Communication device and wireless communication method

Abstract

Disclosed are a communication device and a wireless communication method, the wireless communication method comprising: a terminal device receiving first resource information sent by a computing arrangement node, the first resource information comprising second resource information of at least one wireless access node, the second resource information represents a computing resource state and a wireless resource state, related to the reasoning task, of the wireless access node in a first time period; the terminal equipment determines a first strategy for a first reasoning task to be executed based on the first resource information, the first reasoning task comprises a plurality of sub-tasks, and the first strategy is used for indicating that all or part of the sub-tasks included in the first reasoning task are calculated by a target wireless access node; the target wireless access node is one of the at least one wireless access node; and the terminal equipment executes the first reasoning task based on the first strategy.

Description

Technical Field

[0001] This application relates to the field of mobile communication technology, including but not limited to communication equipment and wireless communication methods. Background Technology

[0002] User equipment (UE) in a network can generate various tasks, but the computing resources available to the UE are limited. Therefore, other servers in the network need to provide appropriate offloading strategies for these tasks to reduce completion time. Optimizing and improving task completion efficiency is a pressing technical problem that needs to be solved. Summary of the Invention

[0003] This application provides a communication device and a wireless communication method that can improve the reliability of emergency data transmission.

[0004] The technical solution of this application embodiment is implemented as follows: This application provides a terminal device, the terminal device including a first transceiver; and A first processor, coupled to the first transceiver; the first processor is configured to: The first resource information sent by the computation orchestration node is received via the first transceiver. The first resource information includes second resource information of at least one wireless access node. The second resource information characterizes the computational resource status and wireless resource status of the wireless access node related to the inference task in a first time period. Based on the first resource information, a first strategy is determined for the first inference task to be executed. The first inference task includes multiple subtasks. The first strategy is used to instruct all or part of the subtasks included in the first inference task to be calculated by a target wireless access node, and the target wireless access node is one of the at least one wireless access node. Based on the first strategy, perform the first inference task.

[0005] In some embodiments, the second resource information includes computing resource status information; the computing resource status information includes at least one of the following: idle status indication information, the idle status information indicating whether the wireless access node is idle during the first time period; the start time of the first batch of processing within the first time period; the end time of the first batch of processing; the period of the first batch of processing; the remaining capacity of subtasks that the first batch of processing can handle; and computing power evaluation information.

[0006] In some embodiments, the second resource information includes wireless resource status information, which includes wireless resource occupancy rate or wireless resource idle rate between a first time and a second time. The first time is the start time of the first time period, and the second time is the start time of the first batch of processing at the beginning of the first time period.

[0007] In some embodiments, the first processor is configured to: receive the first resource information sent by the computation orchestration node via a connection with the computation orchestration node; the connection includes one of the following: a control plane connection, a user plane connection, and a computation plane connection.

[0008] In some embodiments, the first processor is configured to receive the first resource information sent by the computing orchestration node via an air interface control plane message sent by the serving radio access node of the terminal device.

[0009] In some embodiments, the first processor is configured to: generate a second policy based on the first resource information; send the second policy to the computation orchestration node; receive policy feedback information for the second policy sent by the computation orchestration node; and determine the first policy based on the policy feedback information and the second policy.

[0010] In some embodiments, the first processor is configured to: determine, based on the first resource information, whether all or some of the subtasks in the first inference task are to be executed by the wireless access node; if all or some of the subtasks in the first inference task are to be executed by the wireless access node, send task information of the first inference task to the computation orchestration node; and receive the first policy sent by the computation orchestration node.

[0011] In some embodiments, the first processor is configured to: send subtask information of at least one second subtask processed by the target wireless access node in the first inference task to the target wireless access node when the target wireless access node is the serving wireless access node of the terminal device.

[0012] This application provides a computational orchestration node, which includes a second transceiver and a second processor coupled to the second transceiver. The second processor is configured to: receive, via the second transceiver, first resource information transmitted by at least one wireless access node, the first resource information including second resource information of the at least one wireless access node, the second resource information representing the computational resource status and wireless resource status of the wireless access node related to an inference task within a first time period; and transmit the first resource information to a terminal device via the second transceiver, the first resource information being used to determine a first strategy for a first inference task to be executed, the first inference task including multiple subtasks, the first strategy being used to instruct all or part of the subtasks included in the first inference task to be computed by a target wireless access node, the target wireless access node being one of the at least one wireless access nodes.

[0013] This application provides a wireless communication method applied to a terminal device. The method includes: receiving first resource information sent by a computation orchestration node, the first resource information including second resource information of at least one wireless access node, the second resource information representing the computational resource status and wireless resource status of the wireless access node related to an inference task within a first time period; determining a first strategy for a first inference task to be executed based on the first resource information, the first inference task including multiple subtasks, the first strategy being used to instruct all or part of the subtasks included in the first inference task to be computed by a target wireless access node, the target wireless access node being one of the at least one wireless access node; and executing the first inference task based on the first strategy.

[0014] This application provides a wireless communication method applied to a computation orchestration node. The method includes: receiving first resource information sent by at least one wireless access node, the first resource information including second resource information of the at least one wireless access node, the second resource information representing the computational resource status and wireless resource status of the wireless access node related to an inference task within a first time period; sending the first resource information to a terminal device, the first resource information being used to determine a first strategy for a first inference task to be executed, the first inference task including multiple subtasks, the first strategy being used to instruct all or part of the subtasks included in the first inference task to be computed by a target wireless access node, the target wireless access node being one of the at least one wireless access node.

[0015] This application provides a wireless communication device, the device comprising: Memory is used to store executable instructions or computer programs. The processor, when executing computer-executable instructions or computer programs stored in the memory, implements the wireless communication method provided in the embodiments of this application.

[0016] This application provides a computer-readable storage medium storing a computer program or computer-executable instructions for implementing the wireless communication method provided in this application when executed by a processor.

[0017] This application provides a computer program product, including a computer program or computer executable instructions. When the computer program or computer executable instructions are executed by a processor, they implement the wireless communication method provided in this application.

[0018] In this embodiment, a wireless access node is used as a computing node that can be used to offload inference tasks. A computing orchestration node sends first resource information to a terminal device. The first resource information includes second resource information of at least one wireless access node. The second resource information represents the computing resource status and wireless resource status of the wireless access node related to the inference task within a first time period. The terminal device determines a first strategy for executing the first inference task based on the first resource information, thereby determining that all or part of the sub-tasks in the first inference task will be computed by a target wireless access node among at least one wireless access node. Thus, based on the computing resource status and wireless resource status related to the inference task within the first time period, the execution strategy of the first inference task is determined to ensure the reliability of the execution of the first task. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of an optional scenario for the wireless communication method provided in the embodiments of this application; Figure 2 This is an optional flowchart illustrating the wireless communication method provided in the embodiments of this application. Figure 1 ; Figure 3 This is an optional flowchart illustrating the wireless communication method provided in the embodiments of this application. Figure 2 ; Figure 4 This is an optional network diagram of the wireless communication method provided in the embodiments of this application; Figure 5 This is an optional schematic diagram of batch processing provided in the embodiments of this application; Figure 6 This is an optional flowchart illustrating the wireless communication method provided in the embodiments of this application. Figure 3 ; Figure 7 This is an optional flowchart illustrating the wireless communication method provided in the embodiments of this application. Figure 4 ; Figure 8This is an optional flowchart illustrating the wireless communication method provided in the embodiments of this application. Figure 5 ; Figure 9 This is an optional flowchart illustrating the wireless communication method provided in the embodiments of this application. Figure 6 ; Figure 10 This is an optional flowchart illustrating the wireless communication method provided in the embodiments of this application. Figure 7 ; Figure 11 This is an optional schematic diagram of the inference task provided in an embodiment of this application; Figure 12 This is an optional illustration of the computation flow of the inference task provided in the embodiments of this application. Figure 1 ; Figure 13 This is an optional illustration of the computation flow of the inference task provided in the embodiments of this application. Figure 2 ; Figure 14 This is an optional illustration of the computation flow of the inference task provided in the embodiments of this application. Figure 3 ; Figure 15 This is an optional flowchart illustrating the wireless communication method provided in the embodiments of this application. Figure 8 ; Figure 16 This is an optional flowchart illustrating the wireless communication method provided in the embodiments of this application. Figure 9 ; Figure 17 This is an optional flowchart illustrating the wireless communication method provided in the embodiments of this application. Figure 10 ; Figure 18 This is an optional flowchart illustrating the wireless communication method provided in the embodiments of this application. Figure 10 one; Figure 19 This is an optional flowchart illustrating the wireless communication method provided in the embodiments of this application. Figure 10 two; Figure 20 This is an optional flowchart illustrating the wireless communication method provided in the embodiments of this application. Figure 10 three; Figure 21 This is an optional flowchart illustrating the wireless communication method provided in the embodiments of this application. Figure 10 Four; Figure 22 This is a schematic diagram of an optional structure of the communication device provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0022] In the following description, the terms "first," "second," and "third" are used only to distinguish different objects and do not represent a specific order of objects, nor are they constituting a chronological order. It is understood that "first," "second," and "third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0024] Communication and computing convergence for 6th Generation Mobile Networks (6G) refers to the deep integration of communication and computing capabilities. It aims to provide more efficient and flexible services by optimizing communication network architecture and computing resource allocation. 6G is not merely a simple communication system; it will natively support communication, sensing, and computing services, becoming the network information foundation supporting the efficient and sustainable development of future society.

[0025] As a typical use case of converged computing and general computing, 6G network compute offloading services refer to the process of transmitting some or all of the tasks that would otherwise need to be processed locally on user devices (such as smartphones, tablets, augmented reality glasses, etc.) to remote servers or edge nodes for processing via high-speed, low-latency communication networks. The processing results are then quickly transmitted back to the user device. This approach effectively reduces the computational burden on terminal devices, extends battery life, and improves the user experience.

[0026] Below, we will use a specific example to illustrate how to use 6G networks to provide computing offloading services for users: Consider a user watching a virtual concert using lightweight augmented reality (AR) glasses. This scenario demands extremely high graphics quality while also requiring interaction with the live audience, which necessitates significant real-time data processing capabilities.

[0027] The application in the AR glasses recognizes that the currently running virtual concert experience requires complex image rendering. Determining that local resources are insufficient to efficiently complete these rendering tasks, the application decides to offload some or all of the rendering work to remote computing nodes or cloud servers.

[0028] With the support of 6G networks, AR glasses can quickly search for and select the most suitable edge computing nodes or cloud servers for image rendering.

[0029] User motion capture data (such as head rotation, gestures, etc.), environmental information (such as ambient lighting conditions), and the required 3D models and other resources are compressed and efficiently uploaded to the selected remote computing node via a 6G network.

[0030] After receiving data from the AR glasses, the remote computing node utilizes its powerful GPU resources to begin executing image rendering tasks. The rendering process may involve a series of complex computational operations such as lighting simulation, shadow generation, and material mapping.

[0031] Once completed, the rendered frames are transmitted back to the user's AR glasses at high speed via a 6G network in the form of encoded video streams.

[0032] AR glasses decode the received video stream and overlay it onto the user's actual field of vision, providing an immersive visual experience.

[0033] This application provides a wireless communication method in which a terminal device receives first resource information sent by a computation orchestration node. The first resource information includes second resource information of at least one wireless access node, wherein the second resource information characterizes the computational resource status and wireless resource status of the wireless access node related to an inference task within a first time period. Based on the first resource information, the terminal device determines a first strategy for a first inference task to be executed. The first inference task includes multiple subtasks, and the first strategy is used to instruct that all or part of the subtasks included in the first inference task be computed by a target wireless access node, wherein the target wireless access node is one of the at least one wireless access node. Based on the first strategy, the terminal device executes the first inference task.

[0034] The wireless communication method provided in this application embodiment can be applied to... Figure 1 The scene shown. (As shown) Figure 1 As shown, the scenarios for this application can be considered from the following aspects.

[0035] In aspect 1, in 6G networks, Radio Access Network (RAN) nodes act as computing nodes, providing computing offloading services for UEs. Computing nodes can also be called computing task nodes.

[0036] In aspect 2, each computing task may be performed by a multi-layered Artificial Intelligence Markup Language (AIML) model (e.g., Deep Neural Networks (DNN)). AIML models have different requirements regarding size, number of parameters used in computation, and computation latency. For example, a lightweight image classification model (such as MobileNetV2) is approximately 14 MB and requires less than 100 milliseconds for real-time processing. Similarly, a lightweight speech recognition model (such as DeepSpeech 2) is approximately 65 MB, and to maintain a natural conversational feel, latency generally needs to be kept below 200 milliseconds.

[0037] In aspect 3, the UE sends a computation offload request to the 6G network computation orchestration node as needed. Upon receiving the computation offload request, the 6G network computation orchestration node generates a computation offload strategy (e.g., which computation sub-tasks will be computed locally on the UE, and which computation sub-tasks will be offloaded to which 6G network computation node). Each computation task can be broken down into different computation sub-tasks, i.e., subtasks.

[0038] In one example, lightweight models for image classification (such as MobileNetV2) can be broken down into the following subtasks: Front-end (shallow) processing: This typically includes preprocessing of the input image and the first few convolutional layers. These layers are mainly responsible for extracting basic features such as edges and textures; Intermediate layer processing: Continue with more complex feature extraction, such as shape and object parts; Backend (deep) processing: The last few layers focus on learning advanced features and ultimately making classification decisions.

[0039] In one example, a lightweight speech recognition model (such as DeepSpeech 2) can be broken down into the following sub-tasks: Front-end processing includes audio preprocessing (e.g., sample rate conversion, normalization) and feature extraction (e.g., Mel-frequency cepstral coefficients (MFCCs)). This part can be performed on the user's device to reduce the amount of data transmitted. Intermediate processing: The feature vector corresponding to each audio segment is sent to an edge server or the cloud for further processing.

[0040] Backend processing: The final decoding steps, such as Connectionist Temporal Classification (CTC) decoding, can be completed in the cloud and the results returned to the user device.

[0041] This application provides a wireless communication method. In practical applications, the wireless communication method can be implemented by a wireless communication device. The functional entities within the wireless communication device can be collaboratively implemented using the hardware resources of the communication equipment, such as computing resources like processors and communication resources (e.g., resources used to support various communication methods like optical fiber and cellular networks). The communication equipment may include terminal equipment and network equipment. The network equipment may include access network equipment or core network equipment. The access network equipment may be an evolved Node B (eNB or eNodeB) in a Long Term Evolution (LTE) system, a Next Generation Radio Access Network (NG RAN) device, a base station (gNB) in an NR system, a base station in a 6G system, a radio controller in a Cloud Radio Access Network (CRAN), or a network device in a future Public Land Mobile Network (PLMN), etc. The core network equipment may be a 5G Core (5GC) device. In some embodiments, the core network equipment may also be an Evolved Packet Core (EPC) device for an LTE network. During network evolution, the aforementioned core network equipment may also be called by other names, or new network entities may be formed by dividing the functions of the core network; this application embodiment does not impose any restrictions on this.

[0042] Figure 2An exemplary embodiment shows a network device and two terminal devices. Optionally, the communication system 200 may include multiple network devices and each network device may include other numbers of terminal devices within its coverage area. This application embodiment does not limit this.

[0043] Firstly, the wireless communication method provided in the embodiments of this application is applied to a terminal device, such as... Figure 2 As shown, it includes S201 to S203.

[0044] S201. The terminal device receives first resource information sent by the computing orchestration node. The first resource information includes second resource information of at least one wireless access node. The second resource information represents the computing resource status and wireless resource status of the wireless access node related to the inference task in a first time period.

[0045] S202. Based on the first resource information, the terminal device determines a first strategy for a first inference task to be executed. The first inference task includes multiple sub-tasks. The first strategy is used to instruct all or part of the sub-tasks included in the first inference task to be calculated by a target wireless access node, and the target wireless access node is one of the at least one wireless access node.

[0046] S203. The terminal device executes the first inference task based on the first strategy.

[0047] Secondly, the wireless communication method provided in this application embodiment is applied to computing orchestration nodes, such as... Figure 3 As shown, it includes S301 and S302.

[0048] S301. The computation orchestration node receives first resource information sent by at least one wireless access node. The first resource information includes second resource information of the at least one wireless access node. The second resource information characterizes the computational resource status and wireless resource status of the wireless access node related to the inference task in a first time period.

[0049] S302. The computation orchestration node sends the first resource information to the terminal device. The first resource information is used to determine a first strategy for the first inference task to be executed. The first inference task includes multiple sub-tasks. The first strategy is used to instruct all or part of the sub-tasks included in the first inference task to be computed by a target wireless access node. The target wireless access node is one of the at least one wireless access node.

[0050] Below, on Figure 3 or Figure 4 The wireless communication method shown is described.

[0051] The wireless communication method provided in this application embodiment can be applied to... Figure 4 In the communication system shown, such as Figure 4 As shown, the system includes: a terminal device 401, at least one wireless access node 402, and a computing and orchestration node 403. The terminal device 401 connects to one of the at least one wireless access node 403, and the connected wireless access node provides services to the terminal device. The wireless access node to which the terminal device connects can also be referred to as the serving wireless access node.

[0052] Terminal device 401 can be any terminal device capable of communicating with the accessed wireless access node, and can also communicate with the computing orchestration node through the accessed wireless access node. The terminal device can generate inference tasks based on computing needs. An inference task can be understood as a computing task comprising multiple subtasks. An inference task can also be described as a computing task or a computing inference task. In one example, an inference task may include a computing task performed by a multi-layer Artificial Intelligence Markup Language (AIML) model, where a subtask may be one or more layers within the model. The inference task in this embodiment may include at least one of the following: image rendering, object recognition, video processing, user traffic prediction, and user trajectory prediction.

[0053] In this embodiment of the application, the terminal device can be any terminal device within the coverage area of ​​any one of the at least one wireless access nodes.

[0054] A wireless access node, also known as a radio access network (RAN) or RAN node, provides communication coverage for a specific geographical area and can communicate with terminal devices located within that coverage area, i.e., terminal devices connected to the wireless access node. A wireless access node can also act as a computing node to offload inference tasks from terminal devices. This offloading can be understood as distributing or transmitting all or part of the subtasks of the inference task to the wireless access node for computation, instead of computing them locally on the terminal device.

[0055] The computation orchestration node, also known as the computation coordination node, is capable of communicating with at least one wireless access node.

[0056] In this embodiment, the wireless access node and the computation orchestration node can be located on the network side. The computation orchestration node can be deployed on different physical devices than each of the at least one wireless access node, or it can be deployed on the same physical device, such as a base station, as one of the at least one wireless access nodes. In one example, the computation orchestration node is deployed on a core network device, and the wireless access node is deployed on a base station. In another example, the computation orchestration node is deployed in a centralized unit (CU) of the base station, and the wireless access node is deployed in a distributed unit (DU) of the base station.

[0057] For each of at least one wireless access node, the computing resource status and wireless resource status related to the inference task can be evaluated within a first time period to obtain the second resource information of that wireless access node. The second resource information may include computing resource status information and wireless resource status information. For a wireless access node, the computing resource status information characterizes the computing resource status of that wireless access node related to the inference task within the first time period, and the wireless resource status information, also referred to as wireless resource information, characterizes the wireless resource status of that wireless access node related to the inference task within the first time period. The computing resource status related to the inference task can be understood as the status of the computing resources used to compute the inference task. The wireless resource status related to the inference task can be understood as the status of the wireless resources of the air interface used to transmit information for the inference task.

[0058] In this embodiment, the wireless access node can assess the start time as the starting point of a first time period, the duration of the first time period as the first duration, and determine the computing resource status and wireless resource status related to the inference task within the first time period. Understandably, the wireless access node assesses the computing resource status and wireless resource status related to the inference task within the future first time period.

[0059] In one possible implementation, the wireless access node can periodically evaluate the state of computing resources and wireless resources related to the inference task within a first time period. In another possible implementation, the computation orchestration node can send an evaluation command or evaluation signaling to at least one wireless access node, and the wireless access node, in response to the received evaluation command or evaluation signaling, evaluates the state of computing resources and wireless resources related to the inference task within the first time period.

[0060] At least one wireless access node sends the determined second resource information to the computational orchestration node. The computational orchestration node summarizes the second resource information sent by each wireless access node to obtain the first resource information, and then sends the first resource information to the terminal device.

[0061] After receiving the first resource information sent by the computation orchestration node, the terminal device determines a first strategy for the first inference task based on the first resource information. The first inference task can be any inference task of the terminal device. The first inference task can be an already generated inference task or a predicted inference task that will be generated soon.

[0062] The first inference task includes multiple subtasks, and all or some of these subtasks can be offloaded to a target access node within at least one wireless access node. In one example, in Figure 1 In the above, for UE#1, computation task #1 includes four subtasks: subtask #1-1, subtask #1-2, subtask #1-3, and subtask #1-4. Subtask #1-1 and subtask #1-2 are computed locally on UE#1, while subtask #1-3 and subtask #1-4 are offloaded to computation service #1. For UE#2, computation task #2 includes four subtasks: subtask #2-1, subtask #2-2, subtask #2-3, and subtask #2-4. Subtask #2-1, subtask #2-2, and subtask #2-3 are computed locally on UE#2. #2-4 is offloaded to computing service #1; For UE #3, computing task #3 includes 4 subtasks: subtask #3-1, subtask #3-2, subtask #3-3, and subtask #3-4, of which subtask #3-1 and subtask #3-2 are computed locally on UE #3, and subtask #3-3 and subtask #3-4 are offloaded to computing service #2; For UE #4, computing task #4 includes 3 subtasks: subtask #4-1, subtask #4-2, and subtask #4-3, of which subtask #4-1, subtask #4-2, and subtask #4-3 are computed locally on UE #3.

[0063] The first strategy is used to instruct the offloading of all or part of the subtasks in the first inference task to the target wireless access node. The first strategy is an offloading strategy or computation offloading strategy for the first inference task. This strategy describes whether each subtask in the multiple subtasks included in the inference task is computed locally on the terminal device or offloaded to the wireless access node for computation, and the wireless access node undertaking the offloading of the inference task is the target wireless access node. The first strategy can also be called a first offloading strategy or a first computation offloading strategy.

[0064] In one possible implementation, the terminal device can directly generate a first policy based on the first resource information. In another possible implementation, the terminal device can update the policy that needs adjustment based on the first resource information to obtain the first policy. In yet another possible implementation, the terminal device generates a second policy based on the first resource information, sends the second policy to the network device, and obtains the first policy based on the feedback information from the network device regarding the second policy and the second policy itself. In yet another possible implementation, if the terminal device determines, based on the first resource information, that the first inference task should be offloaded to the wireless access node, it sends the task information of the first inference task to the network device, and the network device sends the first policy for the first inference task to the terminal device. In this embodiment, the method for determining the first policy based on the first resource information is not limited.

[0065] In one possible implementation, after determining a first policy, the terminal device performs local computation based on the first policy and sends the first policy to a target wireless access node in the network, allowing the target wireless access node to simultaneously perform offload computation. In another possible implementation, after determining the first policy, the terminal device sends the first policy to a computation orchestration node in the network. After the computation orchestration node approves the transmission of the first policy, the terminal device performs local computation. In yet another possible implementation, the terminal device may further determine a second policy and send the second policy to a computation orchestration node in the network. The computation orchestration node sends feedback information, and after the terminal device determines the first policy based on the second policy and the feedback information, the terminal device performs local computation.

[0066] After generating the first inference task, the terminal device executes the first inference task based on a first strategy. If all subtasks in the first inference task are calculated by the target wireless access node, the terminal device sends the input information of the first inference task to the target wireless access node. The target wireless access node calculates the first inference task based on the input information, obtains output information, and sends the output information to the terminal device, thus obtaining the output information of the first inference task. If some subtasks in the first inference task are calculated by the target wireless access node, the terminal device calculates one or more of the first subtasks locally based on the input information of the first inference task, obtains intermediate information, and sends the intermediate information to the target wireless access node. The target wireless access node calculates one or more second subtasks based on the intermediate information sent by the terminal device, obtains output information, and sends the output information to the terminal device, thus obtaining the output information of the first inference task. The first subtask is the subtask calculated by the terminal device in the first inference task, and can also be called a local calculation task. The second subtask is the subtask calculated by the wireless access node in the first inference task, and can also be called an offloaded calculation task or a calculation offload task.

[0067] The terminal device can directly send the input or intermediate information of the first inference task to the target wireless access node. In this case, the target wireless access node directly sends the output information of the first inference task to the terminal device. Alternatively, the terminal device can send the input or intermediate information of the first inference task to the computation orchestration node. The computation orchestration node then sends the received input or intermediate information of the first inference task to the target wireless access node. In this case, the target wireless access node sends the output information of the first inference task to the computation orchestration node, and the computation orchestration node then sends the received output information of the first inference task to the terminal device.

[0068] In this embodiment, a wireless access node is used as a computing node that can be used to offload inference tasks. A computing orchestration node sends first resource information to a terminal device. The first resource information includes second resource information of at least one wireless access node. The second resource information represents the computing resource status and wireless resource status of the wireless access node related to the inference task within a first time period. The terminal device determines a first strategy for executing the first inference task based on the first resource information, thereby determining that all or part of the sub-tasks in the first inference task will be computed by a target wireless access node among at least one wireless access node. Thus, based on the computing resource status and wireless resource status related to the inference task within the first time period, the execution strategy of the first inference task is determined to ensure the reliability of the execution of the first task.

[0069] In some embodiments, the second resource information includes computing resource status information; the computing resource status information includes at least one of the following: idle status indication information, the idle status information indicating whether the wireless access node is idle during the first time period; the start time of the first batch of processing included in the first time period; the end time of the first batch of processing; the period of the first batch of processing; the remaining capacity of subtasks that the first batch of processing can carry; and computing power evaluation information.

[0070] The second resource information of a wireless access node includes computing resource status information, which can be used to determine from the perspective of computing resources whether it can support the offloading of the first inference task.

[0071] Within the first time period, a wireless access node may or may not have a batch process that has started. Here, the batch process that starts within the first time period can be referred to as the first batch of processes. It is understood that the start time of the first batch of processes is within the first time period.

[0072] like Figure 5As shown, batch processing can be understood as the time period during which a wireless access node processes subtasks, also known as the inference time period or computation time period. In this embodiment, for the wireless access node, batch processing may include periodic batch processing or triggered batch processing. Periodic batch processing is batch processing that starts based on a fixed period. If there are subtasks to be computed before the start of a batch processing, these subtasks to be computed are computed within this batch processing. If there are no subtasks to be computed before the start of the first batch processing, the batch processing remains idle. For triggered batch processing, no fixed batch processing time or period is set. Computational resources are temporarily configured to compute subtasks based on the unloading request of the terminal device.

[0073] One or more batch processing sessions can begin within the first time period. In this embodiment, the first batch processing may not include batch processing sessions that have already started within the first time period and batch processing sessions whose start time is less than a first time threshold. Batch processing sessions whose start time is less than the first time threshold can be understood as batch processing sessions that will start immediately. The size of the first time threshold can be set according to requirements.

[0074] In one example, such as Figure 5 As shown, for RAN node #1, the first batch of processing within the first time period includes a2; for RAN node #2, the first batch of processing within the first time period includes b2; and for RAN node #3, the first batch of processing within the first time period includes c1.

[0075] The idle status indication information is used to indicate whether the wireless access node is idle during a first time period, i.e., whether there is an inference task that needs to be computed. Here, the inference task to be processed can be a sub-task computed by the wireless access node that is included in other inference tasks.

[0076] One or more first batch processes may be started within the first time period. The computing resource status information may include the start time of the first batch process that starts within the first time period, or the start time of each of the first batch processes that start within the first time period.

[0077] The end time of the first batch of processing can be the estimated latest end time of the first batch of processing, and the actual end time of the first batch of processing can be earlier than or equal to the latest end time of the first batch of processing.

[0078] When the first batch of processing is periodic batch processing, the computing resource status information may include the period of the first batch of processing.

[0079] The remaining capacity of the first batch of processing can be understood as the number of subtasks that can be handled before the first batch of processing ends, excluding subtasks that have already been reserved. In one example, the first batch of processing can handle 10 subtasks, and 3 subtasks have already been reserved. Therefore, the remaining capacity of the first batch of processing can be any number from 0 to 7.

[0080] Computational power assessment information is used to characterize the computing power of the evaluated wireless access node. In one example, the computing power assessment information may include the latency corresponding to the subtask. Here, the latency corresponding to the subtask can be understood as the latency added by adding one more subtask to the subtask computed by the wireless access node.

[0081] In this embodiment, the terminal device can understand the computing capabilities of the wireless access node related to the inference task in the first time period based on the computing resource status information, thereby assessing whether the first inference task can be offloaded to the wireless access node, and how many sub-tasks can be offloaded to the wireless access node, and determining the offloading method from the perspective of computing resources.

[0082] In some embodiments, the second resource information includes wireless resource status information, which includes wireless resource occupancy rate or wireless resource idle rate between a first time and a second time. The first time is the start time of the first time period, and the second time is the start time of the first batch of processing within the first time period.

[0083] The second resource information of a wireless access node includes wireless resource status information, which can be used to determine whether the offloading of the first inference task can be supported from the perspective of the wireless resources of the air interface.

[0084] Radio resource status information can be used to indicate the remaining radio resources, i.e., idle radio resources, of a radio access node within a first time period.

[0085] If there are one or more first-batch processing sessions within the first time period, there may be one or more second time periods. In this case, the radio resource status information may include the radio resource occupancy rate or radio resource idle rate between each of the first time period and the one or more second time periods.

[0086] In one example, for Figure 5The batch processing shown includes, for RAN node #1, the radio resource status information includes the radio resource occupancy rate or radio resource idle rate between the start time of the first time period and the start time of a2; for RAN node #2, the radio resource status information includes the radio resource occupancy rate or radio resource idle rate between the start time of the first time period and the start time of b2; and for RAN node #3, the radio resource status information includes the radio resource occupancy rate or radio resource idle rate between the start time of the first time period and the start time of c1.

[0087] The wireless resource occupancy rate or wireless resource idle rate between the first time and the second time can be the wireless resource occupancy rate or wireless resource idle rate of the serving cell of the wireless access node.

[0088] The radio resource occupancy rate or radio resource idle rate of the serving cell can be the average radio resource occupancy rate or radio resource idle rate of the serving cell. The average radio resource occupancy rate or radio resource idle rate of the serving cell can include an integer value. The value can range from 1 to 10 or from 1 to 100. When the value ranges from 1 to 10, 10 represents that the radio resources are fully occupied or idle; when the value ranges from 1 to 100, 100 represents that the radio resources are fully occupied or idle. The embodiments of this application do not limit the range of this value.

[0089] The radio resource occupancy rate or radio resource idle rate of a serving cell may include: the radio resource occupancy rate or radio resource idle rate of each serving cell in at least one serving cell of a radio access node; or, the radio resource occupancy rate or radio resource idle rate of each synchronization signal in at least one synchronization signal of a radio access node; or, the radio resource occupancy rate or radio resource idle rate of each physical resource block (PRB) in at least one physical resource block (PRB) of a radio access node. Synchronization signals may include synchronization signal blocks (SSBs).

[0090] The radio resource occupancy rate or radio resource idle rate of a serving cell can be represented by an integer value. For example, the value ranges from 1 to 100, where 100 indicates that the radio resources of the corresponding serving cell are fully occupied or idle.

[0091] The radio resource occupancy rate or radio resource idle rate of a synchronization signal can be represented by an integer value. For example, the value ranges from 1 to 100, where 100 indicates that the radio resources of the corresponding synchronization signal are fully occupied or idle.

[0092] The radio resource occupancy rate or radio resource idle rate of a PRB can be represented by an integer value. For example, the value ranges from 1 to 10, where 10 indicates that the radio resource of the corresponding PRB is fully occupied or idle.

[0093] In some embodiments, the wireless resource status information may further include: historical resource occupancy information. Historical resource occupancy information may include at least one of the following: historical occupancy rate or historical idle rate of wireless resources, and historical statistical information of wireless resources. Historical statistical information of wireless resources may include: variance, prediction accuracy, etc.

[0094] In this embodiment, the radio resource status information can be used to evaluate the success rate of the first inference task accessing the first batch of processing. Here, the access of the first inference task to the first batch of processing can be understood as the input information or intermediate information of the first inference task being sent to the radio access node before the start of the first batch of processing, so that the first batch of processing can perform calculations on the second subtask in the first inference task.

[0095] In some embodiments, for a terminal device, S201 receiving first resource information sent by a computing orchestration node may include: receiving the first resource information sent by the computing orchestration node through a connection with the computing orchestration node; the connection includes one of the following: a control plane connection, a user plane connection, and a computing plane connection.

[0096] In some embodiments, for a computation orchestration node, S302, sending the first resource information to the terminal device may include: sending the first resource information to the terminal device through a connection with the terminal device; the connection includes one of the following: a control plane connection, a user plane connection, and a computation plane connection.

[0097] The computation orchestration node can establish a connection with the terminal device. Understandably, this link between the computation orchestration node and the terminal device is an end-to-end logical connection, established by the computation orchestration node through a serving radio access node accessed by the terminal device. This connection may include a control plane connection, a user plane connection, and / or a computation plane connection. When a control plane connection is established between the computation orchestration node and the terminal device, the computation orchestration node can send first resource information to the terminal device via the control plane connection, and the terminal device can receive the first resource information via the control plane connection. When a user plane connection is established between the computation orchestration node and the terminal device, the computation orchestration node can send first resource information to the terminal device via the user plane connection, and the terminal device can receive the first resource information via the user plane connection. When a computation plane connection is established between the computation orchestration node and the terminal device, the computation orchestration node can send first resource information to the terminal device via the computation plane connection, and the terminal device can receive the first resource information via the computation plane connection.

[0098] Understandably, the connection between the compute orchestration node and the terminal device physically passes through the serving radio access node. When the first resource information is sent to the compute orchestration node through this connection, the receiving end is the terminal device, and the serving radio access node does not need to parse the sent content.

[0099] In this embodiment of the application, the computational orchestration node can directly send the first resource information to the terminal device through the established connection between the computational orchestration node and the terminal device, which can reduce the transmission latency of the first resource information and improve the transmission reliability of the first resource information.

[0100] In some embodiments, for a terminal device, S201 receiving first resource information sent by a computing orchestration node may include: receiving the first resource information sent by the computing orchestration node through an air interface control plane message sent by the serving radio access node of the terminal device.

[0101] In some embodiments, for the computation orchestration node, S302, sending the first resource information to the terminal device may include: sending the first resource information to the serving radio access node of the terminal device to send the first resource information through control plane messages sent by the serving radio access node of the terminal device.

[0102] like Figure 6 As shown, the computation orchestration node can first send the first resource information to the serving radio access node. After receiving the first resource information, the serving radio access node sends the first resource information to the terminal device through the control plane message of the air interface.

[0103] The computation orchestration node can transmit first resource information through messages supported by the interface between the computation orchestration node and the service wireless access node. These messages may include Hypertext Transfer Protocol (HTTP) / Hypertext Transfer Protocol Secure (HTTPS) messages.

[0104] The serving radio access node can send first resource information to the terminal device via control plane messages over the air interface. Control plane messages over the air interface may include system messages and Radio Resource Control (RRC) messages. System messages may include: Master Information Block (MIB) and System Information Block (SIB).

[0105] exist Figure 6In this embodiment, taking the example of the computing orchestration node sending an HTTP message to the serving radio access node and the serving radio access node sending an SIB to the terminal device, the types of messages sent by the computing orchestration node to the serving radio access node and the types of control plane messages sent by the serving radio access node to the terminal device are not limited.

[0106] Understandably, when the computation orchestration node sends the first resource information to the serving radio access node, the receiving end is the serving radio access node. The serving radio access node parses the received content to obtain the first resource information, encapsulates the first resource information again, and sends it to the terminal device in a control message. The receiving end of the first resource information sent by the serving radio access node is the terminal device.

[0107] In this embodiment, the computation orchestration node can first send the first resource information to the serving radio access node. After receiving the first resource information, the serving radio access node sends the first resource information to the terminal device through the control plane message of the air interface. It is not necessary to establish an end-to-end connection between the computation orchestration node and the terminal device, thereby not limiting the scenarios for the transmission of the first resource information and improving the applicability of the solution.

[0108] In some embodiments, for a terminal device, S202, determining a first strategy for a first inference task to be executed based on the first resource information may include: generating a second strategy based on the first resource information; sending the second strategy to the computation orchestration node; receiving strategy feedback information for the second strategy sent by the computation orchestration node; and determining the first strategy based on the strategy feedback information and the second strategy.

[0109] In some embodiments, for a computational orchestration node, the wireless communication method provided in this application further includes: receiving a second strategy sent by the terminal device; sending strategy feedback information for the second strategy to the terminal device; the strategy feedback information and the second strategy are used to determine the first strategy.

[0110] like Figure 7 As shown, the process includes steps S701 to S705. S701: The terminal device generates a second policy based on the first resource information. S702: The terminal device sends the second policy to the computational orchestration node. S703: The computational orchestration node determines policy feedback information for the second policy. S704: The computational orchestration node sends the policy feedback information for the second policy to the terminal device. S705: The terminal device determines a first policy based on the second policy and the policy feedback information.

[0111] Here, the second policy generated by the terminal device based on the first resource information can be understood as a policy suggested by the terminal device. The terminal device sends the second policy to the computational orchestration node, which determines whether to adopt the second policy and returns policy feedback information for the second policy. The terminal device can generate the second policy directly based on the first resource information, or it can generate the second policy based on the first resource information and a policy that needs to be adjusted. Here, the policy that needs to be adjusted can be a preset initial policy, or it can be a policy that the network device does not allow the terminal device to adopt, which can also be generated based on the first resource information.

[0112] In this embodiment, the terminal device can determine a second strategy for the first inference task based on first resource information and first general computing resource information. The first general computing resource information characterizes the general computing resources of the terminal device. General computing resources include communication resources and computing resources. Communication resources are resources used for transmitting data and may include bandwidth, rate, latency, reliability, etc. Computing resources are resources used for processing or calculating data.

[0113] In this embodiment, the terminal device autonomously determines the offloading strategy, which can be based on its own implementation and the first resource information related to the inference task of the wireless access node. This makes the determined strategy more in line with the implementation of the terminal device and improves the fit between the first strategy and the terminal device.

[0114] In some embodiments, the second strategy includes at least one of the following: The subtask identifier of at least one first subtask processed by the terminal device in the first inference task; The subtask identifier of at least one second subtask processed by the wireless access node in the first inference task; A subtask segmentation identifier, the segmentation identifier being used to identify the at least one first subtask and the at least one second subtask in the first inference task.

[0115] The first subtask can be understood as a subtask calculated locally by the terminal device, and the second subtask can be understood as a subtask calculated by the wireless access node.

[0116] The second strategy includes task allocation information for determining at least one first subtask and at least one second subtask. The task allocation information characterizes the allocation method of the first inference task. The task allocation information may include at least one of the following: a subtask identifier of at least one first subtask in the first inference task; a subtask identifier of at least one second subtask in the first inference task; and a subtask segmentation identifier.

[0117] A subtask identifier for at least one first subtask can be used to identify at least one first subtask. A subtask identifier for at least one second subtask can be used to identify at least one second subtask. A subtask split identifier can identify either the first second subtask among at least one second subtask or the last first subtask among at least one first subtask. When a subtask split identifier identifies the first second subtask among at least one second subtask, the subtask preceding the subtask identified by the split identifier is considered a first subtask, and the subtask identified by the split identifier and the subtask following it are considered second subtasks. When a subtask split identifier identifies the last first subtask among at least one first subtask, the subtask identified by the split identifier and the subtask preceding it are considered first subtasks, and the subtask following it are considered second subtasks.

[0118] In this embodiment, the subtask identifier may include the ID or index of the subtask. The subtask segmentation identifier may include the ID or index of the first second subtask, or the ID or index of the last first subtask.

[0119] In this embodiment, the second strategy may designate all or some of the subtasks as second subtasks. When the second strategy designates all subtasks as second subtasks, the second strategy does not include the subtask identifier of the first subtask.

[0120] In this embodiment of the application, the second strategy may further include a task identifier for the first inference task. The task identifier for the first inference task may include the ID or index of the first inference task.

[0121] In some embodiments, the second strategy further includes at least one of the following: the node identifier of the first candidate wireless access node; the start time and / or end time of the wireless access node processing the first inference task or the at least one second subtask.

[0122] The first candidate wireless access node can be understood as the wireless access node determined by the terminal device for calculating the second sub-task or the suggested target wireless access node.

[0123] The start time and / or end time of the wireless access node processing the first inference task or the at least one second sub-task can be understood as the time during which the first candidate wireless access node established by the terminal device processes at least one second sub-task.

[0124] In this embodiment of the application, the second strategy includes the content suggested by the terminal device based on the first resource information for processing the first inference task. The terminal device provides a suggestion on the allocation method of sub-tasks in the first inference task, and may also provide a suggestion on unloading the wireless access node of the first inference task and / or a suggestion on the time of calculating at least one second sub-task in the first inference task, thereby providing a suggestion on unloading the orchestration node for the first inference task.

[0125] In some embodiments, the policy feedback information includes at least one of the following: First indication information, the first indication information being used to indicate whether the second strategy is permitted; The node identifier or IP address of the target wireless access node.

[0126] The first indication can be understood as a binary feedback, indicating whether the second strategy is allowed or not. In one possible implementation, the first indication can indicate whether the second strategy is allowed based on different values.

[0127] If the second strategy does not include the first candidate wireless access node, and the first indication information indicates that the second strategy is not allowed, then the second strategy may be used as a strategy that needs to be adjusted, or the terminal device may terminate the second strategy and generate a new second strategy based on the first resource information.

[0128] If the second strategy does not include the first candidate wireless access node, the first indication information indicates that the second strategy is allowed. The strategy feedback information may also include the node identifier or IP address of the target wireless access node. Then the second strategy and the node identifier or IP address of the target wireless access node constitute the first strategy.

[0129] If the second strategy includes the first candidate wireless access node, and the first indication information indicates that the second strategy is allowed, then the second strategy is the first strategy.

[0130] If the second strategy includes the first candidate wireless access node, the first indication information indicates that the second strategy is not allowed, and the strategy feedback information does not include the node identifier or IP address of the target wireless access node. The second strategy can be used as a strategy that needs to be adjusted, or the terminal device can terminate the second strategy and generate a new second strategy based on the first resource information.

[0131] When the second strategy includes the first candidate wireless access node, the first indication information indicates that the second strategy is not allowed. The strategy feedback information includes the node identifier or IP address of the target wireless access node. The content of the second strategy other than the node identifier of the first candidate wireless access node and the node identifier or IP address of the target wireless access node constitute the first strategy. In this case, it can be considered that replacing the first candidate wireless access node in the second strategy with the target wireless access node yields the first strategy.

[0132] In this embodiment, the terminal device determines the final first strategy based on the second strategy and the strategy feedback information fed back by the computing orchestration node for the second strategy. Taking into account the terminal device's own implementation, the network device determines the final first strategy based on scheduling considerations.

[0133] In some embodiments, the computational orchestration node can evaluate second general computing resource information, which represents the general computing resources of each wireless access node in at least one wireless node. Based on the second general computing resource information, the node evaluates the second policy to determine whether to allow the second policy and the target wireless access node. Based on the determined result, policy feedback information is fed back to the terminal device.

[0134] In some embodiments, for a computational orchestration node, the wireless communication method provided in this application further includes: sending the second strategy to at least one second candidate wireless access node; receiving an acceptance preference degree for the second strategy sent by the at least one second candidate wireless access node; and determining the target wireless access node from the at least one second candidate wireless access node based on the received acceptance preference degree.

[0135] like Figure 8 As shown, the process includes steps S801 to S804. S801: The computational orchestration node sends a second policy to the second candidate radio access node. S802: The second candidate radio access node evaluates the second policy and determines its acceptance preference. S803: The second candidate radio access node sends its acceptance preference to the computational orchestration node. S804: The computational orchestration node determines the target radio access node based on the received acceptance preference.

[0136] Figure 8 The diagram illustrates the interaction between a second candidate wireless node and a computational orchestration node. In practice, it may include at least one second candidate wireless access node. The interaction between each second candidate wireless access node and the computational orchestration node can be found in [reference needed]. Figure 8 .

[0137] At least one second candidate radio access node may include all or some of the at least one radio access node. After receiving the second policy, the orchestration node sends the second policy to at least one second candidate radio access node. If the second policy does not include the first candidate radio access node, the second candidate radio access node determines its acceptance preference for the second policy. If the second policy includes the first candidate radio access node, the second candidate radio access node determines its acceptance preference for the content of the second policy other than that of the first candidate radio access node.

[0138] Here, the acceptance preference can be understood as the second candidate wireless access node's acceptance or lack thereof, or degree of acceptance, of the allocation method and / or computation time for the first inference task described by the second strategy. In one possible implementation, the acceptance preference may include two values, corresponding to acceptance or non-acceptance respectively. In one example, the acceptance preference is a Boolean value (0 or 1), where 0 represents non-acceptance and 1 represents acceptance. In one possible implementation, the acceptance preference may include a first number of values, which may be greater than 2, with different values ​​representing different degrees of acceptance. In one example, the acceptance preference ranges from 1 to 100, with a higher value indicating a higher degree of acceptance of the second strategy.

[0139] If the second strategy does not include the first candidate radio access node, the orchestration node calculates the acceptance preference of at least one second candidate radio access node received from the second candidate radio access nodes and selects the second candidate radio access node that accepts the second strategy as the target radio access node, or selects the second candidate radio access node with the highest acceptance level as the target radio access node. In this case, the first strategy includes the second strategy and the target radio access node.

[0140] When the second strategy includes the first candidate radio access node, the orchestration node can determine whether the acceptance preference of the first candidate radio access node indicates acceptance. If the acceptance preference of the first candidate radio access node indicates acceptance, the first candidate radio access node is determined as the target radio access node; in this case, the second strategy is the first strategy. If the acceptance preference of the first candidate radio access node indicates non-acceptance, the second candidate radio access node that accepts the second strategy is selected as the target radio access node; in this case, the first strategy is the second strategy of replacing the first candidate radio access node with the target radio access node.

[0141] When the second strategy includes the first candidate wireless access node, the orchestration node can determine whether the acceptance preference of the first candidate wireless access node is higher than the preference threshold. If the acceptance preference of the first candidate wireless access node is higher than the preference threshold, the first candidate wireless access node is determined as the target wireless access node, and in this case, the second strategy is the first strategy. If the acceptance preference of the first candidate wireless access node is not higher than the preference threshold, the second candidate wireless access node with the highest acceptance is selected as the target wireless access node, and in this case, the first strategy is the second strategy of replacing the first candidate wireless access node with the target wireless access node.

[0142] In practical applications, when the second strategy includes the first candidate radio access node, the orchestration node can first send the second strategy to the first candidate radio access node. If the first candidate radio access node's acceptance preference indication is accepted or higher than the acceptance threshold, the first candidate radio access node is determined as the target radio access node. If the first candidate radio access node's acceptance preference indication is not accepted or is not higher than the acceptance threshold, the second strategy is then sent to other second candidate radio access nodes, and the target radio access node is selected from these other second candidate radio access nodes.

[0143] In this embodiment of the application, when the computation orchestration node determines the target wireless access node, a second indication information can be sent to the target wireless access node. The second indication information is used to instruct the target wireless access node to unload the first inference task.

[0144] In some embodiments, for a terminal device, S202 determines a first strategy for a first inference task to be executed based on the first resource information, which may include: determining, based on the first resource information, whether a wireless access node should execute all or part of the sub-tasks in the first inference task; if a wireless access node executes all or part of the sub-tasks in the first inference task, sending task information of the first inference task to the computation orchestration node; and receiving the first strategy sent by the computation orchestration node.

[0145] In some embodiments, for computational orchestration nodes, the wireless communication method provided in this application further includes: Receive task information of the first inference task sent by the terminal device; determine the first strategy based on the task information; and send the first strategy to the terminal device.

[0146] like Figure 9 As shown, it includes S901 to S904.

[0147] S901, The terminal device determines whether to offload the first inference task to the wireless access node based on the first resource information.

[0148] If it is determined that the unloading is performed by the wireless access node, which will calculate some or all of the subtasks, then S902 is executed.

[0149] S902, The terminal device sends the task information of the first inference task to the computing orchestration node.

[0150] S903, the computation orchestration node determines the first strategy for the first inference task based on the task information.

[0151] S904, the computational orchestration node sends the first strategy to the terminal device.

[0152] Here, the terminal device determines whether to offload the first inference task to the wireless access node based on the first resource information. If the terminal device determines not to offload the first inference task to the wireless access node, the terminal device computes all subtasks included in the first inference task locally. If the terminal device determines to offload the first inference task to the wireless access node, the terminal device sends the task information of the first inference task to the computation orchestration node. Upon receiving the task information of the first inference task, the computation orchestration node determines a first strategy based on the task information of the first inference task.

[0153] In this embodiment, the task information of the first inference task may include: first task information related to the first inference task and second task information related to the subtasks within the first inference task. The first task information may include at least one of the following: a task identifier for the first inference task, a task description for the first inference task, complete AI model parameters or files used by the first inference task, and the Quality of Service (QoS) requirements for the first inference task. The task description for the first inference task may include an application description of the application to which the first inference task belongs. QoS requirements may include latency requirements, computing power requirements, etc. The second task information may include at least one of the following: a task identifier for a subtask, a task description for the subtask, complete AI model parameters or files used by the subtask, and the Quality of Service (QoS) requirements for the subtask. The task description for the subtask may include the type of model corresponding to the subtask.

[0154] In this embodiment, the computation orchestration node can determine a first strategy for the first inference task based on first resource information and / or second general computing resource information. The second general computing resource information represents the general computing resources of each wireless access node in at least one wireless node.

[0155] In this embodiment of the application, the computational orchestration node can directly determine the first strategy based on the first resource information and task information, or it can first determine the third strategy based on the first resource information and task information, and then determine the first strategy through the evaluation of the third strategy.

[0156] In this embodiment, the computing orchestration node autonomously determines the offloading strategy, which can be based on the computing resources of the wireless access nodes it schedules, thereby reducing the processing steps of the terminal device and reducing the computing load of the terminal device.

[0157] In some embodiments, the first strategy includes the node identifier or IP address of the target wireless access node and at least one of the following: a subtask identifier of at least one first subtask processed by the terminal device in the first inference task; a subtask identifier of at least one second subtask processed by the wireless access node in the first inference task; and a segmentation identifier for segmenting the at least one first subtask and the at least one second subtask in the first inference task.

[0158] In this embodiment, the first strategy may include the node identifier or IP address of the target access node and task allocation information for determining at least one first subtask and at least one second subtask. A description of the task allocation information included in the first strategy in this embodiment can be found in the description of task allocation information in the second strategy, and will not be repeated here.

[0159] The first strategy may also include a task identifier for the first inference task. The task identifier for the first inference task may include the ID or index of the first inference task.

[0160] In some embodiments, the computation orchestration node may evaluate second general computing resource information, which characterizes the general computing resources of each wireless access node in at least one wireless node, and determine a first strategy based on the second general computing resource information.

[0161] In some embodiments, for a computation orchestration node, determining the first strategy based on the task information may include: generating a third strategy based on the task information; sending the third strategy to at least one second candidate radio access node; receiving an acceptance preference for the third strategy sent by the at least one second candidate radio access node; determining the target radio access node from the at least one second candidate radio access node based on the received acceptance preference; and determining the first strategy based on the target radio access node and the third strategy.

[0162] like Figure 10As shown, the process includes steps S1001 to S1004. S1001: The computational orchestration node determines a third strategy based on task information. S1002: The computational orchestration node sends the third strategy to the second candidate radio access node. S1003: The second candidate radio access node evaluates the third strategy and determines its acceptance preference. S1004: The second candidate radio access node sends its acceptance preference to the computational orchestration node. S1005: The computational orchestration node determines the target radio access node based on the received acceptance preference. Figure 10 The diagram illustrates the interaction between a second candidate wireless node and a computational orchestration node. In practice, it may include at least one second candidate wireless access node. The interaction between each second candidate wireless access node and the computational orchestration node can be found in [reference needed]. Figure 10 .

[0163] In this embodiment, the content of the third strategy can be found in the content of the second strategy. The difference between the third strategy and the yield measurement is that the third strategy is determined by the computational orchestration node based on the first resource information and / or the second general computing resource, while the second strategy is determined by the terminal device based on the first resource information.

[0164] The third strategy may include the node identifier of the first candidate wireless access node, or it may not include the node identifier of the first candidate wireless access node.

[0165] After the orchestration node generates the third policy, it sends the third policy to at least one second candidate radio access node. If the third policy does not include the first candidate radio access node, the second candidate radio access node determines its acceptance preference for the third policy. If the third policy includes the first candidate radio access node, the second candidate radio access node determines its acceptance preference for the content of the second policy other than that of the first candidate radio access node.

[0166] Here, the acceptance preference can be understood as the second candidate wireless access node's acceptance or lack thereof, or degree of acceptance, of the allocation method and / or computation time for the first inference task described by the third strategy. In one possible implementation, the acceptance preference may include two values, corresponding to acceptance or non-acceptance respectively. In one example, the acceptance preference is a Boolean value (0 or 1), where 0 represents non-acceptance and 1 represents acceptance. In one possible implementation, the acceptance preference may include a first number of values, which may be greater than 2, with different values ​​representing different degrees of acceptance. In one example, the acceptance preference ranges from 1 to 100, with a higher value indicating a greater degree of acceptance of the second strategy.

[0167] If the third strategy does not include the first candidate radio access node, the orchestration node calculates the acceptance preference of at least one second candidate radio access node received from the third strategy and selects the second candidate radio access node that accepts the third strategy as the target radio access node, or selects the second candidate radio access node with the highest acceptance level as the target radio access node. The first strategy includes the third strategy and the target radio access node.

[0168] When the third strategy includes the first candidate radio access node, the orchestration node can determine whether the acceptance preference of the first candidate radio access node indicates acceptance. If the acceptance preference of the first candidate radio access node indicates acceptance, the first candidate radio access node is determined as the target radio access node, and the third strategy becomes the first strategy. If the acceptance preference of the first candidate radio access node indicates non-acceptance, the second candidate radio access node that accepts the third strategy is selected as the target radio access node. The first strategy includes all the contents of the third strategy except for the first candidate radio access node and the target radio access node. The first strategy can also be understood as the third strategy that updates the first candidate radio access node to the target radio access node.

[0169] When the third strategy includes the first candidate wireless access node, the orchestration node can determine whether the acceptance preference of the first candidate wireless access node is higher than the preference threshold. If the acceptance preference of the first candidate wireless access node is higher than the preference threshold, the first candidate wireless access node is determined as the target wireless access node, and the third strategy is the first strategy. When the acceptance preference of the first candidate wireless access node is not higher than the preference threshold, the second candidate wireless access node with the highest acceptance is selected as the target wireless access node. The first strategy includes the content of the third strategy other than the first candidate wireless access node and the target wireless access node. The first strategy can also be understood as the third strategy of updating the first candidate wireless access node to the target wireless access node.

[0170] In practical applications, when the third strategy includes the first candidate wireless access node, the orchestration node can first send the third strategy to the first candidate wireless access node. If the first candidate wireless access node's acceptance preference indication is accepted or higher than the acceptance threshold, the first candidate wireless access node is determined as the target wireless access node. If the first candidate wireless access node's acceptance preference indication is not accepted or is not higher than the acceptance threshold, the third strategy is then sent to other second candidate wireless access nodes, and the target wireless access node is selected from these second candidate wireless access nodes.

[0171] In this embodiment of the application, when the computation orchestration node determines the target wireless access node, a second indication information can be sent to the target wireless access node. The second indication information is used to instruct the target wireless access node to unload the first inference task.

[0172] In this embodiment, after determining the first strategy, the terminal device determines at least one second subtask in the first inference task and a target access node. It can send subtask information of at least one second subtask to the target wireless access node, and the target wireless access node executes at least one second subtask based on the received subtask information. The subtask information can also be understood as second task information.

[0173] In some embodiments, for a terminal device, the wireless communication method provided in this application further includes: If the target wireless access node is the serving wireless access node of the terminal device, subtask information of at least one second subtask processed by the target wireless access node in the first inference task is sent to the target wireless access node.

[0174] When the target radio access node is the serving radio access node of the terminal device, the terminal device can directly communicate with the target radio access node using non-IP technology. In this case, the terminal device sends subtask information of at least one second subtask to the target radio access node. The terminal device and the target radio access node can send subtask information of at least one second subtask through protocol layer messages between the UE and the RAN.

[0175] In some embodiments, for a terminal device, the wireless communication method provided in this application further includes: If the target wireless access node is a non-serving wireless access node of the terminal device, switch to the target wireless access node; send subtask information of at least one second subtask processed by the target wireless access node in the first inference task to the target wireless access node.

[0176] If the target wireless access node is neither the serving wireless access node nor a non-serving wireless access node of the terminal device, the terminal device cannot directly communicate with the target wireless access node without IP. In this case, the terminal device performs a handover process, switching the serving wireless access node to the target wireless access node. When the terminal device switches to the target wireless access node, it sends at least one subtask of the second subtask to the target wireless access node.

[0177] The terminal device and the target radio access node can send subtask information of at least one second subtask through protocol layer messages between the UE and the RAN.

[0178] In some embodiments, for a terminal device, the wireless communication method provided in this application further includes: Using the IP address of the target wireless access node, send subtask information of at least one second subtask processed by the target wireless access node in the first inference task to the target wireless access node.

[0179] Here, it is not important whether the target wireless access node is a serving wireless access node; the target wireless access node can be either a serving or non-serving wireless access node. The terminal device sends at least one subtask information of the second subtask to the target wireless access node based on the target wireless access node's IP address.

[0180] In this embodiment of the application, when sending subtask information of at least one second subtask based on the IP address, it is not necessary to pay attention to whether the target wireless access node is the serving wireless access node of the terminal device, thereby reducing the implementation complexity of the scheme.

[0181] The wireless communication method provided in the embodiments of this application will now be described.

[0182] In related technologies, a framework is proposed to jointly optimize DNN inference task offloading and offloading task scheduling for multi-user collaborative inference on edge servers with batch processing capabilities. It systematically addresses the problem of minimizing user energy consumption under inference latency constraints in both offline and online scenarios. Specifically, we propose Independent Per-user Subtask Scheduling and Allocation (IP-SSA), which independently offloads subtasks for each user and schedules all identical subtasks in the same batch for tasks with the same latency constraints. We also propose an Offloading Grouper (OG) that groups tasks with different latency constraints into tasks with similar latency constraints. Experimental results show that IP-SSA and OG can significantly reduce user energy consumption through batch processing. Furthermore, for online scenarios, we propose a Deep Deterministic Policy Gradient-Offloading Grouper (DDPG-OG), where training a reinforcement learning (RL) agent controls the trade-off between serving arrived tasks and reserving resources for future tasks through a proposed two-dimensional control. Future work could further explore large-scale edge inference systems with multiple servers, where low-complexity distributed algorithms for user association, load balancing, and batch queue scheduling may prove important. This approach proposes that in multi-user compute offloading scenarios, batch processing of similar computational tasks reduces overall computational latency.

[0183] In related technologies, a task offloading method with topological dependencies in mobile scenarios is proposed. This method includes: obtaining the deadlines of each subtask of a computation task based on the dependency structure within the computation task in a mobile edge network and the task's deadline; and formulating a computation offloading decision for each subtask based on the dependency structure, the network structure of the mobile edge network, potential offloading opportunities arising from mobility within the mobile edge network, and the subtask's deadline, thereby completing the offloading of the computation task. In this scheme, a computation offloading strategy for each subtask is formulated based on the task's deadline.

[0184] In the embodiments of this application, Figure 1In the scenario shown, in order to ensure that different computing tasks from different UEs can be completed and returned to the UE within their respective latency requirements, the 6G network computing orchestration node needs to consider the following factors to determine the computing offloading strategy and orchestrate different computing subtasks (for batch processing): Factor 1, latency of transmitting intermediate computing variables in the air interface uplink; Factor 2, latency of transmitting computing results in the air interface downlink; Factor 3, computing latency of computing subtasks in the 6G network computing node.

[0185] To achieve the above objectives, in the wireless communication method proposed in this application, the UE can autonomously generate a suggested computation offloading strategy based on the computation resource status and radio resource availability of the computation task RAN ​​node provided by the network, and provide it to the computation orchestration node for reference. The computation task RAN ​​node and the computation orchestration node can interact and negotiate the final computation offloading strategy, including selecting a suitable computation task RAN ​​node.

[0186] In collaborative inference scenarios of Mobile Edge Computing (MEC), multiple user-generated inference tasks, such as... Figure 11 The task shown consists of a series of sub-tasks, also known as AI reasoning tasks. Figure 11 In this example, taking an inference task comprising four subtasks (subtask 1, subtask 2, subtask 3, and subtask 4), the input to the inference task is fed into the first subtask, subtask 1. The output of the previous subtask becomes the input of the next subtask, and so on, until the output of the last subtask (subtask 4). The output of the last subtask is the output of the inference task, i.e., the computation result. Taking the Mobilenet-v2 model as an example, the complete model can be divided into eight parts: subtask 1 to subtask 8. Conv, B, and CLS are abbreviations for convolutional layer, bottleneck module, and classification layer, respectively. In one computation offloading strategy, such as... Figure 12 As shown, subtasks #1 to #3 are calculated locally on the UE, and the generated intermediate variables (of size [28, 28, 32]) are uploaded from the UE to the network via the air interface as input for subtask #4. Subsequently, subtasks #4-8 are calculated on the network side, and the generated calculation results, i.e., the output, are sent back to the UE via downlink.

[0187] Mobile devices are capable of autonomously executing the initial subtasks of an inference job, while offloading subsequent subtasks to the server. These initial subtasks can be understood as those executed locally on the mobile device. The server, equipped with a graphics processing unit (GPU), can batch similar subtasks for simultaneous processing. This batch processing method significantly reduces the total time consumed by inference across multiple tasks. Figure 13 An example of a reasoning task processing procedure is shown. Figure 13 In the process, mobile device 1 (MD1) locally computes subtasks 1 and 2, then uploads (uplinks) the output data of subtask 2 to the server. Simultaneously, mobile device 2 (MD2) locally computes subtasks 1, 2, and 3, and uploads (uplinks) the output data of subtask 3 to the server. The server receives subtasks 3 and 4 from mobile device 1 and subtask 4 from mobile device 2, and computes them sequentially. During this process, the server merges subtask 4 from mobile devices 1 and 2 into a batch for batch processing (concurrent processing). Finally, the server returns the inference results to the corresponding MDs.

[0188] From a network architecture perspective, this application embodiment considers the following two deployment scenarios: deployment scenario 1 and deployment scenario 2.

[0189] Deployment Scenario 1: Computing Orchestration Nodes are deployed in the core network or Operation Administration and Maintenance (OAR), while Computing Task Nodes are deployed in the base station, i.e., the RAN.

[0190] Deployment Scenario 2: The compute orchestration node is deployed at a specific node of the base station, i.e., the RAN, such as the Centralized Unit (CU); the compute node is deployed at a specific node of the base station, i.e., the RAN, such as the Distributed Unit (DU).

[0191] For different computing nodes in the network, such as Figure 14 As shown, there are two calculation methods: Method 1: Periodic model batch processing. Each batch of inference has a fixed start time. Before startup, the inference tasks for that batch on that node are unloaded, and these tasks are automatically processed after startup. If no tasks are unloaded, the batch remains idle during its original inference cycle. Method 2: Triggered model batch processing. The computing nodes are not set with fixed batch processing time and period. Computing resources are temporarily configured to perform batch inference based on user uninstallation requests.

[0192] exist Figure 14 In the diagram, UE#1's computation task #1 comprises four subtasks. Subtasks #1 and #2 of computation task #1 are processed locally, while subtasks #3 and #4 are offloaded to the batch processing of node #2. UE#2's computation task #2 also comprises four subtasks. Subtasks #1, #2, and #3 of computation task #2 are processed locally, while subtask #4 is offloaded to the batch processing of node #2. Similarly, UE#3's computation task #3 comprises four subtasks. Subtasks #1 and #2 of computation task #3 are processed locally, while subtasks #3 and #4 are offloaded to the batch processing of node #n. The batch processing in nodes #1 and #2 is periodic batch processing, while the batch processing in node #n is triggered batch processing.

[0193] The compute orchestration node is configured with a network compute resource orchestrator. Under this network compute resource orchestrator, the inference tasks that users uninstall are allocated to any batch of inference on any compute node in the network.

[0194] Each inference task needs to be completed within a predetermined latency constraint. The time required for inference includes local processing latency, uplink transmission latency, and server processing latency. To minimize the latency violation rate of inference tasks, carefully designing the offloading strategy for each task is crucial. Furthermore, considering that mobile devices may generate new inference tasks at any time, the server can improve inference efficiency by merging more subtasks into a single processing cycle while waiting for more tasks to be uploaded to the cache.

[0195] The wireless communication methods provided in this application may include, but are not limited to, the following embodiments 1 and 2.

[0196] Example 1: The user triggers a task unloading request based on network-side resource information. In Example 1, the orchestration node periodically broadcasts network resource information. This resource information indicates the status of computing communication resources. After receiving the resource information, the user selects an unloading method based on the broadcast information, triggering a task unloading request. For example... Figure 15 As shown, it includes: S1501, a connection is established between the UE and the computational orchestration node.

[0197] The connection established between the UE and the computing orchestration node may be a control plane (CP) connection, a user plane (UP) connection, or a computing plane (Computing Plane) connection.

[0198] S1502, RAN periodically evaluates the ID information, computing resource status information and radio resource status information of computing nodes.

[0199] Here, RAN is a computing node. Therefore, it can be understood that the computing node periodically evaluates the computing node's ID information, computing resource status information, and radio resource availability information.

[0200] Computing resource status information may include the following information 1 and information 2: Information 1: Batch inference information to be initiated within a future time window (excluding batch processing that has already started or is about to start, as uplink transmission is required for unloading), including the start time, end time, and remaining inference tasks. The remaining inference tasks is the number of n subtasks (subtask(Nn) - subtask N) that can be handled before the end of the batch inference cycle, excluding tasks already reserved. When an inference task includes N subtasks, n can take any of N values ​​from 0 to N-1. To reduce the state space of the remaining inference tasks, n can be selected from several values ​​based on the actual situation. In one example, if an inference task includes 10 subtasks and 3 tasks are currently reserved, then the remaining inference tasks can be any value from 0 to 7. The duration of this time window can be the first duration, and the corresponding time period is the first time period.

[0201] Information 2, the node status of trigger-based batch processing, may also include the server idle status that can provide inference services, and the server computing power assessment (the time required to complete the inference of a fully unloaded task).

[0202] In one example, such as Figure 16 As shown, there are three computing nodes. Computing nodes 1 and 2 use a fixed-periodic processing method, and will periodically start a "batch processing" operation regardless of whether there is an inference task. The start time and duration of the "batch processing" are different for each node. Computing node 3 uses triggered batch processing, and will autonomously perform batch processing based on the unloaded computing tasks. Figure 16 As shown, for periodic batch processing, the computing resource status information may include: period start time, period end time, and inference task remaining capacity. For triggered batch processing, the computing resource status information may include: server idle status and server computing power assessment.

[0203] For compute node 1 and compute node 2, batch processing a1 of compute node 1 and batch processing b1 of compute node 2 have already started. Therefore, these compute nodes only send information about a2 and b2 contained in the time window to the compute orchestration node. The information about a2 and b2 includes the start time of a2 and b2, the latest end time of the batch processing of a2 and b2, and the cycle of each compute node's batch processing (for compute node 1, this is the duration between the start time of a3 and the start time of a2).

[0204] For compute node 3, it is in an idle state, but because it has reserved computing resources for other users' inference tasks, it has proactively scheduled batch processing c1. The information that needs to be sent to the compute orchestration node at this time is the start time and the latest end time of batch processing c1. The latest end time of c1 is the latest value of the deadline for processing c1. Additionally, the server's computing capacity also needs to be sent, i.e., how much the average inference latency will increase for each additional fully unloaded inference task. If no batch processing is scheduled within the time window, only the computing resource status and the server's computing capacity are sent, i.e., the time required for the server to complete a fully unloaded inference task.

[0205] Understandably, for compute nodes that trigger batch processing, if batch processing is scheduled, sending the start time and latest end time of the batch processing indicates that the compute node has unloaded tasks that need to be processed; if no batch processing is scheduled, it indicates that the status is idle.

[0206] The wireless resource status information can be referred to as the wireless resource reserve information. The wireless resource reserve information includes the following information 3 and information 4.

[0207] Information 3: Considering the batch inference processes that will begin within a future time window, predict the average radio resource occupancy / idle rate of the serving cell from the current time to the start time of these batch processes, and the corresponding assessment accuracy (e.g., variance). Serving cell radio resource occupancy / idleness may be represented by an integer value, such as 0 to 100, where 100 means the air interface resource is completely idle / occupied. Serving cell radio resource occupancy / idleness can also be represented by the occupancy of Physical Resource Blocks (PRBs) per cell or per SSB (0-100).

[0208] like Figure 16 As shown, the wireless resource status information includes: wireless resource occupancy information and historical resource occupancy information from the current time to the start time of batch processing.

[0209] In one example, for Figure 16 In the scenario shown, batch processing a1 of computing node 1 and batch processing b1 of computing node 2 have already started. Therefore, these computing nodes only send radio resource information for a2 and b2 within the time window. The start time of the periodic batch processing is fixed, and the uplink transmission radio resource of a2 is the bandwidth usage from the current time to the start time of a2. Since this is reference information for the user, the content not only considers the currently uplink transmission task, or the determined bandwidth usage of tasks that will occupy uplink bandwidth in the future, but also considers the statistical information (variance, prediction accuracy) of historical usage for the user's reference.

[0210] S1503 and RAN send the evaluation results to the computational orchestration node.

[0211] Compute nodes can send evaluation results to compute orchestration nodes via HTTPS messages.

[0212] S1504. The computation orchestration node summarizes the computational resource status and wireless resource status information of all computation nodes.

[0213] Here, the computing resource status and wireless resource reserves of a computing node can be understood as the computing resource information of the computing node (corresponding to the second resource information).

[0214] S1505, The computing orchestration node sends the computing resource status and radio resource status information of the computing node to the UE.

[0215] The combined computing resource status and wireless resource status information of all computing nodes can be referred to as resource information (corresponding to the first resource information).

[0216] The computational orchestration node can send resource information to the UE using either transmission method 1 or transmission method 2.

[0217] Transmission Method 1: The computational orchestration node sends resource information to the UE via the connection established with the UE in S1501. If it's a control plane connection, the information may be NAS information. If it's a user plane connection, the information may be an IP packet. If it's a computing plane connection, the information may be an IP / non-IP message for the protocol dedicated to computing tasks.

[0218] Method 2: The computation orchestration node first sends the resource information to the computation node (e.g., via HTTP / HTTPS messages), and the computation node then sends the resource information to the UE via air interface control plane messages (e.g., SIB / RRC).

[0219] S1506. The UE determines the offloading method for the inference task based on the resource information.

[0220] The UE can generate or predict the generation of inference tasks. Inference tasks may include tasks from third-party applications, such as image rendering, target recognition, and video processing. Inference tasks may also include tasks from communication-related applications, such as user traffic prediction and user trajectory prediction.

[0221] Each inference task can be divided into multiple subtasks. For example, if the inference task is performed by a multi-layer AI model, the subtasks may be one or more layers of that model.

[0222] The UE determines the offloading method for inference tasks based on resource information. The offloading method here can be understood as whether to offload to a compute node or the offloading strategy.

[0223] In this embodiment of the application, the UE can consider whether the latency of the inference task can be met when the current task is unloaded to a computing node based on the computing resource status information.

[0224] In one example, with Figure 16 Taking the reported start times of a2 and b2, the latest end times of batch processing of a2 and b2, and the cycle of batch processing of each node as an example, when the UE uses the received information to consider unloading the inference task to a2, it will compare the latency of the inference task with the latest end time of a2. Although in reality a2 may only batch process a few inference tasks, the actual end time of batch processing of a2 is earlier than the latest end time.

[0225] In one example, with Figure 16 Taking the reported start time and latest end time of c1 and the server's computing power as an example, when the user considers unloading the inference task to the computing node c1 using the received information, the user knows the proportion of the computing volume of different subtasks of the inference task to the total computing volume (which is an imprecise value). The user can roughly decide the split point and assess whether unloading to the computing node c1 will affect the latency of the original task of c1.

[0226] In this embodiment of the application, the UE can consider the success rate of offloading subtasks to access batch processing based on the availability of radio resources.

[0227] In one example, with Figure 16 Taking the reported wireless resource availability information of a3 as an example, if batch processing of a3 will start after 1 second, then to access the batch processing of a3, it is necessary to consider the success rate of sending the relevant computational offloading subtasks to node 1 within 1 second.

[0228] In this embodiment, the UE can decide the offloading strategy independently, or it can report computing task information to the computing orchestration node, which will then decide the offloading strategy.

[0229] When the UE decides its own offloading strategy, it can be as follows: Figure 17 As shown, it includes: S1506A, UE determines the offloading strategy.

[0230] The offloading policy indicates to the UE which computational subtasks will be completed locally and which will be offloaded to the compute node. This offloading policy corresponds to the second policy. The offloading policy may be represented in the following form: The UE selects the compute node information, such as compute node ID / Index, RAN ID, cell ID, etc.; TaskID / Index + a list of subtask IDs / Index to be computed locally, or TaskID / Index + a list of subtask IDs / Index to be computed in the compute task node, or TaskID / Index + a list of subtask IDs / Index to be computed locally + a list of subtask IDs / Index to be computed in the compute task node, or TaskID / Index + subtask split ID or index (a subtask split ID / Index representing the split point, e.g., a subtask with ID / Index below (or equal to) the split ID / Index will be executed at the UE locally); the start and end times of one or more compute tasks or compute subtasks on the compute task node.

[0231] In this embodiment, the computing node information selected by the UE is optional.

[0232] S1507A, UE sends the offload policy to the computation orchestration node.

[0233] Here, the UE sends the determined offloading policy as a suggested offloading policy to the computation orchestration node.

[0234] The UE can also send computing task information to the computing orchestration node.

[0235] S1508A, the computational orchestration node and RAN node interact to determine the binary decision to be fed back.

[0236] Interaction between compute orchestration nodes and RAN nodes can be used to assess current compute resources and determine a binary decision to be fed back based on the actual situation of these resources. This binary decision indicates whether to permit or deny the offloading policy. The compute orchestration node can also decide which compute node will undertake the compute offloading task. In one example, if the UE's current serving RAN node is selected as the compute node supporting the UE to undertake the compute offloading task, the UE and the serving RAN may further interact on the compute offloading task through the UE-RAN end-to-end protocol layer (none-IP). In another example, if a non-serving RAN node is selected as the compute node supporting the UE to undertake the compute offloading task, the UE is first handovered to the corresponding RAN node, and further interact on the compute offloading task through the UE-RAN end-to-end protocol layer (none-IP). In yet another example, the compute orchestration node provides the IP address information of the selected compute node to undertake the compute offloading task to the UE, and the UE will further interact on the compute offloading task through the given IP address.

[0237] S1509A, the computational orchestration node feeds back binary decisions to the UE.

[0238] Binary decision is used to indicate whether the UE is allowed to send offload policies, where "allow" can also be described as "permitted" and "disallowed" can also be described as "prohibited".

[0239] When the computation orchestration node determines the computation node to undertake the computation offloading task, the computation orchestration node can also send the computation node information (corresponding to the target radio access node) to the UE to carry out the computation offloading task.

[0240] S1510A: The UE can choose to continue the current offloading policy, adjust the current offloading policy, or terminate the current offloading policy based on the binary decision.

[0241] When the UE reports computing task information to the computing orchestration node, and the computing orchestration node determines the offload strategy, such as Figure 18As shown, it includes: S1506B: The UE decides whether to unload the inference task based on the resource information.

[0242] Based on the resource information (corresponding to the first resource information), the user decides locally whether to uninstall the inference task. If the user chooses to perform the computation entirely locally, the computation task information will no longer be reported; otherwise, S1507B will be executed.

[0243] In one example, the UE determines whether to perform computation offload based on the first resource information. If there are computing and radio resources available on the network, then offload is performed, triggering an offload request.

[0244] S1507B, UE sends computing task information to the computing orchestration node.

[0245] The computation task information here corresponds to the task information. The UE can also send service quality, computing power requirements, and local computing resource status to the computation orchestration node. The computation task information can be included in the offload request.

[0246] The computing task information sent by the UE to the computing orchestration node may include the following information related to the computing task (corresponding to the first task information): a) computing task ID / Index; b) computing task description, such as application description; c) complete AI model parameters or files used by the computing task; d) computing task QoS requirements, such as latency requirements (e.g., inmilli-seconds or seconds) and overall computing power requirements (e.g., number of processing units, number of FLOPS, or an abstract value of computing power level 1-100).

[0247] The computing task information sent by the UE to the computing orchestration node may include the following information related to the computing subtask (corresponding to the second task information): a) Subtask ID / Index; b) Description of the computing subtask, such as the type of model layer; c) Complete AI model parameters or files used by the computing subtask; d) QoS requirements of the computing subtask, such as latency requirements (e.g., in milliseconds or seconds) and overall computing power requirements (e.g., number of processing units, number of FLOPS, or an abstract value of computing powerlevel 1-100).

[0248] The UE can also send local computing resource status information to the computing orchestration node.

[0249] S1508B, the computation orchestration node, and the RAN node interact with each other to determine the offloading strategy.

[0250] The uninstallation strategy here corresponds to the third strategy.

[0251] The computation orchestration node and the RAN node interact to evaluate the current network computing resources. Based on the actual situation of the current network computing resources, they will decide on the UE offloading strategy. Here, the computation orchestration node can send the determined third strategy to the RAN node, which will then evaluate it based on the actual network computing resources to determine the final offloading strategy (the corresponding first strategy).

[0252] Here, the computation orchestration node can also decide which computation node (corresponding to the target radio access node) will undertake the computation offloading task. In one example, if the UE's current serving RAN node is selected as the computation node supporting the UE to undertake the computation offloading task, the UE and the serving RAN may further interact regarding the computation offloading task through the UE-RAN end-to-end protocol layer (none-IP). In another example, if a non-serving RAN node is selected as the computation node supporting the UE to undertake the computation offloading task, the UE is first handovered to the corresponding RAN node, and further interacts regarding the computation offloading task through the UE-RAN end-to-end protocol layer (none-IP). In yet another example, the computation orchestration node will provide the UE with the IP address information of the selected computation node undertaking the computation offloading task, and the UE will further interact regarding the computation offloading task through the given IP address.

[0253] S1509B: The computation orchestration node sends the offload policy to the UE.

[0254] S1510B and UE perform inference tasks based on the offloading strategy.

[0255] exist Figure 18 In the communication method shown, the network side determines the user's offloading strategy based on the computing resources and sends it to the user, while reserving computing resources. The user then performs the offloading of the inference task according to the offloading strategy issued by the network side.

[0256] In this embodiment, S1506 may include S1506A or S1506B.

[0257] Example 2: Offloading Strategy for Collaborative Decision-Making and Reasoning Tasks of Computation Orchestration Nodes and Computation Nodes In Example 2, the computation orchestration node evaluates the current computing resource information based on the information reported by the user in S1507A or S1507B of Example 1 and responds accordingly. Specifically, in S1506A of Example 1, the UE reports its self-generated offload policy, and the computation orchestration node responds with acceptance / rejection information; in S1506B of Example 1, the UE provides computing task information, and the computation orchestration node responds with the generated offload policy.

[0258] The evaluation process in S1508A or S1508B is as follows: Figure 19 As shown, it includes: S8-1, The compute orchestration node sends the offload policy to the compute node.

[0259] Upon receiving offload policies or computing task information reported by the UE, the computing orchestration node sends the relevant offload policies to the candidate computing task nodes. Figure 17 In the illustrated process, the offloading strategy is generated by the UE (i.e., the second strategy). Figure 18 In the illustrated process, the unloading strategy is generated by the computation orchestration node (i.e., the third strategy). The computation orchestration node may send computation task information to candidate nodes, or it may choose not to send computation task information to candidate computation nodes.

[0260] S8-2. Candidate computing nodes assess their preference for accepting computing tasks based on their own computing resources and air interface resource status.

[0261] For example, if the current RAN node has a low utilization rate of computing resources and a low utilization rate of air interface resources, it is more likely to accept computing offload tasks.

[0262] S8-3, Candidate computing nodes feed back their acceptance preference for computing to the computing orchestration nodes.

[0263] Acceptance preference can be represented in ways including but not limited to the following: Representation method 1, Boolean indication, such as yes / no; Representation method 2, percentage value, such as 1 to 100.

[0264] S8-4. The computation orchestration node integrates the acceptance preferences of multiple candidate computation nodes for the inference task and determines the final unloading strategy.

[0265] The final offloading strategy corresponds to the first strategy. The computation orchestration node, by considering the acceptance preferences of multiple candidate computation nodes for the inference task, can also determine the final computation node to undertake the computation offloading task. In this embodiment, the computation orchestration node can send the same computation subtask to multiple computation nodes for querying. If, for the same computation subtask, computation node 1 has a higher acceptance rate than computation node 2, the computation subtask may ultimately be offloaded to computation node 1.

[0266] After the computation orchestration nodes determine the final computation offloading strategy, in Figure 17 The system feeds back binary decision-making to the UE. Figure 18 The system feeds back the offloading strategy to the UE.

[0267] Optionally, the compute orchestration node may also send the final offload decision to the selected compute task node.

[0268] After the UE receives feedback (binary feedback or offload strategy) from the computation orchestration node, when it completes local computation and uploads intermediate variables, the network side has... Figure 20 and Figure 21 The two receiving methods are shown.

[0269] like Figure 20 As shown, the UE first sends the intermediate information of the model inference to the computation orchestration node, which then forwards it to the corresponding computation task node. After completing the computation offloading task, the computation task node returns the computation result to the UE through the computation orchestration node.

[0270] like Figure 21 As shown, the UE communicates directly with the computing task node and exchanges intermediate computing information and computing results.

[0271] In the wireless communication method provided in this application embodiment, for the UE: - receive information related to the offloading of computing tasks (e.g., network computing resources and air interface radio resources) from the computing orchestration node; generate a computing offloading strategy suggestion or computing task information for the UE and send it to the computing orchestration node; receive the final computing offloading strategy decision from the computing orchestration node.

[0272] In some embodiments, the computing resource status information received by the UE includes batch inference information to be started within a future time window (considering that uplink transmission is still required for offloading, batch processing that has already started or is about to start is not considered), including the start time, end time, and inference task reserve. The inference task reserve is the number of inference tasks (subtask(Nn) - subtaskN) that can be offloaded from n subtasks before the end of the batch inference cycle, excluding tasks that have already been reserved.

[0273] In some embodiments, the air interface radio resource information received by the UE includes, considering batch inference that will be initiated within a future time window, the prediction of the average radioresource occupancy / idle rate of the serving cell from the current time to the start time of these batches, and the corresponding evaluation accuracy (e.g., variance). The representation of the serving cell radioresource occupancy / idleness may include: Method 1, an integer value, e.g., 0-100, where 100 means the air interface resource is completely idle / occupied; Method 2, representing it using the occupancy status of the Physical Resource Block (PRB) per cell or per SSB (0-100).

[0274] In some embodiments, if the UE’s current serving RAN node is selected as the computing node supporting the UE to undertake computing offloading tasks, the UE and the serving RAN may further interact on computing offloading tasks through the UE-RAN end-to-end protocol layer (none-IP).

[0275] In some embodiments, if a non-serving RAN node is selected as the computing node supporting the UE to undertake the computing offloading task, the UE is first handover switched to the corresponding RAN node, and the computing offloading task is further interacted through the UE-RAN end-to-end protocol layer (none-IP).

[0276] In some embodiments, the compute orchestration node provides the IP address information of the selected compute node to the UE, and the UE will further interact with the compute offloading task through the given IP address.

[0277] In this embodiment of the application, when the RAN is used as a computing task node, the following method is used to provide computing offloading services to the UE: on the one hand, the UE autonomously generates a suggested computing offloading strategy based on the computing resource status and radio resource availability of the computing task RAN ​​node provided by the network, and provides it to the computing orchestration node for reference; on the other hand, the computing task RAN ​​node and the computing orchestration node interact and negotiate the final computing offloading strategy, including selecting a suitable computing task RAN ​​node.

[0278] Thirdly, to implement the above-mentioned wireless communication method, an embodiment of this application provides a device 2200 (terminal device or computing orchestration node) that may include at least one processor 2201 and at least one transceiver 2202 coupled to the at least one processor 2201. The transceiver 2202 may include at least one separate receiving circuit system and a transmitting circuit system, or at least one integrated receiving circuit system and transmitting circuit system. The at least one processor 2201 may be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP), or a field-programmable gate array (FPGA), etc.

[0279] According to some embodiments of this application, when device 2200 is a terminal device, processor 2201 may be referred to as a first processor, and transceiver 2202 may be referred to as a first transceiver.

[0280] The first processor is configured to: receive first resource information transmitted by a computation orchestration node via the first transceiver, the first resource information including second resource information of at least one wireless access node, the second resource information representing the computational resource status and wireless resource status of the wireless access node related to the inference task within a first time period; determine a first strategy for a first inference task to be executed based on the first resource information, the first inference task including multiple subtasks, the first strategy being used to instruct all or part of the subtasks included in the first inference task to be computed by a target wireless access node, the target wireless access node being one of the at least one wireless access node; and execute the first inference task based on the first strategy.

[0281] In some embodiments, the second resource information includes computing resource status information; the computing resource status information includes at least one of the following: idle status indication information, the idle status information indicating whether the wireless access node is idle during the first time period; the start time of the first batch of processing within the first time period; the end time of the first batch of processing; the period of the first batch of processing; the remaining capacity of subtasks that the first batch of processing can handle; and computing power evaluation information.

[0282] In some embodiments, the second resource information includes wireless resource status information, which includes wireless resource occupancy rate or wireless resource idle rate between a first time and a second time. The first time is the start time of the first time period, and the second time is the start time of the first batch of processing within the first time period.

[0283] In some embodiments, the first processor is configured to: receive the first resource information sent by the computation orchestration node via a connection with the computation orchestration node; the connection includes one of the following: a control plane connection, a user plane connection, and a computation plane connection.

[0284] In some embodiments, the first processor is configured to receive the first resource information sent by the computing orchestration node via an air interface control plane message sent by the serving radio access node of the terminal device.

[0285] In some embodiments, the first processor is configured to: generate a second policy based on the first resource information; send the second policy to the computation orchestration node; receive policy feedback information for the second policy sent by the computation orchestration node; and determine the first policy based on the policy feedback information and the second policy.

[0286] In some embodiments, the second strategy includes at least one of the following: a subtask identifier of at least one first subtask processed by the terminal device in the first inference task; a subtask identifier of at least one second subtask processed by the wireless access node in the first inference task; and a subtask segmentation identifier, the segmentation identifier being used to identify the at least one first subtask and the at least one second subtask in the first inference task.

[0287] In some embodiments, the second strategy further includes at least one of the following: the node identifier of the first candidate wireless access node; the start time and / or end time of the wireless access node processing the first inference task or the at least one second subtask.

[0288] In some embodiments, the policy feedback information includes at least one of the following: first indication information, which indicates whether the second policy is allowed; the node identifier or IP address of the target wireless access node.

[0289] In some embodiments, the first processor is configured to: determine, based on the first resource information, whether all or some of the subtasks in the first inference task are to be executed by the wireless access node; if all or some of the subtasks in the first inference task are to be executed by the wireless access node, send task information of the first inference task to the computation orchestration node; and receive the first policy sent by the computation orchestration node.

[0290] In some embodiments, the first strategy includes the node identifier or IP address of the target wireless access node and at least one of the following: a subtask identifier of at least one first subtask processed by the terminal device in the first inference task; a subtask identifier of at least one second subtask processed by the wireless access node in the first inference task; and a segmentation identifier for segmenting the at least one first subtask and the at least one second subtask in the first inference task.

[0291] In some embodiments, the first processor is configured to: send subtask information of at least one second subtask processed by the target wireless access node in the first inference task to the target wireless access node when the target wireless access node is the serving wireless access node of the terminal device.

[0292] In some embodiments, the first processor is configured to: switch to the target wireless access node when the target wireless access node is a non-serving wireless access node of the terminal device; and send to the target wireless access node subtask information of at least one second subtask processed by the target wireless access node in the first inference task.

[0293] In some embodiments, the first processor is configured to: send subtask information of at least one second subtask processed by the target wireless access node in the first inference task to the target wireless access node via the IP address of the target wireless access node.

[0294] According to some embodiments of this application, when device 2200 is a computing orchestration node, processor 2201 may be referred to as a second processor, and transceiver 2202 may be referred to as a second transceiver.

[0295] The second processor is configured to: receive, via the second transceiver, first resource information transmitted by at least one wireless access node, the first resource information including second resource information of the at least one wireless access node, the second resource information representing the computational resource status and wireless resource status of the wireless access node related to the inference task within a first time period; and transmit, via the second transceiver, the first resource information to a terminal device, the first resource information being used to determine a first strategy for a first inference task to be executed, the first inference task including multiple subtasks, the first strategy being used to instruct all or part of the subtasks included in the first inference task to be computed by a target wireless access node, the target wireless access node being one of the at least one wireless access node.

[0296] In some embodiments, the second resource information includes computing resource status information; the computing resource status information includes at least one of the following: idle status indication information, the idle status information indicating whether the wireless access node is idle during the first time period; the start time of the first batch of processing within the first time period; the end time of the first batch of processing; the period of the first batch of processing; the remaining capacity of subtasks that the first batch of processing can handle; and computing power evaluation information.

[0297] In some embodiments, the second resource information includes wireless resource status information, which includes wireless resource occupancy rate or wireless resource idle rate between a first time and a second time. The first time is the start time of the first time period, and the second time is the start time of the first batch of processing within the first time period.

[0298] In some embodiments, the second processor is configured to: send the first resource information to the terminal device via a connection with the terminal device; the connection includes one of the following: a control plane connection, a user plane connection, and a compute plane connection.

[0299] In some embodiments, the second processor is configured to: send the first resource information to the serving radio access node of the terminal device, and send the first resource information via an air interface control plane message sent by the serving radio access node of the terminal device.

[0300] In some embodiments, the second processor is configured to: receive a second policy sent by the terminal device; send policy feedback information for the second policy to the terminal device; the policy feedback information and the second policy are used to determine the first policy.

[0301] In some embodiments, the second strategy includes at least one of the following: a subtask identifier of at least one first subtask processed by the terminal device in the first inference task; a subtask identifier of at least one second subtask processed by the wireless access node in the first inference task; and a subtask segmentation identifier, the segmentation identifier being used to identify the at least one first subtask and the at least one second subtask in the first inference task.

[0302] In some embodiments, the second strategy further includes at least one of the following: the node identifier of the first candidate wireless access node; the start time and / or end time of the wireless access node processing the first inference task or the at least one second subtask.

[0303] In some embodiments, the policy feedback information includes at least one of the following: first indication information, which indicates whether the second policy is allowed; the node identifier or IP address of the target wireless access node.

[0304] In some embodiments, the second processor is configured to: send the second policy to at least one second candidate wireless access node; receive an acceptance preference for the second policy sent by the at least one second candidate wireless access node; and determine the target wireless access node from the at least one second candidate wireless access node based on the received acceptance preference.

[0305] In some embodiments, the first processor is configured to: receive task information of the first inference task sent by the terminal device; determine the first strategy based on the task information; and send the first strategy to the terminal device.

[0306] In some embodiments, the first strategy includes the node identifier or IP address of the target wireless access node and at least one of the following: a subtask identifier of at least one first subtask processed by the terminal device in the first inference task; a subtask identifier of at least one second subtask processed by the wireless access node in the first inference task; and a segmentation identifier for segmenting the at least one first subtask and the at least one second subtask in the first inference task.

[0307] In some embodiments, the second processor is configured to: generate a third policy based on the task information; send the third policy to at least one second candidate radio access node; receive an acceptance preference for the third policy sent by the at least one second candidate radio access node; determine the target radio access node from the at least one second candidate radio access node based on the received acceptance preference; and determine the first policy based on the target radio access node and the third policy.

[0308] The descriptions of the above device embodiments are similar to those of the above method embodiments, and have similar beneficial effects. For technical details not disclosed in the device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0309] It should be noted that, in the embodiments of this application, if the above-described wireless communication method is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the related technology, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware and software combination.

[0310] Fourthly, to implement the above-mentioned wireless communication method, embodiments of this application provide an electronic device, including a memory and a processor. The memory stores a computer program that can run on the processor, and when the processor executes the program, it implements the steps in the wireless communication method provided in the above embodiments.

[0311] Fifthly, embodiments of this application provide a storage medium, namely a computer-readable storage medium, on which a computer program is stored, which, when executed by a processor, implements the steps in the wireless communication method provided in the above embodiments.

[0312] It should be noted that the descriptions of the storage medium and device embodiments above are similar to those of the method embodiments above, and have similar beneficial effects. For technical details not disclosed in the storage medium and device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0313] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in some embodiments" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0314] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0315] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0316] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0317] In addition, each functional unit in the various embodiments of this application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0318] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0319] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0320] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A terminal device, the terminal device comprising a first transceiver; and A first processor, coupled to the first transceiver; the first processor is configured to: The first resource information sent by the computation orchestration node is received via the first transceiver. The first resource information includes second resource information of at least one wireless access node. The second resource information characterizes the computational resource status and wireless resource status of the wireless access node related to the inference task in a first time period. Based on the first resource information, a first strategy is determined for the first inference task to be executed. The first inference task includes multiple subtasks. The first strategy is used to instruct all or part of the subtasks included in the first inference task to be calculated by a target wireless access node, wherein the target wireless access node is one of the at least one wireless access node. Based on the first strategy, perform the first inference task.

2. The terminal device according to claim 1, wherein the second resource information includes computing resource status information; the computing resource status information includes at least one of the following: Idle status indication information, the idle status information indicating whether the wireless access node is idle during the first time period; The start time of the first batch of processing within the first time period; The end time of the first batch of processing; The cycle of the first batch of processing; The remaining capacity of the first batch of processing to handle sub-tasks; Computing power assessment information.

3. The terminal device according to claim 1, wherein the second resource information includes wireless resource status information, the wireless resource status information including wireless resource occupancy rate or wireless resource idle rate between a first time and a second time, the first time being the start time of the first time period, and the second time being the start time of the first batch of processing within the first time period.

4. The terminal device according to claim 1, wherein the first processor is configured to: The system receives the first resource information sent by the computation orchestration node through a connection with the computation orchestration node; the connection includes one of the following: control plane connection, user plane connection, and computation plane connection.

5. The terminal device according to claim 1, wherein the first processor is configured to: The terminal device receives the first resource information sent by the computing and orchestration node through the air interface control plane message sent by the serving wireless access node.

6. The terminal device according to claim 1, wherein the first processor is configured to: Based on the first resource information, a second strategy is generated; Send the second strategy to the computation orchestration node; Receive policy feedback information for the second policy sent by the computation orchestration node; Based on the policy feedback information and the second policy, the first policy is determined.

7. The terminal device according to claim 1, wherein the first processor is configured to: Based on the first resource information, determine whether the wireless access node should execute all or part of the sub-tasks in the first inference task; In the case where all or part of the subtasks in the first inference task are executed by the wireless access node, the task information of the first inference task is sent to the computation orchestration node. Receive the first strategy sent by the computation orchestration node.

8. The terminal device according to claim 1, wherein the first processor is configured to: If the target wireless access node is the serving wireless access node of the terminal device, subtask information of at least one second subtask processed by the target wireless access node in the first inference task is sent to the target wireless access node.

9. A computational orchestration node, the computational orchestration node comprising a second transceiver; and A second processor, coupled to the second transceiver; the second processor is configured to: The transceiver receives first resource information sent by at least one wireless access node, the first resource information including second resource information of the at least one wireless access node, the second resource information representing the computing resource status and wireless resource status of the wireless access node related to the inference task in a first time period. The first resource information is sent to the terminal device via the second transceiver. The first resource information is used to determine a first strategy for the first inference task to be executed. The first inference task includes multiple subtasks. The first strategy is used to instruct all or part of the subtasks included in the first inference task to be calculated by a target wireless access node, which is one of the at least one wireless access node.

10. A wireless communication method applied to a terminal device, the method comprising: Receive first resource information sent by a computation orchestration node, the first resource information including second resource information of at least one wireless access node, the second resource information representing the computational resource status and wireless resource status of the wireless access node related to the inference task in a first time period; Based on the first resource information, a first strategy is determined for the first inference task to be executed. The first inference task includes multiple subtasks. The first strategy is used to instruct all or part of the subtasks included in the first inference task to be calculated by a target wireless access node, wherein the target wireless access node is one of the at least one wireless access node. Based on the first strategy, perform the first inference task.

Images