WO2023193579A1 - Data transmission method and apparatus - Google Patents

Abstract

Provided in the embodiments of the present application are a data transmission method and apparatus. The method comprises: receiving a general packet radio service tunneling protocol-user plane (GTP-U) data packet, wherein a load of the GTP-U data packet comprises a neural network data packet, and a packet header comprises indication information for indicating the priority of the neural network data packet; and transmitting the neural network data packet according to the indication information. By means of carrying, in a GTP-U data packet, indication information for indicating the priorities of neural network data packets, an access network device, which receives the GTP-U data packet, can transmit the neural network data packets according to the indication information, so as to realize differentiated transmission of neural network data packets with different priorities.

Description

Data transmission methods and devices

This application claims priority to the Chinese patent application filed with the China Patent Office on April 8, 2022, with the application number 202210365108.5 and the application title "Data transmission method and device", the entire content of which is incorporated into this application by reference.

Technical field

The embodiments of the present application relate to the field of communications, and more specifically, to a data transmission method and device.

Background technique

In recent years, with the continuous advancement and improvement of extended reality (XR) technology, related industries have developed vigorously. Today, XR technology has entered various fields closely related to people's production and life, such as education, entertainment, military, medical care, environmental protection, transportation, and public health. Among them, XR is the general term for various reality-related technologies, including: virtual reality (VR), augmented reality (AR) and mixed reality (MR). Through the rendering of vision and hearing, it brings users an "immersive experience" of virtual scenes and real scenes.

In XR technology, providing high-resolution images (or videos) improves user experience. For example, low-resolution images can be converted into high-resolution images through super resolution (SR) technology. SR technology based on neural network (NN) has received widespread attention because of its remarkable image restoration effect. Users convert low-resolution images into high-resolution images based on neural networks, so how to transmit neural network data packets has become a focus The problem.

Contents of the invention

Embodiments of the present application provide a communication method in order to realize differentiated transmission of neural network data packets of different priorities and improve user experience.

The first aspect provides a communication method, which can be executed by an access network device, or can be executed by a chip or circuit configured in the access network device, or can also be executed by a device that can realize all or part of the connection. The logic module or software execution of the network access device function is not limited in this application. For convenience of description, the following description takes execution by the access network device as an example.

The method includes: receiving a general packet radio service tunneling protocol-user plane (GTP-U) data packet, the payload of the GTP-U data packet includes a neural network data packet, and the GTP-U data packet The packet header includes indication information, the indication information is used to indicate the priority of the neural network data packet; the neural network data packet is transmitted according to the indication information.

Based on the above technical solution, the header of the GTP-U data packet received by the access network device includes instruction information indicating the neural network data packet, so that the access network device can read the instruction from the header of the GTP-U data packet. information, and transmit and process the neural network data packet based on the priority of the neural network data packet indicated by the indication information. It can be understood that the access network device takes the priority of the neural network data packets into consideration when transmitting the neural network data packets. It facilitates differentiated transmission of neural network data packets of different priorities.

For example, the access network device receives two GTP-U data packets (GTP-U data packet #1 and GTP-U data packet #2). Among them, the payload of GTP-U data packet #1 includes a neural network data packet. #1, the header of GTP-U packet #1 includes indication information #1, the payload of GTP-U packet #2 includes neural network packet #2, the header of GTP-U packet #2 includes indication information #2. The access network device differentially transmits neural network data packet #1 and neural network data packet #2 according to instruction information #1 and instruction information #2. For example, instruction information #1 indicates that the priority of neural network data packet #1 is high priority. Level, indication information #2 indicates that the priority of neural network data packet #2 is low priority. The access network device can transmit neural network data packet #1 first, and then transmit neural network data after the transmission of neural network data packet #1 is completed. Package #2.

In connection with the first aspect, in some implementations of the first aspect, transmitting the neural network data packet according to the indication information includes: when the indication information indicates that the neural network data packet is a high priority, transmitting the neural network data packet first Neural network data packet; or, when the indication information indicates that the neural network data packet is of low priority, delay the transmission of the neural network data packet; or, when the indication information indicates that the neural network data packet is of low priority , and when the network status is congested, the transmission of the neural network data packet is given up.

Based on the above technical solution, the access network equipment can determine the transmission mode of different neural network data packets (such as priority transmission or abandonment of transmission, etc.) according to the priority of the neural network data packet indicated by the indication information, thereby improving the flexibility of the solution.

The second aspect provides a communication method, which can be executed by a core network device, or can be executed by a chip or circuit configured in the core network device, or can also be executed by a core network device that can implement all or part of the core network device. Functional logic modules or software execution are not limited in this application. For convenience of description, the following description takes execution by the core network device as an example.

The method includes: obtaining a neural network data packet, the neural network data packet including indication information, the indication information being used to indicate the priority of the neural network data packet; sending a General Packet Wireless Service Tunneling Protocol user plane GTP to the access network device -U data packet, the payload of the GTP-U data packet includes the neural network data packet, and the header of the GTP-U data packet includes the indication information.

Based on the above technical solution, after the core network device receives the neural network data packet carrying the indication information indicating the priority of the neural network data packet, it can read the priority of the neural network data packet from the neural network data packet. level indication information, and encapsulates the indication information into the GTP-U data packet header according to the GTP-U protocol, and uses the neural network data packet as the payload of the GTP-U data packet. So that the access network device that receives the GTP-U data packet can read the indication information from the header of the GTP-U data packet, and based on the priority of the neural network data packet indicated by the indication information, the neural network Data packets are processed for transmission. It can be understood that the access network device takes the priority of the neural network data packet into consideration when transmitting the neural network data packet, so as to achieve differentiated transmission of neural network data packets of different priorities.

The third aspect provides a communication method, which can be executed by a server, or can be executed by a chip or circuit configured in the server, or can also be executed by a logic module or software that can realize all or part of the server functions. execution, this application does not limit this. For the convenience of description, the following description takes execution by the server as an example.

The method includes: generating a neural network data packet, the neural network data packet including indication information, the indication information being used to indicate the priority of the neural network data packet; and sending the neural network data packet.

Based on the above technical solution, when the server generates a neural network data packet, it can carry indication information indicating the priority of the neural network data packet in the neural network data packet, so as to facilitate the core network equipment that receives the neural network data packet. The device can read the indication information used to indicate the priority of the neural network data packet from the neural network data packet, and learn the priority of the neural network data packet. In order to achieve differentiated transmission of neural network data packets of different priorities.

A fourth aspect provides a communication device, which is used to perform the method provided in the first aspect. The device may be access network equipment, or may be a component of access network equipment (such as a processor, chip, or chip system, etc.), or may be a logic module or software that can realize all or part of the functions of the access network equipment, The device includes:

The interface unit is configured to receive a General Packet Radio Service Tunneling Protocol user plane GTP-U data packet. The payload of the GTP-U data packet includes a neural network data packet. The header of the GTP-U data packet includes indication information. The indication The information is used to indicate the priority of the neural network data packet; the processing unit is used to control the device to transmit the neural network data packet according to the indication information.

In connection with the fourth aspect, in some implementations of the fourth aspect, the processing unit controls the device to transmit the neural network data packet according to the indication information, including: when the indication information indicates that the neural network data packet is a high priority In this case, the processing unit controls the device to transmit the neural network data packet with priority; or, in the case where the indication information indicates that the neural network data packet is of low priority, the processing unit controls the device to delay transmission of the neural network data. packet; or, when the indication information indicates that the neural network data packet is of low priority and the network status is congested, the processing unit controls the device to give up transmitting the neural network data packet.

Specifically, the communication device may include units and/or modules for executing the method provided by any implementation of the first aspect, such as a processing unit and an interface unit.

In one implementation, the interface unit may be a transceiver, or an input/output interface; the processing unit may be at least one processor. Alternatively, the transceiver may be a transceiver circuit. Alternatively, the input/output interface may be an input/output circuit.

In another implementation, the interface unit may be an input/output interface, interface circuit, output circuit, input circuit, pin or related circuit on the chip, chip system or circuit, etc.; the processing unit may be at least one processor , processing circuits or logic circuits, etc.

The beneficial effects of the device shown in the above fourth aspect and its possible designs may be referred to the beneficial effects of the first aspect and its possible designs.

In a fifth aspect, a communication device is provided, which is used to perform the method provided in the second aspect. The device can be a core network device, or a component of the core network device (such as a processor, a chip, or a chip system, etc.), or a logic module or software that can realize all or part of the functions of the core network device. The device includes :

The interface unit is used to obtain a neural network data packet, the neural network data packet includes indication information, and the indication information is used to indicate the priority of the neural network data packet;

The interface unit is also used to send a General Packet Wireless Service Tunneling Protocol user plane GTP-U data packet to the access network device. The payload of the GTP-U data packet includes the neural network data packet. The GTP-U data packet contains This instruction is included in the header.

Specifically, the communication device may include units and/or modules for executing the method provided by any implementation of the second aspect, such as a processing unit and an interface unit.

In one implementation, the interface unit may be a transceiver, or an input/output interface; the processing unit may be at least one processor. Alternatively, the transceiver may be a transceiver circuit. Alternatively, the input/output interface may be an input/output circuit.

In another implementation, the interface unit may be an input/output interface on the chip, chip system or circuit, Interface circuit, output circuit, input circuit, pin or related circuit, etc.; the processing unit can be at least one processor, processing circuit or logic circuit, etc.

The beneficial effects of the device shown in the fifth aspect above can be referred to the beneficial effects of the second aspect.

In a sixth aspect, a communication device is provided, which is used to perform the method provided in the third aspect. The device can be a server, or a component of the server (such as a processor, a chip, or a chip system, etc.), or a logic module or software that can realize all or part of the server functions. The device includes:

The processing unit is used to generate a neural network data packet. The neural network data packet includes indication information, and the indication information is used to indicate the priority of the neural network data packet. The interface unit is used to send the neural network data packet.

Specifically, the communication device may include units and/or modules for executing the method provided by any implementation of the third aspect, such as a processing unit and an interface unit.

In one implementation, the interface unit may be a transceiver, or an input/output interface; the processing unit may be at least one processor. Alternatively, the transceiver may be a transceiver circuit. Alternatively, the input/output interface may be an input/output circuit.

In another implementation, the interface unit may be an input/output interface, interface circuit, output circuit, input circuit, pin or related circuit on the chip, chip system or circuit, etc.; the processing unit may be at least one processor , processing circuits or logic circuits, etc.

The beneficial effects of the device shown in the sixth aspect above can be referred to the beneficial effects of the third aspect.

In some implementations of the first to sixth aspects, the priority of the neural network data packet is related to the effect of the neural network recovery data corresponding to the neural network data packet. Among them, the effect of neural network on data recovery can be indicated by any of the following indicators:

Peak signal to noise ratio (PSNR), structural similarity (structural similarity, SSIM) or video multimethod assessment fusion (VMAF), etc.

For example, when the effect of a neural network on recovering data meets expectations (for example, VAMF is greater than a preset threshold), the data of the neural network (for example, the parameter information of the neural network, the coefficients of the neural network, etc.) are encoded by the neural network. The priority of the packet is high priority.

For another example, when the effect of the neural network on recovering data does not meet expectations (for example, the VAMF is less than or equal to a preset threshold), the priority of the neural network data packet obtained by encoding the data of the neural network is low priority.

Based on the above technical solution, in this application, the priority of the neural network data packets of the neural network can be determined according to the effect of the neural network on data recovery, so that the neural network data packets of the neural network with good data recovery effect are preferentially transmitted, so that the user You can give priority to using neural networks with good data recovery effects to restore data and improve user experience.

In some implementations of the first to sixth aspects, the priority of the neural network data packet is also related to the effect of the preset algorithm on recovering the data. Among them, the effect of the preset algorithm on recovering the data can be indicated by any of the following indicators: PSNR, SSIM or VMAF, etc.

For example, the effect of the neural network on recovering the data is higher than expected compared to the effect of the preset algorithm on recovering the data (for example, the VAMF corresponding to the neural network is greater than a preset value compared to the VAMF corresponding to the preset algorithm). In this case, the priority of the neural network data packet obtained by encoding the neural network data is high priority.

For example, the effect of the neural network on recovering the data is lower than expected compared to the effect of the preset algorithm on recovering the data (for example, the VAMF corresponding to the neural network is less than or equal to a preset value compared to the VAMF corresponding to the preset algorithm. value), the priority of the neural network data packet obtained by encoding the data of the neural network is low priority.

Based on the above technical solution, in this application, the priority of the neural network data packets of the neural network can be determined based on the effect of restoring data by the neural network and the effect of restoring the data by the preset algorithm, and the effect of restoring data in multiple neural networks can be achieved. If the data recovery effect is better than the preset algorithm, the neural network that recovers the data better than the preset algorithm will be given priority, so that the user can give priority to using the neural network with the best data recovery effect to recover the data. Improve user experience.

In some implementations of the first to sixth aspects, the neural network data packet is used to reconstruct the neural network corresponding to the neural network data packet, and the priority of the neural network data packet is related to the effect of reconstructing the neural network. Related. Wherein, the neural network data packet used to reconstruct the neural network corresponding to the neural network data packet can be understood as: the data of the neural network data packet is the data used to reconstruct the neural network. For example, the data of the neural network data packet is the data of the neural network coefficients, and the neural network coefficients are used to reconstruct the neural network.

Based on the above technical solution, in this application, the priority of the neural network data packets of the neural network can be determined according to the effect of the reconstructed neural network, so as to preferentially transmit the neural network data packets with good effect of the reconstructed neural network, and improve user reuse. efficiency of the neural network.

In some implementations of the first to sixth aspects, the neural network data packet includes data of coefficients of the neural network, the data of the coefficients are used to obtain the coefficients, and the coefficients are used to reconstruct the neural network, the The priority of a neural network packet is related to the impact of that coefficient's data on that coefficient.

For example, when the difference between coefficient #1 and the coefficient calculated by the coefficient data meets expectations (for example, the difference between coefficient #1 and the coefficient is less than or equal to the preset threshold), determine the value of the neural network data packet The priority is high priority.

For another example, in the case where the difference between coefficient #1 and the coefficient calculated by the coefficient data does not meet expectations (for example, the difference between coefficient #1 and the coefficient is greater than a preset threshold), determine the value of the neural network data packet The priority is low priority.

Based on the above technical solution, in this application, the priority of the neural network data packet can be determined based on the difference between the value calculated from the data of the neural network data packet and the coefficient of the neural network, so that the priority transmission can be calculated and compared with the neural network. The neural network data packet with the closest coefficient value enables the user to quickly reconstruct the neural network based on the received neural network data packet and improves the efficiency of the user in reconstructing the neural network.

In some implementations of the first to sixth aspects, the coefficient is represented by a plurality of bits, the values of the plurality of bits are used to calculate the absolute value of the coefficient, and the data of the coefficient is represented by at least one bit. , the at least one bit belongs to the plurality of bits.

Based on the above technical solution, the coefficients of the neural network can be represented by multiple bits, the data of the neural network data packet can be represented by at least one bit, and at least one bit belongs to the above-mentioned multiple bits, by using the bit form to represent Coefficient data is helpful for users to calculate the coefficients of the neural network based on the received neural network data packets.

In some implementations of the first to sixth aspects, the plurality of bits includes a sign part, an exponent part and a fraction part, and the at least one bit is the sign part, the exponent part and the fraction part. In the case of a part, the priority of the neural network data packet is the first priority; in the case where the at least one bit is the second part of the fractional part, the scheduling priority corresponding to the neural network is determined to be the second priority. , wherein the first priority is higher than the second priority, the first part of the fractional part is the high-order data part of the fractional part, and the second part of the fractional part is the low-order data part of the fractional part.

Based on the above technical solution, the coefficients of the neural network can be represented by the symbolic part, the exponential part and the fractional part, which is helpful to determine the degree of influence of different parts on the absolute value of the coefficient of the neural network, so that the neural network data packet can be obtained The data represented by those parts of the coefficient determine the degree of influence of the data of the neural network packet on the absolute value of the neural network coefficient, making the scheme more concise.

In some implementations of the first to sixth aspects, the indication information is carried in a header of the neural network data packet.

In some implementations of the first to sixth aspects, the indication information is located between the User Datagram Protocol UDP field and the Real-Time Transport Protocol RTP field in the neural network data packet header.

In a seventh aspect, this application provides a processor for executing the methods provided in the above aspects.

For operations such as sending and getting/receiving involved in the processor, if there is no special explanation, or if it does not conflict with its actual role or internal logic in the relevant description, it can be understood as processor output, reception, input and other operations. , can also be understood as the transmitting and receiving operations performed by the radio frequency circuit and the antenna, which is not limited in this application.

In an eighth aspect, a computer-readable storage medium is provided. The computer-readable storage medium stores a program code for device execution. The program code includes a method for executing any one of the above-mentioned first to third aspects. Methods.

A ninth aspect provides a computer program product containing instructions, which when the computer program product is run on a computer, causes the computer to execute the method provided by any one of the above implementations of the first to third aspects.

In a tenth aspect, a chip is provided. The chip includes a processor and a communication interface. The processor reads instructions stored in the memory through the communication interface and executes the method provided by any one of the above-mentioned implementations of the first to third aspects.

Optionally, as an implementation manner, the chip also includes a memory, in which computer programs or instructions are stored. The processor is used to execute the computer programs or instructions stored in the memory. When the computer program or instructions are executed, the processor is used to execute The method provided by any one of the above implementations of the first aspect to the third aspect.

In an eleventh aspect, a data transmission device is provided. The device includes a processor configured to perform any of the possible implementations of the first to fourth aspects and the first to third aspects. method.

A twelfth aspect provides a communication system, including the data transmission device of the fourth aspect to the data transmission device of the sixth aspect.

For the beneficial effects of the above seventh aspect to the twelfth aspect, reference can be made to the description of the beneficial effects of the first aspect to the third aspect, and will not be repeated.

Description of the drawings

(a) to (c) in Figure 1 are schematic diagrams of application scenarios applicable to embodiments of the present application.

Figure 2 is a functional block diagram of the DNN-based SR transmission mode and the traditional transmission mode provided by the embodiment of the present application.

Figure 3 is a schematic diagram of XR video transmission based on the SR method provided by the embodiment of the present application.

Figure 4 is a schematic diagram of the architecture of the QoS guarantee mechanism provided by the embodiment of the present application.

Figure 5 is a schematic diagram of QoS flow mapping applicable to the embodiment of the present application.

Figure 6 is a schematic diagram of the bit structure of floating point type data.

Figure 7 is a schematic flow chart of a communication method provided by an embodiment of the present application.

Figure 8 is a schematic diagram of a neural network data packet provided by an embodiment of the present application.

Figure 9 is a schematic diagram of generating neural network data packets provided by an embodiment of the present application.

Figure 10 is a schematic diagram of a header of a GTP-U data packet provided by an embodiment of the present application.

Figure 11 is a schematic block diagram of a communication device suitable for embodiments of the present application.

Figure 12 is a structural block diagram of a communication device suitable for embodiments of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be described below with reference to the accompanying drawings.

The technical solutions of the embodiments of this application can be applied to various communication systems, such as: fifth generation (5G) systems or new radio (NR), long term evolution (LTE) systems, LTE frequency Frequency division duplex (FDD) system, LTE time division duplex (TDD), etc. The technical solution provided by this application can also be applied to future communication systems, such as the sixth generation mobile communication system. The technical solutions of the embodiments of this application can also be applied to device-to-device (D2D) communication, vehicle-to-everything (V2X) communication, machine-to-machine (M2M) communication, and machine-to-machine (M2M) communication. Type communication (machine type communication, MTC), and Internet of things (Internet of things, IoT) communication system or other communication systems.

In order to facilitate understanding of the embodiments of the present application, the communication system applicable to the embodiments of the present application is briefly introduced with reference to (a) to (c) in Figure 1 .

The technical solutions of the embodiments of the present application can be applied to the communication systems shown in (a) to (c) in Figure 1. Of course, they can also be used in future network architectures, such as the sixth generation (6th generation, 6G) network architecture, etc. , the embodiments of this application do not specifically limit this.

The communication system applicable to the embodiment of the present application will be illustrated below with reference to (a) to (c) in Figure 1 . It should be understood that the communication system described herein is only an example and should not constitute any limitation on this application.

As an exemplary illustration, (a) in FIG. 1 shows a schematic architectural diagram of a communication system 100a applicable to the embodiment of the present application. As shown in (a) in Figure 1, the network architecture may include but is not limited to the following network elements (also known as functional network elements, functional entities, nodes, devices, etc.):

User equipment (UE), access network equipment (AN) (or radio access network equipment (RAN)), access and mobility management function , AMF) network element, session management function (SMF) network element, user plane function (UPF) network element, policy control function (PCF) network element, application function (application function) , AF) network element, capability exposure function (network exposure function, NEF) network element, data network (data network, DN) and server.

The following is a brief introduction to each network element shown in (a) in Figure 1:

1. UE: A terminal that communicates with (R)AN. It can also be called terminal equipment, access terminal, user unit, user station, mobile station, mobile station (MS), mobile terminal (mobile). terminal, MT), remote station, remote terminal, mobile device, user terminal, terminal, wireless communications equipment, user agent or user device. The terminal device may be a device that provides voice/data connectivity to the user, such as a handheld device, a vehicle-mounted device, etc. with wireless connectivity capabilities. Currently, some examples of terminals include: mobile phones, tablets, computers with wireless transceiver functions (such as laptops, handheld computers, etc.), mobile internet devices (MID), virtual reality (virtual reality, VR) equipment, augmented reality (AR) equipment, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in remote medical Terminals, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities Terminals, wireless terminals in smart homes, cellular phones, cordless phones, session initiation protocol (SIP) phones, wireless local loop (WLL) stations, personal digital assistants assistant (PDA)), handheld devices with wireless communication capabilities, computing devices or other processing devices connected to wireless modems, vehicle-mounted devices, wearable devices (such as head-mounted extended reality (XR) glasses), video players , holographic projectors and other equipment, terminal equipment in the 5G network or terminal equipment in the future evolved public land mobile communication network (public land mobile network, PLMN), etc.

In addition, the terminal device can also be a terminal device in an Internet of things (IoT) system. For example, wireless terminals in driverless driving, wireless terminals in remote medical, wireless terminals in smart grid, wireless terminals in transportation safety, and wireless terminals in smart city. wireless terminals, wireless terminals in smart homes, wearable terminal devices, etc. IoT is an important part of the future development of information technology. Its main technical feature is to connect objects to the network through communication technology, thereby realizing an intelligent network of human-computer interconnection and object interconnection. IoT technology can achieve massive connections, deep coverage, and terminal power saving through narrowband (NB) technology, for example.

It should be understood that the terminal device can be any device that can access the network. Terminal equipment and access network equipment can communicate with each other using some air interface technology.

Optionally, the user equipment can be used to act as a base station. For example, user equipment may act as a scheduling entity that provides sidelink signals between user equipments in V2X or D2D, etc. For example, cell phones and cars use sidelink signals to communicate with each other. Cell phones and smart home devices communicate between each other without having to relay communication signals through base stations.

2. AN: Used to provide network access functions for authorized user equipment in a specific area, and can use transmission tunnels with different service qualities according to the level of user equipment, business needs, etc.

AN can manage wireless resources, provide access services for user equipment, and then complete the forwarding of control signals and user equipment data between user equipment and the core network. AN can also be understood as a base station in a traditional network.

Illustratively, the access network device in the embodiment of the present application may be any communication device with wireless transceiver functions used to communicate with user equipment. The access network equipment includes but is not limited to: evolved node B (evolved node B, eNB), wireless network controller (radio network controller, RNC), node B (node B, NB), base station controller (base station controller) , BSC), base transceiver station (BTS), home base station (home evolved node B, HeNB, or home node B, HNB), base band unit (base band unit, BBU), wireless fidelity, WIFI) system access point (access point, AP), wireless relay node, wireless backhaul node, transmission point (transmission point, TP) or sending and receiving point (transmission and reception point, TRP), etc., can also be A gNB in a 5G, e.g., NR, system, or a transmission point (TRP or TP), one or a group (including multiple antenna panels) of antenna panels of a base station in a 5G system, or may also constitute a gNB or transmission Point network nodes, such as baseband unit (BBU), or distributed unit (DU), etc.

In some deployments, gNB may include centralized units (CUs) and DUs. The gNB may also include an active antenna unit (AAU). CU implements some functions of gNB, and DU implements some functions of gNB. For example, the CU is responsible for processing non-real-time protocols and services, and implementing radio resource control (RRC) and packet data convergence protocol (PDCP) layer functions. DU is responsible for processing physical layer protocols and real-time services, and implementing the functions of the radio link control (RLC) layer, media access control (MAC) layer and physical (physical, PHY) layer. AAU implements some physical layer processing functions, radio frequency processing and active antenna related functions. Since RRC layer information will eventually become PHY layer information, or transformed from PHY layer information, in this architecture, high-level signaling, such as RRC layer signaling, can also be considered to be sent by DU , or sent by DU+AAU. It can be understood that the access network device may be a device including one or more of a CU node, a DU node, and an AAU node. In addition, the CU can be divided into access network equipment in the access network (radio access network, RAN), or the CU can be divided into access network equipment in the core network (core network, CN). This application does not Make limitations.

3. User plane network element: used for packet routing and forwarding and quality of service (QoS) processing of user plane data.

As shown in (a) in Figure 1, the user plane network element can be a UPF network element, which can include an intermediate user plane function (I-UPF) network element and an anchor user plane function (PDU session anchor). user plane function, PSA-UPF) network element. In future communication systems, user plane network elements can still be UPF network elements, or they can have other names, which are not limited in this application.

4. Data network: Provides operator services, Internet access or third-party services, including servers, which implement video source encoding and rendering on the server side.

In future communication systems, the data network may still be a DN, or may have other names, which are not limited in this application.

In the 5G communication system, after the terminal device is connected to the network, it can establish a protocol data unit (PDU) session and access the DN through the PDU session. It can communicate with the application function network elements (application function network elements such as application function network elements) deployed in the DN. for application server) interaction. As shown in (a) in Figure 1, depending on the DN that the user accesses, the network can select the UPF of the access DN as the PDU session anchor (PSA) according to the network policy, and access it through the N6 interface of the PSA Application function network element.

5. Access and mobility management network element: Mainly used for mobility management and access management, etc., and can be used to implement other functions in the mobility management entity (MME) function except session management. For example, functions such as lawful interception and access authorization/authentication.

As shown in (a) in Figure 1, the access management network element may be an AMF network element. In future communication systems, the access management network element can still be an AMF network element, or it can also have other names, which is not limited in this application.

6. Session management network element: Mainly used for session management, network interconnection protocol (IP) address allocation and management of terminal equipment, selection of endpoints for manageable terminal equipment plane functions, policy control and charging function interfaces, and downlink Data notifications, etc.

As shown in (a) in Figure 1, the session management network element can be an SMF network element and can include an intermediate session management function (I-SMF) network element and an anchor session management function (anchor session management function). function, A-SMF) network element. In future communication systems, the session management network element can still be an SMF network element, or it can also have other names, which is not limited in this application.

7. Policy control network element: A unified policy framework used to guide network behavior and provide policy rule information for control plane functional network elements (such as AMF, SMF network elements, etc.).

The policy control network element may be a policy and charging rules function (PCRF) network element. As shown in (a) in Figure 1, the policy control network element may be a PCF network element. In future communication systems, the policy control network element can still be a PCF network element, or it can also have other names, which is not limited in this application.

8. Application function network element: Application function network element can interact with the 5G system through the application function network element and be used for access network Network open function network elements or interact with the policy framework for policy control, etc.

As shown in (a) in Figure 1, the application function network element may be an application function, AF network element. In the future communication system, the application function network element can still be an AF network element, or it can also have other names, which is not limited in this application.

9. Network opening function network element: used to provide customized functions for network opening.

As shown in (a) in Figure 1, the network exposure function network element can be a network exposure function (NEF) network element. In future communication systems, the network exposure function network element can still be an NEF network element. Alternatively, there may be other names, which are not limited in this application.

10. Server: Can provide application service data. For example, video data, audio data, or other types of data can be provided. The data types of application services provided by the server are only used as examples in this application and are not limited.

It can be understood that the above network element or function can be a network element in a hardware device, a software function running on dedicated hardware, or a virtualization function instantiated on a platform (for example, a cloud platform). The above network elements or functions can be divided into one or more services. Furthermore, there may also be services that exist independently of network functions. In this application, instances of the above functions, or instances of services included in the above functions, or service instances that exist independently of network functions can be called service instances.

Further, the AF network element may be abbreviated as AF, the NEF network element may be abbreviated as NEF, and the AMF network element may be abbreviated as AMF. That is, the AF described later in this application can be replaced by the application function network element, the NEF can be replaced by the network opening function network element, and the AMF can be replaced by the access and mobility management network element.

In the architecture shown in (a) in Figure 1, the interface names and functions between each network element are as follows:

1), N1: The interface between AMF and the terminal, which can be used to transmit QoS control rules to the terminal.

2), N2: The interface between AMF and RAN, which can be used to transmit wireless bearer control information from the core network side to the RAN.

3), N3: The interface between RAN and UPF, mainly used to transmit uplink and downlink user plane data between RAN and UPF.

4), N4: The interface between SMF and UPF can be used to transfer information between the control plane and the user plane, including controlling the delivery of user-oriented forwarding rules, QoS control rules, traffic statistics rules, etc., as well as the user plane Report information.

5), N5: The interface between AF and PCF, which can be used to issue application service requests and report network events.

6), N6: The interface between UPF and DN, used to transmit uplink and downlink user data flows between UPF and DN.

7), N7: The interface between PCF and SMF can be used to deliver protocol data unit (PDU) session granularity and business data flow granularity control policy.

8), N11: The interface between SMF and AMF can be used to transfer PDU session tunnel information between RAN and UPF, transfer control messages sent to the terminal, transfer radio resource control information sent to RAN, etc.

The meaning of these interface serial numbers is not limited here.

It should also be understood that service-oriented interfaces can be used between certain network elements in the system, which will not be described again here.

As an exemplary illustration, (b) in FIG. 1 shows a schematic architectural diagram of a communication system 100b applicable to the embodiment of the present application. As shown in (b) in Figure 1, the architecture is a terminal-network-terminal architecture scenario. This scenario can be a tactile Internet (TI). One terminal interfaces with the main domain tactile user and the artificial system, and the other end is subject to Remote control robots or remote operators in the control domain, network transmission core network and access network include LTE, 5G or next-generation air interface 6G. The main domain receives audio/video feedback signals from the controlled domain. The main domain and the controlled domain are connected through two-way communication links on the network domain with the help of various commands and feedback signals, thus forming a global control loop.

As shown in (b) in Figure 1, the network architecture may include but is not limited to the following network elements (also known as functional network elements, functional entities, nodes, devices, etc.):

UE#1, AN#1, UPF, UE#2, AN#2.

The following is a brief introduction to each network element shown in (b) in Figure 1:

1. UE#1: It can interface between the main domain tactile user and the artificial system, and receive video, audio and other data from the controlled domain. It can include various handheld devices with wireless communication functions, vehicle-mounted devices, wearable devices such as head-mounted glasses, computing devices or other processing devices connected to wireless modems, as well as various forms of terminals, mobile stations (MS) , user equipment, soft terminals, etc., such as video playback equipment, holographic projectors, etc. The embodiments of the present application are not limited to this.

2. AN#1: used to provide network access functions for authorized terminal equipment (such as UE#1) in a specific area, and can use transmission tunnels of different qualities according to the level of the terminal equipment, business requirements, etc.

3. UPF: used for packet routing and forwarding and QoS processing of user plane data. Refer to the description of user plane network elements in (a) of Figure 1, which will not be described again here.

4. AN#2: Used to provide network access functions for authorized terminal equipment (such as UE#2) in a specific area, and can use transmission tunnels of different qualities according to the level of the terminal equipment, business requirements, etc.

5. UE#2: It is a remote control robot or remote operator in the controlled domain. Video, audio and other data can be sent to the main domain. It can include various handheld devices with wireless communication functions, vehicle-mounted devices, wearable devices such as head-mounted glasses, computing devices or other processing devices connected to wireless modems, as well as various forms of terminals, mobile stations (MS) , user equipment, soft terminals, etc., such as video playback equipment, holographic projectors, etc. The embodiments of the present application are not limited to this.

As an exemplary illustration, (c) in FIG. 1 shows a schematic architectural diagram of a communication system 100c applicable to the embodiment of the present application. As shown in (c) in Figure 1, the architecture is a WiFi scenario. In this scenario, the cloud server transmits XR media data or ordinary video to the terminal (XR device) through the fixed network, WiFi router/AP/set-top box.

As shown in (c) in Figure 1, the network architecture may include but is not limited to the following network elements (also known as functional network elements, functional entities, nodes, devices, etc.):

Server, fixed network, WiFi router/WiFi access point (AP) and UE.

The following is a brief introduction to each network element shown in (c) in Figure 1:

1. Server: Can provide application service data. For example, video data, audio data, or other types of data can be provided. The data types of application services provided by the server are only used as examples in this application and are not limited.

2. Fixed network: A network that transmits signals through solid media such as metal wires or optical fiber lines.

In this application, application service data such as video data and audio data can be transmitted to the WiFi router/WiFi AP through the fixed network.

3. WiFi router/WiFi AP: Can convert wired network signals and mobile network signals into wireless signals for reception by UEs with wireless communication capabilities.

4. UE: It can include various handheld devices, vehicle-mounted devices, wearable devices, computing devices or other processing devices connected to wireless modems with wireless communication functions, such as head-mounted glasses, video playback devices, and holographic projectors. etc. The embodiments of the present application are not limited to this.

It should be understood that the network architecture to which the embodiments of the present application can be applied are only illustrative. The network architecture applicable to the embodiments of the present application is not limited to this. Any network architecture that can realize the functions of each of the above network elements is applicable to this application. Application examples.

It should also be understood that the AMF, SMF, UPF, PCF, NEF, etc. shown in (a) of Figure 1 can be understood as network elements used to implement different functions, and can, for example, be combined into network slices as needed. These network elements can be independent devices, or they can be integrated into the same device to implement different functions, or they can be network elements in hardware devices, software functions running on dedicated hardware, or platforms (for example, cloud The virtualization function instantiated on the platform), this application does not limit the specific form of the above network elements.

It should also be understood that the above nomenclature is only defined to facilitate the differentiation of different functions and should not constitute any limitation on this application. This application does not rule out the possibility of using other naming in 5G networks and other future networks. For example, in a 6G network, some or all of the above network elements may use the terminology used in 5G, or may adopt other names.

It should also be understood that the interface name between each network element in (a) in Figure 1 is just an example. In specific implementation, the name of the interface may be other names, and this application does not specifically limit this. In addition, the names of the messages (or signaling) transmitted between the various network elements are only examples and do not constitute any limitation on the function of the messages themselves.

It should be understood that the methods provided by the embodiments of the present application can be applied to 5G communication systems, for example, the communication systems shown in (a) to (c) in Figure 1 . However, the embodiments of the present application do not limit the scenarios in which this method can be applied. For example, it is also applicable to other network architectures including network elements that can implement corresponding functions. Another example is the sixth generation communication (the6th generation, 6G) system architecture, etc. Moreover, the names of each network element used in the above embodiments of the present application may maintain the same functions in future communication systems, but the names will change.

In order to facilitate understanding of the technical solutions of the embodiments of the present application, some terms or concepts that may be involved in the embodiments of the present application are briefly described.

1. XR technology: In recent years, with the continuous progress and improvement of XR technology, related industries have developed vigorously. Today, extended reality technology has entered various fields closely related to people's production and life, such as education, entertainment, military, medical care, environmental protection, transportation, and public health. Extended reality is a general term for various reality-related technologies, including: VR, AR, MR, etc.

VR technology mainly refers to the rendering of visual and audio scenes to simulate as much as possible the visual and audio stimulation of the user in the real world. VR technology usually requires users to wear a head-mounted display (HMD) to simulate The visual component completely replaces the user's field of view while requiring the user to wear headphones to provide accompanying audio to the user. In addition, it is usually necessary to perform some kind of head and movement tracking on the user in VR to update the simulated visual and audio content in a timely manner so that the visual and audio content of the user experience is consistent with the user's movements.

AR technology mainly refers to providing additional visual or auditory information or artificially generated content in the real environment perceived by the user. Among them, the user's acquisition of the real environment can be direct, that is, without intermediate sensing, processing, and rendering, or it can It is indirect, that is, it is transmitted through sensors and other methods, and further enhanced processing is performed.

MR technology is an advanced form of AR. One of its implementation methods is to insert some virtual elements into the physical scene, with the purpose of providing users with an immersive experience in which these elements are part of the real scene.

2. Super resolution (SR): refers to the technology of improving the resolution of the original image/video through hardware or software methods, and obtaining high-resolution images through low-resolution images. At present, SR technology based on NN has received widespread attention because of its remarkable picture restoration effect. Among them, NN can be a deep neural network (deep neural network). network, DNN).

For ease of understanding, the DNN-based SR technology is introduced in detail with reference to Figures 2 and 3.

Figure 2 is a principle block diagram of the DNN-based SR transmission mode and the traditional transmission mode provided by the embodiment of the present application. The DNN-based SR transmission mode specifically includes the following four steps:

Step 1: On the server side, the high definition (HD) XR video frame is spatially divided into blocks (tiles) (or slices, for ease of description, collectively referred to as blocks below).

For example, the entire video frame has a resolution of 4K (3840*1920) and can be divided into small blocks. The resolution of a small block is (192*192). Then the entire video can be divided into segments in time. For example, 1-2 seconds can be divided into a segment, or a video frame can be divided into a segment.

The purpose of step one is to amortize the processing load and speed up processing through parallel operations.

Step 2: On the server side, the HD block is downsampled (i.e., sampled in the spatial domain or frequency domain) to obtain low definition (LD) blocks.

For example, the resolution of a block is (192*192), and the resolution after downsampling is (24*24). After downsampling, traditional video compression techniques such as high efficiency video coding (HEVC) technology can be used to further compress into ultra-low resolution (ULR) blocks.

Step 3: On the server side, use the ULR block as the input of the DNN, use the original HD content as the target output of the DNN, use these two as the training set of the DNN, and use PSNR as the loss function for DNN training, so that you can get Adaptive neural network matching application layer services. The ULR block is then sent to the user together with the DNN. The reason for transmitting the information of the neural network is that the receiving end does not know the original video, and the source end needs to generate the neural network based on the original video and transmit it to the receiving end.

Step 4: On the user side, the ULR block can be used as the input of DNN, and the output is high-definition video content.

Using SR technology in the current transmission architecture is to divide a frame of XR video into dozens of Internet Protocol (IP) packets at the network transmission layer, such as 50 IP packets, and NN The data is also encoded into multiple IP packets, and then these IP packets are transmitted to the fixed network and/or core network, and then the IP data packets are transmitted to the UE through the RAN. The specific process is shown in Figure 3. Figure 3 is the implementation of this application The example provides a schematic diagram of XR video transmission based on the SR method.

3. Protocol data unit (PDU) session: It is an association between the terminal device and the DN, used to provide a PDU connection service.

4. Quality of service (QoS) flow mechanism: The current standard stipulates that QoS flow is the minimum QoS control granularity, and QoS flow has corresponding QoS configuration.

For ease of understanding, the architecture of the QoS guarantee mechanism involved in this application will be described below with reference to Figure 4. Figure 4 is a schematic diagram of the architecture of the QoS guarantee mechanism provided by the embodiment of the present application.

QoS is a guarantee mechanism for service transmission quality. Its purpose is to provide end-to-end service quality assurance for the different needs of various businesses. In a protocol data unit (PDU) session, QoS flow is the smallest granularity that distinguishes QoS. In the 5G system, the QoS flow identifier (QFI) is used to identify the QoS flow, and the QFI must be unique within a PDU session. In other words, a PDU session can have multiple (up to 64) QoS flows, and different QoS flows have different QFIs. In a PDU session, user plane service flows with the same QFI use the same service forwarding processing method (such as scheduling).

In terms of configuration granularity, one PDU session can correspond to multiple radio bearers (radio bearers, RBs). a nothing A line bearer can contain multiple QoS flows.

For a PDU session, there is still a single NG-U channel between the 5GC and the AN. The wireless bearer is used between the AN and the UE, and the AN controls which bearer the QoS flow is mapped to.

Figure 5 is a schematic diagram of QoS flow mapping applicable to the embodiment of the present application. 5GC and AN ensure quality of service by mapping data packets to appropriate QoS flows and wireless bearers.

UPF implements the mapping of Internet Protocol (IP) flows to QoS flows, and AN implements the mapping of QoS flows to RBs. QoS mapping can include three parts: UPF mapping, AN mapping and UE mapping:

UPF mapping: After UPF receives downlink data, it maps it to the corresponding QoS flow using allocation and reservation priorities. Then perform QoS control of the QoS flow and mark the data through QFI. The data is sent to the AN through the N3 interface corresponding to the QoS flow.

AN mapping: After receiving the downlink data, AN determines the RB and QoS flow corresponding to the QFI. Then perform QoS control corresponding to the QoS flow, and send the data to the UE through the RB. Or, after receiving the uplink data, the AN determines the QoS flow corresponding to the QFI. Then perform QoS control corresponding to the QoS flow, and send the data to UPF through the N3 interface corresponding to the QoS flow.

UE mapping: When the UE wants to send uplink data, it is mapped to the corresponding QoS flow according to QoS rules. The uplink data is then sent through the RB corresponding to the QoS flow.

It should be understood that SMF is responsible for the control of QoS flows. When establishing a PDU session, SMF can configure response QoS parameters for UPF, AN, and UE. QoS flows can be established and modified through PDU sessions or defined through preconfiguration. The configuration response parameters of a QoS flow include three parts:

1) QoS configuration (QoS profile): SMF can provide QoS configuration to AN through the N2 interface, or it can also be pre-configured in the AN.

It should be understood that the QoS configuration of a certain QoS flow can also be called a QoS configuration file. The specific parameters of QoS configuration are shown in Table 1.

Table 1

5QI is a scalar used to index a 5G QoS characteristic. 5QI can be standardized, preconfigured, or dynamically defined. The attributes of 5QI are shown in Table 2 below.

Table 2

2) QoS rules: SMF can provide QoS rules to UE through the N1 interface. Alternatively, the UE can derive it through the QoS mechanism.

It should be understood that the UE performs classification and marking of uplink user plane data services, that is, mapping uplink data to corresponding QoS flows according to QoS rules. These QoS rules can be explicitly provided to the UE (that is, explicitly configured to the UE through signaling during the PDU session establishment/modification process); or they can be pre-configured on the UE; or the UE can use the reflection QoS mechanism. Implicitly derived. QoS rules have the following characteristics:

A QoS rule includes: QFI associated with the QoS flow, packet filter set (a filter list), and priority.

A QoS flow can have multiple QoS rules.

A PDU session must be configured with a default QoS rule, and the default QoS rule is associated with a QoS flow.

3). Upstream and downstream packet detection rules (packet detection rules, PDR): SMF provides PDR(s) to UPF through the N4 interface.

5. Group of picture (GoP): consists of multiple types of video frames. The first frame in the GoP is an I frame (intra frame), which can contain multiple P frames (predicted frames) later. The I frame is an intra-frame reference frame. Usually the amount of data is large, and it is restored based on the data of this frame during decoding. Image, errors have a great impact on video quality; P frame is a predictive coding frame, usually with a small amount of data. It is used to represent the data that is different from the previous frame. When decoding, it is necessary to superimpose the previously cached picture on the frame defined by this frame. Differentially generated images, errors have relatively little impact on video quality. Therefore, the data packet can be scheduled according to the type of video frame to which the data packet belongs. For example, since I frames are more important than P frames, the data packets to which I frames belong have a higher scheduling priority, and the data packets to which P frames belong have a lower scheduling priority. .

6. Video quality evaluation indicators: The current mainstream video quality evaluation indicators mainly include two categories: one is objective evaluation indicators, such as PSNR and SSIM, which are values obtained by calculating the difference or correlation between each pixel; The second category is subjective evaluation indicators, such as VMAF, which reflects the impact of different image distortions on the user's subjective experience, with scores ranging from 0 to 100 points. Specifically, the higher the VMAF score, the less image distortion and the better the user's subjective experience.

7. Bit structure of floating-point type data: The bits of a floating-point number include the sign part, the exponent part and the fraction part.

Taking the 32-bit float type as an example, the bit structure of floating-point type data is introduced in conjunction with Figure 6. Figure 6 is a schematic diagram of the bit structure of floating-point type data, including a 1-bit sign part: 0 represents a positive number, 1 represents Negative number; 8-bit exponent: the exponent ranges from -127 to +127; 23-bit fraction: the minimum precision is 1/(2^23). Specifically, the absolute value of the floating point type data can be calculated based on the following formula:

Therefore, it can be seen that the low-order bits of the fractional part have a small impact on the absolute value of the coefficient, while the high-order bits of the sign part, the exponent part, and the fractional part have a greater impact on the absolute value. It should be noted that the low-order bits of the fraction part and the high-order bits of the fraction part involved in the embodiments of this application can be understood as the fraction part is composed of high-order bits and low-order bits, and the fraction part except the high-order bits are all low-order bits. Bits, for example, the fraction part has a total of 23 bits, the first bit is a high-order bit, and the remaining 22 bits are low-order bits; or the first x bits are high-order bits, and the remaining 23-x are low-order bits.

From the QoS flow mechanism introduced above, we can set different QoS requirements for NN data packets and video frame data packets according to different business types. However, for the NN data packet itself, there is currently a lack of a mechanism to distinguish the importance of different NN data packets.

In addition, it can be seen from the GoP introduced above that in order to distinguish the importance of video frame data packets, the data packets of different types of video frames can be distinguished according to the structural characteristics of video coding. However, for NN, the current coding structure has no importance partitioning, so it cannot Use the method of measuring the importance of video frames to measure the importance of NN packets.

In order to solve the shortcomings of the current NN data packet transmission, this application provides a communication method by carrying indication information indicating the priority of the neural network data packet in the GTP-U data packet, so that the GTP-U data packet is received The access network device can transmit neural network data packets according to the instruction information, in order to achieve differentiated transmission of neural network data packets of different priorities.

The scenarios in which the embodiments of the present application can be applied are introduced above with reference to (a) to (c) in Figure 1 , and the basic concepts involved in the present application are also briefly introduced. Below, the communication provided by the present application will be introduced in detail with reference to the accompanying drawings. method.

The embodiments shown below do not specifically limit the specific structure of the execution body of the method provided by the embodiment of the present application, as long as it can be provided according to the embodiment of the present application by running a program that records the code of the method provided by the embodiment of the present application. It suffices to communicate by a method. For example, the execution subject of the method provided by the embodiment of the present application may be the core network device, or a functional module in the core network device that can call the program and execute the program.

In order to facilitate understanding of the embodiments of the present application, the following points are explained.

First, in this application, "for indicating" may include direct instructions and indirect instructions. When describing certain information to indicate A, it may include that the information directly indicates A or indirectly indicates A, but it does not mean that the information must contain A.

The information indicated by the information is called information to be indicated. In the specific implementation process, there are many ways to indicate the information to be indicated. For example, but not limited to, the information to be indicated can be directly indicated, such as the information to be indicated itself or the information to be indicated. Index of information, etc. The information to be indicated may also be indirectly indicated by indicating other information, where there is an association relationship between the other information and the information to be indicated. It is also possible to indicate only part of the information to be indicated, while other parts of the information to be indicated are Some are known or agreed in advance. For example, the indication of specific information can also be achieved by means of a pre-agreed (for example, protocol stipulated) arrangement order of each piece of information, thereby reducing the indication overhead to a certain extent. At the same time, the common parts of each piece of information can also be identified and indicated in a unified manner to reduce the instruction overhead caused by indicating the same information individually.

Second, the first, second and various numerical numbers (for example, "#1", "#2", etc.) shown in this application are only for convenience of description and are used to distinguish objects, and are not used to limit this application. Scope of Application Embodiments. For example, distinguish different data packets, etc. It is not used to describe a specific order or sequence. It is to be understood that objects so described are interchangeable where appropriate to enable description of aspects other than the embodiments of the present application.

Third, in this application, "preconfigured" may include predefined, for example, protocol definitions. Among them, "pre-definition" can be realized by pre-saving corresponding codes, tables or other methods that can be used to indicate relevant information in the device (for example, including each network element). This application does not limit its specific implementation method.

Fourth, the “save” involved in the embodiments of this application may refer to saving in one or more memories. The one or more memories may be provided separately, or may be integrated in an encoder or decoder, a processor, or a communication device. The one or more memories may also be partially provided separately and partially integrated in the decoder, processor, or communication device. The type of memory can be any form of storage medium, and this application is not limited thereto.

Fifth, the term "and/or" in this article is just an association relationship that describes related objects, indicating that there can be three relationships. For example, A and/or B can mean: A alone exists, and A and B exist simultaneously. , there are three situations of B alone. In addition, the character "/" in this article generally indicates that the related objects are an "or" relationship.

Sixth, the "protocol" involved in the embodiments of this application may refer to standard protocols in the communication field, which may include, for example, 5G protocols, new radio (NR) protocols, and related protocols applied in future communication systems. There are no restrictions on this application.

In the following, without loss of generality, the communication method provided by the embodiment of the present application will be described in detail by taking the interaction between devices as an example.

Figure 7 is a schematic flow chart of a communication method provided by an embodiment of the present application. It can be understood that in Figure 7, the server, the core network device and the access network device are used as the execution subjects of the interactive representation as an example to illustrate the method, but this application does not limit the execution subjects of the interactive representation. For example, the server in Figure 7 can also be a chip, chip system, or processor that supports the server to implement the method, or can be a logic module or software that can realize all or part of the server functions; the core network equipment in Figure 7 can also be It is a chip, chip system, or processor that supports core network equipment to implement this method, or it can be a logic module or software that can realize all or part of the core network equipment functions; the access network equipment in Figure 7 can also be an access network equipment that supports access. The chip, chip system, or processor of the network device that implements the method may also be a logic module or software that can realize all or part of the functions of the access network device. The method includes the following steps:

S710, the server generates neural network data packets.

The neural network data packet includes indication information, and the indication information is used to indicate the priority of the neural network data packet.

The priority of the neural network data packet involved in the embodiment of the present application may refer to the scheduling (or transmission) priority of the neural network data packet. For example, the priority of the neural network data packet is used to determine whether to prioritize the scheduling of the neural network data packet. , for example, when network congestion occurs, neural network data packets with higher priority can be transmitted to the user side in time.

or,

The priority of the neural network data packet may refer to the processing priority of the neural network data packet, where the processing method includes but is not limited to any one of the following processing methods: The priority of the neural network data packet is used to determine the physical layer on the user side In the order in which neural network data packets are submitted to the application layer, neural network data packets with higher priority can be submitted to the application layer in time to restore the transmission data.

As can be seen from the above, in this embodiment, when the server generates a neural network data packet, it can carry indication information indicating the priority of the neural network data packet in the neural network data packet, so as to receive the core of the neural network data packet. The network device can read the instruction information used to indicate the priority of the neural network data packet from the neural network data packet, and learn the priority of the neural network data packet, in order to achieve differentiated transmission of neural network data packets of different priorities. .

For example, the indication information included in the neural network data packet may include the indication information in the header of the neural network data packet.

For example, the server can add indication information to the packet at the transport layer or higher. For example, the indication information can be added to the user datagram protocol (UDP) field and the real-time transport protocol (RTP) field. ) fields.

For ease of understanding, how the indication information is carried in the header of the neural network data packet will be described with reference to FIG. 8 , which is a schematic diagram of a neural network data packet provided by an embodiment of the present application.

It should be understood that Figure 8 only illustrates that the indication information can be carried in the header of the neural network data packet, and does not constitute any limitation on the protection scope of the present application. The indication information can also be added at other locations in the header of the neural network data packet. For example, between the IP field and the UDP field, or after the RTP field, etc., I will not give examples one by one here.

Specifically, the neural network data packet generated by the server corresponds to a certain neural network. For example, in the scenario shown in Figure 2 above, the server uses the ULR block as the input of the DNN, uses the original HD block as the target output of the DNN, uses the two as the training set of the DNN, and uses PSNR as the loss function for training. DNN, ULR block and DNN need to be sent to the user together. Among them, the DNN data can be encoded into multiple IPs, and the multiple IP packets are transmitted to the core network, and then the IP data packets are transmitted to the UE through the wireless access network. In this embodiment, the IP packet obtained by encoding the data of the neural network is called a neural network data packet.

It should be understood that the data of a neural network can be encoded into one or more neural network data packets. The neural network data packet generated by the above-mentioned server may be one or more neural network data packets obtained by encoding data of a certain neural network.

It should also be understood that the neural network data packet generated by the server is used to carry the parameter information of the neural network corresponding to the neural network data packet (for example, the coefficients of the neural network).

Specifically, the neural network corresponding to the neural network data packet is used to process data (eg, restore data), where the data can be video frame data, audio frame data, or image data, or other types of data. The embodiments of this application do not limit the type of data processed.

For ease of description, the following description takes the example that the data is video frame data or image data. The neural network is used to restore low-resolution images to obtain high-resolution images.

It can be seen from the above that the neural network data packet generated by the server includes indication information, and the indication information is used to indicate the priority of the neural network data packet, or it can be understood that the indication information is used to indicate the importance of the neural network data packet.

As a possible implementation method: the priority of a neural network data packet is related to the effect of the neural network recovering data corresponding to the neural network data packet.

In this implementation, the server can determine the priority of the neural network data packets obtained by encoding the data of the neural network based on the effect of the neural network on recovering data.

In this embodiment, the data restored by the neural network may be image or video data, and the effect of the data restored by the neural network may be a picture restoration effect. For example, in the super-resolution transmission mode of XR, neural networks are used to restore low-definition video frames to high-definition video frames. Different users have different restoration effects on videos corresponding to the neural network models at different times. In this implementation, the restoration effect of the neural network on the video is used as a criterion to measure the priority of the neural network data packets obtained by encoding the data of the neural network.

For example, the effect of the neural network on recovering data can be indicated by any of the following indicators: PSNR, SSIM or VMAF, etc.

For example, when the effect of the neural network on recovering data meets expectations (for example, VAMF is greater than a preset threshold), the priority of the neural network data packet obtained by encoding the data of the neural network is high priority.

For another example, when the effect of the neural network on recovering data does not meet expectations (for example, the VAMF is less than or equal to a preset threshold), the priority of the neural network data packet obtained by encoding the data of the neural network is low priority.

It should be understood that in the embodiments of the present application, there is no limit on how to determine the values of the recovery effect indicators (such as PSNR, SSIM or VMAF, etc.) corresponding to different neural networks. The determination method can be based on the current determination methods recorded in the related art, including but not limited to : Obtain the values of PSNR and SSIM by calculating the difference or correlation between the pixels of the image restored using the neural network and the original high-definition image; or,

The value of VMAF corresponding to the neural network is determined based on the experience effect of user feedback.

Optionally, take the VMAF corresponding to different neural networks to represent the effect of different neural networks on data recovery as an example to explain how the server determines the priority of neural network data packets in this implementation.

For example, the server needs to determine the neural network data packet (e.g., the data of neural network #1) encoded by 4 neural networks (e.g., neural network #1, neural network #2, neural network #3, and neural network #4). The encoded neural network data packet is neural network data packet #1, the encoded neural network data packet of neural network #2 is neural network data packet #2, and the encoded neural network data packet of neural network #3 is The neural network data packet obtained by encoding the data of neural network data packet #3 and neural network #4 is the priority of neural network data packet #4).

When the VMAFs corresponding to different neural networks represent the effects of data recovery by different neural networks, the server can determine the priority of the neural network data packets obtained by encoding the data of the four neural networks respectively based on the VMAFs corresponding to the four neural networks. .

Table 3 below gives an exemplary standard based on the VMAF (score range is 0 to 100 points) evaluation mechanism to measure the priority of neural network data packets obtained by data encoding of different neural networks. There are 4 levels in total. Among them, the larger the priority value, the higher the priority.

table 3

For example, if the VMAF score of neural network #1 is <25 points, then the priority of the neural network data packet obtained by encoding the data of neural network #1 is 1; the VMAF score of neural network #2 is [25 points, 50 points], Then the priority of the neural network data packet obtained by encoding the data of neural network #2 is 2; the VMAF score of neural network #3 is (50 points, 75 points), then the neural network data packet obtained by encoding the data of neural network #3 has a priority of 3; the VMAF score of Neural Network #4 is [75 points, 100 points], then the priority of the neural network data packet obtained by encoding the data of neural network #4 is 4.

It should be understood that Table 3 is only an example and does not constitute any limitation on the scope of protection of the present application. An evaluation mechanism different from that shown in Table 3 can also be developed based on VMAF to measure the data encoding of different neural networks to obtain the priority of neural network data packets.

It should also be understood that the above-mentioned use of the VMAF score of the neural network to represent the effect of the neural network on data recovery is only an example, and does not constitute any limitation on the protection scope of the present application. Other indicators (such as PSNR, SSIM, etc.) can also be used to represent the neural network. The effect of recovering data (for example, dividing the indicator value into multiple ranges, and each range corresponds to the degree of effect of a neural network on recovering data). The specific expression method is similar to the above-mentioned VMAF, and will not be described again here.

The representation of the priority of the neural network data packets shown in this implementation explains: the priority of the neural network data packets of the neural network can be determined according to the effect of the neural network on recovering data, so that the neural network with the best effect of recovering data can be transmitted first. The neural network data package of the network is used, so that users can prioritize the use of neural networks with good data recovery effects to restore data and improve user experience.

As another possible implementation method: the priority of a neural network data packet is related to the effect of the neural network recovering the data corresponding to the neural network data packet and the effect of the preset algorithm (or traditional algorithm) recovering the data. Among them, the preset algorithm can understand the algorithm for recovering data that the user can directly use, such as the algorithm predefined by the protocol.

In this implementation, the server can determine the priority of the neural network data packets obtained by encoding the data of the neural network based on the effect of the neural network on recovering the data and the effect of the preset algorithm on recovering the data.

In this implementation, considering that in the super-resolution transmission mode of XR, in addition to using neural networks to restore high-definition videos, users can also use some traditional algorithms (such as bilinear interpolation algorithms) To restore high-definition videos, although the traditional algorithm is simple and can be used directly by users, the restoration effect is poor compared to neural networks. In this implementation, the degree of improvement in the video recovery effect of the neural network compared to the data recovery effect of the preset algorithm is used as a criterion to measure the priority of the neural network data packets obtained by encoding the data of the neural network.

For example, the effect of the neural network on recovering the data and the effect of the preset algorithm on recovering the data can be indicated by any of the following indicators: PSNR, SSIM or VMAF, etc.

For example, the effect of the neural network on recovering the data is higher than expected compared to the effect of the preset algorithm on recovering the data (for example, the VAMF corresponding to the neural network is greater than a preset value compared to the VAMF corresponding to the preset algorithm). In this case, the priority of the neural network data packet obtained by encoding the neural network data is high priority.

For example, the effect of the neural network on recovering the data is lower than expected compared to the effect of the preset algorithm on recovering the data (for example, the VAMF corresponding to the neural network is less than or equal to a preset value compared to the VAMF corresponding to the preset algorithm. value), the priority of the neural network data packet obtained by encoding the data of the neural network is low priority.

Optionally, take the VMAF corresponding to different neural networks to represent the effect of different neural networks on data recovery, and take the VMAF corresponding to the preset algorithm to represent the effect of the preset algorithm to restore the data as an example to illustrate how the server determines the neural network data in this implementation. The priority of the package.

For example, the server needs to determine the neural network data packet (e.g., the data of neural network #1) encoded by 4 neural networks (e.g., neural network #1, neural network #2, neural network #3, and neural network #4). The encoded neural network data packet is neural network data packet #1, the encoded neural network data packet of neural network #2 is neural network data packet #2, and the encoded neural network data packet of neural network #3 is The neural network data packet obtained by encoding the data of neural network data packet #3 and neural network #4 is the priority of neural network data packet #4).

When the VMAF corresponding to different neural networks represents the effect of different neural networks on data recovery, the corresponding VMAF of the preset algorithm When VMAF indicates the effect of the preset algorithm on recovering the data, the server can determine the neural network data packet obtained by encoding the data of the four neural networks respectively based on the VMAF corresponding to the four neural networks and the VMAF corresponding to the preset algorithm. priority.

The server can use the video quality evaluation improvement information of a certain neural network compared with the traditional algorithm as a criterion for measuring the priority of the neural network. The higher the image enhancement effect of the neural network, the higher the priority of the neural network data packets obtained by its data encoding.

Table 4 below gives an exemplary range evaluation mechanism based on the VMAF difference (the difference between the VMAF corresponding to the neural network and the VMAF corresponding to the preset algorithm) to measure the data encoding of different neural networks to obtain the priority of neural network data packets. There are 4 levels of standards. Among them, the larger the priority value, the higher the priority.

Table 4

For example, if the VMAF score of neural network #1 is improved by <5 points compared to the VMAF score of the traditional algorithm, then the priority of the neural network data packet obtained by encoding the data of neural network #1 is 1; the VMAF score of neural network #2 is the same. Compared with the VMAF score of the traditional algorithm, the VMAF score is improved [5 points, 10 points], then the priority of the neural network data packet obtained by encoding the data of neural network #2 is 2; the VMAF score of neural network #3 is compared with the VMAF of the traditional algorithm If the score increases (10 points, 20 points), then the priority of the neural network data packet obtained by the data encoding of neural network #3 is 3; the VMAF score of neural network #4 is improved by ≥20 points compared with the VMAF score of the traditional algorithm. Then the priority of the neural network data packet obtained by encoding the data of neural network #4 is 4.

It should be understood that Table 4 is only an example and does not constitute any limitation on the scope of protection of the present application. An evaluation mechanism different from that shown in Table 4 can also be developed based on VMAF to measure the data encoding of different neural networks to obtain the priority of neural network data packets.

It should also be understood that the above-mentioned VMAF score of the neural network represents the effect of the neural network on recovering the data, and the VMAF corresponding to the preset algorithm represents the effect of the preset algorithm on recovering the data. This is only an example and does not constitute any limitation on the scope of protection of the present application. It can also be Other indicators (such as PSNR, SSIM, etc.) represent the effect of the neural network on recovering the data and the effect of the preset algorithm on recovering the data. The specific expression method is similar to the above-mentioned VMAF, and will not be described again here.

The representation of the priority of the neural network data packets shown in this implementation explains: the priority of the neural network data packets of the neural network can be determined based on the effect of the neural network on recovering the data and the effect of the preset algorithm on recovering the data. When the data recovery effect of multiple neural networks is better than that of the preset algorithm, the neural network that recovers the data better than the preset algorithm can be prioritized so that the user can use the recovered data first. The effective neural network restores data and improves user experience.

As another possible implementation method: the priority of the neural network data packet is related to the effect of reconstructing the neural network, and the neural network data packet is used to reconstruct the neural network corresponding to the neural network data packet.

In this implementation, the server can determine the priority of the neural network data packet based on the effect of reconstructing the neural network from the neural network data packet.

For example, in the data reconstruction of the neural network packet, the neural network #1 is obtained and the neural network trained by the server is When the network gap meets expectations (for example, the number of layers of the constructed neural network #1 and the neural network that needs to be transmitted to the user is the same), determine the priority of the neural network data packet as high priority.

For example, the gap between the neural network #1 obtained by data reconstruction of the neural network data packet and the neural network obtained by server training does not meet expectations (for example, the number of layers of the constructed neural network #1 and the neural network that needs to be transmitted to the user If they are not the same), determine the priority of the neural network data packet as low priority.

The representation of the priority of the neural network data packets shown in this implementation shows that the priority of the neural network data packets of the neural network can be determined according to the effect of the reconstructed neural network, so that the effect of the reconstructed neural network is good for priority transmission. neural network data package to improve the efficiency of users in reconstructing neural networks.

As another possible implementation method: the priority of the neural network data packet is related to the degree of influence of the neural network coefficient data included in the neural network data packet on the neural network coefficient.

In this implementation manner, the server may determine the priority of the neural network data packet based on the degree of impact of the data on the coefficients of the neural network included in the neural network data packet on the coefficients of the neural network.

For example, when the difference between coefficient #1 and the coefficient calculated by the coefficient data meets expectations (for example, the difference between coefficient #1 and the coefficient is less than or equal to the preset threshold), determine the value of the neural network data packet The priority is high priority.

For another example, in the case where the difference between coefficient #1 and the coefficient calculated by the coefficient data does not meet expectations (for example, the difference between coefficient #1 and the coefficient is greater than a preset threshold), determine the value of the neural network data packet The priority is low priority.

The representation of the priority of the neural network data packet shown in this implementation explains: the priority of the neural network data packet can be determined based on the difference between the value calculated from the data of the neural network data packet and the coefficient of the neural network, so as to facilitate Priority is given to transmitting neural network data packets that can calculate the values closest to the coefficients of the neural network, so that users can quickly reconstruct the neural network based on the received neural network data packets, and improve the efficiency of the user's reconstruction of the neural network.

Exemplarily, the coefficient is represented by a plurality of bits, and the values of the plurality of bits are used to calculate the absolute value of the coefficient. The data of the coefficient is represented by at least one bit, and the at least one bit belongs to the plurality of bits. Bit.

Specifically, the plurality of bits includes a sign part, an exponent part and a fraction part. In the case where the at least one bit is the first part of the sign part, the exponent part and the fraction part, the priority of the neural network data packet is the first priority; in the case where the at least one bit is the second part of the fractional part, the scheduling priority corresponding to the neural network is determined to be the second priority, wherein the first priority is higher than the third Second priority, the first part of the fractional part is the high-order data part of the fractional part, and the second part of the fractional part is the low-order data part of the fractional part.

It can be seen from the above that the coefficients of the neural network can be represented by the symbolic part, the exponential part and the fractional part, which is helpful to determine the degree of influence of different parts on the absolute value of the coefficients of the neural network, so that the data of the neural network data packet can be obtained by those of the coefficients Part of it indicates the degree of influence of the data of the neural network packet on the absolute value of the coefficient of the neural network, making the solution more concise.

For ease of understanding, how to determine the priority of a neural network data packet based on the degree of impact of the neural network coefficient data included in the neural network data packet on the neural network coefficient will be explained with reference to a specific example.

Example 1: The coefficients of the neural network are represented by 32-bit floating point data (as shown in Figure 6). Based on the structural characteristics of the floating point data mentioned above, the bits that will have a large impact on the absolute value of the coefficients of the neural network Bits are regarded as high-priority data, and bits that have a small impact on the absolute value of the coefficient of the neural network are regarded as low-priority data.

For example, the sign bit, exponent bit and high-order bit of the fractional part of the floating-point number corresponding to the coefficient of the neural network can be compared to The remaining fractional low-order bits are treated as data with low priority.

When the server generates neural network data packets of the neural network, the different bits are placed in different data packets according to the degree of influence they have on the absolute value of the coefficients of the neural network. As shown in Figure 9, Figure 9 is a schematic diagram of generating neural network data packets provided by an embodiment of the present application.

As can be seen from Figure 9, the sign bit, exponent bit and high-order bits of the fractional part of the floating point number corresponding to the neural network coefficient are placed in a neural network data packet, and the priority in the neural network data packet is set is "H"; and the low-order bits of the fractional part of the floating point number corresponding to the coefficient of the neural network are placed in another neural network data packet, and the priority in the neural network data packet is set to "L", where " H" means high priority, "L" means low priority.

Further, in order to transmit the neural network data packet to the user, the server can send the generated neural network data packet to the core network device. The method flow shown in Figure 7 also includes:

S720: The server sends the neural network data packet to the core network device, or the core network device receives the neural network data packet from the server.

It should be understood that the core network device receiving the neural network data packet from the server is only a way for the core network device to obtain the neural network data packet.

For example, core network equipment can obtain neural network data packets in the following ways:

Method 1: The core network device receives the neural network data packet from the server.

For example, the server directly sends the neural network data packet to the core network device. For another example, the server indirectly sends the neural network data packet to the core network device through other devices.

In this method one, the core network equipment and the server can be connected through a fixed network.

Method 2: The core network device obtains the neural network data packet generated by the server from the memory (or internal interface).

In the second method, the core network device and the server can be combined into one device (for example, in a mobile edge computing (MEC) scenario), and the neural network data packets generated by the server can be cached in the memory of the joint device , and the core network device obtains the neural network data packet from the memory; or the core network device and the server in the joint device transmit the data packet through the internal interface, and the core network device can obtain the neural network data from the server through the internal interface Bag.

It should be understood that the above-mentioned methods 1 and 2 for the core network device to obtain neural network data packets are only examples and do not limit the scope of protection of the present application. In this embodiment, the core network device can also obtain the neural network data through other methods. Bag. For example, the neural network data packet is generated according to the received parameter information of the neural network, which will not be described again here.

In this embodiment, after receiving the neural network data packet, the core network device may read indication information indicating the priority of the neural network data packet from the neural network data packet. The indication information will then be encapsulated into the GTP-U data packet header according to the GTP-U protocol so that the access network device can read it.

The method flow shown in Figure 7 also includes:

S730, the core network device generates GTP-U data packets.

The payload of the GTP-U data packet includes a neural network data packet, and the header of the GTP-U data packet includes indication information. The indication information is used to indicate the priority of the neural network data packet.

In this embodiment, after the core network device receives the neural network data packet carrying the indication information indicating the priority of the neural network data packet, it can read the priority of the neural network data packet from the neural network data packet. class The instruction information is encapsulated into the GTP-U data packet header according to the GTP-U protocol, and the neural network data packet is used as the payload of the GTP-U data packet. So that the access network device that receives the GTP-U data packet can read the indication information from the header of the GTP-U data packet, and based on the priority of the neural network data packet indicated by the indication information, the neural network Data packets are transmitted and processed. It can be understood that the access network device takes the priority of the neural network data packet into consideration when transmitting the neural network data packet, so as to achieve differentiated transmission of neural network data packets of different priorities.

In this embodiment, the indication information included in the header of the GTP-U data packet for indicating the priority of the neural network data packet (which may be called indication information #1 for distinction), and the indication information included in the neural network data packet for indicating The specific form of the indication information of the priority of the neural network data packet (which may be called indication information #2 for distinction) may be the same or different.

Specifically, in this embodiment, after the core network device receives the neural network data packet, it can read the instruction information #2 used to indicate the priority of the neural network data packet from the neural network data packet, and then can follow the GTP-U The protocol encapsulates the indication information #2 into the GTP-U data packet header, which is called indication information #1, so that the access network equipment station can read it. The instruction information #2 encapsulated according to the GTP-U protocol may still be called instruction information #2, or may be called instruction information #1 for ease of distinction. That is to say, in this embodiment, the specific forms of the indication information #1 and the indication information #2 are not limited, and they can be used to indicate the priority of the neural network data packet. For ease of description, they can be collectively called instruction information.

For example, a new field can be added to the GTP-U data packet header to indicate the priority of the neural network data packet.

For ease of understanding, it is explained that the GTP-U data packet header includes indication information with reference to FIG. 10 . FIG. 10 is a schematic diagram of a GTP-U data packet header provided by an embodiment of the present application.

As can be seen from Figure 10, a new byte can be added to the GTP-U data packet to add indication information.

It should be understood that Figure 10 only illustrates that the indication information can be carried in the header of the GTP-U data packet, and does not constitute any limitation on the protection scope of the present application. The indication information can also be added to other parts of the header of the GTP-U data packet. The position, for example, in the fourth bit of the GTP-U packet, will not be explained one by one here.

Further, the core network device sends the GTP-U data packet to the access network device. The method flow shown in Figure 7 also includes:

S740: The core network device sends the GTP-U data packet to the access network device, or the access network device receives the GTP-U data packet from the core network device.

S750: The access network device transmits the neural network data packet according to the instruction information.

The header of the GTP-U data packet received by the access network device includes indication information indicating the neural network data packet, so that the access network device can read the indication information from the header of the GTP-U data packet, and based on the The priority of the neural network data packet indicated by the indication information, and the neural network data packet is transmitted and processed. It can be understood that the access network device takes the priority of the neural network data packet into consideration when transmitting the neural network data packet, so as to achieve differentiated transmission of neural network data packets of different priorities.

For example, the access network device receives two GTP-U data packets (GTP-U data packet #1 and GTP-U data packet #2). Among them, the payload of GTP-U data packet #1 includes a neural network data packet. #1, the header of GTP-U packet #1 includes indication information #1, the payload of GTP-U packet #2 includes neural network packet #2, the header of GTP-U packet #2 includes indication information #2. The access network device differentially transmits neural network data packet #1 and neural network data packet #2 according to instruction information #1 and instruction information #2. For example, instruction information #1 indicates that the priority of neural network data packet #1 is high priority. class, Instruction information #2 indicates that the priority of neural network data packet #2 is low priority. The access network device can transmit neural network data packet #1 first, and then transmit neural network data packet #1 after the transmission of neural network data packet #1 is completed. 2.

Exemplarily, the access network device transmits the neural network data packet according to the indication information, including:

When the indication information indicates that the neural network data packet is of high priority, the access network device preferentially transmits the neural network data packet; or,

When the indication information indicates that the neural network data packet is of low priority, the access network device delays transmission of the neural network data packet; or,

When the indication information indicates that the neural network data packet is of low priority and the network status is congested, the access network device gives up transmitting the neural network data packet.

Specifically, the access network device can calculate the corresponding data based on the neural network packet priority information (such as the above-mentioned indication information) and other related parameters of air interface scheduling, including but not limited to historical rate, instantaneous rate, user level, etc. transmission priority.

For example, traditional proportional fair scheduling priority meets the following conditions:
factor1＝R _i ÷R _h

Among them, factor1 represents the proportional fair scheduling priority, R _i represents the user's instantaneous rate. The better the user's channel condition, the higher the instantaneous rate; R _i represents the user's historical rate, which represents the average rate of the channel within a period of time.

Specifically, the access network device can determine the scheduling priority of the neural network data packet based on the proportional fair scheduling algorithm and the priority of the neural network data packet. For example, the scheduling priority of the neural network data packet satisfies the following conditions: factor2= f(N)×R _i ÷R _h

Among them, factor2 represents the scheduling priority of the neural network data packet, N represents the priority of the neural network data packet, and f can be an increasing linear or exponential function.

It should be understood that the specific examples shown in FIG. 7 in the embodiments of the present application are only to help those skilled in the art better understand the embodiments of the present application, but are not intended to limit the scope of the embodiments of the present application. It should also be understood that the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its functions and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

For example, the embodiment shown in Figure 7 records that determining the priority of neural network data packets is implemented by the server. Another possibility is that the above-mentioned action of determining the priority of neural network data packets can be implemented by the core network device. Specifically determine The method is similar to the method determined by the server, except that the execution subject is the core network device.

It should also be understood that in the various embodiments of the present application, if there are no special instructions or logical conflicts, the terms and/or descriptions between different embodiments are consistent and can be referenced to each other. The technical features in different embodiments New embodiments can be formed based on their internal logical relationships.

It should also be understood that in some of the above embodiments, network elements in the existing network architecture are mainly used as examples for illustrative explanations (such as AF, AMF, SMF, etc.). It should be understood that the specific form of the network element is The application examples are not limiting. For example, network elements that can implement the same functions in the future are applicable to the embodiments of this application.

It can be understood that in the above method embodiments, the methods and operations implemented by devices (such as servers, core network equipment, and access network equipment) can also be implemented by components (such as chips or circuits) that can be used in network equipment.

Above, the communication method provided by the embodiment of the present application is described in detail with reference to FIG. 7 . The above communication methods are mainly introduced from the perspective of interaction between various network elements. It can be understood that, in order to implement the above functions, each network element includes a corresponding hardware structure and/or software module to perform each function.

Those skilled in the art should realize that the present application can be implemented in the form of hardware or a combination of hardware and computer software with the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein. Whether a function is performed by hardware or computer software driving the hardware depends on the specific application and design constraints of the technical solution. Skilled artisans may use different methods to implement the described functionality for specific applications, but such implementations should not be considered beyond the scope of this application.

Hereinafter, the communication device provided by the embodiment of the present application will be described in detail with reference to FIG. 11 and FIG. 12 . It should be understood that the description of the device embodiments corresponds to the description of the method embodiments. Therefore, for content that is not described in detail, please refer to the above method embodiments. For the sake of brevity, some content will not be described again.

Figure 11 is a schematic block diagram of a communication device provided by an embodiment of the present application. As shown in Figure 11, the device 1100 may include an interface unit 1110 and a processing unit 1120. The interface unit 1110 can communicate with the outside, and the processing unit 1120 is used for data processing. The interface unit 1110 may also be called a communication interface, communication unit or transceiver unit.

Optionally, the device 1100 may also include a storage unit, which may be used to store instructions and/or data, and the processing unit 1120 may read the instructions and/or data in the storage unit, so that the device implements the foregoing method embodiments. .

The device 1100 can be used to perform the actions performed by the transceiver equipment (such as servers, core network equipment, and access network equipment) in the above method embodiments. In this case, the device 1100 can be a transceiver device or can be configured on the transceiver device. Components, the interface unit 1110 is used to perform operations related to the transceiver of the transceiver device in the above method embodiment, and the processing unit 1120 is used to perform operations related to the processing of the transceiver device in the above method embodiment.

As a design, the device 1100 is used to perform the actions performed by the access network equipment in the above method embodiment.

The interface unit 1110 is configured to receive a General Packet Radio Service Tunneling Protocol user plane GTP-U data packet. The payload of the GTP-U data packet includes a neural network data packet. The header of the GTP-U data packet includes indication information. The indication information is used to indicate the priority of the neural network data packet;

The processing unit 1120 is configured to control the device to transmit the neural network data packet according to the instruction information.

The apparatus 1100 may implement steps or processes corresponding to those executed by the access network equipment in the method embodiments of the present application, and the apparatus 1100 may include units for executing the methods executed by the access network equipment in the method embodiments. . Moreover, each unit in the device 1100 and the above-mentioned other operations and/or functions are respectively intended to implement the corresponding processes of the method embodiment in the access network equipment in the method embodiment.

When the device 1100 is used to perform the method in FIG. 7, the interface unit 1110 can be used to perform the sending and receiving steps in the method, such as step S740; the processing unit 720 can be used to perform the processing steps in the method, such as step S750.

It should be understood that the specific process of each unit performing the above corresponding steps has been described in detail in the above method embodiments, and will not be described again for the sake of brevity. In addition, the beneficial effects brought about by each unit performing the above corresponding steps have been described in detail in the above method embodiments, and will not be described again here.

As another design, the device 1100 is used to perform the actions performed by the core network equipment in the above method embodiment.

The interface unit 1110 is used to obtain a neural network data packet. The neural network data packet includes indication information, and the indication information is used to indicate the priority of the neural network data packet;

The interface unit 1110 is also used to send a General Packet Wireless Service Tunneling Protocol user plane GTP-U data packet to the access network device. The payload of the GTP-U data packet includes the neural network data packet. The GTP-U data packet Includes this instruction information in the packet header.

The device 1100 may implement steps or processes corresponding to those executed by the core network equipment in the method embodiments of the present application. The device 1100 may include a unit for executing the method executed by the core network equipment in the method embodiments. Yuan. Moreover, each unit in the device 1100 and the above-mentioned other operations and/or functions are respectively intended to implement the corresponding processes of the method embodiment in the core network equipment in the method embodiment.

When the device 1100 is used to perform the method in Figure 7, the interface unit 1110 can be used to perform the sending and receiving steps in the method, such as step S720; the processing unit 720 can be used to perform the processing steps in the method, such as step S730.

It should be understood that the specific process of each unit performing the above corresponding steps has been described in detail in the above method embodiments, and will not be described again for the sake of brevity. In addition, the beneficial effects brought about by each unit performing the above corresponding steps have been described in detail in the above method embodiments, and will not be described again here.

As yet another design, the device 1100 is used to perform the actions performed by the server in the above method embodiment.

The processing unit 1120 is configured to generate a neural network data packet. The neural network data packet includes indication information, and the indication information is used to indicate the priority of the neural network data packet;

The interface unit 1110 is used to send the neural network data packet.

The device 1100 may implement steps or processes corresponding to those executed by the server in the method embodiments of the embodiments of the present application, and the device 1100 may include a unit for executing the method executed by the server in the method embodiments. Moreover, each unit in the device 1100 and the above-mentioned other operations and/or functions are respectively intended to implement the corresponding processes of the method embodiment in the server in the method embodiment.

When the device 1100 is used to perform the method in Figure 7, the interface unit 1110 can be used to perform the sending and receiving steps in the method, such as step S720; the processing unit 720 can be used to perform the processing steps in the method, such as step S710.

It should be understood that the specific process of each unit performing the above corresponding steps has been described in detail in the above method embodiments, and will not be described again for the sake of brevity. In addition, the beneficial effects brought about by each unit performing the above corresponding steps have been described in detail in the above method embodiments, and will not be described again here.

The processing unit 1120 in the above embodiments may be implemented by at least one processor or processor-related circuit. The interface unit 1110 may be implemented by a transceiver or transceiver related circuitry. The storage unit may be implemented by at least one memory.

As shown in Figure 12, this embodiment of the present application also provides a device 1200. The apparatus 1200 includes a processor 1210 and may also include one or more memories 1220. The processor 1210 is coupled to the memory 1220. The memory 1220 is used to store computer programs or instructions and/or data. The processor 1210 is used to execute the computer programs or instructions and/or data stored in the memory 1220, so that the method in the above method embodiment be executed. Optionally, the device 1200 includes one or more processors 1210 .

Optionally, the memory 1220 can be integrated with the processor 1210 or provided separately.

Optionally, as shown in Figure 12, the device 1200 may also include a transceiver 1230, which is used for receiving and/or transmitting signals. For example, the processor 1210 is used to control the transceiver 1230 to receive and/or transmit signals.

As a solution, the device 1200 is used to implement the operations performed by the transceiver equipment (such as server, core network equipment, and access network equipment) in the above method embodiment.

Embodiments of the present application also provide a computer-readable storage medium on which are stored computer instructions for implementing the method executed by the transceiver device (such as a server, a core network device, and an access network device) in the above method embodiment.

For example, when the computer program is executed by a computer, the computer can implement the method executed by the transceiver device (such as a server, core network device, and access network device) in the above method embodiment.

Embodiments of the present application also provide a computer program product containing instructions. When the instructions are executed by a computer, the computer implements the method executed by the transceiver device (such as a server, a core network device, and an access network device) in the above method embodiment.

An embodiment of the present application also provides a communication system, which includes the server, core network equipment, and access network equipment in the above embodiment.

For explanations of relevant content and beneficial effects of any of the devices provided above, please refer to the corresponding method embodiments provided above, and will not be described again here.

It should be understood that the processor mentioned in the embodiments of this application may be a central processing unit (CPU), or other general-purpose processor, digital signal processor (DSP), or application-specific integrated circuit (ASIC). application specific integrated circuit (ASIC), off-the-shelf programmable gate array (field programmable gate array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

It should also be understood that the memory mentioned in the embodiments of this application may be a volatile memory and/or a non-volatile memory. Among them, non-volatile memory can be read-only memory (ROM), programmable ROM (PROM), erasable programmable read-only memory (erasable PROM, EPROM), electrically removable memory. Erase electrically programmable read-only memory (EPROM, EEPROM) or flash memory. Volatile memory can be random access memory (RAM). For example, RAM can be used as an external cache. As an example and not a limitation, RAM may include the following forms: static random access memory (static RAM, SRAM), dynamic random access memory (dynamic RAM, DRAM), synchronous dynamic random access memory (synchronous DRAM, SDRAM) , double data rate synchronous dynamic random access memory (double data rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (synchlink DRAM, SLDRAM) and Direct memory bus random access memory (direct rambus RAM, DR RAM).

It should be noted that when the processor is a general-purpose processor, DSP, ASIC, FPGA or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, the memory (storage module) can be integrated in the processor.

It should also be noted that the memories described herein are intended to include, but are not limited to, these and any other suitable types of memories.

Those of ordinary skill in the art will appreciate that the units and steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of protection of this application.

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

The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or they may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to implement the solution provided by this application.

In addition, each functional unit in each embodiment of the present application can be integrated into one unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. For example, the computer may be a personal computer, a server, or a network device. The computer instructions may be stored in or transmitted from one computer-readable storage medium to another, e.g., the computer instructions may be transferred from a website, computer, server, or data center Transmission to another website, computer, server or data center by wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains one or more available media integrated. The available media may be magnetic media (such as floppy disks, hard disks, magnetic tapes), optical media (such as DVDs), or semiconductor media (such as solid state disks (SSD)), etc. For example, the aforementioned available media may include But it is not limited to: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk and other media that can store program code.

The above are only specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application. should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims (22)

1. A communication method, characterized by including:

   Receive a General Packet Radio Service Tunneling Protocol user plane GTP-U data packet. The payload of the GTP-U data packet includes a neural network data packet. The header of the GTP-U data packet includes indication information. The indication information is To indicate the priority of the neural network data packet;

   Transmitting the neural network data packet according to the indication information.
2. The method of claim 1, wherein transmitting the neural network data packet according to the indication information includes:

   When the indication information indicates that the neural network data packet is of high priority, the neural network data packet is transmitted first; or,

   When the indication information indicates that the neural network data packet is of low priority, delay the transmission of the neural network data packet; or,

   When the indication information indicates that the neural network data packet is of low priority and the network status is congested, the transmission of the neural network data packet is given up.
3. A communication method, characterized by including:

   Obtain a neural network data packet, the neural network data packet includes indication information, the indication information is used to indicate the priority of the neural network data packet;

   Send a General Packet Radio Service Tunneling Protocol user plane GTP-U data packet to the access network device, the payload of the GTP-U data packet includes the neural network data packet, and the header of the GTP-U data packet includes the instructions.
4. The method according to any one of claims 1 to 3, characterized in that the priority of the neural network data packet is related to the effect of the neural network recovering data corresponding to the neural network data packet.
5. The method according to claim 4, characterized in that:

   The priority of the neural network data packet is also related to the effect of the preset algorithm on recovering the data.
6. The method according to any one of claims 1 to 3, characterized in that the neural network data packet is used to reconstruct the neural network corresponding to the neural network data packet,

   The priority of the neural network data packet is related to the effect of reconstructing the neural network.
7. The method according to claim 6, characterized in that the neural network data packet includes data of coefficients of the neural network, the data of the coefficients are used to obtain the coefficients, and the coefficients are used to reconstruct the Neural Networks,

   The priority of the neural network data packet is related to the impact of the data of the coefficient on the coefficient.
8. The method according to any one of claims 1 to 7, characterized in that the indication information is carried in a header of the neural network data packet.
9. The method according to claim 8, characterized in that the indication information is located between the User Datagram Protocol UDP field and the Real-time Transport Protocol RTP field in the neural network data packet header.
10. A communication device, characterized by including:

    An interface unit configured to receive a General Packet Radio Service Tunneling Protocol user plane GTP-U data packet, the payload of the GTP-U data packet includes a neural network data packet, and the header of the GTP-U data packet includes indication information, The indication information is used to indicate the priority of the neural network data packet;

    A processing unit, configured to control the device to transmit the neural network data packet according to the instruction information.
11. The device according to claim 10, wherein the processing unit controls the device to transmit the neural network data packet according to the instruction information, including:

    In the case where the indication information indicates that the neural network data packet is of high priority, the processing unit controls the device to transmit the neural network data packet with priority; or,

    When the indication information indicates that the neural network data packet is of low priority, the processing unit controls the device to delay transmission of the neural network data packet; or,

    When the indication information indicates that the neural network data packet is of low priority and the network status is congested, the processing unit controls the device to give up transmitting the neural network data packet.
12. A communication device, characterized by including:

    An interface unit, configured to obtain a neural network data packet, where the neural network data packet includes indication information, and the indication information is used to indicate the priority of the neural network data packet;

    The interface unit is also configured to send a General Packet Wireless Service Tunneling Protocol user plane GTP-U data packet to the access network device. The payload of the GTP-U data packet includes the neural network data packet, and the GTP-U data packet contains the neural network data packet. The header of the U data packet includes the indication information.
13. The device according to any one of claims 10 to 12, characterized in that the priority of the neural network data packet is related to the effect of the neural network recovering data corresponding to the neural network data packet.
14. The device according to claim 13, characterized in that:

    The priority of the neural network data packet is also related to the effect of the preset algorithm on recovering the data.
15. The device according to any one of claims 10 to 12, characterized in that the neural network data packet is used to reconstruct the neural network corresponding to the neural network data packet,

    The priority of the neural network data packet is related to the effect of reconstructing the neural network.
16. The device according to claim 15, characterized in that the neural network data packet includes data of coefficients of the neural network, the data of the coefficients are used to obtain the coefficients, and the coefficients are used to reconstruct the Neural Networks,

    The priority of the neural network data packet is related to the impact of the data of the coefficient on the coefficient.
17. The device according to any one of claims 10 to 16, characterized in that the indication information is carried in a header of the neural network data packet.
18. The device according to claim 17, wherein the indication information is located between a User Datagram Protocol (UDP) field and a Real-time Transport Protocol (RTP) field in the header of the neural network data packet.
19. A communication device, characterized in that it includes a processor, the processor is coupled to a memory, the memory is used to store computer programs or instructions, and the processor is used to execute the computer program or instructions in the memory, so that the The device performs the method according to any one of claims 1 to 9.
20. A computer-readable storage medium, characterized in that a computer program or instructions are stored on the computer-readable storage medium, and when the computer program or instructions are run on a computer, the computer is caused to execute the steps as claimed in claims 1 to 1. The method described in any one of 9.
21. A chip system, characterized by comprising: a processor, configured to call and run a computer program from a memory, so that a communication device installed with the chip system executes the method described in any one of claims 1 to 9.
22. A computer program product, characterized in that, when the computer program product is run on a computer, it causes the computer to execute the method according to any one of claims 1 to 9.