may acquire at least one of the channel type, the first path location, or the relative azimuth angle of the UE and determine the location information of the UE based on the acquired parameters. In this implementation, the positioning information acquisition model can assist the LMF in implementing positioning functionality for the UE. In this disclosure, the relative azimuth angle represents the relative azimuth angle between the base station and the UE, e.g., the azimuth angle of the UE relative to the base station, or the azimuth angle of the base station relative to the UE.

[0142] Figure 5A is a schematic diagram of the system architecture of the positioning method according to this disclosure. When the AMF decides to initiate positioning services for the UE, or when the AMF receives a request for positioning service relations for the UE from another network element, such as the UE or a gateway mobile location center (GMLC), the AMF sends a positioning service request to the LMF. The LMF obtains the UE's location information based on the positioning service request and sends the location information to the AMF. In scenarios where the AMF receives a request for positioning service relations for the UE from another network element, the AMF sends the UE's location information to this other network element. The AMF and LMF may be connected by a wired or wireless method, and are usually connected by a wired method. The positioning service request sent by the AMF to the LMF may include information relating to the UE. This information may be sent to the AMF by a base station in the RAN, or by the UE to the AMF. The RAN may include base stations based on one or more access technologies. For example, Figure 5A shows an LTE base station eNB and an NR base station gNB. The base station and the AMF may be connected by wire or wireless means, and are usually connected by wire. Optionally, when the UE sends UE-related information to the AMF, or when the AMF sends UE location information to the UE, data transmission may be carried out through a logical interface between the UE and the AMF.

[0143] In this disclosure, data (or information) between the UE and core network elements, such as the AMF or LMF, may be used as non-access stratum (NAS) data and is transmitted through the base station's air interface. For example, a base station may use data from the AMF as NAS data and transmit the data to the UE through the air interface. The UE may also transmit data sent to the AMF and used as NAS data to the base station through the air interface, and the base station transmits the data to the AMF.

[0144] Figure 5A illustrates a possible system framework according to embodiments of the present application and is not intended to limit the present application. For example, unlike the system framework shown in Figure 5A, there may be a logical interface between the UE and the LMF, and the UE may exchange data with the LMF through this logical interface without using the AMF. For example, the UE may send a positioning service request to the LMF, and the LMF may send the UE location information to the UE. Similar to the above methods for data exchange between the UE and the AMF, the data between the UE and the LMF may be used as NAS data and transmitted by the base station. In another example, unlike the system framework shown in Figure 5A, there is an interface for data exchange between the base station and the LMF. In this case, the base station and the LMF may be connected by wired or wireless means, and are usually connected by wired means. In another example, unlike the communication system shown in Figure 5A, the functions of the AMF and the LMF may be integrated into the same module, or the positioning functions of the AMF and the positioning functions of the LMF may be integrated into the same module. For example, as shown in Figure 5B, the device on which that module is located may be called a positioning server. Examples are not provided individually in this disclosure.

[0145] This disclosure is illustrated by using an example in which an LMF implements a positioning method. This disclosure may also be applicable to other examples in which a different network element implements a positioning method. In this case, the LMF may be interchangeable with the other network element. The network element configured to implement a positioning method may be called an LMF, or it may have another name, for example, a first network element or another name, but is not limited to this. In this disclosure, as illustrated in Figure 5A, an AMF may assist the LMF in implementing the positioning functionality of a UE. The AMF may have another name, for example, a second network element, but is not limited to this.

[0146] Figure 6 is a flowchart of the first positioning method according to this disclosure. This method includes the following operations.

[0147] Optionally, operation S601: UE sends training data to the AI ​​function network element. Optionally, operation S602: LMF sends training data to the AI ​​function network element. Optionally, operation S603: AI function network element performs model training using the training data.

[0148] AI functional network elements can perform model training by using training data to acquire channel feature extraction models and positioning information acquisition models. Training data can be sent to the AI ​​functional network elements by the UE and / or LMF.

[0149] In this disclosure, when a UE exchanges information with an AI functional network element, for example, when the UE sends training data to an AI functional network element, or when an AI functional network element sends information about a channel feature extraction model to the UE, the AI ​​functional network element communicates with the UE directly, either wired or wirelessly, or through transmission by another network element (e.g., a core network element and / or a base station).

[0150] In this disclosure, when the LMF exchanges information with an AI function network element, for example, when the LMF sends training data to an AI function network element, or when the AI ​​function network element sends model information to the LMF, the AI ​​function network element communicates directly with the LMF via a wired or wireless method, or the AI ​​function network element communicates with the LMF through transmission by another network element (for example, but not limited to, a base station and / or the AMF or another core network element).

[0151] Operations S601, S602, and S603 are optional. For example, the channel feature extraction model and the positioning information acquisition model are agreed upon in the protocol after offline training. Alternatively, the channel feature extraction model and the positioning information acquisition model are downloaded from a third-party website. In this case, S601, S602, and S603 may be skipped.

[0152] Optionally, operations S601 and S603 may be performed but operation S602 may not be performed, or operations S602 and S603 may be performed but operation S601 may not be performed, or operations S601, S602, and S603 may all be performed. In operation S601, the training data sent by the UE to the AI ​​function network elements may be measured by the UE and / or sent to the UE by the base station using signaling. In operation S602, the training data sent by the LMF to the AI ​​function network elements may be sent to the LMF by the UE and / or the base station.

[0153] The training data acquired by the AI ​​functional network element includes the training data used to determine the inputs to the channel feature extraction model and the labels corresponding to these inputs. The labels corresponding to the inputs may be understood as the target output or ideal output of the positioning information acquisition model. For example, when the inputs to the channel feature extraction model are determined based on training data A, the target output of the positioning information acquisition model is the labels corresponding to training data A.

[0154] Optionally, the training data used to determine the input for the channel feature extraction model includes at least one of the following: - Channel response between the UE and the base station. Optionally, the channel response between the UE and the base station is measured by the UE and then sent to the AI ​​functional network element or LMF. For example, as shown in Figure 7A, P base stations (in Figure 7A, three are shown as an example for illustrative purposes, but this is not limited to this case) send Y downlink reference signals to the UE. The reference signals sent to the UE by the base stations are downlink reference signals, which are sometimes abbreviated as reference signals in this disclosure. The UE receives Y reference signals. Y is an integer greater than or equal to 1, typically an integer greater than 1, e.g., 3, 4, 5, or a number greater than 1. P is an integer greater than or equal to 1 and less than or equal to Y (in Figure 7A, the case where Y is equal to P is used as an example for illustrative purposes, but this is not limited to this case). Optionally, in this disclosure, when a UE receives a reference signal, the LMF or base station may constitute at least one of the following for the UE of the reference signal: namely, bandwidth, time-domain resources, frequency-domain resources, transmit count, antenna port, spatial correlation, etc. In this disclosure, a single base station may manage one or more cells with different coverage. Within a single cell, the base station may operate in one or more frequency bands. The base station may send multiple reference signals to the UE in different cells, with one cell corresponding to one reference signal, and / or the base station may send multiple reference signals to the UE in different frequency bands within the same cell, with one frequency band corresponding to one reference signal. In this method, one reference signal may be considered to correspond to one channel. Optionally, different frequency bands may be further considered to be different cells. For example, base station A sends reference signal 1 and reference signal 2 to the UE in cell 1 and cell 2, respectively. In this case, base station A may be considered to send reference signal 1 to the UE through channel 1 in cell 1, and reference signal 2 to the UE through channel 2 in cell 2. The amount of reference signals sent to the UE by different base stations may be the same or different, but is not limited to this. Optionally, within a single cell, the base station may send multiple reference signals to the UE by using multiple different beams, with each beam corresponding to one reference signal. In this disclosure, the downlink reference signal may be a downlink synchronization signal, a positioning reference signal (PRS), or another signal whose transmission sequence is known information. This is not limited to the above. The known information may be agreed upon in the protocol or may be communicated to the UE in advance by the base station using signaling. For the y-th reference signal among Y reference signals, the UE can estimate or calculate the channel response of the channel through which the reference signal passes, based on the known transmission sequence of the reference signal and the sequence of the reference signal received by the UE. The channel response corresponds to the channel between the UE and the base station sending the y-th reference signal. For example, if P is equal to Y and each base station sends one reference signal to the UE, the channel response corresponds to the channel between the UE and the y-th base station among the Y base stations. The value of y ranges from 1 to Y. For example, the UE sends a total of Y channel responses to an AI functional network element or LMF. The Y channel responses are in a one-to-one correspondence with the Y channels between the UE and the P base stations, and each channel response uniquely corresponds to one of the channels. Each channel response can be represented in the form of a multidimensional array. For example, if the quantity of the transmitting antenna is Nt, the quantity of the receiving antenna is Nr, there are K subcarriers in the frequency domain, and there are L orthogonal frequency division multiplexing (OFDM) symbols in the time domain, then the array dimension of the channel response is Nt × Nr × K × L. Nt, Nr, K, and L are positive integers, and each element represents the channel response between the UE and the base station, at the index corresponding to that element. For example, Nt, Nr, K, and L correspond to the (j1)th transmitting antenna, the (j2)th receiving antenna, the (j3)th subcarrier, and the (j4)th OFDM symbol, respectively. For simplicity, noise is not considered. The response h between the UE and the base station satisfies the equation Srs = Stx * h, where Srs represents the time-domain received signal, Stx represents the time-domain transmitted signal, * represents the convolution, and h represents the time-domain channel response. Alternatively, the response h between the UE and the base station satisfies the equation Srs = Stx × h, where Srs represents the frequency-domain received signal, Stx represents the frequency-domain transmitted signal, × represents multiplication, and h represents the frequency-domain channel response. Srs, Stx, and h can be real or complex numbers.j1 is greater than or equal to 1 and less than or equal to Nt, j2 is greater than or equal to 1 and less than or equal to Nr, j3 is greater than or equal to 1 and less than or equal to K, and j4 is greater than or equal to 1 and less than or equal to L. For the sake of brevity of explanation, the following example uses one base station sending one reference signal to the UE. Optionally, the channel response between the UE and the base station is measured by the base station and then sent to the UE or LMF. For example, as shown in Figure 7B, the UE sends uplink reference signals to each of P base stations (in Figure 7B, the case where three are shown is used as an illustrative example), for a total of Y uplink reference signals. The reference signals sent by the UE to the base stations are uplink reference signals, which are sometimes abbreviated as reference signals in this disclosure. Each of the P base stations receives an uplink reference signal. Y is an integer greater than or equal to 1, typically an integer greater than 1, such as 3, 4, 5, or a number greater than 1. P is an integer greater than or equal to 1 and less than or equal to Y (in Figure 7B, the case where Y is equal to P is used as an illustrative example, but is not limited to this). Similar to the description of downlink reference signals above, the UE may send multiple reference signals to a single base station in different cells, with each cell corresponding to one reference signal, and / or the UE may send multiple reference signals to a single base station in different frequency bands of the same cell, with each frequency band corresponding to one reference signal. In this method, one reference signal may be considered to correspond to one channel. Optionally, different frequency bands may be further considered as different cells. For example, the UE sends reference signal 1 and reference signal 2 to base station A in cell 1 and cell 2, respectively. In this case, the UE may be considered to send reference signal 1 to base station A through channel 1 in cell 1, and reference signal 2 to base station A through channel 2 in cell 2. Optionally, within a single cell, a base station may receive multiple reference signals from the UE by using multiple different beams, with each beam corresponding to one reference signal. The amount of reference signals sent by the UE to different base stations may be the same or different; this is not limited. In this disclosure, the uplink reference signal may be a random access preamble, a sounding reference signal (SRS), or another signal whose send sequence is known information. This is not limited to the above.Known information may be agreed upon in the protocol or may be communicated to the UE in advance by the base station using signaling. After receiving the yth uplink reference signal, the corresponding base station among the P base stations may estimate or calculate the channel response of the channel through which the reference signal passes, based on the reference signal transmission sequence known to the base station and the reference signal sequence received by the base station. The channel response is described as corresponding to the channel between the UE and the base station receiving the yth uplink reference signal, or corresponding to the channel through which the yth uplink reference signal passes. The base station sends the channel response to the UE or AI functional network element. The value of y ranges from 1 to Y. The UE or AI functional network element receives a total of Y channel responses from the P base stations. For the sake of brevity of explanation, the following example uses a case where the UE sends one reference signal to one base station for illustrative purposes. Optionally, in this disclosure, when a base station receives a reference signal, the LMF may constitute at least one of the following for the base station of the reference signal: namely, bandwidth, time-domain resources, frequency-domain resources, transmit count, antenna port, spatial correlation, etc. In this disclosure, both uplink reference signals and downlink reference signals are sometimes referred to simply as reference signals, but whether a reference signal is an uplink reference signal or a downlink reference signal may be determined based on the transmitting and / or receiving units. In this disclosure, uplink reference signals and downlink reference signals differ in at least one of the following: the transmission direction of the reference signals, time-domain resources, frequency-domain resources, or sequence values. Optionally, in the training data, the "channel response between the UE and the base station" may be further replaced with at least one of the following: the amplitude of the channel response, the phase of the channel response, the real part of the channel response, the imaginary part of the channel response, the phase difference of the channel responses of different receiving or transmitting antennas, or the phase difference between multiple channel responses. The channel response and these variations of the channel response may be collectively referred to as channel information. Optionally, the channel information may further include reference signal received power (RSRP), reference signal time difference (RSTP), and the type of environment in which the UE is located. - RSRP for the channel between the UE and the base station. Optionally, the RSRP is measured by the UE and then sent to the AI-enabled network element or LMF. For example, based on the above description of Figure 7A, we use the example where P is equal to Y and each base station sends one reference signal to the UE. After estimating the channel response of the channel between the UE and the yth base station, the UE may obtain the RSRP of the channel based on the channel response. The value of y ranges from 1 to Y. For example, the UE sends a total of Y RSRPs to the AI-enabled network element or LMF, and the Y RSRPs are in a one-to-one correspondence with the Y channels between the UE and the Y base stations, with each RSRP uniquely corresponding to one of the channels. The case where P is not equal to Y is similar to the above description. Further details are not described herein. Optionally, information is measured by the base station and then sent to the UE or LMF. For example, based on the above description of Figure 7B, an example is used where P is equal to Y and the UE sends one reference signal to each base station. After estimating the channel response of the channel between the UE and the yth base station, the yth base station may obtain the RSRP of the channel based on the channel response. The yth base station sends the RSRP to the LMF or UE. The value of y ranges from 1 to Y. The UE or AI functional network element receives a total of Y RSRPs from the Y base stations. The case where P is not equal to Y is similar to the above description. Further details are not described herein. - RSTP channel between the UE and the base station. The information is measured by the UE and then sent to the AI ​​function network element or LMF. For example, based on the above description in Figure 7A, the UE may obtain the RSTD of different reference signals through estimation. For example, P is equal to Y, and each base station sends one reference signal to the UE. The UE may measure the RSTD of the reference signal of the yth base station and the RSTD of the reference signal of the first base station to obtain a total of Y-1 RSTDs, and the UE sends Y-1 RSTDs to the AI ​​function network element or LMF. Alternatively, the UE may measure the RSTD of the reference signal of the yth base station and the RSTD of the reference signal of the (y-1)th base station to obtain a total of Y-1 RSTDs, and the UE sends Y-1 RSTDs to the AI ​​function network element or LMF. Specific RSTD reporting methods are not limited in this disclosure. The value of y ranges from 2 to Y. The case where P is not equal to Y is as described above. Further details are not described herein. - The type of environment in which the UE is located. The type of environment in which the UE is located may be a factory environment, an office environment, a high-speed rail environment, a subway environment, a shopping mall environment, a residential area environment, or any other possible environment. This is not limited to these. Information may be reported by the UE or base station to an AI functional network element or LMF. In this disclosure, training data corresponding to different environments may be used to train a model that corresponds to different environments. Optionally, the target output (training data labels) of the positioning information acquisition model may include at least one of the following: - UE location information. UE location information includes the longitude, latitude, and altitude of the UE location, as well as the offset of the UE location relative to a reference location (e.g., the location of a reference base station or a reference building). Optionally, the UE is a beacon UE. The UE knows its location information and sends it to an AI function network element or LMF. Alternatively, the base station knows the UE's location information and sends it to the UE or LMF. Alternatively, the LMF knows the UE's location information and sends it to an AI function network element. The base station is at least one of P base stations. Optionally, the LMF may obtain the UE's location information by using a non-AI-based positioning method and send the location information to the AI ​​function network element. For example, in this disclosure, the non-AI-based positioning method may be a downlink (DL)-time difference of arrival (TDOA) positioning method, an uplink (UL)-DOA positioning method, an UL-AoA positioning method, a DL-AOD positioning method, or another positioning method described in the 3rd generation partnership project (3GPP) 38.305 protocol, but is not limited to these. Optionally, to increase accuracy, the LMF may obtain multiple estimated location information for the UE through multiple estimations, use a weighted average of the multiple estimated location information as the UE's location information, and send this location information to the AI ​​function network element. - The type of channel between the UE and the base station. For a given base station, the channel type between the UE and the base station can be either LOS or NLOS. Optionally, an example is used where P is equal to Y, and UE is a beacon UE. The UE knows the channel type of the channel between the UE and base station y, and the UE sends the channel type to an AI function network element or LMF, and the UE sends a total of Y channel types. Alternatively, base station y knows the channel type of the channel between the UE and base station y, and base station y sends the channel type to the UE or LMF, and a total of Y base stations send the channel type to the UE or LMF, and the UE or LMF receives a total of Y channel types. Alternatively, the LMF knows the channel type of the channel between the UE and base station y, and the LMF sends the channel type to an AI function network element, and the LMF sends a total of Y channel types. The value of y ranges from 1 to Y. When P is not equal to Y, this method is similar. For example, for base station A, base station A, together with the UE, transmits (sends or receives) reference signal 1 and reference signal 2 in cell 1 and cell 2, respectively. In this case, for channel 1 corresponding to reference signal 1 and channel 2 corresponding to reference signal 2, the UE or base station A sends the channel type for channel 1 and the channel type for channel 2, respectively. In other words, a total of Y channel types are determined for Y channels between P base stations and UEs, and each channel corresponds to one reference signal. Similar cases are not described again below. Optionally, an example where P equals Y is used. The LMF obtains the channel type of the channel between the UE and the yth base station by using a non-AI-based channel type determination method and sends the channel type to the AI ​​function network element. The LMF sends a total of Y channel types. The value of y ranges from 1 to Y. Optionally, to increase accuracy, the LMF may obtain the channel type of the channel between the UE and the yth base station through multiple estimations and use the channel type that appears more frequently or with a higher probability as the channel type of the channel between the UE and the yth base station, and send the channel type to the AI ​​function network element. For example, the LMF obtains the channel type of the channel between the UE and the yth base station through 5 estimations, and the channel types are LOS, NLOS, LOS, LOS, and LOS, respectively. Since LOS appears more frequently or with a higher probability, the LMF considers the channel type between the UE and the yth base station to be LOS. The method is similar when P is not equal to Y. For example, regarding base station A, base station A, together with the UE, transmits (sends or receives) reference signal 1 and reference signal 2 in cell 1 and cell 2, respectively. In this case, for channel 1 corresponding to reference signal 1 and channel 2 corresponding to reference signal 2, the LMF determines the channel type of channel 1 and the channel type of channel 2, respectively. In other words, a total of Y channel types are determined for Y channels between P base stations and UEs, and each channel corresponds to one reference signal. Similar cases are not described again below. Optionally, in this disclosure, a non-AI-based channel type determination method may determine whether the channel type is LOS or NLOS by using the rice factor of the channel response. Typically, the rice factor of an LOS signal is higher than that of an NLOS signal. For example, the rice factor of the channel response may be compared to a threshold. When the rice factor is greater than or equal to the threshold, the channel type is LOS; otherwise, the channel type is NLOS. For a channel response, the rice factor of the channel response is

number

number

number

number

number

[0155] Optionally, when measuring training data, the UE, base station, and / or LMF may, in order to improve measurement accuracy, repeatedly measure a single training data set multiple times and use the combined value of the multiple repeated measurements, e.g., mean, weighted sum, or another possible value, as the training data. This method can improve the reliability of the training data. Optionally, when measuring training data, the UE, base station, and / or LMF may further mark the reliability and / or generation time of the training data for reference during the model training process. For example, training data that is close to and / or highly reliable with respect to its generation time during training may be used for model training.

[0156] Optionally, in this disclosure, an AI functional network element acquires multiple training data sets corresponding to at least one UE. These multiple training data sets may be considered as a single training dataset. The AI ​​functional network element performs model training by using the training dataset to acquire a channel feature extraction model and a positioning information acquisition model. When a single training dataset corresponds to multiple UEs, two different training data sets in the training dataset may correspond to two different UEs, measurement data for the same UE at two different geographic locations, measurement data for the same UE at two different moments in time, or measurement data for the same UE in two different frequency bands. When a single training dataset corresponds to one UE, two different training data sets in the training dataset may correspond to measurement data for the same UE at two different geographic locations, measurement data for the same UE at two different moments in time, or measurement data for the same UE in two different frequency bands.

[0157] In this disclosure, one training dataset may correspond to one environment type. For example, an AI functional network element acquires multiple training data corresponding to at least one UE in an environment. Multiple training data may be considered as a single training dataset. The AI ​​functional network element performs model training by using the training dataset to acquire a channel feature extraction model and a positioning information acquisition model corresponding to the environment type. As described above, when one training dataset corresponds to multiple UEs, two different training data in the training dataset may correspond to two different UEs in the same environment type, or to measurement data for the same UE in the same environment type but at two different geographic locations, different moments, or different frequency bands. When one training dataset corresponds to one UE, two different training data in the training dataset may correspond to measurement data for the same UE in the same environment type but at two different geographic locations, different moments, or different frequency bands. As described above, the environment type may be a factory environment, an office environment, a high-speed rail environment, a subway environment, a shopping mall environment, a residential area environment, or another possible environment, but is not limited to these. This method allows for the acquisition of a model that better matches each environment type by training a separate model for each environment type. Therefore, during model application, the corresponding model can be matched and applied based on the UE's environment.

[0158] In this disclosure, the UE configured to perform model training and the UE configured to perform model inference may be the same or different (as shown in the figures of this disclosure), but this is not limited. For example, the first UE performs operation S601, and the second UE performs operations S604, S606a, S606b, S607, and S608. The positioning information of the second UE is determined in operation S609. The first UE and the second UE may be the same or different. Regardless of whether the first UE and the second UE are the same, the base station configured to perform model training and the base station configured to perform model inference may be the same or different (as shown in the figures of this disclosure), but this is not limited. For example, the model training process (e.g., operation S601) relates to a first group of P base stations, and the model application process (e.g., operation S607) relates to a second group of P base stations. The first group of P base stations and the second group of P base stations may be the same or different. When the first group of P base stations and the second group of P base stations are different, at least one base station in the first group of P base stations is not included in the second group of P base stations. Optionally, the first UE and the second UE are the same, and at least one identical base station exists in both the first group of P base stations and the second group of P base stations. This identical base station communicates with the UE in the first cell when it is in the first group of P base stations, and communicates with the UE in the second cell when it is in the second group of P base stations. The first cell and the second cell may be the same or different.

[0159] After the training dataset has been collected, the AI ​​functional network elements can acquire a channel feature extraction model and a positioning information acquisition model through training using the training data and the labels corresponding to the training data. During model training, the model parameters may be adjusted, and the objective of model training is as follows: When the inputs to the channel feature extraction model are determined based on the training data in the training dataset, and the inputs to the positioning information acquisition model are acquired based on the output of the channel feature extraction model, the output of the positioning information acquisition model will be as close as possible to the labels of the training data. Methods for training AI models are not described in detail or limited in this application. For example, mature industry methods for training neural networks, such as stochastic gradient descent (SGD), batch gradient descent (BGD), adaptive gradient algorithm (AdaGrad), momentum algorithm, and Newton's method, can be used to acquire the most optimal neural network parameters possible through iterative training. Model training methods obtained through future industry research may also be applicable to this disclosure, but are not limited to these.

[0160] Optionally, the AI function network element may further collect a new training dataset and perform model update training by using this new training dataset to obtain an updated channel feature extraction model and an updated positioning information acquisition model. For example, the update training is periodic. The AI function network element uses the data collected within the period as the training dataset and performs update training by using the training dataset. In another example, the update training is triggered by an event. For example, when the error of the measurement result of the positioning information acquisition model exceeds a threshold, the AI function network element performs update training on the channel feature extraction model and the positioning information acquisition model based on the new training dataset.

[0161] FIG. 8A, FIG. 8B, and FIG. 8C are structural diagrams for training a channel feature extraction model and a positioning information acquisition model for positioning. The following explains how the AI function network element trains the channel feature extraction model and the positioning information acquisition model based on the training dataset according to FIGS. 8A, 8B, and 8C.

[0162] For one training data A in the training set, the process of training the channel feature extraction model and the positioning information acquisition model by using the training data A may be referred to as the training process.

[0163] Optionally, as shown in FIG. 8A, the training data A includes Y channel responses H1, H2,..., H Y For each of the Y channel responses, or for the y-th channel response H y the AI function network element determines the input Feature_model_In y of the channel feature extraction model. The value of y ranges from 1 to Y. The AI function network element uses the input Feature_model_In yBased on the channel feature extraction model, the output of the channel feature extraction model is Feature_model_Out. y Obtain Feature_model_Out here, y This is the channel feature S y This demonstrates that the AI ​​functional network element is a channel feature extraction model and a total of Y channel features S1, S2, ..., S Y This method can obtain the channel response H y For each channel, the channel feature extraction model extracts the channel features S corresponding to the channel response. y Used to obtain.

[0164] Optionally, channel feature S y This can be a real number, a vector, a matrix, or a multidimensional array (larger than one dimension). It is not limited to these.

[0165] The AI ​​functional network element determines the input to the positioning information acquisition model based on the output of the channel feature extraction model. For example, the AI ​​functional network element determines the input to the positioning information acquisition model based on Y channel features. The AI ​​functional network element outputs positioning information based on the input and the positioning information acquisition model. For example, the AI ​​functional network element takes channel features S1, S2, ..., S Y Based on this, the B inputs of the positioning information acquisition model are Position_model_In b It is possible to obtain the following. The value of b ranges from 1 to B, and B is an integer greater than or equal to 1. The AI ​​function network element takes B inputs into the positioning information acquisition model and outputs B outputs Position_model_Out b Obtain the value of b, which ranges from 1 to B. Position_model_In b The corresponding output is Position_model_Out b That is the case.

[0166] Position_model_Out bThis indicates the location information of the UE, or at least one of the following for each of the Y channels: namely, the channel type of the UE, the first path location, or the relative azimuth angle (e.g., AOA and / or AOD of the reference signal communicated through the channel). In the model training algorithm, the channel feature extraction model and / or positioning information acquisition model output Position_model_Out b The output is adaptively adjusted by comparing it to the labels (target output) of training data A. The channel feature extraction model and positioning information acquisition model obtained in one training process can be used for subsequent iterations of training. In subsequent iterations, the channel feature extraction model and positioning information acquisition model obtained through training based on training data A are used as base or initial models for update training. After multiple iterations of training, the difference between the output of the positioning information acquisition model and the target output is made as small as possible. For example, for the number of training iterations N train N can be defined. train The model obtained after a certain number of training iterations is used as the model obtained throughout the training process. train is a positive integer. In another example, a loss function may be defined. The loss function is used to represent the difference between the output of the positioning information acquisition model and the target output. After multiple iterations using multiple training data from the training dataset, when the value of the loss function becomes as small as possible or smaller than (or equal to) a threshold, the model acquired through iterative training is used as the model acquired through training.

[0167] Optionally, as shown in Figure 8B, the training data A consists of Y channel responses H1, H2, ..., H YThe AI ​​functional network element determines the input to a channel feature extraction model based on Y channel responses, and obtains the output to the channel feature extraction model based on the input and the channel feature extraction model, where the output represents one channel feature S. Optionally, S can be a real number, a vector, a matrix, or a multidimensional array (greater than one dimension). This is not limited to this. The channel feature S can be considered to correspond to Y channels between the UE and P base stations. The AI ​​functional network element determines the input to a positioning information acquisition model based on the channel feature S. The AI ​​functional network element outputs the positioning information of the UE based on the input and the positioning information acquisition model. For example, the AI ​​functional network element may obtain the input Position_model_In to the positioning information acquisition model based on the channel feature S. The AI ​​functional network element inputs Position_model_In to the positioning information acquisition model and outputs Position_model_Out.

[0168] Position_model_Out indicates the location information of the UE, or at least one of the following for each of the Y channels: namely, the channel type of the UE, the first path location, or the relative azimuth angle (e.g., the AOA and / or AOD of the reference signal communicated through the channel). As described above in Figure 8A, the channel feature extraction model and the position information acquisition model may be obtained after multiple iterative training sessions using multiple training data in the training dataset.

[0169] Optionally, as shown in Figure 8C, the training data A consists of Y channel responses H1, H2, ..., H YThe AI ​​functional network element can determine a total of Q inputs to a channel feature extraction model based on Y channel responses. Each input is determined by T channel responses out of Y channel responses, where Q and T are positive integers and Q × T = Y. For the q-th input out of Q inputs, the AI ​​functional network element obtains the output of the channel feature extraction model based on the input and the channel feature extraction model, where the output is the channel feature S q This shows that the value of q ranges from 1 to Q. Optionally, channel features S q This can be a real number, a vector, a matrix, or a multidimensional array (larger than one dimension). This is not limited to this. The AI ​​functional network element is a channel feature extraction model with a total of Q channel features S1, S2, ..., S Q It is possible to obtain Q channel features that can be considered to correspond to Y channels. In this method, for every T channel responses, the channel feature extraction model obtains one channel feature S corresponding to the T channel responses. Q Used to obtain. The AI ​​function network elements are S1, S2, ..., S Q The input to the positioning information acquisition model is determined based on this. The AI ​​function network element outputs the positioning information of the UE based on the input and the positioning information acquisition model. For example, the AI ​​function network element has channel features S1, S2, ..., S Q Based on the N inputs of the positioning information acquisition model, Position_model_In n The following can be obtained, where the value of n ranges from 1 to N, and N is an integer greater than or equal to 1. The AI ​​function network element has N inputs to the positioning information acquisition model, Position_model_In n Enter the following and output N Position_model_Out n Obtain the following: The value of n ranges from 1 to N. Position_model_In n The corresponding output is Position_model_Out n That is the case.

[0170] Position_model_Out n This indicates the location information of the UE, or at least one of the following for each of the Y channels: namely, the channel type of the UE, the first path location, or the relative azimuth angle (e.g., the AOA and / or AOD of the reference signal communicated through the channel). As described above in Figure 8A, after multiple iterative training sessions using multiple training data in the training dataset are completed, the channel feature extraction model and the positioning information acquisition model may be obtained.

[0171] Optionally, in this disclosure, the input and / or output dimensions of a model in a model training algorithm may be predefined or predetermined. Thus, the channel feature extraction model obtained through training can satisfy requirements for feedback overhead and positioning accuracy. For example, a smaller output dimension of the channel feature extraction model indicates lower feedback overhead. A larger output dimension of the channel feature extraction model indicates higher positioning accuracy.

[0172] In this disclosure, determining the input to a channel feature extraction model based on a channel response includes, but is not limited to, using the channel response as the input to a channel feature extraction model; separating the real part of a channel response of dimension Z from the imaginary part, combining the separated real part matrix of dimension Z and the separated imaginary part matrix of dimension Z into a 2Z-dimensional matrix, and using the 2Z-dimensional matrix as the input to a channel feature extraction model; separating the amplitude of a channel response of dimension Z from the phase, combining the separated amplitude matrix of dimension Z and the separated phase matrix of dimension Z into a 2Z-dimensional matrix, and using the 2Z-dimensional matrix as the input to a channel feature extraction model; combining the real part matrix, the imaginary part matrix, the amplitude matrix, and the phase matrix into a 4Z-dimensional matrix, and using the 4Z-dimensional matrix as the input to a channel feature extraction model; or combining information about the channel response (at least one of the real part, imaginary part, amplitude, or phase) with other feedback information (e.g., RSRP and / or RSTD) as the input to a channel feature extraction model.

[0173] In this disclosure, determining the inputs of a positioning information acquisition model based on the outputs of a channel feature extraction model includes, but is not limited to, determining one input of a positioning information acquisition model based on one or more outputs of a channel feature extraction model. For example, one output of a channel feature extraction model may be used as one input of a positioning information acquisition model, or multiple outputs of a channel feature extraction model may be connected or combined to form one input of a positioning information acquisition model, or one or more outputs of a channel feature extraction model and other feedback information (e.g., RSRP and / or RSTD) may be combined into one input of a positioning information acquisition model. For example, S channel features may be S1, S2, ..., S Y It can be expressed as, and one input to the positioning information acquisition model is, [S1,S 2, ...,S Y ] or [S 1, S2,...,S Y ] TIt may include. For example, T channel features could be S, S2, ..., S T It can be expressed as, and one input to the positioning information acquisition model is, [S1,S 2, ...,S T ] or [S 1, S2,...,S T ] T It may include. [ ] T represents the transposition of a matrix. Similar to determining the input to a channel feature extraction model based on channel responses, the input to a positioning information acquisition model may be further obtained by separating the real part of the channel features from the imaginary part and / or separating the amplitude of the channel features from the phase. Further details are again not explained. Similar methods may be further used to determine the input to a channel feature extraction model based on multiple channel responses. Further details are again not explained.

[0174] Optionally, in this disclosure, the above H y Regarding H y This can be further divided into multiple subchannel responses, each of which corresponds to a single antenna port, and the input to the channel feature extraction model is determined based on each of the subchannel responses. For example, H yThe dimensions are 512 × 4 × 16, where 512 represents the frequency domain dimension (e.g., corresponding to 512 subcarriers), 4 represents the time domain dimension (e.g., corresponding to 4 OFDM symbols), and 16 represents the spatial domain dimension (e.g., corresponding to 16 antenna ports), where 512, 4, and 16 are used only as examples and can be replaced with other integer values. A channel feature extraction model can be used to perform channel feature extraction for each channel response of an antenna port. For example, the input to a channel feature extraction model is the channel response of one antenna port, and the dimensions of the input are 512 × 4. Alternatively, a single channel feature extraction model can be used for channel feature extraction for 16 antenna ports. For example, the input to a channel feature extraction model is the channel responses of multiple antenna ports, and the dimensions of the input are 512 × 4 × 16. Different methods affect the input dimensions of the channel feature extraction model, but the methods of model training and model application are similar. Similarly, in the frequency domain and / or time domain, partitioning can also be carried out in a manner similar to that of antenna ports. The granularity of the partitioning in the frequency domain may be an integer number of subcarriers, an integer number of resource blocks (RBs), an integer number of resource block groups (RBGs), or other possible cases. The granularity of the partitioning in the time domain may be an integer number of OFDM symbols, an integer number of slots, or other possible cases.

[0175] Optionally, multiple pairs of models (channel feature extraction model + positioning information acquisition model) can be trained, and the appropriate model can be selected and applied as needed. The application of the AI ​​model may further consider the compromise between complexity and performance. A neural network is used as an example; more layers and neurons in the network indicate higher complexity and higher performance of the AI ​​model, but require more computing resources to be consumed. Fewer layers and neurons in the network indicate lower complexity and lower performance of the AI ​​model, but require less computing resources to be consumed. The AI ​​model can be selected by taking into account the actual application scenario. For example, a more complex AI model may be used for UEs with high computing power, such as mobile phones, tablet computers, or in-vehicle terminals. For UEs with low computing power, such as Internet of Things terminals or mMTC terminals, a slightly simpler AI model or an AI model with fewer parameters may be used. Accordingly, the appropriate AI model may also be selected based on the computing power of the LMF.

[0176] Optionally, the input to the channel feature extraction model may further include the RSRP of the channel between the UE and the base station and / or the RSTP of the channel between the UE and the base station. Alternatively, optionally, in addition to the input to the positioning information acquisition model determined based on the output of the channel feature extraction model, the input to the positioning information acquisition model may further include the RSRP of the channel between the UE and the base station and / or the RSTP of the channel between the UE and the base station.

[0177] Optionally, operation S604:UE sends UE capability information to LMF.

[0178] UE capability information indicates at least one of the following pieces of information about the UE: - Will UE support AI-based positioning methods? For example, when the UE supports an AI-based positioning method, the LMF may determine the UE's positioning information by using the AI-based positioning method. For instance, to perform positioning, the UE uses the channel feature extraction model provided in this disclosure, and the LMF uses the positioning information acquisition model provided in this disclosure. When the UE does not support an AI-based positioning method, the LMF may determine the UE's location information by using a non-AI-based positioning method. The question "whether the UE supports AI-based positioning methods" can be replaced with the type of positioning method supported by the UE. The type can be either an AI-based positioning method or a non-AI-based positioning method. Optionally, in this disclosure, AI-based positioning methods may be further described as machine learning algorithm-based positioning methods, AI model-based positioning methods, or by any other name. This is not limited to the above. - UE computing capability information. UE computing power information may indicate the amount of operation supported by the UE per unit time, or it may indicate the UE's computing power level. For example, multiple computing power levels may be agreed upon in the protocol or pre-configured for the UE by the network side (e.g., base station, AMF, or LMF), with each computing power level corresponding to the amount of operation supported by the UE per unit time. The UE may determine its computing power level based on the amount of operation supported by the UE per unit time and multiple computing power levels. The network side (for example, the LMF or AI functional network elements described below) can configure an appropriate channel feature extraction model for the UE based on the UE's computing power information. According to this method, the UE can operate successfully. For example, the computing resources used by the UE when it uses the channel feature extraction model to perform inference do not exceed the UE's computing power level, or do not occupy too much of the UE's computing power. Occupying too much of the UE's computing power includes occupying more than 50% of the UE's computing resources or any other percentage of the UE's computing resources. - Information about models (AI models) supported by UE. The information indicates information about AI models supported by the UE. This information may indicate whether the UE supports a particular type of AI model, whether the UE describes an AI model, or the type of AI model supported by the UE. For example, the information may indicate whether the UE supports at least one of the following types of AI models: fully connected neural networks, convolutional neural networks, recurrent neural networks, neural networks supporting attention mechanisms, or different neural network types. According to this method, the network can construct a suitable model for the UE.

[0179] In this disclosure, the following methods may be used when the UE sends information to the LMF, for example, when sending information about the UE's capabilities.

[0180] Optionally, when the UE sends information to the LMF, if there is a logical interface between the UE and the LMF, the information may be sent by the UE to the base station in the form of a NAS message, forwarded by the base station to core network element A (e.g., the AMF or another network element), and then forwarded by core network element A to the LMF, or the information may be sent by the UE to the base station in the form of a NAS message, and then forwarded by the base station to the LMF.

[0181] Optionally, when the UE sends information to the LMF, if there is no logical interface between the UE and the LMF, the information may be sent by the UE to the base station in the form of a NAS message, forwarded by the base station to core network element B (e.g., AMF or another network element), and then notified to the LMF by core network element B; or the information may be sent by the UE to the base station in the form of a NAS message, forwarded by the base station to core network element A, forwarded by core network element A to core network element B, and then notified to the LMF by core network element B. If there is a logical interface between the UE and core network element B, the logical interface is used to communicate the above information. If there is no logical interface between the UE and core network element A, or if there is a logical interface between the UE and core network element A, the above information is not interpreted and cannot be communicated at the logical interface layer.

[0182] Core network element A can be either a single network element (one-hop transfer) or multiple network elements (multi-hop transfer).

[0183] In this disclosure, when an LMF sends information to a UE, for example, information about a channel feature extraction model, the LMF may send the information by using a path opposite to that described above. For example, when an LMF sends information to a UE, if there is a logical interface between the UE and the LMF, the information may be sent by the LMF to core network element A in the form of a NAS message, forwarded by core network element A to a base station, and then forwarded by the base station to the UE, or the information may be sent by the LMF to a base station in the form of a NAS message, and then forwarded by the base station to the UE. For the sake of brevity of explanation, further details are not described herein.

[0184] Optionally, sending information to the LMF by the UE may involve querying (or requesting) the LMF. For example, the LMF sends UE capability query information to the UE. After receiving the query information, the UE reports its capability information to the LMF.

[0185] Optionally, a UE may pre-send its capability information to the LMF. For example, when a UE accesses a network, the UE may pre-send its capability information to the LMF, or when a time-domain resource (e.g., a period and an offset within the period) for reporting capability information by the UE is agreed upon in the protocol or pre-configured by the LMF, the UE may send its capability information to the LMF on the time-domain resource, or when capability information changes or a network handover occurs, the UE may send its capability information to the LMF. The reasons for a UE sending its capability information to the LMF are not limited in this application.

[0186] Action S604 is optional. For example, if the capabilities of a UE are agreed upon in the protocol, S604 does not need to be performed. For example, if there are multiple capability types of UEs in the system, or if the capabilities of a UE can change autonomously, a UE may perform S604.

[0187] Optionally, operation S605 (not shown): A positioning information acquisition model is configured. Optionally, operation S606 (not shown): A channel feature extraction model is configured. S605 and S606 can be collectively referred to as model configuration. By performing operation S605 and operation S606, the AI function network element can deploy the models obtained through training on the UE and the LMF.

[0188] The model configuration can be implemented by using model configuration method 1 or model configuration method 2 as follows.

[0189] Model configuration method 1:

[0190] Optionally, operation S605a: The AI function network element sends information about the positioning information acquisition model to the LMF. Optionally, operation S606a: The AI function network element sends information about the channel feature extraction model to the UE.

[0191] Operation S605a is optional. For example, the positioning information acquisition model can be agreed upon in the protocol after offline training, or the positioning information acquisition model can be downloaded from a third-party website. Therefore, S605a can be skipped.

[0192] Operation S606a is optional. For example, the channel feature extraction model can be agreed upon in the protocol after offline training, or the channel feature extraction model can be downloaded from a third-party website. Therefore, S606a can be skipped.

[0193] In the above method, both operation S605a and operation S606a can be performed, or neither operation S605a nor operation S606a may be performed, or operation S605a can be performed but S606a is not performed, or operation S605a may not be performed but S606a is performed. This is not limited.

[0194] Model configuration method 2:

[0195] Optionally, in operation S605b: the AI ​​function network element sends information about the positioning information acquisition model and information about the channel feature extraction model to the LMF. In operation S606b: the LMF sends information about the channel feature extraction model to the UE.

[0196] In possible implementations, operations S605b and / or S606b are optional. For example, the positioning information acquisition model and the channel feature extraction model may be agreed upon in the protocol after offline training, or the positioning information acquisition model and the channel feature extraction model may be downloaded from a third-party website, or either the positioning information acquisition model or the channel feature extraction model may be agreed upon in the protocol after offline training, and the other may be downloaded from a third-party website.

[0197] In this disclosure, when model information is exchanged between different network elements (indicated as network element C and network element D), for example, when information regarding a positioning information acquisition model or a channel feature extraction model is exchanged between an AI function network element and an LMF, or when information regarding a channel feature extraction model is exchanged between an AI function network element and an UE, or when information regarding a channel feature extraction model is exchanged between an LMF and an UE, or when information regarding a channel feature extraction model is exchanged between an LMF and a base station as described below, or when information regarding a channel feature extraction model is exchanged between an AI function network element and a base station as described below, an example is used in which network element C sends model information to network element D, and network element C may send at least one of the following pieces of information about the model to network element D. - Dimensions of the model's input parameters. - The type of input parameter for the model. For example, input parameter types include channel response, channel RSRP, channel RSTP, etc. - Gradient information of the model. For example, the information represents the gradient information of the currently sent model relative to a base model or to a previously sent model. For instance, the model information previously sent to network element D by network element C is information A, and network element D may obtain information B based on information A and the gradient information, and information B gives the new model information of the model. In other words, the model can be updated or reconfigured by using the model's gradient information. - The number of layers in the model. - The connection relationships between layers of the model. - The offset of the model's neurons. - The weights of the neurons in the model. - The model's index (or identifier). - The model's validity period. The model must be available within its validity period. Otherwise, the model is unavailable.

[0198] Optionally, when an AI function network element is an LMF, or when the AI ​​function network element and the LMF are located in the same device, the AI ​​function network element sending model information to the LMF can be implemented as follows: The LMF reads the model information.

[0199] Operation S607: The UE extracts channel features using a channel feature extraction model. Operation S608: The UE sends the channel features to the LMF. Operation S609: The LMF acquires the UE's positioning information using a positioning information acquisition model.

[0200] For example, as shown in Figure 7A, and similar to the above description of model training, P base stations send Y downlink reference signals to the UE. The UE receives Y reference signals, where Y is an integer greater than or equal to 1, typically an integer greater than 1, such as 3, 4, 5, or a number greater than 1. For the yth reference signal among the Y reference signals, the UE may, through measurement, obtain the channel response of the channel through which the reference signal passes. The channel response corresponds to the channel between the UE and the base station sending the reference signal. The value of y ranges from 1 to Y.

[0201] After obtaining Y channel responses, the UE can determine the input to a channel feature extraction model based on the Y channel responses, and can obtain X channel features through inference by using the channel feature extraction model. The UE sends the X channel features to the LMF. This method can be described as follows: The UE sends X channel features to the LMF, where the X channel features correspond to the Y channels of the UE, the X channel features are obtained by using a channel feature extraction model, the input to the channel feature extraction model is determined based on the Y channel responses, the Y channel responses have a one-to-one correspondence with the Y channels, Y is a positive integer greater than or equal to 1, and X is an integer greater than or equal to 1.

[0202] Y channels are the Y channels between the UE and P base stations. Optionally, there are E channels between each of the P base stations and the UE, where E is an integer greater than or equal to 1. Optionally, when E is greater than 1, different channels among the E channels correspond to different cells and / or frequency bands. For different base stations, the value of E may be the same or different; this is not limited.

[0203] The LMF receives X channel features and determines the UE's positioning information by using the X channel features and a positioning information acquisition model. For example, the LMF determines the input to the positioning information acquisition model based on the X channel features and acquires the UE's positioning information through inference by using the positioning information acquisition model.

[0204] In this method, positioning is performed using an AI model to implement intelligent positioning, and therefore the positioning performance becomes closer to the actual channel environment, thereby implementing more accurate positioning.

[0205] In the case of arbitrary selection, X is a positive integer less than or equal to Y. When fewer than Y(X) channel features are reported compared to reporting Y channel responses, the amount of information reported can be reduced, and signaling overhead can be reduced.

[0206] Optionally, the total dimension of X channel features is less than the total dimension of Y channel responses. This method is a possible way to reduce the amount of information reported and can reduce the signaling overhead between the UE and the LMF. For example, compared to sending high-dimensional channel responses to the LMF to determine the UE's positioning information, this method sends low-dimensional channel features to the LMF to determine the UE's positioning information, which can reduce the signaling overhead between the UE and the LMF. For example, the total dimension of Y channel responses is 512 × 4 × 16, and the total dimension of X channel features can be reduced to 16. For example, 16 real or complex numbers represent channel features corresponding to channel responses. This example is used solely to illustrate the problem and is not intended to limit the present disclosure.

[0207] Figures 9A, 9B, and 9C are structural diagrams showing how the UE and LMF perform positioning using a channel feature extraction model and a positioning information acquisition model. Figures 9A, 9B, and 9C are similar to Figures 8A, 8B, and 8C, respectively. Figures 8A, 8B, and 8C are used for model training, while Figures 9A, 9B, and 9C are used for model application or model inference. In Figures 8A, 8B, and 8C, and in Figures 9A, 9B, and 9C, the channel feature extraction model and the positioning information acquisition model operate in the same manner. The difference is that in Figures 8A, 8B, and 8C, the model information of the channel feature extraction model and / or the positioning information acquisition model may be updated throughout training, and finally, the model information of the trained channel feature extraction model and / or the trained positioning information acquisition model is output, while in Figures 9A, 9B, and 9C, inference is performed by using the trained channel feature extraction model and / or the trained positioning information acquisition model, and in this process, the model information of the channel feature extraction model and / or the positioning information acquisition model does not change.

[0208] Optionally, as shown in Figure 9A, Y channel responses H1, H2, ..., H Y After obtaining the Y channel responses, for each of the Y channel responses, or for the y-th channel response H y Regarding this, UE uses the input Feature_model_In for the channel feature extraction model. y Determines the value of y, which ranges from 1 to Y. The UE is the input Feature_model_In y Based on the channel feature extraction model, the output of the channel feature extraction model, Feature_model_Out, is obtained through inference. y Obtain Feature_model_Out y This is the channel feature S y This shows that UE is a channel feature extraction model and a total of Y channel features S1, S2, ..., S Y This method can obtain the channel response H yFor each channel, the channel feature extraction model extracts the channel features S corresponding to the channel response. y Used to obtain the channel feature S. Optionally, y The channel features can be real numbers, vectors, matrices, or multidimensional arrays (larger than one dimension). This is not limited to these. In actual processing, the UE may have (or store) one channel feature extraction model, and all Y channel features are obtained through inference by using the channel feature extraction model. This method can conserve the UE's memory resources. Alternatively, the UE may have (or store) multiple (e.g., Y) channel feature extraction models, and at least two of the Y channel features are obtained through inference by using the same channel feature extraction model. This method can improve the UE's processing speed.

[0209] The method by which the UE determines the input to the channel feature extraction model based on the channel response is similar to the corresponding explanation in Figure 8A. Further details are again not described herein.

[0210] UE is obtained from Y channel features S1, S2, ..., S in LMF. Y The LMF sends the following: The LMF determines the input to the positioning information acquisition model based on Y channel features. The LMF acquires the positioning information of the UE through inference based on the input and the positioning information acquisition model. For example, the LMF takes channel features S1, S2, ..., S Y Based on this, the B inputs of the positioning information acquisition model are Position_model_In b It is possible to obtain the following. LMF has B inputs to the positioning information acquisition model: Position_model_In b Inputting this information, the inference process yields a total of B outputs: Position_model_Out b Obtain the following: The value of b ranges from 1 to B, where B is an integer greater than or equal to 1. Position_model_In b The corresponding output is Position_model_Out bIt is as follows. When B is greater than 1, in actual processing, LMF may have (or store) one positioning information acquisition model, and multiple input Position_model_Out b are all obtained through inference by using the positioning information acquisition model. This method can save the memory resources of LMF. Alternatively, LMF may have (or store) multiple (for example, B) positioning information acquisition models, and multiple Position_model_Out b At least two of the Position_model_Out b are obtained through inference by using the same positioning information acquisition model. This method can improve the processing speed of LMF.

[0211] Optionally, the positioning information of the UE indicates the location information of the UE.

[0212] Example A1:

[0213] B is equal to 1, and Position_model_Out b indicates at least one of the longitude, latitude, and altitude of the UE.

[0214] Example B1:

[0215] B is equal to 1, and Position_model_Out b indicates the offset of the UE with respect to the reference location. Optionally, the input of the positioning information acquisition model may further include the reference location.

[0216] Optionally, in Example A1 and Example B1, Position_model_In b may include Y elements, where the y-th element is S y and the value of y ranges from 1 to Y. Alternatively, similar to the above description, Position_model_In b is the real part, imaginary part, amplitude, and y separated from Y S phaseincludes at least one of them.

[0217] Optionally, the positioning information of the UE indicates at least one of the following for each of the Y channels, namely, the channel type of the UE, the first path location, or the relative azimuth angle. The channel type, the first path location, and / or the relative azimuth angle of the UE are used to determine the location information of the UE.

[0218] Example C1:

[0219] B is greater than or equal to 1, and Position_model_Out b , b = 1,..., B represent the first path location [First_Path1, First_Path2,..., First_Path Y of each of the Y channels.

[0220] For example, B is equal to 1, and Position_model_In b contains Y elements, where the y-th element is S y and the value of y ranges from 1 to Y, and Position_model_Out b represents the first path location of each of the Y channels.

[0221] In another example, B is greater than 1, for example, B is equal to 2, Position_model_In1 contains Y1 elements, where the (y1)-th element is S y1 and the value of y1 ranges from 1 to Y1, Position_model_Out1 represents the first path location of each of the first channel to the (Y1)-th channel, Position_model_In2 contains Y2 elements, where the (y2)-th element is S Y1+y2The value of y2 ranges from 1 to Y2, and Position_model_Out2 indicates the first path location for each channel from the (Y1+1) channel to the (Y1+Y2) channel. The fact that B is equal to 2 is used as an example only and is not intended to limit this disclosure. Position_model_In b This is determined by using the channel features of the channel group, and Position_model_Out b It can be understood that this indicates the first route location of the group of channels.

[0222] Optionally, Position_model_Out b , b=1, ..., B is the time-domain location First_Path for each of the Y channels within the same time length (e.g., the length of one OFDM symbol). y This shows that, where the value of y ranges from 1 to Y. After obtaining the first route location for each channel, the LMF may obtain the location information of the UE based on the first route locations of at least three channels.

[0223] For example, Figure 10 shows an example of estimating the location of a UE based on a first route location. In Figure 10, base station 1 sends a reference signal P1 to the UE through channel 1, with arrival time t1 for P1; base station 2 sends a reference signal P2 to the UE through channel 2, with arrival time t2 for P2; and base station 3 sends a reference signal P3 to the UE through channel 3, with arrival time t3 for P3. t1, t2, and t3 are all greater than 0. The UE may, through measurements, obtain the arrival time difference t1-t2 or t2-t1 (t1-t2 is used as an example in the figure and below) between P1 and P2, and the arrival time difference t1-t3 or t3-t1 (t1-t3 is used as an example in the figure and below) between P1 and P3. The UE may report the arrival time differences t1-t2 and t1-t3 to the LMF. The departure times for P1 and P2 are assumed to be P1_Tx_t and P2_Tx_t, respectively. P1_Tx_t and P2_Tx_t are greater than 0. In possible implementations, P1_Tx_t and P2_Tx_t are the same as those agreed upon in the protocol. In another possible implementation, P1_Tx_t and P2_Tx_t can be configured flexibly. In this case, P1_Tx_t and P2_Tx_t may be the same or different. Base station 1 may send P1_Tx_t to the LMF, and base station 2 may send P2_Tx_t to the LMF. The departure times for P1 and P3 are assumed to be P1_Tx_t and P3_Tx_t, respectively. P3_Tx_t is greater than 0. In possible implementations, P1_Tx_t and P3_Tx_t are the same as those agreed upon in the protocol. In another possible implementation, P1_Tx_t and P3_Tx_t can be configured flexibly. In this case, P1_Tx_t and P3_Tx_t may be the same or different. Base station 1 may send P1_Tx_t to the LMF, and base station 3 may send P3_Tx_t to the LMF.The LMF may determine that the arrival time difference between the first paths of P1 and P2 is (t1 - P1_Tx_t + First_Path1) - (t2 - P2_Tx_t + First_Path2), and may determine that the arrival time difference between the first paths of P1 and P3 is (t1 - P1_Tx_t + First_Path1) - (t3 - P3_Tx_t + First_Path3). When P1_Tx_t and P2_Tx_t are the same, (t1 - P1_Tx_t + First_Path1) - (t2 - P2_Tx_t + First_Path2) can be simplified to (t1 + First_Path1) - (t2 + First_Path2). When P1_Tx_t and P3_Tx_t are the same, (t1 - P1_Tx_t + First_Path1) - (t3 - P3_Tx_t + First_Path3) can be simplified to (t1 + First_Path1) - (t3 + First_Path3). Based on the arrival time difference between the first paths of P1 and P3, the LMF may determine that the location of the UE satisfying the arrival time difference is curve 1, and based on the arrival time difference between the first paths of P1 and P2, the LMF may determine that the location of the UE satisfying the arrival time difference is curve 2. Therefore, the LMF may consider that the intersection of curve 1 and curve 2 is the location of the UE, or rather, the intersection of curve 1 and curve 2 may be regarded as the estimated location for the UE.

[0224] Optionally, Position_model_Out b , b = 1,..., B where First_Path is more than three y When indicating, the LMF may select three channels corresponding to the three First_Path y and estimate the location of the UE according to the above method. Alternatively, the LMF may estimate multiple locations for the UE and determine the average value of the multiple estimated locations as the location of the UE. Each of the estimated locations of the UE is obtained through an estimation based on the three channels corresponding to the three First_Path y

[0225] ​This disclosure does not limit the specific algorithms used by the LMF to determine the location of the UE based on the first route location, or in other words, it does not limit how the LMF uses the output of the positioning information acquisition model to help implement the positioning capabilities of the UE.

[0226] Example D1:

[0227] B is 1 or greater, and Position_model_Out b , b=1, ..., B is the first path location for each of the Y channels [First_Path1, First_Path2, ..., First_Path Y This indicates the channel type. Similar to example C1, B can be equal to 1 or greater than 1. Position_model_In b This is determined by using the channel features of the channel group, and Position_model_Out b It can be understood that this indicates the first route location and channel type of the channel group.

[0228] As shown in Figure 11, the base station sends a signal to the UE. The signal can reach the UE via multiple paths. However, the UE cannot measure the signal on the LOS path because the LOS path between the base station and the UE is blocked by a tree. The first path measured by the UE is assumed to be the signal on the NLOS path, which is reflected by a wall. If the UE's location is estimated based on the arrival time of the signal on the NLOS path, the distance between the UE and the base station can be estimated to be d2 + d3. However, the actual distance between the UE and the base station is d1 on the LOS path. Therefore, to improve the accuracy of the UE's location estimation, when the first path between the UE and the base station is the LOS path, the UE's location can be estimated based on the location of the first path.

[0229] In Example D1, the LMF has a channel type of Position_model_Out b Based on the first path location of the channel, which is the LOS indicated by b=1, ..., B, the location of the UE can be determined by using a method similar to that in Example C1. For example, Position_model_Out b ,b=1, ...,B represents [First_Path1,First_Path2, ...,First_Path8] and Position_model_Out b , b=1, ..., B indicates that channels 1, 2, 4, 5, 7, and 8 are LOS paths, and channels 3 and 6 are NLOS paths. In this case, the LMF can determine the location of the UE based on First_Path1, First_Path2, First_Path4, First_Path5, First_Path7, and First_Path8 by using the method described in Example C1.

[0230] Example E1:

[0231] B is 1 or greater, and Position_model_Out b , b=1, ..., B represents the channel type of each of the Y channels. Similar to Example C1, B can be equal to 1 or greater than 1. Position_model_In b This is determined by using the channel features of the channel group, and Position_model_Out b It can be understood that this indicates the channel type of the group of channels.

[0232] Example E1 is similar to Example D1. Unlike Example D1, the first route location for each of the Y channels is Position_model_Out b, b=1, ..., not shown by B, and obtained by another method. The specific method is not limited. For example, to simplify the calculation, the first route locations for various channels may be the same as those agreed upon in the protocol, or the first route locations may be reported to the LMF by the UE or the corresponding base station. For example, the UE may report the first route location for each of the Y channels, or each of the Y base stations corresponding to the Y channels may report the first route location for the channel between the base station and the UE.

[0233] Example F1:

[0234] B is 1 or greater, and Position_model_Out b , b=1, ..., B is the relative azimuth angle of the UE corresponding to each of the Y channels [Direction1, Direction 2,..., Direction Y This indicates ].

[0235] For example, B is equal to 1, and Position_model_In b It contains Y elements, where the yth element is S y And the value of y ranges from 1 to Y, and Position_model_Out b This indicates the azimuth angle of the UE corresponding to each of the Y channels.

[0236] In another example, B is greater than 1, for example, B is equal to 2, and Position_model_In1 contains Y1 elements, where the (y1)th element is S y1 The value of y1 ranges from 1 to Y1, Position_model_Out1 indicates the relative azimuth angle of the UE corresponding to each of the first channels from the (Y1) channel, Position_model_In2 contains two elements, where the (y2) element is S Y1+y2The value of y2 ranges from 1 to Y2, and Position_model_Out2 indicates the relative azimuth angle of the UE corresponding to each of the channels from the (Y1+1) to the (Y1+Y2). The fact that B is equal to 2 is used as an example only and is not intended to limit this disclosure. Position_model_In b This is determined by using the channel features of the channel group, and Position_model_Out b It can be understood that this indicates the relative azimuth angle of the UE corresponding to the group of channels.

[0237] Optionally, Position_model_Out b , b=1, ..., B is the relative azimuth angle Direction of the UE corresponding to each of the Y channels. y This shows that, where the value of y ranges from 1 to Y. After obtaining the relative azimuth angles of the UE corresponding to the channels, the LMF may obtain the location information of the UE based on the relative azimuth angles of the UE corresponding to at least two of the channels.

[0238] For a channel, the relative azimuth angle of the UE corresponding to the channel may be AOA and / or AOD. For how the LMF determines the location information of the UE by using AOA and AOD, see the methods described in the 3GPP 38.805 protocol, for example, the DL (downlink)-AOD positioning method and the UL-AOA (uplink) positioning method described in the protocol. This disclosure may further use other positioning methods based on AOA and AOD, but is not limited to these.

[0239] Optionally, Position_model_Out b Directions where b=1, ..., B is greater than 2 y When indicating, LMF is Direction yA portion of the channel corresponding to the UE can be selected, and the UE's location can be estimated according to the method described above. Alternatively, the LMF can estimate multiple locations for the UE and determine the average of the multiple estimated locations as the UE's location.

[0240] This disclosure does not limit the specific algorithms used by the LMF to determine the location of a UE based on its azimuth angle, or, in other words, does not limit how the LMF uses the output of a positioning information acquisition model to help implement the positioning capabilities of a UE.

[0241] Optionally, as shown in Figure 9B, Y channel responses H1, H2, ..., H Y After obtaining the channel responses, the UE determines one input Feature_model_In of the channel feature extraction model based on the Y channel responses, and obtains one output Feature_model_Out of the channel feature extraction model through inference based on the input and the channel feature extraction model. The output represents the channel feature S. The UE may obtain a total of one channel feature S based on the channel feature extraction model and the Y channel responses. Optionally, S can be a real number, a vector, a matrix, or a multidimensional array (greater than one dimension). This is not limited. The UE sends the channel feature S to the LMF. In this method, the channel feature S can be considered to correspond to the Y channels.

[0242] The method by which the UE determines the input to the channel feature extraction model based on the channel response is similar to the corresponding explanation in Figure 8B. Further details are again not described herein.

[0243] The LMF determines the input to the positioning information acquisition model based on the channel features S. The LMF acquires the UE's positioning information through inference based on the input and the positioning information acquisition model. For example, the LMF may acquire the input Feature_model_In to the positioning information acquisition model based on the channel features S. The LMF acquires the output Feature_model_Out through inference based on the input and the positioning information acquisition model. Feature_model_Out represents the UE's positioning information.

[0244] Optionally, the UE positioning information indicates the UE's location information, or the channel type and / or first route location for each of the Y channels. The channel type and / or first route location are used to determine the UE's location information. The channel type and / or first route location are used to assist the LMF in determining the UE's location information. The description of Feature_model_Out is the same as that in Figure 9A. Details are again not explained. In Figures 9A and 9B, the inputs and outputs of the channel feature extraction models are different, but their processing ideas are similar. In Figure 9A, for Y channels, the LMF may perform processing by using one or more positioning information acquisition models. In Figure 9B, for Y channels, the LMF may perform processing by using one positioning information acquisition model.

[0245] Optionally, as shown in Figure 9C, Y channel responses H1, H2, ..., H Y After obtaining the channel responses, the UE can determine a total of Q inputs to the channel feature extraction model based on the Y channel responses, where each input is determined by T channel responses out of the Y channel responses, and Q and T are positive integers, with Q × T = Y. For the q-th input out of the Q inputs, the UE obtains the output of the channel feature extraction model based on the input and the channel feature extraction model, where the output is the channel feature S q This shows that the value of q ranges from 1 to Q.

[0246] Optionally, channel feature S q The UE can be a real number, a vector, a matrix, or a multidimensional array (larger than one dimension). This is not limited to this. The UE is a channel feature extraction model with Y channel responses, and a total of Q channel features S1, S2, ..., S Q It is possible to obtain Q channel features that can be considered to correspond to Y channels. In this method, for every T channel responses, the channel feature extraction model obtains one channel feature S corresponding to the T channel responses. Q This is used to obtain the channel feature extraction model. In actual processing, the UE may store one channel feature extraction model, and each of the Q inputs is sequentially inferred based on the channel feature extraction model. Alternatively, the UE may store multiple (e.g., Q) channel feature extraction models, and at least two of the Q inputs are inferred based on different channel feature extraction models. This is not limited to this.

[0247] UE has Q channel features S1, S2, ..., S in the LMF. Q Send. LMF is S1, S2, ..., S Q The input to the positioning information acquisition model is determined based on the input. The LMF acquires the positioning information of the UE through inference based on the input and the positioning information acquisition model. For example, the LMF uses channel features S1, S2, ..., S Q Based on the N inputs of the positioning information acquisition model, Position_model_In n The LMF obtains N inputs to the positioning information acquisition model, where n is an integer greater than or equal to 1, and n is a value between 1 and N. n Input, and through inference, obtain N outputs Position_model_Out n Obtain the following: The value of n ranges from 1 to N. Position_model_In n The corresponding output is Position_model_Out n That is the case.

[0248] Similar to the explanation in Figure 9A, N=1 and Position_model_Out n This indicates the location information of the UE. Alternatively, N is greater than or equal to 1, and Position_model_Out n , n=1, ..., N represents the channel type and / or first route location for each of the Y channels. The channel type and / or first route location are used to help LMF determine the location information of the UE. Position_model_Out n The explanation is the same as that in Figure 9A. Details are again not explained. In Figures 9A and 9C, the inputs and outputs of the channel feature extraction model are different, but the processing idea is the same.

[0249] Figures 8A, 8B, 8C, 9A, 9B, and 9C are used to illustrate the architecture of the channel feature extraction model and the positioning information acquisition model and are not intended to limit the present disclosure.

[0250] Figure 12 is a flowchart of the second positioning method according to this disclosure. This method includes the following operations.

[0251] Optionally, in operation S1201: the base station sends training data to the AI ​​function network element. Optionally, in operation S1202: the LMF sends training data to the AI ​​function network element. Optionally, in operation S1203: the AI ​​function network element performs model training using the training data.

[0252] AI functional network elements can acquire channel feature extraction models and positioning information acquisition models by undergoing training through a model training process using training data. Training data can be sent to the AI ​​functional network elements by base stations and / or LMFs.

[0253] In this disclosure, when a base station exchanges information with an AI functional network element, for example, when the base station sends training data to the AI ​​functional network element, or when the AI ​​functional network element sends information about a channel feature extraction model to the base station, the AI ​​functional network element communicates with the base station either directly by wired or wireless means, or by transmission through another network element (e.g., a core network element, but not limited to).

[0254] Operations S1201, S1202, and S1203 are optional. For example, the channel feature extraction model and the positioning information acquisition model are agreed upon in the protocol after offline training. Alternatively, the channel feature extraction model and the positioning information acquisition model are downloaded from a third-party website. In this case, S1201, S1202, and S1203 may be skipped.

[0255] Optionally, operations S1201 and S1203 may be performed but S1202 may not be performed, or operations S1202 and S1203 may be performed but S1201 may not be performed, or operations S1201, S1202, and S1203 may all be performed. In operation S1202, training data sent to the AI ​​functional network element by the base station may be measured by the base station and / or reported to the base station by the UE using signaling. In operation S1202, training data sent to the AI ​​functional network element by the LMF may be sent to the LMF by the UE and / or the base station.

[0256] The training data acquired by the AI ​​functional network elements includes training data used to determine the inputs of the channel feature extraction model and the labels corresponding to those inputs. The labels corresponding to the inputs can be understood as the target output or ideal output of the positioning information acquisition model. For example, when the inputs of the channel feature extraction model are determined based on training data A, the target output of the positioning information acquisition model is the labels corresponding to training data A.

[0257] Optionally, the training data used to determine the input for the channel feature extraction model includes at least one of the following: - Channel response between the UE and the base station. Optionally, information is measured by the UE and then sent to the base station or LMF, which then sends it to the AI ​​functional network element. For specific measurement methods, please refer to the corresponding explanation in Figure 6. Optionally, information is measured by the base station and then sent to the LMF or AI functional network element, or optionally, sent to the AI ​​functional network element by the LMF. For specific measurement methods, please refer to the corresponding explanation in Figure 6. - RSRP for the channel between the UE and the base station. Optionally, information is measured by the UE and then sent to the LMF or base station, which then sends it to the AI ​​functional network element. For specific measurement methods, please refer to the corresponding explanation in Figure 6. Optionally, information is measured by the base station and then sent to the LMF or AI functional network element, or optionally, sent to the AI ​​functional network element by the LMF. For specific measurement methods, please refer to the corresponding explanation in Figure 6. - RSTP channel between the UE and the base station. The information is measured by the UE and then sent to the LMF or base station, which then sends it to the AI ​​functional network elements. For specific measurement methods, please refer to the corresponding explanation in Figure 6. - The type of environment in which the UE is located. For a description of the environment types, please refer to the corresponding description in Figure 6. Information may be reported by the UE to the LMF or base station, sent to the AI ​​functional network element by the LMF or base station, or information may be sent to the AI ​​functional network element or LMF by the base station, and optionally sent to the AI ​​functional network element by the LMF. Optionally, the target output of the positioning information acquisition model in the training data includes at least one of the following: - UE location information. For an explanation of the location information, please refer to the corresponding explanation in Figure 6. Optionally, the UE is a beacon UE. The UE knows its location information and reports it to the LMF or base station, which then sends the location information to the AI ​​functional network element. Alternatively, the base station knows the UE's location information and sends the location information to the AI ​​functional network element or LMF, which then optionally sends the location information to the AI ​​functional network element. Alternatively, the LMF knows the UE's location information and sends the location information to the AI ​​functional network element. Optionally, the LMF obtains the UE's location information using a non-AI-based positioning method and sends this location information to the AI-enabled network element. For a description of the non-AI-based positioning method, please refer to the corresponding explanation in Figure 6. - The type of channel between the UE and the base station. Optionally, an example is used where P is equal to Y, and UE is a beacon UE. UE knows the channel type of the channel between UE and base station y, UE sends the channel type to LMF or base station, UE sends a total of Y channel types, and then LMF or base station sends Y channel types to AI functional network element. Alternatively, base station y knows the channel type of the channel between UE and base station y, base station y sends the channel type to AI functional network element or LMF, a total of Y base stations send the channel type to AI functional network element or LMF, then AI functional network element or LMF receives a total of Y channel types, and optionally, LMF sends Y channel types to AI functional network element. Alternatively, LMF knows the channel type of the channel between UE and base station y, LMF sends the channel type to AI functional network element, and LMF sends a total of Y channel types. The value of y ranges from 1 to Y. Similar to the corresponding explanation in Figure 6, this method is the same when P is not equal to Y. Optionally, an example is used where P is equal to Y. The LMF obtains the channel type of the channel between the UE and the yth base station by using a non-AI-based positioning method and sends the channel type to the AI-enabled network element. The LMF sends a total of Y channel types. The value of y ranges from 1 to Y. For a description of the non-AI-based positioning method, see the corresponding description in Figure 6. As with the corresponding description in Figure 6, the method is the same when P is not equal to Y. - The first route location of the channel between the UE and the base station. The method for determining and reporting the first route location of the channel between the UE and the base station is the same as described above in "Channel Types Between the UE and the Base Station," where the channel type is replaced by the first route location. Further details are again not provided herein.

[0258] As described in the corresponding section of Figure 6, the AI ​​functional network element acquires multiple training data corresponding to at least one UE. These multiple training data can be considered a single training dataset. Model training is performed using the training dataset to acquire a channel feature extraction model and a positioning information acquisition model.

[0259] In this disclosure, as in the corresponding description in Figure 6, the UE configured to perform model training and the UE configured to perform model inference may be the same or different (as shown in the figures of this disclosure), but are not limited to this. For example, the UE described in operations S1201, S1202, and S1203 are the first UE, and the UE described in operations S1207, S1208, and S1209 are the second UE. The first UE and the second UE may be the same or different. Regardless of whether the first UE and the second UE are the same, the base station configured to perform model training and the base station configured to perform model inference may be the same or different (as shown in the figures of this disclosure), but are not limited to this. For example, the model training process (e.g., operation S1201) relates to a first group of P base stations, and the model application process (e.g., operation S1207) relates to a second group of P base stations. For descriptions of the first group of P base stations and the second group of P base stations, please refer to the corresponding descriptions in Figure 6. Further details are again not provided herein. As with the corresponding descriptions in Figure 6, the AI ​​functional network elements may perform model training or model update training by using the training dataset. In the method shown in Figure 12, the process by which the AI ​​functional network elements perform model training or model update training by using the training dataset is the same as that shown in Figure 8A. Further details are again not provided herein.

[0260] Optionally, as described in Figure 6, multiple pairs of models (channel feature extraction model + positioning information acquisition model) may be trained, and the appropriate model is selected and applied as needed. The AI ​​model can be selected by taking into account the actual application scenario. For example, a more complex AI model may be used for base stations with high computing power, such as macro base stations. A slightly simpler AI model or an AI model with fewer parameters may be used for base stations with low computing power, such as small cells or micro base stations. Accordingly, the appropriate AI model may also be selected based on the computing power of the LMF.

[0261] Optionally, in operation S1204: the base station sends base station capability information to the LMF.

[0262] Base station capability information indicates at least one of the following pieces of information about the base station: - Will the base station support AI-based positioning methods? For example, when a base station supports an AI-based positioning method, the LMF may determine the UE's positioning information by using the AI-based positioning method. For example, to perform positioning, the base station uses the channel feature extraction model provided in this disclosure, and the LMF uses the positioning information acquisition model provided in this disclosure. When a base station does not support an AI-based positioning method, the LMF may determine the UE's location information by using a non-AI-based positioning method. The question "whether the base station supports AI-based positioning methods" can be replaced with the type of positioning method supported by the base station. The type can be an AI-based positioning method or a non-AI-based positioning method. - Information on the computing capabilities of the base station. Similar to the computing capability information of the UE in the method shown in Figure 6, the UE is replaced by a base station. - An AI model that supports base station capability information. Similar to the AI ​​model supporting the UE capability information in the method shown in Figure 6, the UE can be replaced with a base station.

[0263] In this disclosure, the following methods may be used when a base station sends information to an LMF, for example, when sending base station capability information.

[0264] Optionally, when a base station sends information to an LMF, if there is an interface between the base station and the LMF, the base station can send information to the LMF. If there is no interface between the base station and the LMF, the LMF can send information to a core network element E (for example, an AMF or another network element), and the core network element E then forwards the information to the LMF. The core network element E can be a single network element (one-hop forwarding) or multiple network elements (multi-hop forwarding).

[0265] In this disclosure, when an LMF sends information to a base station, for example, information about a channel feature extraction model, the LMF may send the information by using a route opposite to the method described above. For example, when an LMF sends information to a base station, if there is an interface between the base station and the LMF, the LMF sends the information to the base station. If there is no interface between the base station and the LMF, the LMF may send the information to a core network element E (for example, an AMF or another network element), and the core network element E then forwards the information to the base station.

[0266] Optionally, a base station sending information to the LMF may be a query (or request) based on the LMF. For example, the LMF sends base station capability query information to the base station. After receiving the query information, the base station reports its capability information to the LMF. Alternatively, the base station may pre-send its capability information to the LMF. For example, during network deployment, the base station may pre-send its capability information to the LMF, or send it to the LMF when its capability information changes. This is not limited to these cases.

[0267] Operation S1204 is optional. For example, if the capabilities of a base station are agreed upon in the protocol, S1204 does not need to be performed. For example, a base station may perform S1204 if there are base stations of multiple capability types in the system, or if the capabilities of a base station can change autonomously.

[0268] Optionally, operation S1205 (not shown) configures a positioning information acquisition model. Optionally, operation S1206 (not shown) configures a channel feature extraction model. S1205 and S1206 can be collectively referred to as the model configuration.

[0269] The model configuration can be implemented by using either Model Configuration Method 1 or Model Configuration Method 2 as described below.

[0270] Model configuration method 1:

[0271] Optionally, operation S1205a: The AI ​​function network element sends information about the positioning information acquisition model to the LMF. Optionally, operation S1206a: The AI ​​function network element sends information about the channel feature extraction model to the base station.

[0272] Optionally, the AI ​​functional network element and the base station are connected by a wired or wireless method, or they communicate with each other through transmission by another network element (e.g., a core network element, but not limited to one).

[0273] For an explanation of the optionality of operations S1205a and S1206a, please refer to the corresponding explanation in Figure 6. Further details are again not provided herein.

[0274] Model configuration method 2:

[0275] Optionally, in operation S1205b: the AI ​​function network element sends information about the positioning information acquisition model and information about the channel feature extraction model to the LMF. In operation S1206b: the LMF sends information about the channel feature extraction model to the base station.

[0276] For an explanation of the optionality of operations S1205b and S1206b, please refer to the corresponding explanation in Figure 6. Further details are again not provided herein.

[0277] Operation S1207: The base station extracts channel features using a channel feature extraction model. Operation S1208: The base station sends the channel features to the LMF. Operation S1209: The LMF determines the positioning information of the UE using a positioning information acquisition model.

[0278] For example, as shown in Figure 7B, an example is used in which P is equal to Y and the UE sends one reference signal to each base station. The UE sends uplink reference signals to Y base stations (the example shown in Figure 7B, with three, is used as an example for illustrative purposes). Each of the Y base stations receives an uplink reference signal. Y is an integer greater than or equal to 1, and is typically an integer greater than 1, such as 3, 4, 5, or a number greater than 1. After receiving the uplink reference signal, the y-th base station among the Y base stations determines the channel response H of the channel through which the reference signal passes, based on the transmission sequence value of the reference signal known to the y-th base station and the sequence value of the reference signal received by the base station. y The channel response H can be estimated or calculated. y P corresponds to the channel between the UE and the y-th base station. The value of y ranges from 1 to Y. The case where P is not equal to Y is similar to the description above. Further details are again not described herein. When P is not equal to Y, the P base stations obtain a total of Y channel responses for the Y channels between the P base stations and the UE through estimation.

[0279] After obtaining E channel responses corresponding to each base station, each base station may obtain E channel features based on the channel responses and a channel feature extraction model, where each channel feature corresponds to one channel response. The base station sends E channel features to the LMF. For each base station, the method can be described as follows: The base station sends E channel features to the LMF, where the E channel features correspond to E channels between the UE and the base station, each channel feature is obtained by using a channel feature extraction model, the input to the channel feature extraction model is determined based on the channel responses, and the channel responses correspond to the channels corresponding to the channel features. E is an integer greater than or equal to 1, and the value of E corresponding to different base stations may be the same or different. A total of P base stations send y channel features to the LMF.

[0280] The LMF receives a total of Y channel features from P base stations and determines the UE's positioning information by using the Y channel features and a positioning information acquisition model. For example, the LMF determines the input to the positioning information acquisition model based on the Y channel features and obtains the UE's positioning information through inference by using the positioning information acquisition model.

[0281] In this method, positioning is performed using an AI model to implement intelligent positioning, and therefore, positioning becomes closer to the actual channel environment, thereby implementing more accurate positioning.

[0282] Optionally, the dimension of the channel features sent to the LMF by the base station is smaller than the dimension of the corresponding channel response. This method can reduce the signaling overhead between the base station and the LMF.

[0283] An example is used where P is equal to Y. Figure 9A is a structural diagram showing how the base station and LMF perform positioning by using a channel feature extraction model and a positioning information acquisition model.

[0284] As shown in Figure 9A, the yth base station (or base station y) has a channel response H y Obtain H y Based on the input Feature_model_In of the channel feature extraction model y Determine the value of y, where the value of y ranges from 1 to Y. The base station y is Feature_model_In y Based on the base station's channel feature extraction model, the output of the channel feature extraction model is Feature_model_Out. y Obtain Feature_model_Out y This is the channel feature S y This indicates that the base station uses S in LMF. y The LMF sends a total of Y channel features from Y base stations. The Y channel features are S1, S2, ..., S Y It is shown as follows. LMF is channel features S1, S2, ..., S Y Based on this, the input to the positioning information acquisition model is obtained, and the positioning information of the UE is obtained through inference by using the positioning information acquisition model. LMF is S1, S2, ..., S Y The method for determining the UE's positioning and location information based on this is the same as the corresponding explanation in Figure 6. Further details are not provided again.

[0285] In the methods shown in Figures 6 and 12, training data is collected, and the channel feature extraction model and the positioning information acquisition model are trained using the collected data to obtain positioning information. The disclosure may further provide a reference dataset and a reference model (including the reference channel feature extraction model and the reference positioning information acquisition model). The UE, base station, and LMF may perform inference using the reference model to obtain positioning information for the UE. Alternatively, the UE or base station may perform update training on the reference channel feature extraction model using an offline or online training method to obtain an updated channel feature extraction model, and / or the LMF may perform update training on the reference positioning information acquisition model using an offline or online training method to obtain an updated positioning information acquisition model. Positioning information for the UE is obtained through inference using the updated channel feature extraction model and the reference positioning information acquisition model, or using the reference channel feature extraction model and the updated positioning information acquisition model, or using the updated channel feature extraction model and the updated positioning information acquisition model.

[0286] Figure 13 is a flowchart of the third positioning method according to this disclosure. This method includes the following operations.

[0287] Operation S1301: UE and LMF determine the first reference model.

[0288] The first reference model includes a first reference channel feature extraction model and a first reference positioning information acquisition model. The first reference model can be used by the UE and LMF to determine the positioning information of the UE.

[0289] In this disclosure, the first reference model may be determined by using one of the following methods:

[0290] Method 1 for determining the first reference model:

[0291] The first reference model is agreed upon in the protocol. The first reference model can be considered agreed upon in the protocol after offline training. Optionally, the reference dataset, loss function, and first threshold are agreed upon in the protocol.

[0292] For a detailed explanation of the reference dataset, loss function, and first threshold, please refer to operation S1302 below.

[0293] Method 2 for determining the first reference model:

[0294] The AI ​​Functional Network element obtains a first reference model through training by using the methods described in operations S601, S602, and S603. The AI ​​Functional Network element sends information about the first reference model to the UE and the LMF, or the AI ​​Functional Network element sends information about the first reference model to the LMF, and the LMF sends information about the first reference model to the UE. Optionally, the AI ​​Functional Network element further provides the UE and the LMF with at least one of the following: a reference dataset, a loss function, or a first threshold. For example, the AI ​​Functional Network element provides the reference dataset to the UE and the LMF, or the AI ​​Functional Network element provides the reference dataset to the LMF, and the LMF provides the first reference dataset to the UE, with the loss function and first threshold being agreed upon in the protocol. Other possible cases are not provided one by one.

[0295] Method 3 for determining the first standard model:

[0296] The UE determines a first reference model from W reference models, where W is an integer greater than or equal to 1. The UE sends the index of the first reference model to the LMF, which then determines the first reference model from the W reference models based on the index.

[0297] Optionally, the UE selects a first reference model from W reference models based on the UE's computing power information. For an explanation of the UE's computing power information, see the corresponding explanation in Figure 6. For example, the W reference models differ in structural complexity, with more complex structures requiring higher computing power. Based on the UE's computing power, the UE selects a first reference model from the W reference models that does not require higher computing power than the UE's computing power.

[0298] Optionally, each of the W reference models (referred to as reference model A) corresponds to a group of parameter configurations, each group of parameter configurations including at least one of the following: transmit antenna port quantity corresponding to channel response, receive antenna port quantity corresponding to channel response, bandwidth corresponding to channel response, or OFDM symbol quantity corresponding to channel response. The channel response is used to determine the input to the channel feature extraction model of reference model A. As shown in some examples in Tables 1 to 5, there is at least one different parameter configuration among two groups of parameter configurations corresponding to two different reference models. These examples do not constitute a limitation to this disclosure. The UE selects an appropriate first reference model from the W reference models based on the UE's parameter configuration (e.g., transmit antenna port quantity, receive antenna port quantity, bandwidth capability, and / or maximum amount of time-domain OFDM symbols processed), or the UE selects an appropriate first reference model from the W reference models based on the UE's measurement configuration (e.g., measured antenna port quantity, measured bandwidth, and / or measured amount of time-domain OFDM symbols). [Table 1] [Table 2] [Table 3] [Table 4] [Table 5]

[0299] Optionally, each of the W reference models (referred to as Reference Model A) corresponds to a group of application scenarios, each group of application scenarios including at least one of the following scenarios: eMBB, URLLC, mMTC, D2D, V2X, MTC, IoT, virtual reality, augmented reality, industrial control, autonomous driving, telemedicine, smart grid, smart furniture, smart office, smart wearable, intelligent transport, smart city, etc. As shown in Table 6, there is at least one different application scenario within the group of application scenarios corresponding to each reference model. The UE selects a first reference model from the W reference models based on the UE's application scenarios. [Table 6]

[0300] Optionally, each of the W reference models (referred to as Reference Model A) corresponds to a group of application environments, each group of application environments including at least one of the following environments: factory environment, office environment, high-speed rail environment, subway environment, etc. As shown in Table 7, each reference model has at least one different application environment within its group of application environments. The UE selects a first reference model from the W reference models based on the environment in which the UE is located. [Table 7]

[0301] Optionally, W reference models may be agreed upon in the protocol. Optionally, a reference dataset, loss function, and first threshold corresponding to each reference model may be further agreed upon in the protocol. Optionally, the W reference models may correspond to the same loss function and / or the same first threshold.

[0302] Optionally, W reference models may be indicated to the UE and LMF by AI Functional Network elements. The indication method is similar to the corresponding description in the first reference model determination method 2. Optionally, the AI ​​Functional Network elements further indicate to the UE and LMF at least one of the following: a reference dataset, a loss function, or a first threshold, corresponding to each of the reference models. Optionally, the loss function and / or first threshold corresponding to each of the reference models are agreed upon in the protocol. Optionally, the W reference models correspond to the same loss function and / or the same first threshold.

[0303] Optionally, operation S1302:UE performs model update training based on the first reference model and the first reference dataset to obtain an updated channel feature extraction model.

[0304] The first reference dataset includes multiple training data and corresponding labels for each training data. For explanations of the training data and labels, see the corresponding explanations in Figure 6. The UE obtains an updated channel feature extraction model through training using the model training method described in operation S603. In operation S603, training data is collected from the UE and / or LMF, and during the training process, the parameters of both the channel feature extraction model and the positioning information acquisition model may be updated. In operation S1302, as shown in Figure 14A, the training data is provided by the reference dataset, and during the training process, the parameters of the channel feature extraction model may be updated, where the initial model of the channel feature extraction model is the first reference channel feature extraction model, but the parameters of the positioning information acquisition model are not updated; in other words, the positioning information acquisition model is the first reference positioning information acquisition model. During the training process, iterative training is performed on the first reference channel feature extraction model using the training data in the reference dataset to obtain an updated channel feature extraction model. The purpose of training is to determine the inputs to the updated channel feature extraction model by using training data in a reference dataset, and to ensure that when the updated channel feature extraction model and the first reference positioning information acquisition model are used in a consistent manner, the loss function between the output of the first reference positioning information acquisition model and the labels in the training data is less than or equal to a first threshold.

[0305] Optionally, operation S1303:LMF performs model update training based on the first reference model and the first reference dataset to obtain an updated positioning information acquisition model.

[0306] The first reference dataset includes multiple training data and corresponding labels for each training data. For explanations of the training data and labels, see the corresponding explanations in Figure 6. The LMF obtains a positioning information acquisition model updated through training by using the model training method described in operation S603. In operation S603, training data is collected from the UE and / or LMF, and during the training process, the parameters of both the channel feature extraction model and the positioning information acquisition model may be updated. In operation S1303, as shown in Figure 14B, the training data is provided by the reference dataset, and during the training process, the parameters of the channel feature extraction model are not updated; in other words, the channel feature extraction model is the first reference channel feature extraction model, but the positioning information acquisition model may be updated, where the initial model of the positioning information acquisition model is the first reference positioning information acquisition model. During the training process, iterative training is performed on the first reference positioning information acquisition model using the training data in the reference dataset to obtain an updated positioning information acquisition model. The purpose of training is to determine the input to the first reference channel feature extraction model by using training data in the reference dataset, and to ensure that when the first reference channel feature extraction model and the updated positioning information acquisition model are used in a consistent manner, the loss function between the output of the updated positioning information acquisition model and the labels in the training data is less than or equal to the first threshold.

[0307] Given that there are W reference models of arbitrary choice, the LMF may first determine a first reference model and then train a channel feature extraction model based on the first reference model. Alternatively, the LMF may obtain a pre-determined corresponding channel feature extraction model through training based on each of the W reference models. When the first reference model is determined, a channel feature extraction model trained based on the first reference model may be obtained.

[0308] Operation S1304: UE extracts channel features using a channel feature extraction model. Operation S1305: UE sends the channel features to the LMF.

[0309] In this operation, if S1302 is not performed, the channel feature extraction model is either the first reference channel feature extraction model, or if S1302 is performed, the channel feature extraction model is the updated channel feature extraction model.

[0310] Given that there are W reference models of any choice, the UE may first determine a first reference model and then train a channel feature extraction model based on the first reference model. Alternatively, the UE may obtain a pre-determined corresponding channel feature extraction model through training based on each of the W reference models. When the first reference model is determined, a channel feature extraction model trained on the first reference model may be obtained.

[0311] Specific implementations of S1304 and S1305 are similar to the corresponding descriptions in Figure 6. Further details are again not described herein.

[0312] Operation S1306: The LMF determines the positioning information of the UE by using the positioning information acquisition model.

[0313] In this operation, if S1303 is not performed, the positioning information acquisition model is the first reference positioning information acquisition model, or if S1303 is performed, the positioning information acquisition model is the updated positioning information acquisition model. A specific implementation of S1306 is similar to the corresponding description in Figure 6. Further details are again not described herein.

[0314] Figure 15 is a flowchart of the fourth positioning method according to this disclosure. This method includes the following operations.

[0315] Operation S1501: The base station and LMF determine the first reference model.

[0316] The first reference model includes a first reference channel feature extraction model and a first reference positioning information acquisition model. The first reference model is used by the base station and the LMF to determine the positioning information of the UE.

[0317] The base station and LMF can determine the first reference model by using one of the following methods:

[0318] Method 1 for determining the first reference model: Same as the corresponding explanation in Figure 13.

[0319] Method 2 for determining the first reference model: Similar to the corresponding explanation in Figure 13, where UE is replaced with base station.

[0320] Method 3 for determining the first standard model:

[0321] The base station determines a first reference model from W reference models, where W is an integer greater than 1. The base station sends the index of the first reference model to the LMF, which then determines the first reference model from the W reference models based on the index.

[0322] Optionally, the base station selects a first reference model from W reference models based on the base station's computing capability information. For an explanation of the base station's computing capability information, please refer to the corresponding explanation in Figure 12. For example, the W reference models differ in their structural complexity, with more complex structures requiring higher computing power. Based on the base station's computing power, the base station selects a first reference model from the W reference models that does not require higher computing power than the base station's computing power.

[0323] Optionally, each of the W reference models (referred to as reference model A) corresponds to a group of parameter configurations, each group of parameter configurations containing at least one of the following: transmit antenna port quantity corresponding to the channel response, receive antenna port quantity corresponding to the channel response, bandwidth corresponding to the channel response, or OFDM symbol quantity corresponding to the channel response. The channel response is used to determine the input to the channel feature extraction model of reference model A. As shown in some examples in Tables 1 to 5, there is at least one different parameter configuration in the group of parameter configurations corresponding to each reference model. The base station selects an appropriate first reference model from the W reference models based on the base station's parameter configurations (e.g., transmit antenna port quantity, receive antenna port quantity, bandwidth capability, and / or maximum amount of time-domain OFDM symbols processed), or the base station selects an appropriate first reference model from the W reference models based on the base station's measured information (e.g., measured antenna port quantity, measured bandwidth, and / or measured amount of time-domain OFDM symbols).

[0324] Optionally, each of the W reference models (indicated as Reference Model A) corresponds to a group of UE application scenarios. For descriptions of the application scenarios, see the corresponding descriptions in Figure 13. As shown in Table 6, there is at least one different application scenario within the group of application scenarios corresponding to each reference model. The base station selects a first reference model from the W reference models based on the UE application scenarios. Optionally, the UE application scenarios learned by the base station are determined, for example, based on the UE service type or reported to the base station by the UE, but are not limited to these.

[0325] Optionally, each of the W reference models (denoted as Reference Model A) corresponds to a group of application environments. For descriptions of application scenarios, see the corresponding descriptions in Figure 13. As shown in Table 7, there is at least one different application environment within the group of application environments corresponding to each reference model. The base station selects a first reference model from the W reference models based on the environment in which the UE is located. Optionally, the environment in which the UE is located, as learned by the base station, may be obtained, for example, through estimation based on measurement information corresponding to the UE, or reported to the base station by the UE. This is not limited to these methods.

[0326] Optionally, W reference models may be agreed upon in the protocol. Optionally, a reference dataset, loss function, and first threshold corresponding to each reference model may be further agreed upon in the protocol. Optionally, the W reference models may correspond to the same loss function and / or the same first threshold.

[0327] Optionally, W reference models may be indicated to the base station and LMF by an AI functional network element. The indication method is similar to the corresponding description in the first reference model determination method 2. Optionally, the AI ​​functional network element further indicates to the base station and LMF at least one of the following: a reference dataset, a loss function, or a first threshold, corresponding to each of the reference models. Optionally, the loss function and / or first threshold corresponding to each of the reference models are agreed upon in the protocol. Optionally, the W reference models correspond to the same loss function and / or the same first threshold.

[0328] Optionally, in operation S1502, the base station performs model update training based on the first reference model and the first reference dataset to obtain an updated channel feature extraction model.

[0329] The first reference dataset includes multiple training data and corresponding labels for each of the training data. For a description of the training data and labels, see the corresponding description in Figure 12. As shown in Figure 14A, the base station obtains an updated channel feature extraction model through training using the model training method described in operation S1203. The difference between operations S1502 and S1203 is similar to the difference between operations S1302 and S603. Further details are again not described herein.

[0330] Optionally, operation S1503:LMF performs model update training based on the first reference model and the first reference dataset to obtain an updated channel feature extraction model.

[0331] The first reference dataset includes multiple training data and corresponding labels for each of the training data. See the corresponding descriptions in Figure 12 for explanations of the training data and labels. As shown in Figure 14B, the LMF obtains an updated channel feature extraction model through training using the model training method described in operation S1203. The difference between operation S1503 and S1203 is similar to the difference between operation S1303 and S603. Further details are again not described herein.

[0332] Operation S1504: The base station extracts channel features using a channel feature extraction model. Operation S1505: The base station sends the channel features to the LMF.

[0333] In this operation, if S1502 is not performed, the channel feature extraction model is either the first reference channel feature extraction model, or if S1502 is performed, the channel feature extraction model is the updated channel feature extraction model.

[0334] Specific implementations of S1304 and S1305 are similar to the corresponding descriptions in Figure 12. Further details are again not described herein.

[0335] Operation S1506: The LMF determines the positioning information of the UE by using the positioning information acquisition model.

[0336] In this operation, if S1503 is not performed, the positioning information acquisition model is the first reference positioning information acquisition model, or if S1503 is performed, the positioning information acquisition model is the updated positioning information acquisition model. A specific implementation of S1506 is similar to the corresponding description in Figure 12. Further details are again not described herein.

[0337] To implement the functions in the above method, it can be understood that the base station, UE, LMF, and AI function network elements include corresponding hardware structures and / or software modules for performing the functions. Those skilled in the art will readily recognize, by referring to the units and method steps in the examples described herein, that this application can be implemented by hardware or a combination of hardware and computer software. Whether the functions are performed by hardware or by hardware driven by computer software depends on the specific application scenario and the design constraints of the technical solution.

[0338] Figures 16 and 17 are schematic diagrams of possible communication device structures according to this disclosure. These communication devices may be configured to implement the functions of the base station, UE, LMF, and AI functional network elements in the above method, and thus can also implement the beneficial effects of the above method.

[0339] As shown in Figure 16, the communication device 900 includes a processing unit 910 and a transceiver unit 920. The communication device 900 is configured to implement the methods shown in Figures 6, 12, 13, or 15.

[0340] When the communication device 900 is configured to implement the functions of the UE in the manner shown in Figure 6 or Figure 13, the transceiver unit 920 is configured to send X channel features to the LMF, and the processing unit 910 is configured to obtain X channel features based on Y channel responses and a channel feature extraction model. X and Y are integers greater than or equal to 1. The X channel features correspond to Y channels between the UE and P base stations, and the Y channel responses are in a one-to-one correspondence with the Y channels.

[0341] When the communication device 900 is configured to implement the functions of a base station in the manner shown in Figure 12 or Figure 15, the transceiver unit 920 is configured to send E channel features to the LMF, where E is an integer greater than or equal to 1, and the processing unit 910 is configured to obtain channel features corresponding to each of the E channel responses based on a channel feature extraction model. The E channel features correspond to the E channels between the UE and the base station.

[0342] When the communication device 900 is configured to implement the LMF functions in the manner shown in Figure 6, Figure 12, Figure 13, or Figure 15, the transceiver unit 920 is configured to receive X channel features, where the X channel features correspond to Y channels between the UE and P base stations, and X and Y are integers greater than or equal to 1, and the processing unit 910 is configured to acquire positioning information of the UE based on the X channel features and a positioning information acquisition model.

[0343] When the communication device 900 is configured to implement the functions of the AI ​​functional network elements in the manner shown in Figure 6, Figure 12, Figure 13, or Figure 15, the transceiver unit 920 is configured to receive training data, and the processing unit 910 is configured to acquire a channel feature extraction model and a positioning information acquisition model through training based on the training data.

[0344] For a more detailed description of the processing unit 910 and the transceiver unit 920, please refer to the relevant descriptions in the methods shown in Figures 6, 12, 13, or 15 to obtain the explanation directly. Further details are again not provided herein.

[0345] As shown in Figure 17, the communication device 1000 includes a processor 1010 and an interface circuit 1020. The processor 1010 and the interface circuit 1020 are coupled to each other. It can be understood that the interface circuit 1020 may be a transceiver, a pin, or an input / output interface. Optionally, the communication device 1000 may further include a memory 1030 configured to store at least one of the following: instructions executed by the processor 1010, input data required by the processor 1010 to execute the instructions, or data generated after the processor 1010 has executed the instructions.

[0346] When the communication device 1000 is configured to implement the method described above, the processor 1010 is configured to implement the functions of the processing unit 910 described above, and the interface circuit 1020 is configured to implement the functions of the transceiver unit 920 described above.

[0347] When the communication device is a chip used in the UE, the UE chip implements the functions of the UE in the manner described above. The UE chip receives information from another module in the UE (e.g., a radio frequency module or antenna), where the information is sent to the UE by a base station, LMF, AI function network element, etc. Alternatively, the UE chip sends information to another module in the UE (e.g., a radio frequency module or antenna), where the information is sent to a base station, LMF, AI function network element, etc., by the UE.

[0348] When the communication device is a module used in a base station, the base station module implements the functions of the base station in the manner described above. The base station module receives information from another module in the base station (e.g., a radio frequency module or an antenna), where the information is sent to the base station by a UE, LMF, or AI functional network element. Alternatively, the base station module sends information to another module in the base station (e.g., a radio frequency module or an antenna), where the information is sent to a UE, LMF, AI functional network element, etc., by the base station. The base station module in this specification may be a baseband chip in the base station, or it may be a quasi-real-time RIC, CU, DU, or another module. The quasi-real-time RIC, CU, and DU in this specification may be a quasi-real-time RIC, CU, and DU in an O-RAN architecture.

[0349] When a communication device is used in an LMF module, the LMF module implements the functions of the LMF in the manner described above. The LMF module receives information from another module in the LMF (e.g., a radio frequency module or antenna), where the information is sent to a base station by the UE, base station, or AI functional network element. Alternatively, the LMF module sends information to another module in the LMF (e.g., a radio frequency module or antenna), where the information is sent to the UE, base station, AI functional network element, etc., by the LMF.

[0350] In this application, the processor may be a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field-programmable gate array or another programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component that can implement or carry out the methods, steps, and logic block diagrams of this disclosure. The general-purpose processor may be a microprocessor, any conventional processor, etc. The steps of the methods of this disclosure may be carried out directly by a hardware processor or by a combination of hardware and software modules in the processor.

[0351] In this application, memory may be non-volatile memory, such as a hard disk drive (HDD) or solid-state drive (SSD), or volatile memory, such as random access memory (RAM). Memory may be any other medium that can carry or store program code expected in the form of instructions or data structures and that can be accessed by a computer, but is not limited to such medium. Alternatively, memory in this application may be a circuit or any other device that can implement a storage function and is configured to store program instructions and / or data.

[0352] The methods described in this application may be implemented entirely or partially by software, hardware, firmware, or any combination thereof. When software is used to implement the methods, the methods may be implemented entirely or partially in the form of a computer program product. A computer program product includes one or more computer programs or instructions. When a computer program or instruction is loaded onto a computer and executed, the processing or function described in this application is performed entirely or partially. The computer may be a general-purpose computer, a dedicated computer, a computer network, an access network device, a terminal device, a core network device, an AI function network element, or another programmable device. The computer programs or instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, a computer program or instruction may be transmitted wired or wirelessly from one website, computer, server, or data center to another website, computer, server, or data center. The computer-readable storage medium may be any available medium accessible by a computer or a data storage device that integrates one or more available media, such as a server or data center. The usable media may be magnetic media, such as floppy disks, hard disk drives, or magnetic tapes; optical media, such as digital video discs; or semiconductor media, such as solid-state drives. Computer-readable storage media may be volatile or non-volatile storage media, or may include both volatile and non-volatile storage media.

[0353] The above description is merely a specific implementation of the present invention and is not intended to limit the scope of protection of the present invention. Any modification or substitution that is readily conceivable by a person skilled in the art within the technical scope disclosed herein shall fall within the scope of protection of the present invention. Accordingly, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A communication method performed by a terminal device or a chip system for said terminal device, The step includes sending X channel features to the Location Management Function (LMF), The X channel features correspond to the Y channels of the terminal device, the X channel features are obtained by using an artificial intelligence (AI) model called a channel feature extraction model, the input to the channel feature extraction model is determined based on the Y channel responses, the Y channel responses have a one-to-one correspondence with the Y channels, and X and Y are integers of 1 or greater. The aforementioned method, A communication method further comprising the step of determining the channel feature extraction model based on a first reference model, wherein the first reference model includes a first reference channel feature extraction model for mapping the channel response to the channel features.

2. The method according to claim 1, wherein X is less than or equal to Y, and / or the total dimension of the X channel features is less than the total dimension of the Y channel responses.

3. The method according to claim 1, wherein the X channel features are used to determine the inputs of a positioning information acquisition model for obtaining location information of the terminal device, and the output of the positioning information acquisition model includes positioning information of the terminal device.

4. The positioning information indicates the location information of the terminal device, or The positioning information indicates at least one of the following for the Y channels: a channel type including line of sight (LOS) or non-line of sight (NLOS), a first path location including the time-domain location of the first path in the channel in one OFDM symbol, or the relative azimuth angle of the terminal device. The method according to claim 3.

5. A communication method performed by an access network device or a chip system for said access network device, The process includes the step of sending E channel features to the Location Management Function (LMF), where E is an integer greater than or equal to 1. The E channel features correspond to the E channels between the terminal device and the access network device, each of the E channel features is obtained by using an artificial intelligence (AI) model, which is a channel feature extraction model, the input to the channel feature extraction model is determined based on the channel response, and the channel response corresponds to the channel corresponding to the channel feature. The aforementioned method, A communication method further comprising the step of determining the channel feature extraction model based on a first reference model, wherein the first reference model includes a first reference channel feature extraction model for mapping the channel response to the channel features.

6. The method according to claim 5, wherein the dimension of the channel feature is smaller than the dimension of the channel response.

7. The method according to claim 5, wherein the E channel features are used to determine the inputs of a positioning information acquisition model for obtaining location information of the terminal device, and the output of the positioning information acquisition model includes positioning information of the terminal device.

8. The positioning information indicates the location information of the terminal device, or The positioning information indicates at least one of the following for the E channels: a channel type including line of sight (LOS) or non-line of sight (NLOS), a first path location including the time-domain location of the first path in the channel in one OFDM symbol, or the relative azimuth angle of the terminal device. The method according to claim 7.

9. The method according to claim 8, wherein at least one of the channel type, the first route location, or the relative azimuth angle of the terminal device is used to determine the location information of the terminal device.

10. The aforementioned method, step of receiving information about the channel feature extraction model The method according to claim 5, further comprising:

11. The method according to claim 5, wherein the first reference model further includes a first reference positioning information acquisition model for mapping the channel features to positioning information of the terminal device, and update training is performed on the first reference channel feature extraction model and the first reference positioning information acquisition model, respectively.

12. The method according to claim 11, wherein when the channel feature extraction model and the first reference positioning information acquisition model are used, the value of the loss function between the output of the first reference positioning information acquisition model and the labels of the training data is less than or equal to a first threshold.

13. A communication method performed by the Location Management Function (LMF), A step of receiving X channel features, wherein the X channel features correspond to Y channels of a terminal device, and X and Y are positive integers of 1 or more. The steps include: acquiring positioning information of the terminal device based on the X channel features and a positioning information acquisition model which is an artificial intelligence (AI) model; Includes, The aforementioned method, A communication method further comprising the step of determining the positioning information acquisition model based on a first reference model, wherein the first reference model includes a first reference positioning information acquisition model for mapping the channel features to positioning information of the terminal device.

14. The method according to claim 13, wherein X is less than or equal to Y, and / or the total dimension of the X channel features is less than the total dimension of the Y channel responses.

15. The positioning information indicates the location information of the terminal device, or The positioning information indicates at least one of the following for the Y channels: a channel type including line of sight (LOS) or non-line of sight (NLOS), a first path location including the time-domain location of the first path in the channel in one OFDM symbol, or the relative azimuth angle of the terminal device. The method according to claim 13.

16. The step of determining the location information of the terminal device based on at least one of the channel type, the first route location, or the relative azimuth angle of the terminal device. The method according to claim 15, further comprising:

17. The aforementioned method, step of receiving information regarding the positioning information acquisition model The method according to claim 13, further comprising:

18. The method according to claim 13, wherein the first reference model further includes a first reference channel feature extraction model for mapping channel responses to channel features, and update training is performed on the first reference channel feature extraction model and the first reference positioning information acquisition model, respectively.

19. The method according to claim 18, wherein, when the positioning information acquisition model and the first reference channel feature extraction model are used, the value of the loss function between the output of the first reference positioning information acquisition model and the labels of the training data is less than or equal to a first threshold.

20. The aforementioned method, The process further includes the step of sending information about a channel feature extraction model, wherein the output of the channel feature extraction model includes the channel features. The method according to claim 13.

21. A communication method performed by an artificial intelligence (AI) model training device, A step of receiving a training dataset, wherein each piece of training data in the training dataset represents Y channel responses and Y positioning information for Y channels of a terminal device, and the Y channel responses are in a one-to-one correspondence with the Y positioning information. A step of obtaining a channel feature extraction model and a positioning information acquisition model through training based on the training dataset, wherein the channel feature extraction model and the positioning information acquisition model are artificial intelligence (AI) models, the input of the channel feature extraction model is determined based on at least one of the Y channel responses, and the input of the positioning information acquisition model is determined based on the output of the channel feature extraction model. Includes, The aforementioned method, A communication method further comprising the step of obtaining the channel feature extraction model and the positioning information acquisition model based on a first reference model, wherein the first reference model includes a first reference channel feature extraction model for mapping the channel response to the channel features and a first reference positioning information acquisition model for mapping the channel features to positioning information of the terminal device.

22. A communication device comprising a module configured to implement the method described in any one of claims 1 to 12.

23. A communication device comprising a processor and memory, wherein the processor is coupled to the memory, and the processor is configured to implement the method according to any one of claims 1 to 12.

24. A communication device comprising a module configured to implement the method described in any one of claims 13 to 20.

25. A communication device comprising a processor and memory, wherein the processor is coupled to the memory, and the processor is configured to implement the method according to any one of claims 13 to 20.

26. A communication device comprising a module configured to implement the method described in claim 21.

27. A communication device configured to implement the method described in any one of claims 1 to 12, and a communication device configured to implement the method described in any one of claims 13 to 20, or A communication device configured to implement the method described in any one of claims 1 to 12, and a communication device configured to implement the method described in claim 21, or A communication device configured to implement the method described in any one of claims 13 to 20, and a communication device configured to implement the method described in claim 21, or A communication device configured to implement the method according to any one of claims 1 to 12, a communication device configured to implement the method according to any one of claims 13 to 20, and a communication device configured to implement the method according to claim 21, Communication system.

28. A computer-readable storage medium, wherein the computer-readable storage medium stores instructions, and when the instructions are executed on a computer, the computer is enabled to carry out the method according to any one of claims 1 to 12, or the method according to any one of claims 13 to 20, or the method according to claim 21.

29. A computer program including an instruction, wherein when the instruction is executed on a computer, the instruction causes the computer to execute the method according to any one of claims 1 to 12, or the method according to any one of claims 13 to 20, or the method according to claim 21.

Images