A 5G and MEC coupling method and device for low-latency video services

Abstract

The application provides a 5G and MEC coupling method and device for low-latency video services, which comprises the following steps: determining a next frame size prediction result of a target image frame according to application layer context information; performing real-time link dynamicity estimation on 5G access network state information to determine a wireless link quality estimation result of the 5G access network; fusing the next frame size prediction result and the wireless link quality estimation result into a wireless resource demand rule; and controlling real-time data interaction between the 5G access network and the edge server according to the wireless resource demand rule based on an incentive and acceleration mechanism to adaptively adjust the wireless resource allocation for the video service. The application can accurately determine the wireless resource demand of the video service, adaptively adjust the wireless resource for the video service, and guarantee the reliability of resource allocation, thereby ensuring the latency sensitivity of the video service such as video analysis and video processing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless resource scheduling, and particularly relates to a 5G and MEC coupling method and device for low-latency video services. BACKGROUND

[0002] The rise of 5G has inspired many exciting application scenarios, such as remote medical surgery, cloud gaming, mobile AR / VR, V2X, video analytics, etc. They have put strict requirements on the performance of 5G - consuming a large amount of bandwidth and computing resources and ensuring ultra-low latency response of data services. Especially in latency-sensitive video analytics, 5G is usually required to respond to each image frame within 200 milliseconds, otherwise there is a risk of a decline in user experience quality (QoE) or even more serious risks, such as traffic accidents in the case of vision-assisted autonomous driving.

[0003] In order to meet its QoE requirements, the innovation of 5G NR alone is not enough. Studies have shown that although the packet-level delay between the user terminal and the base station has been reduced to below 2 milliseconds, the end-to-end delay (from the user terminal to the remote cloud) is still as high as tens or even hundreds of milliseconds (depending on the geographical location of the remote cloud). Therefore, the 5G community has proposed a 5G MEC architecture and regards it as a key component of the 5G ecosystem. The basic idea is to deploy computing servers close to the location of the end user (such as the edge side of the 5G network) rather than on the remote cloud, in order to shorten the transmission path and reduce the end-to-end transmission delay. In addition, resource-constrained user terminals can also offload computing-intensive data services with the help of the computing power of the edge cloud.

[0004] Traditional 5G MEC works focus more on architecture design, theory or simulation-driven analysis and small-scale prototype implementation. They usually ignore the actual QoE research of edge-assisted applications, especially the latency-sensitive nature. Through large-scale field measurement experiments on an operational 5G MEC network, we found that the current 5G network and MEC framework are not coordinated, and there are drawbacks of "each for its own" and weak interaction capabilities. This results in long service response delays for latency-sensitive video analytics applications (only 44.11% of image frames can reach the server in time) and a 27.89% decline in inference performance. After in-depth investigation and research, we revealed that the root cause is that the current 5G and MEC are loosely coupled. On the one hand, the application program on the edge server cannot know the wireless instantaneous dynamics of the 5G RAN, so it cannot timely perform sufficient wireless resource adaptation; on the other hand, the 5G RAN does not understand the real wireless resource requirements of the upper-layer application (i.e. how much wireless resource is needed for data transmission and how to efficiently grant them), but only allocates wireless resources through inefficient multiple rounds of "request-grant" operations. Therefore, the above two factors jointly cause the application response delay to be too long. SUMMARY

[0005] In view of this, embodiments of this application provide a 5G and MEC coupling method and apparatus for low-latency video services, in order to eliminate or improve one or more defects existing in the prior art.

[0006] One aspect of this application provides a 5G and MEC coupling method for low-latency video services, comprising:

[0007] Based on the context information of the target image frame corresponding to the low-latency video service obtained from the edge server, the next frame size prediction result of the target image frame is determined;

[0008] Real-time link dynamics estimation is performed on the current status information of the 5G access network obtained by the edge server to determine the wireless link quality estimation result of the 5G access network.

[0009] The next frame size prediction result and the wireless link quality estimation result are fused into the wireless resource demand rule;

[0010] Based on a preset incentive and acceleration mechanism, the 5G access network and the edge server are controlled to perform real-time data interaction according to the wireless resource demand rules in order to adaptively adjust the allocation of wireless resources for the video service.

[0011] In some embodiments of this application, determining the next frame size prediction result of the target image frame based on the context information of the target image frame corresponding to the video service obtained by the edge server includes:

[0012] Based on the context information of the target image frame corresponding to the video service obtained from the edge server, a single-frame feature extraction is performed on the target image frame using a regression learning algorithm to obtain the feature vector of the target image frame.

[0013] The feature vector of the target image frame is combined with the feature vectors of other image frames corresponding to the video service in a temporal sequence to obtain a corresponding temporal feature vector group; the temporal feature vector group is subjected to temporal feature extraction processing to obtain the content change features of the next frame image of the target image frame, and the content change features of the next frame image are used as the size prediction result of the next frame.

[0014] In some embodiments of this application, the step of extracting single-frame features from the target image frame based on a regression learning algorithm, using context information of the target image frame corresponding to the video service obtained from the edge server, to obtain the feature vector of the target image frame, includes:

[0015] The context information of the target image frame corresponding to the video service is obtained locally on the edge server. The context information includes: the size information, content information and semantic information of the target image frame corresponding to the video service sent from the user terminal.

[0016] The size information of the target image frame is input into a first feature extractor, so that the first feature extractor outputs the size feature vector of the target image frame;

[0017] The content information of the target image frame is input into the second feature extractor so that the second feature extractor outputs the frame feature vector of the target image frame itself. The second feature extractor includes a pre-trained ResNet-50 model.

[0018] Furthermore, the semantic information of the target image frame is input into a third feature extractor so that the third feature extractor outputs the number of target objects and the motion state feature vector in the target image frame;

[0019] The feature vector of the target image frame is obtained based on the size feature vector, the frame's own feature vector, and the number of target objects and the motion state feature vector.

[0020] In some embodiments of this application, the step of performing real-time link dynamics estimation on the current state information of the 5G access network obtained by the edge server to determine the radio link quality estimation result of the 5G access network includes:

[0021] The current status information of the 5G access network is obtained by applying the radio network information service corresponding to the edge server;

[0022] Based on a preset likelihood estimation method, the current state information of the 5G access network is used to perform real-time link dynamics estimation, thereby obtaining the next SINR change estimate of the 5G access network, and using the next SINR change estimate as the wireless link quality estimation result.

[0023] In some embodiments of this application, fusing the next frame size prediction result and the radio link quality estimation result into the radio resource demand rule includes:

[0024] Based on the correspondence between the pre-acquired application layer wireless resources and the physical bit capacity, the next frame size prediction result is converted into the corresponding target physical bit capacity.

[0025] Based on the target physical bit capacity, the wireless link quality estimation result, and the preset physical channel overhead, a target PRB requirement is generated as the wireless resource requirement rule.

[0026] In some embodiments of this application, controlling the real-time data interaction between the 5G access network and the edge server according to the wireless resource demand rules based on a preset incentive and acceleration mechanism to adaptively adjust the allocation of wireless resources for the video service includes:

[0027] The real-time PRB authorization status is obtained from the current status information of the 5G access network, and the timer simultaneously determines whether the current resource scheduling unit is less than the deadline threshold.

[0028] If so, the incentive mechanism in the preset incentive and acceleration mechanism is activated, and the video service is adjusted in a fair priority manner based on the target PRB demand between the 5G access network and the edge server.

[0029] If not, the acceleration mechanism in the incentive and acceleration mechanism is activated, and the acceleration priority of the video service is adaptively adjusted between the 5G access network and the edge server based on the target PRB demand.

[0030] Another aspect of this application provides a 5G and MEC coupling device for low-latency video services, the 5G and MEC coupling device for low-latency video services being used to perform the 5G and MEC coupling method for low-latency video services.

[0031] Some embodiments of this application include:

[0032] The cross-layer wireless resource demand customization module is used to determine the next frame size prediction result of the target image frame based on the context information of the target image frame corresponding to the video service obtained by the edge server; to perform real-time link dynamic estimation on the current state information of the 5G access network obtained by the edge server to determine the wireless link quality estimation result of the 5G access network; and to integrate the next frame size prediction result and the wireless link quality estimation result into the wireless resource demand rule.

[0033] The deadline-sensitive resource granting module is used to control the real-time data interaction between the 5G access network and the edge server according to the wireless resource demand rules based on a preset incentive and acceleration mechanism, so as to adaptively adjust the allocation of wireless resources for the video service.

[0034] Another aspect of this application provides an interaction system between a 5G access network and a MEC, comprising: a user plane and a control plane partitioned in the software space of an edge server;

[0035] The user plane is used to implement user plane functions that route or offload traffic to edge applications;

[0036] The control plane is equipped with the 5G and MEC coupling device for low-latency video services.

[0037] In some embodiments of this application, it also includes:

[0038] Each transmission interface is used to enable data interaction between the user plane and the 5G access network inside the interactive system, the user terminal outside the interactive system, and the 5G core network.

[0039] The northbound interface is used to enable data interaction between the coupling device and the edge application.

[0040] The southbound interface is used to enable data interaction between the coupling device and the 5G access network.

[0041] The 5G and MEC coupling method for low-latency video services provided in this application determines the next frame size prediction result of the target image frame based on the context information of the target image frame corresponding to the video service obtained by the edge server. This effectively reveals the actual wireless resource requirements of the user terminal and provides a more accurate and effective data foundation for wireless resource scheduling for the user terminal. By performing real-time link dynamic estimation on the current state information of the 5G access network obtained by the edge server, the method effectively predicts the wireless link quality of the 5G access network, further improving the accuracy and effectiveness of wireless resource scheduling for the user terminal. By integrating the next frame size prediction result and the wireless link quality estimation result into the wireless resource requirement rule, the method effectively improves the customization... The system enhances the accuracy and intelligence of application service requirements, effectively improving the user experience of video analytics applications. By controlling real-time data interaction between the 5G access network and the edge server according to the wireless resource requirement rules based on preset incentive and acceleration mechanisms, it adaptively schedules wireless resources for video services. This enables adaptive adjustment of wireless resources and ensures reliable allocation, further improving the intelligence and effectiveness of the wireless resource scheduling process. This allows the system to meet the diverse application scenarios of video analytics, particularly those requiring latency, effectively improving the user experience for latency-sensitive applications, such as enhancing the driver's experience in vision-assisted autonomous driving use cases.

[0042] Additional advantages, objectives, and features of this application will be set forth in part in the description which follows, and will in part become apparent to those skilled in the art upon review of the following text, or may be learned by practice of the application. The objectives and other advantages of this application can be realized and obtained by means of the structures specifically pointed out in the specification and drawings.

[0043] Those skilled in the art will understand that the purposes and advantages that can be achieved with this application are not limited to those specifically described above, and that the above and other purposes that this application can achieve will be more clearly understood from the following detailed description. Attached Figure Description

[0044] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, do not constitute a limitation thereof. The components in the drawings are not drawn to scale but are merely for illustrating the principles of this application. For ease of illustration and description of certain parts of this application, corresponding portions in the drawings may be enlarged, i.e., may appear larger relative to other components in an exemplary device actually manufactured according to this application. In the drawings:

[0045] Figure 1 This is a schematic diagram of the overall process of a 5G and MEC coupling method for low-latency video services according to an embodiment of this application.

[0046] Figure 2 This is a schematic diagram of a specific process of a 5G and MEC coupling method for low-latency video services according to an embodiment of this application.

[0047] Figure 3 This is a schematic diagram of the specific process of step 210 in the 5G and MEC coupling method for low-latency video services in one embodiment of this application.

[0048] Figure 4 This is a schematic diagram of the structure of a 5G and MEC coupling device for low-latency video services according to another embodiment of this application.

[0049] Figure 5 This is a schematic diagram of the interaction system between the 5G access network and MEC in another embodiment of this application.

[0050] Figure 6 This is a schematic diagram of the workflow of the software-defined 5G and MEC coupled architecture provided in the application example of this application.

[0051] Figure 7 This is a schematic diagram of the context-aware application layer content change prediction algorithm provided in the application example of this application.

[0052] Figure 8 This is a schematic diagram of a deadline-sensitive resource granting method provided in an application example of this application.

[0053] Figure 9 This is a schematic diagram of a standard-compliant interaction protocol provided in the application example of this application. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and their descriptions are used to explain this application, but are not intended to limit it.

[0055] It should also be noted that, in order to avoid obscuring this application with unnecessary details, only the structures and / or processing steps closely related to the solution according to this application are shown in the accompanying drawings, while other details that are not closely related to this application are omitted.

[0056] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0057] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.

[0058] In the following description, embodiments of the present application will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0059] Mobile edge computing (MEC), a key component of the 5G ecosystem, is designed to support emerging 5G applications with stringent latency requirements. The basic idea is to deploy servers closer to end users, such as at the network edge rather than in the traditional remote cloud. While this is conceptually sound, we have found that commercial 5G networks have failed to coordinate with existing MEC architectures, resulting in intolerable long response times.

[0060] Traditional 5G MEC work focuses more on architecture design, theoretical or simulation-driven analysis, and niche prototype implementations. They typically neglect research into the actual quality of user experience (QoE) of edge-assisted applications, especially those with latency-critical requirements. Specifically:

[0061] (1) 5G and MEC Optimization. Traditional work has devoted much effort to 5G MEC performance measurement, control plane design, and mobility management. Orion, LL-MEC, FlexRIC, and 5G-EmPower have established programmable SD-RAN platforms to support 5G edge computing services. However, these works have not considered the impact of physical layer link quality dynamics and application layer content changes on the QoE of edge applications.

[0062] (2) Delaying the guarantee of critical application requirements. Existing methods for guaranteeing application requirements include network slicing, which includes resource reservation, resource virtualization, function virtualization, and slice isolation. However, these methods rely on operators to modify the network infrastructure, including user terminals, access networks, and core networks. This approach is costly and has slow iteration updates.

[0063] This application proposes a software-defined 5G and MEC coupled architecture or device implementation, which can be referred to as "Tutti". To achieve this coupling capability, Tutti first integrates instantaneous radio dynamic information from the 5G RAN with edge-assisted application layer content change information to precisely tailor radio resource requirements. Then, Tutti facilitates real-time interaction between the 5G RAN and the edge server to ensure reliable resource allocation.

[0064] The following examples will provide a detailed description.

[0065] This application provides a 5G and MEC coupling method for low-latency video services. See [link to relevant documentation]. Figure 1 The 5G and MEC coupling method for low-latency video services, which can be implemented by a 5G and MEC coupling device for low-latency video services, specifically includes the following:

[0066] Step 100: Determine the next frame size prediction result of the target image frame based on the context information of the target image frame corresponding to the video service obtained by the edge server.

[0067] It is understood that the predicted size of the next frame can specifically be the size or dimensions of the next image frame of the target image frame, which can be denoted as F. n+1 .

[0068] Step 200: Perform real-time link dynamics estimation on the current status information of the 5G access network obtained by the edge server, and determine the wireless link quality estimation result of the 5G access network.

[0069] It is understood that the wireless link quality estimation result can be: wireless link quality, which can be labeled as I. est .

[0070] It is understood that steps 100 and 200 can be executed sequentially or synchronously, depending on the actual application scenario and the processing capabilities of the 5G and MEC coupling device.

[0071] Step 300: Integrate the next frame size prediction result and the wireless link quality estimation result into the wireless resource demand rule.

[0072] It is understood that the aforementioned wireless resource demand rules can refer to wireless resource demand rule data, specifically manifested as a quantitative result. For example, physical channel overhead.

[0073] Step 400: Based on a preset incentive and acceleration mechanism, control the 5G access network and the edge server to perform real-time data interaction according to the wireless resource demand rules in order to adaptively adjust the allocation of wireless resources for the video service.

[0074] The video service corresponding to the user terminal continuously sends radio resource requests to the base station via the 5G network and waits for grants. Once the user terminal obtains the resources allocated by the base station, it immediately uploads the data packets (image frames) in its buffer to the edge server for calculation. The user terminal then uses the inference results fed back from the server for subsequent processing. The data packets sent by the user terminal are the context information of the target image frame corresponding to the video service mentioned in one or more embodiments of this application. Furthermore, the video analysis application can specifically refer to latency-sensitive video analysis applications, such as visual assistance analysis applications in autonomous driving.

[0075] In one or more embodiments of this application, the context information of the target image frame may also be referred to as application layer context information.

[0076] In one or more embodiments of this application, the 5G access network can be written as 5G RAN, and the edge server refers to a mobile edge computing (MEC) device.

[0077] In one or more embodiments of this application, the wireless link quality estimation result may be referred to as the wireless instantaneous dynamic estimation result.

[0078] The 5G and MEC coupling method for low-latency video services provided in this application can meet the needs of latency-critical applications. To achieve this coupling capability, the 5G and MEC coupling method for low-latency video services first deeply understands the network dynamics of the 5G RAN and the content changes of upper-layer applications to accurately customize the service requirements of applications. Then, through a series of lightweight, real-time interactive "micro" operations between the 5G RAN and edge servers, it adaptively adjusts the allocation of service resources. For example, when the edge server knows that an image frame is about to reach its critical deadline, it immediately urges the base station to prioritize its resource allocation. More importantly, current 5G operators are using network slicing technology to divide dedicated communication, network, and computing resource slices to cater to the performance of 5G niche use cases. The design concept of the 5G and MEC coupling method for low-latency video services is included in this, and its cross-layer optimization and lightweight implementation methods also provide operators with feasible network slicing examples.

[0079] As described above, the 5G and MEC coupling method for low-latency video services provided in this application determines the next frame size prediction result of the target image frame based on the context information of the target image frame corresponding to the video service obtained by the edge server. This effectively reveals the actual wireless resource requirements of the user terminal and provides a more accurate and effective data foundation for wireless resource scheduling for the user terminal. By performing real-time link dynamic estimation on the current state information of the 5G access network obtained by the edge server, the wireless link quality of the 5G access network can be effectively predicted, further improving the accuracy and effectiveness of wireless resource scheduling for the user terminal. By integrating the next frame size prediction result and the wireless link quality estimation result into the wireless resource requirement rule, it can achieve... This system effectively improves the accuracy and intelligence of customized application service requirements and enhances the user experience of video analytics applications. By controlling real-time data interaction between the 5G access network and the edge server according to the wireless resource requirement rules based on preset incentive and acceleration mechanisms, it adaptively schedules wireless resources for video services. This enables adaptive adjustment of wireless resources and ensures reliable allocation of wireless resources, further improving the intelligence and effectiveness of the wireless resource scheduling process. This allows it to meet the diverse application scenarios of video analytics, especially those of latency-sensitive video analytics applications, effectively improving the user experience of such applications, for example, enhancing the driver's experience in vision-assisted autonomous driving use cases.

[0080] To further improve the accuracy, reliability, and effectiveness of wireless resource demand forecasting, in a 5G and MEC coupling method for low-latency video services provided in this application embodiment, see [link to relevant documentation]. Figure 2 Step 100 in the 5G and MEC coupling method for low-latency video services specifically includes the following:

[0081] Step 110: Based on the context information of the target image frame corresponding to the video service obtained from the edge server, perform single-frame feature extraction on the target image frame using a regression learning algorithm to obtain the feature vector of the target image frame.

[0082] Step 120: Combine the feature vector of the target image frame with the feature vectors of other image frames corresponding to the video service in a temporal sequence to obtain the corresponding temporal feature vector group; perform temporal feature extraction processing on the temporal feature vector group to obtain the content change features of the next frame image of the target image frame, and use the content change features of the next frame image as the size prediction result of the next frame.

[0083] In step 120, the input time-series feature set is used to predict the size of the next frame (i.e., F). n+1 Each feature group contains feature vectors from m historical serialized frames. The core algorithm in this stage is based on the GRU (Gate Recurrent Unit) feature extractor. It is essentially an iterator. It combines a new feature group with past historical feature groups to generate intermediate results (i.e.,...). To deduce the next image size value (i.e., the content change value), we can use the following:

[0084]

[0085] and

[0086] To further improve the reliability, effectiveness, and accuracy of single-frame feature extraction, a 5G and MEC coupling method for low-latency video services is provided in this application embodiment, see [link to relevant documentation]. Figure 3 Step 110 of the 5G and MEC coupling method for low-latency video services specifically includes the following:

[0087] Step 111: Obtain the context information of the target image frame corresponding to the video service locally on the edge server. The context information includes: the size information, content information, and semantic information of the target image frame corresponding to the video service sent from the user terminal.

[0088] Step 112: Input the size information of the target image frame into the first feature extractor so that the first feature extractor outputs the size feature vector of the target image frame.

[0089] Step 113: Input the content information of the target image frame into the second feature extractor so that the second feature extractor outputs the frame feature vector of the target image frame itself, wherein the second feature extractor includes: a pre-trained ResNet-50 model.

[0090] Step 114: Input the semantic information of the target image frame into the third feature extractor so that the third feature extractor outputs the number of target objects and the motion state feature vector in the target image frame.

[0091] It is understandable that steps 112 to 114 can be executed in any order, such as sequentially or synchronously, depending on the actual application scenario and the processing capabilities of the 5G and MEC coupled devices for low-latency video services.

[0092] Step 115: Obtain the feature vector of the target image frame based on the size feature vector, the frame itself feature vector, and the number of target objects and the motion state feature vector.

[0093] It's understandable that ResNet stands for Deep Residual Network. The "50" in the "ResNet-50" model refers to the 50 layers.

[0094] Specifically, the scene diversity of video analytics applications is mainly affected by three factors: background complexity, the number of object elements, and the motion state of the objects (reflecting the degree of scene change). We use regression learning algorithms to understand these three factors. In particular, (1) the image frames uploaded by the user terminal are essentially serialized. This can help Tutti understand the degree of change in background complexity. (2) The number of object elements and motion state can be detected from the inferred semantic information. For example, we can count the bounding boxes detected by the video analytics model as the number of object elements and judge the degree of scene change by tracking the position changes of related objects in the serialized frames.

[0095] The image frame size (F) of the i-th frame in history i Image frame content (C) i ) and semantic information (S i The input is fed into the corresponding feature extractor, and the output is a feature vector (V). i We formulate them as follows:

[0096]

[0097] in, This refers to the concatenation operation. MF(·) and MS(·) are embedding layers that quantify the latent serialization features of image frame size, number of objects, and object movement states, respectively. Tutti also utilizes an identity-related encoder through ResNet-50 (i.e., R(·)) to obtain the frame's own features. Notably, F... i It is a floating-point variable; C i It is a 3D matrix (we resize each .JPEG frame to 224×224, i.e., with three RGB channels for easier model input); S i This includes the detected bounding boxes and the motion state of object elements.

[0098] To further improve the reliability and effectiveness of real-time link dynamics estimation, in a 5G and MEC coupling method for low-latency video services provided in this application embodiment, see [link to relevant documentation]. Figure 2Step 200 in the 5G and MEC coupling method for low-latency video services specifically includes the following:

[0099] Step 210: Use the radio network information service corresponding to the edge server to obtain the current status information of the 5G access network.

[0100] Step 220: Based on a preset likelihood estimation method, perform real-time link dynamics estimation on the current state information of the 5G access network to obtain the next SINR change estimate of the 5G access network, and use the next SINR change estimate as the wireless link quality estimation result.

[0101] SINR refers to the signal-to-interference-plus-noise ratio.

[0102] Specifically, traditional wireless link quality estimation algorithms are mainly based on wireless channel measurements, whereby user terminals test the quality of the wireless link between themselves and the serving base station and upload the measurement results as a whole to the base station. However, the upload time of this method is not fixed, easily leading to delays of hundreds of milliseconds, and it cannot perceive link status information with fine granularity. Therefore, Tutti adopts a prediction-based approach, that is, whenever we want to know the channel quality, we can directly estimate it without waiting for the channel measurement results.

[0103] We first observe that the signal-to-interference-plus-noise ratio (SINR) data approximately follows a truncated Gaussian distribution in the short term. This inspires us to use likelihood estimation to predict the next SINR change. First, we formulate the likelihood model as:

[0104] I est =L({I1,…,I i ,..,I m}).

[0105] Among them, I est This is the estimated SINR value at the next time step. in, This is the set of historical SINR values ​​for the short term. L(·) is the likelihood function. To reduce estimation error, we set L(·) to... The mathematical expectation E(·):

[0106]

[0107] f(·) is the probability density function of a truncated Gaussian distribution. φ(·) is the probability density function of the standard Gaussian distribution; Φ(·) is the probability distribution function of the correlation. We define...

[0108] To further improve the reliability and effectiveness of information fusion applications, in a 5G and MEC coupling method for low-latency video services provided in this application embodiment, see [link to relevant documentation]. Figure 2 Step 300 in the 5G and MEC coupling method for low-latency video services specifically includes the following:

[0109] Step 310: Based on the pre-acquired correspondence between application layer wireless resources and physical bit capacity, convert the next frame size prediction result into the corresponding target physical bit capacity;

[0110] Step 320: Based on the target physical bit capacity, the wireless link quality estimation result, and the preset physical channel overhead, generate the target PRB demand as the wireless resource demand rule.

[0111] The PRB refers to the Authorized Physical Resource Block.

[0112] Specifically, after understanding the changes in application layer content and dynamic information of the radio link, the next step is to translate these into specific radio resource requirement rules during the base station licensing process. In other words, Tutti needs to convert the image frame size F... n+1 and wireless link quality I est Integrate into the resource wireless resource demand rules (denoted as) ).

[0113] However, to achieve this, two problems need to be solved:

[0114] (1) Bridging the gap between the amount of data at the application layer and the amount of bits at the physical layer. We know that F n+1 This refers to the application layer frame size, but the PRB configuration is based on the physical layer bit size. This capacity difference is mainly determined by protocol cost, packet retransmission, control signaling, etc. We represent the physical bit size as... Assume there are J data packets transmitted to the (n+1)th frame, and each data packet is s in size. j (1≤j≤J), therefore we obtain Furthermore, considering header cost and packet retransmission, we express the physical bit capacity as:

[0115]

[0116] Where, r j This indicates whether the j-th data packet eventually reached the edge server. If it arrived successfully, r... j =0; otherwise, 1. k j This represents the number of retransmissions for the j-th data packet. ∈ represents the protocol cost. We set ∈ = 0.068. This represents the proportion of control signaling. In an ideal situation (i.e., r...),... j =0 and k j =1, indicating that all data packets arrived successfully on the first try. We infer that:

[0117]

[0118] (2) Filling Physical Channel Overhead. When base stations allocate Licensed Physical Resource Blocks (PRBs), they typically isolate some redundant PRBs to prevent carrier interference. We look up the 3GPP 5G NR specification (Rel-15 TS 38.306) and... Quantified as:

[0119]

[0120] Where OH is the physical channel overhead. In the sub-6GHz NR band, its value is 0.08. M(·) is the channel modulation function, used to determine the modulation type and order, coding rate, and efficiency.

[0121] To further improve the reliability and intelligence of adaptive adjustment of wireless resource allocation for video services, a 5G and MEC coupling method for low-latency video services is provided in this application embodiment, see [link to relevant documentation]. Figure 2 Step 400 in the 5G and MEC coupling method for low-latency video services specifically includes the following:

[0122] Step 410: Obtain the real-time PRB authorization status from the current status information of the 5G access network, and the timer simultaneously determines whether the current resource scheduling unit is less than the deadline threshold; wherein, the deadline threshold can be set to 100 milliseconds.

[0123] If so, proceed to step 420: activate the incentive mechanism in the preset incentive and acceleration mechanism, and perform fair priority adaptive adjustment of wireless resources for the video service between the 5G access network and the edge server based on the target PRB demand.

[0124] If not, proceed to step 430: activate the acceleration mechanism in the incentive and acceleration mechanism, and adaptively adjust the acceleration priority of the video service for wireless resources between the 5G access network and the edge server based on the target PRB demand.

[0125] Understandably, while acquiring the real-time PRB grant status (recorded in the RAN status message), a timer (initially 200 milliseconds) determines whether the current TTI (Resource Scheduling Unit) is approaching its deadline. We set the threshold for this determination to 100 milliseconds. On one hand, if there is sufficient time (e.g., the current TTI is less than 100 milliseconds), we activate an incentive mechanism to perform a fair priority adjustment—Tutti allocates as many resources as possible to latency-critical users to support rapid data uploads without sacrificing the QoE of other users. On the other hand, if the deadline is imminent, Tutti urges an acceleration mechanism to exponentially accelerate priority growth. Furthermore, regardless of the mechanism, Tutti diagnoses whether the currently allocated PRB meets the resource guarantee threshold, i.e., the minimum PRB requirement to maintain application operation. For example, if the allocated resources are below the PRB guarantee threshold, Tutti will update the priority to a higher level in the next TTI. Finally, Tutti returns a PRB grant priority indicator to the serving base station in a resource configuration message.

[0126] From a software-defined perspective, this application also provides a 5G and MEC coupling device for performing all or part of the 5G and MEC coupling method for low-latency video services, see [link to relevant documentation]. Figure 4 The 5G and MEC coupling device for low-latency video services specifically includes the following components:

[0127] The cross-layer wireless resource demand customization module 10 is used to determine the next frame size prediction result of the target image frame based on the context information of the target image frame corresponding to the video service obtained by the edge server; to perform real-time link dynamic estimation on the current state information of the 5G access network obtained by the edge server to determine the wireless link quality estimation result of the 5G access network; and to integrate the next frame size prediction result and the wireless link quality estimation result into the wireless resource demand rule.

[0128] The deadline-sensitive resource granting module 20 is used to control the real-time data interaction between the 5G access network and the edge server according to the wireless resource demand rules based on a preset incentive and acceleration mechanism, so as to adaptively adjust the allocation of wireless resources for the video service.

[0129] The embodiments of the 5G and MEC coupling device for low-latency video services provided in this application can be used to execute the processing flow of the embodiments of the 5G and MEC coupling method for low-latency video services in the above embodiments. Its functions will not be repeated here, but can be referred to the detailed description of the embodiments of the 5G and MEC coupling method for low-latency video services described above.

[0130] As described above, the 5G and MEC coupling device for low-latency video services provided in this application embodiment determines the next frame size prediction result of the target image frame based on the context information of the target image frame corresponding to the video service obtained by the edge server. This effectively reveals the actual wireless resource requirements of the user terminal and provides a more accurate and effective data foundation for wireless resource scheduling for the user terminal. By performing real-time link dynamic estimation on the current state information of the 5G access network obtained by the edge server, the wireless link quality of the 5G access network can be effectively predicted, further improving the accuracy and effectiveness of wireless resource scheduling for the user terminal. By integrating the next frame size prediction result and the wireless link quality estimation result into the wireless resource requirement rules, it is possible to achieve... This system effectively improves the accuracy and intelligence of customized application service requirements and enhances the user experience of video analytics applications. By controlling real-time data interaction between the 5G access network and the edge server according to the wireless resource requirement rules based on preset incentive and acceleration mechanisms, it adaptively schedules wireless resources for video services. This enables adaptive adjustment of wireless resources and ensures reliable allocation of wireless resources, further improving the intelligence and effectiveness of the wireless resource scheduling process. This allows it to meet the diverse application scenarios of video analytics, especially those of latency-sensitive video analytics applications, effectively improving the user experience of such applications, for example, enhancing the driver's experience in vision-assisted autonomous driving use cases.

[0131] Based on the above embodiments of the 5G and MEC coupling device and the 5G and MEC coupling method for low-latency video services, this application also provides an embodiment of an interaction system between a 5G access network and MEC, see [link to embodiment]. Figure 5 The interaction system between the 5G access network and MEC specifically includes the following:

[0132] The user plane and control plane are divided in the software space of the edge server;

[0133] The user plane is used to implement user plane functions that route or offload traffic to edge applications;

[0134] The control plane is equipped with the 5G and MEC coupling device for low-latency video services.

[0135] In one embodiment of an interaction system between a 5G access network and a MEC, the interaction system further includes:

[0136] Each transmission interface is used to enable data interaction between the user plane and the 5G access network inside the interactive system, the user terminal outside the interactive system, and the 5G core network; the transmission interfaces include: GTP-U interface, N4 interface, N5 interface, and N6 interface.

[0137] The northbound interface is used to enable data interaction between the coupling device and the edge application.

[0138] The southbound interface is used to enable data interaction between the coupling device and the 5G access network.

[0139] To further illustrate this solution, this application also provides a specific application example of a software-defined 5G RAN and MEC coupling architecture for implementing the aforementioned 5G and MEC coupling method for low-latency video services, to meet the needs of latency-critical applications. To achieve this coupling capability, Tutti first deeply understands the network dynamics of the 5G RAN and the content changes of upper-layer applications to precisely tailor the service requirements of the applications. Then, Tutti adaptively adjusts service resource allocation through a series of lightweight, real-time interactive "micro" operations between the 5G RAN and edge servers. For example, when an edge server knows that an image frame is about to reach its critical deadline, it immediately urges the base station to prioritize its resource allocation. More importantly, current 5G operators are using network slicing technology to divide dedicated communication, network, and computing resource slices to cater to the performance of 5G niche use cases. Tutti's design philosophy is among these approaches, and its cross-layer optimization and lightweight implementation methods also provide operators with a feasible network slicing example.

[0140] See Figure 6Tutti's operation process: The user terminal continuously sends radio resource requests to the serving base station and waits for grant; once the resources allocated by the base station are obtained, the user terminal immediately uploads the data packets (image frames) in the buffer to the edge server for calculation; the user terminal then uses the inference results fed back from the server for subsequent processing. In the above process, ensuring that the user terminal obtains sufficient radio resources in a timely manner is the key to improving application QoE. To this end, Tutti takes 5G RAN status information and application layer context information (obtained from the edge server) as system inputs and outputs resource configuration through two main modules to guide the base station's radio resource grant: (1) Cross-layer radio resource demand customization method. This method is used to deeply understand the radio resource situation required for the next frame delivery. Specifically, it predicts the changes in application layer content (i.e., frame size) through regression learning and uses a statistical likelihood model to estimate the radio link dynamics that change over time. Then, Tutti integrates the intermediate results into the demand rules to calibrate the amount of resources required by the upper layer applications. (2) To ensure that these radio resources are granted before the deadline, Tutti adopts a critical deadline-sensitive radio resource granting method. Specifically, Tutti employs an incentive and acceleration mechanism to dynamically determine the priority of resource allocation—if the current time is far from the deadline, it will allocate lower-priority resources. Otherwise, it will urgently accelerate resource allocation.

[0141] The specific explanation is as follows:

[0142] (I) Customization of Cross-Layer Wireless Resource Requirements

[0143] Accurate understanding of the wireless resource requirements for video analytics is affected by changes in application layer content and wireless channel quality. Fortunately, Tutti can address these challenges by fusing cross-layer information. In short, (1) because Tutti is located on an edge server, it can directly obtain application layer context information, including frame data sent from user terminals and semantic information inferred from video analytics models (e.g., bounding boxes of identified objects and trajectory tracking trajectories). (2) The MEC standard introduces a new network function entity, Radio Network Information Service (RNIS). It can decode physical layer and network layer link feedback information and expose millisecond-granular RAN status information (such as PRB grants and packet transmission recorders) to upper-layer services.

[0144] Tutti can leverage this cross-layer information to customize radio resource requirements. Specifically, Tutti first inputs the application layer context into the application layer content change prediction module to obtain the next frame size. Second, the real-time radio link dynamic estimation algorithm uses fine-grained RAN state information to predict radio link quality. Finally, Tutti integrates these into the radio resource requirement rules to determine the amount of resources required for the next frame transmission.

[0145] 1.1 Context-Aware Application-Layer Content Change Prediction Algorithm

[0146] The diversity of scenarios in video analytics applications is mainly influenced by three factors: background complexity, the number of object elements, and the motion state of objects (reflecting the degree of scene change). We use regression learning algorithms to understand these three factors. Specifically, (1) the image frames uploaded by the user terminal are essentially serialized. This helps Tutti understand the degree of change in background complexity. (2) The number of object elements and motion state can be detected from the inferred semantic information. For example, we can count the bounding boxes detected by the video analytics model as the number of object elements and judge the degree of scene change by tracking the positional changes of related objects in the serialized frames.

[0147] Algorithm execution flow. This algorithm consists of... Figure 7 The process consists of two stages: (1) Single-frame feature extraction stage. It extracts the image frame size (F) of the i-th frame from the past. i Image frame content (C) i ) and semantic information (S i The input is fed into the corresponding feature extractor, and the output is a feature vector (V). i We formulate them as follows:

[0148]

[0149] in, This refers to the concatenation operation. MF(·) and MS(·) are embedding layers that quantify the latent serialization features of image frame size, number of objects, and object movement states, respectively. Tutti also utilizes an identity-related encoder through ResNet-50 (i.e., R(·)) to obtain the frame's own features. Notably, F... i It is a floating-point variable; C i It is a 3D matrix (we resize each .JPEG frame to 224×224, i.e., with three RGB channels for easier model input); S i This includes the detected bounding boxes and the motion state of object elements. (2) Temporal feature extraction stage. Input the time series feature set to predict the size of the next frame (i.e., F). n+1 Each feature group contains feature vectors from m historical serialized frames. The core algorithm in this stage is based on the GRU (Gate Recurrent Unit) feature extractor. It is essentially an iterator. It combines a new feature group with past historical feature groups to generate intermediate results (i.e.,...). To deduce the next image size value (i.e., the content change value), we can use the following:

[0150]

[0151] and

[0152] Model Training. We trained the prediction model on a subset of the ILSVRC2015-VID dataset containing 1281K images. 60% of the dataset was used for training, and the remainder for model testing. During training, we used the L1 loss function and Adam stochastic gradient descent algorithm to optimize the model parameters. Furthermore, we set the learning rate to 1e-6. The batch size was set to 32 by default. All training operations were performed on a single NVIDIA Tesla V100 GPU.

[0153] 1.2 Real-time Wireless Link Dynamic Estimation Algorithm

[0154] Traditional wireless link quality estimation algorithms are primarily based on wireless channel measurements, whereby user terminals test the quality of their wireless link with the serving base station and upload the measurement results as a whole to the base station. However, the upload time of this method is not fixed, easily leading to delays of hundreds of milliseconds, and it cannot perceive link status information with fine granularity. Therefore, Tutti adopts a prediction-based approach, meaning that whenever we want to know the channel quality, we can directly estimate it without waiting for the channel measurement results.

[0155] We first observe that the signal-to-interference-plus-noise ratio (SINR) data approximately follows a truncated Gaussian distribution in the short term. This inspires us to use likelihood estimation to predict the next SINR change. First, we formulate the likelihood model as:

[0156] I est =L({I1,…,I i ,..,I m}).

[0157] Among them, I est This is the estimated SINR value at the next time step. in, This is the set of historical SINR values ​​for the short term. L(·) is the likelihood function. To reduce estimation error, we set L(·) to... The mathematical expectation E(·):

[0158]

[0159] f(·) is the probability density function of a truncated Gaussian distribution. φ(·) is the probability density function of the standard Gaussian distribution; Φ(·) is the probability distribution function of the correlation. We define...

[0160] 1.3 Demand Rule Generation Algorithm Based on Information Fusion

[0161] After understanding the changes in application layer content and dynamic information of the radio link, the next step is to translate these into specific radio resource requirement rules during the base station licensing process. In other words, Tutti needs to change the image frame size F... n+1 and wireless link quality I est Integrate into the resource wireless resource demand rules (denoted as) ).

[0162] However, to achieve this, two problems need to be solved: (1) bridging the gap between the amount of data in the application layer and the amount of bits in the physical layer. We know that F n+1 This refers to the application layer frame size, but the PRB configuration is based on the physical layer bit size. This capacity difference is mainly determined by protocol cost, packet retransmission, control signaling, etc. We represent the physical bit size as... Assume there are J data packets transmitted to the (n+1)th frame, and each data packet is s in size. j (1≤j≤J), therefore we obtain Furthermore, considering header cost and packet retransmission, we express the physical bit capacity as:

[0163]

[0164] Where, r j This indicates whether the j-th data packet eventually reached the edge server. If it arrived successfully, r... j =0; otherwise, 1. k j This represents the number of retransmissions for the j-th data packet. ∈ represents the protocol cost. We set ∈ = 0.068. This represents the proportion of control signaling. In an ideal situation (i.e., r...),... j =0 and k j =1, indicating that all data packets arrived successfully on the first try. We infer that:

[0165]

[0166] (2) Filling Physical Channel Overhead. When base stations allocate Licensed Physical Resource Blocks (PRBs), they typically isolate some redundant PRBs to prevent carrier interference. We look up the 3GPP 5G NR specification (Rel-15 TS 38.306) and... Quantified as:

[0167]

[0168] Where OH is the physical channel overhead. In the sub-6GHz NR band, its value is 0.08. M(·) is the channel modulation function, used to determine the modulation type and order, coding rate, and efficiency.

[0169] (ii) Deadline-Sensitive Wireless Resource Granting Methods

[0170] After determining the rules for radio resource requirements, Tutti needs to ensure that they are faithfully enforced within the deadline. One straightforward approach is for the base station to reserve sufficient PRBs for a given user terminal and grant them promptly upon receiving a request. However, this approach has two drawbacks: (1) it relies on the reconfiguration of the network infrastructure (e.g., resource pool management and network entity virtualization). This violates the operator's encapsulation licensing authority. (2) Forced subscriptions may create unfair competition for other users.

[0171] Therefore, Tutti employs a deadline-sensitive approach to radio resource provisioning, enforcing radio resource demand rules in a lightweight and fair manner. The basic idea is that Tutti continuously monitors the PRB (Programming Restriction Board) grant status of base stations and adaptively adjusts resource allocation priorities for delay-critical users while maintaining fairness for other users.

[0172] We are Figure 8 The diagram illustrates its operational flow. While acquiring the real-time PRB grant status (recorded in the RAN status message), a timer (initially 200 milliseconds) determines if the current TTI (Resource Scheduling Unit) is approaching its deadline. We set the threshold for this determination to 100 milliseconds. On one hand, if there is sufficient time (e.g., the current TTI is less than 100 milliseconds), we activate an incentive mechanism to perform a fair priority adjustment—Tutti allocates as many resources as possible to latency-critical users to support rapid data uploads without sacrificing the QoE of other users. On the other hand, if the deadline is imminent, Tutti urges an acceleration mechanism to exponentially accelerate priority growth. Furthermore, regardless of the mechanism, Tutti diagnoses whether the currently allocated PRB meets the resource guarantee threshold, i.e., the minimum PRB requirement to maintain application operation. For example, if the allocated resources are below the PRB guarantee threshold, Tutti will update the priority to a higher level in the next TTI. Finally, Tutti returns a PRB grant priority indicator to the serving base station in a resource configuration message.

[0173] Incentive mechanisms for granting fair shares. We first define a metric S. t (where t is the TTI identifier), to represent the current TTI. The service satisfaction of the assigned PRB is:

[0174]

[0175] This indicates that the data was assigned to the k-th user. Can the safety net be met? Therefore, signal function It has three value types: 1 (when ), 0 (when ), or -1 (when ). S represents t Gain, which is determined by the service priority window (P) k )and (Right now, Confirmed. Furthermore, we will also mark the service satisfaction of key users who experience delays as... In summary, we adjusted the service priority window for critical users (i.e., This guides the allocation of resources to base stations. We describe it as...

[0176]

[0177] For example, (1) And S t <0 indicates that although It can meet the resource needs of latency-sensitive users, but it is unfair to other users (S t <0). Therefore, Tutti will reduce its service priority window (e.g., prioritize the last one). (2) When And S t When ≥0, all users have sufficient resources. Tutti can encourage granting more resources to users whose latency is critical. (3) (regardless of S) t How to represent The basic resource needs of latency-critical users cannot be met. Tutti should prioritize their wireless resource requirements. However, to avoid unfairness caused by blindly increasing resources, Tutti adopts linear growth. The increment step size is set to...

[0178] An acceleration mechanism to meet the final critical deadline. While the above incentive mechanism can guarantee resource allocation, it still carries a risk—it cannot guarantee that a single frame will be fully delivered by the deadline (i.e.,...). For example, even While linearly increasing resources is possible, resources might not be allocated first to latency-critical users due to other higher-priority users. Therefore, a "bottom-line approach" is necessary to ensure data transmission is completed before the deadline. Intuitively, the acceleration mechanism is that the base station should prioritize providing radio resources to users as the current TTI approaches its deadline. We increase resources exponentially. To achieve this, that is

[0179]

[0180] in, This indicates how close the deadline is. It refers to the resource supply status.

[0181] 2.2 Standard-compliant interaction protocol

[0182] To facilitate communication between the RAN and the edge server, we designed a standard-compliant interaction protocol that covers processes such as traffic forwarding, application layer context passing, and RAN status / resource configuration message exchange.

[0183] like Figure 9 As shown, following the principles of software-defined networking and network function virtualization, we divide the software space of the edge server into a user plane and a control plane, ensuring backward compatibility with 3GPP LTE / 5G and ETSI MEC standards. The user plane function (UPF) aims to route or offload traffic to edge applications, while the control plane primarily refers to functional entities related to Tutti. On one hand, we developed GTP-U / N4 / N5 / N6 interfaces to ensure reliable user plane connections for packet transmission. On the other hand, we designed a northbound API (for internal interaction between Tutti and edge applications) and a southbound API (between Tutti and the RAN) to communicate control plane messages. For example, Tutti can subscribe to application layer context information via the northbound API and transmit RAN status and resource configuration messages via the southbound API.

[0184] Importantly, to reduce the additional time spent on interface communication, we first defined a set of RESTful operations to unify message types on the northbound / southbound APIs, such as echo requests (GET) and replies (POST). Secondly, we standardized the message format, starting with a common header (8 bytes) that includes the protocol version, message type (GET / POST), flags, and length fields. For example, flag fields determine the message body content, such as application layer context (00), RAN status (01), resource configuration (10), and blank (11).

[0185] This application proposes and implements a software-defined 5G and MEC coupled architecture. First, a cross-layer radio resource demand customization method is designed, which customizes the radio resource requirements for each frame by deeply understanding and integrating application layer content changes and the instantaneous dynamics of the physical link. Second, a critical deadline-sensitive radio resource granting method is designed to facilitate interaction between the 5G RAN and edge servers, enabling on-demand resource supply and ensuring timely allocation of the designated resources.

[0186] Those skilled in the art can implement the various exemplary components, systems, and methods described herein in conjunction with the embodiments disclosed herein, either in software or hardware, or a combination of both. The specific implementation method depends on the particular application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave.

[0187] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the academic ideas of this application.

[0188] In this application, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0189] The above description is merely a preferred embodiment of this application and is not intended to limit this application. For those skilled in the art, various modifications and variations can be made to the embodiments of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A 5G and MEC coupling method for low-latency video services, characterized in that, include: Based on the context information of the target image frame corresponding to the low-latency video service obtained from the edge server, the next frame size prediction result of the target image frame is determined; Real-time link dynamics estimation is performed on the current status information of the 5G access network obtained by the edge server to determine the wireless link quality estimation result of the 5G access network. Based on the pre-acquired correspondence between application layer radio resources and physical bit capacity, the next frame size prediction result is converted into the corresponding target physical bit capacity; based on the target physical bit capacity, the radio link quality estimation result, and the preset physical channel overhead, a target PRB requirement is generated as a radio resource requirement rule. The real-time PRB authorization status is obtained from the current status information of the 5G access network. At the same time, the timer determines whether the current resource scheduling unit is less than the deadline threshold. If so, the incentive mechanism in the preset incentive and acceleration mechanism is activated, and the wireless resource priority of the video service is adaptively adjusted between the 5G access network and the edge server based on the target PRB demand. If not, the acceleration mechanism in the incentive and acceleration mechanism is activated, and the acceleration priority of the video service is adaptively adjusted between the 5G access network and the edge server based on the target PRB demand.

2. The 5G and MEC coupling method for low-latency video services according to claim 1, characterized in that, The step of determining the next frame size prediction result of the target image frame based on the context information of the target image frame corresponding to the video service obtained from the edge server includes: Based on the context information of the target image frame corresponding to the video service obtained from the edge server, a single-frame feature extraction is performed on the target image frame using a regression learning algorithm to obtain the feature vector of the target image frame. The feature vector of the target image frame is combined with the feature vectors of other image frames corresponding to the video service in a temporal sequence to obtain a corresponding temporal feature vector group; the temporal feature vector group is subjected to temporal feature extraction processing to obtain the content change features of the next frame image of the target image frame, and the content change features of the next frame image are used as the size prediction result of the next frame.

3. The 5G and MEC coupling method for low-latency video services according to claim 2, characterized in that, The step of extracting single-frame features from the target image frame based on the context information of the target image frame corresponding to the video service obtained from the edge server, using a regression learning algorithm, to obtain the feature vector of the target image frame includes: The context information of the target image frame corresponding to the video service is obtained locally on the edge server. The context information includes: the size information, content information and semantic information of the target image frame corresponding to the video service sent from the user terminal. The size information of the target image frame is input into a first feature extractor, so that the first feature extractor outputs the size feature vector of the target image frame; The content information of the target image frame is input into the second feature extractor so that the second feature extractor outputs the frame feature vector of the target image frame itself. The second feature extractor includes a pre-trained ResNet-50 model. Furthermore, the semantic information of the target image frame is input into a third feature extractor so that the third feature extractor outputs the number of target objects and the motion state feature vector in the target image frame; The feature vector of the target image frame is obtained based on the size feature vector, the frame's own feature vector, and the number of target objects and the motion state feature vector.

4. The 5G and MEC coupling method for low-latency video services according to claim 1, characterized in that, The step of performing real-time link dynamics estimation on the current state information of the 5G access network obtained by the edge server to determine the radio link quality estimation result of the 5G access network includes: The current status information of the 5G access network is obtained by applying the radio network information service corresponding to the edge server. Based on a preset likelihood estimation method, the current state information of the 5G access network is used to perform real-time link dynamics estimation, thereby obtaining the next SINR change estimate of the 5G access network, and using the next SINR change estimate as the wireless link quality estimation result.

5. A 5G and MEC coupling device for low-latency video services, characterized in that, The 5G and MEC coupling device for low-latency video services is used to perform the 5G and MEC coupling method for low-latency video services as described in any one of claims 1 to 4; The device includes: The cross-layer wireless resource demand customization module is used to determine the next frame size prediction result of the target image frame based on the context information of the target image frame corresponding to the video service obtained by the edge server; to perform real-time link dynamic estimation on the current state information of the 5G access network obtained by the edge server to determine the wireless link quality estimation result of the 5G access network; and to integrate the next frame size prediction result and the wireless link quality estimation result into the wireless resource demand rule. The deadline-sensitive resource granting module is used to control the real-time data interaction between the 5G access network and the edge server according to the wireless resource demand rules based on a preset incentive and acceleration mechanism, so as to adaptively adjust the allocation of wireless resources for the video service.

6. An interaction system between a 5G access network and MEC, characterized in that, include: The user plane and control plane are divided in the software space of the edge server; The user plane is used to implement user plane functions that route or offload traffic to edge applications; The control plane is equipped with a 5G and MEC coupling device as described in claim 5 for low-latency video services.

7. The interaction system between the 5G access network and MEC according to claim 6, characterized in that, Also includes: Each transmission interface is used to enable data interaction between the user plane and the 5G access network inside the interactive system, the user terminal outside the interactive system, and the 5G core network. The northbound interface is used to enable data interaction between the coupling device and the edge application. The southbound interface is used to enable data interaction between the coupling device and the 5G access network.

Images