Target wake-up time scheduling method and device, equipment and storage medium
By using AI prediction models and federated learning optimization techniques, the target wake-up time is dynamically adjusted, solving the problem that existing technologies cannot cope with sudden traffic surges, thereby extending device battery life and improving user experience.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-24
- Publication Date
- 2026-03-24
AI Technical Summary
Existing target wake-up time scheduling methods cannot cope with sudden traffic surges, resulting in insufficient energy efficiency and increased latency.
By employing an AI prediction model that combines a long short-term memory network and a lightweight deep learning network, the system dynamically predicts the target wake-up time based on device traffic information, environmental information, and link information. The model parameters are then optimized through federated learning to achieve closed-loop feedback adjustment.
It improves the rationality of target wake-up time scheduling, extends device battery life, and enhances the user experience.
Smart Images
Figure CN121728541A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wireless communication technology, and in particular to a target wake-up time scheduling method, apparatus, device, and storage medium. Background Technology
[0002] Target Wake Time (TWT), an energy-saving mechanism introduced in Wi-Fi 6 / 7, negotiates wake-up times between devices and access points (APs) to reduce unnecessary standby power consumption and significantly extend battery life. Essentially, the device and access point agree on a specific wake-up time, disabling Wi-Fi functionality during non-wake periods to avoid continuous channel monitoring and thus extend battery life. However, existing technologies often use static configuration, which cannot handle sudden traffic surges, leading to insufficient energy efficiency and increased latency. Summary of the Invention
[0003] In view of this, the purpose of this invention is to provide a target wake-up time scheduling method, apparatus, device, and storage medium, which can improve the rationality of target wake-up time scheduling, thereby improving device battery life and energy saving effect, as well as user experience. The specific solution is as follows:
[0004] In a first aspect, this application discloses a target wake-up time scheduling method, comprising:
[0005] Obtain basic information; the basic information includes traffic information and environmental information of the target device, as well as link information between the target device and the access point;
[0006] The basic information is input into the AI prediction model, and the first target wake-up time is predicted based on the model output. The target device and the access point are configured according to the first target wake-up time, and the actual user experience quality value is calculated after configuration.
[0007] The second target wake-up time is calculated based on the actual user experience quality value. The final target wake-up time is determined based on the first target wake-up time and the second target wake-up time. The target device and the access point are then configured.
[0008] Optionally, the AI prediction model includes a first sub-network built on a long short-term memory network and an attention mechanism, and a second sub-network built on a lightweight deep learning network.
[0009] The first target wake-up time is predicted based on the model output, including:
[0010] Based on the basic information, future load information is predicted using the first sub-network, and scene state information is predicted using the second sub-network based on the basic information.
[0011] The first target wake-up time is determined based on the load information and the scene state information.
[0012] Optionally, determining the first target wake-up time based on the load information and the scene state information includes:
[0013] Based on the future load information, the predicted latency and predicted packet loss rate are determined, and the predicted user experience quality value is calculated based on the predicted latency and the predicted packet loss rate.
[0014] The adjustment coefficient is determined based on the predicted user experience quality value and the target user experience quality value;
[0015] The target wake-up time threshold is determined based on the scenario status information;
[0016] The first target wake-up time is determined based on the current target wake-up time, the adjustment coefficient, and the target wake-up time threshold.
[0017] Optionally, determining the final target wake-up time based on the first target wake-up time and the second target wake-up time includes:
[0018] Target weights are generated based on the confidence level of the first target wake-up time and the historical error index;
[0019] The final target wake-up time is obtained by weighted averaging based on the first target wake-up time, the second target wake-up time, and the target weight.
[0020] Optionally, calculating the second target wake-up time based on the actual user experience quality value includes:
[0021] Compare the actual user experience quality value with the experience quality threshold;
[0022] If the actual user experience quality value is less than the experience quality threshold, then the second target wake-up time is determined according to the shortening strategy;
[0023] If the actual user experience quality value is not less than the experience quality threshold, then the second target wake-up time is determined according to the extended optimization strategy.
[0024] Optionally, the training process of the AI prediction model includes:
[0025] The initial model parameters of the initial model are sent to multiple client devices. The client devices train the model based on the initial model parameters to calculate local model parameters, and add noise to the local model parameters to obtain processed local model parameters.
[0026] The processed local model parameters uploaded by different client devices are obtained, and the initial model is updated after weighted averaging to obtain the AI prediction model.
[0027] Optionally, the AI prediction model is trained with the goal of minimizing a cost function, which is constructed based on a trade-off factor, the overall energy consumption of the device, and the user experience quality value. The trade-off factor is dynamically determined according to usage.
[0028] Optionally, the target wake-up time scheduling method further includes:
[0029] The received signal strength of the target device is obtained. If the received signal strength is less than a preset strength threshold, a new access point is selected for the target device.
[0030] Coordination information is generated based on the current target wake-up time and the latest model output corresponding to the target device;
[0031] The coordination information is sent to the new access point, and after sending, the link of the target device is switched to the link with the new access point.
[0032] Secondly, this application discloses a target wake-up time scheduling device, comprising:
[0033] The information acquisition module is used to acquire basic information, including traffic information and environmental information of the target device, as well as link information between the target device and the access point.
[0034] The first target wake-up time determination module is used to input the basic information into the AI prediction model, predict the first target wake-up time according to the model output, configure the target device and the access point according to the first target wake-up time, and calculate the actual user experience quality value after configuration.
[0035] The final target wake-up time determination module is used to calculate a second target wake-up time based on the actual user experience quality value, determine the final target wake-up time based on the first target wake-up time and the second target wake-up time, and configure the target device and the access point.
[0036] Thirdly, this application discloses an electronic device, including:
[0037] Memory, used to store computer programs;
[0038] A processor is used to execute the computer program to implement the aforementioned target wake-up time scheduling method.
[0039] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein the computer program, when executed by a processor, implements the aforementioned target wake-up time scheduling method.
[0040] In this application, basic information is obtained, including traffic information and environmental information of the target device, as well as link information between the target device and the access point. This basic information is input into an AI prediction model, and a first target wake-up time is predicted based on the model output. The target device and the access point are configured according to the first target wake-up time, and an actual user experience quality value is calculated after configuration. A second target wake-up time is calculated based on the actual user experience quality value. The final target wake-up time is determined based on the first and second target wake-up times, and the target device and the access point are configured again. It is evident that by predicting the first target wake-up time currently adapted to the target device based on the device's traffic information, environmental information, and corresponding link information, and combining this with dynamic calculation using multimodal information, it can adapt to various usage scenarios, cope with traffic surges, and improve the accuracy of the first target wake-up time. Simultaneously, by adjusting the target wake-up time based on the second target wake-up time calculated using the actual user experience quality value, a closed-loop feedback adjustment is achieved, improving the rationality of target wake-up time scheduling, thereby improving device battery life and energy efficiency, as well as the user experience. Attached Figure Description
[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0042] Figure 1 A flowchart of a target wake-up time scheduling method provided in this application; Figure 2 A flowchart of a specific target wake-up time scheduling method provided in this application; Figure 3 This application provides a specific timing diagram for coordinating multi-link operation and target wake-up time scheduling; Figure 4 A schematic diagram of a target wake-up time scheduling device provided in this application; Figure 5 This application provides a structural diagram of an electronic device. Detailed Implementation
[0043] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0044] In existing technologies, static configuration is often used, which cannot cope with sudden traffic surges, resulting in insufficient energy efficiency and increased latency. To overcome the above technical problems, this application proposes a target wake-up time scheduling method, which can improve the rationality of target wake-up time scheduling, thereby improving device battery life and energy saving effect, as well as user experience.
[0045] This application discloses a target wake-up time scheduling method. See also Figure 1 As shown, the method may include the following steps:
[0046] Step S11: Obtain basic information; the basic information includes traffic information and environmental information of the target device, as well as link information between the target device and the access point.
[0047] First, basic information is acquired in real time. This basic information includes, but is not limited to, traffic information of the target device, environmental information, and link information of the link formed between the target device and the access point. Traffic information may include bandwidth requirements and burst rate; bandwidth requirements refer to the amount of data to be transmitted, and burst rate measures the degree of fluctuation or instability of the data stream. Link information may include, but is not limited to, Received Signal Strength Indicator (RSSI), latency, and jitter. RSSI reflects the basic strength of the link, latency reflects the response speed of the link, and jitter reflects the stability of the link.
[0048] Environmental information can include Channel State Information (CSI), dynamic spectral entropy, and sensor status. It should be noted that sensor status refers not only to the physical on / off state of the sensor but also includes multi-dimensional information such as its operational mode, data quality state, hardware health, connectivity state, event-level state, and abnormal behavior. Operational modes include, but are not limited to: active acquisition mode, low-power acquisition mode, sleep / standby mode, and triggered acquisition mode. Data quality state includes, but is not limited to: noise level, signal stability, data loss rate, and data anomalies. Hardware status includes, but is not limited to: battery level, sensor temperature, self-check result, and hardware fault. Connectivity state includes, but is not limited to: channel occupancy, link quality (RSSI / SNR), and reporting interval. Event-level state includes, but is not limited to: human activity detection, environmental change events, and trigger event occurrence time.
[0049] Step S12: Input the basic information into the AI prediction model, predict the first target wake-up time based on the model output, configure the target device and the access point based on the first target wake-up time, and calculate the actual user experience quality value after configuration.
[0050] The acquired basic information is input into the AI prediction model. This AI prediction model, an artificial intelligence model, can predict future device usage and link conditions based on existing basic information. It then uses this learned information to predict the first target wake-up time for the target device within a certain period, achieving advance prediction of the target wake-up time. The prediction output of the AI prediction model is an indispensable direct input variable for TWT parameter calculation. Without model prediction, stable TWT scheduling parameters cannot be obtained, making it impossible to simultaneously meet energy consumption and latency constraints. Based on the determined first target wake-up time, the target device and access point are configured so that the target device wakes up at the corresponding time. Simultaneously, after configuration, the actual Quality of Experience (QoE) value is calculated, and this QoE value is then used to adjust the target wake-up time.
[0051] In a preferred embodiment, the model output includes future load information and scenario status information. The future load information is the service load within the future TWT window predicted based on features such as historical traffic sequences, service types, and interaction patterns, such as predicting the peak probability of the next TWT window. The scenario status information is the environmental status inferred from multimodal features such as CSI waveforms, sensor status, channel load, and location activity. Compared to related technologies that adjust the wake-up time based solely on historical traffic fluctuations, combining future load information and scenario status information to cross-determine the first target wake-up time can identify real-time services, determine whether a user has left, and predict the wake-up time in advance, significantly reducing scheduling latency and saving device power.
[0052] In a specific embodiment, the AI prediction model includes a first sub-network constructed based on a Long Short-Term Memory (LSTM) network and an attention mechanism, and a second sub-network constructed based on a lightweight deep learning network. The model predicts a first target wake-up time based on its output, including: predicting future load information using the first sub-network based on the basic information; predicting scene state information using the second sub-network based on the basic information; and determining the first target wake-up time based on the load information and the scene state information. Specifically, the first sub-network structure uses a Short Short-Term Memory (LSTM) network (hidden=128) combined with an attention mechanism, while the second sub-network structure uses a lightweight Transformer. The first sub-network outputs predicted future load information (y_hat), which is the business load within the future TWT window predicted by the first sub-network based on features such as historical traffic sequences, business types, and interaction patterns. Examples of this include: data volume, peak business probability, packet arrival rate, and queue growth trends over a future period. The second subnetwork predicts the scene state information (env_hat) for a future period. Specifically, it infers the environmental state based on multimodal features such as CSI waveforms, sensor status, channel load, and location activity. Scene state information can include: unmanned state, low-activity state with personnel, high-activity state, real-time service interaction state, or abnormal state. It also outputs corresponding confidence scores: the confidence score conf_y for future load information and the confidence score conf_env for scene state information.
[0053] Of course, in addition to using a dual-network AI prediction model, a single-network AI prediction model can also be used. For example, a single AI prediction model can be constructed by using a network structure with a shared encoder and two prediction heads, with the two prediction heads outputting future load information and scene state information respectively.
[0054] Furthermore, determining the first target wake-up time based on the load information and the scene state information includes the following steps:
[0055] S201: Determine the predicted latency and predicted packet loss rate based on the future load information, and calculate the predicted user experience quality value based on the predicted latency and the predicted packet loss rate; Predicted User Experience Quality Score The calculation formula is as follows: ; in, To predict delays, To predict packet loss rate, , Here is the weighting constant. This is a parameter used to prevent division by zero. Specifically, The value range can be [0.5, 5]. The value range can be [1, 50], because packet loss rate is usually small but has a significant impact. Of course, the first subnet can also be designed to directly output the predicted user experience quality value.
[0056] S202: Determine the adjustment coefficient based on the predicted user experience quality value and the target user experience quality value;
[0057] The target user experience quality value can be determined based on historical averages or strategy values. The adjustment factor is determined based on the predicted user experience quality value and the target user experience quality value: Where k is a sensitivity constant, the larger the k is, the more aggressive the adjustment;
[0058] S203: Determine the target wake-up time threshold value based on the scene state information;
[0059] S204: Determine the first target wake-up time based on the current target wake-up time, the adjustment coefficient, and the target wake-up time threshold. Specifically, extract the model-predicted future latency and future packet loss rate from the future load information output by the model, and then calculate the predicted user experience quality value according to the formula.
[0060] Based on the scene state information predicted by the second sub-network, the target wake-up time threshold is determined by querying the mapping relationship between scene state information and threshold value. , Then, based on the current target wake-up time, the adjustment coefficient, and the target wake-up time threshold, the first target wake-up time is determined: ; Here, `clamp` is used to constrain the target's wake-up time within a specified range. That is, if... Greater than and less than Then choose As the primary target wake-up time; if Less than Then choose As the primary target wake-up time; if Greater than Then select it as the first target wake-up time.
[0061] Furthermore, after calculating the first target wake-up time, the current target wake-up time can be used to filter and smooth it. For example, exponential smoothing / low-pass filtering can be used: ; in, Wake-up time for the current target; The adoption rate is defined as the value in the range of (0,1], which can be selected based on whether the strategy is conservative or aggressive.
[0062] In some embodiments, the training process of the above-mentioned AI prediction model may include the following steps:
[0063] S301: Send the initial model parameters of the initial model to multiple client devices; the server (Federated Optimization Center) first sends the initial model parameters to each client device, and different devices train the model through the Federated Optimization Center.
[0064] S302: The client device performs model training based on the initial model parameters to calculate the local model parameters, and adds noise to the local model parameters to obtain the processed local model parameters;
[0065] This involves using federated learning to train the model on multiple client devices. Furthermore, to protect data security, differential privacy is achieved by adding noise to the local model parameters.
[0066] S303: Obtain the processed local model parameters uploaded by different client devices, and update the initial model after weighted averaging to obtain the AI prediction model.
[0067] The server performs a weighted average of the processed local model parameters uploaded by each client device to obtain the averaged model parameters. It then uses these parameters to update the initial model. This process is repeated until the model converges to obtain the AI prediction model.
[0068] Furthermore, the AI prediction model is trained with the objective of minimizing a cost function. This cost function is constructed based on a tradeoff factor, overall device energy consumption, and user experience quality value. The tradeoff factor is dynamically determined according to usage. Cost function: ; Where P represents the device’s total energy consumption during the current communication cycle, including idle listening, data reception, transmission, wake-up, and multimodal sensing overhead; As a trade-off between energy consumption and user experience, , The larger the value, the more energy-efficient it is. A smaller value indicates a greater emphasis on ensuring user experience. .further, It can be dynamically adjusted based on application type, network latency, user level, and environmental conditions (such as personnel leaving / area being unoccupied).
[0069] A specific model training process is as follows:
[0070] Global initialization: Initialize global model W0; Initialize a global AI model with parameters W0 as the starting point for federated learning. All devices will begin local training from this model. Set learning rate Set the learning rate for model training. The learning rate controls the step size of each parameter update; QoE_threshold; Sets the experience quality threshold; RSSI_threshold=-80dBm; This sets the threshold for the received signal strength, meaning that a signal strength below -80dBm is considered weak and the connection may be unstable.
[0071] Global training rounds: For each global round t = 1...T; the entire federated learning process will consist of T global rounds. In each round, the server coordinates all devices to perform local training and then aggregates the results. Local training phase of the device: For each device i in parallel: Wi = Wt; Initialize to the global model. Each device i downloads the latest global model parameters Wt from the server and uses them as the initial state Wi of its local model. for epoch in local_epochs; The device trains on its own local data for multiple epochs (local_epochs). y_hat,env_hat=Model(Wi,Ft,Et,Lt); Ft,Et,Lt are traffic information, environment information, and link information, respectively. The model outputs future load information y_hat and scene status information env_hat. The client device is trained based on the initial model parameters, with the loss function being Li, where MSE represents the mean squared error and CE represents the cross-entropy loss. and It's a hyperparameter. , This indicates future actual load information and actual scenario status information; Update the device's local model parameters Wi using gradient descent. It is the gradient of the loss function with respect to the model parameters; After local training is complete, the device calculates the update amounts of the local model and the initial global model: Noise is noise used to achieve differential privacy, and then it is sent to the server.
[0072] Server federation aggregation: The server collects noisy model updates from all devices; it executes a federated averaging algorithm to aggregate these updates to generate a new global model Wt+1; ni is the amount of data from device i, and N is the total amount of data from all devices.
[0073] Finally, the server broadcasts the updated global model Wt+1 to all devices, thus initiating the next global training round.
[0074] In some embodiments, the training process of the AI prediction model includes: deploying a lightweight model on a client device, where the client device trains the model; and periodically exchanging the latest model summary with different clients so that the clients can update their models based on the received latest model summary. In other words, besides employing federated learning, model training can also involve peer-to-peer communication between clients, eliminating reliance on a central server for weighted averaging, and saving communication costs by only exchanging model summaries.
[0075] In some embodiments, before configuring the target device and the access point according to the first target wake-up time, the method further includes: comparing the first target wake-up time with the current target wake-up time; if the difference between the first target wake-up time and the current target wake-up time exceeds a first error threshold, then performing the operation of configuring the target device and the access point according to the first target wake-up time. That is, to avoid resource consumption caused by frequent updates to the target wake-up time, a slight error in the target wake-up time is allowed; if the difference between the first target wake-up time and the current target wake-up time does not exceed the first error threshold, no processing is performed.
[0076] Step S13: Calculate the second target wake-up time based on the actual user experience quality value, determine the final target wake-up time based on the first target wake-up time and the second target wake-up time, and configure the target device and the access point.
[0077] After configuration, the actual user experience quality value is calculated. The calculation of the actual user experience quality value is similar to that of the predicted user experience quality value, except that the actual user experience quality value is calculated based on actual latency and actual packet loss rate. Then, a second target wake-up time is calculated based on the actual user experience quality value. Finally, the final target wake-up time is determined based on the first and second target wake-up times, i.e., the target wake-up time is adjusted with reference to the actual user experience quality value. In other words, by making forward-looking adjustments using predicted values (y_hat, env_hat) and performing closed-loop correction using the actual user experience quality value, the reasonableness of the target wake-up time is improved.
[0078] The calculation of the second target wake-up time based on the actual user experience quality value includes: comparing the actual user experience quality value with an experience quality threshold; if the actual user experience quality value is less than the experience quality threshold, then the second target wake-up time is determined according to a shortening strategy; if the actual user experience quality value is not less than the experience quality threshold, then the second target wake-up time is determined according to an extension optimization strategy. For example... Figure 2 As shown, it's understandable that the Quality of Experience (QoE) threshold is a preset, rather poor threshold. If the actual user experience quality value is lower than this threshold, the current state is unacceptable, and an emergency shortening strategy can be adopted. For example, the current target wake-up time can be halved to obtain a second target wake-up time, but at the same time, a target wake-up time threshold can be used as a constraint to ensure that the target wake-up time is not lower than the minimum value. If the actual user experience quality value is not less than the experience quality threshold, indicating that the current situation is good, a gentle extension strategy can be adopted; for example, 1.05 times the previous target wake-up time can be used as the second target wake-up time to gradually explore the possibility of more power saving while maintaining stability.
[0079] As can be seen, the final target wake-up time in this application is determined through a joint decision based on predicted load, predicted scenario state, and user experience quality value, and is dynamically controlled under multiple constraints. For example, when the business predicted load is high but the environment is predicted to be unattended, aggressive shortening of TWT is not performed; when the environment is predicted to be a real-time business scenario but QoE has not yet decreased, small-step TWT adjustments are made in advance; when real-time QoE suddenly deteriorates but the AI prediction model fails to identify the business peak, only limited adjustments are made to avoid oscillations. The entire process does not rely on a fixed TWT table or static rules, but rather dynamically adjusts TWT based on predicted future usage.
[0080] In a preferred embodiment, the formula for calculating the wake-up time of the second target is: ; in, Let f(y_hat) represent the current target wake-up time, f(y_hat) represent the quantifiable adjustment range based on future load information (y_hat), and g(env_hat) represent the quantifiable adjustment range based on scene state information (env_hat). , For adaptive step size, The experience quality threshold (QoE_threshold); where, and Specifically, the step size parameter can be obtained by adaptively updating based on historical QoE fluctuations.
[0081] Before determining the final target wake-up time based on the first target wake-up time and the second target wake-up time, the method further includes: if the difference between the first target wake-up time and the second target wake-up time exceeds a second error threshold, then the final target wake-up time is determined based on the first target wake-up time and the second target wake-up time; otherwise, no modification is made. That is, if the difference between the first target wake-up time and the second target wake-up time is small, no processing is required to avoid consuming scheduling resources.
[0082] Of course, if the first target wake-up time is calculated based on the predicted user experience quality value, before calculating the second target wake-up time based on the actual user experience quality value, the following steps are also included: comparing the predicted user experience quality value with the actual user experience quality value; if the difference between the predicted user experience quality value and the actual user experience quality value exceeds the experience quality error threshold, then the second target wake-up time is calculated based on the actual user experience quality value; if it does not exceed the threshold, then the first target wake-up time can still be used to avoid consuming scheduling resources.
[0083] In some embodiments, determining the final target wake-up time based on the first target wake-up time and the second target wake-up time includes: generating a target weight based on the confidence level of the first target wake-up time and the historical error index; and obtaining the final target wake-up time by weighted averaging based on the first target wake-up time, the second target wake-up time, and the target weight. That is, the first target wake-up time and the second target wake-up time are combined through a fusion of confidence level and weight to ultimately generate the final target wake-up time. Specifically, in each federated / local inference cycle, QoEpred can be calculated based on y_hat and env_hat, and a suggested first target wake-up time can be given accordingly to allow for advance adjustments to cope with upcoming load / scenario changes. After performing configuration based on the first target wake-up time, the actual latency and packet loss rate are obtained in subsequent runtime windows to calculate the actual user experience quality value QoEmeas. After each adjustment, a weighted fusion of prediction and actual measurements can be used to determine the final target wake-up time, with the fusion weight determined by the confidence level and historical error.
[0084] For example, first determine the confidence level of the first target wake-up time. If the first target wake-up time is calculated based on the future load information predicted by the first sub-network, then the confidence level at this time is conf_y. If the first target wake-up time is the target wake-up time threshold, then the confidence score uses conf_env. Then, the historical prediction error index (EMA_error) is calculated. The historical prediction error index can be the average relative error over the past N predictions, used to adjust the confidence level. Finally, the target weights (w_pred, w_meas) are calculated. ; in, A settable coefficient is used to enable... .
[0085] Weighted average to obtain the final target wake-up time: ; in, The wake-up time for the first target The wake-up time for the second objective is given. It is evident that if the prediction confidence is high and historical accuracy is good, the wake-up time for the first objective is given higher weight. This allows for early prevention and mitigation through prediction, followed by adjustments and stabilization based on measured values.
[0086] This application combines artificial intelligence prediction, multimodal environment perception, multi-link operation coordination, and federated learning mechanisms to optimize the target wake-up time under the IEEE 802.11bn (Wi-Fi 8) standard, improve the rationality of target wake-up time scheduling, thereby improving device battery life and energy saving, as well as user experience.
[0087] As can be seen from the above, this embodiment acquires basic information, including traffic information and environmental information of the target device, as well as link information between the target device and the access point. The basic information is input into an AI prediction model, and a first target wake-up time is predicted based on the model output. The target device and the access point are configured based on the first target wake-up time, and an actual user experience quality value is calculated after configuration. A second target wake-up time is calculated based on the actual user experience quality value. The final target wake-up time is determined based on the first and second target wake-up times, and the target device and the access point are configured again. It is evident that by predicting the first target wake-up time currently adapted to the target device based on the device's traffic information, environmental information, and corresponding link information, and by dynamically calculating using multimodal information to improve the accuracy of the first target wake-up time, and by adjusting the target wake-up time based on the second target wake-up time calculated using the actual user experience quality value, a closed-loop feedback adjustment is achieved, improving the user experience.
[0088] In some embodiments, the target wake-up time scheduling method described above may further include: if the target device is a real-time interactive device, the target wake-up time is initially configured as a third target wake-up time, which can be a value less than 10ms; the real-time interactive device is an XR / VR device, etc., to ensure the normal operation of such devices. If the target device is a low-power device, the target wake-up time is initially configured as a fourth target wake-up time, which can be a value greater than 500ms; adjustments are made if the actual user experience quality value shows extreme degradation during subsequent operation.
[0089] In some embodiments, the target wake-up time scheduling method described above may further include: obtaining a link quality score of the link between the target device and the current access point; if the link quality score is less than a score threshold, then selecting a new access point for the target device; generating coordination information based on the current target wake-up time and the latest model output corresponding to the target device; sending the coordination information to the new access point, and switching the link of the target device to the link with the new access point after sending. That is, when the link quality deteriorates, timely link switching is performed to ensure normal service operation; link quality can be determined by comprehensively considering factors such as received signal strength, signal-to-noise ratio, and bit error rate.
[0090] Furthermore, to ensure that the target wake-up time configuration can still be used between the new access point and the target device, coordination information is generated for the target device, including the current target wake-up time and the latest model output of the AI prediction model for that target device, and sent to the new access point. That is, the coordination information is used to synchronize future load information, scene status information, and the current target wake-up time scheduling configuration across access points. In a preferred embodiment, control frames are used as the transmission method. Control frames are generated based on the coordination information and sent to the new access node. These control frames include, but are not limited to, the following fields: the latest model output fields (i.e., future load information (y_hat), scene status information (env_hat)), the current link quality index field (LINK_QUALITY_INDEX), and the current target wake-up time scheduling configuration fields (i.e., the current final target wake-up time, wake-up cycle, etc.). This ensures that multiple links associated with a device can select a consistent target wake-up time scheduling strategy under the same environmental perception results, thereby avoiding additional wake-ups, increased latency, or increased energy consumption caused by inconsistent strategies between links. By adding control frames to extend the MAC layer capabilities, collaborative energy-saving decisions across APs are achieved, thereby improving overall energy efficiency and QoE. To achieve coordination between Multi-Link Operation (MLO) and target wake-up time scheduling.
[0091] In addition to generating dedicated control frames, information elements (such as IE (Information Element) fields) can be added to Beacon or Probe Response frames. These information elements are similar to the fields included in the control frames, including but not limited to the current target wake-up time of the target device and the latest model output.
[0092] By synchronizing information such as the current wake-up time configuration and the latest model output across access points, the coordination between multi-link operation and target wake-up time scheduling is achieved, solving the problem of inconsistent wake-up times and lack of synchronization after switching links under multi-link operation in the prior art.
[0093] In some embodiments, the target wake-up time scheduling method described above may further include: acquiring the received signal strength of the target device; if the received signal strength is less than a preset strength threshold, then selecting a new access point for the target device; generating coordination information based on the current target wake-up time and the latest model output corresponding to the target device; sending the coordination information to the new access point, and switching the link of the target device to the link with the new access point after sending. Specifically, received signal strength can be used as a link quality evaluation indicator; for example, if RSSI is less than 80dBm and the new link quality meets the condition, link migration is triggered. Alternatively, a hybrid strategy of old and new links can be used, such as distributing the current target wake-up time across two links.
[0094] For example Figure 3 The diagram shows a specific multi-link operation and target wake-up time scheduling coordination sequence. The terminal device (i.e. the target device) sends an initial TWT request to the primary link AP, and the primary link AP confirms the TWT. When RSSI is less than 80dBm, the primary link sends a control frame to the backup link. After the backup link allocates a new TWT window to the device, the primary link is switched to the backup link, and the backup link enters the sleep / wake-up cycle according to the control frame.
[0095] Accordingly, embodiments of this application also disclose a target wake-up time scheduling device, see [link to relevant documentation]. Figure 4 As shown, the device includes:
[0096] The information acquisition module 11 is used to acquire basic information, including traffic information and environmental information of the target device, as well as link information between the target device and the access point.
[0097] The first target wake-up time determination module 12 is used to input the basic information into the AI prediction model, predict the first target wake-up time according to the model output, configure the target device and the access point according to the first target wake-up time, and calculate the actual user experience quality value after configuration.
[0098] The final target wake-up time determination module 13 is used to calculate the second target wake-up time based on the actual user experience quality value, determine the final target wake-up time based on the first target wake-up time and the second target wake-up time, and configure the target device and the access point.
[0099] As can be seen from the above, this embodiment acquires basic information, including traffic information and environmental information of the target device, as well as link information between the target device and the access point. The basic information is input into an AI prediction model, and a first target wake-up time is predicted based on the model output. The target device and the access point are configured based on the first target wake-up time, and an actual user experience quality value is calculated after configuration. A second target wake-up time is calculated based on the actual user experience quality value. The final target wake-up time is determined based on the first and second target wake-up times, and the target device and the access point are configured again. It is evident that by predicting the first target wake-up time currently adapted to the target device based on the device's traffic information, environmental information, and corresponding link information, and by dynamically calculating using multimodal information to improve the accuracy of the first target wake-up time, and by adjusting the target wake-up time based on the second target wake-up time calculated using the actual user experience quality value, a closed-loop feedback adjustment is achieved, improving the user experience.
[0100] In some specific embodiments, the AI prediction model includes a first sub-network built on a long short-term memory network and an attention mechanism, and a second sub-network built on a lightweight deep learning network;
[0101] The first target wake-up time determination module 12 may specifically include:
[0102] The prediction unit is used to predict future load information based on the basic information using the first sub-network, and to predict scene state information based on the basic information using the second sub-network.
[0103] The first target wake-up time determination unit is used to determine the first target wake-up time based on the load information and the scene state information.
[0104] In some specific embodiments, the first target wake-up time determination unit may specifically include:
[0105] The predictive user experience quality value calculation unit is used to determine the predicted latency and predicted packet loss rate based on the future load information, and calculate the predicted user experience quality value based on the predicted latency and the predicted packet loss rate.
[0106] An adjustment coefficient determination unit is used to determine an adjustment coefficient based on the predicted user experience quality value and the target user experience quality value.
[0107] The target wake-up time threshold determination unit is used to determine the target wake-up time threshold based on the scene state information.
[0108] The first target wake-up time determination unit is used to determine the first target wake-up time based on the current target wake-up time, the adjustment coefficient, and the target wake-up time threshold.
[0109] In some specific embodiments, the final target wake-up time determination module 13 may specifically include:
[0110] The weight generation unit is used to generate target weights based on the confidence level of the first target wake-up time and the historical error index.
[0111] The final target wake-up time calculation unit is used to obtain the final target wake-up time by weighted averaging based on the first target wake-up time, the second target wake-up time, and the target weight.
[0112] In some specific embodiments, the training process of the AI prediction model includes:
[0113] The initial model parameters of the initial model are sent to multiple client devices. The client devices train the model based on the initial model parameters to calculate local model parameters, and add noise to the local model parameters to obtain processed local model parameters.
[0114] The processed local model parameters uploaded by different client devices are obtained, and the initial model is updated after weighted averaging to obtain the AI prediction model.
[0115] In some specific embodiments, the AI prediction model is trained with the goal of minimizing a cost function, which is constructed based on a trade-off factor, the overall energy consumption of the device, and the user experience quality value. The trade-off factor is dynamically determined according to usage.
[0116] In some specific embodiments, the target wake-up time scheduling device may further include:
[0117] The new access point filtering unit is used to obtain the received signal strength of the target device. If the received signal strength is less than a preset strength threshold, a new access point is filtered for the target device.
[0118] A coordination information generation unit is used to generate coordination information based on the current target wake-up time and the latest model output corresponding to the target device;
[0119] The link switching unit is used to send the coordination information to the new access point, and after sending, switch the link of the target device to the link with the new access point.
[0120] Furthermore, this application also discloses an electronic device, see [link to relevant documentation]. Figure 5As shown, the content in the figure should not be considered as any limitation on the scope of use of this application.
[0121] Figure 5 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the target wake-up time scheduling method disclosed in any of the foregoing embodiments.
[0122] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.
[0123] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored on it include operating system 221, computer program 222 and data 223 including basic information, etc., and the storage method can be temporary storage or permanent storage.
[0124] The operating system 221 manages and controls the various hardware devices on the electronic device 20 and the computer program 222 to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. The operating system 221 can be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the target wake-up time scheduling method disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.
[0125] Furthermore, this application also discloses a computer storage medium storing computer-executable instructions. When the computer-executable instructions are loaded and executed by a processor, they implement the target wake-up time scheduling method steps disclosed in any of the foregoing embodiments.
[0126] Furthermore, embodiments of this application also disclose a computer program product, including a computer program that, when executed by a processor, implements the target wake-up time scheduling method steps disclosed in any of the foregoing embodiments.
[0127] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.
[0128] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.
[0129] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0130] The above provides a detailed description of the target wake-up time scheduling method, apparatus, device, and storage medium provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A target wake-up time scheduling method, characterized in that, include: Obtain basic information; the basic information includes traffic information and environmental information of the target device, as well as link information between the target device and the access point; The basic information is input into the AI prediction model, and the first target wake-up time is predicted based on the model output. The target device and the access point are configured according to the first target wake-up time, and the actual user experience quality value is calculated after configuration. The second target wake-up time is calculated based on the actual user experience quality value. The final target wake-up time is determined based on the first target wake-up time and the second target wake-up time. The target device and the access point are then configured.
2. The target wake-up time scheduling method according to claim 1, characterized in that, The AI prediction model includes a first sub-network built on a long short-term memory network and an attention mechanism, and a second sub-network built on a lightweight deep learning network. The first target wake-up time is predicted based on the model output, including: Based on the basic information, future load information is predicted using the first sub-network, and scene state information is predicted using the second sub-network based on the basic information. The first target wake-up time is determined based on the load information and the scene state information.
3. The target wake-up time scheduling method according to claim 2, characterized in that, Determining the first target wake-up time based on the load information and the scene state information includes: Based on the future load information, the predicted latency and predicted packet loss rate are determined, and the predicted user experience quality value is calculated based on the predicted latency and the predicted packet loss rate. The adjustment coefficient is determined based on the predicted user experience quality value and the target user experience quality value; The target wake-up time threshold is determined based on the scenario status information; The first target wake-up time is determined based on the current target wake-up time, the adjustment coefficient, and the target wake-up time threshold.
4. The target wake-up time scheduling method according to claim 1, characterized in that, The step of determining the final target wake-up time based on the first target wake-up time and the second target wake-up time includes: Target weights are generated based on the confidence level of the first target wake-up time and the historical error index; The final target wake-up time is obtained by weighted averaging based on the first target wake-up time, the second target wake-up time, and the target weight.
5. The target wake-up time scheduling method according to claim 1, characterized in that, The step of calculating the second target wake-up time based on the actual user experience quality value includes: Compare the actual user experience quality value with the experience quality threshold; If the actual user experience quality value is less than the experience quality threshold, then the second target wake-up time is determined according to the shortening strategy; If the actual user experience quality value is not less than the experience quality threshold, then the second target wake-up time is determined according to the extended optimization strategy.
6. The target wake-up time scheduling method according to claim 1, characterized in that, The training process of the AI prediction model includes: The initial model parameters of the initial model are sent to multiple client devices. The client devices train the model based on the initial model parameters to calculate local model parameters, and add noise to the local model parameters to obtain processed local model parameters. The processed local model parameters uploaded by different client devices are obtained, and the initial model is updated after weighted averaging to obtain the AI prediction model.
7. The target wake-up time scheduling method according to claim 6, characterized in that, The AI prediction model is trained with the goal of minimizing the cost function, which is constructed based on trade-off factors, overall device energy consumption, and user experience quality values. The trade-off factors are dynamically determined according to usage.
8. The target wake-up time scheduling method according to any one of claims 1 to 7, characterized in that, Also includes: The received signal strength of the target device is obtained. If the received signal strength is less than a preset strength threshold, a new access point is selected for the target device. Coordination information is generated based on the current target wake-up time and the latest model output corresponding to the target device; The coordination information is sent to the new access point, and after sending, the link of the target device is switched to the link with the new access point.
9. A target wake-up time scheduling device, characterized in that, include: The information acquisition module is used to acquire basic information; The basic information includes the target device's traffic information, environmental information, and the link information between the target device and the access point; The first target wake-up time determination module is used to input the basic information into the AI prediction model, predict the first target wake-up time according to the model output, configure the target device and the access point according to the first target wake-up time, and calculate the actual user experience quality value after configuration. The final target wake-up time determination module is used to calculate a second target wake-up time based on the actual user experience quality value, determine the final target wake-up time based on the first target wake-up time and the second target wake-up time, and configure the target device and the access point.
10. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the target wake-up time scheduling method as described in any one of claims 1 to 8.
11. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein the computer programs, when executed by a processor, implement the target wake-up time scheduling method as described in any one of claims 1 to 8.