A handover prediction system and method for LEO satellite networks based on segmented page hierarchical caching

Abstract

This invention discloses a LEO satellite network handover prediction system and method based on segment-page hierarchical caching, relating to the field of satellite communication technology, and solving the problem of high computational complexity in handover prediction under large-scale user scenarios. The system adopts a segment-page hierarchical caching architecture, dividing the global geospatial space into segment-level and page-level grids to construct a two-level geographic segmentation structure; parallel processing is achieved through a dual-process prediction processor. The system includes an active region maintenance module, a segment satellite mapping module, a high-precision pre-caching module, a page matching module, and a batch binary backtracking module. Experimental results show that when the number of users increases from 10,000 to 200,000, the decision process time remains stable, the number of predictions exhibits a saturation effect, and the computational overhead is decoupled from the number of users, significantly improving the efficiency and scalability of LEO satellite network handover prediction.

Description

Technical Field

[0001] This invention relates to the field of satellite communication technology, and more specifically, to a handover prediction system and method for LEO satellite networks based on segment-page hierarchical caching, particularly a system and method designed to solve the problem of computational complexity in handover prediction under large-scale user scenarios. Background Technology

[0002] Low-Earth orbit (LEO) satellite communication networks have broad development prospects due to their global coverage and low transmission latency. However, the high-speed movement of LEO satellites around the Earth necessitates frequent network handovers for terrestrial user equipment (UEs). Traditional mobility management mechanisms largely follow terrestrial network strategies, relying on time-consuming signaling interactions between the terminal and the network to complete the handover. In scenarios involving high-speed satellite movement, this model can severely impact service quality due to signaling delays.

[0003] To address this challenge, existing advanced technologies have proposed pre-calculating the UE's access satellites on the core network side by leveraging the determinism and predictability of satellite orbits, thereby avoiding or reducing time-consuming signaling interactions between the core network and the access network (i.e., satellites) during handover. This type of predictive handover scheme can significantly reduce handover latency and represents the current direction of technological development.

[0004] A typical implementation of this type of predictive handover scheme requires the core network to periodically and independently predict the future access satellites for each UE. This process usually involves the following steps: UE location acquisition, orbit prediction, traversal calculation, and optimal satellite selection. However, it is precisely this mode of independent traversal calculation for each UE that introduces a new and severe technical challenge to this scheme: huge computational overhead. In future LEO networks, a single core network may need to serve millions or even more UEs. Performing the above calculation process independently and periodically for each UE will place an unbearable computational burden on the core network. Especially when the number of users is large, the prediction task may not be completed within the required decision-making cycle, thus causing the entire prediction mechanism to fail. Therefore, how to fundamentally solve the problem of the strong coupling between the computational complexity of the prediction algorithm and the user scale without sacrificing prediction accuracy is the key bottleneck for realizing the practical application and large-scale deployment of predictive handover schemes. Summary of the Invention

[0005] The purpose of this invention is to address the problems of high computational complexity, low cache utilization, and coarse geographic area management in existing LEO satellite network handover prediction algorithms, and to provide a LEO satellite network handover prediction system and method based on segment-page hierarchical caching. This method significantly improves computational efficiency and prediction accuracy through techniques such as a two-level geographic segmentation structure, high-precision orbit pre-caching, and batch binary backtracking.

[0006] The technical solution of the present invention is: a LEO satellite network handover prediction system based on segment and page hierarchical caching. The system is a dual-process prediction processor, including an active region maintenance module, a segment satellite mapping module, a high-precision orbit pre-caching module, a page matching module, and a batch binary backtracking module.

[0007] The prediction process, as process 1, is responsible for prediction and high-precision pre-caching. Among them: the active area maintenance module dynamically updates the list of active segments and active pages according to the location distribution of user equipment (UE) and establishes a parent-child relationship mapping between pages and segments; the segment satellite mapping module establishes candidate satellite mappings for active segments; and the high-precision orbit caching module performs orbit prediction on the candidate satellites identified in the segment satellite mapping and stores the prediction results.

[0008] The decision-making process, as process 2, is responsible for page-level matching and batch binary search backtracking. Specifically, the page matching module determines the optimal serving satellite for each active page based on the satellite position data provided by the high-precision orbit cache module; the batch binary search backtracking module performs batch binary search on the UEs that need to be switched, determines the switching time point, and outputs a list of switching events, including user identifier, source satellite, target satellite, and time point.

[0009] Furthermore, the prediction process is activated at the current time T of a decision period N to obtain the prediction result; the decision process is activated at the next time T+t to complete page matching, switch recognition and batch binary backtracking search using the prediction result data already in the cache, and finally outputs the switch result of period N.

[0010] Furthermore, at the beginning of each prediction period, the segment satellite mapping module determines a list of potential service satellites for each active segment and performs coarse screening using segment-level coverage radius. The active segment is the segment where user equipment exists, and the side length of the segment depends on the current satellite's scanning radius.

[0011] Furthermore, the high-precision orbital caching module adopts a two-level geographic segmentation structure, dividing the global geographic space into segment-level grids and page-level grids for geographic region management. The segment-level and page-level grids differ by 10-50 orders of magnitude.

[0012] Furthermore, the optimal service satellite is determined based on the received signal strength indication value.

[0013] Furthermore, the batch binary backtracking module adopts a hierarchical batch processing mechanism. Each iteration includes three stages: demand collection, batch prediction, and decision diversion. In the demand collection stage, prediction demands from all active contexts are collected in a unified manner. In the batch prediction stage, orbit predictions for all unique time point-satellite pairs are completed at once. In the decision diversion stage, the binary search interval is updated based on the RSSI comparison results.

[0014] This invention also provides a method for predicting LEO satellite network handover based on segment-page hierarchical caching, comprising the following steps:

[0015] Step 1: Global segment satellite mapping and candidate set generation; At the beginning of the prediction period, the satellite constellation is first spatially discretized to establish a dynamic mapping relationship between satellites and segment IDs; Using the aforementioned mapping relationship, all satellites in the parent segment where the active page is located and in adjacent segments are retrieved to obtain a candidate satellite list;

[0016] Step 2: Page matching and handover requirement identification; Based on the candidate satellite list, for each page-level grid with a UE, determine its group optimal serving satellite based on the received signal strength indicator value, and then traverse all UEs in the page, compare its currently connected serving satellite with the elected group optimal serving satellite, determine the handover requirement, and if a handover is required, generate a pending handover event for it.

[0017] Step 3: Batch binary backtracking; After identifying all UEs with handover requirements, start the batch binary backtracking method to determine the handover time point for each UE;

[0018] Step 4: Iterative convergence and event generation; Batch binary backtracking method iterative convergence, when the UE's search interval width is less than the preset accuracy threshold or the preset maximum number of iterations is reached, the iteration stops, and the final handover event list is generated for subsequent network execution;

[0019] Step 2 is described in detail below:

[0020] Sampling point layout and weighting: Within each page-level grid containing users, a nine-point sampling model is used to deploy nine fixed virtual samples, and differentiated weighting coefficients w are assigned to sampling points at different locations. i ;

[0021] Quality of Service (QoS) score calculation: For each satellite s in the candidate satellite list, calculate its page QoS score, which is the weighted sum of the received signal strength indication values ​​of the satellite at all sampling points. The calculation formula is as follows:

[0022] ;

[0023] Where, p i It is the i-th sampling point, w i It is its corresponding weight, RSSI(s,p) i ) represents the signal strength of satellite s at that point; the satellite with the highest service quality score Q(s) is selected as the group's optimal service satellite s∗;

[0024] The system iterates through all UEs on the page and compares their currently connected serving satellite with the elected group's best serving satellite s*. If they do not match, the system determines that the UE has a potential handover requirement and generates a pending handover event for it. This event includes the UE identifier, source satellite, target satellite, and initial search interval. The initial search interval is defined as [t_low, t_high], where t_low represents the start time of the search interval and t_high represents the end time of the search interval.

[0025] Step 3 is as follows:

[0026] Step 3.1: Requirements Collection and Integration: At the beginning of the Gth iteration, the system first traverses all pending switching events, extracts their source satellite, target satellite and corresponding midpoint time mid_t, and generates two specific orbit prediction task units: <source satellite ID, mid_t> and <target satellite ID, mid_t>.

[0027] Initialize a hash set specifically for storing unique <satellite ID, mid_t> tuples. For each generated <satellite ID, mid_t> task unit, the hash function calculates a unique storage address based on the tuple's content. This storage address allows you to instantly locate its storage position in the hash table. Simply compare it within the hash table to check if there is an identical item. If it already exists, the current addition operation will be automatically ignored.

[0028] After traversing all pending switching events, the hash set contains all and unique <satellite ID, mid_t> combinations required for prediction in this iteration. This set constitutes the satellite-time point combination table for this iteration, serving as the input for the next batch orbit prediction.

[0029] Step 3.2: Batch Orbit Prediction and Caching; Based on the satellite-time point combination table, the system sends a one-time batch prediction request to the orbit predictor; After receiving this request table, the orbit predictor will traverse each satellite-time point combination and calculate the high-precision status of the specified satellite at the specified time; After the calculation is completed, all returned position information will be updated or stored in the shared orbit cache, and indexed by satellite ID and time point for quick querying in subsequent steps;

[0030] Step 3.3: Parallel Received Signal Strength Indicator Comparison and Decision; For each UE, retrieve the positions of its current satellite and target satellite at the current countermeasure time from the cache, and calculate two corresponding received signal strength indication values ​​in combination with the UE's own geographical location. By comparing the magnitude of the values, determine whether the signal of the target satellite is better than that of the current satellite at the midpoint time.

[0031] Step 3.4: Search interval update; Based on the comparison results of the previous step, update the time interval of the binary search for each UE independently;

[0032] If at time mid_t, the RSSI value of the target satellite is greater than the RSSI value of the current satellite, the upper limit of the time search interval for the UE is updated to mid_t, and the new search interval becomes [t_low, mid_t].

[0033] Conversely, if the target satellite signal is weaker than or equal to the current satellite signal, the lower limit of the search interval is updated to mid_t, and the new interval becomes [mid_t, t_high].

[0034] This method also includes anomaly handling and rollback mechanisms; in actual operation, whether it is because the UE is located at the page boundary and selects a non-predicted satellite with better signal, or because of sudden changes in user trajectory or orbit prediction deviation, any situation that causes the satellite actually accessed by the UE to be inconsistent with the core network prediction result is regarded as prediction failure.

[0035] The present invention, through the above-described method and system, has at least the following advantages:

[0036] Computational overhead is decoupled from the number of users: the complexity of the core decision-making process is only related to the number of active pages and does not increase linearly with the total number of user experiences (UEs). Experimental data shows that when the number of users increases from 10,000 to 200,000, the computation time of the traditional algorithm increases from approximately 409.27ms to approximately 7628.72ms, while the decision-making process time of this invention remains basically stable.

[0037] Prediction overhead tends to be constant: The two-level geographic grid design exhibits a "saturation effect" in high-density areas. When the number of users reaches 50,000, the number of predictions hardly increases, and the performance overhead can be locked at a controllable level.

[0038] Highly efficient batch processing mechanism: Through batch processing and deduplication of binary backtracking, the search for switching time points of massive UEs is transformed from serial calculation to parallel batch processing, which greatly reduces redundant calculations and improves concurrent processing capabilities. Attached Figure Description

[0039] Figure 1 This is a diagram of the overall system architecture.

[0040] Figure 2 This is a flowchart of a dual-process workflow;

[0041] Figure 3 This is a segmented page model diagram;

[0042] Figure 4 This is a diagram of a nine-point sampling model;

[0043] Figure 5 Here is a flowchart of the batch binary backtracking algorithm;

[0044] Figure 6 Sample images for matching results;

[0045] Figure 7 The figure shows the results of the performance comparison experiment.

[0046] Figure 8 This is a flowchart of the dual-process satellite switching process. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of the present invention clearer, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0048] This invention provides a LEO satellite network handover prediction system based on segment-page hierarchical caching, specifically a dual-process prediction processor. Process 1 performs active region maintenance and high-precision orbit pre-caching, while process 2 performs page-level matching and batch binary backtracking, achieving parallel processing of computational tasks. When process 2 is in period N, process 1 has already started prediction for period N+1.

[0049] The system comprises five core modules: an active region maintenance module, a segment satellite mapping module, a high-precision orbit pre-caching module, a page matching module, and a batch binary backtracking module. These modules work collaboratively to achieve a complete predictive process from geographic region management to precise switchover time determination. For example... Figure 1 and Figure 8 As shown, the specific modules are as follows:

[0050] Process 1 is responsible for prediction and high-precision pre-caching; the active area maintenance module dynamically updates the list of active segments and pages according to the UE location distribution, and establishes a parent-child relationship mapping between pages and segments; the segment satellite mapping module establishes candidate satellite mappings for active segments. Specifically, at the beginning of each prediction period, it determines a list of potential service satellites for each active segment and performs coarse screening using a segment-level coverage radius of 2000km; the high-precision orbit caching module performs orbit predictions on the candidate satellites identified in the segment satellite mapping and stores the results in a cache structure with a time point index; specifically, the high-precision orbit caching module adopts a two-level geographic segmentation structure, dividing the global geographic space into segment-level grids (1000km) and page-level grids (20km) to achieve refined geographic area management.

[0051] Process 2 is responsible for page-level matching and batch binary search backtracking. The page matching module determines the optimal serving satellite for each active page based on satellite position data provided by the high-precision orbit caching module, selecting the best satellite by comparing the RSSI values ​​of candidate satellites at the center of the page. The batch binary search backtracking module performs batch binary search on UEs requiring handover, determining the precise handover time point through multiple iterations. Specifically, it adopts a hierarchical batch processing mechanism, converting the traditional serial binary search into parallel batch processing. Each iteration includes three stages: demand collection, batch prediction, and decision routing. In the demand collection stage, prediction demands from all active contexts are collected uniformly. In the batch prediction stage, orbit predictions for all unique time-satellite pairs are completed at once. In the decision routing stage, the binary search interval is updated based on the RSSI comparison results. The output is a list of handover events, including user identifier, source satellite, target satellite, and time point.

[0052] This invention proposes a LEO satellite network handover prediction system based on segment-page hierarchical caching. Its core lies in employing a dual-process parallel architecture, such as... Figure 2 As shown, this architecture decouples the entire switching prediction task into two independent processes that work together through a pipeline mechanism: the prediction process (Process1_Predict) and the decision process (Process2_Match).

[0053] The prediction process (Process1_Predict) performs forward-looking pre-computation. Activated at the start time T of a decision cycle N, it first updates the list of active areas based on the real-time geographic distribution of user equipment (UEs), then filters candidate satellites for these areas, and predicts the precise orbital states of these satellites at the next decision point (T+5s). Finally, this high-precision data is stored in the shared orbital cache OrbitCache_N. This process is computationally intensive, but by executing it in advance, it avoids blocking subsequent real-time decisions.

[0054] The decision process (Process2_Match) is time-critical. After receiving the data-ready signal from the prediction process, this process does not execute immediately but waits for the time to advance to the precise decision point T+5 seconds. At that time, the decision process will use the OrbitCache_N data already in the cache to quickly complete page matching, switch recognition, and the core batch binary backtracking search, finally outputting the switch result for period N.

[0055] like Figure 2As shown, the pipeline characteristic of this system is reflected in the fact that while the decision-making process is processing the task of cycle N at time T+5s, the prediction process has already started the prediction task of the next cycle N+1 (preparing for time T+10s). This parallel and overlapping operation mode ensures seamless connection between data preparation and decision-making, greatly improving the system's processing throughput and resource utilization.

[0056] This invention also provides a LEO satellite network handover prediction method based on segment-page hierarchical caching. Its core lies in utilizing the segment-page hierarchical geospatial structure to construct a filtering funnel from macro to micro levels, thereby decoupling computational complexity from the total number of users. The entire method can be broken down into the following steps:

[0057] Step 1: Global segment satellite mapping and candidate set generation; such as Figure 3 This stage is the foundation of the entire prediction algorithm, aiming to greatly narrow the computational scope for subsequent precise matching through coarse-grained macro-screening.

[0058] Global Satellite-Segment Mapping: At the beginning of the prediction period, the system first performs spatial discretization on the satellite constellation. It traverses all satellites in the constellation and dynamically assigns each satellite to its corresponding 1000km × 1000km geographical segment based on its real-time geographic location, thereby establishing a dynamic mapping relationship between satellites and segment IDs.

[0059] Candidate Set Generation: When it's necessary to determine serving satellites for an active page (e.g., a 20km x 20km area), the algorithm avoids traversing all satellites. It utilizes the mapping established in the previous step, retrieving only the parent segment containing the page and its surrounding adjacent segments (e.g., forming a 3x3 grid area). All satellites contained within these segments collectively constitute a candidate satellite list for subsequent calculations, such as... Figure 3 As shown.

[0060] Through this stage, only satellites that are physically most likely to provide services to the target area are taken into consideration, thus completing the crucial computational dimensionality reduction.

[0061] Step 2: Page Matching and Switching Request Identification; At the decision-making time (e.g., T+5s), the decision-making process, based on the candidate satellite list generated in Step 1, determines the optimal serving satellite for each page-level grid with users and identifies potential switching requests. This determination process utilizes geospatial partitioning to reduce computational complexity and employs a weighted multi-point service quality assessment mechanism, as follows:

[0062] Sampling point layout and weighting: such as Figure 4Within each page-level grid containing a user, the system employs a nine-point sampling model to deploy nine fixed virtual samples, assigning differentiated weight coefficients w to the sampling points at different locations. i This scientifically reflects the importance of different areas within the page. The center point has the highest weight, followed by the midpoints of the edges, and the corner points have the lowest weight.

[0063] Quality of Service (QoS) Score Calculation: The system calculates a page QoS score Q(s) for each satellite s in the candidate satellite list. This score is a weighted sum of the Received Signal Strength Indications (RSSI) of the satellite across all sampling points, calculated using the following formula:

[0064] ;

[0065] Where, p i It is the i-th sampling point, w i It is its corresponding weight, RSSI(s,p) i () represents the signal strength of satellite s at that point. This formula comprehensively evaluates the overall service capability of the satellite across the entire page-weighted area.

[0066] Optimal satellite election: The system finally selects the group's best serving satellite s∗, which is the satellite with the highest service quality score Q(s);

[0067] Subsequently, the system iterates through all UEs on the page, comparing their currently connected serving satellite with the elected group-optimal serving satellite s∗. If they do not match, the system determines that the UE has a potential handover requirement and generates a Pending Handover Event. This event is a data structure containing the UE identifier, source satellite, target satellite (i.e., group-optimal serving satellite s∗), and initial search interval, which serves as input for subsequent precise calculations. This initial search interval is defined as [t_low, t_high], where t_low represents the start time of the search interval and t_high represents the end time of the search interval. The midpoint time (mid_t) mentioned in subsequent steps is precisely the midpoint of this dynamically updated interval.

[0068] Step 3: Batch Binary Backtracking; After identifying all UEs with handover needs, the system initiates the core batch binary backtracking algorithm to accurately determine the optimal handover time for each UE. This algorithm is an iterative process, and its logic is as follows: Figure 5 As shown, it includes the following steps:

[0069] Step 3.1: Requirements Collection and Integration: At the start of the Gth iteration, the system first traverses all pending handover contexts (i.e., pending handover events), extracts their source satellite, target satellite, and corresponding midpoint time mid_t, and generates two specific orbit prediction task units: <Source Satellite ID, mid_t> and <Target Satellite ID, mid_t>. Different users' handover contexts may need to query the status of the same satellite at the same time, which will generate a large number of prediction task units with identical content.

[0070] To avoid redundant trajectory predictions for these identical task units, the system executes an efficient deduplication and integration process. Specifically:

[0071] The system initializes a hash set specifically for storing unique <satellite ID, mid_t> tuples.

[0072] For each <satellite ID, mid_t> task unit generated in step 2, the system attempts to add it to the hash set. The hash function calculates a unique storage address based on the tuple's content (containing both the satellite ID and the time point). The system can use this address to instantly locate its storage position in the hash table, i.e., the hash bucket. The system only needs to compare within this "bucket" to check if there is an identical item. If it already exists, the addition operation will be automatically ignored. This mechanism avoids global traversal, making the average time complexity of the deduplication operation close to constant time O(1), ensuring extremely high efficiency in the integration phase.

[0073] After traversing all pending switching events, the hash set contains all unique <satellite ID, mid_t> combinations required for this iteration. This set constitutes the satellite-time point combination table for this iteration. It will serve as input for the next batch orbit prediction.

[0074] Step 3.2: Batch Orbit Prediction and Caching; Based on the satellite-time point combination table integrated in the previous step, the system initiates a one-time batch prediction request to the orbit predictor. After receiving this request table, the orbit predictor iterates through each "satellite-time point" combination and accurately calculates the high-precision status of the specified satellite at the specified time. After the calculation is completed, all returned position information will be updated or stored in the shared orbit cache, using the satellite ID and time point as indexes for quick lookup in subsequent steps. Through this batch operation, the orbit predictions that previously needed to be performed independently for each UE and may have a large number of duplicates are efficiently merged into a single batch calculation, thereby significantly reducing the overall computational overhead.

[0075] Step 3.3: Parallel RSSI Comparison and Decision; After all the necessary orbital data has been cached, the algorithm iterates through all "handover contexts" again. For each UE, it retrieves the positions of its current satellite and target satellite at the current countermeasure time from the cache, and calculates two corresponding Received Signal Strength Indication (RSSI) values ​​based on the UE's own geographical location. By comparing the magnitudes of these two RSSI values, the system can determine whether the target satellite's signal is superior to the current satellite's at that midpoint time.

[0076] Step 3.4: Search interval update; Based on the comparison results of the previous step, update the time interval of the binary search for each UE independently.

[0077] If, at time mid_t, the RSSI value of the target satellite is greater than the RSSI value of the current satellite, it indicates that the optimal handover time is earlier than mid_t. Therefore, the algorithm updates the upper limit of the UE's time search interval to mid_t, and the new search interval becomes [t_low, mid_t].

[0078] Conversely, if the target satellite signal is weaker than or equal to the current satellite signal, it indicates that the switch should occur after mid_t or exactly at this moment. The lower limit of the search interval is updated to mid_t, and the new interval becomes [mid_t, t_high].

[0079] Step 4: Iterative Convergence and Event Generation; The system will repeatedly execute the iterations of Phase 3. Each iteration halves the handover time search interval for all UEs, thus gradually approaching the precise handover moment. When the search interval width of a UE is less than a preset accuracy threshold (e.g., 0.01 seconds), or when the preset maximum number of iterations is reached, the search process for that UE is considered converged. Its final determined handover time point will be recorded. After the handover times of all UEs have converged, the system will generate a final handover event list, which includes the UE identifier, source satellite, target satellite, and the handover execution time accurate to the millisecond level, for subsequent network execution.

[0080] Step 5: Unified Anomaly Handling and Robustness Guarantee; This invention designs a unified anomaly handling and rollback mechanism to ensure the absolute reliability of the system. In actual operation, whether the UE selects a non-predicted satellite with better signal due to its location at the page boundary, or due to sudden changes in user trajectory, orbit prediction deviation, or other reasons, any situation that causes the satellite actually accessed by the UE to differ from the core network's prediction result is considered a prediction failure.

[0081] In such cases, since the target satellite being accessed has not received the pre-configured forwarding path from the core network, it will automatically trigger a standard 5G NTN handover procedure, sending control signaling to the core network to request the establishment of a data channel. Upon receiving the signaling, the core network will update the user plane routing, completing a handover. Therefore, the cost of predictive failure is equivalent to the overhead of performing a standard 5G NTN handover. This architecture, which prioritizes predictive optimization and uses standard procedures as a fallback, ensures that the system achieves both extreme performance and carrier-grade stability and reliability.

[0082] This invention employs a self-developed C++ simulation platform for performance verification. The simulation platform includes an `AlgorithmPerformanceStats` structure for recording performance metrics, used to compare key performance parameters of the proposed method (Improved Segment-Page Algorithm) with existing technologies. These parameters include: total computation time, total number of orbit predictions, switching decision success rate, and cache hit rate.

[0083] This invention employs a dual-process parallel architecture, where process 1 is responsible for background pre-caching of track information, and process 2 is responsible for real-time user terminal (UE) handover decisions. Since processes 1 and 2 run in parallel, experimental statistics only record the execution time of process 2 to accurately reflect the response latency of the terminal decision.

[0084] The main parameters of the experimental scenario are as follows: geographical coverage range is latitude -80° to 80° and longitude -170° to 170°; segment-level grid size is 1000km × 1000km; page-level grid size is 20km × 20km; segment-level coverage radius is 2000km; decision cycle is 5s; UE scale is 10,000 to 200,000; random distribution of UEs is 40% (the remaining 60% are urban clustered distribution). Under the above configuration, the comparison results are shown in Table 1. Note that the time data for this method only includes the execution latency of process 2; the pre-cached calculation of process 1 does not affect the real-time response performance.

[0085] Figure 6 The sampled data for the matching results is presented in a table format, showing the final handover prediction data generated by the system for a subset of users. Each row in the table represents the prediction result for one user, where "ue_id" is the user identifier, "satellite_id" is the identifier of the satellite currently providing service to that user, and "predicted_time_s" lists the precise predicted handover time. The sampled data demonstrates that the system can not only calculate the handover time to the second (e.g., 15.0098 seconds) for users requiring a handover, but also determine the status of users with stable connections and no need for handover within the prediction period. This table is a concrete example of the final generated "handover event list."

[0086] Figure 7 The figure shows the performance comparison results. It visually illustrates the comparison of the total computation time of the "LEO satellite network handover prediction method based on segmented page hierarchical caching" of this invention with existing methods under different numbers of users. As can be seen from the figure, although the computation time of both methods increases with the number of users, the growth slope of the original method is much greater than that of the method of this invention. This indicates that the present invention has a significant advantage in computational overhead, verifying its ability to handle large-scale user scenarios more efficiently, solving the problem of strong coupling between computational complexity and user scale, and possessing excellent scalability.

[0087] Table 1: Simulation Comparison Results under Different Methods

[0088] Number of UEs Original method time (ms) Original method prediction number The process time for this method is 2 milliseconds. This method predicts the number of times. 10,000 409.271 394,975 4,147,740 16,384,528 20,000 995.335 835,835 8,220,680 16,445,080 50,000 2,075.760 1,405,480 20,494,920 16,445,964 80,000 3,410.760 2,297,650 33,013,380 16,446,468 100,000 3,749.640 2,767,960 40,818,040 16,446,504 200,000 7,628.720 5,115,140 82,268,980 16,446,758

[0089] The performance advantages and technical effects of this invention are demonstrated through prediction data of the saturation effect after a certain number of trials, as detailed below:

[0090] With a 1000km×1000km segment-level grid, the number of predictions essentially stops increasing after the number of user objects (UEs) reaches 20,000.

[0091] 10,000 UEs: 16,384,528 predictions;

[0092] 20,000 UEs: 16,445,080 forecasts (an increase of 0.37%).

[0093] 50,000 UEs: 16,445,964 predictions (only a 0.005% increase from 20,000 UEs);

[0094] 200,000 UEs: 16,446,758 predictions (only a 0.38% increase from 10,000).

[0095] In a 500km×500km segment-level grid, saturation occurs after approximately 50,000 UEs:

[0096] 10,000 UEs: 50,842,524 predictions;

[0097] 20,000 UEs: 62,585,168 forecasts (an increase of 23.1%);

[0098] 50,000 UEs: 65,646,096 forecasts (a 4.9% increase from 20,000 UEs);

[0099] 200,000 UEs: 65,694,686 forecasts (only a 0.07% increase from 50,000).

[0100] With a 50km×50km segment-level grid, a higher UE scale is required to achieve saturation.

[0101] 10,000 UEs: 110,674,432 predictions;

[0102] 20,000 UEs: 207,298,977 forecasts (an increase of 87.3%);

[0103] 50,000 UEs: 482,735,856 predictions (an increase of 133.0%).

[0104] 200,000 UEs: 1,682,388,640 forecasts (a 248.4% increase from 50,000).

[0105] After reducing the geographic segment size from 1000km to 500km and 50km, the prediction overhead curve of the improved algorithm exhibits a significant characteristic of first growing rapidly and then stabilizing. Specifically, when the number of user equipment (UEs) is small, newly added randomly distributed UEs will activate a large number of unused geographic segments, leading to a significant increase in the number of predictions; however, when the number of UEs reaches a certain scale (approximately 50,000 or more), the available geographic segments in the simulation area are basically covered, and almost all new UEs fall into existing segments. The number of predictions and computational overhead tend to saturate as the number of UEs increases.

[0106] This invention transforms the most expensive trajectory prediction overhead from being linearly correlated with the total number of UEs to being correlated with the spatial distribution range of UEs. This causes the number of predictions to increase with the number of segment activations when the number of UEs is small, and to enter a saturation state after the number of UEs reaches a certain scale, locking the prediction overhead at a stable level. This avoids the explosive growth of the original algorithm in large-scale user scenarios, thus achieving a significant improvement in overall performance and excellent scalability with a controllable pre-computation cost.

[0107] The experimental results above demonstrate that the present invention can significantly reduce the computational load of large-scale satellite-to-ground communication handover prediction, improve decision-making efficiency, and support larger-scale terminal access scenarios while maintaining high prediction accuracy.

[0108] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A LEO satellite network handover prediction system based on segment-page hierarchical caching, characterized in that, The system is a dual-process prediction processor, including an active region maintenance module, a segment satellite mapping module, a high-precision orbit pre-caching module, a page matching module, and a batch binary backtracking module; The prediction process, as process 1, is responsible for prediction and high-precision pre-caching. Among them: the active area maintenance module dynamically updates the list of active segments and active pages according to the location distribution of user equipment (UE) and establishes a parent-child relationship mapping between pages and segments; the segment satellite mapping module establishes candidate satellite mappings for active segments; and the high-precision orbit caching module performs orbit prediction on the candidate satellites identified in the segment satellite mapping and stores the prediction results. The decision-making process, as process 2, is responsible for page-level matching and batch binary search backtracking. Specifically, the page matching module determines the optimal serving satellite for each active page based on the satellite position data provided by the high-precision orbit cache module; the batch binary search backtracking module performs batch binary search on the UEs that need to be switched, determines the switching time point, and outputs the final switching event list, including user identifier, source satellite, target satellite, and time point. The prediction process is activated at the current time T of a decision period N, where T is the prediction time, and obtains the prediction result. The decision process, at the next time T+t, where T+t is the decision time, uses the prediction result data already in the cache to complete page matching, switch recognition, and batch binary backtracking search, and finally outputs the switch result of period N. The high-precision orbit cache module adopts a two-level geographic segmentation structure, dividing the global geographic space into segment-level grids and page-level grids for geographic region management. The segment-level and page-level grids differ by 10-50 orders of magnitude.

2. The LEO satellite network handover prediction system based on segment-page hierarchical caching according to claim 1, characterized in that, At the beginning of each prediction period, the segment satellite mapping module determines a list of potential serving satellites for each active segment and performs coarse screening using segment-level coverage radius. The active segment is the segment with user equipment, and the side length of the segment depends on the current satellite's scanning radius.

3. The LEO satellite network handover prediction system based on segment-page hierarchical caching according to claim 1, characterized in that, The optimal service satellite is determined based on the Received Signal Strength Indicator (RSSI) value.

4. The LEO satellite network handover prediction system based on segment-page hierarchical caching according to claim 1, characterized in that, The batch binary search backtracking module adopts a hierarchical batch processing mechanism. Each iteration includes three stages: demand collection, batch prediction, and decision diversion. In the demand collection stage, prediction demands of all active contexts are collected in a unified manner. In the batch prediction stage, orbit predictions of all unique time point-satellite pairs are completed at once. In the decision diversion stage, the binary search interval is updated based on the RSSI value comparison results.

5. A method for predicting handover in LEO satellite networks based on segmented page hierarchical caching, characterized in that, Includes the following steps: Step 1: Global Segment Satellite Mapping and Candidate Set Generation; A two-level geographic segmentation structure is adopted to divide the global geographic space into segment-level grids and page-level grids for geographic region management. The segment-level and page-level grids differ by 10-50 orders of magnitude. At the beginning of the prediction period, the satellite constellation is first spatially discretized to establish a dynamic mapping relationship between satellites and segment IDs. Using the aforementioned mapping relationship, all satellites in the parent segment of the active page and its adjacent segments are retrieved to obtain a candidate satellite list. Step 2: Page matching and switching requirement identification; Based on the candidate satellite list, for each page-level grid with a UE, its group-optimal serving satellite is determined based on the received signal strength indicator value. Then, all UEs in the page are traversed, and their currently connected serving satellite is compared with the elected group-optimal serving satellite to determine the handover requirement. If a handover is required, a pending handover event is generated for it. Step 3: Batch binary backtracking; After identifying all UEs with handover requirements, start the batch binary backtracking method to determine the handover time point for each UE; Step 4: Iterative convergence and event generation; Batch binary backtracking method iterative convergence, when the UE's search interval width is less than the preset accuracy threshold or the preset maximum number of iterations is reached, the iteration stops, and the final handover event list is generated for subsequent network execution.

6. The LEO satellite network handover prediction method based on segment-page hierarchical caching according to claim 5, characterized in that, Step 2 is described in detail below: Sampling point layout and weighting: Within each page-level grid containing users, a nine-point sampling model is used to deploy nine fixed virtual samples, and differentiated weight coefficients w are assigned to sampling points at different locations. i ; Service quality score calculation; For each satellite s in the candidate satellite list, calculate its page service quality score, which is the weighted sum of the received signal strength indication values ​​of the satellite at all sampling points, as follows: ； Where, p i It is the i-th sampling point, w i It is its corresponding weight, RSSI(s,p) i ) represents the signal strength of satellite s at that point; the satellite with the highest service quality score Q(s) is selected as the group's optimal service satellite s∗; The system iterates through all UEs on the page and compares their currently connected serving satellite with the elected group's best serving satellite s*. If they do not match, the system determines that the UE has a potential handover requirement and generates a pending handover event for it. This event includes the UE identifier, source satellite, target satellite, and initial search interval. The initial search interval is defined as [t_low, t_high], where t_low represents the start time of the search interval and t_high represents the end time of the search interval.

7. The LEO satellite network handover prediction method based on segment-page hierarchical caching according to claim 6, characterized in that, Step 3 is as follows: Step 3.1: Requirements Collection and Integration: At the beginning of the Gth iteration, the system first traverses all pending switching events, extracts their source satellite, target satellite and corresponding midpoint time mid_t, and generates two specific orbit prediction task units: <source satellite ID, mid_t> and <target satellite ID, mid_t>. Initialize a hash set specifically for storing unique <satellite ID, mid_t> tuples. For each generated <satellite ID, mid_t> task unit, the hash function calculates a unique storage address based on the tuple's content. This storage address instantly locates its storage position in the hash table. Simply compare the tuple within the hash table to check if an identical item exists. If it already exists, the current addition operation will be automatically ignored. After traversing all pending switching events, the hash set contains all and unique <satellite ID, mid_t> combinations required for prediction in this iteration. This set constitutes the satellite-time point combination table for this iteration, serving as the input for the next batch orbit prediction. Step 3.2: Batch orbit prediction and caching; Based on the satellite-time point combination table, the system sends a one-time batch prediction request to the orbit predictor; After receiving this request table, the orbit predictor will traverse each satellite-time point combination and calculate the high-precision status of the specified satellite at the specified time. Once the calculation is complete, all returned position information will be updated or stored in a shared orbital cache, indexed by satellite ID and time point for quick lookup in subsequent steps; Step 3.3: Parallel Received Signal Strength Indicator Comparison and Decision; For each UE, retrieve the positions of its current satellite and target satellite at the current countermeasure time from the cache, and calculate two corresponding received signal strength indication values ​​in combination with the UE's own geographical location. By comparing the magnitude of the values, determine whether the signal of the target satellite is better than that of the current satellite at the midpoint time. Step 3.4: Search interval update; Based on the comparison results of the previous step, update the time interval of the binary search for each UE independently; If at time mid_t, the RSSI value of the target satellite is greater than the RSSI value of the current satellite, the upper limit of the time search interval for the UE is updated to mid_t, and the new search interval becomes [t_low, mid_t]. Conversely, if the target satellite signal is weaker than or equal to the current satellite signal, the lower limit of the search interval is updated to mid_t, and the new interval becomes [mid_t, t_high].

8. The LEO satellite network handover prediction method based on segment-page hierarchical caching according to claim 7, characterized in that, The method also includes anomaly handling and rollback mechanisms; in actual operation, whether it is because the UE is located at the page boundary and selects a non-predicted satellite with better signal, or because of sudden changes in user trajectory or orbit prediction deviation, any situation that causes the satellite actually accessed by the UE to be inconsistent with the prediction result of the core network is regarded as prediction failure.

Images