Network resource scheduling method and device, computer device and storage medium

Abstract

The application provides a network resource scheduling method and device, computer equipment and storage medium, and relates to the technical field of communication, wherein the method comprises the following steps: classifying the service scene where the user is based on the multi-source perception data of the user, obtaining the context label of the user; calculating the current QoE degradation index of the user based on the real-time network performance data of the user; determining the current service level of the user based on the preset service level classification function, the context label of the user and the current QoE degradation index of the user; determining the current resource adjustment parameter of the resource granularity based on the current service level of the user and the preset resource granularity adaptive scheduling strategy; and performing resource allocation based on the resource adjustment parameter. The technical scheme provided by the application can actively adjust the resource strategy according to the current subjective experience quality of the user, and overcomes the defect that there is a lack of dynamic linkage mechanism between the current service level and the user perception.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a network resource scheduling method, a network resource scheduling device, a computer device, and a computer-readable storage medium. Background Technology

[0002] With the development of communication technology, mobile networks face increasingly complex service demands and diverse user experience requirements. Users are no longer satisfied with just network-side metrics such as bandwidth, latency, and packet loss rate for Quality of Service (QoS), but are paying more attention to the Quality of Experience (QoE), that is, the degree of satisfaction users feel from network services. Typical services such as high-definition video playback, AR (Augmented Reality) / VR (Virtual Reality) interaction, and cloud gaming have stronger requirements for real-time performance and perceptual differentiation in QoE.

[0003] To improve network resource utilization efficiency and ensure user experience, operators are increasingly focusing on achieving differentiated service guarantees through intelligent scheduling and policy optimization. However, network resource scheduling methods in related technologies suffer from a technical problem: the lack of a dynamic linkage mechanism between service levels and user perception. Summary of the Invention

[0004] This invention was completed to at least partially solve the technical problem of the lack of a dynamic linkage mechanism between service level and user perception in network resource scheduling methods in related technologies.

[0005] According to one aspect of the present invention, a network resource scheduling method is provided, comprising periodically performing the following steps: Based on the user's multi-source perception data, the user's business scenario is classified to obtain the user's context label. The multi-source perception data represents the user's real-time behavior status and communication activity characteristics, and the context label represents the user's current business scenario. The user's current QoE degradation index is calculated based on the user's real-time network performance data. The QoE degradation index represents the degree of user experience degradation. Based on a preset service level classification function, the user's current service level is determined using the user's context labels and the user's current QoE degradation index. The current resource adjustment parameters for the resource granularity are determined based on the user's current service level and the preset resource granularity adaptive scheduling strategy. Resource allocation is performed based on the aforementioned resource adjustment parameters.

[0006] Optionally, the step of classifying the user's business scenario based on the user's multi-source perception data to derive the user's contextual labels includes: Construct a multi-source perception feature vector based on the user's multi-source perception data; The multi-source perception feature vectors are mapped to user context labels using a trained context label classification model.

[0007] Optionally, the multi-source sensing data includes at least one of the following: foreground application package name, screen touch frequency, user motion state, screen on / off state, TCP connection parameters, uplink / downlink traffic patterns, and time period classification; the context label represents the user's business scenario, including at least one of the following: background business, video playback business, real-time interactive game business, video conferencing business, and browsing business.

[0008] Optionally, the user's current QoE degradation index is calculated using the following formula: ; in, Let be the QoE degradation index of the i-th user at time t; Let be the time delay change of the i-th user at time t, and , Let be the round-trip delay for the i-th user at time t. Let $t$ be the round-trip delay for the $i$-th user at time $t-1$. Let be the packet loss rate of the i-th user at time t; Let be the latency jitter for the i-th user at time t; Let be the throughput of the i-th user at time t; t is the current time. , , , This is the weighting coefficient, which is dynamically set based on the context label of the i-th user.

[0009] Optionally, the service level classification function is: ; in, Let i be the service level of the i-th user at time t; Let be the context label of the i-th user at time t; Let be the QoE degradation index of the i-th user at time t; t is the current time. For logical mapping functions, a rule is used to map the user's contextual labels and experience degradation markers obtained by comparing the QoE degradation index with the QoE degradation threshold to the service level. Where, if If the experience degradation is marked as yes, then the experience degradation is marked as yes. If so, the experience degradation flag will be "No". This is the QoE degradation threshold, and it is dynamically set based on the context label of the i-th user.

[0010] Optionally, after determining the user's current service level, the network resource scheduling method further includes: If the experience degradation flag obtained by comparing the QoE degradation index with the QoE degradation threshold is "yes" for N1 consecutive periods, the user's current service level will be upgraded by one level. If the experience degradation flag obtained by comparing the QoE degradation index with the QoE degradation threshold is negative for N2 consecutive cycles, the user's current service level will be reduced by one level. If the number of changes to a user's context label exceeds a preset threshold within N3 consecutive periods, the response speed for service level adjustments will be reduced. In response to the situation where the user's context label remains unchanged for N4 consecutive periods, the QoE degradation index rises continuously, and the increment in adjacent periods exceeds the preset change threshold, the response speed of service level adjustment is accelerated.

[0011] Optionally, the resource granularity includes at least one of scheduling cycle granularity, frequency domain resource granularity, modulation and coding level granularity, beam resource granularity, and uplink / downlink ratio granularity. The process of determining the current resource adjustment parameters based on the user's current service level and a preset resource granularity adaptive scheduling strategy includes: For scheduling cycle granularity, based on the user's current service level, QoE deviation, and preset reference scheduling cycle, minimum scheduling cycle, and maximum scheduling cycle, the user's scheduling cycle is calculated using a preset cycle scheduling function and used as the current cycle adjustment parameter; and / or, For frequency domain resource granularity, based on the user's current service level, QoE deviation, and network load rate of the cell, a preset frequency domain resource scheduling function is used to calculate the user's frequency domain resource scheduling factor; based on a preset reference resource amount and the frequency domain resource scheduling factor, the user's actual allocated resource amount is obtained as the current frequency domain resource adjustment parameter; and / or, For modulation and coding level granularity, based on the user's current service level, QoE deviation, and the MCS index obtained from the CQI mapping reported by the user and the highest MCS index supported by the base station, the highest MCS index allowed by the user is calculated using a preset modulation and coding level scheduling function as the current modulation and coding level adjustment parameter; and / or, For beam resource granularity, based on the user's current service level, QoE deviation, movement speed, and a preset user movement speed threshold, the user's beam pattern is calculated using a preset beam resource scheduling function as the current beam resource adjustment parameter; and / or, For the uplink / downlink ratio granularity, based on the current service level of each user in the same cell, the estimated uplink / downlink traffic demand, and the preset ratio mode set and the number of uplink / downlink time slots in each ratio mode, the cell-level uplink / downlink time slot ratio is calculated using the preset uplink / downlink ratio scheduling function as the current uplink / downlink ratio adjustment parameter. The QoE deviation is defined as the difference between the user's current QoE degradation index and the QoE target value corresponding to the user's current service level.

[0012] According to another aspect of the present invention, a network resource scheduling apparatus is provided, comprising: The business scenario classification module is configured to classify the business scenarios in which the user is located based on the user's multi-source perception data and obtain the user's context label. The multi-source perception data represents the user's real-time behavior status and communication activity characteristics, and the context label represents the user's current business scenario. The experience degradation calculation module is configured to calculate the user's current QoE degradation index based on the user's real-time network performance data. The QoE degradation index represents the degree of user experience degradation. The service level determination module is configured to determine the user's current service level based on a preset service level classification function, using the user's context labels and the user's current QoE degradation index. The parameter adjustment determination module is configured to determine the current resource adjustment parameters based on the user's current service level and a preset resource granularity adaptive scheduling strategy. The resource allocation module is configured to allocate resources based on the resource adjustment parameters.

[0013] According to another aspect of the present invention, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the aforementioned network resource scheduling method.

[0014] According to another aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor performs the aforementioned network resource scheduling method.

[0015] The technical solution provided by this invention may include the following beneficial effects: The network resource scheduling method and apparatus provided by this invention deeply integrates real-time user behavior perception with network performance perception. It dynamically determines the current service level through a preset service level classification function and maps it to specific resource adjustment parameters. This allows network resource allocation to move beyond passive responses based on static QoS thresholds or simple network load, proactively adjusting resource strategies according to each user's current subjective experience quality. This achieves experience-driven, refined scheduling, effectively solving the technical problem of the lack of a dynamic linkage mechanism between service level and user perception in related network resource scheduling methods. Simultaneously, periodic execution ensures that the scheduling strategy is updated in real time according to user status and network environment, significantly improving the balance between resource utilization efficiency and user satisfaction.

[0016] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description, claims, and drawings. Attached Figure Description

[0017] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0018] Figure 1 A flowchart illustrating a network resource scheduling method provided in an embodiment of the present invention; Figure 2 A flowchart illustrating another network resource scheduling method provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the network resource scheduling device provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the specific implementation methods of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific implementation methods described herein are for illustration and explanation only and are not intended to limit the present invention.

[0020] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a set order or sequence; furthermore, in the absence of conflict, the embodiments and features in the embodiments of this invention can be arbitrarily combined with each other. In the following description, the use of suffixes such as "module," "component," or "unit" to represent elements is only for the convenience of the description of this invention and has no inherent meaning. Therefore, "module," "component," or "unit" can be used interchangeably.

[0021] In related technologies, network resource scheduling methods such as "allocating resources on demand" and "adjusting service levels according to business priorities" are gradually becoming the mainstream trend. However, achieving "maximum resource efficiency" while ensuring QoE awareness still faces significant challenges.

[0022] Typical resource scheduling schemes include the following two categories: 1) Fixed service level matching mechanism based on QoS thresholds: In this network, fixed service levels are configured for different service types, and resources are allocated based on static QoS indicators (such as bandwidth requirements and latency requirements). For example, high-definition video services are allocated high-bandwidth, low-latency QoS guarantees, while web browsing services are allocated lower service levels. This mechanism has the following shortcomings: it cannot respond in real time to fluctuations in user-perceived quality, ignoring the dynamic nature of user experience; static policies are difficult to cope with sudden network congestion or resource shortages, easily leading to resource waste or a decline in experience; and it lacks the ability to perceive personalized differences among users, failing to achieve refined, differentiated guarantees.

[0023] 2) Centralized resource scheduling mechanism based on network load: A centralized resource scheduling mechanism based on network load assesses the current load at the network control plane or edge nodes and dynamically adjusts service scheduling strategies according to preset rules, such as limiting bandwidth for low-priority services when the load is too high. This mechanism has the following shortcomings: most solutions only focus on network-side metrics, and QoE feedback is not involved in the scheduling loop; it is difficult to achieve granular adaptive adjustments (such as flexible scheduling at the subcarrier, time slice, or cell level); and it lacks a dynamic mapping mechanism between service levels and user perception, affecting the accuracy and feasibility of the scheduling strategy.

[0024] To address the aforementioned issues, this invention provides a comprehensive resource scheduling scheme that integrates intelligent user context awareness, QoE feedback modeling, dynamic service level adjustment, and adaptive resource granularity scheduling. This scheme establishes a dynamic mapping mechanism between QoE awareness and network service levels, intelligently adjusts service levels, and achieves a closed-loop optimization between service levels and user perception, enabling network scheduling to balance user satisfaction with resource utilization. Specific embodiments are described in detail below.

[0025] Figure 1 This is a flowchart illustrating a network resource scheduling method provided in an embodiment of the present invention. Figure 1 As shown, the method includes periodically executing the following steps S101 to S105, each period being T, and T can range from 2 to 5 seconds.

[0026] S101. Classify the business scenarios in which the user is located based on the user's multi-source perception data to obtain the user's context label. The multi-source perception data represents the user's real-time behavioral state and communication activity characteristics, and the context label represents the user's current business scenario. S102. Calculate the user's current QoE degradation index based on the user's real-time network performance data, wherein the QoE degradation index represents the degree of user experience degradation; S103. Based on a preset service level classification function, determine the user's current service level using the user's context labels and the user's current QoE degradation index; S104. Determine the current resource adjustment parameters for the resource granularity based on the user's current service level and the preset resource granularity adaptive scheduling strategy; S105. Allocate resources based on the resource adjustment parameters.

[0027] It should be noted that in step S101, the user's multi-source sensing data is collected according to a sliding time window (length T1 seconds), and in step S102, the user's real-time network performance data is also collected according to a sliding time window (length T2 seconds). The period T should be less than the lengths T1 and T2 of the sliding time window, usually set to 1 / 5 to 1 / 2 of the sliding window length, and T1 ≥ T2.

[0028] In this embodiment, real-time user behavior awareness (contextual labels) and network performance awareness (QoE degradation index) are deeply integrated. The current service level is dynamically determined through a preset service level classification function and mapped to specific resource adjustment parameters. This allows network resource allocation to move beyond passive responses based on static QoS thresholds or simple network load, proactively adjusting resource strategies according to each user's current subjective experience quality. This achieves experience-driven, fine-grained scheduling, effectively solving the technical problem of lacking a dynamic linkage mechanism between service level and user perception in related network resource scheduling methods. Simultaneously, periodic execution ensures that the scheduling strategy is updated in real-time according to user status and network environment, significantly improving the balance between resource utilization efficiency and user satisfaction.

[0029] In one specific embodiment, step S101 specifically includes the following steps S1011 and S1012.

[0030] S1011. Construct a multi-source sensing feature vector based on the user's multi-source sensing data; S1012. Using a trained context label classification model, map the multi-source perception feature vector to the user's context label.

[0031] The context label classification model can be the LightGBM model. LightGBM (Light Gradient Boosting Machine) is a machine learning algorithm based on the gradient boosting framework. The final result is obtained by accumulating the outputs of multiple decision trees, resulting in more accurate output.

[0032] In this embodiment, by constructing multi-source sensing feature vectors and mapping context labels using a trained classification model, massive and heterogeneous raw sensing data can be efficiently transformed into scene labels with clear business semantics. This classification method does not require manual definition of complex rules, can automatically adapt to the personalized behavior patterns of different users, and lightweight models (such as LightGBM) can perform online real-time inference, ensuring the low-latency characteristics of the scheduling method. Compared with directly using raw data, this feature vector plus classification model architecture significantly improves the accuracy and robustness of context recognition, providing a reliable prerequisite for subsequent service level adjustments.

[0033] In one specific implementation, the multi-source sensing data includes at least one of the following: foreground application package name, screen touch frequency, user motion state, screen on / off state, TCP (Transmission Control Protocol) connection parameters, uplink / downlink traffic patterns, and time period classification; the context label represents the user's business scenario, including at least one of the following: background business, video playback business, real-time interactive game business, video conferencing business, and browsing business.

[0034] Specifically, the Foreground App Package Name refers to the unique identifier of the application currently running on the screen and being interacted with by the user (i.e., the currently active foreground application); screen touch frequency refers to the number of times the user touches the screen per unit of time; user motion state refers to the user's motion state determined by accelerometer, gyroscope, or base station positioning; screen on / off state refers to whether the screen is on; TCP connection parameters refer to the number of active TCP connections, port numbers, and corresponding application layer protocols; uplink / downlink traffic pattern refers to the long distribution, burstiness, and intermittent characteristics of uplink / downlink traffic data packets; and time period category refers to the time period to which the current time belongs.

[0035] Specifically, background services refer to data transmission tasks that are not sensitive to latency and packet loss and usually run in the backend, including instant messaging message synchronization, cloud backup, and other services; browsing services refer to a type of service whose main purpose is to obtain information and read content, with intermittent user interaction and bursty traffic patterns, including web browsing and document reading services.

[0036] In this embodiment, the multi-source sensing data set takes into account both the terminal side and the network side, reflecting both user subjective behavior (such as touch and motion) and network communication characteristics (such as TCP connections and traffic patterns), thus providing sufficient information for context classification. Simultaneously, the scene category division covers the vast majority of mobile internet services, ensuring the universality of the scheduling strategy across different application types. Furthermore, it allows for flexible addition or removal of data types or scene categories based on actual deployment needs, exhibiting good scalability.

[0037] In one specific implementation, in step S102, the user's current QoE degradation index is calculated using the following formula: ; in, Let be the QoE degradation index of the i-th user at time t; Let be the time delay change of the i-th user at time t, and , Let be the round-trip delay for the i-th user at time t. Let $t$ be the round-trip delay for the $i$-th user at time $t-1$. Let be the packet loss rate of the i-th user at time t; Let be the latency jitter for the i-th user at time t; Let be the throughput of the i-th user at time t; t is the current time. , , , This is the weighting coefficient, which is dynamically set based on the context label of the i-th user.

[0038] In this embodiment, a weighted linear combination formula is used to calculate the QoE degradation index. This formula aggregates four key network performance indicators—latency variation, packet loss rate, jitter, and throughput—into a single value, and dynamically sets weighting coefficients based on context labels. This formula quantifies the degree of degradation in the user's subjective experience, and the weighting coefficients adaptively change with the service scenario (e.g., increasing the weight of latency and jitter in gaming scenarios, and increasing the weight of throughput and packet loss rate in video scenarios). This allows for differentiated representation of the experience impact of the same network degradation phenomenon on different services. Compared to the fixed-weight MOS model, this dynamic weighting method more closely reflects actual user perception, providing a precise quantitative basis for subsequent service level adjustments.

[0039] In one specific implementation, in step S103, the service level classification function is: ; in, Let i be the service level of the i-th user at time t; Let be the context label of the i-th user at time t; Let be the QoE degradation index of the i-th user at time t; t is the current time. For logical mapping functions, a rule is used to map the user's contextual labels and experience degradation markers obtained by comparing the QoE degradation index with the QoE degradation threshold to the service level. Where, if If the experience degradation is marked as yes, then the experience degradation is marked as yes. If so, the experience degradation flag will be "No". This is the QoE degradation threshold, and it is dynamically set based on the context label of the i-th user.

[0040] For example, the QoE degradation threshold for real-time interactive gaming services is lower than that for video playback services, and the QoE degradation threshold for video playback services is lower than that for browsing services.

[0041] In practical applications, service levels can be divided into three categories from high to low: high protection priority (Class-H), medium protection priority (Class-M), and basic protection priority (Class-L). Of course, this invention is not limited to this, and it can also be divided into more or fewer service levels, which can be set by those skilled in the art according to actual needs.

[0042] In this embodiment, a logical mapping function is used to map context labels and experience degradation markers obtained by thresholding the QoE degradation index to service levels. This constructs a ternary decision rule from business scenarios and experience status to service levels, solving the one-sided problem in related technologies that either only consider business type (static level) or only consider network load (no user perception). The mapping rule can be obtained in advance through expert knowledge or machine learning, and it supports independent configuration of the QoE degradation threshold according to different business scenarios. This ensures that the service level classification conforms to business priorities and can respond promptly to experience degradation, forming a dynamic and interconnected core decision-making link.

[0043] In one specific embodiment, after step S103, the network resource scheduling method further includes the following steps S103a to S103d.

[0044] S103a. In response to the experience degradation flag obtained by comparing the QoE degradation index with the QoE degradation threshold for N1 consecutive cycles being "yes", the user's current service level will be upgraded by one level; S103b. If the experience degradation flag obtained by comparing the QoE degradation index with the QoE degradation threshold is negative for N2 consecutive cycles, the user's current service level will be reduced by one level; S103c. In response to the fact that the number of changes in the user's context label exceeds a preset change threshold within N3 consecutive periods, the response speed of service level adjustment is reduced; S103d. In response to the user's context label remaining unchanged for N4 consecutive periods, the QoE degradation index continuously increasing and the increment in adjacent periods exceeding the preset change threshold, the response speed of service level adjustment is accelerated.

[0045] The N1, N2, N3, and N4 mentioned above can be set by those skilled in the art according to actual needs.

[0046] In step S103a, the upper limit for service level promotion is the highest service level (e.g., Class-H).

[0047] In step S103b, the lower limit for reducing the service level is the lowest service level (e.g., Class-L).

[0048] In steps S103c and S103d, speeding up / reducing the response speed of service level adjustment refers to dynamically adjusting the sensitivity of service level change judgment, which is specifically achieved by changing the cycle length or the number of consecutive judgments of experience degradation flag.

[0049] Normal response speed: Service level changes are performed according to the default cycle (e.g., the experience degradation flag is evaluated once every 5 seconds) and the default number of consecutive judgments (e.g., the experience degradation flag needs to be judged as yes 3 times in a row to upgrade the service level).

[0050] Speed ​​up response time: shorten the cycle (e.g., from 5 seconds to 2 seconds) or reduce the number of consecutive checks (e.g., from 3 consecutive checks to 2 consecutive checks) to trigger service level changes faster.

[0051] Reduce response speed: Extend the cycle (e.g., from 5 seconds to 10 seconds) or increase the number of consecutive checks (e.g., from 3 consecutive checks to 5 consecutive checks) to slow down service level changes and avoid frequent service level switching.

[0052] In this embodiment, determining the service level for N1 / N2 consecutive cycles avoids frequent service level switching due to momentary network jitter, ensuring the stability of the scheduling strategy. When the context changes frequently for N3 consecutive cycles, the response speed is reduced to prevent unnecessary level changes caused by users briefly switching applications. When the context is stable for N4 consecutive cycles but the experience continues to deteriorate, the response speed is accelerated to ensure that service levels are quickly upgraded for users who truly need protection. This mechanism achieves a good balance between stability and sensitivity, preventing the "ping-pong effect" while ensuring timely response in emergency scenarios, significantly improving the robustness of the scheduling strategy.

[0053] In one specific implementation, the resource granularity includes at least one of scheduling cycle granularity, frequency domain resource granularity, modulation and coding level granularity, beam resource granularity, and uplink / downlink ratio granularity.

[0054] Accordingly, for the scheduling cycle granularity, step S104 includes: Based on the user's current service level, QoE deviation, and preset reference scheduling period, minimum scheduling period, and maximum scheduling period, the user's scheduling period is calculated using a preset periodic scheduling function and used as the current periodic adjustment parameter.

[0055] The QoE deviation is defined as the difference between the user's current QoE degradation index and the QoE target value corresponding to the user's current service level.

[0056] Specifically, the periodic scheduling function can be: .

[0057] in, The scheduling period for the i-th user; The minimum preset scheduling period (i.e., the minimum scheduling period, such as 1 frame, to ensure the low latency requirements of high-priority services). The preset maximum scheduling period (i.e., the maximum scheduling period, such as 8 frames, to control the resource overhead of low-priority services). The preset reference scheduling period (i.e., the baseline value, such as 4 frames). This is the rating value for the current service level of the i-th user. For example, the rating value for Class-H is 3, the rating value for Class-M is 2, and the rating value for Class-L is 1. , Let be the QoE degradation index of the i-th user at time t. Let be the QoE target value corresponding to the current service level of the i-th user, representing the standard QoE value that the corresponding service level should achieve. To ensure the contribution of this parameter, QoE deviation, to other parameters (…), , The contributions of the two are on the same order of magnitude, and the difference between the QoE degradation index and the QoE target value can be normalized and used as the QoE deviation; k1 and k2 are weighting coefficients.

[0058] It should be noted that scheduling periods are typically measured in frames (10ms), subframes (1ms), or slots (usually 0.5ms or less). This invention uses "frames" as the unit of measurement for scheduling periods. A reference scheduling period of 4 frames indicates that, by default, the system allocates resources to a user every 4 radio frames (i.e., 40ms). Of course, the reference scheduling period can be shortened or extended as needed by the system.

[0059] For frequency domain resource granularity, step S104 includes: Based on the user's current service level, QoE deviation, and network load rate of the cell, the user's frequency domain resource scheduling factor is calculated using a preset frequency domain resource scheduling function. Based on the preset reference resource amount and the frequency domain resource scheduling factor, the user's actual allocated resource amount is obtained as the current frequency domain resource adjustment parameter.

[0060] Specifically, the frequency domain resource scheduling function can be: .

[0061] in, Let be the frequency domain resource scheduling factor for the i-th user; This is the rating value for the current service level of the i-th user. For example, the rating value for Class-H is 3, the rating value for Class-M is 2, and the rating value for Class-L is 1. Let be the QoE deviation of the i-th user, and , Let be the QoE degradation index of the i-th user at time t. Let be the QoE target value corresponding to the current service level of the i-th user, representing the standard QoE value that the corresponding service level should achieve. This parameter contributes to ensuring the QoE deviation from other parameters. , The contributions of the two are on the same order of magnitude, and the difference between the QoE degradation index and the QoE target value can be normalized and used as the QoE deviation. Let be the network load rate of the cell to which the i-th user belongs, representing the resource consumption of the corresponding cell, such as PRB utilization rate, CPU utilization rate, etc. , , These are the weighting coefficients.

[0062] For modulation coding level granularity, step S104 includes: Based on the user's current service level, QoE deviation, and the MCS index (Modulation and Coding Scheme) obtained by mapping the CQI (Channel Quality Indicator) reported by the user, as well as the highest MCS index supported by the base station, the highest MCS index allowed by the user is calculated using a preset modulation and coding level scheduling function as the current modulation and coding level adjustment parameter.

[0063] Specifically, the modulation and coding level scheduling function can be: .

[0064] in, The highest MCS index allowed for the i-th user; The highest MCS index supported by the base station (e.g., 27); The MCS index is obtained from the CQI mapping reported by the i-th user; This is the rating value for the current service level of the i-th user. For example, the rating value for Class-H is 3, the rating value for Class-M is 2, and the rating value for Class-L is 1. Let be the QoE deviation of the i-th user, and , Let be the QoE degradation index of the i-th user at time t. Let be the QoE target value corresponding to the current service level of the i-th user, representing the standard QoE value that the corresponding service level should achieve. This parameter contributes to ensuring the QoE deviation from other parameters. The contributions of the two are on the same order of magnitude, and the difference between the QoE degradation index and the QoE target value can be normalized and used as the QoE deviation; round(·) is the rounding function; , These are the weighting coefficients.

[0065] It should be noted that CQI is the integer (usually 0-15) reported by the UE to the base station after measuring the downlink channel quality, while MCS is the modulation order and code rate selected by the base station based on the CQI. The 5G NR standard (3GPP TS 38.214) defines a mapping table from CQI to MCS. The mapping rule is: each CQI index corresponds to a specific modulation order, code rate, and efficiency, and the base station selects the appropriate MCS index accordingly. In actual implementation, a predefined mapping table is usually used (for example, CQI=12 may correspond to 64QAM and a code rate of 0.5, with a corresponding MCS index of 18). Therefore, The CQI reported by user i can be retrieved through a preset CQI-MCS mapping table.

[0066] For beam resource granularity, step S104 includes: Based on the user's current service level, QoE deviation, movement speed, and preset user movement speed threshold, the user's beam pattern is calculated using a preset beam resource scheduling function as the current beam resource adjustment parameter.

[0067] Specifically, the beam resource scheduling function can be: .

[0068] in, For the i-th user, the beam pattern is (2 = multi-beam joint transmission, 1 = user-level dedicated beam, 0 = cell-level broadcast beam). This is the rating value for the current service level of the i-th user. For example, the rating value for Class-H is 3, the rating value for Class-M is 2, and the rating value for Class-L is 1. The rating for the highest service level; The rating value is for an intermediate service level that falls between the highest and lowest service levels. Let be the moving speed (km / h) of the i-th user. The preset low-speed threshold; This is the preset medium speed threshold; Let be the QoE deviation of the i-th user, and , Let be the QoE degradation index of the i-th user at time t. Let be the QoE target value corresponding to the current service level of the i-th user, representing the QoE standard value that the corresponding service level should achieve. Specifically, the difference between the QoE degradation index and the QoE target value can be normalized and then used as the QoE deviation. The threshold for significant deviation of QoE (e.g., 0.5).

[0069] For the uplink / downlink ratio granularity, step S104 includes: Based on the current service level of each user in the same cell, the estimated uplink and downlink traffic demand, and the preset allocation mode set and the number of uplink and downlink time slots in each allocation mode, the cell-level uplink and downlink time slot allocation is calculated using the preset uplink and downlink ratio scheduling function as the current uplink and downlink ratio adjustment parameter.

[0070] Specifically, for a TDD (Time Division Duplex) system, the uplink / downlink ratio scheduling function can be: .

[0071] in, The uplink / downlink time slot allocation ratio at the cell level (e.g., 2:3, 3:2, 4:1) corresponds to the cell to be adjusted; R This refers to a set of preset proportioning patterns, or a set of optional proportioning patterns. For the i-th user, the estimated downlink traffic demand (bps); For the estimated uplink traffic demand (bps) of the i-th user; Let ∑ be the rating value of the current service level of the i-th user, for example, the rating value of Class-H is 3, the rating value of Class-M is 2, and the rating value of Class-L is 1; i (·) is the formula for summing over all users in the same community; Under the matching mode r, the number of downlink time slots within one matching cycle; For the ratio mode r, the number of uplink time slots within one ratio cycle; This indicates a set of preset proportioning patterns. R In the expression, find a pattern r such that the absolute value within the parentheses is minimized.

[0072] In this embodiment, adaptive adjustment parameter calculation methods based on service level and QoE deviation are designed for five different resource granularities: scheduling period, frequency domain resources, modulation and coding, beamforming, and uplink / downlink ratio. Each resource granularity finely adjusts resource allocation from different perspectives: shortening or extending the scheduling period in the time domain, allocating more or fewer PRBs in the frequency domain, raising or lowering the MCS upper limit in the modulation domain, switching beamforming modes in the spatial domain, and dynamically adjusting the uplink / downlink time slot ratio in a TDD system. These strategies can be used independently or in combination, achieving comprehensive resource collaborative optimization across five dimensions: time, frequency, modulation and coding, space, and frame structure. Compared to single-dimensional scheduling methods, this multi-granularity joint adaptive mechanism can more fully tap the potential of wireless resources, prioritizing the experience of high-level users in high-load scenarios and avoiding resource waste in low-load scenarios, ultimately achieving dual optimization of resource efficiency and user perception.

[0073] The network resource scheduling method provided in this invention constructs a complete closed loop from user subjective experience to physical resource scheduling through periodically executed context awareness, QoE degradation index calculation, dynamic service level determination, multi-granularity resource parameter generation, and resource allocation. This method deeply integrates business scenario identification and experience degradation monitoring, using a service level classification function to achieve joint decision-making on "what service" and "how good the experience is," overcoming the fundamental deficiency of related technologies that lack a dynamic linkage mechanism between service level and user perception. Simultaneously, it achieves a good balance between stability and sensitivity through upgrade / downgrade buffering and adaptive adjustment of response speed. Furthermore, it designs adaptive strategies based on service level and QoE deviation in five dimensions: time domain (scheduling cycle), frequency domain (PRB quantity), modulation domain (MCS level), spatial domain (beam pattern), and frame structure domain (uplink / downlink ratio), realizing coordinated and refined scheduling of multi-dimensional resources. Overall, this method ensures rapid response and resource allocation when user experience deteriorates, while avoiding unnecessary scheduling overhead and ping-pong effects, significantly improving the balance between network resource utilization efficiency and user satisfaction.

[0074] Figure 2 This is a flowchart illustrating another network resource scheduling method provided in an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S201 to S204.

[0075] S201. User context intelligent awareness.

[0076] The purpose of this step is to enable the network to understand each user's current usage scenario (e.g., whether they are playing games, watching videos, or downloading in the background) in order to allocate resources on demand, rather than treating everyone the same or allocating resources statically. However, in related technologies, the network layer only sees IP packets and QCI levels, and cannot see what the user is actually doing. Therefore, the goal of this step is to perform real-time contextual label classification for each user through multi-source feature extraction and AI intelligent classification.

[0077] The input information for this step includes: sensor data reported by the user terminal, including screen operation frequency, APP foreground status, screen status, motion status, etc., as well as data collected by the base station / core network side, including TCP connection information, traffic patterns, etc.

[0078] The features shown in Table 1 below were extracted from the acquired data: Table 1

[0079] Based on the extracted features, the following feature vector is constructed: X i (t)=[App i (t), Touchi (t), Motion i (t),Screen i (t), TCP i (t), Packet i (t), Time i (t), …].

[0080] Using a trained lightweight classifier, such as LightGBM, map the feature vector Xi(t) to context labels: .

[0081] Where X i (t) is the context feature vector of the i-th user at time t, C i f(t) is the predicted context label (numerical or vector) for the i-th user at time t, where t is the current time. θ (·) represents a trained classification model that maps user context feature vectors to context labels. The feature vectors are obtained by sliding a time window (T1 seconds in length) and used as input to the classification model.

[0082] Context label settings (example) are as follows: C i (t)∈{Background, Video, Gaming, VoIP, Browsing}.

[0083] The description of each context tag is shown in Table 2 below: Table 2

[0084] Finally, a context label (corresponding to a business scenario) is identified for each user, and the corresponding structured output result (JSON format) is shown in the following example: { "user_id":"UE_0234567", "time":"2025-07-21 15:05:00", "context":"Gaming", "confidence": 0.92 } This output will be used as input for the next step. The obtained context label, such as context=Gaming, will be used as the business scenario identification result.

[0085] S202. QoE back feedback modeling.

[0086] Currently, cellular network scheduling relies solely on low-level protocol metrics (such as SINR, CQI, and packet loss rate). These metrics are disconnected from the user's subjective experience (QoE) and cannot accurately reflect situations such as video stuttering despite good channel performance, game latency fluctuations despite normal RSRP, and frequent packet loss during conferences despite no decrease in CQI.

[0087] Therefore, the purpose of this step is to aggregate multiple physical / protocol metrics from the network layer and transport layer into a dynamic "user experience degradation index" to assist in subsequent service level adjustments and resource allocation optimization.

[0088] Real-time network performance metrics for each user UE are obtained from the network and the terminal using a sliding time window (window length T2 seconds), as shown in Table 3 below: Table 3

[0089] Define the QoE degradation index: .

[0090] in, Let be the QoE degradation index of the i-th user at time t. The larger the value, the worse the user experience or the worsening it is. Let be the time delay change of the i-th user at time t, and , Let be the round-trip delay for the i-th user at time t. Let $t$ be the round-trip delay for the $i$-th user at time $t-1$. Let be the packet loss rate of the i-th user at time t; Let be the latency jitter for the i-th user at time t; Let be the throughput of the i-th user at time t; t is the current time. , , , The weighting coefficients for each indicator (set through historical training or business scenarios) can be dynamically selected based on user context tags.

[0091] For example, if context = Gaming, it is more sensitive to latency and jitter → improves w1 and w3; if context = Video, it is more sensitive to throughput and packet loss → improves w2 and w4.

[0092] The relationship between the various parameters / indicators involved in the QoE degradation index and QoE is shown in Table 4 below: Table 4

[0093] If the Q of the i-th user i(t) exceeds a certain threshold Q th (QoE degradation threshold) indicates that the user experience is deteriorating. Different context labels can use different QoE thresholds. th Thresholds are used as the judgment criteria.

[0094] Finally, a QoE degradation index and experience degradation flag are calculated for each user. An example of the corresponding structured output (JSON format) is shown below: { "user_id":"UE_0234567", "time":"2025-07-21 15:10:00", "Qi(t)":0.87, "degraded":true, "delta_RTT":45, "PLR":2.3, "Tput": 1.5 } S203. Service level is dynamically adjusted.

[0095] Currently, the service levels of cellular networks (such as GBR / Non-GBR, QCI 1~9) are static and cannot respond to changes in the user's actual experience. For example, video may stutter but the QCI level remains unchanged; game latency may increase but it will still be scheduled with low priority.

[0096] The goal of this step is to dynamically adjust the user's service level by analyzing the user's current business scenario (step S201) and the degree of experience degradation (step S202) in real time, and thereby affect the priority and granularity of the scheduler in allocating network resources.

[0097] The input for this step comes from the results of the previous two steps: The result of step S201: For example, context: "Gaming" is used to indicate what the user's current business scenario is (game, video, webpage, etc.). The result of step S202: Qi(t): 0.87, degraded: true, is used to indicate whether the user's current experience has deteriorated (e.g., lag, latency, packet loss, etc.).

[0098] This embodiment defines three dynamic service levels (Class-H / M / L) to guide the scheduler's resource allocation. The descriptions of these three service levels are shown in Table 5 below: Table 5

[0099] Service level adjustment judgment logic: Construct a service level classification function: .

[0100] in, It is the context label output in step S201; It is the QoE degradation index calculated in step S202; g(·) is a logistic mapping function (which can be learned through the model).

[0101] An example of the context label and QoE degradation index linkage mapping rule table (g function table) is shown in Table 6 below: Table 6

[0102] To avoid frequent fluctuations, the following jitter buffering mechanism can also be configured: If the QoE is determined to be degraded for N1 consecutive QoE assessment cycles (e.g., N1=3), the current service level will be upgraded by one level until it reaches Class-H. If the QoE is determined to be non-degraded for N2 consecutive QoE assessment cycles (e.g., N2=5), the current service level will be downgraded by one level until it reaches Class-L.

[0103] And combine it with context stability strategies: If business scenarios change frequently (such as switching apps) → postpone service level adjustments; If the business scenario remains stable but the user experience deteriorates sharply, accelerate the response speed for service level adjustments.

[0104] Finally, a service level is dynamically assigned to each user, along with a reason for the assignment. An example of the structured output (JSON format) is shown below: { "user_id":"UE_0234567", "time":"2025-07-21 15:15:00", "assigned_class":"Class-H", "reason":"Gaming+QoE degraded" } This output will directly serve as input for the next step: resource-granular adaptive scheduling, guiding actual operations such as wireless resource block allocation and scheduling priority selection.

[0105] Therefore, step S201 determines whether it is a sensitive service, step S202 determines whether the user experience is poor, and based on these two judgment results, it is decided whether the user should optimize the scheduling.

[0106] S204. Resource-granularity adaptive scheduling.

[0107] The goal of this step is to dynamically adjust the resource allocation of the access network at the granular level based on the Service Level (SL) dynamically adjusted in step S203, ensuring that resource allocation is both accurate and efficient, avoiding resource waste, and meeting user QoE requirements. Specifically, based on the already refined and layered Service Level, this step further optimizes the resource units corresponding to each user (such as bandwidth, PRB, MCS, time slots, scheduling cycles, etc.), making resource scheduling both granular and dynamically variable.

[0108] In traditional scheduling mechanisms, resource allocation granularity is fixed, such as allocating a fixed number of PRBs, bandwidth, or scheduling cycles to each user. This can lead to resource waste (too coarse-grained) and poor user experience (too fine-grained, failing to meet the needs of high-priority users) when traffic fluctuates dynamically. In this step, however, resource granularity is adaptively adjusted and dynamically changes based on service level.

[0109] Input information for this step: Each user's current service level is determined from step S203; The target QoE value for each service level can be set by the operator according to their own policies. Current network status of the user (RSRP, SINR, CQI, historical throughput, etc.); Current system resource status (total PRB utilization, etc.).

[0110] Defining "resource granularity" can include: Scheduling cycle granularity: Whether to schedule once every 1 frame / 2 frames / 4 frames; Frequency domain resource granularity (PRB allocation unit): the minimum number of PRBs allocated each time; Modulation coding level granularity (MCS): Whether a higher-order MCS can be used; Beam or antenna resource granularity: Whether the multi-beam scheduling capability is refined to the cell, sub-cell, or user level; Uplink / downlink ratio granularity: Adaptively adjust the uplink / downlink resource ratio according to the business scenario (especially for TDD systems).

[0111] Table 7 below shows examples of adaptive scheduling strategies for resource granularity based on service level: Table 7

[0112] The following example uses frequency domain resource granularity to design a frequency domain resource scheduling factor γ. i (This represents the frequency domain resource granularity adjustment coefficient), and its calculation expression is as follows, used to guide the dynamic allocation of resources: .

[0113] Where, γ i The frequency domain resource scheduling factor for the i-th user represents a dynamic adjustment value for the priority or weight of the frequency domain resources required by the i-th user. It determines whether the system should prioritize the service experience of the user when network resources are limited. The higher the value, the more the user should be prioritized for QoE or resource enhancement. SL i (Service Level): Indicates the current service level of the i-th user, such as high, medium, or low; ΔQoE i : QoE deviation of the i-th user, ΔQoE i = Q i (t)-Q target,i Q i (t) represents the current QoE degradation index of the i-th user, Q i-target Let be the QoE target value corresponding to the current service level of the i-th user. It measures the deviation of the current actual experience quality (such as download speed, latency, and stuttering rate) from the QoE standard value that should be achieved for this service level. The larger the value, the worse the user experience, and the more resources the system should provide. ρ i : Network load rate of the cell to which the i-th user currently belongs, representing the resource usage of the cell (such as PRB utilization, CPU utilization, etc.).

[0114] When the load is high, even if the user experience deviates significantly, resources must be carefully balanced across multiple users according to service levels.

[0115] The function f(·) is not a fixed function, but a dynamic decision model or weighted function, combining weighted scores or rule-based reasoning from three input parameters. It can be formalized as a linear combination (for example), or it can be a complex nonlinear function implemented using a neural network or decision tree, through mapping: .

[0116] Among them, w(SL) i ): The rating value of the current service level of the i-th user (e.g., high priority rating value = 3, medium priority rating value = 2, low priority rating value = 1); ΔQoE i : QoE deviation of the i-th user; ρi: Network load rate of the cell to which the i-th user currently belongs, where 1 represents 100% load; α, β, θ: These are the weighting coefficients configured in the system.

[0117] The final amount of resources allocated to the i-th user is: .

[0118] This formula represents the amount of resources allocated to the i-th user, which is equal to a basic resource allocation R. base Multiply by the user's frequency domain resource scheduling factor γ i This formula is the core computational logic of adaptive frequency domain resource scheduling. It flexibly adjusts the frequency domain resource allocation strategy based on the user's dynamic needs and system feedback, ensuring a balance between maximizing user experience and maximizing system resource utilization.

[0119] in, : Represents the amount of resources actually allocated to the i-th user (such as bandwidth, number of time slots, etc.); This represents the default or minimum basic resource unit that each user in the system can obtain. It is a fixed reference value that reflects the minimum service guarantee of the system. γ i (Gamma Factor): Represents the resource scheduling factor for the i-th user, typically with a value ≥ 0. It comprehensively reflects: the QoE degradation index and its deviation from the target value in step S202 (QoE backfeedback modeling); and the service level determined in step S203 (service level dynamic adjustment).

[0120] Here is a specific example: Assume the system is defined as follows: R base =100 kbps; User A's QoE feedback is good, γ A =1.2; User B's QoE feedback was poor, γ B =0.6.

[0121] Therefore, the resource allocation result is: =100×1.2=120 kbps (more resources are allocated preferentially). =100×0.6=60 kbps (reducing resources).

[0122] Final output: a resource scheduling granularity configuration table for each user, and control parameters for the base station scheduler (e.g., PRB allocation policy table, MCS range, scheduling period length, etc.).

[0123] The network resource scheduling method provided in this embodiment of the invention is specifically a scheduling strategy that integrates AI user context awareness, QoE dynamic feedback loop, and resource granularity reconstruction mechanism. It dynamically adjusts the service level of users by modeling the real-time context of users and combining it with network QoS feedback. It also predicts and reconstructs the granularity of resource blocks and allocation time slots based on resource utilization efficiency, thereby achieving fine-grained and highly matched intelligent scheduling.

[0124] Figure 3 This is a schematic diagram of the network resource scheduling device provided in an embodiment of the present invention. Figure 3 As shown, the network resource scheduling device includes: a service scenario classification module 301, an experience degradation calculation module 302, a service level determination module 303, an adjustment parameter determination module 304, and a resource allocation module 305.

[0125] The service scenario classification module 301 is configured to classify the user's service scenario based on the user's multi-source perception data to obtain the user's context label. The multi-source perception data represents the user's real-time behavior status and communication activity characteristics, and the context label represents the user's current service scenario. The experience degradation calculation module 302 is configured to calculate the user's current QoE degradation index based on the user's real-time network performance data. The QoE degradation index represents the degree of user experience degradation. The service level determination module 303 is configured to determine the user's current service level based on a preset service level classification function, using the user's context label and the user's current QoE degradation index. The adjustment parameter determination module 304 is configured to determine the current resource adjustment parameters based on the user's current service level and a preset resource granularity adaptive scheduling strategy. The resource allocation module 305 is configured to allocate resources based on the resource adjustment parameters.

[0126] Moreover, the corresponding operations of the business scenario classification module 301, experience degradation calculation module 302, service level determination module 303, adjustment parameter determination module 304, and resource allocation module 305 are executed periodically.

[0127] In one specific implementation, the business scenario classification module 301 specifically includes a feature vector construction unit and a feature vector mapping unit.

[0128] The feature vector construction unit is configured to construct a multi-source perception feature vector based on the user's multi-source perception data; the feature vector mapping unit is configured to use a trained context label classification model to map the multi-source perception feature vector to the user's context label.

[0129] In one specific implementation, the multi-source sensing data includes at least one of the following: foreground application package name, screen touch frequency, user motion state, screen on / off state, TCP connection parameters, uplink / downlink traffic patterns, and time period classification; the context label represents the user's business scenario, including at least one of the following: background business, video playback business, real-time interactive game business, video conferencing business, and browsing business.

[0130] In one specific implementation, the experience degradation calculation module 302 calculates the user's current QoE degradation index using the following formula: ; in, Let be the QoE degradation index of the i-th user at time t; Let be the time delay change of the i-th user at time t, and , Let be the round-trip delay for the i-th user at time t. Let $t$ be the round-trip delay for the $i$-th user at time $t-1$. Let be the packet loss rate of the i-th user at time t; Let be the latency jitter for the i-th user at time t; Let be the throughput of the i-th user at time t; t is the current time. , , , This is the weighting coefficient, which is dynamically set based on the context label of the i-th user.

[0131] In one specific implementation, the service level determination module 303 uses the following service level classification function: ; in, Let i be the service level of the i-th user at time t; Let be the context label of the i-th user at time t; Let be the QoE degradation index of the i-th user at time t; t is the current time. For logical mapping functions, a rule is used to map the user's contextual labels and experience degradation markers obtained by comparing the QoE degradation index with the QoE degradation threshold to the service level. Where, if If the experience degradation is marked as yes, then the experience degradation is marked as yes. If so, the experience degradation flag will be "No". This is the QoE degradation threshold, and it is dynamically set based on the context label of the i-th user.

[0132] In one specific embodiment, the network resource scheduling device further includes a service level adjustment module.

[0133] The service level adjustment module is set as follows: If the experience degradation flag obtained by comparing the QoE degradation index with the QoE degradation threshold is "yes" for N1 consecutive periods, the user's current service level will be upgraded by one level. If the experience degradation flag obtained by comparing the QoE degradation index with the QoE degradation threshold is negative for N2 consecutive cycles, the user's current service level will be reduced by one level. If the number of changes to a user's context label exceeds a preset threshold within N3 consecutive periods, the response speed for service level adjustments will be reduced. In response to the situation where the user's context label remains unchanged for N4 consecutive periods, the QoE degradation index rises continuously, and the increment in adjacent periods exceeds the preset change threshold, the response speed of service level adjustment is accelerated.

[0134] In one specific implementation, the resource granularity includes at least one of scheduling cycle granularity, frequency domain resource granularity, modulation and coding level granularity, beam resource granularity, and uplink / downlink ratio granularity.

[0135] The specific settings for parameter adjustment module 304 are as follows: For scheduling cycle granularity, based on the user's current service level, QoE deviation, and preset reference scheduling cycle, minimum scheduling cycle, and maximum scheduling cycle, the user's scheduling cycle is calculated using a preset cycle scheduling function as the current cycle adjustment parameter. For frequency domain resource granularity, based on the user's current service level, QoE deviation, and network load rate of the cell, the user's frequency domain resource scheduling factor is calculated using a preset frequency domain resource scheduling function; based on the preset reference resource amount and the frequency domain resource scheduling factor, the user's actual allocated resource amount is obtained as the current frequency domain resource adjustment parameter. For modulation and coding level granularity, based on the user's current service level, QoE deviation, and the MCS index obtained from the CQI mapping reported by the user and the highest MCS index supported by the base station, the highest MCS index allowed by the user is calculated using the preset modulation and coding level scheduling function as the current modulation and coding level adjustment parameter. For beam resource granularity, based on the user's current service level, QoE deviation, movement speed, and preset user movement speed threshold, the user's beam pattern is calculated using a preset beam resource scheduling function as the current beam resource adjustment parameter. For the uplink / downlink ratio granularity, based on the current service level of each user in the same cell, the estimated uplink / downlink traffic demand, and the preset ratio mode set and the number of uplink / downlink time slots in each ratio mode, the cell-level uplink / downlink time slot ratio is calculated using the preset uplink / downlink ratio scheduling function as the current uplink / downlink ratio adjustment parameter. The QoE deviation is defined as the difference between the user's current QoE degradation index and the QoE target value corresponding to the user's current service level.

[0136] The network resource scheduling device provided in this invention constructs a complete closed loop from user subjective experience to physical resource scheduling through periodically executed context awareness, QoE degradation index calculation, dynamic service level determination, multi-granularity resource parameter generation, and resource allocation. This method deeply integrates business scenario identification and experience degradation monitoring, using a service level classification function to achieve joint decision-making on "what service" and "how good the experience is," overcoming the fundamental deficiency of related technologies that lack a dynamic linkage mechanism between service level and user perception. Simultaneously, it achieves a good balance between stability and sensitivity through upgrade / downgrade buffering and adaptive adjustment of response speed. Furthermore, it designs adaptive strategies based on service level and QoE deviation from five dimensions: time domain (scheduling cycle), frequency domain (PRB quantity), modulation domain (MCS level), spatial domain (beam pattern), and frame structure domain (uplink / downlink ratio), realizing coordinated and refined scheduling of multi-dimensional resources. Overall, this method ensures rapid response and resource allocation when user experience deteriorates, while avoiding unnecessary scheduling overhead and ping-pong effects, significantly improving the balance between network resource utilization efficiency and user satisfaction.

[0137] Based on the same technical concept, embodiments of the present invention also provide a computer device, such as... Figure 4 As shown, the computer device includes a memory 401 and a processor 402. The memory 401 stores a computer program. When the processor 402 runs the computer program stored in the memory 401, the processor 402 executes the aforementioned network resource scheduling method.

[0138] Based on the same technical concept, embodiments of the present invention also provide a computer-readable storage medium storing a computer program thereon, wherein when the computer program is executed by a processor, the processor executes the aforementioned network resource scheduling method.

[0139] In summary, the network resource scheduling method, apparatus, computer equipment, and storage medium provided in this embodiment of the invention achieve real-time quantification and precise response to user experience quality by constructing a dynamic closed-loop linkage mechanism between user perception and network resources. It can proactively and differentiatedly adjust resource allocation strategies according to business scenarios and experience status, thereby significantly improving the overall network resource utilization efficiency, scheduling stability, and user satisfaction while ensuring the service quality of high-priority services and users with degraded experiences. This effectively overcomes the problem of the disconnect between service level and subjective perception in traditional scheduling methods.

[0140] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A network resource scheduling method, characterized in that, This includes periodically performing the following steps: Based on the user's multi-source perception data, the user's business scenario is classified to obtain the user's context label. The multi-source perception data represents the user's real-time behavior status and communication activity characteristics, and the context label represents the user's current business scenario. The user's current QoE degradation index is calculated based on the user's real-time network performance data. The QoE degradation index represents the degree of user experience degradation. Based on a preset service level classification function, the user's current service level is determined using the user's context labels and the user's current QoE degradation index. The current resource adjustment parameters for the resource granularity are determined based on the user's current service level and the preset resource granularity adaptive scheduling strategy. Resource allocation is performed based on the aforementioned resource adjustment parameters.

2. The network resource scheduling method according to claim 1, characterized in that, The process of classifying the user's business scenario based on multi-source perception data to derive the user's contextual tags includes: Construct a multi-source perception feature vector based on the user's multi-source perception data; The multi-source perception feature vectors are mapped to user context labels using a trained context label classification model.

3. The network resource scheduling method according to claim 1 or 2, characterized in that, The multi-source sensing data includes at least one of the following: foreground application package name, screen touch frequency, user motion state, screen on / off state, TCP connection parameters, uplink / downlink traffic patterns, and time period classification; the context label represents the user's business scenario, including at least one of the following: background business, video playback business, real-time interactive game business, video conferencing business, and browsing business.

4. The network resource scheduling method according to claim 1, characterized in that, The user's current QoE degradation index is calculated using the following formula: ； in, Let be the QoE degradation index of the i-th user at time t; Let be the time delay change of the i-th user at time t, and , Let be the round-trip delay for the i-th user at time t. Let $t$ be the round-trip delay for the $i$-th user at time $t-1$. Let be the packet loss rate of the i-th user at time t; Let be the latency jitter for the i-th user at time t; Let be the throughput of the i-th user at time t; t is the current time. , , , This is the weighting coefficient, which is dynamically set based on the context label of the i-th user.

5. The network resource scheduling method according to claim 1, characterized in that, The service level classification function is: ； in, Let i be the service level of the i-th user at time t; Let be the context label of the i-th user at time t; Let be the QoE degradation index of the i-th user at time t; t is the current time. For logical mapping functions, a rule is used to map the user's contextual labels and experience degradation markers obtained by comparing the QoE degradation index with the QoE degradation threshold to the service level. Where, if If the experience degradation is marked as yes, then the experience degradation is marked as yes. If so, the experience degradation flag will be "No". This is the QoE degradation threshold, and it is dynamically set based on the context label of the i-th user.

6. The network resource scheduling method according to claim 5, characterized in that, After determining the user's current service level, the network resource scheduling method further includes: If the experience degradation flag obtained by comparing the QoE degradation index with the QoE degradation threshold is "yes" for N1 consecutive periods, the user's current service level will be upgraded by one level. If the experience degradation flag obtained by comparing the QoE degradation index with the QoE degradation threshold is negative for N2 consecutive cycles, the user's current service level will be reduced by one level. If the number of changes to a user's context label exceeds a preset threshold within N3 consecutive periods, the response speed for service level adjustments will be reduced. In response to the situation where the user's context label remains unchanged for N4 consecutive periods, the QoE degradation index rises continuously, and the increment in adjacent periods exceeds the preset change threshold, the response speed of service level adjustment is accelerated.

7. The network resource scheduling method according to claim 1, characterized in that, The resource granularity includes at least one of scheduling cycle granularity, frequency domain resource granularity, modulation and coding level granularity, beam resource granularity, and uplink / downlink ratio granularity. The process of determining the current resource adjustment parameters based on the user's current service level and a preset resource granularity adaptive scheduling strategy includes: For scheduling cycle granularity, based on the user's current service level, QoE deviation, and preset reference scheduling cycle, minimum scheduling cycle, and maximum scheduling cycle, the user's scheduling cycle is calculated using a preset cycle scheduling function and used as the current cycle adjustment parameter; and / or, For frequency domain resource granularity, based on the user's current service level, QoE deviation, and network load rate of the cell, a preset frequency domain resource scheduling function is used to calculate the user's frequency domain resource scheduling factor; based on a preset reference resource amount and the frequency domain resource scheduling factor, the user's actual allocated resource amount is obtained as the current frequency domain resource adjustment parameter; and / or, For modulation and coding level granularity, based on the user's current service level, QoE deviation, and the MCS index obtained from the CQI mapping reported by the user and the highest MCS index supported by the base station, the highest MCS index allowed by the user is calculated using a preset modulation and coding level scheduling function as the current modulation and coding level adjustment parameter; and / or, For beam resource granularity, based on the user's current service level, QoE deviation, movement speed, and a preset user movement speed threshold, the user's beam pattern is calculated using a preset beam resource scheduling function as the current beam resource adjustment parameter; and / or, For the uplink / downlink ratio granularity, based on the current service level of each user in the same cell, the estimated uplink / downlink traffic demand, and the preset ratio mode set and the number of uplink / downlink time slots in each ratio mode, the cell-level uplink / downlink time slot ratio is calculated using the preset uplink / downlink ratio scheduling function as the current uplink / downlink ratio adjustment parameter. The QoE deviation is defined as the difference between the user's current QoE degradation index and the QoE target value corresponding to the user's current service level.

8. A network resource scheduling device, characterized in that, include: The business scenario classification module is configured to classify the business scenarios in which the user is located based on the user's multi-source perception data and obtain the user's context label. The multi-source perception data represents the user's real-time behavior status and communication activity characteristics, and the context label represents the user's current business scenario. The experience degradation calculation module is configured to calculate the user's current QoE degradation index based on the user's real-time network performance data. The QoE degradation index represents the degree of user experience degradation. The service level determination module is configured to determine the user's current service level based on a preset service level classification function, using the user's context labels and the user's current QoE degradation index. The parameter adjustment determination module is configured to determine the current resource adjustment parameters based on the user's current service level and a preset resource granularity adaptive scheduling strategy. The resource allocation module is configured to allocate resources based on the resource adjustment parameters.

9. A computer device, characterized in that, The system includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the network resource scheduling method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, the processor performs the network resource scheduling method according to any one of claims 1 to 7.

Images