WO2023143267A1 - 一种模型配置方法及装置 - Google Patents
一种模型配置方法及装置 Download PDFInfo
- Publication number
- WO2023143267A1 WO2023143267A1 PCT/CN2023/072691 CN2023072691W WO2023143267A1 WO 2023143267 A1 WO2023143267 A1 WO 2023143267A1 CN 2023072691 W CN2023072691 W CN 2023072691W WO 2023143267 A1 WO2023143267 A1 WO 2023143267A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- model
- information
- access network
- network device
- oam
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 143
- 238000013507 mapping Methods 0.000 claims abstract description 125
- 230000006870 function Effects 0.000 claims description 200
- 238000013473 artificial intelligence Methods 0.000 claims description 77
- 238000004891 communication Methods 0.000 claims description 74
- 238000012549 training Methods 0.000 claims description 68
- 230000008859 change Effects 0.000 claims description 49
- 230000015654 memory Effects 0.000 claims description 41
- 238000004590 computer program Methods 0.000 claims description 8
- 238000012423 maintenance Methods 0.000 abstract description 3
- 210000004027 cell Anatomy 0.000 description 90
- 238000013461 design Methods 0.000 description 61
- 230000000875 corresponding effect Effects 0.000 description 55
- 230000008569 process Effects 0.000 description 36
- 238000013528 artificial neural network Methods 0.000 description 32
- 238000012545 processing Methods 0.000 description 22
- 210000002569 neuron Anatomy 0.000 description 21
- 238000010586 diagram Methods 0.000 description 13
- 230000004913 activation Effects 0.000 description 11
- 238000004364 calculation method Methods 0.000 description 11
- 239000013256 coordination polymer Substances 0.000 description 11
- 238000005259 measurement Methods 0.000 description 11
- 238000010801 machine learning Methods 0.000 description 9
- 238000012360 testing method Methods 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 8
- 230000009286 beneficial effect Effects 0.000 description 6
- 230000005540 biological transmission Effects 0.000 description 6
- 238000012795 verification Methods 0.000 description 6
- 230000033001 locomotion Effects 0.000 description 5
- 238000007726 management method Methods 0.000 description 5
- 238000003672 processing method Methods 0.000 description 5
- 230000009471 action Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 3
- 230000003190 augmentative effect Effects 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 238000010295 mobile communication Methods 0.000 description 3
- 238000004171 remote diagnosis Methods 0.000 description 3
- 238000012706 support-vector machine Methods 0.000 description 3
- 230000006978 adaptation Effects 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 2
- 235000019580 granularity Nutrition 0.000 description 2
- 238000012417 linear regression Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000002787 reinforcement Effects 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 102100022734 Acyl carrier protein, mitochondrial Human genes 0.000 description 1
- 101000678845 Homo sapiens Acyl carrier protein, mitochondrial Proteins 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 210000005013 brain tissue Anatomy 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000001356 surgical procedure Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/445—Program loading or initiating
- G06F9/44505—Configuring for program initiating, e.g. using registry, configuration files
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/445—Program loading or initiating
Definitions
- the present application relates to the field of communication technologies, and in particular to a model configuration method and device.
- a wireless communication network such as a mobile communication network
- services supported by the network are becoming more and more diverse, and therefore requirements to be met are becoming more and more diverse.
- the network needs to be able to support ultra-high speed, ultra-low latency, and/or very large connections.
- This feature makes network planning, network configuration, and/or resource scheduling increasingly complex.
- These new requirements, new scenarios, and new features have brought unprecedented challenges to network planning, O&M, and efficient operations.
- artificial intelligence technology can be introduced into the wireless communication network to realize network intelligence. Based on this, how to configure the artificial intelligence model to the access network equipment in the network is a problem worth studying.
- the present application provides a model configuration method and device, so as to configure a first model matching an application scenario or a performance level for a first access network device, and improve the performance and speed of model prediction.
- a model configuration method is provided.
- the execution subject of the method is the operation management and maintenance OAM, and may also be a component (processor, chip or other) configured in the OAM, or may be a software module, etc., including:
- a first model request message from a first access network device; determine a first model according to the first model request message and a first mapping relationship, where the first mapping relationship includes a model, a model application scenario, and a model function or the first mapping relationship includes a mapping relationship between a model, a model performance level, and a model function; and sending information about the first model to the first access network device.
- the mapping relationship between the model, the model performance level, and the model function may be replaced by: the mapping relationship between the model, the model performance index, and the model function.
- the OAM stores a first mapping relationship, and the first mapping relationship is a mapping relationship of model+model application scenario+model function.
- the model corresponding to the scenario can be allocated, so that the model can meet the performance index requirements in the scenario corresponding to the model.
- the same model is used for model inference.
- the OAM establishes a first mapping relationship, where the first mapping relationship is a mapping relationship of model+model performance level (or model performance index)+model function.
- OAM can configure different models for different model performance indicators in different application scenarios according to the first mapping relationship.
- using the same model for model reasoning can improve prediction performance and prediction speed.
- the first model request message is used to indicate a model application scenario and a model function of the first model.
- the first model request message is used to indicate the model function and model performance level of the first model, or to indicate the model function and model performance index of the first model.
- the information of the first model is used to indicate at least one of the following of the first model: model index, model structure information, model parameters, model input format, model output format, model Performance indicators, model application scenarios, model functions, or training parameters.
- the OAM sends the information of the first model to the first access network device.
- the first access network device may determine or restore the first model according to the information of the first model.
- the information of the first model may include training parameters, and the first access network device may continue to train the first model according to the training parameters to meet various requirements for the model.
- the method further includes: receiving first feedback information from the first access network device; and updating the first model according to the first feedback information.
- the first feedback information is used to indicate the actual accuracy rate of the prediction information output by the first model
- the updating of the first model according to the feedback information includes: When the actual accuracy rate of the prediction information output by the first model is less than a threshold, retrain the first model and update the first model.
- the first access network device may perform model reasoning according to the first model, and the reasoning result may be called prediction information.
- the first feedback information is determined according to the prediction information, the first feedback information includes the actual accuracy rate of the prediction information output by the first model, and the first feedback information is fed back to the OAM; the OAM calculates the first model according to the first feedback information Updates are performed to improve the accuracy of the inference results of the first model.
- the first feedback information is used to indicate a scenario change indication, a model performance level change indication, or a model performance index change indication
- the first model Updating including: selecting a second model for the terminal device according to the scene change indication, model performance level change indication, or model performance index change indication; sending information about the second model to the first access network device , the second model is used to update the first model.
- the difference is that in this design, when the application scenario, model performance level or model performance index changes, the first access network device sends a message to the OAM indicating the application scenario change, model performance level change or model For the indication information of the performance index change, the OAM re-assigns the model to the first access network device for model inference according to the changed application scenario, model performance level, or model performance index, which can ensure that the first access network device The assigned model meets the requirements of the application scenario, model performance level, or model performance index.
- the method further includes: receiving a second model request message from the second access network device, where the second model request message is used to indicate the index of the first model; according to the first model an index of the first model, and determine the first model; and send the information of the first model to the second access network device.
- the second aspect provides a model configuration method, the method corresponds to the first access network device side of the first aspect, the beneficial effect can be referred to the first aspect above, the method is executed by the first access network device, and It may be a component (processor, chip or other) configured in the first access network device, or may be a software module, etc., including:
- the first model request message before sending the first model request message to OAM, it also includes: determining whether To provide enhanced services based on artificial intelligence AI.
- the information of the first model is used to indicate at least one of the following of the first model: model index, model structure information, model parameters, model input format, model output format, model Performance indicators, model application scenarios, model functions, or training parameters.
- the method further includes: performing model reasoning according to the first model to obtain prediction information.
- the method further includes: sending the prediction information to the terminal device, where the prediction information includes a prediction result.
- the prediction information also includes a prediction accuracy rate of the prediction result.
- the method further includes: receiving second feedback information from the terminal device, where the second feedback information is used to indicate the actual accuracy rate of the prediction information.
- the method further includes: sending first feedback information to the OAM, where the first feedback information is used to indicate the actual accuracy rate of the prediction information, an indication of an application scenario change, an indication of a model performance level change, or Model performance indicator change indication.
- the above-mentioned first model may provide artificial intelligence AI or machine learning ML reasoning service for the terminal device, or provide AI or ML reasoning service for the first access network device.
- the first model when used to provide AI or ML reasoning services for the terminal device, it also includes: the first RAN node sends the reasoning information (or prediction information) of the first model to the terminal device; the terminal device, according to the reasoning information, Sending second feedback information to the first access network device, where the second feedback information may be used to indicate the actual accuracy of the reasoning information, and the first access network device may directly forward the second feedback information to the OAM; or, the second An access network device acquires the actual accuracy rate of the reasoning information according to the second feedback information, and sends the first feedback information indicating the actual accuracy rate of the reasoning information to the OAM.
- the OAM can update the first model according to the actual accuracy of inference information, and improve the inference performance and inference speed of the first model.
- the method further includes: determining a target cell of the terminal device; and sending an index of the first model to a second access network device corresponding to the target cell.
- the method further includes: sending first indication information to the second access network device corresponding to the target cell, where the first indication information is used to indicate the type information of the input data of the first model .
- the method further includes: sending historical data information required by the model reasoning to the second access network device.
- the method further includes: receiving second indication information from the second access network device, where the second indication information is used to indicate at least one of the following: the terminal is handed over to the target cell .
- the first access network device deletes the first model, or the first access network device releases corresponding computing resources; deleting the first model releases corresponding computing resources.
- a model configuration method is provided.
- the first access network device side of the method corresponds to the first aspect.
- the execution subject of the method is the second access network device, and It may be a component (processor, chip or other) configured in the second access network device, or may be a software module, etc., including:
- the information of the first model is used to indicate at least one of the following of the first model: model index, model structure information, model parameters, model input format, model output format, model Performance indicators, model application scenarios, model functions, or training parameters.
- the method further includes: receiving first indication information from the first access network device, where the first indication information is used to indicate type information of input data of the first model.
- the method further includes: receiving historical data information required for model reasoning from the first access network device; in the model reasoning, the historical data information is used to determine the first The input data for the model.
- the method further includes: sending second indication information to the first access network device, where the second indication information is used to indicate at least one of the following: handover of the terminal to the target cell, The first access network device deletes the first model, or the first access network device releases corresponding computing resources.
- a device is provided.
- the device may be an OAM, or a device configured in the OAM, or a device that can be used in conjunction with the OAM.
- the device includes a one-to-one unit for performing the method/operation/step/action described in the first aspect, and the unit may be a hardware circuit, or software, or a combination of hardware circuit and software.
- the device may include a processing unit and a communication unit, and the processing unit and the communication unit may perform corresponding functions in any design example of the first aspect above, specifically:
- a communication unit configured to receive a first model request message from a first access network device; a processing unit, configured to determine a first model according to the first model request message and a first mapping relationship, and the first mapping relationship Including the mapping relationship between models, model application scenarios, and model functions; or the first mapping relationship includes the mapping relationship between models, model performance levels, and model functions; the communication unit is also used to communicate to the first access network device Send information about the first model.
- a processing unit configured to determine a first model according to the first model request message and a first mapping relationship, and the first mapping relationship Including the mapping relationship between models, model application scenarios, and model functions; or the first mapping relationship includes the mapping relationship between models, model performance levels, and model functions; the communication unit is also used to communicate to the first access network device Send information about the first model.
- the device includes a memory, configured to implement the method described in the first aspect above.
- the apparatus may also include memory for storing instructions and/or data.
- the memory is coupled to the processor, and when the processor executes the program instructions stored in the memory, the method described in the first aspect above can be implemented.
- the device may also include a communication interface for the device to communicate with other devices.
- the communication interface may be a transceiver, a circuit, a bus, a module, a pin or other types of communication interfaces, and other devices may be network devices and the like.
- the device includes:
- a communication interface configured to receive a first model request message from the first access network device
- a processor configured to determine a first model according to the first model request message and a first mapping relationship, where the first mapping relationship includes a mapping relationship between a model, a model application scenario, and a model function; or the first mapping Relationships include mapping relationships between models, model performance levels, and model functions;
- the communication interface is further configured to send the information of the first model to the first access network device.
- the device can be the first access network device, or a device configured in the first access network device, or can be connected with the first access network device. network equipment to match the device used.
- the device includes a one-to-one unit for performing the methods/operations/steps/actions described in the second aspect.
- the unit may be a hardware circuit, or software, or a combination of hardware circuit and software.
- the device may include a processing unit and a communication unit, and the processing unit and the communication unit may perform corresponding functions in any design example of the second aspect above, specifically:
- a communication unit configured to send a first model request message to OAM, where the first model request message is used to indicate the model application scenario and model function of the first model; or, the first model request message uses for indicating model capabilities and model performance levels of the first model; and receiving information from the OAM for the first model.
- a processing unit configured to determine the first model according to the information of the first model.
- the device includes a memory, configured to implement the method described in the second aspect above.
- the apparatus may also include memory for storing instructions and/or data.
- the memory is coupled to the processor, and when the processor executes the program instructions stored in the memory, the method described in the second aspect above can be implemented.
- the device may also include a communication interface for the device to communicate with other devices.
- the communication interface may be a transceiver, circuit, bus, module, pin or other types of communication interface, and the other device may be a terminal or the like.
- the device includes:
- a communication interface configured to send a first model request message to OAM, where the first model request message is used to indicate the model application scenario and model function of the first model; or, the first model request message uses for indicating model capabilities and model performance levels of the first model; and receiving information from the OAM for the first model.
- a processor configured to determine the first model according to the information of the first model.
- the device may be the second access network device, or a device configured in the second access network device, or capable of communicating with the second access network device. network equipment to match the device used.
- the device includes a one-to-one unit for performing the methods/operations/steps/actions described in the third aspect.
- the unit may be a hardware circuit, or software, or a combination of hardware circuit and software.
- the device may include a processing unit and a communication unit, and the processing unit and the communication unit may perform corresponding functions in any design example of the third aspect above, specifically:
- a communication unit configured to receive the index of the first model from the first access network device; send a second model request message to OAM, where the second model request message is used to indicate the index of the first model; and receiving information from said first model of said OAM.
- a processing unit configured to determine the first model according to the information of the first model.
- the device includes a memory, configured to implement the method described in the third aspect above.
- the apparatus may also include memory for storing instructions and/or data.
- the memory is coupled to the processor, and when the processor executes the program instructions stored in the memory, the method described in the third aspect above can be implemented.
- the device may also include a communication interface for the device to communicate with other devices.
- the communication interface may be a transceiver, circuit, bus, module, pin or other types of communication interface, and the other device may be a terminal or the like.
- the device includes:
- a communication interface configured to receive the index of the first model from the first access network device; send a second model request message to OAM, where the second model request message is used to indicate the index of the first model; and receiving information from said first model of said OAM.
- a processor configured to determine the first model according to the information of the first model.
- the present application also provides a computer-readable storage medium, including instructions, which, when run on a computer, cause the computer to execute the method in any one of the first aspect, the second aspect, or the third aspect.
- the present application further provides a system-on-a-chip, which includes a processor and may further include a memory, for implementing the method in any one of the first aspect, the second aspect, or the third aspect.
- the system-on-a-chip can be constructed from a chip components, and can also include chips and other discrete components.
- the present application further provides a computer program product, including instructions, which, when run on a computer, cause the computer to execute the method of any one of the first aspect, the second aspect, or the third aspect.
- the present application further provides a system, which includes: the device of the fourth aspect and the device of the fifth aspect; or, the device of the fourth aspect, the device of the fifth aspect, and the device of the sixth aspect.
- FIG. 1 is a schematic diagram of a communication system provided by the present application.
- FIG. 1 Figure 2 and Figure 3 are schematic diagrams of model deployment provided by this application.
- FIG. 4a and Figure 4b are schematic diagrams of the architecture of the communication system provided by the present application.
- Figure 4c is a schematic diagram of the application of the AI model provided by this application.
- Fig. 5 is the schematic diagram of the neuron provided by the present application.
- Fig. 6 is the schematic diagram of the neural network provided by the present application.
- FIG. 7 and Figure 8 are schematic flow charts provided by the present application.
- 9 and 10 are schematic structural diagrams of the device provided by the present application.
- FIG. 1 is a schematic structural diagram of a communication system 1000 to which the present application can be applied.
- the communication system includes a radio access network 100 and a core network 200 , and optionally, the communication system 1000 may also include the Internet 300 .
- the radio access network 100 may include at least one access network device (such as 110a and 110b in FIG. 1 ), and may also include at least one terminal device (such as 120a-120j in FIG. 1 ).
- the terminal device is connected to the access network device in a wireless manner, and the access network device is connected to the core network in a wireless or wired manner.
- the core network device and the access network device can be independent and different physical devices, or the functions of the core network device and the logical functions of the access network device can be integrated on the same physical device, or they can be integrated on one physical device Part of the functions of the core network device and part of the functions of the access network device are specified.
- the terminal device and the terminal device and the access network device and the access network device may be connected to each other in a wired or wireless manner.
- FIG. 1 is only a schematic diagram.
- the communication system may also include other network devices, such as wireless relay devices and wireless backhaul devices, which are not shown in FIG. 1 .
- the access network equipment can be a base station (base station), an evolved base station (evolved NodeB, eNodeB), a transmission reception point (transmission reception point, TRP), and a next-generation base station in the fifth generation (5th generation, 5G) mobile communication system (next generation NodeB, gNB), access network equipment in the open radio access network (open radio access network, O-RAN), next-generation base stations in the sixth generation (6th generation, 6G) mobile communication system, future mobile
- DU distributed unit
- CU control plane centralized unit control plane
- CU-CP centralized unit control plane
- CU user plane centralized unit user plane
- the access network device may be a macro base station (such as 110a in Figure 1), a micro base station or an indoor station (such as 110b in Figure 1), or a relay node or a donor node.
- a macro base station such as 110a in Figure 1
- a micro base station such as 110b in Figure 1
- a relay node or a donor node.
- the device used to realize the function of the access network device may be the access network device; it may also be a device capable of supporting the access network device to realize the function, such as a chip system, a hardware circuit, a software module, or a hardware circuit plus soft software module, the device can be installed in the access network equipment or can be matched with the access network equipment.
- a system-on-a-chip may consist of a chip, or may include a chip and other discrete devices.
- the technical solution provided by the present application will be described below by taking the apparatus for realizing the functions of the access network equipment as the access network equipment and the access network equipment as the RAN node as an example.
- the protocol layer structure may include a control plane protocol layer structure and a user plane protocol layer structure.
- the control plane protocol layer structure may include a radio resource control (radio resource control, RRC) layer, a packet data convergence protocol (packet data convergence protocol, PDCP) layer, a radio link control (radio link control, RLC) layer, a media The access control (media access control, MAC) layer and the function of the protocol layer such as the physical layer.
- the user plane protocol layer structure may include the functions of the PDCP layer, the RLC layer, the MAC layer, and the physical layer.
- the PDCP layer may also include a service data adaptation protocol (service data adaptation protocol). protocol, SDAP) layer.
- the protocol layer structure between the access network device and the terminal device may further include an artificial intelligence (AI) layer, which is used to transmit data related to the AI function.
- AI artificial intelligence
- Access devices may include CUs and DUs. Multiple DUs can be centrally controlled by one CU.
- the interface between the CU and the DU may be referred to as an F1 interface.
- the control plane (control panel, CP) interface may be F1-C
- the user plane (user panel, UP) interface may be F1-U. This application does not limit the specific names of each interface.
- CU and DU can be divided according to the protocol layer of the wireless network: for example, the functions of the PDCP layer and above protocol layers are set in the CU, and the functions of the protocol layers below the PDCP layer (such as RLC layer and MAC layer, etc.) are set in the DU; another example, PDCP The functions of the protocol layer above the layer are set in the CU, and the functions of the PDCP layer and the protocol layer below are set in the DU, without restriction.
- the CU or DU may be divided into functions having more protocol layers, and for example, the CU or DU may also be divided into part processing functions having protocol layers.
- part of the functions of the RLC layer and the functions of the protocol layers above the RLC layer are set in the CU, and the rest of the functions of the RLC layer and the functions of the protocol layers below the RLC layer are set in the DU.
- the functions of the CU or DU can also be divided according to the business type or other system requirements, for example, according to the delay, and the functions whose processing time needs to meet the delay requirement are set in the DU, which does not need to meet the delay
- the required feature set is in the CU.
- the CU may also have one or more functions of the core network.
- the CU can be set on the network side to facilitate centralized management.
- the wireless unit (radio unit, RU) of the DU is set remotely.
- the RU may have a radio frequency function.
- DUs and RUs may be divided in a physical layer (physical layer, PHY).
- the DU can implement high-level functions in the PHY layer
- the RU can implement low-level functions in the PHY layer.
- the functions of the PHY layer may include at least one of the following: adding a cyclic redundancy check (cyclic redundancy check, CRC) code, channel coding, rate matching, scrambling, modulation, layer mapping, precoding, Resource mapping, physical antenna mapping, or radio frequency transmission functions.
- CRC cyclic redundancy check
- the functions of the PHY layer may include at least one of the following: CRC check, channel decoding, de-rate matching, descrambling, demodulation, de-layer mapping, channel detection, resource de-mapping, physical antenna de-mapping, or RF receiving function.
- the high-level functions in the PHY layer may include a part of the functions of the PHY layer, for example, this part of the functions is closer to the MAC layer, and the lower-level functions in the PHY layer may include another part of the functions of the PHY layer, for example, this part of the functions is closer to the radio frequency function.
- high-level functions in the PHY layer may include adding CRC codes, channel coding, rate matching, scrambling, modulation, and layer mapping
- low-level functions in the PHY layer may include precoding, resource mapping, material Antenna mapping, and RF transmission functions;
- high-level functions in the PHY layer may include adding CRC codes, channel coding, rate matching, scrambling, modulation, layer mapping, and precoding
- low-level functions in the PHY layer may include resource mapping , physical antenna mapping, and RF transmit functions.
- the high-level functions in the PHY layer may include CRC check, channel decoding, de-rate matching, decoding, demodulation, and de-layer mapping
- the low-level functions in the PHY layer may include channel detection, resource de-mapping, physical antenna de-mapping, and RF receiving functions
- the high-level functions in the PHY layer may include CRC check, channel decoding, de-rate matching, decoding, demodulation, de-layer mapping, and channel detection
- the low-level functions in the PHY layer may include resource de-mapping , physical antenna demapping, and RF receiving functions.
- the function of the CU may be implemented by one entity, or may also be implemented by different entities.
- the functions of the CU can be further divided, that is, the control plane and the user plane are separated and realized by different entities, namely, the control plane CU entity (ie, the CU-CP entity) and the user plane CU entity (ie, the CU-UP entity) .
- the CU-CP entity and CU-UP entity can be coupled with the DU to jointly complete the functions of the access network equipment.
- any one of the foregoing DU, CU, CU-CP, CU-UP, and RU may be a software module, a hardware structure, or a software module+hardware structure, without limitation.
- the existence forms of different entities may be different, which is not limited.
- DU, CU, CU-CP, and CU-UP are software modules
- RU is a hardware structure.
- the access network device includes CU-CP, CU-UP, DU and RU.
- the execution subject of this application includes DU, or includes DU and RU, or includes CU-CP, DU and RU, or includes CU-UP, DU and RU, without limitation.
- the methods executed by each module are also within the protection scope of the present application.
- the terminal equipment may also be called a terminal, user equipment (user equipment, UE), mobile station, mobile terminal equipment, and the like.
- Terminal devices can be widely used in communication in various scenarios, including but not limited to at least one of the following scenarios: device-to-device (device-to-device, D2D), vehicle-to-everything (V2X), machine-type communication ( machine-type communication (MTC), Internet of Things (IOT), virtual reality, augmented reality, industrial control, automatic driving, telemedicine, smart grid, smart furniture, smart office, smart wear, smart transportation, or intelligence city etc.
- the terminal device can be a mobile phone, a tablet computer, a computer with wireless transceiver function, a wearable device, a vehicle, a drone, a helicopter, an airplane, a ship, a robot, a mechanical arm, or a smart home device, etc.
- This application does not limit the specific technology and specific equipment form adopted by the terminal equipment.
- the device used to realize the function of the terminal device may be a terminal device; it may also be a device capable of supporting the terminal device to realize the function, such as a chip system, a hardware circuit, a software module, or a hardware circuit plus a software module.
- the device can be installed in the terminal equipment or can be matched with the terminal equipment for use.
- the technical solution provided by the present application will be described below by taking the apparatus for realizing the functions of the terminal equipment as the terminal equipment and the terminal equipment as the UE as an example.
- Base stations and terminal equipment can be fixed or mobile.
- Base stations and/or terminal equipment can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; they can also be deployed on water; they can also be deployed on aircraft, balloons and artificial satellites in the air.
- This application does not limit the application scenarios of the base station and terminal equipment.
- the base station and the terminal device can be deployed in the same scene or in different scenarios. For example, the base station and the terminal device are deployed on land at the same time; or, the base station is deployed on land and the terminal device is deployed on water.
- the helicopter or drone 120i in FIG. Device 120i is a base station; however, for base station 110a, 120i is a terminal device, that is, communication between 110a and 120i is performed through a wireless air interface protocol. Communication between 110a and 120i may also be performed through an interface protocol between base stations Yes, at this time, relative to 110a, 120i is also a base station. Therefore, both the base station and the terminal equipment can be collectively referred to as a communication device, 110a and 110b in FIG. 1 can be referred to as a communication device with a base station function, and 120a-120j in FIG. 1 can be referred to as a communication device with a terminal device function.
- an independent network element such as AI network element, or AI node, etc.
- AI network element can be introduced into the communication system shown in Figure 1 to realize AI-related operations, and the AI network element can communicate with the interface in the communication system
- Network access devices are directly connected, or can be indirectly connected through third-party network elements and access network devices.
- the third-party network element can be a core network element such as authentication management function (authentication management function, AMF) or user plane function (user plane function, UPF); or, AI can be configured in other network elements in the communication system Functions, AI modules or AI entities to implement AI-related operations, for example, the other network elements can be access network equipment (such as gNB), core network equipment, or network management (operation, administration and maintenance, OAM), etc., in this
- the network element performing AI-related operations is a network element with a built-in AI function.
- the above-mentioned OAM is used to operate, manage and maintain the access network equipment and/or the core network equipment and the like.
- At least one of the core network equipment, access network equipment, terminal equipment, or OAM can be deployed with an AI model, and use the AI model to implement corresponding functions.
- the AI models deployed in different nodes may be the same or different.
- Different models include at least one of the following differences: different structural parameters of the models, for example, different layers and/or weights of the models; different input parameters of the models; or different output parameters of the models. Wherein, different input parameters of the model and/or different output parameters of the model can be described as different functions of the model.
- the function of the access network device is divided into CU and DU.
- the CU and DU may be CU and DU under the O-RAN architecture.
- One or more AI models can be deployed in the CU. And/or, one or more AI models may be deployed in the DU.
- the CU in FIG. 3 may be further split into CU-CP and CU-UP.
- one or more AI models may be deployed in the CU-CP.
- one or more AI models may be deployed in the CU-UP.
- the OAM of the access network device and the OAM of the core network device may be deployed independently.
- FIG. 4a is a communication system architecture of the present application.
- the access network equipment includes a near real-time access network intelligent controller (RAN intelligent controller, RIC) module for model training and inference.
- RAN intelligent controller RIC
- near real-time RIC can be used to train an AI model and use that AI model for inference.
- the near real-time RIC can obtain network-side and/or terminal-side information from at least one of CU, DU, or RU, and the information can be used as training data or inference data.
- the near real-time RIC may submit the reasoning result to at least one of CU, DU, RU or terminal device.
- the inference results can be exchanged between the CU and the DU.
- the reasoning results can be exchanged between the DU and the RU, for example, the near real-time RIC submits the reasoning result to the DU, and the DU forwards it to the RU.
- the access network device includes a non-real-time RIC (optionally, the non-real-time RIC can be located in the OAM or in the core network device) for model training and reasoning.
- non-real-time RIC is used to train an AI model and use that model for inference.
- the non-real-time RIC can obtain network-side and/or terminal-side information from at least one of CU, DU, or RU, which can be used as training data or inference data, and the inference results can be submitted to CU, DU, or RU or at least one of the terminal devices.
- the inference results can be exchanged between the CU and the DU.
- the reasoning results can be exchanged between the DU and the RU, for example, the non-real-time RIC submits the reasoning result to the DU, and the DU forwards it to the RU.
- the access network equipment includes a near-real-time RIC, and the access network equipment includes a non-real-time RIC (optionally, the non-real-time RIC can be located in the OAM or the core network device).
- non-real-time RIC can be used for model training and inference.
- the near real-time RIC can be used for model training and reasoning.
- the non-real-time RIC performs model training, and the near-real-time RIC can obtain AI model information from the non-real-time RIC, and obtain network-side and/or terminal-side information from at least one of CU, DU or RU, and use the information And the AI model information to get the reasoning result.
- the near real-time RIC may submit the reasoning result to at least one of CU, DU, RU or terminal device.
- the inference results can be exchanged between the CU and the DU.
- the reasoning results can be exchanged between the DU and the RU, for example, the near real-time RIC submits the reasoning result to the DU, and the DU forwards it to the RU.
- near real-time RIC is used to train model A and use model A for inference.
- non-real-time RIC is used to train Model B and utilize Model B for inference.
- the non-real-time RIC is used to train the model C, and the information of the model C is sent to the near-real-time RIC, and the near-real-time RIC uses the model C for inference.
- Fig. 4b is an architecture of another communication system of the present application. Compared with Figure 4a, CU is separated into CU-CP and CU-UP in Figure 4b.
- the AI model is the specific realization of the AI function, and the AI model represents the mapping relationship between the input and output of the model.
- the AI model can be a neural network, a linear regression model, a decision tree model, a support vector machine (SVM), a Bayesian network, a Q-learning model, or other machine learning models.
- the AI function may include at least one of the following: data collection (collection of training data and/or reasoning data), data preprocessing, model training (or model learning), model information publishing (configuration model information), Model verification, model inference, or release of inference results.
- reasoning can also be called prediction.
- the AI model may be referred to as a model for short.
- Figure 4c is a schematic diagram of an application architecture of the AI model.
- Data source data source
- the model training host model training host
- the model training host obtains the AI model by analyzing or training the training data provided by the data source, and deploys the AI model in the model inference host (model inference host).
- the model training node can also update the AI model deployed on the model inference node.
- the model inference node can also feed back information about the deployed model to the model training node, so that the model training node can optimize or update the deployed AI model.
- the model inference node uses the AI model to perform inference based on the inference data provided by the data source, and obtains the inference result.
- This method can also be described as: the model inference node inputs the inference data into the AI model, and obtains an output through the AI model, and the output is the inference result.
- the inference result may indicate: configuration parameters used (executed) by the execution object, and/or operations performed by the execution object.
- the reasoning result can be uniformly planned by the execution (actor) entity, and sent to one or more execution objects (for example, network entities) for execution.
- the execution entity or execution object may feed back the collected parameters or measurements to the data source.
- This process may be called performance feedback, and the fed-back parameters may be used as training data or inference data.
- the execution entity or execution object can also determine feedback information related to model performance according to the inference results output by the model inference node, and feed the feedback information to the model inference node.
- the model inference node can, based on the feedback information, Feedback the performance information of the model to the model training node, so that the model training node optimizes or updates the deployed AI model, etc. This process can be called model feedback.
- AI models can be neural networks or other machine learning models.
- neural network is a specific implementation form of machine learning technology. According to the general approximation theorem, the neural network can theoretically approximate any continuous function, so that the neural network has the ability to learn any mapping. Therefore, neural networks can perform accurate abstract modeling of complex high-dimensional problems.
- Each neuron performs a weighted sum operation on its input values, and passes the weighted sum result through an activation function to generate an output.
- FIG. 5 it is a schematic diagram of a neuron structure.
- the bias of the weighted sum for b the form of the activation function can be diversified.
- the output of the neuron is: x i , w i , and b can be various possible values such as decimals, integers (including 0, positive integers or negative integers, etc.), or complex numbers.
- the activation functions of different neurons in a neural network can be the same or different.
- a neural network generally includes a multi-layer structure, and each layer may include one or more neurons. Increasing the depth and/or width of a neural network can improve the expressive power of the neural network, providing more powerful information extraction and abstract modeling capabilities for complex systems.
- the depth of the neural network may refer to the number of layers included in the neural network, and the number of neurons included in each layer may be referred to as the width of the layer.
- Figure 6 it is a schematic diagram of the layer relationship of the neural network.
- a neural network includes an input layer and an output layer. The input layer of the neural network processes the input received by neurons, and then passes the result to the output layer, and the output layer obtains the output result of the neural network.
- a neural network in another implementation, includes an input layer, a hidden layer, and an output layer.
- the input layer of the neural network processes the input received by neurons, and then passes the result to the hidden layer in the middle, and the hidden layer then passes the calculation result to the output layer or the adjacent hidden layer, and finally the output layer obtains the result of the neural network. Output the result.
- a neural network may include one or more sequentially connected hidden layers without limitation.
- a loss function can be defined. The loss function describes the gap or difference between the output value of the neural network and the ideal target value, and the application does not limit the specific form of the loss function.
- the training process of the neural network is to adjust the parameters of the neural network, such as the number of layers of the neural network, the width, the weight of the neuron, and/or the parameters in the activation function of the neuron, etc., so that the value of the loss function is less than the threshold threshold value Or the process of meeting the target requirements.
- RAN nodes support multiple AI-assisted cases. For example, mobility enhancement, load balancing and network energy saving etc.
- the OAM is used to train or store the AI model, and the RAN node requests the AI model from the OAM, and uses the AI model to perform reasoning. How the OAM configures the corresponding model to the RAN node is a problem to be solved in this application.
- OAM can be replaced by other nodes capable of training or storing AI models, such as cloud servers, core network devices, or other possible devices.
- the process of providing a model configuration method includes at least:
- Step 701 the first RAN node sends a first model request message to the OAM; correspondingly, the OAM receives the first model request message from the first RAN node.
- the model request message sent by the first RAN node to the OAM is referred to as the first model request message.
- the model request message sent by the second RAN node to the OAM is called a second model request message.
- the model request message is used for the RAN node to request a model from the OAM, and the name of the model request message is not limited.
- the first model request message may also be called a first message
- the second model request message may also be called a second message, and so on.
- they are referred to as the first RAN node and the second RAN node.
- the first RAN node and the second RAN node are different RAN nodes, and the relationship between them is not limited.
- the first RAN node may determine service scenarios and functional requirements, and determine the first model request message according to the service scenarios and functional requirements.
- the service scenarios and functional requirements are the service scenarios and functional requirements of the RAN node.
- the first model requested by the first model request message is used for model reasoning related to the RAN node, for example, for cell energy saving and the like.
- the first RAN node may determine the first model request message and the like according to the service scenario and functional requirements of the RAN node.
- the business scenarios and functional requirements are the UE's business scenarios and functional requirements beg.
- the first model requested by the first model request message is used for UE-related model reasoning, for example, for predicting a motion track of the UE.
- the first RAN node may determine the first model request message and the like according to the service scenario and functional requirements of the UE.
- the first RAN node may determine the first model request message and the like according to service scenarios and functional requirements of nodes other than the RAN node and the UE, without limitation.
- the first RAN node may obtain the service scenario and functional requirements of the UE according to the information reported by the UE.
- the UE is in an automatic driving scenario and sends a quality of service (QoS) prediction service request to the RAN node, and the RAN node can obtain the service scenario and functional requirements of the UE as follows: automatic driving + QoS prediction.
- the first RAN node may obtain the service scenario and functional requirements of the UE through analysis.
- the RAN node finds that the UE moves near the edge of the cell, mobility enhancement service is required, and the mobility enhancement service needs to predict the trajectory information of the UE.
- the service scenario and functional requirements of the UE are: mobility enhancement + prediction UE trajectory information.
- the first RAN node may acquire the service scenario and functional requirements of the UE from other devices, the other devices include but not limited to: minimization drive test (MDT) and the like.
- the first RAN node stores the UE's business scenarios and functional requirements, for example, when the UE accesses the first RAN node, it actively reports its own business scenarios and functional requirements to the first RAN node.
- the business scenarios include but are not limited to at least one of the following: mobility enhancement, load balancing, virtual reality (virtual reality, VR), augmented reality (augmented reality, AR), car networking (automatic driving), Wireless robot cloud control, remote diagnosis with force feedback, or high-definition video (such as 8K video), etc.
- the mobility enhancement refers to providing an enhanced service for switching of a serving cell of a mobile UE.
- the RAN node predicts the UE's target cell, handover time, etc. according to at least one of the UE's movement track, measurement report, and cell load conditions.
- the target cell of the UE may refer to a serving cell to which the UE is to be handed over.
- the mobility mode of the UE includes fixed-line mobility and non-fixed-line mobility.
- the mobility enhancement may include the mobility enhancement of vehicle movement on fixed routes, the mobility enhancement of vehicle movement on non-fixed routes, or the mobility enhancement of pedestrian scenes, etc.
- doctors conduct diagnosis or surgery on patients in different places through the network.
- the above-mentioned feedback force refers to force tactile feedback, which is a service with particularly high requirements on delay, and needs to provide delay prediction services. For example, when the predicted delay exceeds a certain threshold, it is necessary to adjust the information transmission scheme or give an alarm to deal with emergencies in advance.
- the functional requirements include but are not limited to at least one of the following: UE trajectory prediction, load prediction, signal to interference plus noise ratio (SINR) prediction, reference signal receiving power , RSRP) prediction, reference signal receiving quality (reference signal receiving quality, RSRQ) prediction, delay prediction, QoS prediction, throughput prediction or cell energy saving, etc.
- the first model request message is used to indicate the service scenario and functional requirements, or described as,
- the first model request message is used to indicate the model application scenario and model function of the requested first model.
- the application scenario of the model corresponds to the above-mentioned business scenario
- the function of the model corresponds to the above-mentioned functional requirements.
- the indications in this application may include implicit indications or display indications, etc., without limitation.
- the first model request message may include the business scenario, functional requirements, and the like.
- the first RAN node may directly send the service scenario and functional requirements to the OAM and the like.
- the combination of business scenarios and functional requirements sent by the first RAN node to OAM may include: mobility enhancement scenario + trajectory prediction, load balancing + trajectory prediction, automatic driving + QoS prediction, VR/AR + QoS prediction, wireless robot cloud Control + delay prediction, remote diagnosis with force feedback + delay prediction, etc.
- the process of the first RAN node determining the first model request message according to the business scenario and functional requirements includes: the first RAN node determining the requested first model request message according to the business scenario and functional requirements.
- Model features and model performance metrics for the model.
- the model functions correspond to the above functional requirements.
- the model The performance index includes at least one of the following: accuracy rate, calculation amount, memory usage, or inference time.
- the accuracy rate refers to the percentage of the number of correctly predicted samples to the total number of samples.
- the calculation amount refers to the number of calculations required by the model.
- the overall calculation amount of the model is equal to the sum of the calculation amount of each operator in the model, which feeds back the model's demand for hardware computing units; or the calculation amount can refer to the running time of the model.
- the memory occupation refers to the memory and video memory size occupied by the model when it is running; or the memory occupation can be referred to as the size of the model, etc.
- the inference time refers to the time required for the model to infer the predicted results.
- the first model request message sent by the first RAN node to the OAM is used to indicate the model function and model performance index of the first model, for example, the first model request message includes the model function and model performance index. Alternatively, the first RAN node may directly send the model function and model performance index to the OAM.
- the model functions may also be referred to as model classes.
- the first RAN node learns that it needs to provide UE with QoS prediction service in the autonomous driving scenario, and the first RAN node can continue to analyze the performance indicators required for the QoS prediction service in the autonomous driving scenario, and the first RAN node will
- the QoS prediction service and corresponding performance indicators are sent to OAM.
- the first RAN node determines the model performance level according to the model performance index, and there is a corresponding relationship between the model performance level and the model performance index, which can be referred to the description below.
- the first model request message sent by the first RAN node to the OAM is used to indicate the requested model function and model performance level of the first model.
- the method may further include: the first RAN node determines whether to provide an AI-based enhanced service. If AI-based enhanced service needs to be provided, perform step 701 to send a first model request message to the OAM; otherwise, the above-mentioned process shown in FIG. 7 is not performed again.
- the source RAN node can use AI to predict the target cell of the UE. But when the UE is not at the edge of the cell, the source RAN node does not need to predict the target cell of the UE. That is, only when the UE is at the edge of the cell, the source RAN node needs to send the first model request message to the OAM for requesting a model for predicting the target cell.
- the source RAN node predicts the target cell of the UE in an AI manner
- the input of the model includes the trajectory information of the UE.
- the trajectory of the UE is very random, the accuracy of the target cell predicted by using the AI method may be greatly reduced.
- the source RAN node may consider no longer using the AI method to predict the target cell of the UE, and no longer send the first model request message to the OAM at this time, but use the traditional method to predict the target cell of the UE.
- Step 702 The OAM determines the first model according to the first model request message and the first mapping relationship.
- the model represents the mapping relationship between the input and output of the model, and is used to realize specific functions, and the name of the model is not limited.
- the model may also be called an AI model, and the AI model may be a neural network, a linear regression model, a decision tree model, an SVM, a Bayesian network, a Q-learning model, or other machine learning models.
- the above-mentioned first mapping relationship represents a mapping or corresponding relationship.
- the first mapping relationship may also be referred to as a corresponding relationship or the like. Subsequent descriptions of similar situations will not be given one by one.
- the OAM may obtain the first mapping relationship, and the first mapping relationship may be established by the OAM or by other nodes, and the OAM obtains the first mapping relationship through other nodes.
- the first mapping relationship includes a mapping relationship between a model, a model performance level, and a model function
- the model performance level may be simply referred to as a performance level.
- model performance index used for classifying models includes accuracy rate, which can be classified according to the accuracy rate. The higher the accuracy rate, the higher the model performance level.
- model performance indicators used for classifying models include inference duration, which can be graded according to the inference duration. The shorter the inference duration, the higher the model performance level.
- the model performance indicators used for classifying models include at least two indicators among accuracy rate, calculation amount, inference duration and memory usage. For example, according to a certain weighting method, for multiple indicators Carry out weighting, determine the weighted value, and classify according to the weighted value.
- Models with different functions have the same or different ways of classifying model performance levels. For example, take the model used for trajectory prediction and the model used for SINR prediction, and classify the model performance according to the accuracy of the model as an example. It is assumed that both the model for trajectory prediction and the model for SINR prediction include 3 model performance levels. As shown in Table 1, for the model used for trajectory prediction, the accuracy rate corresponding to model performance level 1 is [n1, n2], the accuracy rate corresponding to model performance level 2 is (n2, n3], and the accuracy rate corresponding to model performance level 3 is The accuracy rate is (n3, n4], n4 is greater than or equal to n3, n3 is greater than or equal to n2, and n2 is greater than or equal to n1.
- the accuracy rate corresponding to model performance level 1 is [m1 , m2], the accuracy rate corresponding to model performance level 2 is (m2, m3], m3 is greater than or equal to m2, and m2 is greater than or equal to m1.
- the model performance level 1 The corresponding inference duration is [t1, t2], the inference duration corresponding to model performance level 2 is (t2, t3], the inference duration corresponding to model performance level 3 is (t3, t4], t4 is greater than or equal to t3, and t3 is greater than or equal to t2, and t2 is greater than or equal to t1.
- Table 1 Correspondence between model performance grades and model performance indicators
- Table 2 Correspondence between model performance grades and model performance indicators
- Table 3 Correspondence between model performance grades and model performance indicators
- the first RAN node can obtain the application scenario of the model, determine the corresponding model performance index according to the application scenario of the model, determine the model performance level according to the model performance index, and send the first model request message to OAM for Indicates the model performance class and model capabilities.
- the OAM determines a model in the first mapping relationship, which is called the first model.
- the first mapping relationship includes the mapping relationship between the model, the model performance level, and the model function, and reference may be made to the description in Table 6 below.
- the first mapping relationship includes a mapping relationship of a model, a model application scenario, and a model function.
- the first mapping relationship includes: AI model+automatic driving scene+QoS prediction and so on.
- the first RAN node may obtain the model application scenario and model function of the first model, and reference may be made to the description in step 701 .
- the first model request message sent by the first RAN node to the OAM is used to indicate the model application scenario and model function of the first model.
- the OAM may determine an AI model in the first mapping relationship according to the application scenario and model function indicated by the first model request message, and the AI model is called the first model.
- the first mapping relationship includes 4 application scenarios (indexes are 1 to 4 respectively), and 3 model functions (indexes are 1 to 3 respectively).
- the four application scenarios and the three model functions can form 12 mapping relationships, and the indexes of the 12 mapping relationships are 1 to 12 respectively.
- mapping relationship 1 includes application scenario 1 + model function 1 + model 1
- mapping relationship 2 includes application scenario 1 + model function 2 + model 2, etc.
- the corresponding models are different in different scenarios.
- model function 1 in application scenarios 1 to 4, it corresponds to model 1, model 4, model 7 and model 10 respectively.
- OAM can collect corresponding training data and train corresponding AI models in different scenarios.
- OAM can collect training data corresponding to model function 1 in application scenario 1, and use the training data to train model 1, so that model 1 can meet the performance index of application scenario 1.
- machine learning, non-machine learning, or reinforcement learning can be used, without limitation.
- model performance level 1 corresponds to model performance indicator 1
- the corresponding accuracy rate is above 99%
- the calculation amount is below 3%
- the inference time is in Within 1s
- the memory usage is within 500KB, etc.
- Table 5 Correspondence between model performance grades and model performance indicators
- the first RAN node may obtain the model performance index of the requested model, and the first model request message sent to the OAM is used to indicate the model function and model performance index of the requested model.
- the OAM may determine the model performance level according to the model performance index indicated by the first model request message.
- the first RAN node may determine the model performance level according to the model performance index, and the first model request message sent to the OAM is used to indicate the model function and the model performance level.
- the OAM determines a model in the first mapping relationship according to the model function and the model performance level, and the model is called the first model.
- the first mapping relationship includes 12 mapping relationships, and the indexes are 1 to 12 respectively.
- mapping relationship 1 includes: model 1 + model performance level 1 + model function 1
- mapping relationship 2 includes: model 2 + model performance level 1 + model function 2.
- the model performance level in the first mapping relationship may be replaced by the model performance index.
- the first mapping relationship may be established according to the model performance index, for example, the first mapping relationship includes the mapping relationship of the model, the model performance index, and the model function.
- the first model request message is used to indicate the model function and model performance index of the first model.
- the OAM determines a model in the first mapping relationship according to the model function and model performance index indicated by the first model request message, and the model may be called the first model.
- the first mapping relationship includes 12 mapping relationships, and the indexes are 1 to 12 respectively.
- mapping relationship 1 includes: model 1 + model performance indicator 1 + model function 1
- mapping relationship 2 includes: model 2 + model performance indicator 1 + model function 2.
- the corresponding models are different under different model performance levels or model performance indicators.
- the corresponding models are: model 1, model 4, model 7 and model 10.
- OAM can respectively collect corresponding training data and train corresponding models under different model performance levels or model performance indicators.
- the OAM can collect training data corresponding to model function 1 under model performance level 1 or model performance index 1, and train model 1, etc.
- the model performance index includes at least one of the following: accuracy rate, computation amount, memory usage, or inference duration.
- the same model function may have different requirements for model performance indicators.
- the calculation amount in the model performance index is also referred to as computing resource occupation, and the memory occupation is also referred to as model size.
- One possible situation, as shown in Table 8, is that the performance index requirements of the QoS prediction model are different in the VR/AR scenario and the V2X scenario.
- Table 9 in scenarios such as fixed-route vehicles, non-fixed-route vehicles, and pedestrian scenarios, the requirements for the performance indicators of the trajectory prediction models are different.
- Table 8 QoS prediction functions in different scenarios
- the performance index requirements for models with the same function may be different. If the same model is used for reasoning for a certain function no matter what the scenario is, it will be difficult to meet the performance index requirements of the corresponding scenario.
- Table 8 for AR/VR and V2X scenarios, if the same QoS prediction model is used for model inference, it is difficult to meet the requirements for performance indicators in each scenario.
- different models may be configured in different scenarios for the same model function. For example, in a VR/AR scenario, configure a model for QoS prediction; in a V2X scenario, configure another model for QoS prediction, etc.
- mapping relationship shown in Table 6 refers to the mapping relationship shown in Table 6 above, and configure a model corresponding to the performance level according to the performance index requirements of different scenarios, or refer to the mapping relationship shown in Table 7 above, and configure it for different performance indicators Different models, so that each model meets the performance index requirements of the corresponding application scenario.
- Step 703 the OAM sends the information of the first model to the first RAN node.
- the information of the first model is used to indicate at least one of the following of the first model: model index (index), model structure information, model parameters, model input format, model output format, input Data processing methods, output data processing methods, model performance indicators, model application scenarios, model functions, or training parameters, etc.
- model index is used to distinguish different models.
- Model structure information, model parameters, model input format, model output format, etc. are used by the first RAN node to determine the first model. Or described as: the first RAN node determines the first model according to the structural information of the model, model parameters, model input format, model output format, and the like.
- the structural information of the model includes at least one of the following: the model includes the number of layers (also referred to as the depth of the model), the network type of each layer (for example, a fully connected layer network, a convolutional layer network, a long short-term memory (( long short-term memory, LSTM) layer etc.), the number of neurons included in each layer (also known as the width of the model), or the connection relationship between layers, etc.
- the model parameters may include at least one of the following: a weight value of a neuron, an activation function of a neuron, or a bias in an activation function, and the like.
- the input format and output format of the above model may refer to the format that the input data and output data should satisfy when performing model training or inference on the model.
- the above input data processing methods include preprocessing the collected raw data before inputting it into the model, for example, separating the imaginary part and the real part of the raw data, normalizing processing or phase and amplitude separation, etc.
- the above-mentioned output data processing method includes the processing method of the output data when the model outputs the data. For example, intercepting output data and so on.
- model performance indicators, model application scenarios, and model functions please refer to the above description.
- the first RAN node determines the first model based on information such as the model structure, model parameters, model input format, and model output format, it can also continue to train the first model based on the above training parameters.
- the training parameters may include indication information of a loss function and the like.
- the first RAN node may obtain training data, which may come from the first RAN node itself, or from the UE, or from both the first RAN node and the UE, or from other nodes, etc., the first RAN node according to the training data , to train the first model.
- training data is input into the first model.
- the size of the loss function value is determined. If the value of the loss function is smaller than the threshold or meets the target requirement, the training of the first model is ended; otherwise, the parameters of the first model are updated and the training of the first model is continued.
- the updating of the parameters of the first model may include updating at least one of the following information of the first model: the number of layers of the neural network, the width of the neural network, the connection relationship between layers, the weight of neurons, the activation function of neurons, Or bias in activation functions etc.
- the training process may use label learning, non-label learning or reinforcement learning, etc., without limitation.
- the initial model of the first model may be indicated by the information of the above-mentioned first model, or may be randomly generated, or agreed upon in an agreement, without limitation.
- the information about the first model is used to indicate at least one of the following first model: model, model usage instructions, or model index.
- the model includes network architecture, parameter weights, training hyperparameters, optimizer, batch size, learning rate, momentum, etc.
- the first RAN node may perform online training according to the training hyperparameters.
- the model usage instructions include at least one of the model input and output format, model performance indicators, model application scenarios, and model functions; the index of the model refers to the number of the model, which is used to distinguish different models.
- the requirements for performance indicators of models that implement the same function are different.
- the performance index requirements of the model are different in different scenarios.
- the OAM establishes a first mapping relationship, and the first mapping relationship is a mapping relationship of model+model application scenario+model function.
- the model corresponding to the scenario can be configured, so that the model can meet the requirements for performance indicators in the scenario corresponding to the model.
- the same model is used for model inference.
- OAM establishes a first mapping relationship, the first mapping relationship is a mapping relationship of model + model performance level + model function, or the first mapping relationship is a mapping relationship of model + model performance index + model function. Since different application scenarios have different requirements for model performance indicators, and there is a corresponding relationship between model performance indicators and model performance levels, OAM can configure different models for different model performance indicator requirements in different application scenarios according to the first mapping relationship. In contrast, regardless of the model performance index of any scenario, using the same model for model reasoning can improve prediction performance and prediction speed.
- Step 704 The first RAN node performs model reasoning according to the first model to obtain prediction information.
- the first RAN node can collect input data, allocate computing resources and memory resources, etc., for model pushing Reasonably, the input data may come from the first RAN node and/or UE, or from other nodes, etc., which is not limited.
- the first RAN node may collect input data according to the model usage instructions.
- the input data is input to the first model, and the output is predictive information.
- the input data is processed, and the processed input data is input to the first model, and/or the output data of the first model is processed, and the processed output data is prediction information and the like. It can be understood that the input data and output data meet the requirements of the input format and output format of the first model.
- the prediction information includes prediction results.
- the model training node such as OAM, divides the collected data set into training set, verification set and test set when training the model.
- the model training node uses the training set to perform model training to obtain a model, and this process can be called model training; the verification set is used to verify the model, and this process can be called model verification.
- Model validation which usually happens during model training. For example, every time the model is trained for one or more iterations (epoch), the current model can be verified with the verification set to monitor the status of the model training, such as verifying whether it is underfitting, overfitting or has converged, etc., to determine whether End training.
- the hyperparameters of the model can also be adjusted, and the hyperparameters can refer to at least one of the following parameters of the model: the number of layers of the neural network, the number of neurons, the activation function, or the loss function wait.
- this process can be called model testing.
- a test set can be used to test the trained model. For example, evaluate the generalization ability of the model, judge whether the model meets the requirements, determine whether the model is available, or obtain the prediction accuracy of the model, etc.
- the information of the first model sent by the OAM to the first RAN node may also include prediction accuracy and the like.
- the first RAN node may use locally collected test data to test the first model to obtain prediction accuracy and the like.
- the prediction information sent by the first RAN node to the UE may also optionally include a prediction accuracy rate.
- the prediction information sent by the first RAN node to the UE includes: the predicted trajectory information of the UE + the accuracy rate of the predicted trajectory information (such as 98%).
- the first RAN node determines the first feedback information according to the prediction information. For example, the first RAN performs corresponding actions according to the prediction information.
- the prediction information is an energy-saving policy, such as carrier off, symbol off, cell off, or deep sleep, etc.
- the first RAN node deduces the energy-saving policy by using the first model, it can Enforce the energy saving strategy.
- the first RAN node determines the first feedback information according to performance changes before and after the execution of the prediction information (for example, cell throughput, or QoS changes of the UE, etc.).
- the prediction information is UE-related prediction information, and the first RAN node needs to send the prediction information to the UE.
- the UE determines second feedback information according to the prediction information.
- the UE may determine the second feedback information according to the performance change after executing the prediction information.
- the prediction information is a target cell for UE handover, handover time, and the like.
- the UE may perform a corresponding action according to the prediction information, and determine second feedback information according to a performance change after executing the prediction information.
- the UE compares the prediction information with the actual information, and determines the second feedback information and the like.
- the UE may compare the predicted running trajectory of the UE with the actual running trajectory of the UE to determine second feedback information and the like.
- the UE sends second feedback information to the first RAN node.
- the first RAN node sends the first feedback information to the OAM according to the received second feedback information, and the OAM updates the first model according to the first feedback information.
- the first method in the following steps 705 to 706. Description of the design.
- the first RAN node when the first RAN node determines that the application scenario changes and the performance level (or performance index) of the required model changes, it may also send the first feedback information for indicating the scene change or the performance level change to the OAM, and the OAM may according to the
- the scene change instruction or the performance level change instruction is to reconfigure the model for the first RAN node, and the model may be called the second model or the like.
- the first RAN node uses the trajectory prediction model requested by the walking scene to predict the motion trajectory in the fixed-route vehicle driving scene, which may lead to poor prediction accuracy. reduce.
- the first RAN node finds that the scene of the UE changes
- the scene change may be indicated to OAM, etc.
- OAM may configure a model matching the current scene for the first RAN node.
- the first model is a model related to the first RAN node, and the first RAN may determine a change of an application scenario or a change of a performance level by analyzing itself.
- Step 705 the first RAN node sends the prediction information to the UE; correspondingly, the UE receives the prediction information from the first RAN node.
- Step 706 the UE sends second feedback information to the first RAN node.
- the first RAN node receives the second feedback information from the UE.
- the UE when it receives the prediction information, it can compare the prediction information sent by the first RAN node with the actual information to determine the accuracy rate of the prediction information, which is called the actual accuracy rate of the prediction information .
- the prediction information is trajectory information
- the UE can compare the predicted trajectory information with the UE's actual trajectory information, determine the actual accuracy of the prediction information, and send second feedback information to the first RAN node, the second feedback information It is used to indicate the actual accuracy rate of the forecast information.
- the indication may be an implicit indication or a display indication, and the like. Or described as: the actual accuracy rate of the prediction information sent by the UE to the first RAN node.
- the UE may feed back to the first RAN node the actual accuracy rate of prediction information within a period of time, which may be periodic feedback, or event-based trigger feedback, and the like.
- the UE may calculate the actual accuracy rate of the prediction information every 10 seconds at a period of 10 seconds, and feed it back to the first RAN node.
- the UE also calculates the accuracy rate of the prediction information every 10 seconds, and only sends the actual accuracy rate of the prediction information to the first RAN node when the accuracy rate is lower than a threshold (for example, 80%).
- a threshold for example, 80%
- the UE when the UE finds that its application scenario changes or its performance level changes, it can also send second feedback information to the first RAN node, and the second feedback information is used to indicate the application scenario change or performance level change. Alternatively, the UE sends an application scenario change indication, or a performance level change indication, etc. to the first RAN node.
- Step 707 the first RAN node sends first feedback information to the OAM.
- the OAM receives first feedback information from the first RAN node.
- the first RAN node may send the first feedback information to the OAM periodically, or based on event triggering, etc., and the first feedback information may indicate or display the actual accuracy rate of the prediction information. Or described as, the first RAN node sends the actual accuracy rate of the prediction information to the OAM. Similar to step 706 above, the first RAN node may feed back the actual accuracy rate of the prediction information for a period of time to the OAM. For example, the first RAN node may periodically, or based on event triggering, etc., feed back to the OAM the actual accuracy rate of prediction information for a period of time.
- the first RAN node when the application scenario or performance level related to the first model changes, the first RAN node sends the first feedback information to the OAM, and the first feedback information is used to indicate the change of the application scenario or the change of the performance level .
- the first RAN node sends an application scenario change indication or a performance level change indication to the OAM.
- Step 708 The OAM updates the first model according to the first feedback information.
- the OAM updates the first model according to the first feedback information , comprising: retraining the first model and updating the first model when the actual accuracy rate of the prediction information output by the first model is less than a threshold.
- the OAM may collect training data from the first RAN node and/or UE, or other nodes, etc. The OAM can use the collected training data to retrain the first model.
- the process of model training please refer to the above description.
- the first feedback information is used to indicate a scene change indication, or a model performance level change indication, or a model performance index change indication
- the OAM performs an operation on the first feedback information according to the first feedback information.
- Model updating including: determining the second model according to the scene change indication, or the model performance level change indication, or the model performance index change indication; for example, the first mapping relationship includes the mapping relationship between models, model application scenarios, and model functions
- the OAM can determine the changed scene according to the indicated scene change instruction. According to the above-mentioned first mapping relationship, the model corresponding to the changed scene is determined, which is called the second model.
- the first mapping relationship includes a mapping relationship between a model, a model performance level, and a model function
- the OAM may determine the changed model performance level according to the indicated model performance level change instruction.
- the OAM may determine the changed model performance level according to the indicated model performance index change instruction.
- the model corresponding to the changed model performance level is determined, which is called the second model.
- the first mapping relationship includes a mapping relationship among models, model performance indicators, and model functions, and the OAM determines, according to the first mapping relationship, the model corresponding to the indicated changed model performance indicator, which is called the second model.
- Step 709 The OAM sends the updated model information to the first RAN node; correspondingly, the first RAN node receives the updated model information from the OAM.
- the first RAN node may send the first feedback information to the OAM periodically or based on event triggering.
- the OAM evaluates whether the first model needs to be updated according to the first feedback information, so as to update the first model in time to improve the prediction performance and prediction speed of the first model.
- this application provides a flow chart of cell handover, in which the first RAN node is called the source RAN node, and the second RAN node is called the target RAN node, at least including:
- Step 801 The source RAN node determines the target cell.
- the source RAN node may receive a measurement report reported by the UE, where the measurement report includes cell identity information, cell quality information, and the like.
- the cell may be a current serving cell and/or a neighboring cell, etc., without limitation.
- the UE may periodically report measurement reports to the source RAN node. For example, the UE may periodically measure the signal quality of the serving cell and/or the signal quality of the neighboring cell, and then periodically report the measured information to the source RAN node. Alternatively, the UE may periodically measure the signal quality of the serving cell and/or the signal quality of neighboring cells, and report the measured information to the source RAN node when certain conditions are met.
- the measurement report includes identification information of a cell.
- the identification information of the cell includes a cell global identifier (CGI), a physical cell identifier (PCI) and a frequency point, a cell identifier (cell ID), and a non-public network identifier (non-public network identifier) of the cell.
- CGI cell global identifier
- PCI physical cell identifier
- cell ID cell identifier
- non-public network identifier non-public network identifier
- the CGI may include a public land mobile network (public land mobile network, PLMN ID) and a cell ID.
- the identification information of the cell may also include a tracking area code (tracking area code, TAC) and/or identification information of a network device to which the cell belongs, such as a global network device identification.
- TAC tracking area code
- the measurement report includes the signal quality information of the cell.
- the UE may measure the downlink synchronization channel, channel state information reference signal, demodulation reference signal (demodulation reference signal, DMRS), cell-specific reference signal (cell-specific reference signal, CRS) signal, synchronization signal block (synchronization signal block, SSB), synchronization signal/physical broadcast channel block or other downlink signals to obtain the quality of the cell amount of information.
- demodulation reference signal demodulation reference signal
- CRS cell-specific reference signal
- synchronization signal block synchronization signal block
- SSB synchronization signal/physical broadcast channel block or other downlink signals
- the quality information of the cell may include received signal code power (received signal code power, RSCP), reference signal receiving power (reference signal receiving power, RSRP), reference signal receiving quality (reference signal receiving quality, RSRQ), At least one of signal noise ratio (signal noise ratio, SNR), signal to interference plus noise ratio (signal to interference plus noise ratio, SINR), reference signal strength indication (reference signal strength indication, RSSI) or other signal qualities.
- RSCP received signal code power
- RSRP reference signal receiving power
- RSRQ reference signal receiving quality
- SNR signal noise ratio
- SINR signal to interference plus noise ratio
- RSSI reference signal strength indication
- the cell quality information can be cell level, beam level, synchronization signal/physical broadcast channel block (synchronization signal/physical broadcast channel block, SS/PBCH Block) level, channel state information reference signal (channel-state At least one of information reference signal (CSI-RS) level, air interface technology (numerology) level, slicing (slicing) level, or bandwidth part (bandwidth part, BWP) level.
- the level of the cell quality information refers to the granularity with which the cell quality information is measured. For example, the quality information of a cell is at the cell level, which means that the UE measures each of the multiple cells to be measured to obtain the quality of each cell.
- the quality of the cell is at the beam level, which means that the cell includes at least one beam, and the UE obtains the cell quality information by measuring the beams in the cell.
- a cell includes 3 beams, and the UE can obtain the quality information of beam 1 and beam 2 that meet the beam quality conditions by separately measuring beam 1, beam 2 and beam 3 among the above three beams.
- the UE may obtain the quality information of the cell based on the quality information of the beam 1 and the quality information of the beam 2, such as taking a larger value, calculating an average, or a weighted sum, and the like.
- the UE may report the quality information of beam 1 and the quality information of beam 2 at the same time.
- the UE measuring the quality information of the cell based on other granularities it is similar to the above and will not be described again.
- the cell identification information also includes the corresponding At least one of the identification information of the beam, the identification information of the SS/PBCH Block, the identification information of the CSI-RS, the identification information of the air interface technology, the identification information of the slice, or the identification information of the BWP.
- the source RAN node when receiving the measurement report, may determine whether to switch the serving cell for the UE based on the measurement report. For example, if the service quality of the source cell is lower than a certain threshold, it may be determined that the serving cell needs to be switched for the UE.
- the source RAN node can select a target cell for the UE according to the received measurement report. For example, among the adjacent cells of the UE, a cell with the best signal quality greater than the threshold may be selected as the target cell and the like.
- the source RAN node may determine the target cell based on AI.
- the source RAN node can use an AI model or a machine learning (machine learning, ML) model to predict the target cell.
- the input data of the AI model or the ML model includes at least one of the following: the measurement report of the UE, the historical and/or real-time load of the adjacent cell, the historical trajectory information of the UE, the geographic coordinates of the UE at the current moment, the UE's current The running direction, or the moving speed of the UE, etc.
- the output data of the AI model or the ML model includes at least one of the following: a target cell of the UE, a handover time stamp, and the like.
- Step 802 The source RAN node sends model information to the target RAN node corresponding to the target cell.
- the source RAN node may send model information to the target RAN node, where the model information includes an index of the first model.
- the model information may also include at least one of the following: the application scenario of the first model, the function of the first model, or which data the first model needs to collect for model reasoning.
- the RAN node that performs model reasoning needs to request the UE's historical handover success or failure from the RAN node that the UE has accessed before.
- the process that the source RAN node needs to notify the target RAN node which data needs to be collected for model reasoning can also be described as: the RAN node sends first indication information to the target RAN node, and the first indication information is used to indicate Type information of the input data of the first model.
- the source RAN node can notify the target RAN node in advance, which data needs to be collected for model reasoning during the model reasoning process, so that the target source RAN can prepare for model reasoning and improve the efficiency of model reasoning.
- Step 803 the source RAN node sends the historical information of the UE to the target RAN node, where the historical information at least includes historical data and/or historical information collected by the source RAN node from the UE.
- the above historical information of the UE may be information required by the first model for model reasoning, etc., that is to say, when the target RAN node uses the first model to perform model reasoning, the above historical information can be used as the first model part of the input data.
- the above process can also be described as: the source RAN node sends the historical data information required for model reasoning to the target RAN node, and the historical data information is used to determine the input data of the first model.
- the target RAN node can directly use the above historical data as the input of the first model to perform model reasoning; or, the target RAN node can process the above historical data, and use the processed historical data as the input of the first model, etc., without be restricted.
- step 803 may be performed after step 802, or may be performed after step 804, which is not limited.
- step 803 it is described by taking execution after step 802 as an example.
- step 803 may be performed after step 802 .
- step 803 may be performed after step 804 .
- Step 804 The UE is handed over from the source cell corresponding to the source RAN node to the target cell corresponding to the target RAN node. After handover, the target cell serves as the UE's serving cell.
- Step 805 The target RAN node sends second indication information to the source RAN node, where the second indication information is used to indicate at least one of the following: the UE switches to the target cell, the source RAN node deletes the first model, or the source RAN node Nodes release corresponding computing resources.
- the source RAN node allocates 1% of resources for model reasoning to the UE, and when the UE is handed over to the target cell, the source RAN node can release 1% of the computing resources allocated for the UE, delete the first model, and so on.
- Step 806 The source RAN node deletes the first model and releases corresponding computing resources.
- Step 807 The target RAN node sends a second model request message to the OAM, where the second model request message is used to indicate the index of the first model.
- Step 808 The OAM may determine the first model according to the index of the first model.
- the OAM may determine the first model according to the index of the first model.
- Step 809 the OAM sends the information of the first model to the target RAN node.
- Step 8010 The target RAN node performs model reasoning according to the first model to obtain prediction information.
- Step 8011 The target RAN node sends prediction information to the UE.
- Step 8012 UE sends second feedback information to the target RAN node.
- Step 8013 the target RAN node sends the first feedback information to the OAM.
- Step 8014 OAM executes the first model update.
- Step 8015 The OAM sends the updated information of the first model to the target RAN node.
- steps 809 to 8015 For the specific execution process of steps 809 to 8015, refer to the above description in FIG. 7 .
- the source RAN node ensures the continuity of the AI service by sending the index of the first model to the target RAN node; and after the UE handover is successful, the target RAN node notifies the source RAN node Delete model-related information and release computing resources, which can save memory resource occupation and computing resources of RAN nodes occupy.
- the OAM and RAN nodes include hardware structures and/or software modules corresponding to each function.
- the application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a certain function is executed by hardware or computer software drives the hardware depends on the specific application scenario and design constraints of the technical solution.
- 9 and 10 are schematic structural diagrams of possible devices provided by the present application. These communication devices can be used to implement the functions of the OAM or RAN node in the above method, and therefore can also realize the beneficial effects of the above method.
- a communication device 900 includes a processing unit 910 and a transceiver unit 920 .
- the communication device 900 is configured to implement functions of the OAM, source RAN node, or target RAN node in the method shown in FIG. 7 or FIG. 8 above.
- the transceiver unit 920 is used to receive the first model request message from the first access network device; the processing unit 910 is used to The first model request message and the first mapping relationship determine the first model, the first mapping relationship includes the mapping relationship between the model, the model application scenario, and the model function; or the first mapping relationship includes the model, the model performance level, The mapping relationship with the model function; the transceiver unit 920 is further configured to send the information of the first model to the first access network device.
- the model request message is used to indicate the model application scenario and model function of the first model; or, the first model request message is used to indicate the model function and model performance level of the first model; the transceiver unit is also used to Information is received from the first model of the OAM.
- the transceiver unit 920 is configured to receive the index of the first model from the first access network device, and send the second A model request message, the second model request message is used to indicate the index of the first model, and receive the information of the first model from the OAM.
- processing unit 910 and the transceiver unit 920 can be directly obtained by referring to the related descriptions in the method shown in FIG. 5 , FIG. 7 or FIG. 8 , and will not be repeated here.
- a 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.
- the interface circuit 1020 may be a transceiver, an input/output interface, or a pin.
- the communication device 1000 may further include a memory 1030 for storing instructions executed by the processor 1010 or storing input data required by the processor 1010 to execute the instructions or storing data generated by the processor 1010 after executing the instructions.
- the processor 1010 is used to implement the functions of the processing unit 910
- the interface circuit 1020 is used to implement the functions of the transceiver unit 920 .
- the OAM module realizes the function of OAM in the above-mentioned method.
- the OAM module receives information from other modules in the OAM (such as radio frequency modules or antennas), and the information is sent to the OAM by the RAN node; or, the OAM module sends information to other modules in the OAM (such as radio frequency modules or antennas), This information is sent by the OAM to the RAN node.
- the OAM module here may be an OAM baseband chip, or other modules.
- the RAN node module implements the RAN in the above-mentioned method The function of the node.
- the RAN node module receives information from other modules in the RAN node (such as radio frequency modules or antennas), and the information is sent by the terminal to the RAN node; or, the RAN node module sends information to other modules in the RAN node (such as radio frequency modules or antennas) ) to send information, which is sent by the RAN node to the terminal.
- the RAN node module here may be a baseband chip of the RAN node, or may be a DU or other modules, and the DU here may be a DU under an open radio access network (open radio access network, O-RAN) architecture.
- processor in this application can be a central processing unit (central processing unit, CPU), and can also be other general processors, digital signal processors (digital signal processor, DSP), application specific integrated circuits (application specific integrated circuit, ASIC), field programmable gate array (field programmable gate array, FPGA) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof.
- CPU central processing unit
- DSP digital signal processor
- ASIC application specific integrated circuit
- FPGA field programmable gate array
- a general-purpose processor can be a microprocessor, or any conventional processor.
- the memory in this application can be random access memory, flash memory, read-only memory, programmable read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, register, hard disk, mobile hard disk, CD-ROM or any other form of storage media known in the art.
- An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium.
- a storage medium may also be an integral part of the processor.
- the processor and storage medium can be located in the ASIC.
- the ASIC can be located in the base station or the terminal.
- the processor and the storage medium may also exist in the base station or the terminal as discrete components.
- the methods in this application may be fully or partially implemented by software, hardware, firmware or any combination thereof.
- software When implemented using software, it may be implemented in whole or in part in the form of a computer program product.
- the computer program product comprises one or more computer programs or instructions. When the computer programs or instructions are loaded and executed on the computer, the processes or functions described in this application are executed in whole or in part.
- the computer may be a general computer, a special computer, a computer network, a network device, a user device, a core network device, an OAM or other programmable devices.
- the computer program or instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer program or instructions may be downloaded from a website, computer, A server or data center transmits to another website site, computer, server or data center by wired or wireless means.
- the computer-readable storage medium may be any available medium that can be accessed by a computer, or a data storage device such as a server or a data center integrating one or more available media.
- the available medium may be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; it may also be an optical medium, such as a digital video disk; or it may be a semiconductor medium, such as a solid state disk.
- the computer readable storage medium may be a volatile or a nonvolatile storage medium, or may include both volatile and nonvolatile types of storage media.
- “at least one” means one or more, and “multiple” means two or more.
- “And/or” describes the association relationship of associated objects, indicating that there may be three types of relationships, for example, A and/or B, which can mean: A exists alone, A and B exist at the same time, and B exists alone, where A, B can be singular or plural.
- the character “/” generally indicates that the contextual objects are an “or”relationship; in the formulas of this application, the character “/” indicates that the contextual objects are a “division” Relationship.
- “Including at least one of A, B or C” may mean: including A; including B; including C; including A and B; including A and C; including B and C; including A, B, and C.
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
一种模型配置方法及装置,该方法包括:操作管理维护OAM接收来自第一接入网设备的第一模型请求消息;OAM根据所述第一模型请求消息和第一映射关系,确定第一模型,所述第一映射关系包括模型、模型应用场景、和模型功能的映射关系;或者所述第一映射关系包括模型、模型性能等级、和模型功能的映射关系;OAM向所述第一接入网设备发送所述第一模型的信息。采用本申请的方法及装置,可为第一接入网设备配置与应用场景或性能指标相匹配的第一模型,提高模型推理的性能和速度。
Description
相关申请的交叉引用
本申请要求在2022年01月26日提交中国专利局、申请号为202210096175.1、申请名称为“一种模型配置方法及装置”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
本申请涉及通信技术领域,尤其涉及一种模型配置方法及装置。
在无线通信网络中,例如在移动通信网络中,网络支持的业务越来越多样,因此需要满足的需求越来越多样。例如,网络需要能够支持超高速率、超低时延、和/或超大连接。该特点使得网络规划、网络配置、和/或资源调度越来越复杂。这些新需求、新场景和新特性给网络规划、运维和高效运营带来了前所未有的挑战。为了迎接该挑战,可以将人工智能技术引入无线通信网络中,从而实现网络智能化。基于此,如何向网络中的接入网设备配置人工智能模型是一个值得研究的问题。
发明内容
本申请提供一种模型配置方法及装置,以实现为第一接入网设备配置与应用场景或性能等级相匹配的第一模型,提高模型预测的性能和速度。
第一方面,提供一种模型配置方法,该方法的执行主体为操作管理维护OAM,还可以为配置于OAM中的部件(处理器、芯片或其它),或者可以为软件模块等,包括:
接收来自第一接入网设备的第一模型请求消息;根据所述第一模型请求消息和第一映射关系,确定第一模型,所述第一映射关系包括模型、模型应用场景、和模型功能的映射关系;或者所述第一映射关系包括模型、模型性能等级、和模型功能的映射关系;向所述第一接入网设备发送所述第一模型的信息。可选的,模型、模型性能等级、和模型功能的映射关系可以替换为:模型、模型性能指标、和模型功能的映射关系。
通过上述方法,由于在不同应用场景下,对实现同一功能的模型的性能指标要求不同。在本申请中,OAM存储有第一映射关系,该第一映射关系为模型+模型应用场景+模型功能的映射关系。通过该映射关系,针对同一个模型功能,在不同的应用场景下,可分配该场景对应的模型,从而使得该模型满足该模型对应场景中对性能指标的要求。相较于,无论何种场景,都采用同一模型进行模型推理。采用本申请的方案,可提高预测性能和预测速度等。或者,在本申请中,OAM建立第一映射关系,该第一映射关系为模型+模型性能等级(或模型性能指标)+模型功能的映射关系。由于不同应用场景下对模型的模型性能指标要求不同,模型性能指标与性能等级间存在对应关系,OAM可根据第一映射关系,为不同应用场景下的不同模型性能指标,配置不同的模型。相对于,无论何种场景的模型性能指标下,采用同一模型进行模型推理,可提高预测性能和预测速度等。
在一种可能的设计中,所述第一模型请求消息用于指示所述第一模型的模型应用场景和模型功能。
在一种可能的设计中,所述第一模型请求消息用于指示所述第一模型的模型功能和模型性能等级,或者用于指示所述第一模型的模型功能和模型性能指标。
在一种可能的设计中,所述第一模型的信息用于指示所述第一模型的以下至少一项:模型索引、模型的结构信息、模型的参数、模型输入格式、模型输出格式、模型性能指标、模型应用场景、模型功能、或训练参数。
通过上述方法,OAM将第一模型的信息发送给第一接入网设备。第一接入网设备根据第一模型的信息,可确定或恢复第一模型。可选的,第一模型的信息中可包括训练参数,第一接入网设备可后续根据训练参数,继续对第一模型训练,满足了对模型的各种需求。
在一种可能的设计中,还包括:接收来自所述第一接入网设备的第一反馈信息;根据所述第一反馈信息,对所述第一模型更新。
在一种可能的设计中,所述第一反馈信息用于指示所述第一模型所输出的预测信息的实际准确率,所述根据所述反馈信息,对所述第一模型更新,包括:当所述第一模型所输出的预测信息的实际准确率小于阈值时,重新对所述第一模型训练,并更新所述第一模型。
通过上述方法,第一接入网设备在确定第一模型时,可根据第一模型进行模型推理,推理结果可称为预测信息。根据预测信息确定第一反馈信息,该第一反馈信息中包括第一模型输出的预测信息的实际准确率,且将第一反馈信息反馈给OAM;OAM根据该第一反馈信息,对第一模型进行更新,从而提高第一模型推理结果的准确性。
在一种可能的设计中,所述第一反馈信息用于指示场景变更指示,模型性能等级变更指示,或模型性能指标变更指示,所述根据所述第一反馈信息,对所述第一模型更新,包括:根据所述场景变更指示、模型性能等级变更指示或模型性能指标变更指示,为所述终端设备选择第二模型;向所述第一接入网设备发送所述第二模型的信息,所述第二模型用于更新所述第一模型。
与上述方法相似,不同的是,在该设计中,当应用场景、模型性能等级或模型性能指标变更时,第一接入网设备向OAM发送用于指示应用场景变更、模型性能等级变更或模型性能指标变更的指示信息,OAM根据该变更后的应用场景、模型性能等级或模型性能指标,重新为第一接入网设备分配模型用于模型推理,可尽量保证为第一接入网设备所分配的模型满足应用场景、模型性能等级或模型性能指标的要求。
在一种可能的设计中,还包括:接收来自第二接入网设备的第二模型请求消息,所述第二模型请求消息用于指示所述第一模型的索引;根据所述第一模型的索引,确定所述第一模型;向所述第二接入网设备发送所述第一模型的信息。
第二方面,提供一种模型配置方法,该方法与第一方面对应的第一接入网设备侧,有益效果可参见上述第一方面,该方法的执行主体为第一接入网设备,还可以为配置于第一接入网设备中的部件(处理器、芯片或其它),或者可以为软件模块等,包括:
向操作维护管理OAM发送第一模型请求消息,所述第一模型请求消息用于指示所述第一模型的模型应用场景和模型功能;或者,所述第一模型请求消息用于指示所述第一模型的模型功能和模型性能等级,或者用于指示所述第一模型的模型功能和模型性能指标;接收来自所述OAM的第一模型的信息。
在一种可能的设计中,在向OAM发送第一模型请求消息之前,还包括:确定是否需
要提供基于人工智能AI的增强服务。
在一种可能的设计中,所述第一模型的信息用于指示所述第一模型的以下至少一项:模型索引、模型的结构信息、模型的参数、模型输入格式、模型输出格式、模型性能指标、模型应用场景、模型功能、或训练参数。
在一种可能的设计中,还包括:根据所述第一模型,执行模型推理,获得预测信息。
在一种可能的设计中,还包括:向终端设备发送所述预测信息,所述预测信息中包含预测结果。
在一种可能的设计中,所述预测信息中还包含对所述预测结果所预测的准确率。
在一种可能的设计中,还包括:接收来自所述终端设备的第二反馈信息,所述第二反馈信息用于指示所述预测信息的实际准确率。
在一种可能的设计中,还包括:向所述OAM发送第一反馈信息,所述第一反馈信息用于指示所述预测信息的实际准确率、应用场景变更指示、模型性能等级变更指示或模型性能指标变更指示。
应当指出,在本申请中,上述第一模型可为终端设备提供人工智能AI或机器学习ML推理服务,或者为第一接入网设备提供AI或ML推理服务。在第一模型用于为终端设备提供AI或ML推理服务时,还包括:第一RAN节点将第一模型的推理信息(或称为预测信息)发送给终端设备;终端设备根据该推理信息,向第一接入网设备发送第二反馈信息,第二反馈信息中可用于指示推理信息的实际准确率,第一接入网设备可直接将所述第二反馈信息转发给OAM;或者,第一接入网设备根据第二反馈信息,获取推理信息的实际准确率,向OAM发送用于指示推理信息的实际准确率的第一反馈信息等。OAM可根据推理信息的实际准确率对第一模型进行更新,提高第一模型的推理性能和推理速度等。
在一种可能的设计中,还包括:确定所述终端设备的目标小区;向所述目标小区对应的第二接入网设备发送所述第一模型的索引。
在一种可能的设计中,还包括:向所述目标小区对应的第二接入网设备发送第一指示信息,所述第一指示信息用于指示所述第一模型的输入数据的类型信息。
在一种可能的设计中,还包括:向所述第二接入网设备发送所述模型推理所需的历史数据信息。
在一种可能的设计中,还包括:接收来自所述第二接入网设备的第二指示信息,所述第二指示信息用于指示以下至少一项:所述终端切换到所述目标小区、所述第一接入网设备删除所述第一模型,或所述第一接入网设备释放相应的计算资源;删除所述第一模型,释放相应的计算资源。
第三方面,提供一种模型配置方法,该方法与第一方面对应的第一接入网设备侧,有益效果可参见上述第一方面,该方法的执行主体为第二接入网设备,还可以为配置于第二接入网设备中的部件(处理器、芯片或其它),或者可以为软件模块等,包括:
接收来自第一接入网设备的第一模型的索引;向操作维护管理OAM发送第二模型请求消息,所述第二模型请求消息用于指示所述第一模型的索引;接收来自所述OAM的所述第一模型的信息。
在一种可能的设计中,所述第一模型的信息用于指示所述第一模型的以下至少一项:模型索引、模型的结构信息、模型的参数、模型输入格式、模型输出格式、模型性能指标、模型应用场景、模型功能、或训练参数。
在一种可能的设计中,还包括:接收来自所述第一接入网设备的第一指示信息,所述第一指示信息用于指示所述第一模型的输入数据的类型信息。
在一种可能的设计中,还包括:接收来自所述第一接入网设备的模型推理所需的历史数据信息;在所述模型推理中,所述历史数据信息用于确定所述第一模型的输入数据。
在一种可能的设计中,还包括:向所述第一接入网设备发送第二指示信息,所述第二指示信息用于指示以下至少一项:所述终端切换到所述目标小区、所述第一接入网设备删除所述第一模型,或所述第一接入网设备释放相应的计算资源。
第四方面,提供一种装置,有益效果可参见第一方面的记载,该装置可以是操作管理维护OAM、或者配置于OAM中的装置、或者能够和OAM匹配使用的装置。
在一种设计中,该装置包括执行第一方面中所描述的方法/操作/步骤/动作一一对应的单元,该单元可以是硬件电路,也可是软件,也可以是硬件电路结合软件实现。
示例性地,该装置可以包括处理单元和通信单元,且处理单元和通信单元可以执行上述第一方面任一种设计示例中的相应功能,具体的:
通信单元,用于接收来自第一接入网设备的第一模型请求消息;处理单元,用于根据所述第一模型请求消息和第一映射关系,确定第一模型,所述第一映射关系包括模型、模型应用场景、和模型功能的映射关系;或者所述第一映射关系包括模型、模型性能等级、和模型功能的映射关系;通信单元,还用于向所述第一接入网设备发送所述第一模型的信息。上述处理单元和通信单元的具体执行过程可以参考第一方面,这里不再赘述。
示例性地,所述装置包括存储器,用于实现上述第一方面描述的方法。所述装置还可以包括存储器,用于存储指令和/或数据。所述存储器与所述处理器耦合,所述处理器执行所述存储器中存储的程序指令时,可以实现上述第一方面描述的方法。所述装置还可以包括通信接口,所述通信接口用于该装置和其它设备进行通信。示例性地,通信接口可以是收发器、电路、总线、模块、管脚或其它类型的通信接口,其它设备可以为网络设备等。在一种可能的设计中,该装置包括:
存储器,用于存储程序指令;
通信接口,用于接收来自第一接入网设备的第一模型请求消息;
处理器,用于根据所述第一模型请求消息和第一映射关系,确定第一模型,所述第一映射关系包括模型、模型应用场景、和模型功能的映射关系;或者所述第一映射关系包括模型、模型性能等级、和模型功能的映射关系;
通信接口,还用于向所述第一接入网设备发送所述第一模型的信息。
关于通信接口与处理器的具体执行过程,可参见上述第一方面的记载,不再赘述。
第五方面,提供一种装置,有益效果可参见第二方面的记载,该装置可以是第一接入网设备,或者配置于第一接入网设备中的装置,或者能够和第一接入网设备匹配使用的装置。一种设计中,该装置包括执行第二方面中所描述的方法/操作/步骤/动作一一对应的单元,该单元可以是硬件电路,也可是软件,也可以是硬件电路结合软件实现。
示例性地,该装置可以包括处理单元和通信单元,且处理单元和通信单元可以执行上述第二方面任一种设计示例中的相应功能,具体的:
通信单元,用于向操作维护管理OAM发送第一模型请求消息,所述第一模型请求消息用于指示所述第一模型的模型应用场景和模型功能;或者,所述第一模型请求消息用于指示所述第一模型的模型功能和模型性能等级;和接收来自所述OAM的第一模型的信息。
处理单元,用于根据第一模型的信息,确定第一模型。上述处理单元和通信单元的具体执行过程可以参考第二方面,这里不再赘述。
示例性地,所述装置包括存储器,用于实现上述第二方面描述的方法。所述装置还可以包括存储器,用于存储指令和/或数据。所述存储器与所述处理器耦合,所述处理器执行所述存储器中存储的程序指令时,可以实现上述第二方面描述的方法。所述装置还可以包括通信接口,所述通信接口用于该装置和其它设备进行通信。示例性地,通信接口可以是收发器、电路、总线、模块、管脚或其它类型的通信接口,其它设备可以为终端等。在一种可能的设计中,该装置包括:
存储器,用于存储程序指令;
通信接口,用于向操作维护管理OAM发送第一模型请求消息,所述第一模型请求消息用于指示所述第一模型的模型应用场景和模型功能;或者,所述第一模型请求消息用于指示所述第一模型的模型功能和模型性能等级;和接收来自所述OAM的第一模型的信息。
处理器,用于根据所述第一模型的信息,确定第一模型。
关于通信接口与处理器的具体执行过程,可参见上述第二方面的记载,不再赘述。
第六方面,提供一种装置,有益效果可参见第三方面的记载,该装置可以是第二接入网设备,或者配置于第二接入网设备中的装置,或者能够和第二接入网设备匹配使用的装置。一种设计中,该装置包括执行第三方面中所描述的方法/操作/步骤/动作一一对应的单元,该单元可以是硬件电路,也可是软件,也可以是硬件电路结合软件实现。
示例性地,该装置可以包括处理单元和通信单元,且处理单元和通信单元可以执行上述第三方面任一种设计示例中的相应功能,具体的:
通信单元,用于接收来自第一接入网设备的第一模型的索引;向操作维护管理OAM发送第二模型请求消息,所述第二模型请求消息用于指示所述第一模型的索引;和接收来自所述OAM的所述第一模型的信息。处理单元,用于根据第一模型的信息,确定第一模型。上述处理单元和通信单元的具体执行过程可以参考第三方面,这里不再赘述。
示例性地,所述装置包括存储器,用于实现上述第三方面描述的方法。所述装置还可以包括存储器,用于存储指令和/或数据。所述存储器与所述处理器耦合,所述处理器执行所述存储器中存储的程序指令时,可以实现上述第三方面描述的方法。所述装置还可以包括通信接口,所述通信接口用于该装置和其它设备进行通信。示例性地,通信接口可以是收发器、电路、总线、模块、管脚或其它类型的通信接口,其它设备可以为终端等。在一种可能的设计中,该装置包括:
存储器,用于存储程序指令;
通信接口,用于接收来自第一接入网设备的第一模型的索引;向操作维护管理OAM发送第二模型请求消息,所述第二模型请求消息用于指示所述第一模型的索引;和接收来自所述OAM的所述第一模型的信息。
处理器,用于根据所述第一模型的信息,确定第一模型。
关于通信接口与处理器的具体执行过程,可参见上述第三方面的记载,不再赘述。
第七方面,本申请还提供一种计算机可读存储介质,包括指令,当其在计算机上运行时,使得计算机执行第一方面、第二方面或第三方面中任一方面的方法。
第八方面,本申请还提供一种芯片***,该芯片***包括处理器,还可以包括存储器,用于实现第一方面、第二方面、或第三方面中任一方面的方法。该芯片***可以由芯片构
成,也可以包含芯片和其他分立器件。
第九方面,本申请还提供一种计算机程序产品,包括指令,当其在计算机上运行时,使得计算机执行第一方面、第二方面或第三方面任一方面的方法。
第十方面,本申请还提供一种***,该***中包括:第四方面的装置和第五方面的装置;或者,第四方面的装置、第五方面的装置和第六方面的装置。
图1为本申请提供的通信***的示意图;
图2和图3为本申请提供的模型部署的示意图;
图4a和图4b为本申请提供的通信***的架构示意图;
图4c为本申请提供的AI模型的应用示意图;
图5为本申请提供的神经元的示意图;
图6为本申请提供的神经网络的示意图;
图7和图8为本申请提供的流程示意图;
图9和图10为本申请提供的装置的结构示意图。
图1是本申请能够应用的通信***1000的架构示意图。如图1所示,该通信***包括无线接入网100和核心网200,可选的,通信***1000还可以包括互联网300。其中,无线接入网100可以包括至少一个接入网设备(如图1中的110a和110b),还可以包括至少一个终端设备(如图1中的120a-120j)。终端设备通过无线的方式与接入网设备相连,接入网设备通过无线或有线方式与核心网连接。核心网设备与接入网设备可以是独立的不同的物理设备,或者可以是将核心网设备的功能与接入网设备的逻辑功能集成在同一个物理设备上,或者可以是一个物理设备上集成了部分核心网设备的功能和部分的接入网设备的功能。终端设备和终端设备之间以及接入网设备和接入网设备之间可以通过有线或无线的方式相互连接。图1只是示意图,该通信***中还可以包括其它网络设备,如还可以包括无线中继设备和无线回传设备等,在图1中未画出。
接入网设备可以是基站(base station)、演进型基站(evolved NodeB,eNodeB)、发送接收点(transmission reception point,TRP)、第五代(5th generation,5G)移动通信***中的下一代基站(next generation NodeB,gNB)、开放无线接入网(open radio access network,O-RAN)中的接入网设备、第六代(6th generation,6G)移动通信***中的下一代基站、未来移动通信***中的基站或无线保真(wireless fidelity,WiFi)***中的接入节点等;或者可以是完成基站部分功能的模块或单元,例如,可以是集中式单元(central unit,CU)、分布式单元(distributed unit,DU)、集中单元控制面(CU control plane,CU-CP)模块、或集中单元用户面(CU user plane,CU-UP)模块。接入网设备可以是宏基站(如图1中的110a),也可以是微基站或室内站(如图1中的110b),还可以是中继节点或施主节点等。本申请中对接入网设备所采用的具体技术和具体设备形态不做限定。
在本申请中,用于实现接入网设备的功能的装置可以是接入网设备;也可以是能够支持接入网设备实现该功能的装置,例如芯片***、硬件电路、软件模块、或硬件电路加软
件模块,该装置可以被安装在接入网设备中或可以与接入网设备匹配使用。在本申请中,芯片***可以由芯片构成,也可以包括芯片和其他分立器件。为了便于描述,下文以用于实现接入网设备的功能的装置是接入网设备,接入网设备为RAN节点为例,描述本申请提供的技术方案。
(1)协议层结构。
接入网设备和终端设备之间的通信遵循一定的协议层结构。该协议层结构可以包括控制面协议层结构和用户面协议层结构。例如,控制面协议层结构可以包括无线资源控制(radio resource control,RRC)层、分组数据汇聚层协议(packet data convergence protocol,PDCP)层、无线链路控制(radio link control,RLC)层、媒体接入控制(media access control,MAC)层和物理层等协议层的功能。例如,用户面协议层结构可以包括PDCP层、RLC层、MAC层和物理层等协议层的功能,在一种可能的实现中,PDCP层之上还可以包括业务数据适配协议(service data adaptation protocol,SDAP)层。
可选的,接入网设备和终端设备之间的协议层结构还可以包括人工智能(artificial intelligence,AI)层,用于传输AI功能相关的数据。
(2)集中单元(central unit,CU)和分布单元(distributed unit,DU)。
接入设备可以包括CU和DU。多个DU可以由一个CU集中控制。作为示例,CU和DU之间的接口可以称为F1接口。其中,控制面(control panel,CP)接口可以为F1-C,用户面(user panel,UP)接口可以为F1-U。本申请不限制各接口的具体名称。CU和DU可以根据无线网络的协议层划分:比如,PDCP层及以上协议层的功能设置在CU,PDCP层以下协议层(例如RLC层和MAC层等)的功能设置在DU;又比如,PDCP层以上协议层的功能设置在CU,PDCP层及以下协议层的功能设置在DU,不予限制。
上述对CU和DU的处理功能按照协议层的划分仅仅是一种举例,也可以按照其他的方式进行划分。例如可以将CU或者DU划分为具有更多协议层的功能,又例如将CU或DU还可以划分为具有协议层的部分处理功能。在一种设计中,将RLC层的部分功能和RLC层以上的协议层的功能设置在CU,将RLC层的剩余功能和RLC层以下的协议层的功能设置在DU。在另一种设计中,还可以按照业务类型或者其他***需求对CU或者DU的功能进行划分,例如按时延划分,将处理时间需要满足时延要求的功能设置在DU,不需要满足该时延要求的功能设置在CU。在另一种设计中,CU也可以具有核心网的一个或多个功能。示例性的,CU可以设置在网络侧方便集中管理。在另一种设计中,将DU的无线单元(radio unit,RU)拉远设置。可选的,RU可以具有射频功能。
可选的,DU和RU可以在物理层(physical layer,PHY)进行划分。例如,DU可以实现PHY层中的高层功能,RU可以实现PHY层中的低层功能。其中,用于发送时,PHY层的功能可以包括以下至少一项:添加循环冗余校验(cyclic redundancy check,CRC)码、信道编码、速率匹配、加扰、调制、层映射、预编码、资源映射、物理天线映射、或射频发送功能。用于接收时,PHY层的功能可以包括以下至少一项:CRC校验、信道解码、解速率匹配、解扰、解调、解层映射、信道检测、资源解映射、物理天线解映射、或射频接收功能。其中,PHY层中的高层功能可以包括PHY层的一部分功能,例如该部分功能更加靠近MAC层,PHY层中的低层功能可以包括PHY层的另一部分功能,例如该部分功能更加靠近射频功能。例如,PHY层中的高层功能可以包括添加CRC码、信道编码、速率匹配、加扰、调制、和层映射,PHY层中的低层功能可以包括预编码、资源映射、物
理天线映射、和射频发送功能;或者,PHY层中的高层功能可以包括添加CRC码、信道编码、速率匹配、加扰、调制、层映射和预编码,PHY层中的低层功能可以包括资源映射、物理天线映射、和射频发送功能。例如,PHY层中的高层功能可以包括CRC校验、信道解码、解速率匹配、解码、解调、和解层映射,PHY层中的低层功能可以包括信道检测、资源解映射、物理天线解映射、和射频接收功能;或者,PHY层中的高层功能可以包括CRC校验、信道解码、解速率匹配、解码、解调、解层映射、和信道检测,PHY层中的低层功能可以包括资源解映射、物理天线解映射、和射频接收功能。
示例性的,CU的功能可以由一个实体来实现,或者也可以由不同的实体来实现。例如,可以对CU的功能进行进一步划分,即将控制面和用户面分离并通过不同实体来实现,分别为控制面CU实体(即CU-CP实体)和用户面CU实体(即CU-UP实体)。该CU-CP实体和CU-UP实体可以与DU相耦合,共同完成接入网设备的功能。
可选的,上述DU、CU、CU-CP、CU-UP和RU中的任一个可以是软件模块、硬件结构、或者软件模块+硬件结构,不予限制。其中,不同实体的存在形式可以是不同的,不予限制。例如DU、CU、CU-CP、CU-UP是软件模块,RU是硬件结构。这些模块及其执行的方法也在本申请的保护范围内。
一种可能的实现中,接入网设备包括CU-CP、CU-UP、DU和RU。例如,本申请的执行主体包括DU,或者包括DU和RU,或者包括CU-CP、DU和RU,或者包括CU-UP、DU和RU,不予限制。各模块所执行的方法也在本申请的保护范围内。
终端设备也可以称为终端、用户设备(user equipment,UE)、移动台、移动终端设备等。终端设备可以广泛应用于各种场景中的通信,例如包括但不限于以下至少一个场景:设备到设备(device-to-device,D2D)、车物(vehicle to everything,V2X)、机器类通信(machine-type communication,MTC)、物联网(internet of things,IOT)、虚拟现实、增强现实、工业控制、自动驾驶、远程医疗、智能电网、智能家具、智能办公、智能穿戴、智能交通、或智慧城市等。终端设备可以是手机、平板电脑、带无线收发功能的电脑、可穿戴设备、车辆、无人机、直升机、飞机、轮船、机器人、机械臂、或智能家居设备等。本申请对终端设备所采用的具体技术和具体设备形态不做限定。
在本申请中,用于实现终端设备的功能的装置可以是终端设备;也可以是能够支持终端设备实现该功能的装置,例如芯片***、硬件电路、软件模块、或硬件电路加软件模块,该装置可以被安装在终端设备中或可以与终端设备匹配使用。为了便于描述,下文以用于实现终端设备的功能的装置是终端设备,终端设备为UE为例,描述本申请提供的技术方案。
基站和终端设备可以是固定位置的,也可以是可移动的。基站和/或终端设备可以部署在陆地上,包括室内或室外、手持或车载;也可以部署在水面上;还可以部署在空中的飞机、气球和人造卫星上。本申请对基站和终端设备的应用场景不做限定。基站和终端设备可以部署在相同的场景或不同的场景,例如,基站和终端设备同时部署在陆地上;或者,基站部署在陆地上,终端设备部署在水面上等,不再一一举例。
基站和终端设备的角色可以是相对的,例如,图1中的直升机或无人机120i可以被配置成移动基站,对于那些通过120i接入到无线接入网100的终端设备120j来说,终端设备120i是基站;但对于基站110a来说,120i是终端设备,即110a与120i之间是通过无线空口协议进行通信的。110a与120i之间也可以是通过基站与基站之间的接口协议进行通信
的,此时,相对于110a来说,120i也是基站。因此,基站和终端设备都可以统一称为通信装置,图1中的110a和110b可以称为具有基站功能的通信装置,图1中的120a-120j可以称为具有终端设备功能的通信装置。
在本申请中,可在前述图1所示的通信***中引入独立的网元如称为AI网元、或AI节点等)来实现AI相关的操作,AI网元可以和通信***中的接入网设备之间直接连接,或者可以通过第三方网元和接入网设备实现间接连接。其中,第三方网元可以是认证管理功能(authentication management function,AMF)、或用户面功能(user plane function,UPF)等核心网网元;或者,可以在通信***中的其他网元内配置AI功能、AI模块或AI实体来实现AI相关的操作,例如该其他网元可以是接入网设备(如gNB)、核心网设备、或网管(operation,administration and maintenance,OAM)等,在这种情况下,执行AI相关的操作的网元为内置AI功能的网元。其中,上述OAM用于对接入网设备和/或核心网设备等进行操作、管理和维护等。
在本申请中,如图2或图3所示,核心网设备、接入网设备、终端设备或OAM等中的至少一个设备可以部署有AI模型,利用该AI模型实现相应的功能。在本申请中,不同节点中部署的AI模型可以相同或不同。模型不同包括以下至少一项不同:模型的结构参数不同,例如,模型的层数和/或权值等不同;模型的输入参数不同;或模型的输出参数不同等。其中,模型的输入参数和/或模型的输出参数不同可以描述为模型的功能不同。与上述图2不同的是,在图3中,将接入网设备的功能拆分为CU和DU。可选的,CU和DU可以为O-RAN架构下的CU和DU。CU中可以部署有一个或多个AI模型。和/或,DU中可以部署有一个或多个AI模型。可选的,还可以进一步,将图3中的CU拆分为CU-CP和CU-UP。可选的,CU-CP中可以部署有一个或多个AI模型。和/或,CU-UP中可以部署有一个或多个AI模型。可选的,图2或图3中,接入网设备的OAM和核心网设备的OAM可以分别独立部署。
可选的,图4a为本申请的一种通信***的架构。如图4a所示,在第一种设计中,接入网设备中包括近实时接入网智能控制(RAN intelligent controller,RIC)模块,用于进行模型训练和推理。例如,近实时RIC可以用于训练AI模型,利用该AI模型进行推理。例如,近实时RIC可以从CU、DU或RU中的至少一项获得网络侧和/或终端侧的信息,该信息可以作为训练数据或者推理数据。可选的,近实时RIC可以将推理结果递交至CU、DU、RU或终端设备中的至少一项。可选的,CU和DU之间可以交互推理结果。可选的,DU和RU之间可以交互推理结果,例如近实时RIC将推理结果递交至DU,由DU转发给RU。
或者,在第二种设计中,如图4a所示,接入网设备之外包括非实时RIC(可选的,非实时RIC可以位于OAM中或者核心网设备中),用于进行模型训练和推理。例如,非实时RIC用于训练AI模型,利用该模型进行推理。例如,非实时RIC可以从CU、DU或RU中的至少一项获得网络侧和/或终端侧的信息,该信息可以作为训练数据或者推理数据,该推理结果可以被递交至CU、DU、RU或终端设备中的至少一项。可选的,CU和DU之间可以交互推理结果。可选的,DU和RU之间可以交互推理结果,例如非实时RIC将推理结果递交至DU,由DU转发给RU。
或者,在第三种设计中,如图4a所示,接入网设备中包括近实时RIC,接入网设备之外包括非实时RIC(可选的,非实时RIC可以位于OAM中或者核心网设备中)。同上述第
二种设计,非实时RIC可以用于进行模型训练和推理。和/或,同上述第一种设计,近实时RIC可以用于进行模型训练和推理。和/或,非实时RIC进行模型训练,近实时RIC可以从非实时RIC获得AI模型信息,并从CU、DU或RU中的至少一项获得网络侧和/或终端侧的信息,利用该信息和该AI模型信息得到推理结果。可选的,近实时RIC可以将推理结果递交至CU、DU、RU或终端设备中的至少一项。可选的,CU和DU之间可以交互推理结果。可选的,DU和RU之间可以交互推理结果,例如近实时RIC将推理结果递交至DU,由DU转发给RU。例如,近实时RIC用于训练模型A,利用模型A进行推理。例如,非实时RIC用于训练模型B,利用模型B进行推理。例如,非实时RIC用于训练模型C,将模型C的信息发送给近实时RIC,近实时RIC利用模型C进行推理。
图4b为本申请的另一种通信***的架构。相对图4a,图4b中将CU分离成为了CU-CP和CU-UP等。
AI模型是AI功能的具体实现,AI模型表征了模型的输入和输出之间的映射关系。AI模型可以是神经网络、线性回归模型、决策树模型、支持向量机(support vector machine,SVM)、贝叶斯网络、Q学习模型或者其他机器学习模型等。本申请中,AI功能可以包括以下至少一项:数据收集(收集训练数据和/或推理数据)、数据预处理、模型训练(或称为,模型学习)、模型信息发布(配置模型信息)、模型校验、模型推理、或推理结果发布。其中,推理又可以称为预测。本申请中,可以将AI模型简称为模型。
如图4c所示为AI模型的一种应用架构示意图。数据源(data source)用于存储训练数据和推理数据。模型训练节点(model trainning host)通过对数据源提供的训练数据(training data)进行分析或训练,得到AI模型,且将AI模型部署在模型推理节点(model inference host)中。可选的,模型训练节点还可以对已部署在模型推理节点的AI模型进行更新。模型推理节点还可以向模型训练节点反馈已部署模型的相关信息,以使得模型训练节点对已部署的AI模型进行优化或更新等。
其中,通过模型训练节点学习得到AI模型,相当于由模型训练节点利用训练数据学习得到模型的输入和输出之间的映射关系。模型推理节点使用AI模型,基于数据源提供的推理数据进行推理,得到推理结果。该方法还可以描述为:模型推理节点将推理数据输入到AI模型,通过AI模型得到输出,该输出即为推理结果。该推理结果可以指示:由执行对象使用(执行)的配置参数、和/或由执行对象执行的操作。推理结果可以由执行(actor)实体统一规划,并发送给一个或多个执行对象(例如,网络实体)去执行。可选的,执行实体或者执行对象可以将其收集到的参数或测量量反馈给数据源,该过程可以称为表现反馈,所反馈的参数可以作为训练数据或推理数据。可选的,执行实体或者执行对象还可以将根据模型推理节点所输出的推理结果,确定模型性能相关的反馈信息,且将该反馈信息反馈给模型推理节点,模型推理节点可根据该反馈信息,向模型训练节点反馈该模型的性能信息等,以使得模型训练节点对已部署的AI模型进行优化或更新等,该过程可称为模型反馈。
AI模型可以是神经网络或其它机器学习模型。以神经网络为例,神经网络是机器学习技术的一种具体实现形式。根据通用近似定理,神经网络在理论上可以逼近任意连续函数,从而使得神经网络具备学习任意映射的能力。因此神经网络可以对复杂的高维度问题进行准确的抽像建模。
神经网络的思想来源于大脑组织的神经元结构。每个神经元都对其输入值做加权求和运算,将加权求和结果通过一个激活函数产生输出。如图5所示,为神经元结构示意图。假设神经元的输入为x=[x0,x1,…,xn],与各输入对应的权值分别为w=[w,w1,…,wn],加权求和的偏置为b。激活函数的形式可以多样化,假设一个神经元的激活函数为:y=f(z)=max(0,z),该神经元的输出为:再例如一个神经元的激活函数为:y=f(z)=z,该神经元的输出为:
xi、wi、和b可以为小数、整数(包括0、正整数或负整数等)、或复数等各种可能的取值。神经网络中不同神经元的激活函数可以相同或不同。
神经网络一般包括多层结构,每层可包括一个或多个神经元。增加神经网络的深度和/或宽度可以提高该神经网络的表达能力,为复杂***提供更强大的信息提取和抽象建模能力。其中,神经网络的深度可以指神经网络包括的层数,每层包括的神经元个数可以称为该层的宽度。如图6所示,为神经网络的层关系示意图。一种实现中,神经网络包括输入层和输出层。神经网络的输入层将接收到的输入经过神经元处理后,将结果传递给输出层,由输出层得到神经网络的输出结果。另一种实现中,神经网络包括输入层、隐藏层和输出层。神经网络的输入层将接收到的输入经过神经元处理后,将结果传递给中间的隐藏层,隐藏层再将计算结果传递给输出层或者相邻的隐藏层,最后由输出层得到神经网络的输出结果。一个神经网络可以包括一层或多层依次连接的隐藏层,不予限制。神经网络的训练过程中,可以定义损失函数。损失函数描述了神经网络的输出值和理想目标值之间的差距或差异,本申请不限制损失函数的具体形式。神经网络的训练过程就是通过调整神经网络参数,如神经网络的层数、宽度、神经元的权值、和/或神经元的激活函数中的参数等,使得损失函数的值小于阈值门限值或者满足目标需求的过程。
在一种方案中,RAN节点支持多种AI辅助的案例。例如,移动性增强、负载均衡和网络节能等。其中,OAM用于训练或存储AI模型,RAN节点向OAM请求AI模型,并利用该AI模型执行推理。OAM如何向RAN节点配置对应的模型,是本申请待解决的问题。在本申请的描述中,OAM可以替换为其他能够训练或存储AI模型的节点,如云服务器、核心网设备或其他可能的设备等。
如图7所示,提供一种模型配置方法的流程,至少包括:
步骤701:第一RAN节点向OAM发送第一模型请求消息;相应的,OAM接收来自第一RAN节点的第一模型请求消息。
为了便于区分,将第一RAN节点向OAM发送的模型请求消息,称为第一模型请求消息。将第二RAN节点向OAM发送的模型请求消息,称为第二模型请求消息。模型请求消息用于RAN节点向OAM请求模型,对模型请求消息的名称不作限定。例如,第一模型请求消息还可称为第一消息,第二模型请求消息还可称为第二消息等。为了区分不同的RAN节点,称为第一RAN节点和第二RAN节点等,第一RAN节点和第二RAN节点为不同的RAN节点,对两者的关系,不作任何限定。
在一种设计中,第一RAN节点可确定业务场景和功能需求,根据业务场景和功能需求,确定第一模型请求消息。其中,所述业务场景和功能需求为RAN节点的业务场景和功能需求。例如,利用所述第一模型请求消息所请求的第一模型,用于RAN节点相关的模型推理,例如用于小区节能等。第一RAN节点可根据RAN节点的业务场景和功能需求,确定第一模型请求消息等。或者,所述业务场景和功能需求,为UE的业务场景和功能需
求。例如,利用所述第一模型请求消息所请求的第一模型,用于UE相关的模型推理,例如用于预测UE的运动轨迹等。第一RAN节点可根据UE的业务场景和功能需求,确定第一模型请求消息等。或者,第一RAN节点可根据除RAN节点和UE外,其它节点的业务场景和功能需求,确定第一模型请求消息等,不作限定。所述第一RAN节点可根据UE上报的信息,获取UE的业务场景和功能需求。例如,UE处于自动驾驶场景,向RAN节点发送服务质量(quality of service,QoS)预测服务请求,RAN节点可以获取UE的业务场景和功能需求为:自动驾驶+QoS预测。或者,第一RAN节点可通过分析,获取UE的业务场景和功能需求。例如,RAN节点发现UE移动到小区边缘附近,需要移动增强服务,移动性增服务需要预测UE的轨迹信息,UE的业务场景和功能需求为:移动性增强+预测UE的轨迹信息。或者,第一RAN节点可以从其它设备中获取UE的业务场景和功能需求,所述其它设备包括但不限于:最小化路测设备(minimization drive test,MDT)等。或者,第一RAN节点中存储有UE的业务场景和功能需求等,例如,UE在接入到第一RAN节点时,主动向第一RAN节点上报自己的业务场景和功能需求等。
在本申请中,所述业务场景包括但不限于以下至少一项:移动性增强、负载均衡、虚拟现实(virtual reality,VR)、增强现实(augmented reality,AR)、车联网(自动驾驶)、无线机器人云端控制、具备力反馈的远程诊断、或高清视频(例如8K视频)等。示例的,所述移动性增强是指对移动UE的服务小区切换提供增强服务。例如,RAN节点根据UE的运动轨迹、测量报告或小区负载情况等中的至少一个,预测UE的目标小区以及切换时间等。所述UE的目标小区可指UE待切换的服务小区。所述UE的移动方式包括固定线路移动和非固定线路移动等。其中,移动性增强可以包括固定线路的车辆移动的移动性增强、非固定路线的车辆移动的移动性增强、或行人场景的移动性增强等。在远程诊断中,或称为远程会诊中,医生通过网络对身处异地的患者进行诊断或手术等。上述具备反馈力是指力的触觉反馈,这是对时延要求特别高的业务,需要提供时延预测服务。例如,预测时延超过某个阈值时,需要对信息传输方案做出调整或给出告警,以提前应对突发事件。
在本申请中,所述功能需求包括但不限于以下至少一项:UE轨迹预测、负载预测、信号干扰噪声比(signal to interference plus noise ratio,SINR)预测、参考信号接收功率(reference signal receiving power,RSRP)预测、参考信号接收质量(reference signal receiving quality,RSRQ)预测、时延预测、QoS预测、吞吐量预测或小区节能等。
在一种设计中,所述第一RAN节点根据业务场景和功能需求,确定第一模型请求消息后,所述第一模型请求消息用于指示所述业务场景和功能需求,或者描述为,所述第一模型请求消息用于指示所请求的第一模型的模型应用场景和模型功能等。所述模型的应用场景与上述业务场景相对应,所述模型功能与上述功能需求相对应。本申请中的所述指示可包括隐示指示或显示指示等,不作限定。例如,第一模型请求消息中可包含所述业务场景和功能需求等。或者,第一RAN节点可直接将所述业务场景和功能需求发送给OAM等。例如,第一RAN节点向OAM发送的业务场景和功能需求的组合可包括:移动性增强场景+轨迹预测、负载均衡+轨迹预测、自动驾驶+QoS预测、VR/AR+QoS预测、无线机器人云端控制+时延预测、具备力反馈的远程诊断+时延预测等。
在另一种设计中,所述第一RAN节点根据业务场景和功能需求,确定第一模型请求消息的过程,包括:所述第一RAN节点根据业务场景和功能需求,确定所请求的第一模型的模型功能和模型性能指标。所述模型功能与上述功能需求相对应。其中,所述模型性
能指标包含以下至少一项:准确率、计算量、内存占用、或推理时长等。所述准确率是指预测正确的样本数量占总样本数量的百分比。计算量是指模型所需的计算次数,模型的整体计算量等于模型中每个算子的计算量之和,它反馈了模型对硬件计算单元的需求;或者计算量可指模型运行时,所占用的计算资源的大小。内存占用是指模型运行时,所占用的内存和显存大小;或者所述内存占用可称为模型的大小等。推理时长是指模型推理出预测结果所需要的时长等。所述第一RAN节点向OAM发送的第一模型请求消息用于指示所述第一模型的模型功能和模型性能指标,例如,第一模型请求消息中包含模型功能和模型性能指标。或者,第一RAN节点可直接向OAM发送所述模型功能和模型性能指标。所述模型功能还可称为模型类别。例如,第一RAN节点通过分析,获知需要为UE提供自动驾驶场景下的QoS预测服务,第一RAN节点可继续分析在自动驾驶场景下,QoS预测服务所需的性能指标,第一RAN节点将QoS预测服务和相应的性能指标发送给OAM。或者,第一RAN节点根据模型性能指标,确定模型性能等级,模型性能等级与模型性能指标间存在对应关系,可参见下文的说明。第一RAN节点向OAM发送的第一模型请求消息用于指示所请求的第一模型的模型功能和模型性能等级。
可选的,在步骤701之前,还可以包括:第一RAN节点确定是否需要提供基于AI的增强服务。如果需要提供基于AI的增强服务,执行步骤701向OAM发送第一模型请求消息;否则不再执行上述图7所示的流程。比如,在基于AI的移动性增强方案中,源RAN节点可以利用AI的方式,预测UE的目标小区。但是当UE不是处于小区边缘时,并不需要源RAN节点预测该UE的目标小区。即只有当UE处于小区边缘时,源RAN节点才需要向OAM发送第一模型请求消息,用于请求用于预测目标小区的模型。或者,当源RAN节点利用AI方式,预测UE的目标小区时,模型的输入中包括UE的轨迹信息。但当UE的轨迹很随机时,利用AI方式预测的目标小区的准确性可能会大大降低。此时,源RAN节点可考虑不再采用AI方式,预测UE的目标小区,此时不再向OAM发送第一模型请求消息,而是采用传统的方式预测UE的目标小区等。
步骤702:OAM根据所述第一模型请求消息和第一映射关系,确定第一模型。
在本申请中,为了区分不同的模型,将不同的模型分别称为第一模型和第二模型等。模型表征了模型的输入和输出之间的映射关系,用于实现具体的功能,对模型的名称不作限定。例如,模型还可称为AI模型,AI模型可以是神经网络、线性回归模型、决策树模型、SVM、贝叶斯网络、Q学习模型或者其他机器学习模型等。上述第一映射关系是表征一种映射或对应关系等。第一映射关系还可称为对应关系等。后续对于类似的情况,不再一一说明。
在本申请中,OAM可以获取第一映射关系,该第一映射关系可以是OAM建立的,或者是其它节点建立的,OAM通过其它节点获取该第一映射关系。
在一种示例中,第一映射关系中包括模型、模型性能等级、和模型功能的映射关系,模型性能等级可简称为性能等级。
例如,模型性能等级与模型性能指标间存在一对多的对应关系。例如,用于模型等级划分的模型性能指标包括准确率,可根据准确率划分等级,准确率越高,模型性能等级越高。或者,用于模型等级划分的模型性能指标包括推理时长,可根据推理时长划分等级,推理时长越短,模型性能等级越高。或者,用于模型等级划分的模型性能指标包括准确率、计算量、推理时长和内存占用中至少两个指标。例如,按照一定的加权方式,对多个指标
进行加权,确定加权值,按照加权值进行等级划分。不同功能的模型,其划分模型性能等级的方式相同或不同。比如,以用于轨迹预测的模型,与用于SINR预测的模型,根据模型准确率进行模型性能等级划分为例。假设对于用于轨迹预测的模型,和SINR预测的模型,均包括3个模型性能等级。如表1所示,对于用于轨迹预测的模型,模型性能等级1对应的准确率为[n1,n2],模型性能等级2对应的准确率为(n2,n3],模型性能等级3对应的准确率为(n3,n4],n4大于或等于n3,n3大于或等于n2,n2大于或等于n1。如表2所示,对于SINR预测的模型,模型性能等级1对应的准确率为[m1,m2],模型性能等级2对应的准确率为(m2,m3],m3大于或等于m2,m2大于或等于m1。再例如,如表3所示,对于SINR预测的模型,模型性能等级1对应的推理时长为[t1,t2],模型性能等级2对应的推理时长为(t2,t3],模型性能等级3对应的推理时长为(t3,t4],t4大于或等于t3,t3大于或等于t2,t2大于或等于t1。
表1:模型性能等级与模型性能指标的对应关系;
表2:模型性能等级与模型性能指标的对应关系;
表3:模型性能等级与模型性能指标的对应关系;
在该设计中,第一RAN节点可获取模型的应用场景,根据该模型的应用场景,确定对应的模型性能指标,根据模型性能指标确定模型性能等级,向OAM发送的第一模型请求消息用于指示模型性能等级和模型功能。OAM根据第一模型请求消息所指示的模型性能等级和模型功能,在第一映射关系中,确定一个模型,称为第一模型。该第一映射关系中包括模型、模型性能等级、和模型功能的映射关系,可参见下述表6的说明。
在一种示例中,第一映射关系包括模型、模型应用场景和模型功能的映射关系。例如,第一映射关系包括:AI模型+自动驾驶场景+QoS预测等。
例如,第一RAN节点可获取第一模型的模型应用场景和模型功能,可参见步骤701中的说明。第一RAN节点向OAM发送的第一模型请求消息用于指示第一模型的模型应用场景和模型功能。OAM可以根据第一模型请求消息所指示的应用场景和模型功能,在第一映射关系中,确定一个AI模型,该AI模型,称为第一模型。示例的,如表4所示,第一映射关系中包括4个应用场景(索引分别为1至4),和3种模型功能(索引分别为1至3)。该4种应用场景和3种模型功能,可组成12种映射关系,该12种映射关系的索引分别为1至12。例如,映射关系1包括应用场景1+模型功能1+模型1,映射关系2包括应用场景1+模型功能2+模型2等。
表4:第一映射关系;
在表4所示的第一映射关系中,对于同一个模型功能,在不同场景中,对应的模型不同。例如,对于模型功能1,在应用场景1至4中,分别对应模型1、模型4、模型7和模型10。以OAM建立上述表4所示的第一映射关系为例,针对同一个模型功能,OAM可以在不同场景中,分别采集对应的训练数据,训练对应的AI模型。例如,OAM可以在应用场景1中采集模型功能1对应的训练数据,利用训练数据对模型1进行训练,从而使得模型1可满足应用场景1的性能指标。对于训练过程可采用机器学习、非机器学习或强化学习等,不作限定。
或者,在该示例中,模型性能等级与模型性能指标,存在一一对应的关系,每个模型性能等级对应唯一的模型性能指标,每个模型性能指标对应唯一的模型性能等级。例如,如表5所示,模型性能等级1至3,分别对应于模型性能指标1至3。以模型性能指标包括:准确率、计算量、推理时长和内存占用为例,模型性能等级1对应于模型性能指标1,对应的准确率在99%以上,计算量在3%以下,推理时长在1s以内,内存占用在500KB以内等。
表5:模型性能等级与模型性能指标的对应关系;
在该示例中,第一RAN节点可获取所请求模型的模型性能指标,向OAM发送的第一模型请求消息用于指示所请求模型的模型功能和模型性能指标。OAM在接收到第一模型请求消息时,可根据第一模型请求消息所指示的模型性能指标,确定模型性能等级。或者,第一RAN节点可根据模型性能指标,确定模型性能等级,向OAM发送的第一模型请求消息用于指示模型功能和模型性能等级。OAM根据模型功能和模型性能等级,在第一映射关系中,确定一个模型,该模型称为第一模型等。例如,如表6所示,第一映射关系中包括12个映射关系,索引分别为1至12。例如,映射关系1包括:模型1+模型性能等级1+模型功能1,映射关系2包括:模型2+模型性能等级1+模型功能2。
表6:第一映射关系;
应当指出,在上述模型性能等级与模型性能指标存在一一对应关系的示例中,上述第一映射关系中的模型性能等级可替换为模型性能指标。或者,在本申请中,可以根据模型性能指标建立第一映射关系,例如,第一映射关系中包括模型、模型性能指标、和模型功能的映射关系。在该设计中,第一模型请求消息用于指示第一模型的模型功能和模型性能指标。OAM根据第一模型请求消息所指示的模型功能和模型性能指标,在第一映射关系中,确定一个模型,该模型可称为第一模型。例如,如表7所示,第一映射关系中包括12个映射关系,索引分别为1至12。例如,映射关系1包括:模型1+模型性能指标1+模型功能1,映射关系2包括:模型2+模型性能指标1+模型功能2。
表7:第一映射关系;
在上述表6或表7所示的第一映射关系中,对于同一个模型功能,在不同的模型性能等级或模型性能指标下,其对应的模型不同。例如,参见表6或表7,对于模型功能1,
在模型性能等级1至4,或者模型性能指标1至4,其分别对应的模型分别为:模型1、模型4、模型7和模型10。以OAM建立表6或表7所示的第一映射关系为例,OAM可以在不同的模型性能等级或模型性能指标下分别采集对应的训练数据,训练对应的模型。例如,OAM可以在模型性能等级1或模型性能指标1下,采集模型功能1对应的训练数据,对模型1进行训练等。
如前所述,所述模型性能指标包含以下至少一项:准确率、计算量、内存占用、或推理时长等。在不同场景下,同一个模型功能对模型性能指标的要求可能不同。可选的,模型性能指标中的计算量还称为计算资源占用,内存占用还称为模型大小。一种可能的情况,如表8所示,在VR/AR场景、和V2X场景下,其对QoS预测的模型的性能指标要求不同。还有一种可能的情况,如表9所示,在固定路线车辆、非固定路线车辆和行人场景等场景下,其对轨迹预测的模型的性能指标要求不同。
表8:不同场景下的QoS预测功能;
表9:不同场景下的轨迹预测功能;
由于在不同场景中,对于同一个功能的模型的性能指标要求可能不同。如果无论在何种场景下,针对某个功能,都利用同一个模型进行推理,很难满足对应场景的性能指标要求。例如,在表8中,对于AR/VR、和V2X场景,如果都利用同一个QoS预测的模型进行模型推理,很难在每个场景中都满足其对性能指标的要求。在本申请的方案中,可参见上述表4所示的第一映射关系,针对同一个模型功能,在不同的场景中,可以配置不同的模型。例如,在VR/AR场景中,针对QoS预测配置一个模型;在V2X场景中,针对QoS预测再配置一个模型等。或者,可参见上述表6所示的映射关系,针对不同场景对性能指标的要求,为其配置对应性能等级的模型,或者,可参见上述表7所示的映射关系,针对不同的性能指标配置不同的模型,从而使得每个模型满足对应应用场景的性能指标要求。
步骤703:OAM向第一RAN节点发送所述第一模型的信息。
在本申请中,所述第一模型的信息用于指示所述第一模型的以下至少一项:模型索引(index)、模型的结构信息、模型的参数、模型输入格式、模型输出格式、输入数据处理方式、输出数据处理方式、模型性能指标、模型应用场景、模型功能、或训练参数等。
其中,模型索引用于区分不同的模型。模型的结构信息、模型参数、模型输入格式和模型输出格式等,用于第一RAN节点确定第一模型。或者描述为:第一RAN节点根据模型的结构信息、模型参数、模型输入格式和模型输出格式等,确定第一模型。所述模型的结构信息包括以下至少一项:该模型包括层的数量(还称为模型的深度)、每层的网络类型(例如,全连接层网络、卷积层网络、长短期记忆((long short-term memory,LSTM)层
等)、每层包括的神经元数量(还称为模型的宽度)、或层间的连接关系等。所述模型参数可包括以下至少一项:神经元的权重值、神经元的激活函数、或激活函数中的偏置等。上述模型的输入格式和输出格式可指对该模型进行模型训练或推理时,输入数据和输出数据应该满足的格式等。上述输入数据处理方式包括将采集的原始数据输入到模型之前,对原始数据进行预处理的方式,比如,将原始数据的虚部和实部分离、归一化处理或相位和幅度分离等。上述输出数据处理方式包括在模型输出数据时,对该输出数据的处理方式。例如,对输出数据进行截取等。关于模型性能指标、模型应用场景和模型功能等可参见前述说明。在本申请中,第一RAN节点在根据模型结构、模型的参数、模型输入格式和模型输出格式等信息,确定第一模型时,还可根据上述训练参数,对第一模型继续训练。该训练参数中可包括损失函数的指示信息等。例如,第一RAN节点可获取训练数据,该训练数据可来自第一RAN节点本身,或来自UE,或同时来自第一RAN节点和UE等,或来自其它节点等,第一RAN节点根据训练数据,对第一模型进行训练。例如,将训练数据输入到所述第一模型中。根据该第一模型的输出,确定损失函数值的大小。若该损失函数值的大小小于阈值或满足目标需求,结束对第一模型的训练;否则,更新第一模型的参数,继续对第一模型训练。所述更新第一模型的参数,可包括更新第一模型的以下至少一项信息:神经网络的层数、神经网络的宽度、层间的连接关系、神经元的权重、神经元的激活函数、或激活函数中的偏置等。所述训练的过程可采用标签学习、非标签学习或强化学习等,不作限定。其中,训练第一模型时,第一模型的初始模型可以是上述第一模型的信息指示的,或者是随机生成的,或者是协议约定的,不予限制。
或者,所述第一模型的信息用于指示第一模型的以下至少一项:模型、模型使用说明或模型索引。所述模型中包含网络架构、参数权重、训练超参数、优化器(optimizer)、批量大小(batch size)、学习率(learning rate)、冲量(momentum)等。可选的,第一RAN节点可根据训练超参数进行在线训练等。其中,模型使用说明包括模型输入输出格式、模型性能指标、模型应用场景以及模型功能中的至少一种;模型的索引是指模型的编号,用于区分不同模型等。
由于在实际应用中,在不同应用场景下,对实现同一功能的模型的性能指标要求不同。例如,如前述表8或表9所示,对于QoS预测和轨迹预测等,在不同场景下对模型的性能指标要求并不相同。在本申请中,OAM建立第一映射关系,该第一映射关系为模型+模型应用场景+模型功能的映射关系。通过该映射关系,针对同一个模型功能,在不同的应用场景下,可配置该场景对应的模型,从而使得该模型满足该模型对应场景中对性能指标的要求。相较于,无论何种场景,都采用同一模型进行模型推理。采用本申请的方案,可提高预测性能和预测速度等。或者,在本申请中,OAM建立第一映射关系,该第一映射关系为模型+模型性能等级+模型功能的映射关系,或者第一映射关系为模型+模型性能指标+模型功能的映射关系。由于不同应用场景下对模型的性能指标要求不同,模型性能指标与模型性能等级间存在对应关系,OAM可根据第一映射关系,为不同应用场景下的不同模型性能指标要求,配置不同的模型。相对于,无论何种场景的模型性能指标下,采用同一模型进行模型推理,可提高预测性能和预测速度等。
可选的,在上述图7所示的流程中,除包括上述步骤701至步骤703外,还可以包括:
步骤704:第一RAN节点根据所述第一模型,执行模型推理,获得预测信息。
示例的,第一RAN节点可收集输入数据,分配计算资源和内存资源等,用于模型推
理,该输入数据可来自第一RAN节点和/或UE的,或来自其它节点的等,不作限定。可选的,第一RAN节点可根据模型使用说明,收集输入数据。将该输入数据输入到第一模型,输出为预测信息。或者,对输入数据作处理,处理后的输入数据输入到第一模型,和/或,对第一模型的输出数据作处理,处理后的输出数据为预测信息等。可以理解的是,输入数据和输出数据满足第一模型的输入格式和输出格式的要求。所述预测信息中包含预测结果。示例的,模型训练节点,例如OAM在训练模型时,把收集的数据集合划分为训练集、验证集和测试集。其中,模型训练节点利用训练集进行模型训练,得到模型,该过程可称为模型训练;利用验证集对模型进行验证,该过程可称为模型验证。模型验证,通常发生在模型训练过程中。例如,模型每经过一次或多次迭代(epoch)的训练,可以用验证集对当前模型进行验证,来监测模型训练的状态,例如验证是否欠拟合、过拟合或已经收敛等,决定是否结束训练。可选的,在模型验证过程中,还可以调整模型的超参数,所述超参数可以指模型的以下至少一项参数:神经网络的层数、神经元的个数、激活函数、或损失函数等。利用测试集对模型测试,该过程可称为模型测试。例如,在模型训练完成后,可利用测试集对训练好的模型进行测试。例如,评估模型的泛化能力,判断模型是否满足要求,决定模型是否可用,或获取该模型的预测准确率等。在本申请中,OAM向第一RAN节点所发送的第一模型的信息中还可以包括预测准确率等。或者,第一RAN节点在确定第一模型时,可利用本地收集的测试数据,对第一模型进行测试,获得预测准确率等。第一RAN节点向UE发送的预测信息中除包括第一模型推理的预测结果外,可选的,还可以包括预测准确率。比如,在轨迹预测中,第一RAN节点向UE发送的所述预测信息中包括:预测的UE的轨迹信息+预测的轨迹信息的准确率(如98%等)。
在本申请中,若该预测信息是第一RAN节点相关的预测信息,第一RAN节点根据预测信息,确定第一反馈信息。例如,第一RAN根据预测信息,执行相应的动作。例如,在网络节能场景中,预测信息为节能策略,如载波关断、符号关断、小区关断、或深度休眠等,第一RAN节点在利用第一模型推理出所述节能策略后,可以执行所述节能策略。第一RAN节点根据执行所述预测信息的前后性能变化(例如小区吞吐量、或UE的QoS变化等),确定第一反馈信息。或者,所述预测信息为UE相关的预测信息,第一RAN节点需要将该预测信息发送给UE。UE根据该预测信息,确定第二反馈信息。例如,UE可根据执行所述预测信息后的性能变化,确定第二反馈信息。比如,在移动性增强场景中,所述预测信息为UE切换的目标小区和切换时间等。UE可以根据该预测信息执行相应动作,根据执行该预测信息后的性能变化,确定第二反馈信息。或者,UE对比该预测信息与实际的信息,确定第二反馈信息等。例如,在轨迹预测场景中,UE可将预测的UE的运行轨迹与UE实际的运行轨迹作对比,确定第二反馈信息等。UE向第一RAN节点发送第二反馈信息。第一RAN节点根据接收的第二反馈信息,向OAM发送第一反馈信息,OAM根据第一反馈信息,对第一模型进行更新等,具体的可参见下述步骤705至步骤706中第一种设计的说明。或者,当第一RAN节点确定应用场景变化、所需模型的性能等级(或性能指标)变化时,也可以向OAM发送用于指示场景变更或性能等级变更的第一反馈信息,OAM可以根据该场景变更指示或性能等级变更指示,重新为第一RAN节点配置模型,该模型可称为第二模型等。例如,当UE从步行场景变更为固定线路车辆行驶场景时,第一RAN节点利用步行场景所请求的轨迹预测模型,对固定线路车辆行驶场景中的运动轨迹进行预测,可能会导致预测的准确率降低。在本申请中,当第一RAN节点发现UE的场景变
更时,可将该场景变更指示给OAM等,OAM可为第一RAN节点配置与当前场景相匹配的模型等。关于上述场景变更指示或性能等级变更指示,可以是第一RAN节点通过对UE进行分析发现的,或者UE向第一RAN节点上报的,具体的可参见下述步骤706中关于第二种设计的介绍。或者该第一模型是第一RAN节点相关的模型,第一RAN通过对自己分析,可以确定应用场景变更或性能等级变更等。
步骤705:第一RAN节点向UE发送所述预测信息;相应的,UE接收来自第一RAN节点的预测信息。
步骤706:UE向第一RAN节点发送第二反馈信息。相应的,第一RAN节点接收来自UE的第二反馈信息。
在第一种设计中,UE在接收到预测信息时,可将第一RAN节点发送的预测信息与实际的信息作对比,确定预测信息的准确率,该准确率称为预测信息的实际准确率。例如,预测信息为轨迹信息,UE可将预测的轨迹信息与UE实际的轨迹信息作对比,确定该预测信息的实际准确率,且向第一RAN节点发送第二反馈信息,该第二反馈信息用于指示该预测信息的实际准确率。该指示可为隐示指示或显示指示等。或者描述为:UE向第一RAN节点发送预测信息的实际准确率。示例的,UE可以向第一RAN节点反馈在一段时间内的预测信息的实际准确率,可以是周期性反馈、或基于事件触发反馈等。例如,UE可以以10秒为周期,每10秒计算一次预测信息的实际准确率,且反馈给第一RAN节点。或者,UE同样每10秒计算一次预测信息的准确率,当准确率低于阈值(例如80%时),UE才向第一RAN节点发送预测信息的实际准确率等。
在第二种设计中,当UE发现自己的应用场景变更,或者性能等级变更时,也可以向第一RAN节点发送第二反馈信息,该第二反馈信息用于指示应用场景变更,或性能等级变更。或者描述为,UE向第一RAN节点发送应用场景变更指示,或者性能等级变更指示等。
步骤707:第一RAN节点向OAM发送第一反馈信息。相应的,OAM接收来自第一RAN节点的第一反馈信息。
在一种设计中,第一RAN节点可以周期性,或基于事件触发等方式,向OAM发送第一反馈信息,该第一反馈信息可隐示或显示指示预测信息的实际准确率。或者描述为,第一RAN节点向OAM发送预测信息的实际准确率。与上述步骤706相似,第一RAN节点可向OAM反馈一段时间的预测信息的实际准确率。例如,第一RAN节点可周期性,或基于事件触发等方式,向OAM反馈一段时间的预测信息的实际准确率等。
在另一种设计中,当第一模型相关的应用场景或性能等级发生变更时,第一RAN节点向OAM发送第一反馈信息,该第一反馈信息用于指示应用场景变更,或性能等级变更。或者描述为:第一RAN节点向OAM发送应用场景变更指示,或性能等级变更指示等。
步骤708:OAM根据第一反馈信息,对所述第一模型进行更新。
在一种设计中,当所述第一反馈信息用于指示所述第一模型所输出的预测信息的实际准确率时,所述OAM根据所述第一反馈信息,对所述第一模型更新,包括:当所述第一模型所输出的预测信息的实际准确率小于阈值时,重新对所述第一模型训练,并更新所述第一模型。例如,OAM可以收集训练数据,该训练数据来自第一RAN节点和/或UE,或者其它节点等。OAM可利用所收集的训练数据,重新对第一模型训练。关于模型训练的过程,可参见前述说明。
在另一种设计中,所述第一反馈信息用于指示场景变更指示,或模型性能等级变更指示,或模型性能指标变更指示,所述OAM根据所述第一反馈信息,对所述第一模型更新,包括:根据所述场景变更指示,或模型性能等级变更指示,或模型性能指标变更指示,确定第二模型;例如,第一映射关系包括模型、模型应用场景和模型功能的映射关系,OAM可根据所指示的场景变更指示,确定变更后的场景。根据上述第一映射关系,确定变更后的场景对应的模型,称为第二模型。或者,第一映射关系包括模型、模型性能等级和模型功能的映射关系,OAM可根据所指示的模型性能等级变更指示,确定变更后的模型性能等级。或者,OAM可根据所指示的模型性能指标变更指示,确定变更后的模型性能等级。根据上述映射关系,确定变更后的模型性能等级对应的模型,称为第二模型。或者,第一映射关系包括模型、模型性能指标和模型功能的映射关系,OAM根据第一映射关系,确定所指示的变更后的模型性能指标对应的模型,称为第二模型等。
步骤709:OAM向第一RAN节点发送更新后模型的信息;相应的,第一RAN节点接收来自OAM的更新后模型的信息。
关于更新后模型的信息所包括的内容,可参见前述步骤703中第一模型的信息中包括的内容,此处不再赘述。
在本申请中,第一RAN节点可周期性或基于事件触发等方式,向OAM发送第一反馈信息。OAM根据该第一反馈信息,评估是否需要对第一模型进行更新,从而及时对第一模型进行更新,提升第一模型的预测性能和预测速度。
当采用上述图7流程中的方法,为第一RAN节点配置第一模型后。如图8所示,本申请提供一种小区切换流程图,在该流程中第一RAN节点称为源RAN节点,第二RAN节点称为目标RAN节点,至少包括:
步骤801:源RAN节点确定目标小区。
在一种设计中,源RAN节点可以接收UE上报的测量报告,该测量报告中包括小区的标识信息和小区质量信息等。所述小区可以为当前服务小区和/或邻区等,不作限定。在一种设计中,UE可以周期性地向源RAN节点上报测量报告。例如,UE可以周期性测量服务小区的信号质量和/或邻区的信号质量,然后周期性的向源RAN节点上报所测量的信息。或者,UE可以周期性测量服务小区的信号质量和/或邻区的信号质量,在满足一定的条件时,再向源RAN节点上报所测量的信息等。
在本申请中,所述测量报告中包括小区的标识信息。所述小区的标识信息包括小区的小区全局标识(cell global identifier,CGI)、物理小区标识(physical cell identifier,PCI)和频点、小区标识(cell identifier,cell ID)、非公网标识(non-public network identifier,NPN ID)、非陆地网络标识(non-terrestrial network identifier,NTN ID)或者其它小区标识等中的至少一种。其中,CGI可以包括公共陆地移动网络(public land mobile network,PLMN ID)和cell ID等。可选的,小区的标识信息中还可以包括跟踪区码(tracking area code,TAC)和/或小区所属的网络设备的标识信息,比如全局网络设备标识等。
在本申请中,所述测量报告中包括小区的信号质量信息。例如,UE可以通过测量下行同步信道、信道状态信息参考信号、解调参考信号(demodulation reference signal,DMRS)、小区参考信号(cell-specific reference signal,CRS)信号、同步信号块(synchronization signal block,SSB)、同步信号/物理广播信道块或者其它下行信号中的至少一种,获得小区的质
量信息。示例的,所述小区的质量信息可以包括接收信号码功率(received signal code power,RSCP)、参考信号接收功率(reference signal receiving power,RSRP)、参考信号接收质量(reference signal receiving quality,RSRQ)、信噪比(signal noise ratio,SNR)、信号与干扰加噪声比(signal to interference plus noise ratio,SINR)、参考信号强度指示(reference signal strength indication,RSSI)或其它信号质量中的至少一种。
可选的,小区质量信息可以是小区级的,波束级的,同步信号/物理广播信道块(synchronization signal/physical broadcast channel block,SS/PBCH Block)级的,信道状态信息参考信号(channel-state information reference signal,CSI-RS)级的,空口技术(numerology)级的,切片(slicing)级的,或带宽部分(bandwidth part,BWP)级的至少一种。关于小区质量信息是何种级的,是指该小区的质量信息是以何种粒度测量的。比如,小区的质量信息是小区级的,是指UE对待测量的多个小区中的每个小区进行测量,得到各小区的质量。或者,小区的质量是波束级的,是指该小区包括至少一个波束,UE通过对该小区中的波束进行测量,获得的小区质量信息。例如,一小区包括3个波束,UE可对通过对上述3个波束中的波束1,波束2和波束3分别进行测量,得到满足波束质量条件的波束1的质量信息和波束2的质量信息。可选的,之后,UE可基于波束1的质量信息和波束2的质量信息,例如取较大值、求平均或者加权求和等,得到该小区的质量信息。或者,UE可以同时上报波束1的质量信息和波束2的质量信息。关于UE基于其它粒度测量小区的质量信息,与上述相似,不再说明。
需要说明的是,若小区质量信息为波束级的,SS/PBCH Block级的,CSI-RS级的、空口技术级的,切片级的,或BWP级的等,小区的标识信息中还包括对应的波束的标识信息,SS/PBCH Block的标识信息,CSI-RS的标识信息,空口技术的标识信息,切片的标识信息,或BWP的标识信息等中的至少一项。
在一种设计中,源RAN节点在接收到所述测量报告时,可基于所述测量报告判断是否需要为UE切换服务小区。比如,若源小区的服务质量低于一定阈值,可确定需要为UE切换服务小区。源RAN节点可根据所接收的测量报告,为UE选择目标小区。例如,可以在UE的相邻小区中,选择信号质量大于阈值且信号质量最好的小区,作为目标小区等。
在另一种设计中,源RAN节点可基于AI的方式,确定目标小区。源RAN节点可利用AI模型或机器学习(machine learning,ML)模型,预测目标小区。AI模型或ML模型的输入数据中包括以下至少一项:所述UE的测量报告、相邻小区的历史和/或实时负载、UE的历史轨迹信息、UE当前时刻的地理坐标、UE当前时刻的运行方向、或UE的运动速度等。所述AI模型或ML模型的输出数据包括以下至少一项:UE的目标小区、切换时间戳等。
步骤802:源RAN节点向所述目标小区对应的目标RAN节点发送模型信息。
在一种设计中,源RAN节点可向目标RAN节点发送模型信息,该模型信息中包括第一模型的索引。可选的,该模型信息中还可以包括以下至少一项:第一模型的应用场景、第一模型的功能、或该第一模型需要收集哪些数据进行模型推理等。例如,在基于AI的移动性增强方案中,执行模型推理的RAN节点,需要向UE之前接入过的RAN节点请求UE的历史切换成功或失败情况等。在本申请中,源RAN节点需要通知目标RAN节点需要收集哪些数据进行模型推理的过程,还可以描述为:RAN节点向目标RAN节点发送第一指示信息,该第一指示信息用于指示所述第一模型的输入数据的类型信息。在该过程中,
源RAN节点可提前通知目标RAN节点,在模型推理过程中,需要收集哪些数据进行模型推理,以便于目标源RAN为模型推理做好准备,提高模型推理的效率。
步骤803:源RAN节点向目标RAN节点发送UE的历史信息,该历史信息至少包括源RAN节点从UE收集到的历史数据和/或历史信息等。
在一种设计中,上述UE的历史信息可能是第一模型进行模型推理所需的信息等,也就是说目标RAN节点在利用第一模型,进行模型推理时,上述历史信息可作为第一模型的输入数据的一部分。上述过程,还可以描述为:源RAN节点向目标RAN节点发送所述模型推理所需的历史数据信息,所述历史数据信息用于确定所述第一模型的输入数据。例如,目标RAN节点可将上述历史数据直接作为第一模型的输入,进行模型推理;或者,目标RAN节点可对上述历史数据进行处理,将处理后的历史数据作为第一模型的输入等,不予限制。
需要指出的是,步骤803可以在步骤802之后执行,或者可以在步骤804之后执行,不作限定。在图8的流程中,是以在步骤802之后执行为例描述的。比如,若采用AI的方式,确定目标小区,步骤803可以在步骤802之后执行。或者,若采用非AI的方式,确定目标小区,步骤803可以在步骤804之后执行。在本申请中,对所有流程中步骤的前后顺序并不作限定。
步骤804:UE由源RAN节点对应的源小区,切换到目标RAN节点对应的目标小区。切换后,目标小区作为UE的服务小区。
步骤805:目标RAN节点向源RAN节点发送第二指示信息,该第二指示信息用于指示以下至少一项:UE切换到所述目标小区、源RAN节点删除所述第一模型,或源RAN节点释放相应的计算资源。
例如,源RAN节点为UE分配了1%的资源用于模型推理,当UE切换到目标小区后,源RAN节点可以释放为UE分配的1%的计算资源,删除第一模型等。
步骤806:源RAN节点删除所述第一模型,释放相应的计算资源。
步骤807:目标RAN节点向OAM发送第二模型请求消息,所述第二模型请求消息用于指示所述第一模型的索引。
步骤808:OAM可根据所述第一模型的索引,确定所述第一模型。
在本申请中,第一模型的索引与第一模型间存在映射关系。OAM可根据该第一模型的索引,确定第一模型。
步骤809:OAM向目标RAN节点发送所述第一模型的信息。
步骤8010:目标RAN节点根据第一模型,执行模型推理,获得预测信息。
步骤8011:目标RAN节点向UE发送预测信息。
步骤8012:UE向目标RAN节点发送第二反馈信息。
步骤8013:目标RAN节点向OAM发送第一反馈信息。
步骤8014:OAM执行第一模型更新。
步骤8015:OAM向目标RAN节点发送更新后第一模型的信息。
关于步骤809至8015的具体执行过程,可参见上述图7中的说明。
在本申请中,当UE执行小区切换时,源RAN节点通过将第一模型的索引发送给目标RAN节点,保证了AI服务的连续性;且UE切换成功后,目标RAN节点再通知源RAN节点删除模型相关信息,释放计算资源,可以节省RAN节点的内存资源占用和计算资源
占用。
可以理解的是,为了实现上述方法中的功能,OAM和RAN节点包括了执行各个功能相应的硬件结构和/或软件模块。本领域技术人员应该很容易意识到,结合本申请描述的各示例的单元及方法步骤,本申请能够以硬件或硬件和计算机软件相结合的形式来实现。某个功能究竟以硬件还是计算机软件驱动硬件的方式来执行,取决于技术方案的特定应用场景和设计约束条件。
图9和图10为本申请提供的可能的装置的结构示意图。这些通信装置可以用于实现上述方法中OAM或RAN节点的功能,因此也能实现上述方法所具备的有益效果。
如图9所示,通信装置900包括处理单元910和收发单元920。通信装置900用于实现上述图7或图8中所示的方法中OAM、源RAN节点或目标RAN节点的功能。
当通信装置900用于实现图7或图8所示的方法中OAM的功能时:收发单元920用于接收来自第一接入网设备的第一模型请求消息;处理单元910用于根据所述第一模型请求消息和第一映射关系,确定第一模型,所述第一映射关系包括模型、模型应用场景、和模型功能的映射关系;或者所述第一映射关系包括模型、模型性能等级、和模型功能的映射关系;收发单元920还用于向所述第一接入网设备发送所述第一模型的信息。
当通信装置900用于实现图7或图8所示的方法中第一RAN节点或源RAN节点的功能时:收发单元920用于向操作维护管理OAM发送第一模型请求消息,所述第一模型请求消息用于指示所述第一模型的模型应用场景和模型功能;或者,所述第一模型请求消息用于指示所述第一模型的模型功能和模型性能等级;收发单元,还用于接收来自所述OAM的第一模型的信息。
当通信装置900用于实现图8所示的方法中目标RAN节点的功能时:收发单元920,用于接收来自第一接入网设备的第一模型的索引、向操作维护管理OAM发送第二模型请求消息,所述第二模型请求消息用于指示所述第一模型的索引,以及接收来自所述OAM的所述第一模型的信息。
有关上述处理单元910和收发单元920更详细的描述可以直接参考图5、图7或图8所示的方法中相关描述直接得到,这里不加赘述。
如图10所示,通信装置1000包括处理器1010和接口电路1020。处理器1010和接口电路1020之间相互耦合。可以理解的是,接口电路1020可以为收发器、输入输出接口、或管脚等。可选的,通信装置1000还可以包括存储器1030,用于存储处理器1010执行的指令或存储处理器1010运行指令所需要的输入数据或存储处理器1010运行指令后产生的数据。
当通信装置1000用于实现上述方法时,处理器1010用于实现上述处理单元910的功能,接口电路1020用于实现上述收发单元920的功能。
当上述通信装置为应用于OAM的模块时,该OAM模块实现上述方法中OAM的功能。该OAM模块从OAM中的其它模块(如射频模块或天线)接收信息,该信息是RAN节点发送给OAM的;或者,该OAM模块向OAM中的其它模块(如射频模块或天线)发送信息,该信息是OAM发送RAN节点的。这里的OAM模块可以是OAM的基带芯片,也可以是其他模块等。
当上述通信装置为应用于RAN节点的模块时,该RAN节点模块实现上述方法中RAN
节点的功能。该RAN节点模块从RAN节点中的其它模块(如射频模块或天线)接收信息,该信息是终端发送给RAN节点的;或者,该RAN节点模块向RAN节点中的其它模块(如射频模块或天线)发送信息,该信息是RAN节点发送给终端的。这里的RAN节点模块可以是RAN节点的基带芯片,也可以是DU或其他模块,这里的DU可以是开放式无线接入网(open radio access network,O-RAN)架构下的DU。
可以理解的是,本申请中的处理器可以是中央处理单元(central processing unit,CPU),还可以是其它通用处理器、数字信号处理器(digital signal processor,DSP)、专用集成电路(application specific integrated circuit,ASIC)、现场可编程门阵列(field programmable gate array,FPGA)或者其它可编程逻辑器件、晶体管逻辑器件,硬件部件或者其任意组合。通用处理器可以是微处理器,也可以是任何常规的处理器。
本申请中的存储器可以是随机存取存储器、闪存、只读存储器、可编程只读存储器、可擦除可编程只读存储器、电可擦除可编程只读存储器、寄存器、硬盘、移动硬盘、CD-ROM或者本领域熟知的任何其它形式的存储介质。
一种示例性的存储介质耦合至处理器,从而使处理器能够从该存储介质读取信息,且可向该存储介质写入信息。存储介质也可以是处理器的组成部分。处理器和存储介质可以位于ASIC中。另外,该ASIC可以位于基站或终端中。当然,处理器和存储介质也可以作为分立组件存在于基站或终端中。
本申请中的方法可以全部或部分地通过软件、硬件、固件或者其任意组合来实现。当使用软件实现时,可以全部或部分地以计算机程序产品的形式实现。所述计算机程序产品包括一个或多个计算机程序或指令。在计算机上加载和执行所述计算机程序或指令时,全部或部分地执行本申请所述的流程或功能。所述计算机可以是通用计算机、专用计算机、计算机网络、网络设备、用户设备、核心网设备、OAM或者其它可编程装置。所述计算机程序或指令可以存储在计算机可读存储介质中,或者从一个计算机可读存储介质向另一个计算机可读存储介质传输,例如,所述计算机程序或指令可以从一个网站站点、计算机、服务器或数据中心通过有线或无线方式向另一个网站站点、计算机、服务器或数据中心进行传输。所述计算机可读存储介质可以是计算机能够存取的任何可用介质或者是集成一个或多个可用介质的服务器、数据中心等数据存储设备。所述可用介质可以是磁性介质,例如,软盘、硬盘、磁带;也可以是光介质,例如,数字视频光盘;还可以是半导体介质,例如,固态硬盘。该计算机可读存储介质可以是易失性或非易失性存储介质,或可包括易失性和非易失性两种类型的存储介质。
在本申请中,如果没有特殊说明以及逻辑冲突,不同的实施例之间的术语和/或描述具有一致性、且可以相互引用,不同的实施例中的技术特征根据其内在的逻辑关系可以组合形成新的实施例。
本申请中,“至少一个”是指一个或者多个,“多个”是指两个或两个以上。“和/或”,描述关联对象的关联关系,表示可以存在三种关系,例如,A和/或B,可以表示:单独存在A,同时存在A和B,单独存在B的情况,其中A,B可以是单数或者复数。在本申请的文字描述中,字符“/”,一般表示前后关联对象是一种“或”的关系;在本申请的公式中,字符“/”,表示前后关联对象是一种“相除”的关系。“包括A,B或C中的至少一个”可以表示:包括A;包括B;包括C;包括A和B;包括A和C;包括B和C;包括A、B和C。
可以理解的是,在本申请中涉及的各种数字编号仅为描述方便进行的区分,并不用来限制本申请的范围。上述各过程的序号的大小并不意味着执行顺序的先后,各过程的执行顺序应以其功能和内在逻辑确定。
Claims (34)
- 一种模型配置方法,其特征在于,包括:接收来自第一接入网设备的第一模型请求消息;根据所述第一模型请求消息和第一映射关系,确定第一模型,所述第一映射关系包括模型、模型应用场景、和模型功能的映射关系;或者所述第一映射关系包括模型、模型性能等级、和模型功能的映射关系;向所述第一接入网设备发送所述第一模型的信息。
- 如权利要求1所述的方法,其特征在于,所述第一模型请求消息用于指示所述第一模型的模型应用场景和模型功能。
- 如权利要求1所述的方法,其特征在于,所述第一模型请求消息用于指示所述第一模型的模型功能和模型性能等级,或者用于指示所述第一模型的模型功能和模型性能指标。
- 如权利要求1至3中任一项所述的方法,其特征在于,所述第一模型的信息用于指示所述第一模型的以下至少一项:模型索引、模型的结构信息、模型的参数、模型输入格式、模型输出格式、模型性能指标、模型应用场景、模型功能、或训练参数。
- 如权利要求1至4中任一项所述的方法,其特征在于,还包括:接收来自所述第一接入网设备的第一反馈信息;根据所述第一反馈信息,对所述第一模型更新。
- 如权利要求5所述的方法,其特征在于,所述第一反馈信息用于指示所述第一模型所输出的预测信息的实际准确率,所述根据所述反馈信息,对所述第一模型更新,包括:当所述第一模型所输出的预测信息的实际准确率小于阈值时,重新对所述第一模型训练,并更新所述第一模型。
- 如权利要求5所述的方法,其特征在于,所述第一反馈信息用于指示场景变更指示、模型性能等级变更指示、或模型性能指标变更指示,所述根据所述第一反馈信息,对所述第一模型更新,包括:根据所述场景变更指示、模型性能等级变更指示、或模型性能指标变更指示,为所述终端设备选择第二模型;向所述第一接入网设备发送所述第二模型的信息,所述第二模型用于更新所述第一模型。
- 如权利要求1至7中任一项所述的方法,其特征在于,还包括:接收来自第二接入网设备的第二模型请求消息,所述第二模型请求消息用于指示所述第一模型的索引;根据所述第一模型的索引,确定所述第一模型;向所述第二接入网设备发送所述第一模型的信息。
- 一种模型配置方法,其特征在于,包括:向操作维护管理OAM发送第一模型请求消息,所述第一模型请求消息用于指示所述第一模型的模型应用场景和模型功能;或者,所述第一模型请求消息用于指示所述第一模型的模型功能和模型性能等级,或者用于指示所述第一模型的模型功能和模型性能指标;接收来自所述OAM的第一模型的信息。
- 如权利要求9所述的方法,其特征在于,在向OAM发送第一模型请求消息之前,还包括:确定是否需要提供基于人工智能AI的增强服务。
- 如权利要求9或10所述的方法,其特征在于,所述第一模型的信息用于指示所述第一模型的以下至少一项:模型索引、模型的结构信息、模型的参数、模型输入格式、模型输出格式、模型性能指标、模型应用场景、模型功能、或训练参数。
- 如权利要求9至11中任一项所述的方法,其特征在于,还包括:根据所述第一模型,执行模型推理,获得预测信息。
- 如权利要求12所述的方法,其特征在于,还包括:向终端设备发送所述预测信息,所述预测信息中包含预测结果。
- 如权利要求12或13所述的方法,其特征在于,所述预测信息中还包含对所述预测结果所预测的准确率。
- 如权利要求13或14所述的方法,其特征在于,还包括:接收来自所述终端设备的第二反馈信息,所述第二反馈信息用于指示所述预测信息的实际准确率。
- 如权利要求9至15中任一项所述的方法,其特征在于,还包括:向所述OAM发送第一反馈信息,所述第一反馈信息用于指示所述预测信息的实际准确率、应用场景变更指示、模型性能等级变更指示、或模型性能指标变更指示。
- 如权利要求9至16中任一项所述的方法,其特征在于,还包括:确定所述终端设备的目标小区;向所述目标小区对应的第二接入网设备发送所述第一模型的索引。
- 如权利要求17所述的方法,其特征在于,还包括:向所述目标小区对应的第二接入网设备发送第一指示信息,所述第一指示信息用于指示所述第一模型的输入数据的类型信息。
- 如权利要求17或18所述的方法,其特征在于,还包括:向所述第二接入网设备发送所述模型推理所需的历史数据信息。
- 如权利要求17至19中任一项所述的方法,其特征在于,还包括:接收来自所述第二接入网设备的第二指示信息,所述第二指示信息用于指示以下至少一项:所述终端切换到所述目标小区、所述第一接入网设备删除所述第一模型,或所述第一接入网设备释放相应的计算资源;删除所述第一模型,释放相应的计算资源。
- 一种模型配置方法,其特征在于,包括:接收来自第一接入网设备的第一模型的索引;向操作维护管理OAM发送第二模型请求消息,所述第二模型请求消息用于指示所述第一模型的索引;接收来自所述OAM的所述第一模型的信息。
- 如权利要求21所述的方法,其特征在于,所述第一模型的信息用于指示所述第一模型的以下至少一项:模型索引、模型的结构信息、模型的参数、模型输入格式、模型输出格式、模型性能 指标、模型应用场景、模型功能、或训练参数。
- 如权利要求21或22所述的方法,其特征在于,还包括:接收来自所述第一接入网设备的第一指示信息,所述第一指示信息用于指示所述第一模型的输入数据的类型信息。
- 如权利要求21至23中任一项所述的方法,其特征在于,还包括:接收来自所述第一接入网设备的模型推理所需的历史数据信息;在所述模型推理中,所述历史数据信息用于确定所述第一模型的输入数据。
- 如权利要求21至24中任一项所述的方法,其特征在于,还包括:向所述第一接入网设备发送第二指示信息,所述第二指示信息用于指示以下至少一项:所述终端切换到所述目标小区、所述第一接入网设备删除所述第一模型,或所述第一接入网设备释放相应的计算资源。
- 一种通信装置,其特征在于,包括用于实现权利要求1至8中任一项所述方法的单元。
- 一种通信装置,其特征在于,包括处理器和存储器,所述处理器和存储器耦合,所述处理器用于实现权利要求1至8中任一项所述的方法。
- 一种通信装置,其特征在于,包括用于实现权利要求9至20中任一项所述方法的单元。
- 一种通信装置,其特征在于,包括处理器和存储器,所述处理器和存储器耦合,所述处理器用于实现权利要求9至20中任一项所述的方法。
- 一种通信装置,其特征在于,包括用于实现权利要求21至25中任一项所述方法的单元。
- 一种通信装置,其特征在于,包括处理器和存储器,所述处理器和存储器耦合,所述处理器用于实现权利要求21至25中任一项所述的方法。
- 一种通信***,其特征在于,包括权利要求26或27所述的通信装置,权利要求28或29所述的通信装置,和权利要求30或31所述的通信装置。
- 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质上存储有指令,当所述指令在计算机上运行时,使得计算机执行权利要求1至8中任一项所述的方法,或者权利要求9至20中任一项所述的方法,或者权利要求21至25中任一项所述的方法。
- 一种计算机程序产品,其特征在于,包括指令,当所述指令在计算机上运行时,使得计算机执行权利要求1至8中任一项所述的方法,或者权利要求9至20中任一项所述的方法,或者权利要求21至25中任一项所述的方法。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210096175.1 | 2022-01-26 | ||
CN202210096175.1A CN116541088A (zh) | 2022-01-26 | 2022-01-26 | 一种模型配置方法及装置 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2023143267A1 true WO2023143267A1 (zh) | 2023-08-03 |
Family
ID=87449380
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2023/072691 WO2023143267A1 (zh) | 2022-01-26 | 2023-01-17 | 一种模型配置方法及装置 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN116541088A (zh) |
WO (1) | WO2023143267A1 (zh) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117576545B (zh) * | 2024-01-16 | 2024-04-05 | 成都同步新创科技股份有限公司 | 一种多算法全匹配接入适配器接入方法 |
CN117676716B (zh) * | 2024-02-01 | 2024-06-11 | 荣耀终端有限公司 | 通信方法、***及相关设备 |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101146312A (zh) * | 2007-10-17 | 2008-03-19 | 中兴通讯股份有限公司 | 一种无线传播模型的自动选择方法 |
EP2627038A1 (en) * | 2012-02-13 | 2013-08-14 | Alcatel Lucent | Optimization of maintenance activities in communication networks |
CN106663224A (zh) * | 2014-06-30 | 2017-05-10 | 亚马逊科技公司 | 用于机器学习模型评估的交互式界面 |
CN113747462A (zh) * | 2020-05-30 | 2021-12-03 | 华为技术有限公司 | 一种信息处理方法及相关设备 |
CN113839797A (zh) * | 2020-06-23 | 2021-12-24 | 华为技术有限公司 | 数据处理方法和装置 |
-
2022
- 2022-01-26 CN CN202210096175.1A patent/CN116541088A/zh active Pending
-
2023
- 2023-01-17 WO PCT/CN2023/072691 patent/WO2023143267A1/zh unknown
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101146312A (zh) * | 2007-10-17 | 2008-03-19 | 中兴通讯股份有限公司 | 一种无线传播模型的自动选择方法 |
EP2627038A1 (en) * | 2012-02-13 | 2013-08-14 | Alcatel Lucent | Optimization of maintenance activities in communication networks |
CN106663224A (zh) * | 2014-06-30 | 2017-05-10 | 亚马逊科技公司 | 用于机器学习模型评估的交互式界面 |
CN113747462A (zh) * | 2020-05-30 | 2021-12-03 | 华为技术有限公司 | 一种信息处理方法及相关设备 |
CN113839797A (zh) * | 2020-06-23 | 2021-12-24 | 华为技术有限公司 | 数据处理方法和装置 |
Also Published As
Publication number | Publication date |
---|---|
CN116541088A (zh) | 2023-08-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113661727B (zh) | 用于无线网络的无线电接入网(ran)节点的神经网络的配置 | |
US20230209390A1 (en) | Intelligent Radio Access Network | |
US11838787B2 (en) | Functional architecture and interface for non-real-time ran intelligent controller | |
WO2023143267A1 (zh) | 一种模型配置方法及装置 | |
US11496230B2 (en) | Systems and methods for mapping resource blocks to network slices | |
US11310864B2 (en) | Hybrid transmission scheme determination | |
CN113498071A (zh) | 预测无线通信链路的未来服务质量的方法、装置和程序 | |
WO2023155468A1 (zh) | 一种确定根因故障的方法及装置 | |
US20240211770A1 (en) | Communication method and apparatus | |
WO2022087930A1 (zh) | 一种模型配置方法及装置 | |
Aguilar-Garcia et al. | Location-aware self-organizing methods in femtocell networks | |
WO2022038067A1 (en) | Method and apparatus for managing conditional handover of a user equipment | |
Alablani et al. | An SDN/ML-based adaptive cell selection approach for HetNets: a real-world case study in London, UK | |
WO2023070381A1 (en) | Method and apparatus for deploying movable base station | |
Panitsas et al. | Predictive handover strategy in 6g and beyond: A deep and transfer learning approach | |
WO2023011655A1 (zh) | 一种通信方法及装置 | |
US11638171B2 (en) | Systems and methods for dynamic wireless network configuration based on mobile radio unit location | |
Han et al. | The analysis of node planning and control logic optimization of 5G wireless networks under deep mapping learning algorithms | |
CN117597969A (zh) | Ai数据的传输方法、装置、设备及存储介质 | |
WO2024109682A9 (zh) | 一种用于定位的方法及装置 | |
WO2023036280A1 (zh) | 一种模型测试方法及装置 | |
WO2023066346A1 (zh) | 一种通信方法及装置 | |
CN116782323A (zh) | 一种通信网络中的切换方法及装置 | |
EP4258730A1 (en) | Method and apparatus for programmable and customized intelligence for traffic steering in 5g networks using open ran architectures | |
WO2024067248A1 (zh) | 一种获取训练数据集的方法和装置 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 23746149 Country of ref document: EP Kind code of ref document: A1 |