Communication method and communication apparatus

Abstract

Provided in the present application are a communication method and a communication apparatus. The method comprises: a terminal device sending a first request and first data information to an access network device, wherein the first request is used for requesting that the access network device executes inference of a first task; then, the access network device performing speculative inference of the first task on the basis of the first data information so as to obtain first speculative inference information, and sending the first speculative inference information to a server; and the server being capable of performing inference of a second task on the basis of the first speculative inference information so as to obtain first inference information. The method helps to reduce the overall inference delay and improve the inference efficiency.

Description

Communication method and communication apparatus

[0001] The present application claims priority to the Chinese patent application No. 202411299781.9, filed on September 14, 2024, with the State Intellectual Property Office of China, and entitled "A communication method and communication apparatus", the whole content of which is incorporated herein by reference. TECHNICAL FIELD

[0002] The present application relates to the field of communication technology, and in particular to a communication method and communication apparatus. BACKGROUND

[0003] At present, artificial intelligence (AI) large models represented by ChatGPT, with their outstanding natural language and multi-modal understanding capabilities and strong support for generative AI, are showing a flourishing development situation and have achieved remarkable results in the fields of image, text, voice and recommendation system. AI large models go through two stages of training and inference from construction to actual landing application, wherein: the training stage refers to the process of constructing a machine learning model, which learns patterns and features from a large amount of data to optimize model performance, thus requiring a large amount of data and computing power support, which can be completed on a server; the inference stage refers to the process of using a trained model to process tasks, which can be performed in special scenarios or with limited computing power, such as in a cloud data center.

[0004] For inference tasks, a very high requirement for latency is generally proposed. If the inference latency is too long, the task processing efficiency will be greatly reduced, which will seriously affect the user experience in the application scenario of large models. Therefore, how to reduce the inference latency is a problem to be solved at present. SUMMARY

[0005] The embodiments of the present application provide a communication method and communication apparatus, based on the method described in the present application, which is beneficial to reduce the inference latency and improve the inference efficiency.

[0006] In a first aspect, the embodiments of the present application provide a communication method, which can be applied to a network side, such as an access network device of the network side, a module (such as a circuit, a chip or a chip system, etc.) in the access network device, or a logic node, a logic module or software capable of realizing all or part of the functions of the access network device. Taking the case that the method is applied to the access network device, in the method: the access network device receives a first request and first data information from a terminal device, the first request being used to request the access network device to perform inference of a first task; further, the access network device performs guess inference of the first task based on the first data information to obtain first guess inference information; and then, the access network device sends the first guess inference information to a server.

[0007] With the above method, when the terminal device has an inference task, considering that the computing power of the terminal device is limited, the terminal device can allocate the inference task to the access network device and the server for implementation. The access network device can deploy a guess inference model, and the server can deploy an inference model. After the access network device receives the first data information, without waiting for the terminal device to report all the data information and then performing inference of the first task, the access network device can directly infer the inference result according to the limited data information by using the guess inference model; or the access network device can first infer the possible subsequent data information according to the first data information by using the guess inference model, and then perform inference of the first task by using the first data information and the inferred possible subsequent data information; the entire process can be considered as a guess inference process, so that the first guess inference information can be obtained, and then the first guess inference information is sent to the server for inference of the second task. In this way, it is beneficial to reduce the overall inference delay and improve the inference efficiency.

[0008] In a possible design, the first data information is part of the data information corresponding to the first task.

[0009] It can be understood that, due to limited air interface resources and long complete data transmission time, the first data information herein is only part of the data information related to the first task reported by the terminal device. Considering that some tasks can be completed by using only part of the data information, the access network device can perform inference of the first task without waiting for the terminal device to report all the data information, which is beneficial to reduce the inference delay.

[0010] In a possible design, the method further includes: receiving, by the access network device, second data information from the terminal device, the second data information being other data information in the data information corresponding to the first task except the first data information.

[0011] It can be understood that, due to limited air interface resources, the terminal device can send different proportions of data information at different sending durations; correspondingly, the access network device also receives different proportions of data information at different receiving durations. Therefore, after the terminal device sends the first data information to the access network device, the terminal device can continue to send other data information (such as second data information) in the data information corresponding to the first task except the first data information to the access network device. The second data information herein can be all the data information remaining in the data information corresponding to the first task except the first data information, or can be part of the data information remaining in the data information corresponding to the first task except the first data information, which is not limited herein.

[0012] In a possible design, the method further includes: performing, by the access network device, guessed reasoning of the first task based on the first data information and the second data information, to obtain second guessed reasoning information; and sending, by the access network device, first indication information to the server if a similarity between the second guessed reasoning information and the first guessed reasoning information is less than a first preset threshold, where the first indication information is used to indicate re-executing the reasoning of the task, and the first indication information includes the second guessed reasoning information.

[0013] It can be understood that, since the second guessed reasoning information is determined based on more data information, the accuracy of the second guessed reasoning information is higher than that of the first guessed reasoning information, and thus the second guessed reasoning information can be used to verify whether the first guessed reasoning information determined previously is accurate. Since the time required for verification is much less than the time required for reasoning, once the verification is successful, the overall reasoning delay can be greatly reduced; if the verification fails, performing a rollback action also does not affect the overall reasoning delay.

[0014] In a possible design, the method further includes: sending, by the access network device, second indication information to the terminal device, where the second indication information is used to indicate increasing the priority of data information transmission corresponding to the first task.

[0015] It can be understood that, the higher the priority of data information transmission is, the higher the transmission speed or efficiency is, and thus when the first guessed reasoning information fails to pass the verification, in order to enable the terminal device to report the remaining data information as soon as possible, the access network device can indicate the terminal device to increase the priority of data information transmission corresponding to the first task.

[0016] In a possible design, the method further includes: sending, by the access network device, third indication information to the terminal device if a data amount corresponding to the first data information is greater than or equal to a second preset threshold, or a reasoning accuracy corresponding to the first data information is greater than or equal to a third preset threshold, or a ratio between the data amount corresponding to the first data information and a data amount of all data information corresponding to the first task is greater than or equal to a ninth preset threshold.

[0017] The third indication information is used to indicate not sending second data information, where the second data information is other data information than the first data information in the data information corresponding to the first task; or the third indication information is used to indicate reducing the priority of data information transmission corresponding to the first task.

[0018] It can be understood that, considering that some tasks only need part of the data information to complete the reasoning, and some reasoning tasks may have accuracy requirements, under certain conditions, in order to save power consumption and reduce delay, the access network device can instruct the terminal device not to continue to send the subsequent data information, so as to avoid resource waste.

[0019] In a possible design, before the access network device performs the speculative reasoning of the first task based on the first data information, the method further includes: if a first time length is greater than a fourth preset threshold, or a quality of service (QoS) corresponds to a first preset priority, the access network device starts the speculative reasoning function; the first time length is a time length of receiving uplink data estimated based on network load information, and the fourth preset threshold is associated with a reasoning time delay threshold indicated by the terminal device.

[0020] It can be understood that the access network device can determine whether to start the speculative reasoning function according to current network load information, or use the QoS information as a condition for starting the speculative reasoning function, to assist in reducing the reasoning time delay.

[0021] In a possible design, the method further includes: the access network device receives fourth indication information from the server, where the fourth indication information is used to indicate a window length corresponding to the first speculative reasoning information. Based on this manner, the flexibility of the window length corresponding to the first speculative reasoning information is improved.

[0022] In a possible design, the first request is further used to request the server to perform reasoning of a second task; when the access network device sends the first speculative reasoning information to the server, the implementation manner can be: the access network device sends a second request and the first speculative reasoning information to the server, where the second request is used to request the server to perform reasoning of the second task.

[0023] It can be understood that the terminal device requests the server to perform reasoning of the second task in addition to requesting the access network device to perform reasoning of the first task, and therefore the access network device subsequently sends the request of the terminal device to the server when sending the first speculative reasoning information to the server, that is, requests the server to perform reasoning of the second task.

[0024] In a possible design, the first task is associated with a capability of the access network device, and the second task is associated with a capability of the server. Based on this manner, the adaptation degree of the task and the device capability can be ensured.

[0025] In a second aspect, an embodiment of the present application provides a communication method, which can be applied to a network side, for example, a server of the network side or a communication module / processing module in the server, or a circuit or chip (such as a modem chip, also known as a baseband chip, or a system on chip (SoC) chip or a system in package (SIP) chip containing a modem core, or a circuit or chip responsible for processing functions in the server (such as a graphics processing unit (GPU), an artificial intelligence (AI) processor, or an application specific integrated circuit (ASIC)) responsible for communication functions in the server, for example, a server. In the method, the server receives first guess inference information from an access network device; then, based on the first guess inference information, the server performs inference of a second task to obtain first inference information.

[0026] In the embodiments of the present application, the beneficial effects of the possible implementation manners of the second aspect can refer to the beneficial effects of the possible implementation manners of the first aspect, which will not be repeated here.

[0027] In a possible design, the method further includes: the server sends the first inference information to the terminal device.

[0028] In a possible design, the method further includes: the server receives first indication information from the access network device, the first indication information being used to indicate re-execution of inference of a task, the first indication information including second guess inference information; further, based on the second guess inference information, the server re-performs inference of the second task to obtain second inference information; and then, the server sends the second inference information to the terminal device.

[0029] In a possible design, the method further includes: the server sends fourth indication information to the access network device, the fourth indication information being used to indicate a window length corresponding to the first guess inference information.

[0030] In a possible design, when the server receives the first guess inference information from the access network device, the specific implementation manner can be: the server receives a second request and the first guess inference information from the access network device, the second request being used to request the server to perform inference of a second task.

[0031] In a possible design, the second task is associated with a capability of the server.

[0032] In a third aspect, an embodiment of the present application provides a communication method, which can be applied to a terminal device side, for example, a terminal device or a communication module / processing module in the terminal device, or a circuit or chip responsible for communication function in the terminal device (such as a modem chip, also known as a baseband chip, or a system on chip (SoC) chip or a system in package (SIP) chip containing a modem core), or a circuit or chip responsible for processing function in the terminal device (such as a graphics processing unit (GPU), an artificial intelligence (AI) processor, or an application specific integrated circuit (ASIC)), for example, the method is applied to a terminal device, in which: the terminal device sends a first request and first data information to an access network device, the first request is used to request the access network device to perform inference of a first task.

[0033] In the embodiments of the present application, the beneficial effects of the possible implementation manners of the third aspect can refer to the beneficial effects of the possible implementation manners of the first aspect, which will not be repeated here.

[0034] In a possible design, the first data information is part of the data information corresponding to the first task.

[0035] In a possible design, the method further includes: the terminal device sends second data information to the access network device, the second data information being other data information in the data information corresponding to the first task except the first data information.

[0036] In a possible design, the method further includes: the terminal device receives second indication information from the access network device, the second indication information being used to indicate that the priority of the data information corresponding to the first task is increased.

[0037] In a possible design, the method further includes: the terminal device receives third indication information from the access network device, the third indication information being used to indicate that the second data information is not sent, the second data information being other data information in the data information corresponding to the first task except the first data information; or, the third indication information indicates that the priority of the data information corresponding to the first task is decreased.

[0038] In a possible design, the method further includes: the terminal device receives first inference information or second inference information from a server.

[0039] In a possible design, the first request is further used to request the server to perform inference of a second task; the first task is associated with a capability of the access network device, and the second task is associated with a capability of the server.

[0040] In a fourth aspect, an embodiment of the present application provides a communication method, which can be applied to a terminal device side, for example, a terminal device or a communication module / processing module in the terminal device, or a circuit or chip responsible for communication function in the terminal device (such as a modem chip, also known as a baseband chip, or a system on chip (SoC) chip or system in package (SIP) chip containing a modem core), or a circuit or chip responsible for processing function in the terminal device (such as a graphics processing unit (GPU), an artificial intelligence (AI) processor, or an application specific integrated circuit (ASIC)), for example, the method is applied to a terminal device, in which: the terminal device sends a third request to an access network device, the third request being used to request data information corresponding to a third task; further, the terminal device receives third data information from the access network device, the third data information being part of the data information corresponding to the third task; then, the terminal device performs guess inference of the third task based on the third data information, to obtain third guess inference information.

[0041] With the above method, when the terminal device has an inference task, if the computing power of the terminal device is sufficient, the terminal device can also perform the inference task by itself. The terminal device can deploy a guess inference model. Due to limited air interface resources and long complete data transmission time, the third data information is only part of the data information related to the third task sent by the access network device. After receiving the third data information, the terminal device does not need to wait for the access network device to send all the data information before performing inference of the third task, but can directly infer the inference result according to the limited data information by using the guess inference model; or the terminal device can first infer the possible data information after the third data information by using the guess inference model, and then perform inference of the third task by using the third data information and the inferred possible data information after the third data information; the whole process can be considered as a guess inference process, to obtain the third guess inference information, and the terminal device can perform a task according to the third guess inference information, which is beneficial to reduce the overall inference delay and improve the inference efficiency.

[0042] In a possible design, the method further includes: the terminal device performs a task based on the third guess inference information.

[0043] In a possible design, the method further includes: the terminal device receives fourth data information from the access network device, the fourth data information being other data information in the data information corresponding to the third task except the third data information.

[0044] It can be understood that, due to the limited air interface resources, the access network device can send different proportions of data information at different sending durations; correspondingly, the terminal device will also receive different proportions of data information at different receiving durations. Therefore, after the access network device sends the third data information to the terminal device, it can continue to send other data information (such as fourth data information) in addition to the third data information to the terminal device. Among them, the fourth data information here can be all the data information in addition to the third data information in the third task corresponding data information, or part of the data information in addition to the third data information in the third task corresponding data information, which is not limited here.

[0045] In a possible design, the method further includes: the terminal device performs guess inference of the third task based on the third data information and the fourth data information to obtain fourth guess inference information; and then performs the task based on the fourth guess inference information.

[0046] It can be understood that, due to the use of more data information for guess inference, the inference result will be more accurate, so the terminal device can perform the task according to the fourth guess inference information.

[0047] In a possible design, the method further includes: if the data amount corresponding to the third data information is greater than or equal to a sixth preset threshold, or the inference accuracy corresponding to the third data information is greater than or equal to a seventh preset threshold, or the ratio between the data amount corresponding to the third data information and the data amount of all data information corresponding to the third task is greater than or equal to a tenth preset threshold, the terminal device sends fifth indication information to the access network device; wherein the fifth indication information is used to indicate that the fourth data information is not sent, the fourth data information being other data information in addition to the third data information in the third task corresponding data information; or the fifth indication information is used to indicate that the priority of the third task corresponding data information transmission is reduced.

[0048] It can be understood that, considering that some tasks only need part of the data information to complete the inference, and some inference tasks may have accuracy requirements, under certain conditions, in order to save power consumption and reduce latency, the terminal device can instruct the access network device not to continue to send the subsequent data information.

[0049] In a possible design, before the terminal device performs guess inference of the third task based on the third data information, the method further includes: the terminal device receives sixth indication information from the access network device, the sixth indication information being used to indicate to turn on the guess inference function; and then turns on the guess inference function according to the sixth indication information.

[0050] It can be understood that the access network device can determine whether to instruct the terminal device to start the guess inference function according to the current network load information, or take the QoS information as a condition for instructing the terminal device to start the guess inference function, to assist in reducing the inference delay.

[0051] In a fifth aspect, an embodiment of the present application provides a communication method, which can be applied to a network side, for example, an access network device of the network side, a module (for example, a circuit, a chip or a chip system, etc.) in the access network device, or a logic node, a logic module or software capable of realizing all or part of the function of the access network device. Taking the case that the method is applied to the access network device, in the method: the access network device receives a third request from a terminal device, the third request being used to request data information corresponding to a third task; and the access network device sends third data information to the terminal device, the third data information being part of the data information corresponding to the third task.

[0052] In the embodiments of the present application, the beneficial effects of the possible implementation manners of the fifth aspect can be referred to the beneficial effects of the possible implementation manners of the fourth aspect, and details are not described herein.

[0053] In a possible design, the method further includes: the access network device sends fourth data information to the terminal device, the fourth data information being other data information in the data information corresponding to the third task except the third data information.

[0054] In a possible design, the method further includes: the access network device receives fifth indication information from the terminal device; the fifth indication information indicates that the fourth data information is not to be sent, the fourth data information being other data information in the data information corresponding to the third task except the third data information; or the fifth indication information indicates that the priority of the transmission of the data information corresponding to the third task is to be reduced.

[0055] In a possible design, the method further includes: if the second time length is greater than an eighth preset threshold value, or a quality of service (QoS) corresponding priority is a second preset priority, the access network device sends sixth indication information to the terminal device, the sixth indication information being used to instruct to start the guess inference function; the second time length is a time length of receiving downlink data estimated based on network load information, and the eighth preset threshold value is associated with the inference delay threshold value indicated by the terminal device.

[0056] In a sixth aspect, the present application provides a communication apparatus, which has the function of realizing the first aspect, for example, the communication apparatus includes a module or unit or means corresponding to the operation of the first aspect, which can be realized by software, or by hardware, or by the way of software combined with hardware.

[0057] In a seventh aspect, the present application provides a communication apparatus, which has the function of the second aspect, and the communication apparatus includes a module or unit or means corresponding to the operation of the second aspect, which can be implemented by software, or by hardware, or by a combination of software and hardware.

[0058] In an eighth aspect, the present application provides a communication apparatus, which has the function of the third aspect, and the communication apparatus includes a module or unit or means corresponding to the operation of the third aspect, which can be implemented by software, or by hardware, or by a combination of software and hardware.

[0059] In a ninth aspect, the present application provides a communication apparatus, which has the function of the fourth aspect, and the communication apparatus includes a module or unit or means corresponding to the operation of the fourth aspect, which can be implemented by software, or by hardware, or by a combination of software and hardware.

[0060] In a tenth aspect, the present application provides a communication apparatus, which has the function of the fifth aspect, and the communication apparatus includes a module or unit or means corresponding to the operation of the fifth aspect, which can be implemented by software, or by hardware, or by a combination of software and hardware.

[0061] In an eleventh aspect, the present application provides a communication apparatus, which includes an interface circuit and one or more processors. The one or more processors are coupled to a memory. The memory is used to store part or all of the necessary computer programs or instructions for implementing the functions of the first aspect or the fifth aspect. The one or more processors can execute the computer programs or instructions, and when the computer programs or instructions are executed, the communication apparatus implements the method in any possible design or implementation manner of the first aspect or the fifth aspect. The interface circuit is used to implement the communication function within the communication apparatus and / or the communication function of the communication apparatus with other devices or components.

[0062] In a possible design, the processor is configured to communicate with other devices or components through the interface circuit.

[0063] In a possible design, the communication apparatus can further include the memory.

[0064] The communication device can be an access network device, or a module (e.g., a circuit, a chip or a chip system, etc.) in the access network device, or a logic node, a logic module or software capable of implementing all or part of the functions of the access network device.

[0065] In a twelfth aspect, a communication device is provided. The communication device includes an interface circuit and one or more processors. The one or more processors are coupled to a memory. The memory is configured to store part or all of the necessary computer programs or instructions for implementing the functions related to the second aspect. The one or more processors can execute the computer programs or instructions, which, when executed, cause the communication device to implement the method in any possible design or implementation manner of the second aspect. The interface circuit is configured to implement the communication function within the communication device and / or the communication function of the communication device with other devices or components.

[0066] In a possible design, the processor is configured to communicate with other devices or components via the interface circuit.

[0067] In a possible design, the communication device can further include the memory.

[0068] The communication device can be a server, or a communication / processing module in the server, or a chip responsible for the communication function (e.g., a modem chip (also referred to as a baseband chip) or an SoC or SIP chip containing a modem module) in the server, or a circuit or chip responsible for the processing function (e.g., a GPU, an AI processor, or an ASIC) in the server.

[0069] In a thirteenth aspect, a communication device is provided. The communication device includes an interface circuit and one or more processors. The one or more processors are coupled to a memory. The memory is configured to store part or all of the necessary computer programs or instructions for implementing the functions related to the third aspect or the fourth aspect. The one or more processors can execute the computer programs or instructions, which, when executed, cause the communication device to implement the method in any possible design or implementation manner of the third aspect or the fourth aspect. The interface circuit is configured to implement the communication function within the communication device and / or the communication function of the communication device with other devices or components.

[0070] In a possible design, the processor is configured to communicate with other devices or components via the interface circuit.

[0071] In a possible design, the communication device can further include the memory.

[0072] The communication device can be a terminal device, a communication / processing module in the terminal device, a chip responsible for communication function in the terminal device, such as a modem chip (also known as a baseband chip) or a SoC or SIP chip containing a modem module, or a circuit or chip responsible for processing function in the terminal device (such as a GPU, an AI processor, or an ASIC).

[0073] In a fourteenth aspect, the present application provides a communication system, which includes an access network device, a server and a terminal device, the access network device is configured to perform the method in any possible design or implementation manner of the first aspect, the server is configured to perform the method in any possible design or implementation manner of the second aspect, and the terminal device is configured to perform the method in any possible design or implementation manner of the third aspect.

[0074] In a fifteenth aspect, the present application provides a communication system, which includes a terminal device and an access network device, the terminal device is configured to perform the method in any possible design or implementation manner of the fourth aspect, and the access network device is configured to perform the method in any possible design or implementation manner of the fifth aspect.

[0075] In a sixteenth aspect, the present application provides a computer readable storage medium, which stores computer readable instructions, when a computer reads and executes the computer readable instructions, the computer is caused to perform the method in any possible design of the first aspect to the fifth aspect.

[0076] In a seventeenth aspect, the present application provides a computer program product, when a computer reads and executes the computer program product, the computer is caused to perform the method in any possible design of the first aspect to the fifth aspect. BRIEF DESCRIPTION OF DRAWINGS

[0077] FIG. 1 is a schematic diagram of a possible, non-limiting architecture of a communication system according to an embodiment of the present application;

[0078] FIG. 2 is a schematic diagram of an application scenario according to an embodiment of the present application;

[0079] FIG. 3A is a schematic diagram of a possible application framework in a communication system according to an embodiment of the present application;

[0080] FIG. 3B is a schematic diagram of another possible application framework in a communication system according to an embodiment of the present application;

[0081] FIG. 4A is a schematic diagram of a distributed inference according to an embodiment of the present application;

[0082] FIG. 4B is a schematic diagram of an inference model according to an embodiment of the present application;

[0083] FIG. 4C is a schematic diagram of a guessing inference model according to an embodiment of the present application;

[0084] FIG. 4D is a flow diagram of a guessing inference according to an embodiment of the present application;

[0085] FIG. 5 is a flow diagram of a communication method according to an embodiment of the present application;

[0086] FIG. 6A is a graph of video frame data versus inference accuracy according to an embodiment of the present application;

[0087] FIG. 6B is a flow diagram of another communication method according to an embodiment of the present application;

[0088] FIG. 7 is a flow diagram of another communication method according to an embodiment of the present application;

[0089] FIG. 8 is a possible exemplary block diagram of a communication apparatus according to an embodiment of the present application;

[0090] FIG. 9 is a schematic diagram of another communication apparatus according to an embodiment of the present application. DETAILED DESCRIPTION

[0091] To facilitate understanding of the technical solutions of the present application, the present application will be further described below with reference to the drawings.

[0092] The terms "first" and "second" and the like in the specification of the present application, claims, and drawings are merely used to distinguish different objects, and are not intended to describe a particular order. In addition, the terms "comprise" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device, etc. comprising a series of steps or units is not limited to the listed steps or units, but can optionally further comprise steps or units not listed, or can optionally further comprise other steps or units inherent to the process, method, product, or device, etc.

[0093] "Embodiment" mentioned herein means that the specific features, structures, or characteristics described in conjunction with the embodiment can be included in at least one embodiment of the present application. The phrase appears at various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment to other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0094] In the present application, "at least one" means one or more, "multiple" means two or more, "at least two" means two or three or more, and "and / or" is used to describe the relationship between the associated objects, which means that there can be three relationships, for example, "A and / or B" can mean: only A, only B, and A and B exist at the same time, where A and B can be singular or plural. "Or" means that there can be two relationships, such as only A, only B; when A and B are not mutually exclusive, it can also mean that there are three relationships, such as only A, only B, and A and B exist at the same time. The character " / " generally represents an "or" relationship between the associated objects. "At least one of the following" or similar expressions means any combination of these items. For example, at least one of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c".

[0095] In the present application, "sending information" can be understood as one device sending information to another device, or also can be understood as one logical module in a device sending information to another logical module. For example, "the access network device sending information" can be understood as the access network device sending information to another device (such as a terminal), or can be understood as a logical module 1 in the access network device sending information to a logical module 2 in the access network device.

[0096] In the present application, "receiving information" can be understood as one device receiving information from another device, or also can be understood as one logical module in a device receiving information from another logical module. For example, "the access network device receiving information" can be understood as the access network device receiving information from another device (such as a terminal), or can be understood as a logical module 1 in the access network device receiving information from a logical module 2 in the access network device.

[0097] In the present application, "sending information to (for example, a terminal)" or related illustrations in the drawings can be understood as the destination of the information being the terminal. It can include directly or indirectly sending information to the terminal. "Receiving information from (for example, a terminal)" or "receiving information sent by (for example, a terminal)" or related illustrations in the drawings can be understood as the source of the information being the terminal, which can include directly or indirectly receiving information from the terminal. The information between the source and the destination of the information sending can be processed as necessary, such as format change, etc., but the destination can understand the valid information from the source. Similar expressions in the present application can be understood similarly, and will not be repeated here.

[0098] In order to better understand the embodiments of the present application, first, the communication system related to the embodiments of the present application will be introduced as follows:

[0099] The method provided by the embodiments of the present application can be applied to various communication systems, for example, a wireless local area network (WLAN) communication system, a wireless fidelity (Wi-Fi) system, a multiple-in multiple-out (MIMO) communication system, a long term evolution (LTE) system, an internet of things (IoT) system, a narrow band internet of things (NB-IoT) system, an LTE frequency division duplex (FDD) system, an LTE time division duplex (TDD), a 4th generation (4G) system, a 5th generation (5G) system or a new radio (NR), and other future communication systems, for example, a 6th generation (6G) system, and the like. The IoT network may, for example, include but is not limited to a vehicle internet. The communication mode in the vehicle internet system can be collectively referred to as vehicle-to-everything (V2X, X may represent any thing). For example, the V2X may include vehicle-to-vehicle (V2V) communication, vehicle-to-infrastructure (V2I) communication, vehicle-to-pedestrian (V2P) communication or vehicle-to-network (V2N) communication, and the like. The method provided by the embodiments of the present application also supports a communication system of multiple wireless technology fusion, for example, can also be applied to a system of unmanned aerial vehicle, satellite communication system, high altitude platform station (HAPS) communication and the like non-terrestrial network (NTN) fusion ground mobile communication network. In addition, it can also be applicable to low frequency (sub 6 GHz) and high frequency (above 6 GHz) communication scenarios. It can be understood that the system architecture described in the embodiments of the present application is for more clearly illustrating the technical solutions provided by the embodiments of the present application, and does not constitute a limitation on the technical solutions provided by the embodiments of the present application.

[0100] FIG. 1 is a schematic diagram of an architecture of a possible, non-limiting communication system applicable to embodiments of the present application. As shown in FIG. 1, the communication system 10 includes a radio access network (RAN) 100, a core network (CN) 200, and a data network (DN) 300. The RAN 100 includes at least one RAN node (e.g., 110a and 110b in FIG. 1, collectively referred to as 110) and at least one terminal (e.g., 120a-120j in FIG. 1, collectively referred to as 120). Other RAN nodes, such as wireless relay devices and / or wireless backhaul devices (not shown in FIG. 1), etc., can also be included in the RAN 100. The terminal 120 is connected to the RAN node 110 in a wireless manner. The RAN node 110 is connected to the core network 200 in a wireless or wired manner. The core network device in the core network 200 and the RAN node 110 in the RAN 100 can be different physical devices respectively, or can be the same physical device integrated with the logical functions of the core network and the radio access network. The core network 200 can also be connected to the data network 300 in a wireless or wired manner.

[0101] The RAN 100 can be a 3rd generation partnership project (3GPP)-related cellular system, such as a 4G, 5G mobile communication system, or a future-oriented evolved system. The RAN 100 can also be an open radio access network (O-RAN or ORAN), a cloud radio access network (CRAN), or a wireless fidelity (WiFi) system. The RAN 100 can also be a communication system that combines two or more of the above systems.

[0102] The RAN node 110, which can also be referred to as an access network device, a RAN entity, or an access node, etc., forms part of the communication system, and is configured to facilitate the wireless access by the terminals. The RAN nodes 110 in the communication system 10 can be of the same type or can be of different types. In some scenarios, the roles of the RAN node 110 and the terminal 120 are relative, e.g., the network element 120i in Figure 1 can be a helicopter or a drone, which can be configured to move as a mobile base station, and for a terminal 120j accessing the RAN 100 via the network element 120i, the network element 120i is a base station; but for the base station 110a, the network element 120i is a terminal. The RAN nodes 110 and the terminals 120 are sometimes referred to as communication devices, e.g., the network elements 110a and 110b in Figure 1 can be understood as communication devices with base station functionalities, and the network elements 120a-120j can be understood as communication devices with terminal functionalities.

[0103] In a possible scenario, the RAN node can be a base station, an evolved NodeB (eNodeB), an access point (AP), a transmission reception point (TRP), a next generation NodeB (gNB), a base station in a future mobile communication system, or an access node in a WiFi system, etc. The RAN node can be a macro base station (e.g., 110a in Figure 1), a micro base station or an indoor station (e.g., 110b in Figure 1), a relay node or a donor node, or a wireless controller in a CRAN scenario. Optionally, the RAN node can also be a server, a wearable device, a vehicle or a vehicle-mounted device, etc. For example, the access network device in vehicle to everything (V2X) technology can be a road side unit (RSU). All or part of the functions of the RAN node in this application can also be implemented by software functions running on hardware, or by virtualized functions instantiated on a platform (e.g., a cloud platform). The RAN node can also be provided with a communication module, circuit or chip for performing corresponding communication functions, and program instructions for performing corresponding communication functions. The RAN node in this application can also be a logical node, a logical module or software that can implement all or part of the functions of the RAN node.

[0104] In another possible scenario, a terminal is assisted by multiple RAN nodes to implement wireless access, and different RAN nodes respectively implement part of functions of a base station. For example, a RAN node can be a central unit (CU), a distributed unit (DU), a CU-control plane (CP), a CU-user plane (UP), or a radio unit (RU), etc. The CU and the DU can be separately arranged, or can also be included in the same network element, for example, in a baseband unit (BBU). The RU can be included in a radio frequency device or a radio frequency unit, for example, included in a remote radio unit (RRU), an active antenna processing unit (AAU), or a remote radio head (RRH).

[0105] In different systems, the CU (or CU-CP and CU-UP), DU or RU can also have different names, but those skilled in the art can understand their meanings. For example, in an ORAN system, the CU can also be referred to as an O-CU (open CU), the DU can also be referred to as an O-DU, the CU-CP can also be referred to as an O-CU-CP, the CU-UP can also be referred to as an O-CU-UP, and the RU can also be referred to as an O-RU. For the convenience of description, the CU, CU-CP, CU-UP, DU and RU are taken as examples for description in this application. Any one of the CU (or CU-CP, CU-UP), DU and RU in this application can be implemented by a software module, a hardware module, or a combination of a software module and a hardware module.

[0106] A terminal can be a device or module with corresponding communication functions and can access the above-mentioned communication system. The terminal can also be referred to as a terminal device, user equipment (UE), mobile station, mobile terminal, etc. The terminal can be widely used in various scenarios, such as device-to-device (D2D) communication, vehicle-to-everything (V2X) communication, machine-type communication (MTC), internet of things (IoT), extended reality (ER), virtual reality (VR), augmented reality (AR), industrial control, autonomous driving, remote medical treatment, smart power grid, smart furniture, smart office, smart wear, smart transportation, smart city, etc. The terminal can be a mobile phone, a tablet computer, a computer with wireless transceiver function, a wearable device (such as a smart watch, a smart bracelet, a pedometer, etc.), a vehicle, a drone, a helicopter, an airplane, a ship, a robot, a mechanical arm, a smart home device, a transport vehicle with wireless communication function, a communication module, etc. Embodiments of the present application do not limit the device form of the terminal. The terminal usually has a communication module, circuit or chip for performing corresponding communication functions. The terminal also has program instructions for performing corresponding communication functions.

[0107] A core network device refers to a device in a core network that provides service support for a terminal device, mainly responsible for registration, call connection, charging, mobility management, providing user connection, managing users, and completing bearer for services, processing and routing data, etc. The core network device can correspond to different devices in different communication systems. For example, in a 4G communication system, it can correspond to one or more of a mobility management entity (MME), a serving gateway (S-GW), etc. For another example, in a 5G communication system, it can correspond to one or more of an access and mobility management function (AMF) network element, a session management function (SMF) network element, a user plane function (UPF) network element, etc. In a next-generation communication system or a future communication system, it can correspond to one or more network elements, devices or entities that provide service support for terminal devices.

[0108] The data network can be referred to as a packet data network (PDN), and the data network can provide, for example, operator services, Internet access, or third-party services including servers. The servers can provide AI computing power for model inference, etc.

[0109] It should be noted that the communication system shown in FIG. 1 is not limited to only including the terminal device, the access network device, the core network device and the data network shown in the figure, but can also include other devices not represented in the figure. The specific embodiments of the present application will not be listed one by one here.

[0110] The application scenario of the present application will be introduced below.

[0111] Please refer to FIG. 2, which is a schematic diagram of an application scenario provided by an embodiment of the present application. As shown in FIG. 2, the present application can be applied to a network architecture of server-network-UE. The server can provide AI computing power for data network (DN) for model inference, etc. The network is used for transmitting data and can include the above-mentioned core network (CN) and radio access network (RAN). The terminal device (for example, UE) can be intelligent robots, intelligent head-mounted XR glasses, video players, holographic projectors, etc.

[0112] As shown in FIG. 2, the 5G core network can include UPF network elements, AMF network elements, SMF network elements, policy control function (PCF) network elements, application function (AF) network elements, open function (network exposure function, NEF) network elements, etc.

[0113] The UPF network element is mainly responsible for user plane related content, such as data packet routing and transmission, mobility anchor point, uplink classifier to support routing traffic to the data network, branching point to support multi-homing protocol data unit (PDU) session, packet detection, service usage reporting, QoS processing, downlink packet storage, charging information statistics, etc.

[0114] The AMF network element can complete connection management, registration process, mobility management, access authentication and authorization management, reachability management, security context management, selection of SMF network elements, and other access and mobility related functions. In addition, it is also responsible for transmitting user policies between the terminal device and the PCF network element.

[0115] The SMF network element is mainly responsible for session management in a mobile network, selection and control of a UPF network element, selection of a service and session continuity (SSC) mode, roaming, and other session-related functions. Session management can include establishment, modification, release, update, and the like of a session. Session management can also include maintenance of a tunnel between a UPF network element and an access network device.

[0116] The PCF network element can be responsible for unified policy formulation, provision of policy control, and obtaining of subscription information related to policy decision from a unified data repository (UDR) network element, and other policy-related functions. The provision of policy control can include providing service data flow and application detection, gating, QoS, and flow-based charging control, and the like.

[0117] The AF network element mainly supports interaction with a 3GPP core network to provide services or services, to affect service flow routing, access network capability exposure, policy control, and the like, and can interact with a NEF network element and the like.

[0118] The NEF network element is mainly responsible for securely exposing services and capabilities provided by 3GPP network functions, internally or to third parties. It translates or translates information exchanged with an AF network element and information exchanged with internal network functions, such as an AF network element service identifier and internal 5G core network information, such as a data network name (DNN), a single network slice selection assistance information (NSSAI) (single NSSAI, S-NSSAI), and the like.

[0119] In addition, N1, N2, N3, N4, N5, N6, N7, N11, N33, and Uu in FIG. 2 are interface sequence numbers. The meanings of these interface sequence numbers can be found in the meanings defined in the relevant standard protocols, and are not limited herein.

[0120] It should be appreciated that the above is an exemplary illustration of the 5G core network elements, and does not constitute a limitation on the 5G core network elements. For example, the 5G core network can also include a unified data management (UDM) network element, an authentication server function (AUSF) network element, a network data analytics function (NWDAF) network element, a network repository function (NRF) network element, a service communications proxy (SCP) network element, and the like.

[0121] In order to support artificial intelligence (AI) technology in the wireless network, an AI node can also be introduced in the network.

[0122] The AI node can be deployed in one or more of the following positions in the communication system: an access network node (RAN node), a terminal device, or a core network device, etc., or the AI node can also be deployed separately, for example, in a position other than any of the above devices, such as a host or a cloud server of an over the top (OTT) system. The AI node can communicate with other devices in the communication system, which can be one or more of the following: a network device, a terminal device, or a network element of a core network, etc.

[0123] It can be understood that the present application does not limit the number of AI nodes. For example, when there are multiple AI nodes, the multiple AI nodes can be divided based on functions, such as different AI nodes being responsible for different functions.

[0124] It can also be understood that the AI node can be a separate device, or can be integrated into the same device to implement different functions, or can be a network element in a hardware device, or can be a software function running on a dedicated hardware, or a virtualized function instantiated on a platform (e.g., a cloud platform), and the present application does not limit the specific form of the AI node.

[0125] The AI node can be an AI network element or an AI module.

[0126] FIG. 3A is a schematic diagram of a possible application framework in a communication system according to an embodiment of the present application. As shown in FIG. 3A, the network elements in the communication system are connected through interfaces (e.g., NG, Xn) or air interfaces. One or more AI modules (only one is shown in FIG. 3A for clarity) are deployed in one or more of the network element nodes, such as a core network device, an access network node (RAN node), a terminal, or one or more devices in operations administration and maintenance (OAM). The access network node can be a single RAN node or can include multiple RAN nodes, e.g., including a CU and a DU. The CU and / or the DU can also be provided with one or more AI modules. The CU can also be split into a CU-CP and a CU-UP, and the CU-CP and / or the CU-UP can be provided with one or more AI modules.

[0127] The AI module is used to implement a corresponding AI function. The AI modules deployed in different network elements can be the same or different. The model of the AI module can implement different functions according to different parameter configurations. The model of the AI module can be configured based on one or more of the following parameters: a structural parameter (e.g., at least one of a number of neural network layers, a neural network width, a connection relationship between layers, a weight of a neuron, an activation function of a neuron, or a bias in the activation function), an input parameter (e.g., a type of input parameter and / or a dimension of the input parameter), or an output parameter (e.g., a type of output parameter and / or a dimension of the output parameter). The bias in the activation function can also be referred to as a bias of the neural network.

[0128] In one example, the neural network described above can be a deep neural network (DNN), a convolutional neuron network (CNN), a recurrent neural network (RNN), or a generative adversarial network (GAN).

[0129] A DNN is an artificial neural network architecture that has multiple layers of nonlinear transformation units stacked together in a hierarchical structure, forming a deep computational model. Compared with a shallow neural network, a deep neural network has more hidden layers, allowing the network model to capture more complex internal structures of data and high-level abstract features.

[0130] A CNN is a deep neural network with a convolutional structure. A CNN includes a feature extractor composed of convolutional layers and subsampling layers. The feature extractor can be viewed as a filter, and the convolution process can be viewed as convolving an input image or a convolutional feature map with a trainable filter.

[0131] An RNN is a type of recursive neural network that takes sequence data as input, performs recursion in the direction of evolution of the sequence, and all nodes (recurrent units) are connected in a chain.

[0132] A GAN is a deep learning model. It is composed of a generator and a discriminator, and is trained through adversarial learning. The purpose is to estimate the latent distribution of data samples and generate new data samples.

[0133] An AI module can have one or more models. A model can infer an output, which includes a parameter or multiple parameters. The learning process, training process, or inference process of different models can be deployed in different nodes or devices, or can be deployed in the same node or device.

[0134] FIG. 3B is a schematic diagram of another possible application framework in a communication system according to an embodiment of the present application. As shown in FIG. 3B, the communication system includes a RAN intelligent controller (RIC). For example, the RIC can be the AI module shown in FIG. 3A, which is used to implement AI-related functions. The RIC includes a near-real time RIC (near-RT RIC) and a non-real time RIC (Non-RT RIC). The non-real time RIC mainly processes non-real-time information, such as data that is not sensitive to latency, which can be on the order of seconds. The real-time RIC mainly processes near-real-time information, such as data that is relatively sensitive to latency, which can be on the order of tens of milliseconds.

[0135] The near-real-time RIC is used for model training and inference. For example, it is used to train an AI model, and inference is performed using the AI model. The near-real-time RIC can obtain network-side and / or terminal-side information from RAN nodes (such as CUs, CU-CPs, CU-UPs, DUs, and / or RUs) and / or terminals. This information can be used as training data or inference data. The near-real-time RIC can deliver inference results to RAN nodes and / or terminals. The CU and the DU, and / or the DU and the RU, can interact with the inference results. For example, the near-real-time RIC delivers the inference results to the DU, which then sends them to the RU.

[0136] Non-real-time RICs are also used for model training and inference. For example, for training an AI model, inference is performed using the model. The non-real-time RICs can obtain network-side and / or terminal-side information from the RAN nodes (e.g., CUs, CU-CPs, CU-UPs, DUs, and / or RUs) and / or terminals. This information can be used as training data or inference data, and the inference results can be delivered to the RAN nodes and / or terminals. The inference results can be exchanged between CUs and DUs, and / or between DUs and RUs, for example, the non-real-time RICs deliver the inference results to the DUs, which then send them to the RUs.

[0137] The near-real-time RICs and non-real-time RICs can also be separately provided as a network element, respectively. The near-real-time RICs and non-real-time RICs can also be part of other devices, for example, the near-real-time RICs are provided in the RAN nodes (e.g., CUs, DUs), and the non-real-time RICs are provided in the OAM, cloud servers, core network devices, or other network devices.

[0138] To facilitate understanding of the schemes provided by the embodiments of the present application, the related concepts involved in the embodiments of the present application are introduced as follows:

[0139] 1. AI large model

[0140] At present, AI large models represented by ChatGPT, with their outstanding natural language and multi-modal understanding capabilities and strong support for generative AI, are showing a flourishing development trend and have achieved remarkable results in the fields of image, text, voice, and recommendation systems. AI large models go through two stages of training and inference from construction to actual application. The training stage refers to the process of building a machine learning model, which learns patterns and features from a large amount of data to optimize model performance, thus requiring a large amount of data and computing power, which can be completed on a server. The inference stage refers to the process of using a trained model to perform task processing, which can be performed in special scenarios or with limited computing power, for example, in a data center in the cloud.

[0141] For the training stage, the large number of parameters and training data lead to the increasing demand for computing power of large models, and distributed cluster training based on multiple machines and multiple cards becomes a must for large models. How to analyze and optimize the training of the network communication characteristics of large model distributed training is the core element of improving the training speed of large models, reducing the training time, and reducing the training cost.

[0142] For the inference stage, a high requirement for latency is generally proposed. If the inference latency is too long, the task processing efficiency will be greatly reduced, which will seriously affect the user experience in the application scenario of large models. Based on this, in the embodiments of the present application, the idea of distributed training can be used for distributed inference to fully utilize the computing power of different devices to reduce inference latency.

[0143] 2. Distributed inference

[0144] In distributed computing, high and low level tasks can be decomposed to the same or different computing nodes, and the task will depend on the last completion time of different computing subtasks. Therefore, the present application proposes to segment the inference task and allocate the segmented subtasks to different devices to jointly complete the inference task.

[0145] With the continuous development of terminal devices, terminal devices also have certain computing power and can bear certain neural network computing. As an example, the front-end preprocessing computing part can be performed by the terminal device, the lower level and lower computing power part can be performed by the access network device, and the higher level and higher computing power part can be performed by the server. Compared with single device completion of inference, distributed inference can minimize inference latency.

[0146] As shown in FIG. 4A, FIG. 4A includes a robot (i.e., a terminal device), a base station, and a server. Among them, the robot has an inference task. Considering the limited computing power of the robot, the robot can distribute the inference task to the base station and the server for implementation. Taking a navigation task as an example, the robot can send video information to the base station to let the base station deploy computing power to perform object recognition based on the video frame, and then the base station sends the object recognition result (such as the information of the recognized obstacle) to the server, and the server provides a navigation path (i.e., a navigation path planning task) for the robot according to the object recognition result and its own navigation planning capability.

[0147] 3. Guessing inference

[0148] Taking a large language model (LLM) as an example, the inference stage usually needs to use an autoregressive model. Their inference process is quite slow, which needs to be decoded token by token in series, causing the problem of slow overall inference latency.

[0149] The implementation process is shown in FIG. 4B. When performing inference, the large model generates corresponding inference results step by step based on the question. After obtaining a token, the current generated token is used as the input of the next inference to perform the next inference. For example, the first token is input into the autoregressive model (i.e., a single inference model), and token (A) is output. Then, token (A) and the second token are input into the autoregressive model, and token (B) is output. This process is repeated until a special stop token is generated (satisfying a certain condition, such as exceeding a specific length). Assuming that there are N tokens in total, the decoding stage needs to perform N-1 forward propagations, which can only be performed in series and is very inefficient.

[0150] To solve this problem, a speculative inference architecture is proposed. The speculative inference can also be referred to as speculative reasoning. The core idea of speculative inference is that many common words and sentences are very easy to predict, so the large model only needs to guide the small model in the key part, which can bring performance improvement. Unlike the traditional autoregressive inference method (as shown in FIG. 4B), the speculative inference adopts a draft model (usually a smaller model) to perform autoregressive inference, and the original large model judges whether to accept the multiple tokens inferred by the small model based on the inference result of the small model.

[0151] It can be understood that, compared with the current single inference model, an additional speculative model can be introduced, and the previous single inference model can be used as a verification model, and the two models together constitute a speculative inference model. In the embodiments of the present application, the speculative inference model can be used to directly infer the inference result based on the data information; or the speculative inference model can be used to first infer the possible data information after the data information based on the data information, and then perform task inference using the received data information and the inferred possible data information, which is not limited herein.

[0152] As shown in FIG. 4C, token (A) is input into the Efficiently Draft model (i.e., the guessing model), which efficiently "guesses" three tokens that are likely to be generated in the future, namely token (A1), token (B1), and token (C1). Then, token (A) and the predicted tokens (A1), (B1), and (C1) are input into the Verify in parallel model (i.e., the verification model) together, and tokens 1, 2, 3, and 4 are output. Among them, tokens 1 and 2 pass the verification and can be used as the decoding result; token 3 fails the verification and cannot be used as the decoding result. In the embodiments of the present application, "guessing" can also be referred to as "speculating".

[0153] If the "guessing" is accurate enough, multiple tokens can be generated in parallel for each decoding step, thereby achieving LLM inference acceleration. Moreover, the verification process using the Verify in parallel model can theoretically guarantee that the decoding result is completely consistent with the decoding result of the autoregressive model. It can be understood that the Verify in parallel model is a lightweight model of the autoregressive model.

[0154] In addition, considering the Efficiently Draft model, single-step multi-step prediction is achieved, and there may be inaccurate prediction. The longer the prediction window, the less accurate the prediction. Therefore, when the Efficiently Draft model is inaccurate, the Verify in parallel model does not adopt the output of the Efficiently Draft model, but selects to fall back and make a prediction based on the accurate result of its own inference.

[0155] For example, the guessing model outputs multiple guessed tokens according to the request, and the accuracy of different tokens is different. Then, the guessed result is sent to the verification model. When the verification model confirms the correctness, the guessed result can be directly used, and the token inference in the following step can be continued. As shown in FIG. 4D, the question "recite the poem by Li Bai" is input into the guessing model, and the guessed content is "bed", "front", and "light", with the corresponding accuracy of "1.0", "0.8", and "0.6", respectively. Then, the question "recite the poem by Li Bai", the guessed content "bed", "front", and "light", and the corresponding accuracy "1.0", "0.8", and "0.6" are input into the verification model together, and the guessed content "bed", "front", and "light" are output, and the guessed content "bed", "front", and "light" all pass the verification.

[0156] The guess reasoning core described above introduces a prediction reasoning module that attempts to predict multiple tokens later based on limited information. If the prediction is accurate, the time delay of large model reasoning can be reduced. If the prediction is not accurate, the large model verification capability is relied on to select fallback, and the time delay relative to large model reasoning is not lost. In the embodiments of the present application, the reasoning model described in FIG. 4B only has the function of reasoning, and the guess reasoning model described in FIG. 4C has both the function of guessing and the function of reasoning.

[0157] It should be noted that a token refers to the smallest semantic unit in a text. For example, a sentence can be divided into several words, and each word is a token. For example, the sentence "I love you" can be divided into three tokens: "I", "love" and "you". Tokens can help us break down text into more easily processed and analyzed parts.

[0158] Based on the above, although distributed computing can split tasks to different modules for execution, thereby reducing the computing time delay, the air interface resource is limited. When the data volume related to the reasoning task is large, the terminal device needs a long time to send the data to the base station, and the entire reasoning time delay is limited by the air interface resource, and cannot be reduced. Moreover, the guess reasoning scheme is currently only applied to the reasoning task of a cloud large model, and there is no corresponding design in distributed reasoning, so the gain of reducing the overall reasoning time delay by guess reasoning cannot be obtained.

[0159] Therefore, in order to reduce the overall reasoning time delay and improve the reasoning efficiency, the present application provides a communication method and a communication device. The communication method and the communication device provided by the embodiments of the present application are described in detail below.

[0160] FIG. 5 is a flow diagram of a communication method according to an embodiment of the present application. As shown in FIG. 5, the communication method includes the following steps S501-S505. The method execution subject shown in FIG. 5 can be the terminal device, the access network device and the server mentioned above. Alternatively, the method execution subject shown in FIG. 5 can be a chip in the terminal device, a chip in the access network device and a chip in the server, which is not limited in the embodiments of the present application. FIG. 5 takes the terminal device, the access network device and the server as the method execution subject as an example for description.

[0161] It can be understood that the terminal device, the access network device and the server are taken as examples in the present application as the execution subject of the interaction, but the present application does not limit the execution subject of the interaction. For example, the method executed by the access network device in the present application can also be implemented by a module (such as a circuit, a chip or a chip system, etc.) in the access network device, or a logic node, a logic module or software capable of realizing all or part of the function of the access network device; the method executed by the terminal device in the present application can also be implemented by a communication / processing module in the terminal device or a circuit or chip (such as a modem chip (also known as a baseband chip), or a SoC chip / SIP chip containing a modem core, or a GPU / AI processor / ASIC) responsible for communication / processing function in the terminal device; the method executed by the server in the present application can also be implemented by a communication / processing module in the server or a circuit or chip (such as a modem chip (also known as a baseband chip), or a SoC chip / SIP chip containing a modem core, or a GPU / AI processor / ASIC) responsible for communication / processing function in the server.

[0162] S501, the terminal device sends a first request and first data information to the access network device, the first request being used to request the access network device to perform inference of a first task. Correspondingly, the access network device receives the first request and the first data information from the terminal device.

[0163] S502, the access network device performs inference of the first task based on the first data information to obtain first inference information.

[0164] S503, the access network device sends the first inference information to the server. Correspondingly, the server receives the first inference information from the access network device.

[0165] In the embodiments of the present application, when the terminal device has an inference task, considering that the computing power of the terminal device is limited, the terminal device can allocate the inference task to the access network device and the server for implementation. When the access network device has an inference task, the inference task can also be allocated to the terminal device and the server for implementation.

[0166] For the case that the terminal device is the subject of task allocation, the access network device and the server need to notify the terminal device in advance (i.e., the access network device and the server need to align the computing power capability with the terminal device), and the computing power capability possessed by the terminal device itself. For example, the access network device can perform object recognition tasks based on video frames, and the server can perform navigation path planning tasks. Among them, the possible signaling adopted by the access network device when notifying the terminal device includes but is not limited to: medium access control control element (MAC CE), downlink control information (DCI). The possible method for the server to notify the terminal device includes but is not limited to: the server first notifies the core network device through the N33 interface, the core network device then notifies the access network device through the GTP-C (GPRS tunneling protocol-Control) signaling, and the access network device then forwards to the terminal device.

[0167] For the case that the access network device is the subject of task allocation, the terminal device and the server need to notify the access network device in advance (i.e., the terminal device and the server need to align the computing power capability with the access network device), and the computing power capability possessed by the access network device itself. The possible signaling adopted by the terminal device when notifying the access network device includes but is not limited to: information in radio resource control (RRC) signaling, and a newly added MAC CE. The possible method for the server to notify the access network device includes but is not limited to: the server first notifies the core network device through the N33 interface, and the core network device then notifies the access network device through the GTP-C signaling.

[0168] Among them, the access network device can deploy the guess inference model mentioned in the above FIG. 4C (which can be considered as the AI node mentioned above), and the server can deploy the inference model mentioned in the above FIG. 4B (which can be considered as the AI node mentioned above). The initialization of the guess inference model and the inference model are strongly related to the task. For example, the computing power of the access network device is used for object recognition in video frames, and then the access network device can deploy a guess inference model based on video frames. The guess inference model can directly infer the inference result according to the limited video frames; or the guess inference model can be used to guess the subsequent video frames based on the received limited video frames, and then perform object recognition inference to obtain the inference result. For another example, the computing power of the server is used for navigation path planning, and then the server can deploy an inference model based on navigation path planning, which can perform navigation path planning inference based on the received object recognition result (such as obstacle information).

[0169] Optionally, the guessing inference model deployed by the access network device can also have a verification module, that is, the accuracy between the guessing inference information and the actually received information can be compared. When the accuracy is greater than or equal to a preset value, it is considered that the verification is passed, proving that the current guessing inference information is correct and can be applied to subsequent tasks.

[0170] The following embodiments of the present application take the terminal device as an example to perform detailed description:

[0171] Firstly, the access network device and the server need to align the computing power capability with the terminal device, that is, the access network device and the server need to inform the terminal device of the computing power capability possessed by the terminal device itself. Then, the access network device needs to initialize the guessing inference model deployed by the access network device, and the server also needs to initialize the inference model deployed by the server. Next, the terminal device performs task allocation in combination with the computing power capability of the access network device and the server and considering the task demand of the terminal device itself, for example, the terminal device can allocate the inference of the first task to the access network device and the inference of the second task to the server. Specifically, the terminal device can send a first request and first data information to the access network device, where the first request is used to request the access network device to perform the inference of the first task. The first data information here is the data information related to the first task.

[0172] Exemplarily, it is assumed that the terminal device (such as a robot) needs to perform a navigation task. The terminal device can send video information (i.e., the first data information) to the access network device and request the access network device to perform the inference of the object recognition task (i.e., the first task) based on the video frame.

[0173] Further, after receiving the first request and the first data information from the terminal device, the access network device can input the first data information into the guessing inference model for guessing inference, to obtain first guessing inference information, since the access network device has deployed the guessing inference model.

[0174] In a possible design, the first data information is part of the data information corresponding to the first task.

[0175] It can be understood that, due to the limitation of air interface resources, the complete data transmission time is relatively long. The first data information here is only part of the data information related to the first task reported by the terminal device. Considering that some tasks can be completed by using only part of the data information, taking the event recognition task in the video (such as object recognition (such as obstacle recognition), action recognition (such as sitting, walking, standing, etc.), etc.) as an example, since the video frames themselves have correlation (information redundancy), accurate inference can be completed by using part of the data information. Exemplarily, as shown in FIG. 6A, 90% of the video frame data is received to achieve 100% inference accuracy, and 100% of the video frame data does not need to be received.

[0176] Therefore, the access network device can perform inference of the first task without waiting for the terminal device to report all data information, can directly infer the inference result according to limited data information by using the guessing inference model, or can first infer possible subsequent data information according to part of the data information by using the guessing inference model, and then perform inference of the first task by using the part of the data information and the inferred possible subsequent data information. The entire process can be considered as a guessing inference process, so that the inference result can be obtained, and the overall inference delay can be reduced.

[0177] For example, it is assumed that the data information corresponding to the first task is video information, the first data information is part of the video information, and includes video frame 1 and video frame 2. The guessing inference model deployed by the access network device can implement guessing inference of object recognition based on video frames. Then, the access network device inputs the received video frame 1 and video frame 2 to the guessing inference model, can first infer the content of video frame 3 through the guessing inference model, and then performs inference of object recognition (i.e., the first task) based on video frame 1, video frame 2, and the inferred video frame 3, so as to obtain obstacle information (i.e., the first guessing inference information).

[0178] In a possible design, the first request is further used to request the server to perform inference of a second task. When the access network device sends the first guessing inference information to the server, the specific implementation manner can be that the access network device sends a second request and the first guessing inference information to the server, where the second request is used to request the server to perform inference of the second task. Correspondingly, the server receives the second request and the first guessing inference information from the access network device.

[0179] It can be understood that, in addition to requesting the access network device to perform inference of the first task, the terminal device also requests the server to perform inference of the second task. Therefore, the access network device subsequently sends the first guessing inference information to the server, and also sends the request of the terminal device to the server, i.e., requests the server to perform inference of the second task. The first task and the second task are both inference tasks allocated by the terminal device.

[0180] Optionally, the first task is associated with the capability of the access network device, and the second task is associated with the capability of the server. For example, the access network device has an object recognition capability based on video frames, and the first task can be an object recognition task based on video frames. The server has a navigation path planning capability, and the second task can be a navigation path planning task. Based on this manner, the adaptability of the task to the device capability can be ensured.

[0181] In a possible design, the method further includes: the server sending, to the access network device, fourth indication information used to indicate a window length corresponding to the first guess inference information. Correspondingly, the access network device receives the fourth indication information from the server.

[0182] It can be understood that the server can indicate the window length corresponding to the first guess inference information to the access network device. Of course, the window length corresponding to the first guess inference information can be predefined or configured (for example, configured by the device out of the factory or preconfigured by the high-layer signaling of the device), which is not limited herein.

[0183] For example, assuming that the fourth indication information sent by the server to the access network device indicates that the window length corresponding to the first guess inference information is 1 frame, after the access network device obtains the video frame 1 and the video frame 2, the content of the next 1 frame (for example, the content of the video frame 3) can be guessed according to the guess inference model.

[0184] For another example, the fourth indication information sent by the server to the access network device indicates that the window length corresponding to the first guess inference information is 2 frames, after the access network device obtains the video frame 1 and the video frame 2, the content of the next 2 frames (for example, the content of the video frame 3 and the content of the video frame 4) can be guessed according to the guess inference model.

[0185] In a possible design, before the access network device performs the guess inference of the first task based on the first data information, the method further includes: if a first time length is greater than a fourth preset threshold, or a priority corresponding to QoS is a first preset priority, the access network device starts the guess inference function; the first time length is a time length of receiving uplink data estimated based on network load information, and the fourth preset threshold is associated with an inference delay threshold indicated by the terminal device. It should be noted that starting the guess inference function means that the guess inference model can be used to perform the guess inference of the task. The QoS refers to a network security mechanism that can use various basic technologies to provide better service capabilities for specified network communication, and is a technology used to solve network delay and congestion. The guarantee of QoS is very important for a network with limited capacity, especially for streaming multimedia applications, because these applications often require fixed transmission rates and are sensitive to delay.

[0186] It can be understood that the access network device can determine whether to start the guess inference function according to the current network load information, that is, estimate the duration (that is, the first duration) of receiving uplink data based on the network load information, and if the first duration is greater than a fourth preset threshold, the guess inference function is started to assist in reducing the inference delay. The fourth preset threshold is associated with the inference delay threshold indicated by the terminal device. For example, if the inference delay threshold reported by the terminal device is 100 ms, the fourth preset threshold can be set to 60 ms (100 ms*0.6), and if the estimated first duration exceeds 60 ms, the guess inference function is started.

[0187] Alternatively, the server can indicate the QoS information to the access network device as a condition for starting the guess inference function. If the priority corresponding to the current QoS is the first preset priority, the guess inference function is started. If the priority corresponding to the current QoS is other priority, the guess inference function is not started. The way in which the server indicates the QoS information to the access network device includes but is not limited to: the server first notifies the core network device through the N33 interface, and the core network device then notifies the access network device using GTP-C signaling.

[0188] For example, assuming that the first preset priority is 2, if the server indicates to the access network device that the priority corresponding to the QoS is 1, the guess inference function is not started; if the server indicates to the access network device that the priority corresponding to the QoS is 2, the guess inference function is started.

[0189] S504, the server performs inference of a second task based on the first guess inference information to obtain first inference information.

[0190] In the embodiment of the application, after the server receives the first guess inference information sent by the access network device, the server can perform inference of a second task based on the first guess inference information to obtain first inference information.

[0191] Optionally, the method further includes the step S505: the server sends the first inference information to the terminal device. Correspondingly, the terminal device receives the first inference information from the server. Subsequently, the terminal device can perform a task according to the first inference information.

[0192] For example, assuming that the inference model deployed by the server can realize inference of navigation path planning. After the server receives the obstacle information (that is, the first guess inference information) sent by the access network device, the server inputs the obstacle information into the inference model to perform inference of navigation path planning (that is, a second task), and can obtain a planned navigation path (that is, first inference information). The terminal device can perform a navigation task according to the navigation path.

[0193] In a possible design, due to limited air interface resources, the terminal device can send different proportions of data information at different sending durations, and correspondingly, the access network device can receive different proportions of data information at different receiving durations. Therefore, after the terminal device sends the first data information to the access network device, the terminal device can continue to send other data information (for example, second data information) in addition to the first data information to the access network device. Of course, the terminal device can also not continue to send other data information (for example, second data information) in addition to the first data information. The following describes the two cases in detail:

[0194] Case 1: After the terminal device sends the first data information to the access network device, the terminal device continues to send the second data information to the access network device.

[0195] In a specific implementation, if the access network device receives the second data information sent by the terminal device, the access network device can perform guess reasoning of the first task again based on the first data information and the second data information, to obtain second guess reasoning information, and perform verification by using the second guess reasoning information. If the verification succeeds, the first reasoning information is directly sent to the terminal device, so that the terminal device can execute the task according to the first reasoning information. If the verification fails, a rollback action is performed, and the server is instructed to perform reasoning based on the second guess reasoning information again, and reasoning information obtained by re-reasoning is sent to the terminal device to execute the task. The second data information can be all data information in addition to the first data information, or part of the data information in addition to the first data information, and the specific implementation is not limited herein. The following describes the specific implementation.

[0196] As shown in FIG. 6B, after steps S501-S504, the method further includes steps s11-s16.

[0197] s11: The terminal device sends second data information to the access network device, where the second data information is other data information in addition to the first data information.

[0198] s12: The access network device performs guess reasoning of the first task based on the first data information and the second data information, to obtain second guess reasoning information.

[0199] In a specific implementation, after the access network device obtains the second guessed inference information, if the similarity between the second guessed inference information and the first guessed inference information is greater than or equal to a first preset threshold, step s13 is performed; if the similarity between the second guessed inference information and the first guessed inference information is less than the first preset threshold, steps s14-s16 are performed.

[0200] The similarity here can refer to semantic similarity between the second guessed inference information and the first guessed inference information. Specific determination methods include but are not limited to: converting the second guessed inference information and the first guessed inference information into text vectors or word vectors, and using cosine similarity, word vector similarity, similarity calculation based on deep learning (such as BERT model), etc. to determine the semantic similarity between the two. The first preset threshold can be preconfigured (for example, configured by the device out of the factory or preconfigured by the high-level signaling of the device), or can be preset, which is not limited here.

[0201] When the similarity between the second guessed inference information and the first guessed inference information is greater than or equal to the first preset threshold, it can be considered that the first guessed inference information is verified successfully. When the similarity between the second guessed inference information and the first guessed inference information is less than the first preset threshold, it can be considered that the first guessed inference information is verified unsuccessfully.

[0202] s13, the server sends the first inference information to the terminal device. Correspondingly, the terminal device receives the first inference information from the server.

[0203] s14, the access network device sends first indication information to the server, the first indication information being used for indicating re-execution of inference of the task, and the first indication information including the second guessed inference information. Correspondingly, the server receives the first indication information from the access network device.

[0204] The possible signaling corresponding to the first indication information includes but is not limited to: non-access stratum (NAS) signaling.

[0205] Optionally, the access network device sends second indication information to the terminal device, the second indication information being used for indicating increasing priority of data information transmission corresponding to the first task. Correspondingly, the terminal device receives the second indication information from the access network device. It can be understood that the higher the priority of data information transmission is, the higher the speed or efficiency of transmission is. Therefore, when the first guessed inference information is verified unsuccessfully, in order to enable the terminal device to report the remaining data information as soon as possible, the access network device can indicate the terminal device to increase the priority of data information transmission corresponding to the first task.

[0206] s15, the server re-performs the inference of the second task based on the second guess inference information, to obtain second inference information.

[0207] s16, the server sends the second inference information to the terminal device. Correspondingly, the terminal device receives the second inference information from the server.

[0208] It can be understood that after the access network device receives the second data information sent by the terminal device subsequently, the access network device inputs the second data information and the first data information received previously into the guess inference model to re-perform the guess inference of the first task, to obtain second guess inference information. Since the second guess inference information is determined based on more data information, the accuracy thereof is higher than that of the first guess inference information, and thus the second guess inference information can be used to verify whether the first guess inference information determined previously is accurate.

[0209] If the similarity between the second guess inference information and the first guess inference information is greater than or equal to a first preset threshold (i.e., the verification is successful), the access network device does not need to perform any action, and the server can directly send the first inference information to the terminal device.

[0210] If the similarity between the second guess inference information and the first guess inference information is less than the first preset threshold (i.e., the verification fails), a rollback action is performed, that is, the access network device sends first indication information to the server, the first indication information being used to indicate that the inference of the task is re-performed, and the first indication information including the second guess inference information. After receiving the first indication information, the server re-performs the inference of the second task based on the second guess inference information, to obtain second inference information. Then, the terminal device is sent the second inference information which is more accurate.

[0211] Since the time required for verification is much less than the time required for inference, once the verification is successful, the overall inference delay can be greatly reduced; if the verification fails, the rollback action is performed, and the overall inference delay is not affected.

[0212] Exemplarily, taking a navigation task as an example, the terminal device sends video frame 1 and video frame 2 to the access network device, the access network device performs guess inference of object recognition based on the video frame 1 and the video frame 2, to obtain a first object recognition result (such as obstacle information); the access network device sends the first object recognition result to the server, the server performs inference of a navigation path planning based on the first object recognition result, to obtain a first navigation path. Further, the terminal device sends video frame 3 to the access network device, the access network device performs guess inference of object recognition based on the video frame 1, the video frame 2 and the video frame 3, to obtain a second object recognition result.

[0213] If the similarity between the second object recognition result and the first object recognition result is greater than or equal to the first preset threshold (i.e., the verification succeeds), the access network device does not need to perform any action, and the server directly sends the first navigation path to the terminal device. After the terminal device receives the first navigation path, the terminal device can perform the navigation task according to the first navigation path.

[0214] If the similarity between the second object recognition result and the first object recognition result is less than the first preset threshold (i.e., the verification fails), a rollback action is performed, that is, the access network device sends first indication information to the server, where the first indication information is used to instruct to re-perform inference of the task, and the first indication information includes the second object recognition result. After the server receives the first indication information, the server re-performs inference of the navigation path planning based on the second object recognition result, to obtain a second navigation path. Then, the server sends the more accurate second navigation path to the terminal device. After the terminal device receives the second navigation path, the terminal device can perform the navigation task according to the second navigation path.

[0215] Based on this, it is assumed that after the terminal device sends the video frame 1, the video frame 2, and the video frame 3 to the access network device, the access network device performs inference of object recognition based on the video frame 1, the video frame 2, and the video frame 3 (i.e., using the inference model shown in FIG. 4B), and the server performs inference of the navigation path planning based on the inference result of the access network device, and the total time delay is 15 seconds.

[0216] If the method of steps S501-S504 and steps s11-s16 is used, the terminal device needs 2 seconds to send the video frame 1 and the video frame 2 to the access network device, the access network device needs 3 seconds to perform inference of object recognition based on the video frame 1 and the video frame 2 (i.e., using the inference model shown in FIG. 4C), and the server needs 3 seconds to perform inference of the navigation path planning based on the inference result of the access network device; the terminal device needs 1 second to further send the video frame 3 to the access network device, the access network device needs 2 seconds to perform inference of object recognition based on the video frame 3, and the access network device needs 1 second to verify the inference result.

[0217] If the verification succeeds, the total time delay is 12 seconds. As can be seen, since the time required for verification is much less than the time required for inference, once the verification succeeds, compared with sending all the video frames to the access network device for inference, it is more beneficial to save the overall inference time delay.

[0218] If the verification fails, a rollback action is performed, the server re-performs inference of the navigation path planning based on the inference result of the access network device (3 seconds), and the total time delay is also 15 seconds. As can be seen, although the verification fails and the rollback action is performed, the overall inference time delay is not affected.

[0219] Case two: after the terminal device sends the first data information to the access network device, the terminal device does not continue to send the second data information to the access network device.

[0220] In a specific implementation, considering that some tasks only need part of the data information to complete the inference, and considering that some inference tasks can have accuracy requirements, in a case where a certain condition is met, in order to save power consumption and reduce latency, and to avoid waste of resources, the access network device can instruct the terminal device not to continue to send the subsequent data information.

[0221] Specifically, the condition met here can adopt the following three manners:

[0222] Manner one: if the data amount corresponding to the first data information is greater than or equal to a second preset threshold, the access network device sends third indication information to the terminal device; wherein the third indication information is used to indicate that the second data information is not sent, the second data information being other data information in the data information corresponding to the first task except the first data information; or the third indication information is used to indicate that the priority of the data information transmission corresponding to the first task is reduced.

[0223] In a specific implementation, the second preset threshold here can be configured in combination with task requirements (for example, configured by a device out of the factory or pre-configured by a high layer signaling of the device), or can be pre-set for different tasks, which is not limited here. The configuration manner includes but is not limited to RRC signaling configuration. If the data amount corresponding to the first data information is greater than or equal to the second preset threshold, the access network device sends third indication information to the terminal device. The third indication information can indicate that the terminal device does not send the second data information, or can indicate that the priority of the data information transmission corresponding to the first task is reduced. Since the higher the priority of the data information transmission is, the higher the speed or efficiency of the data information transmission will be, the purpose of reducing the priority of the data information transmission corresponding to the first task is to reduce the transmission speed of the second data information, or even not to transmit the second data information.

[0224] Exemplarily, taking a video frame as an example, assuming that 90% of the video frames (i.e. the second preset threshold) are received to complete the inference, when the data amount corresponding to the first data information is greater than or equal to 90%, the access network device can send third indication information to the terminal device, indicating that the terminal device does not send the second data information, or reduces the priority of the data information transmission corresponding to the first task.

[0225] The third indication information can be used to instruct the terminal device not to send the second data information, or to reduce the priority of the data information corresponding to the first task.

[0226] In a specific implementation, the third preset threshold can be configured in combination with a task requirement (for example, configured by a device manufacturer or preconfigured by high-layer signaling of the device), or can be preconfigured for different tasks, which is not limited herein. The configured manner includes but is not limited to RRC signaling configuration. The inference accuracy refers to the accuracy of a guessing inference model in performing an inference task on data information (for example, inference accuracy based on event window statistics), for example, assuming that 99 out of 100 inference tasks are correct, the inference accuracy is 99%.

[0227] If the inference accuracy corresponding to the first data information is greater than or equal to the third preset threshold, the access network device sends third indication information to the terminal device. The third indication information can instruct the terminal device not to send the second data information, or to reduce the priority of the data information corresponding to the first task. Since the higher the priority of data information transmission is, the higher the speed or efficiency of data information transmission will be, the purpose of reducing the priority of the data information corresponding to the first task is to reduce the transmission speed of the second data information, or even not to transmit the second data information.

[0228] For example, taking a video frame as an example, assuming that 90 video frames are received, and the third preset threshold is 90%, if the inference accuracy is greater than or equal to 90% at this time, the access network device can send third indication information to the terminal device, instructing the terminal device not to send the second data information, or to reduce the priority of the data information corresponding to the first task.

[0229] The third indication information can be used to instruct the terminal device not to send the second data information, or to reduce the priority of the data information corresponding to the first task.

[0230] In a specific implementation, the ninth preset threshold value herein can be configured in combination with the task requirement (for example, configured out of the factory of the device or pre-configured by the high-layer signaling of the device), or pre-set for different tasks, which is not limited herein. The configuration manner includes but is not limited to RRC signaling configuration. If the ratio between the data amount of the first data information and the data amount of the entire data information corresponding to the first task is greater than or equal to the ninth preset threshold value, the access network device sends third indication information to the terminal device. The third indication information can indicate that the terminal device does not send the second data information, or can indicate to reduce the priority of the data information transmission corresponding to the first task. Since the higher the priority of the data information transmission is, the higher the speed or efficiency of the data information transmission will be, the purpose of reducing the priority of the data information transmission corresponding to the first task is to reduce the transmission speed of the second data information, or even not to transmit the second data information.

[0231] Exemplarily, taking a video frame as an example, assuming that 90 video frames (i.e., the first data information) are received, the entire video frame corresponding to the first task is 100 frames (i.e., the entire data information), and the ninth preset threshold value is 0.9, at this time, the ratio between the first data information and the entire data information is 0.9 (greater than or equal to the ninth preset threshold value), then the access network device can send third indication information to the terminal device, indicating that the terminal device does not send the second data information, or reduces the priority of the data information transmission corresponding to the first task.

[0232] It can be seen that based on the method described in FIG. 5, when the terminal device has an inference task, considering that the computing power of the terminal device is limited, the terminal device can allocate the inference task to the access network device and the server for implementation. The access network device can deploy a guess inference model, and the server can deploy an inference model. After the access network device receives the first data information, it does not need to wait for the terminal device to report the entire data information before performing the inference of the first task, but can directly infer the inference result according to the limited data information by using the guess inference model; or can first infer the possible data information after the first data information by using the guess inference model, and then perform the inference of the first task by using the first data information and the inferred possible data information after the first data information, the entire process can be considered as a guess inference process, so that the first guess inference information can be obtained, and then the first guess inference information is sent to the server for the inference of the second task. In this way, it is beneficial to reduce the overall inference delay and improve the inference efficiency.

[0233] FIG. 7 is a flow diagram of another communication method according to an embodiment of the present application. As shown in FIG. 7, the communication method includes steps S701-S703. The method execution subject shown in FIG. 7 can be the terminal device and the access network device mentioned above. Alternatively, the method execution subject shown in FIG. 7 can be a chip in the terminal device and a chip in the access network device, which is not limited in the embodiments of the present application. FIG. 7 takes the terminal device and the access network device as the method execution subject for example.

[0234] It can be understood that the terminal device and the access network device are taken as the method execution subject for example in the present application, but the present application does not limit the method execution subject. For example, the method executed by the access network device in the present application can also be implemented by a module (such as a circuit, a chip or a chip system, etc.) in the access network device, or a logic node, a logic module or software capable of realizing all or part of the function of the access network device; the method executed by the terminal device in the present application can also be implemented by a communication / processing module in the terminal device or a circuit or a chip (such as a modem chip (also known as a baseband chip), or a SoC chip / SIP chip containing a modem core, or a GPU / AI processor / ASIC) responsible for communication / processing function in the terminal device.

[0235] S701. The terminal device sends a third request to the access network device, where the third request is used to request data information corresponding to a third task.

[0236] S702. The access network device sends third data information to the terminal device, where the third data information is part of the data information corresponding to the third task.

[0237] S703. The terminal device performs guess inference of the third task based on the third data information to obtain third guess inference information.

[0238] In the embodiments of the present application, when the terminal device has an inference task, if the computing power of the terminal device is sufficient, the terminal device can also execute the inference task by itself. The terminal device can deploy the guess inference model mentioned in FIG. 4C (which can be regarded as the AI node mentioned above), and the initialization of the guess inference model is strongly related to the task. For example, if the computing power deployed by the terminal device is navigation path planning, the terminal device can deploy a guess inference model based on navigation path planning, and can perform guess inference of navigation path planning based on the received environment perception information.

[0239] In a possible design, the method further includes: the terminal device performs the task based on the third guess inference information.

[0240] The following embodiments of the present application take the terminal device itself performing the navigation path planning task as an example for detailed description:

[0241] Firstly, the terminal device needs to initialize the guess reasoning model deployed by it, and then the terminal device sends a third request to the access network device, where the third request is used to request data information corresponding to the navigation path planning task (i.e., the third task). After receiving the third request from the terminal device, the access network device can send part of the sensing information (i.e., the third data information) to the terminal device. Since the terminal device is deployed with the guess reasoning model, the guess reasoning model can be used to perform the guess reasoning of the navigation path planning according to the part of the sensing information provided by the access network device, and the planned navigation path (i.e., the third guess reasoning information) is obtained. Subsequently, the terminal device can perform the navigation task according to the navigation path.

[0242] In a possible design, before the terminal device performs the guess reasoning of the third task based on the third data information, the method further includes: if the second time length is greater than an eighth preset threshold value, or the priority corresponding to the QoS is a second preset priority, the access network device sends sixth indication information to the terminal device, where the sixth indication information is used to indicate to start the guess reasoning function; the second time length is a time length of receiving downlink data estimated based on the network load information, and the eighth preset threshold value is associated with the reasoning time delay threshold value indicated by the terminal device. Correspondingly, the terminal device receives the sixth indication information from the access network device, and then starts the guess reasoning function according to the sixth indication information. It should be noted that starting the guess reasoning function means that the guess reasoning model can be used to perform the guess reasoning of the task.

[0243] It can be understood that the access network device can determine whether to start the guess reasoning function according to the current network load information, that is, estimate the time length (i.e., the second time length) of receiving downlink data based on the network load information. If the second time length is greater than the eighth preset threshold value, the access network device sends the sixth indication information to the terminal device to indicate the terminal device to start the guess reasoning function to assist in reducing the reasoning time delay. The eighth preset threshold value is associated with the reasoning time delay threshold value indicated by the terminal device. For example, if the reasoning time delay threshold value reported by the terminal device is 100 ms, the eighth preset threshold value set can be 70 ms (100 ms*0.7), and if the estimated second time length exceeds 70 ms, the access network device sends the sixth indication information to the terminal device to indicate the terminal device to start the guess reasoning function.

[0244] Alternatively, the server can indicate the QoS information to the access network device as a condition for starting the guess inference function. If the priority corresponding to the current QoS is the second preset priority, the sixth indication information is sent to the terminal device to instruct the terminal device to start the guess inference function. If the priority corresponding to the current QoS is other priorities, the terminal device does not need to be instructed to start the guess inference function. The way in which the server indicates the QoS information to the access network device includes but is not limited to that the server first notifies the core network device through the N33 interface, and the core network device then notifies the access network device through GTP-C signaling.

[0245] For example, assuming that the second preset priority is 2, if the priority corresponding to the QoS indicated by the server to the access network device is 1, the access network device does not need to instruct the terminal device to start the guess inference function; if the priority corresponding to the QoS indicated by the server to the access network device is 2, the access network device sends the sixth indication information to the terminal device to instruct the terminal device to start the guess inference function.

[0246] In a possible design, due to limited air interface resources, the access network device can send different proportions of data information at different sending durations; correspondingly, the terminal device also receives different proportions of data information at different receiving durations. Therefore, after the access network device sends the third data information to the terminal device, the access network device can continue to send other data information (i.e., fourth data information) of the data information corresponding to the third task to the terminal device in addition to the third data information. Of course, the access network device can also not continue to send other data information (such as fourth data information) of the data information corresponding to the third task to the terminal device in addition to the third data information. The fourth data information here can be all the data information of the data information corresponding to the third task in addition to the third data information, or can be part of the data information of the data information corresponding to the third task in addition to the third data information, which is not limited here. The following describes the two cases in detail:

[0247] Case 1: After the access network device sends the third data information to the terminal device, the access network device continues to send the fourth data information to the terminal device.

[0248] In a specific implementation, if the terminal device receives the fourth data information sent by the access network device subsequently, the terminal device can perform guess inference for the third task again based on the third data information and the fourth data information to obtain fourth guess inference information. Since more data information is used for guess inference, the inference result is more accurate, and therefore the terminal device can execute the task according to the fourth guess inference information.

[0249] Specifically, the method further includes steps s21-s23:

[0250] s21, the access network device sends fourth data information to the terminal device, the fourth data information being other data information than the third data information among the data information corresponding to the third task; correspondingly, the terminal device receives the fourth data information from the access network device.

[0251] s22, the terminal device performs guess inference of the third task based on the third data information and the fourth data information, and obtains fourth guess inference information.

[0252] s23, the terminal device performs the task based on the fourth guess inference information.

[0253] Exemplarily, taking a navigation task as an example, the access network device sends the terminal device the perception information 1, and the access network device performs guess inference of the navigation path planning task based on the perception information 1 to obtain a third navigation path; further, if the access network device continues to send the terminal device the perception information 2, the access network device will again perform guess inference of the navigation path planning task based on the perception information 1 and the perception information 2 to obtain a fourth navigation path. The terminal device performs the navigation task based on the fourth navigation path.

[0254] Case 2: After the access network device sends the terminal device the third data information, the access network device does not continue to send the terminal device the fourth data information.

[0255] In specific implementation, considering that some tasks only need part of data information to complete inference, and some inference tasks may have accuracy requirements, under certain conditions, in order to save power consumption and reduce time delay, the terminal device can instruct the access network device not to continue to send the subsequent data information, so as to avoid waste of resources.

[0256] Specifically, the conditions met here can adopt the following three ways:

[0257] Way 1: if the data amount corresponding to the third data information is greater than or equal to a sixth preset threshold, the terminal device sends fifth indication information to the access network device; wherein, the fifth indication information is used to indicate that the fourth data information is not sent, the fourth data information being other data information than the third data information among the data information corresponding to the third task; or, the fifth indication information is used to indicate that the priority of transmission of the data information corresponding to the third task is reduced.

[0258] In specific implementation, the sixth preset threshold value herein can be configured in combination with task requirements (for example, configured out of the factory or pre-configured by high-layer signaling of the device), or pre-set for different tasks, which is not limited herein. The configuration manner includes, but is not limited to, RRC signaling configuration. If the data amount corresponding to the third data information is greater than or equal to the sixth preset threshold value, the access network device sends fifth indication information to the terminal device. The fifth indication information can indicate that the access network device does not send the fourth data information, or can indicate to reduce the priority of the data information transmission corresponding to the third task. Since the higher the priority of data information transmission is, the higher the transmission speed or efficiency will be, the purpose of reducing the priority of the data information transmission corresponding to the third task is to reduce the transmission speed of the fourth data information, or even not to transmit the fourth data information.

[0259] Exemplarily, taking the perception information as an example, assuming that 90% of the perception information (i.e., the sixth preset threshold value) is received to complete the inference, when the data amount corresponding to the first data information is greater than or equal to 90%, the terminal device can send the fifth indication information to the access network device, indicating that the access network device does not send the fourth data information, or reduces the priority of the data information transmission corresponding to the third task.

[0260] Mode 2: If the inference accuracy of the third data information is greater than or equal to the seventh preset threshold value, the terminal device sends the fifth indication information to the access network device; wherein, the fifth indication information is used to indicate not to send the fourth data information, the fourth data information being other data information in the data information corresponding to the third task except the third data information; or, the fifth indication information is used to indicate to reduce the priority of the data information transmission corresponding to the third task.

[0261] In specific implementation, the seventh preset threshold value herein can be configured in combination with task requirements (for example, configured out of the factory or pre-configured by high-layer signaling of the device), or pre-set for different tasks, which is not limited herein. The configuration manner includes, but is not limited to, RRC signaling configuration. The inference accuracy herein refers to the accuracy of the guess inference model in performing the inference task on the data information (which can be the inference accuracy based on event window statistics), for example, assuming that 99 out of 100 inference tasks are correct, then the inference accuracy is 99%.

[0262] If the inference accuracy corresponding to the third data information is greater than or equal to a seventh preset threshold, the terminal device sends fifth indication information to the access network device. The fifth indication information can indicate that the access network device does not send the fourth data information, or can indicate that the priority of the data information transmission corresponding to the third task is reduced. Since the higher the priority of the data information transmission is, the higher the speed or efficiency of the data information transmission will be, the purpose of reducing the priority of the data information transmission corresponding to the third task is to reduce the transmission speed of the fourth data information, or even not to transmit the fourth data information.

[0263] Exemplarily, taking the perception information as an example, assuming that 90% of the perception information is received, and the seventh preset threshold is 90%, if the inference accuracy is greater than or equal to 90% at this time, the terminal device can send the fifth indication information to the access network device, indicating that the access network device does not send the fourth data information, or reduces the priority of the data information transmission corresponding to the third task.

[0264] Mode 3: If the ratio between the data amount of the third data information and the data amount of all data information corresponding to the third task is greater than or equal to a tenth preset threshold, the terminal device sends third indication information to the access network device; wherein the fifth indication information is used to indicate that the fourth data information is not sent, the fourth data information being other data information in the data information corresponding to the third task except the third data information; or the fifth indication information is used to indicate that the priority of the data information transmission corresponding to the third task is reduced.

[0265] In a specific implementation, the tenth preset threshold herein can be configured in combination with task requirements (for example, configured by a device out of the factory or pre-configured by a high-layer signaling of the device), or can be pre-set for different tasks, which is not limited herein. The configuration manner includes but is not limited to RRC signaling configuration. If the ratio between the data amount of the third data information and the data amount of all data information corresponding to the third task is greater than or equal to the tenth preset threshold, the terminal device sends the fifth indication information to the access network device. The fifth indication information can indicate that the access network device does not send the fourth data information, or can indicate that the priority of the data information transmission corresponding to the third task is reduced. Since the higher the priority of the data information transmission is, the higher the speed or efficiency of the data information transmission will be, the purpose of reducing the priority of the data information transmission corresponding to the third task is to reduce the transmission speed of the fourth data information, or even not to transmit the fourth data information.

[0266] Exemplarily, taking a video frame as an example, it is assumed that 90 video frames (i.e., third data information) are received, the total video frames corresponding to the third task are 100 frames (i.e., total data information), the tenth preset threshold is 0.9, and the ratio between the third data information and the total data information is 0.9 (greater than or equal to the tenth preset threshold) at this time. Then, the terminal device can send fifth indication information to the access network device, indicating that the access network device does not send the fourth data information, or the priority of the data information corresponding to the third task is reduced.

[0267] It can be seen that, based on the method described in FIG. 7, when the terminal device has an inference task, if the computing power of the terminal device is sufficient, the terminal device can also execute the inference task by itself. Among them, the terminal device can deploy a guess inference model. Due to limited air interface resources and long complete data transmission time, the third data information here is only part of the data information related to the third task sent by the access network device. After receiving the third data information, the terminal device does not need to wait for the access network device to send the complete data information before executing the inference of the third task, but can directly infer the inference result according to the limited data information by using the guess inference model; or the terminal device can first guess the possible data information after the third data information by using the guess inference model, and then execute the inference of the third task by using the third data information and the guessed possible data information after the third data information. The whole process can be considered as a guess inference process, and the third guess inference information is obtained. The terminal device can execute the task according to the third guess inference information. In this way, it is beneficial to reduce the overall inference delay and improve the inference efficiency.

[0268] The following will introduce the apparatus provided by the embodiments of the present application.

[0269] FIG. 8 shows a possible exemplary block diagram of a communication apparatus involved in the embodiments of the present application. As shown in FIG. 8, the communication apparatus 800 can include modules or units for implementing the above-mentioned method embodiments. In one possible design, the communication apparatus 800 includes a processing unit 802 and a communication unit 803. Optionally, the communication apparatus 800 can further include a storage unit 801 for storing apparatus program code and / or data.

[0270] The communication apparatus 800 can be a network side apparatus in the above-mentioned embodiments, for example, an access network device or a communication module in the access network device, or a circuit or chip responsible for communication function in the access network device.

[0271] For example, in one embodiment, the communication unit 803 is configured to receive a first request and first data information from a terminal device, the first request being used to request the access network device to execute inference of a first task.

[0272] The processing unit 802 is configured to perform guess inference on the first task based on the first data information, to obtain first guess inference information.

[0273] The communication unit 803 is further configured to send the first guess inference information to the server.

[0274] In a possible design, the first data information is part of the data information corresponding to the first task.

[0275] In a possible design, the communication unit 803 is further configured to receive second data information from the terminal device, the second data information being other data information in the data information corresponding to the first task and excluding the first data information.

[0276] In a possible design, the processing unit 802 is further configured to perform guess inference on the first task based on the first data information and the second data information, to obtain second guess inference information; and the communication unit 803 is further configured to send first indication information to the server if a similarity between the second guess inference information and the first guess inference information is less than a first preset threshold, the first indication information being used to instruct to re-perform inference on the task, and the first indication information including the second guess inference information.

[0277] In a possible design, the communication unit 803 is further configured to send second indication information to the terminal device, the second indication information being used to instruct to increase a priority of transmission of the data information corresponding to the first task.

[0278] In a possible design, the communication unit 803 is further configured to send third indication information to the terminal device if a data amount corresponding to the first data information is greater than or equal to a second preset threshold, or an inference accuracy corresponding to the first data information is greater than or equal to a third preset threshold, or a ratio between the data amount corresponding to the first data information and a data amount of all the data information corresponding to the first task is greater than or equal to a ninth preset threshold; wherein the third indication information is used to instruct not to send second data information, the second data information being other data information in the data information corresponding to the first task and excluding the first data information; or the third indication information is used to instruct to reduce the priority of transmission of the data information corresponding to the first task.

[0279] In a possible design, before the processing unit 802 performs guess inference on the first task based on the first data information, the processing unit 802 is further configured to start a guess inference function by the access network device if a first time length is greater than a fourth preset threshold or a priority corresponding to QoS is a first preset priority, the first time length being a time length of receiving uplink data estimated based on network load information, and the fourth preset threshold being associated with an inference time delay threshold indicated by the terminal device.

[0280] In a possible design, the communication unit 803 is further configured to receive fourth indication information from the server, where the fourth indication information is used to indicate a window length corresponding to the first guess inference information.

[0281] In a possible design, the first request is further used to request the server to perform inference of a second task; and the communication unit 803 is specifically configured to send, when sending the first guess inference information to the server, a second request and the first guess inference information to the server, where the second request is used to request the server to perform inference of the second task.

[0282] In a possible design, the first task is associated with a capability of the access network device, and the second task is associated with a capability of the server.

[0283] In a possible design, when the communication apparatus 800 is an access network device or a communication module in the access network device, the function of the processing unit 802 can be implemented by one or more processors. The function of the communication unit 803 can be implemented by a transceiver circuit.

[0284] In a possible design, when the communication apparatus 800 is a circuit or chip responsible for communication functions in an access network device, the function of the processing unit 802 can be implemented by a circuit system including one or more processors or processor cores in the chip. The function of the communication unit 803 can be implemented by an interface circuit or a data transceiver circuit on the chip.

[0285] In a possible design, when the communication apparatus 800 is an access network device or a processing module in the access network device, the function of the processing unit 802 can be implemented by one or more processors. Specifically, the processor can include an AI processor. The function of the communication unit 803 can be implemented by a transceiver circuit.

[0286] In a possible design, when the communication apparatus 800 is a circuit or chip responsible for processing functions in an access network device, the function of the processing unit 802 can be implemented by a circuit system including one or more processors or processor cores in the chip. The function of the communication unit 803 can be implemented by an interface circuit or a data transceiver circuit on the chip.

[0287] The communication apparatus 800 can be a network side device in the above-described embodiments, for example, a server or a communication module in the server, or a circuit or chip responsible for communication functions in the server.

[0288] For example, in an embodiment, the communication unit 803 is configured to receive first guess inference information from an access network device.

[0289] The processing unit 802 is configured to perform inference of the second task based on the first guess inference information, to obtain first inference information.

[0290] In a possible design, the communication unit 803 is further configured to send, to the terminal device, the first inference information.

[0291] In a possible design, the communication unit 803 is further configured to receive, from the access network device, first indication information, where the first indication information is used to instruct to re-perform inference of the task, and the first indication information includes second guess inference information; and the processing unit 802 is further configured to re-perform inference of the second task based on the second guess inference information, to obtain second inference information; and the communication unit 803 is further configured to send, to the terminal device, the second inference information.

[0292] In a possible design, the communication unit 803 is further configured to send, to the access network device, fourth indication information, where the fourth indication information is used to instruct a window length corresponding to the first guess inference information.

[0293] In a possible design, when the communication unit 803 receives the first guess inference information from the access network device, the communication unit 803 is specifically configured to receive, from the access network device, a second request and the first guess inference information, where the second request is used to request the server to perform inference of the second task.

[0294] In a possible design, the second task is associated with a capability of the server.

[0295] In a possible design, when the communication apparatus 800 is a server or a communication module in a server, the function of the processing unit 802 can be implemented by one or more processors. Specifically, the processor can include a modem chip, or a system on chip (SoC) chip or a SIP chip including a modem core. The function of the communication unit 803 can be implemented by a transceiver circuit.

[0296] In a possible design, when the communication apparatus 800 is a circuit or chip responsible for communication functions in a server, such as a modem chip or a system on chip (SoC) chip or a SIP chip including a modem core, the function of the processing unit 802 can be implemented by a circuit system including one or more processors or processor cores in the chip. The function of the communication unit 803 can be implemented by an interface circuit or a data transceiver circuit on the chip.

[0297] In a possible design, when the communication apparatus 800 is a server or a processing module in a server, the function of the processing unit 802 can be implemented by one or more processors. Specifically, the processor can include a GPU, or a system on chip (SoC) chip or a SIP chip including a GPU. Alternatively, the processor can include an AI processor, or a SoC chip or a SIP chip including an AI processor. Alternatively, the processor can include an ASIC, or a SoC chip or a SIP chip including an ASIC. The function of the communication unit 803 can be implemented by transceiver circuitry.

[0298] In a possible design, when the communication apparatus 800 is a circuit or chip responsible for processing functions in a server, such as a GPU or a system on chip (SoC) chip or a SIP chip including a GPU, an AI processor or a SoC chip or a SIP chip including an AI processor, or an ASIC or a SoC chip or a SIP chip including an ASIC, the function of the processing unit 802 can be implemented by circuitry including one or more processors or processor cores in the chip. The function of the communication unit 803 can be implemented by interface circuitry or data transceiver circuitry on the chip.

[0299] The communication apparatus 800 can be a terminal device side apparatus in the embodiments, for example, a terminal device or a communication module in a terminal device, or a circuit or chip responsible for communication functions in a terminal device.

[0300] For example, in an embodiment, the communication unit 803 is configured to: send, to an access network device, a first request and first data information, where the first request is used to request the access network device to perform reasoning of a first task.

[0301] In a possible design, the first data information is part of data information corresponding to the first task.

[0302] In a possible design, the communication unit 803 is further configured to: send, to the access network device, second data information, where the second data information is other data information than the first data information in the data information corresponding to the first task.

[0303] In a possible design, the communication unit 803 is further configured to: receive second indication information from the access network device, where the second indication information is used to indicate that a priority of transmission of the data information corresponding to the first task is to be improved.

[0304] In a possible design, the communication unit 803 is further configured to receive third indication information from the access network device, where the third indication information is used to indicate that the second data information, which is data information corresponding to the first task and excluding the first data information, is not to be transmitted, or the third indication information is used to indicate that the priority of transmission of the data information corresponding to the first task is to be reduced.

[0305] In a possible design, the communication unit 803 is further configured to receive the first inference information or the second inference information from the server.

[0306] In a possible design, the first request is further used to request the server to perform inference of a second task, where the first task is associated with a capability of the access network device, and the second task is associated with a capability of the server.

[0307] In a possible design, when the communication apparatus 800 is a terminal device or a communication module in a terminal device, the function of the processing unit 802 can be implemented by one or more processors. Specifically, the processor can include a modem chip, or a system on chip (SoC) chip or a SIP chip that includes a modem core. The function of the communication unit 803 can be implemented by a transceiver circuit.

[0308] In a possible design, when the communication apparatus 800 is a circuit or chip responsible for communication functions in a terminal device, such as a modem chip or a system on chip (SoC) chip or a SIP chip that includes a modem core, the function of the processing unit 802 can be implemented by a circuit system that includes one or more processors or processor cores in the chip. The function of the communication unit 803 can be implemented by an interface circuit or a data transceiver circuit on the chip.

[0309] In a possible design, when the communication apparatus 800 is a terminal device or a processing module in a terminal device, the function of the processing unit 802 can be implemented by one or more processors. Specifically, the processor can include a GPU, or a system on chip (SoC) chip or a SIP chip that includes a GPU. Alternatively, the processor can include an AI processor, or a SoC chip or a SIP chip that includes an AI processor. Alternatively, the processor can include an ASIC, or a SoC chip or a SIP chip that includes an ASIC. The function of the communication unit 803 can be implemented by a transceiver circuit.

[0310] In a possible design, when the communication apparatus 800 is a circuit or chip responsible for processing functions in a terminal device, such as a GPU or a system on chip (SoC) chip or a system in package (SIP) chip containing a GPU, an AI processor or a SoC chip or a SIP chip containing an AI processor, or an ASIC or a SoC chip or a SIP chip containing an ASIC, the function of the processing unit 802 can be implemented by circuitry including one or more processors or processor cores in the chip. The function of the communication unit 803 can be implemented by an interface circuit or a data transceiver circuit on the chip.

[0311] The communication apparatus 800 can be a terminal device side apparatus in the above-described embodiments, for example, a terminal device or a communication module in a terminal device, or a circuit or chip responsible for communication functions in a terminal device.

[0312] For example, in an embodiment, the communication unit 803 is configured to: send, to an access network device, a third request, where the third request is used to request data information corresponding to a third task; and receive, from the access network device, third data information, where the third data information is part of the data information corresponding to the third task.

[0313] The processing unit 802 is configured to: perform, based on the third data information, guess inference of the third task to obtain third guess inference information.

[0314] In a possible design, the processing unit 802 is further configured to: perform a task based on the third guess inference information.

[0315] In a possible design, the communication unit 803 is further configured to: receive, from the access network device, fourth data information, where the fourth data information is other data information in the data information corresponding to the third task except the third data information.

[0316] In a possible design, the processing unit 802 is further configured to: perform, by the terminal device, guess inference of the third task based on the third data information and the fourth data information to obtain fourth guess inference information; and then perform a task based on the fourth guess inference information.

[0317] In a possible design, the communication unit 803 is further configured to: if the data amount corresponding to the third data information is greater than or equal to a sixth preset threshold, or the inference accuracy corresponding to the third data information is greater than or equal to a seventh preset threshold, or the ratio between the data amount corresponding to the third data information and the data amount of all data information corresponding to the third task is greater than or equal to a tenth preset threshold, send fifth indication information to the access network device; and the fifth indication information is used to indicate that fourth data information is not to be sent, the fourth data information being other data information in the data information corresponding to the third task except the third data information; or the fifth indication information is used to indicate that the priority of transmission of the data information corresponding to the third task is to be reduced.

[0318] In a possible design, before the processing unit 802 performs the speculative inference of the third task based on the third data information, the communication unit 803 is further configured to: receive sixth indication information from the access network device, the sixth indication information being used to indicate that the speculative inference function is to be started; and then start the speculative inference function according to the sixth indication information.

[0319] In a possible design, when the communication apparatus 800 is a terminal device or a communication module in a terminal device, the function of the processing unit 802 can be implemented by one or more processors. Specifically, the processor can include a modem chip, or a system on chip (SoC) chip or a SIP chip containing a modem core. The function of the communication unit 803 can be implemented by a transceiver circuit.

[0320] In a possible design, when the communication apparatus 800 is a circuit or chip responsible for communication functions in a terminal device, such as a modem chip or a system on chip (SoC) chip or a SIP chip containing a modem core, the function of the processing unit 802 can be implemented by a circuit system including one or more processors or processor cores in the chip. The function of the communication unit 803 can be implemented by an interface circuit or a data transceiver circuit on the chip.

[0321] In a possible design, when the communication apparatus 800 is a terminal device or a processing module in a terminal device, the function of the processing unit 802 can be implemented by one or more processors. Specifically, the processor can include a GPU, or a system on chip (SoC) chip or a SIP chip containing a GPU. Alternatively, the processor can include an AI processor, or a SoC chip or a SIP chip containing an AI processor. Alternatively, the processor can include an ASIC, or a SoC chip or a SIP chip containing an ASIC. The function of the communication unit 803 can be implemented by a transceiver circuit.

[0322] In a possible design, when the communication apparatus 800 is a circuit or chip responsible for processing functions in a terminal device, such as a GPU or a system on chip (SoC) chip or a system in package (SIP) chip containing a GPU, an AI processor or a SoC chip or a SIP chip containing an AI processor, or an ASIC or a SoC chip or a SIP chip containing an ASIC, the function of the processing unit 802 can be implemented by circuitry including one or more processors or processor cores in the chip. The function of the communication unit 803 can be implemented by interface circuitry or data transceiver circuitry on the chip.

[0323] The communication apparatus 800 can be a network-side device in the above-described embodiments, for example, an access network device or a communication module in an access network device, or a circuit or chip responsible for communication functions in an access network device.

[0324] For example, in an embodiment, the communication unit 803 is configured to: receive a third request from the terminal device, where the third request is used to request data information corresponding to a third task; and send, to the terminal device, third data information, where the third data information is part of the data information corresponding to the third task.

[0325] In a possible design, the communication unit 803 is further configured to: send, to the terminal device, fourth data information, where the fourth data information is other data information in the data information corresponding to the third task than the third data information.

[0326] In a possible design, the communication unit 803 is further configured to: receive fifth indication information from the terminal device, where the fifth indication information indicates that the fourth data information is not to be sent, or the fifth indication information indicates that a priority of transmission of the data information corresponding to the third task is to be lowered.

[0327] In a possible design, the communication unit 803 is further configured to: if a second time length is greater than an eighth preset threshold value, or a priority corresponding to the QoS is a second preset priority, send, to the terminal device, sixth indication information, where the sixth indication information is used to indicate that the guessing and reasoning function is started, and the second time length is a time length of receiving downlink data estimated based on network load information, and the eighth preset threshold value is associated with a reasoning time delay threshold value indicated by the terminal device.

[0328] In a possible design, when the communication apparatus 800 is an access network device or a communication module in an access network device, the function of the processing unit 802 can be implemented by one or more processors. The function of the communication unit 803 can be implemented by transceiver circuitry.

[0329] In a possible design, when the communication apparatus 800 is a circuit or a chip responsible for communication functions in an access network device, the function of the processing unit 802 can be implemented by circuitry including one or more processors or processor cores in the chip. The function of the communication unit 803 can be implemented by interface circuitry or data transceiver circuitry on the chip.

[0330] In a possible design, when the communication apparatus 800 is an access network device or a processing module in an access network device, the function of the processing unit 802 can be implemented by one or more processors. Specifically, the processor can include an AI processor. The function of the communication unit 803 can be implemented by transceiver circuitry.

[0331] In a possible design, when the communication apparatus 800 is a circuit or a chip responsible for processing functions in an access network device, the function of the processing unit 802 can be implemented by circuitry including one or more processors or processor cores in the chip. The function of the communication unit 803 can be implemented by interface circuitry or data transceiver circuitry on the chip.

[0332] Referring to FIG. 9, shown is a structural schematic diagram of a communication apparatus 900 provided by an embodiment of the present application. The communication apparatus 900 can correspond to the access network device shown in FIG. 1, and be configured to implement the operations of the access network device in the above embodiments. Alternatively, the communication apparatus 900 can correspond to the server shown in FIG. 1, and be configured to implement the operations of the server in the above embodiments. Alternatively, the communication apparatus 900 can correspond to the communication apparatus shown in FIG. 1, and be configured to implement the operations of the communication apparatus in the above embodiments. As shown in FIG. 9, the communication apparatus includes one or more antennas 910, a radio frequency processing system 920, and a processor system 930.

[0333] In the downlink or sidelink direction, the radio frequency processing system 920 receives radio frequency signals through the antenna 910, and sends the signals processed by radio frequency to the processor system 930 for further processing. In the uplink or sidelink direction, the processor system 930 performs signal processing on information of the communication apparatus, and sends the signals to the radio frequency processing system 920, which processes the signals by radio frequency and transmits the signals through the antenna 910.

[0334] In one example, the radio frequency processing system 920, as a communication interface for the communication device to communicate with the outside, can include a radio frequency front end 921 (RFFE) and a radio frequency transceiver 922. The RFFE 921 is mainly used for one or more of shaping, passband selection, or gain processing of RF signals received by an antenna or to be transmitted through an antenna, and can include one or more of radio frequency switches, duplexers, filters, power amplifiers, antenna tuning, and low noise amplifiers. The RFFE 921 can be circuitry composed of a plurality of discrete devices, or can be integrated and packaged in one or more chips. The radio frequency transceiver 922 is used to process the RF signals received by the RFFE into baseband / intermediate frequency signals for further processing by the processor system 930, and to process the baseband / intermediate frequency signals provided by the processor system 930 into RF signals for transmission to the RFFE 921. The baseband / intermediate frequency signals transmitted between the radio frequency transceiver 922 and the processor system 930 can be digital signals or analog signals. The radio frequency transceiver 922 can be implemented by one or more chips, which are usually referred to as radio frequency chips (RFIC).

[0335] In one example, the processor system 930 can include one or more processors for processing signals and executing one or more communication protocols. Optionally, the processor system 930 can also include a memory 936. In one example, the one or more processors include at least one baseband processor 931 (also referred to as a modem processor). The memory 936 is used to store data and / or computer program instructions. Optionally, the processor system 930 can also include one or more application processors 932 for implementing processing of the operating system and application layer of the communication device. The application processor 932 can include, for example, a GPU, an AI processor, or an ASIC. Optionally, the processor system 930 can also include one or more of a voice subsystem 933, a multimedia subsystem 934, or an interface circuit 935. The voice subsystem 933 is used to process voice signals, the multimedia subsystem 934 is used to process multimedia related operations such as video encoding and decoding, image processing, etc., and the interface circuit 935 is used to implement communication with other communication device components such as a display 940, an input device 950, a memory 960, etc. The above-mentioned components in the processor system 930 can communicate with each other through a bus or a communication interface circuit.

[0336] In one example, the processor system 930 can be packaged as a processor chip, such as a SoC chip or a SIP chip. In one example, the processor system 930 can be a system composed of multiple chips, for example, the baseband processor 931 can be packaged as a separate chip, or packaged as a chip together with part or all of the circuit of the radio frequency processing system.

[0337] In one example, the memory 936 can be an on-chip memory, i.e., located on the chip of the processor system 930. In one example, the memory 960 can be an off-chip memory, i.e., located off the chip of the processor system 930.

[0338] In one example, the baseband processor 931 can include one or more processor cores 9311 and an interface circuit 9314. The one or more processor cores 9311 are configured to process signals and perform one or more communication protocols. Optionally, the baseband processor 931 can further include a memory 9312 configured to store at least part of corresponding computer program instructions and / or data. In one example, the one or more processor cores 9311 implement the relevant operations in the above-described method embodiments by executing the computer program instructions stored in the memory 9312. In this application, the memory 9312 configured to store corresponding computer program instructions and / or data can mean that the memory 9312 is configured to store all corresponding computer program instructions and / or data for execution by the processor core 9311; or can mean that the memory 9312 is configured to store part of corresponding computer program instructions and / or data, which includes computer program instructions and / or data currently needed for execution by the processor core 9311, and the memory 9312 can store different parts of computer program instructions and / or data for execution by the processor core 9311 multiple times to implement the relevant operations in the above-described method embodiments. The interface circuit 9314 serves as a communication interface to communicate with other components, such as transmitting signals with the radio frequency processing system 920, communicating with other subsystems and related components of the processor system 930 through a bus, such as transmitting data control signals with the application processor 932 and transmitting data or computer program instructions with the memory 936 or the memory 960. Optionally, in order to reduce the load of the processor core, a baseband signal processing circuit 9313 can be further provided to implement at least part of the processing of the baseband signals, including one or more of demodulation, modulation, encoding or decoding of the signals.

[0339] In one example, the communication apparatus provided in this application can be the communication apparatus 900, including the communication module of the processor system 930 and the radio frequency processing system 920, the processor system 930, or the baseband processor 931.

[0340] The processor, processor system, application processor, baseband processor, processor circuit, or processor core can be collectively referred to as a processor, which can include one or a combination of a central processing unit (CPU), a digital signal processor (DSP), a microprocessor unit (MPU), a microcontroller unit (MCU), a graphics processing unit (GPU), a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), an artificial intelligence processor (AI processor), or a neural processing unit (NPU).

[0341] The above-mentioned memory can include one or more of the following storage media: random access memory (RAM), static RAM (SRAM), dynamic RAM (DRAM), phase-change memory (PCM), resistive RAM (ReRAM), magnetoresistive RAM (MRAM), ferroelectric RAM (FRAM), cache, register, read-only memory (ROM), flash memory, erasable programmable ROM (EPROM), hard disk, etc. In one example, computer program instructions for implementing the above-embodied examples can be stored on a non-volatile memory, such as at least part of the above-mentioned memory 960 (e.g., one or more of ROM, flash memory, EPROM, or hard disk). During operation of the communication device, the corresponding computer program instructions can be loaded in whole or in part into a memory with faster transmission speed than the processor, such as at least part of the above-mentioned memory 936 and / or memory 9312 (e.g., one or more of RAM, SRAM, DRAM, PCM, RERAM, MRAM, FRAM, cache, or register), for execution by the processor to implement the steps in the above-mentioned method embodiments.

[0342] In one example, the radio frequency transceiver 922 and the radio frequency front end 921 can also be packaged in one chip. In one example, the radio frequency transceiver 922, the radio frequency front end 921, and the baseband processor 931 can also be packaged in one chip.

[0343] The embodiments of the present application also provide a communication system, which includes an access network device, a server and a terminal device, and the access network device, the server and the terminal device can be used to execute the method in any of the above-embodied examples.

[0344] The embodiments of the present application also provide a communication system, which includes an access network device and a terminal device, and the access network device and the terminal device can be used to execute the method in any of the above-embodied examples.

[0345] In addition, the present application also provides a computer program for implementing the operations and / or processes performed by various devices in the methods provided by the present application.

[0346] The application further provides a computer readable storage medium, wherein computer code is stored in the computer readable storage medium, and when the computer code is run on a computer, the computer code causes the computer to perform operations and / or processes performed by various devices in the method provided by the application.

[0347] The application further provides a computer program product, which comprises computer code or a computer program, and when the computer code or the computer program is run on a computer, operations and / or processes performed by various devices in the method provided by the application are performed.

[0348] The terms "system" and "network" in the embodiments of the application can be used interchangeably. "At least one" refers to one or more, and "multiple" refers to two or more. "And / or" describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent the cases of A alone, A and B together, and B alone, where A and B can be singular or plural. The character " / " generally represents that the associated objects before and after are in an "or" relationship. "At least one of the following" or the like refers to any combination of these items, including any combination of single or multiple items. For example, "at least one of A, B, or C" includes A, B, C, AB, AC, BC, or ABC, and "at least one of A, B, and C" can also be understood to include A, B, C, AB, AC, BC, or ABC. In addition, unless otherwise specified, the ordinal numbers "first", "second", etc. mentioned in the embodiments of the application are used to distinguish multiple objects, and are not used to limit the order, time sequence, priority, or importance of the multiple objects.

[0349] Those skilled in the art should understand that the embodiments of the application can be provided as a method, a system, or a computer program product. Therefore, the application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, optical storage, etc.) containing computer usable program code.

[0350] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks.

[0351] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks.

[0352] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks.

[0353] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A communication method characterized by comprising: The method comprises: receiving a first request and first data information from a terminal device, the first request being used to request an access network device to perform inference of a first task; performing inference of the first task based on the first data information to obtain first inference information; sending the first inference information to a server.

2. The method of claim 1, wherein, The first data information is part of data information corresponding to the first task.

3. The method according to claim 1 or 2, characterized in that, The method further comprises: receiving second data information from the terminal device, the second data information being other data information than the first data information in the data information corresponding to the first task.

4. The method of claim 3, wherein, The method further comprises: performing inference of the first task based on the first data information and the second data information to obtain second inference information; if a similarity between the second inference information and the first inference information is less than a first preset threshold, sending first indication information to the server, the first indication information being used to indicate that inference of the task is to be re-executed, and the first indication information comprising the second inference information.

5. The method of claim 4, wherein, The method further comprises: sending second indication information to the terminal device, the second indication information being used to indicate that a priority of transmission of the data information corresponding to the first task is to be increased.

6. The method of claim 1 or 2, wherein, The method further comprises: if a data amount corresponding to the first data information is greater than or equal to a second preset threshold, or an inference accuracy corresponding to the first data information is greater than or equal to a third preset threshold, or a ratio between the data amount corresponding to the first data information and a data amount of all data information corresponding to the first task is greater than or equal to a ninth preset threshold, sending third indication information to the terminal device; wherein the third indication information is used to indicate that the second data information is not to be sent, the second data information being other data information than the first data information in the data information corresponding to the first task; or the third indication information is used to indicate that the priority of transmission of the data information corresponding to the first task is to be reduced.

7. The method according to any one of claims 1 to 6, characterized in that, The first request is also used to request the server to perform inference of a second task; The method comprises: sending second request and the first inference information to the server, the second request being used to request the server to perform inference of a second task.

8. A communication method characterized by comprising: The method comprises: receiving first inference information from an access network device; performing inference of a second task based on the first inference information to obtain first inference information.

9. The method of claim 8, wherein, The method further comprises: sending the first inference information to a terminal device.

10. The method of claim 8, wherein, The method further comprises: receiving first indication information from the access network device, the first indication information being used to indicate that inference of the task is to be re-executed, and the first indication information comprising second inference information; re-performing inference of the second task based on the second inference information to obtain second inference information; sending the second inference information to a terminal device.

11. The method according to any one of claims 8-10, characterized in that, The receiving first inference information from an access network device comprises: receiving a second request from the access network device, the second request being for requesting the server to perform an inference of a second task.

12. A communications device, characterized by The communication apparatus comprises modules or units for performing the method of any one of claims 1-7.

13. A communications device, characterized by The communication apparatus comprises modules or units for performing the method of any one of claims 8-11.

14. A computer-readable storage medium, characterized in that, The computer readable storage medium is configured to store a computer program, which, when executed, causes the method of any one of claims 1-7 to be performed, or the method of any one of claims 8-11 to be performed.

15. A computer program product, characterised in that, The computer program product, when executed, causes the method of any one of claims 1-7 to be performed, or the method of any one of claims 8-11 to be performed.

16. A communications device, characterized by The apparatus comprises one or more processors coupled with a memory, the memory configured to store a computer program or instructions, which, when executed by the one or more processors, cause the apparatus to implement the method of any one of claims 1-7.

17. A communications device, characterized by The apparatus comprises one or more processors coupled with a memory, the memory configured to store a computer program or instructions, which, when executed by the one or more processors, cause the apparatus to implement the method of any one of claims 8-11.

18. The apparatus of claim 16 or 17, wherein, The interface circuitry is configured to implement communication functions within the apparatus and / or communication functions of the apparatus with other apparatuses or components.

Images