Beam guiding optimization method based on position prediction in unmanned aerial vehicle communication

Abstract

The invention discloses a beam guidance optimization method based on position prediction in unmanned aerial vehicle communication, and relates to the technical field of unmanned aerial vehicle communication optimization, and the method comprises the steps: obtaining the flight path data of an unmanned aerial vehicle, carrying out the adaptive segmentation according to the flight path data, and constructing a state evolution model for each flight path; performing short-term prediction and medium-term residual error calibration prediction according to the state evolution model, and constructing a prediction correction term to generate a prediction point set; mapping the prediction point set to a direction coordinate system of a ground communication array to construct an angle evolution tensor, and screening to obtain a candidate beam direction set; and screening a standby beam direction with the maximum path coverage redundancy from the candidate beam direction set. According to the method, a beam guiding optimization mechanism with state identification, adaptive prediction, space tolerance, historical correction and multi-path dynamic regulation and control capabilities can be realized, and the beam control precision, the failure recovery capability and the communication link stability in a complex airspace motion state during communication of the unmanned aerial vehicle are remarkably improved.

Description

Technical Field

[0001] This invention relates to a beam guidance optimization method based on position prediction in unmanned aerial vehicle (UAV) communication, belonging to the field of UAV communication optimization technology. Background Technology

[0002] With the widespread application of drones in fields such as inspection and monitoring, emergency communication, and logistics delivery, higher requirements are placed on the stability, real-time performance, and spatial coverage of communication links in dynamic airspace environments. Especially in scenarios using high-frequency bands (such as millimeter waves) for long-distance, high-speed data transmission, the system typically relies on beamforming technology to form a high-gain directional signal in space by adjusting the phase and direction of the transmitting array, thereby achieving precise aiming at drone targets and stable communication.

[0003] In existing technologies, beam-guided control methods mainly rely on two types of information input: one is the channel state information transmitted back in real time from the terminal, and the other is the position information reported by the UAV's own positioning system (such as GPS or IMU). The former is used to periodically evaluate link quality and adjust beam direction, while the latter can be used to estimate the current spatial position and expected direction of motion of the UAV, and perform static beam alignment or select the optimal direction from the beam codebook accordingly.

[0004] In recent years, to improve the foresight and timeliness of beamguide, some studies have begun to introduce position prediction mechanisms based on historical trajectories. These mechanisms estimate the spatial position of the UAV in future time slots by constructing trajectory prediction models, and then combine this with base station array structure to estimate beam angles, thereby shortening beam switching latency and reducing blind scan overhead. However, these prediction methods still have many limitations in scenarios involving high-speed flight, multiple disturbances, and frequent obstructions. On the one hand, traditional trajectory prediction models are mostly based on linear regression, constant-speed motion assumptions, or sliding window extrapolation strategies, which cannot effectively model the real motion trajectory of UAVs that are non-uniform, irregular, and have sudden maneuvering characteristics. Especially in flight states such as sudden turns, abrupt stops, or vertical climbs, prediction errors accumulate significantly, leading to severe beam pointing deviations. On the other hand, existing beamguide control mostly adopts a single beam selection mechanism, lacking dynamic tolerance for prediction errors and spatial redundancy configuration strategies. This makes the communication link highly sensitive to offset mismatches, easily causing communication interruptions when the prediction error exceeds the main lobe coverage area. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a beam guidance optimization method based on position prediction in UAV communication. This method achieves precise control and robust adaptation of beamforming direction, maintains spatial redundancy of beam coverage even when prediction errors exist, and dynamically fine-tunes the beam direction configuration of the next time slot based on historical link failure feedback information. It constructs a closed-loop optimization path from prediction to compensation to feedback and finally to correction, effectively improving the stability of communication links and spatial orientation accuracy of UAVs in highly dynamic flight conditions, enhancing the system's tolerance to prediction offsets and directional disturbances, and is suitable for air-to-ground communication environments with high beam alignment requirements, such as millimeter wave and terahertz waves.

[0006] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0007] On one hand, this invention provides a beam steering optimization method based on position prediction in unmanned aerial vehicle (UAV) communication, comprising:

[0008] Acquire flight trajectory data of the drone during a continuous communication cycle;

[0009] The trajectory is adaptively segmented based on the flight trajectory data of the UAV during the continuous communication cycle, and a state evolution model is constructed for each segment of the flight trajectory.

[0010] Short-term and medium-term residual calibration predictions are performed based on the state evolution model of each flight trajectory segment, and prediction correction terms are constructed to generate a prediction point set;

[0011] The predicted point set is mapped to the directional coordinate system of the ground communication array to construct an angle evolution tensor, and the candidate beam direction set is obtained by filtering based on the angle evolution tensor;

[0012] The backup beam direction with the maximum path coverage redundancy is selected from the candidate beam direction set to achieve dynamic beam switching and link stability enhancement.

[0013] Furthermore, the flight trajectory data of the UAV during the continuous communication cycle is collected after the UAV and the ground communication array establish a synchronized time reference. The flight trajectory data includes position, speed and attitude information. The position includes three-dimensional spatial position and the UAV's own body coordinate orientation. The speed includes a speed vector. The attitude information includes roll angle, pitch angle and heading angle. The three-dimensional spatial position is based on a fixed ground reference coordinate system.

[0014] Furthermore, the adaptive segmentation of the trajectory based on the UAV's flight trajectory data within a continuous communication cycle, and the construction of a state evolution model for each flight trajectory segment, includes:

[0015] Construct a flight trajectory sequence matrix, wherein the flight trajectory sequence matrix is ​​flight trajectory data under multiple consecutive timestamps;

[0016] Generate velocity change rate curves, heading change angle sequences, and attitude stability parameters based on the flight trajectory sequence matrix;

[0017] Traverse the velocity change rate curve, heading change angle sequence, and attitude stability parameters, and compare them with the velocity change rate threshold, heading change angle threshold, and attitude stability threshold, respectively. If at least one of them exceeds the corresponding threshold and the duration meets the minimum duration, segmentation is triggered, and each duration is taken as a flight trajectory segment.

[0018] For each flight trajectory segment, if the high-value regions of the velocity change rate curve, heading change angle sequence, and attitude stability parameters in the flight trajectory segment overlap in time or any two of them appear consecutively alternately, then the flight trajectory segment is marked as a disturbance segment; otherwise, it is marked as a stable segment. The region in the flight trajectory segment where the heading change angle sequence exceeds the heading angle threshold and the duration is not less than the minimum duration is defined as the heading change region, and the start and end times, peak change angle, and interval between adjacent regions of the heading change region are recorded.

[0019] A corresponding state evolution model is constructed for each flight trajectory segment based on the disturbance segment, the stable segment, and the region of sudden change in direction.

[0020] Furthermore, the process of performing short-term prediction and medium-term residual calibration prediction for each flight trajectory segment, and constructing prediction correction terms to generate a prediction point set, includes:

[0021] S1. Obtain the velocity change range, heading change range, and attitude stability baseline from the state evolution model corresponding to the flight trajectory segment, and calculate the instantaneous state change trend by combining the flight trajectory data of the current timestamp and the previous timestamp. The instantaneous state change trend includes the rate of change of velocity vector angle, the magnitude of heading angle change, and the real-time evaluation results of attitude stability parameters.

[0022] S2. Using the velocity vector direction at the current timestamp as the reference direction, determine the direction adjustment direction by referring to the heading angle change range, and determine the single-step displacement range based on the velocity change range.

[0023] S3. Combine the direction adjustment and single-step displacement amplitude to obtain the short-term predicted position of the current timestamp;

[0024] S4. Repeat S1~S3 to obtain the short-term predicted position of the flight trajectory segment for all timestamps within the continuous communication period.

[0025] S5. For a timestamp, the spatial offset vector between the short-term predicted position of the flight trajectory segment and the actual position of the timestamp in the previous communication cycle is used as the offset residual of the timestamp.

[0026] S6. Combine the offset residuals of all timestamps into an offset residual sequence in chronological order, and perform direction consistency inspection and amplitude evolution inspection on the offset residual sequence to obtain the inspection results.

[0027] S7. Construct a prediction correction term based on the inspection results, and use the prediction correction term to correct the short-term prediction position of the timestamp in the current communication cycle to obtain the medium-term corrected prediction position.

[0028] S8. Repeat S5~S7 to obtain the mid-term corrected predicted positions of all timestamps. Then, associate the timestamps with their mid-term corrected predicted positions, position tolerance intervals, corresponding trajectory segments, correction source labels, and direction change regions as prediction points. Combine all prediction points to obtain the prediction point set for the flight trajectory segment.

[0029] Furthermore, the position tolerance range includes a radial tolerance boundary and an angular tolerance boundary. The radial tolerance boundary is determined by attitude stability parameters, and the angular tolerance boundary is determined by the heading angle variation amplitude, the velocity vector angle variation rate, and the peak change angle and start and end time in the direction change region.

[0030] The construction of prediction correction terms based on the inspection results includes:

[0031] The direction of the forecast correction term is set from the short-term forecast position to the actual position;

[0032] When the direction of the offset residual is consistent and the amplitude shows a continuous diverging trend, the amplitude is limited to the upper limit of the allowable displacement correction.

[0033] When the direction of the offset residual alternates and the amplitude fluctuates back and forth, the amplitude of the prediction correction term is limited to the 40-60% amplitude range between the upper limit of the allowable displacement correction and the lower limit of the allowable displacement correction.

[0034] When the magnitude of the offset residual shows a gradual convergence trend, the magnitude of the prediction correction term is limited to no more than 10% of the lower bound of the allowable displacement correction.

[0035] The upper and lower bounds of the allowable displacement correction are derived based on the allowable velocity variation range and the attitude stability baseline.

[0036] Furthermore, the step of mapping the predicted point set to the directional coordinate system of the ground communication array to construct an angle evolution tensor, and filtering the candidate beam direction set based on the angle evolution tensor, includes:

[0037] A directional coordinate system is constructed with the array phase center of the ground communication array as the origin;

[0038] The predicted points are mapped one by one to azimuth and pitch angles. The azimuth angle is obtained by comparing the horizontal projection direction of the array phase center to the predicted point with the ground horizontal reference axis as the starting reference direction. The pitch angle is obtained by comparing the angle between the three-dimensional direction of the array phase center to the predicted point and the horizontal plane with the ground horizontal plane as the reference plane.

[0039] An angle evolution tensor is constructed for each prediction point. Local perturbation intensity index is established in each time slice of the angle evolution tensor, and prediction confidence is calculated based on the corrected source label, the association label of the direction change region, and the attitude stability parameter.

[0040] An initial direction vector sequence is generated based on the angle evolution tensor, and direction continuity detection is performed. The initial direction vector sequence that meets the detection conditions is retained as the direction sequence.

[0041] The direction sequence is mapped to the beam codebook of the ground communication array, and the direction sequence closest to the beam center of the codebook is selected as the candidate beam direction sequence. The available fine-tuning margin of the candidate beam direction sequence within the angular tolerance boundary is retained.

[0042] Calculate the matching priority factor and stability confidence factor for each candidate beam direction sequence;

[0043] The matching priority factor and stability confidence factor are compared with the matching priority factor threshold and stability confidence factor threshold, respectively. Candidate beam direction sequences with matching priority factors not lower than the matching priority factor threshold and stability confidence factors not lower than the stability confidence factor threshold are selected to form a candidate beam direction set.

[0044] Furthermore, the step of selecting the backup beam direction with the maximum path coverage redundancy from the candidate beam direction set includes:

[0045] Acquire beam configuration and channel quality feedback data of the UAV in the current and historical communication cycles;

[0046] Based on the beam configuration and channel quality feedback data of the UAV in the current and historical communication cycles, a beam offset residual path map is constructed for the candidate beam direction set.

[0047] The offset correction vector is extracted from the beam offset residual path map, and the offset correction vector is used to compensate and adjust the candidate beam direction set in the next communication cycle to obtain the candidate beam direction set for compensation and adjustment.

[0048] A redundant path map is constructed based on the candidate beam direction set for compensation control;

[0049] A candidate beam direction is selected from the redundant path diagram as the starting node of the main communication path, and adjacent nodes are marked on the redundant path diagram to form an initial set of backup path sequences.

[0050] During the operation of the main communication path, beam switching control is triggered based on the communication quality degradation trend to determine the backup beam direction with the maximum path coverage redundancy.

[0051] Furthermore, the channel quality feedback data includes a signal-to-noise ratio sequence, a received power fluctuation sequence, a bit error rate change rate, and a delay jitter indication;

[0052] The method for constructing the beam offset residual path map includes:

[0053] Based on the azimuth, elevation, effective time period, timestamp, and trajectory segment of the candidate beam direction sequence, communication quality degradation sections and beam failure events are marked using channel quality feedback data, including:

[0054] Within the minimum observation window, if the signal-to-noise ratio sequence shows a continuous decreasing trend and the received power fluctuation sequence continues to decrease, or the bit error rate change rate continuously increases at adjacent observation points in the candidate beam direction sequence, it is marked as a communication quality degradation segment.

[0055] If the main lobe coverage time period of a candidate beam direction sequence coincides with the communication quality degradation section, and the candidate beam direction sequence cannot maintain the stability confidence factor above the stability confidence factor threshold within the angular tolerance boundary, then the candidate beam direction sequence is determined to be a beam failure event.

[0056] For each beam failure event, the actual effective direction is estimated by the direction of the line connecting the center of the ground communication array to the predicted position after mid-term correction of the timestamp corresponding to the beam failure event.

[0057] The actual effective direction estimate is compared with the beam center direction at the time of failure by angular difference, and a direction offset rate sequence is formed in time order within a fixed-length time window;

[0058] The beam offset residual path diagram is obtained from the timestamp, beam center direction, communication failure marker, and direction offset rate sequence.

[0059] The step involves extracting an offset correction vector based on the beam offset residual path map, and then using the offset correction vector to compensate and adjust the candidate beam direction set in the next communication cycle, resulting in a compensated and adjusted candidate beam direction set, including:

[0060] The path segments that match the candidate beam set are retrieved from the beam offset residual path map. The matching conditions are that the trajectory segments are consistent, the direction change region is consistent, and the timestamps are adjacent or the time interval between the timestamps is within the preset time alignment tolerance.

[0061] The inspection results are obtained by inspecting the path segments for directional consistency and amplitude evolution.

[0062] Construct an offset correction vector based on the inspection results:

[0063] If the direction of the path segments is consistent and the amplitude continues to increase, the direction of the offset correction vector is set to the candidate beam center direction pointing to the actual effective direction estimate, and the amplitude is limited to the upper bound of the angular correction.

[0064] If the direction of the path segment alternates and the amplitude fluctuates repeatedly, the amplitude of the offset correction vector is limited to 40-60% of the amplitude range between the upper bound and the lower bound of the angular correction.

[0065] If the directional offset rate of a path segment continues to decrease, the direction of the offset correction vector is set to fine-tuning, and the amplitude is limited to a very small amplitude.

[0066] In the next communication cycle, the offset correction vector is used to compensate and adjust the candidate beam direction set to obtain the compensated and adjusted candidate beam direction set.

[0067] The upper and lower bounds of the angular correction are derived based on the angular tolerance boundary of the time slice corresponding to the candidate direction sequence, the local perturbation intensity of the angular evolution tensor, and the prediction confidence.

[0068] Furthermore, the construction of the redundant path map based on the candidate beam direction set for compensation control includes:

[0069] Candidate beam direction entries are used as nodes in the redundant path graph, and the main lobe overlap relationship between two nodes is used as an edge.

[0070] When the main lobe coverage of two candidate beam directions has a spatial intersection at the same timestamp and the continuous length of the intersection on the time axis reaches the minimum switching window, a directed edge is constructed between the corresponding nodes of the two candidate beam directions.

[0071] Calculate the complementary structural strength of each side to obtain the redundancy path diagram;

[0072] The nodes corresponding to the candidate beam directions with high stability confidence factors and high matching priority factors are taken as the starting nodes of the main communication path, and the nodes directly connected to the main communication path with strong or medium complementary strength of the connecting edge are taken as adjacent nodes.

[0073] During the operation of the main communication path, beam switching control is triggered based on the communication quality degradation trend to determine the backup beam direction with the maximum path coverage redundancy, including:

[0074] During the operation of the main communication path, the signal-to-noise ratio sequence, bit error rate change rate, and received power fluctuation sequence are continuously acquired and aligned.

[0075] When, within the minimum observation window, the signal-to-noise ratio (SNR) sequence shows a continuous downward trend and the received power fluctuation sequence continues to decrease, or the bit error rate (BER) change rate continuously increases at adjacent observation points, and simultaneously the SNR sequence, BER change rate, and received power fluctuation sequence remain consistent and the slope sign is stable at most observation points, then a communication quality degradation trend is confirmed and beam switching control is triggered. The beam switching control includes:

[0076] Taking the starting node of the main communication path in the redundant path graph as the starting point, first search for the edge that is connected to the starting point and has complementary direction and strong structure strength, and then search for the adjacent node pointed to by the edge. The starting point and the adjacent nodes form a candidate sequence of backup beam directions arranged in the order of timestamps.

[0077] The backup beam direction is selected from the candidate sequence of backup beam directions in order of the following criteria: longest continuous coverage time within the minimum switching window, earliest expected effective time, smallest minimum switching distance with the main communication path, and best match with the current direction change area marker.

[0078] Furthermore, the strength of the directional complementary structure is divided into strong, medium, and weak levels, wherein,

[0079] When the main lobes of the two candidate beam directions overlap throughout the entire period of the minimum switching window, and the angle difference between the two candidate beam directions is within the available fine-tuning margin, and the stability confidence factors are not lower than the stability confidence factor threshold, the strength of the complementary direction structure is strong.

[0080] When the main lobes of the two candidate beam directions overlap for more than 50% of the time period of the minimum switching window, or when the angle difference between the two candidate beam directions is at the boundary value of the available fine-tuning margin, and at the same time the stability confidence factor drops to the stability confidence factor threshold boundary for a short time, the strength of the complementary structure is at a medium level.

[0081] When the main lobe overlaps between two candidate beam directions within less than 10% of the minimum switching window, the angle difference between the two candidate beam directions exceeds the available fine-tuning margin, or the stability confidence factor does not meet the threshold, the strength of the directional complementary structure is low.

[0082] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0083] This invention achieves a comprehensive improvement in the predictability, fault tolerance, and adaptive capability of beam pointing control under the dynamic flight state of UAVs by constructing a five-level closed-loop control chain consisting of trajectory segmentation identification, residual correction prediction, orientation tensor modeling, failure feedback correction, and redundant path switching.

[0084] This invention introduces velocity change rate, heading change angle sequence, and attitude stability parameters into the trajectory sequence as segmentation criteria to construct differentiated state evolution models for stable and disturbed segments. This enables the system to identify and respond to nonlinear flight characteristics and provides semantic constraints and structural priors for subsequent predictions, thereby significantly reducing the accumulation rate of prediction errors. In the position prediction stage, a short-term state trend-driven and medium-term residual correction mechanism is introduced to avoid equal-weighted dependence on historical trajectories. Through nonlinear modeling of the residual trend direction and amplitude, a prediction point set with position tolerance interval is constructed, enabling beamguide to have spatial redundancy and anti-migration capability, effectively alleviating the trajectory misalignment problem caused by high-speed flight.

[0085] In the direction mapping and beam selection process, this invention constructs an angle evolution tensor to express the evolution trend of the azimuth and elevation angles of the prediction point on the time axis. Combined with direction continuity detection and direction abrupt change region constraints, it selects direction sequences with continuous main lobe coverage capability. Furthermore, based on the dual constraints of matching priority factors and stability confidence factors for candidate directions, a candidate beam direction set is established, significantly improving the beam selection accuracy and execution feasibility of the system under conditions of coexisting prediction errors and direction disturbances. On this basis, a beam offset residual path graph and offset correction vector construction mechanism are introduced to perform time-series structural modeling of beam failure events and direction offset trends in historical communication cycles, and use this model as fine-tuning control. The instruction-driven candidate direction correction for the new cycle realizes a closed loop of reverse adjustment of the guidance strategy by communication behavior, which strengthens the emergency correction capability of the system when unexpected directional disturbances occur. Furthermore, by constructing a redundant path map and a direction complementary structure strength discrimination mechanism, this invention identifies direction pairs with redundant switching relationships in the candidate beam direction set, and performs backup beam direction selection and dynamic switching control when communication quality degradation trends occur. This enables the main communication path to smoothly exit and switch to an acceptable alternative path when unpredictable environmental changes or error exceedance occur. This realizes the integration of beam guidance from predictive control to multi-path redundancy scheduling, effectively improving link continuity and communication reliability.

[0086] This invention enables a beam guidance optimization mechanism with state recognition, adaptive prediction, spatial tolerance, historical correction, and multi-path dynamic control capabilities in UAV air-to-ground communication environments characterized by high-speed maneuverability, frequent directional disturbances, and unavoidable channel feedback delays. It solves the core problems of traditional methods, such as "large prediction errors, difficulty in directional alignment, delayed correction after failure, and lack of structural redundancy in switching," significantly improving beam control accuracy, failure recovery capability, and communication link stability in complex airspace motion states during UAV communication. It has broad engineering application value and deployment feasibility. Attached Figure Description

[0087] Figure 1 This is a flowchart illustrating a beam guidance optimization method based on position prediction in UAV communication, as described in one embodiment of the present invention. Detailed Implementation

[0088] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0089] like Figure 1 As shown, this embodiment of the invention provides a beam guidance optimization method based on position prediction in UAV communication, including the following steps:

[0090] Step 1: Trajectory Sequence Acquisition and Segmentation Modeling

[0091] After the UAV establishes a synchronized time reference with the ground communication array, flight trajectory data, including position, velocity, and attitude information, is continuously collected. Position includes three-dimensional spatial position and the UAV's own body coordinate orientation; velocity includes the velocity vector; and attitude information includes roll angle, pitch angle, and yaw angle. The three-dimensional spatial position is based on a fixed ground reference coordinate system.

[0092] In this embodiment, to ensure data comparability, time synchronization, coordinate reference system 1, and outlier removal need to be completed:

[0093] Time synchronization is calibrated using a reference clock unified with the communication array. The coordinate reference system uses a fixed ground reference system to represent the three-dimensional spatial position, and records the UAV's own body coordinates for reference system conversion when necessary. Outlier removal is based on continuity and physical reachability; if abrupt changes occur between adjacent samples that are inconsistent with the flight platform's capabilities, linear interpolation is performed using stable data from the neighboring time period. The UAV's own body coordinates use a body coordinate system, which can be converted to and from the fixed ground reference system via Euler angles in the ZYX order.

[0094] Subsequently, a trajectory sequence matrix is ​​constructed, where each row corresponds to an observation entry with a single timestamp, and the observation entry represents the data for each flight trajectory. Based on the flight trajectory sequence matrix, velocity rate of change curves, heading change angle sequences, heading rate of change, and attitude stability parameters are generated.

[0095] The velocity change rate curve is calculated by comparing the velocity magnitudes of adjacent time stamps with their corresponding time intervals, sequentially calculating the rate of velocity change and forming a sequence that evolves over time. The heading angle change sequence is obtained by comparing the heading angle differences of adjacent time stamps and combining them with adjacent time intervals to reflect the magnitude of heading angle changes within a short period. The heading change rate is obtained by performing angle unwrapping (eliminating jumps across ±180°) on the heading angles of adjacent time stamps, followed by finite difference analysis at a fixed time step. Median filtering or exponential smoothing is used within a short window to suppress noise if necessary. Attitude stability parameters are obtained by statistically analyzing the range of roll, pitch, and heading angle changes and the number of consecutive exceedances within a fixed-duration sliding window. The range of changes characterizes the jitter amplitude, and the number of consecutive exceedances characterizes the jitter frequency. These two types of statistics together constitute the attitude stability parameters.

[0096] In this embodiment, to ensure the feasibility of the parameter thresholds, stable flight samples are collected during a short-term test flight phase before the mission. Based on the upper bounds of the distributions of the rate of change of velocity, heading angle difference, rate of change of heading, and attitude angle fluctuations in the stable flight samples, and with a safety margin, the thresholds for the rate of change of velocity, the angle of abrupt heading change, the rate of change of heading, and attitude stability are determined. During the actual mission execution, these thresholds can be adjusted in a bounded manner according to the strength of the external wind field and the flight phase.

[0097] After completing the trajectory sequence matrix construction and threshold setting, adaptive segmentation and non-stationary motion recognition are implemented. In this embodiment, adaptive segmentation is performed by combining sliding window traversal and hysteresis control.

[0098] Within any window, if the velocity rate of change curve exceeds the velocity rate of change threshold for several consecutive samples, or the heading change angle sequence exceeds the heading change angle threshold for several consecutive samples, or the heading rate of change exceeds the heading rate of change threshold for several consecutive samples, or the attitude stability parameters indicate that both the jitter amplitude and jitter frequency are higher than the attitude stability threshold, then a segment boundary candidate point is triggered. To avoid excessive segment fragmentation due to short-term noise, hysteresis control is used, requiring the triggering condition to meet a minimum duration before the segment boundary can be confirmed.

[0099] After determining the segment boundaries, a non-stationary motion identification function is applied to each flight trajectory segment. The determination method for the identification function is disclosed as follows:

[0100] If, within a given segment, high-value regions of the velocity change rate curve, the heading change angle sequence, the heading change rate, and the attitude stability parameter occur simultaneously and overlap temporally, or if any two types of high-value regions alternate continuously within a short period, then the segment is marked as a disturbance segment (i.e., a non-stationary motion segment). If none of the aforementioned high-value regions appear, or only appear briefly and without persistence, then the segment is marked as a stable segment. Simultaneously, the identification of directional change regions is performed within the segment. The identification method involves searching for continuous intervals in the heading change angle sequence that exceed the heading change angle threshold and whose duration is not less than the minimum duration. These continuous intervals are designated as directional change regions, and the start and end times, peak change angle, and intervals between adjacent regions are recorded for subsequent steps in candidate beam direction selection and redundancy layout.

[0101] After the above processing, the flight trajectory segment is divided into a stable segment and a perturbation segment connected in sequence, and each segment is accompanied by data on the direction change region, providing a complete prior for subsequent dynamic position prediction, direction space mapping and beam candidate domain generation.

[0102] Next, an independent state evolution model is constructed for each flight trajectory segment. The state evolution model explicitly discloses three aspects: the velocity trend within the segment, the heading change trend, and the attitude stability baseline (the heading change angle is unpredictable and is therefore not considered in the modeling):

[0103] For the stable segment, the velocity trend is mainly characterized by a smooth and gradual change. The model provides the gradual direction of velocity change over time and the allowable range of small fluctuations on the time axis. The heading change trend is mainly characterized by a monotonically gradual change. The model provides the gradual direction of heading angle shift over time and the maximum acceptable short-term deviation. The attitude stability baseline is mainly characterized by small-amplitude jitter. The model records the typical fluctuation range of roll angle, pitch angle, and heading angle within the sliding window to limit the allowable range of attitude changes for subsequent predictions.

[0104] For the disturbance segment, the velocity trend allows for obvious acceleration or deceleration segments, and the model marks the start and end positions and transition duration of acceleration and deceleration on the time axis; the heading change trend allows for abrupt directional changes, and the model links each abrupt directional change region to an event on the time axis, recording the peak angle of the abrupt change and the transition duration to a stable offset; the attitude stability baseline allows for larger jitter amplitude and higher jitter frequency within the disturbance segment, and the model provides the acceptable range of attitude changes by using an upper limit constraint.

[0105] Step 2: Dynamic position prediction based on offset tolerance constraints:

[0106] Within a continuous communication cycle, the currently processed flight trajectory segment is traversed using a unified time reference. The instantaneous state change trend, allowable speed variation range, allowable heading variation range, and attitude stability baseline are obtained from the state evolution model corresponding to the flight trajectory segment.

[0107] The instantaneous state change trend includes the rate of change of the velocity vector angle, the amplitude of the heading angle change, and the real-time evaluation results of attitude stability parameters. The rate of change of the velocity vector angle refers to the change in the angle between velocity vectors at adjacent timestamps per unit time, used to characterize the turning speed of the velocity direction. The amplitude of the heading angle change is the increase or decrease of the heading angle after angle unwrapping in a very short time, used to reflect the instantaneous change in heading direction. The real-time evaluation results of attitude stability parameters are given an immediate "acceptable / boundary / unacceptable" rating within a short window based on the range of changes in roll angle, pitch angle, and heading angle, and the number of consecutive exceedances.

[0108] Execute short-term forecasts:

[0109] Using the velocity vector direction at the current timestamp as the reference direction, the direction of fine-tuning is determined by referring to the variation in heading angle. The upper and lower bounds of the "single-step displacement amplitude" are determined based on the allowable velocity variation range given by the state evolution model. If the traversed timestamp falls near the beginning of a direction abrupt change region, the direction is gradually adjusted in multiple small steps according to the peak abrupt change angle and transition duration recorded in the state evolution model, without making a large-scale turn all at once, ensuring the continuity and controllability of the direction adjustment. After completing the joint determination of the direction and amplitude, the short-term predicted position corresponding to that timestamp is obtained.

[0110] Simultaneously, to preserve the basis for generating the short-term predicted position for subsequent corrections, the real-time evaluation results of the velocity vector angle change rate, heading angle change magnitude and attitude stability parameters, as well as the velocity vector direction at the current moment, are recorded together with the velocity vector direction at the current moment after the short-term predicted position for this timestamp is calculated.

[0111] After obtaining the short-term predicted position, the intermediate residual correction prediction process begins. The offset residual sequence and residual trend are established to construct the prediction correction term. It should be noted that intermediate correction is not required in the first communication cycle, and it begins from the second communication cycle.

[0112] The offset residual is defined as the spatial offset vector between the short-term predicted position and the actual position at the same timestamp within the previous communication cycle. The actual position refers to the fused positioning result obtained by the positioning and attitude acquisition module outputting the data and recursively estimating it through state observations (e.g., error state Kalman filtering or equivalent inertial navigation / satellite navigation tightly coupled fusion) under the same unified time reference. The sensor source can include global satellite positioning. This spatial offset vector is stored in the offset residual sequence buffer and maintained in chronological order.

[0113] In this embodiment, to adapt to sampling jitter, a time alignment rule is adopted, using the current timestamp as the key for exact matching; if no exact match is found, nearest neighbor matching is used within the time alignment tolerance Δt; if there is a predicted sample before and after the current sample and both are within Δt, linear interpolation can be performed between these two samples to obtain the aligned predicted position; if neither condition is met, the unaligned sample is marked, no residual is generated at this moment, and the intermediate correction is skipped. All successfully aligned predicted positions and their timestamps are written together into the offset residual sequence buffer.

[0114] The offset residuals are examined for directional consistency and amplitude evolution to obtain the results. Directional consistency is examined to determine whether the direction of the offset residual vector has a stable consistency over the most recent samples; amplitude evolution is examined to determine whether the amplitude of the offset residual vector gradually converges, gradually diverges, or exhibits alternating fluctuations over the most recent samples.

[0115] Construct prediction correction terms based on the inspection results:

[0116] The direction of the prediction correction term is set from the short-term predicted position to the actual position. When the direction is consistent and the amplitude shows a continuous diverging trend, the amplitude of the prediction correction term is limited to within the upper limit of the displacement correction. When the direction alternates and the amplitude fluctuates between the upper and lower limits, the amplitude of the prediction correction term is limited to a medium amplitude range (40~60%) between the upper and lower limits of the displacement correction to avoid excessive overshoot. When the offset residual shows a gradual convergence trend, only the direction fine adjustment is retained and the amplitude is set to a very small amplitude, that is, the amplitude of the prediction correction term is limited to no more than 10% of the allowable lower limit of the displacement correction to prevent new disturbances to the already stabilized short-term prediction.

[0117] The above upper and lower bounds of the amplitude are derived from the allowable range of velocity variation and attitude stability baseline given by the state evolution model for this trajectory segment: in the stable segment, the upper and lower bounds of displacement correction are relatively compact; in the perturbation segment, the upper and lower bounds of displacement correction are appropriately widened.

[0118] In this embodiment, if the current timestamp is located in the direction change region, the prediction correction term must not disrupt the transition rhythm of the direction change region. The system only makes fine adjustments within the direction band allowed by the transition rhythm to prevent the correction action from forcibly pulling the direction back to the direction before the change. The prediction correction term is superimposed on the short-term prediction position to obtain the mid-term corrected prediction position.

[0119] For each mid-term corrected predicted position, a position tolerance interval is calculated and attached. The position tolerance interval is composed of radial tolerance boundary and angular tolerance boundary, which is used to provide spatial redundancy and direction selection prior in subsequent candidate beam direction selection and redundant path construction.

[0120] The radial tolerance boundary is determined based on the attitude stability parameters: a smaller allowable range of radial fluctuations is taken in the stable segment to cover the displacement uncertainty caused by positioning noise and small attitude jitters; in the disturbed segment, the upper limit of jitter amplitude and acceleration / deceleration transition time recorded by the state evolution model are appropriately widened to cover the displacement uncertainty caused by gusts, control corrections and acceleration / deceleration; if it is in the starting or ending neighborhood of the direction change region, a wider radial tolerance is maintained within the transition time recorded in the direction change region, and after the transition ends, it is gradually narrowed back to the level of the stable segment according to the recovery rhythm of the state evolution model.

[0121] The angular tolerance boundary is determined by the magnitude of the change in the heading angle and the rate of change of the angle between the velocity vectors: when both are small and the attitude stability baseline shows small fluctuations, the angular tolerance is narrowed to improve the screening directionality; when either indicator continues to rise or a change in direction event is detected, the angular tolerance is widened according to the peak change angle and start and end time of the event to cover the uncertainty of short-term heading adjustment, and gradually narrows according to the recovery rhythm after the transition ends.

[0122] Finally, the predicted position after mid-term correction, radial tolerance boundary, angular tolerance boundary, prediction timestamp, trajectory segment identifier, correction source label, and directional change region association marker are uniformly encapsulated into a prediction point record and written into the prediction point set. This ensures that the prediction point set contains both the position result and the tolerance prior, facilitating subsequent steps to directly select candidate beam directions and construct redundant paths based on the tolerance range. The correction source label indicates the specific type of mid-term residual correction (uniform directional divergence / alternating directional fluctuation / gradual convergence), and the directional change region association marker indicates the relationship between the prediction point and the directional change region (whether it belongs to the directional change region).

[0123] The second step of data processing is as follows:

[0124] First, make minor course adjustments: at the timestamp At this point, the current position vector is denoted as The velocity vector is denoted as The time step is recorded as The heading angle is denoted as , For timestamp The heading angle at that point, the heading increment is The system rotates the velocity vector around the vertical axis by a small angle increment according to the heading to obtain the fine-tuned velocity vector. Its calculation formula is ,in Let be the rotation matrix about the vertical axis. The difference in heading angle between adjacent timestamps. and All are three-dimensional vectors; in obtaining Then, press The short-term extrapolated position is obtained, and the heading fine-tuning is broken down into multiple small steps in the neighborhood of the direction change region to ensure continuous and controllable direction changes. Then, residual correction is performed: a sequence of offset residuals is maintained, where the offset residual is the spatial offset vector between the predicted and actual positions at the same timestamp. The correction magnitude is determined by the consistency of direction and the evolution of magnitude of the most recent residuals, and is limited to the upper bound of displacement correction allowed by the state evolution model, resulting in the predicted position after mid-term correction. To support subsequent directional spatial mapping, the array phase center is used as the reference point. For reference, construct in timestamp The pointing vector at the location And calculate the azimuth and elevation angles from the pointing vector: , ,in The three-dimensional coordinate vector of the array phase center. For timestamp The three-dimensional vector pointing from the array phase center to the predicted position. They are respectively The three components, For timestamp The magnitude of the pointing vector at that point. For timestamp The azimuth of the horizontal plane at that location. For timestamp The vertical pitch angle at the location; the system will The corresponding radial tolerance boundary, angular tolerance boundary, and associated markers are written into the prediction point record, so that in subsequent steps, the candidate direction set within the angular tolerance range and the position coverage within the radial tolerance range are used as constraints to complete beamguide optimization.

[0125] Step 3: Target direction spatial mapping and beam candidate domain generation:

[0126] First, a directional coordinate system is established with the array phase center of the ground communication array as the origin, and the predicted point set is mapped one by one into azimuth and elevation angles. The azimuth angle is obtained by comparing the horizontal projection direction of the array phase center to the predicted point with the ground horizontal reference axis as the starting reference direction. The elevation angle is obtained by comparing the angle between the three-dimensional direction of the array phase center to the predicted point and the horizontal plane with the ground horizontal plane as the reference plane.

[0127] For each prediction point, the angular tolerance boundary and radial tolerance boundary recorded in the second step are combined to expand it into an angle interval entry, and a prediction timestamp, trajectory segment identifier, correction source label and direction change region association mark are attached. Then, the angle evolution tensor is constructed.

[0128] To characterize the degree of dynamic stability, a local perturbation intensity index is established within each time slice of the angle evolution tensor. The local perturbation intensity index is defined by statistically analyzing whether the maximum change in the azimuth rate of change and the maximum change in the pitch curvature occur simultaneously and in what order within a fixed-length window near that time slice: if both increase synchronously or nearly synchronously, it is marked as a high perturbation; if only one increases or the two fluctuate alternately with small amplitudes, it is marked as a medium perturbation; if both are at a low level and there is no sign of directional reversal, it is marked as a low perturbation.

[0129] Simultaneously, the prediction confidence is calculated, which is jointly determined by the correction source label, the orientation change region association label, and the attitude stability parameters: if the correction source label indicates that the offset residual gradually converges and is not in the orientation change region, and the attitude stability parameters are within the baseline range of the stable segment, then the prediction confidence is marked as high; if there are alternating orientation fluctuations or a transition period in the orientation change region, then it is marked as medium; if there is consistent orientation divergence and the attitude stability parameters exceed the allowable upper limit of the disturbance segment, then it is marked as low.

[0130] An initial direction vector sequence is generated based on the angle evolution tensor, and direction continuity detection is performed to meet the exclusion requirements for angle continuity, abrupt changes in angle, and reversals. Direction continuity detection compares the azimuth and elevation angles of adjacent time stamps sequentially. If there is a sudden change in azimuth direction within a short period of time, a sudden change in elevation angle from rising to falling without a transition interval, or a reversal marker in the angle evolution tensor, the direction vector is excluded. Only direction sequences that are within the monotonic trend segment and can continuously cover the angle interval of adjacent prediction points with their main lobe are retained.

[0131] The retained direction sequence is mapped to the beam codebook of the ground communication array to obtain the direction sequence closest to the beam center of the codebook as the candidate beam direction sequence, while retaining the fine-tuning margin of the candidate beam direction sequence within the angular tolerance boundary.

[0132] For each candidate beam direction sequence, calculate the matching priority factor and stability confidence factor:

[0133] The matching priority factor is constructed based on the angular deviation from the prediction point to the main lobe center of the candidate beam direction sequence. When the angular deviation falls near the center of the angular tolerance boundary, the matching priority factor is high; when it is close to the edge of the angular tolerance boundary, the matching priority factor is medium; and when it exceeds the angular tolerance boundary, the candidate direction sequence is eliminated.

[0134] The stability confidence factor is determined jointly based on the local perturbation intensity and prediction confidence of the angle evolution tensor. If the local perturbation intensity is low and the prediction confidence is high, the stability confidence factor is high; if both are at a medium level, the stability confidence factor is medium; if either is low or high perturbation and low confidence occur simultaneously, the stability confidence factor is low and marked as "used only in redundant paths".

[0135] The thresholds are set based on the baseline thresholds of stable flight samples formed during the test flight phase and are adjusted in a bounded manner in combination with the attributes of the current trajectory segment: the stable segment adopts a stricter threshold for the matching priority factor and the stability confidence factor, while the perturbation segment is moderately relaxed but requires the candidate direction sequence to maintain continuous coverage of the main lobe at adjacent timestamps.

[0136] A dual constraint rule is applied to the above two factors: only direction entries that simultaneously satisfy the matching priority factor not lower than the threshold and the stability confidence factor not lower than the threshold can enter the candidate beam direction set.

[0137] For each candidate beam direction sequence in the candidate beam direction set, a structured record is established. The record includes the main lobe center direction of the candidate beam direction, the available fine-tuning margin, the timestamp interval of the coverage, the overlap mark with the direction change region, the trajectory segment identifier, the matching priority factor level and the stability confidence factor level, and generates a direction continuous segment identifier for subsequent beam switching strategy calls.

[0138] The third step of data processing is as follows:

[0139] First, the predicted position after mid-term correction is mapped onto the directional coordinate system of the ground communication array, and the continuity and stability of the directional change are evaluated in time series.

[0140] At any timestamp The predicted position vector after mid-term correction is taken as The array phase center position vector is Then the direction vector from the array to the predicted position is defined as: ,in, Represents timestamp Predicted three-dimensional position coordinates Represents the three-dimensional coordinates of the array phase center. This represents the spatial direction vector pointing from the array to the prediction point. The azimuth and elevation angles of the prediction direction are calculated from this direction vector, which are used to characterize the pointing relationship in the horizontal and vertical directions, respectively. , ,in, To predict the azimuth angle of the direction, For timestamps The pitch angle, Direction vectors The three components in the ground coordinate system.

[0141] To characterize the changes in orientation over time, the system calculates the rate of change of azimuth and the bending trend of pitch between adjacent time stamps for continuity and stability assessment. The rate of change of azimuth is defined as: ,in, Represents timestamp The rate of change of azimuth angle is used to reflect how quickly the direction rotates over a short period of time. The bending trend of the pitch angle is determined using discrete curvature. ,in, Represents timestamp The degree of pitch angle curvature, For timestamps pitch angle, For timestamps pitch angle, For timestamps The pitch angle. When its symbol flips between adjacent timestamps, it is determined that there is a risk of directional reversal or abrupt change. The system makes a comprehensive evaluation within a fixed-length sliding window. fluctuation range The magnitude of the changes and the sign flipping of the values ​​form a local disturbance intensity level, which is used to describe the dynamic stability of the current prediction direction. If the azimuth rate of change and the pitch angle deflection increase significantly at the same time within the window, and the deflection signs flip frequently, it is judged as a high disturbance; if only one of them changes significantly, it is judged as a medium disturbance; if both are at a low level and there is no sign flipping, it is judged as a low disturbance.

[0142] After completing the evaluation of directional continuity and disturbance intensity, the system maps the predicted direction to the array beam codebook. Let the nth beam in the codebook be... The main lobe center direction of each beam is... , For the first The direction angle of each beam For the first The elevation angle of each beam is given by the angular deviation between the predicted direction and the codebook beam, which is then defined as: ,in, The unit vector for predicting the direction. For the first The unit vector in the direction of the codebook beam center. It represents the spatial angle between the two.

[0143] Based on the angular tolerance boundary given in step two With codebook main lobe half-power angle Filter candidate directions: when If the codebook falls within the angular tolerance boundary and can maintain continuous main lobe coverage in adjacent timestamps, the codebook direction is determined to pass the matching priority factor screening at the current timestamp.

[0144] The stability confidence factor is jointly determined based on the local perturbation intensity level and the prediction confidence level: when the local perturbation intensity is low and the prediction confidence is high, the stability confidence factor is high; when both are at a moderate level, the stability confidence factor is medium; when a high perturbation occurs or the prediction confidence is low, the stability confidence factor is low, and it is only allowed as a redundant path candidate. A dual constraint rule is applied to the above two factors: only when the matching priority factor passes and the stability confidence factor is not low can the corresponding codebook direction enter the candidate beam direction set.

[0145] Step 4: Select the backup beam direction with the maximum path coverage redundancy from the candidate beam direction set:

[0146] Acquire beam configuration and channel quality feedback data of UAV in the current and historical communication cycles. Establish a "direction-quality comparison table" for each executed beam configuration, and record the azimuth and elevation angles of the beam center direction, the effective time period of the beam, the timestamp of the beam, the identifier of the trajectory segment, whether it is associated with the direction change region, the matching priority factor level and the stability confidence factor level of the candidate beam direction sequence. At the same time, collect channel quality feedback data, including signal-to-noise ratio sequence, received power fluctuation sequence, bit error rate change rate and delay jitter indication, and write them into the quality record after aligning with a unified time reference.

[0147] Based on channel quality feedback data, degraded communication quality sections and beam failure events are marked:

[0148] Within the minimum observation window, if the signal-to-noise ratio sequence shows a continuous downward trend and the received power fluctuation sequence continues to decrease or the bit error rate change rate increases continuously at adjacent observation points, it is marked as a communication quality degradation segment.

[0149] If the communication quality degradation segment coincides with the main lobe coverage time period of a candidate beam direction sequence, and the candidate beam direction sequence cannot maintain a stability confidence factor above the stability confidence factor threshold within the angular tolerance boundary, then the candidate beam direction sequence is determined to be a beam failure event.

[0150] For each beam failure event, the directional offset rate is calculated: the direction of the line connecting the array phase center to the predicted position after mid-term correction is used as the actual effective direction estimate. The angular difference between this direction and the beam center direction at the time of failure is compared, and a directional offset rate sequence is formed in chronological order within a fixed-length time window. The entries of the timestamp, beam center direction, communication failure mark and directional offset rate are connected to form a time-ordered beam offset residual path diagram. This path diagram provides historical offset patterns and trend basis for subsequent corrections.

[0151] Retrieve path segments that match the candidate beam set from the beam offset residual path map. The matching conditions are that the trajectory segments are consistent, the direction change regions are consistent, and the timestamps are adjacent or similar. Similar means that the time interval between the timestamp corresponding to the candidate beam direction sequence and the timestamp of the path segment in the beam offset residual path map is within the time alignment tolerance Δt mentioned in the second step.

[0152] Two types of trend inspections are performed on the matched path segments: one is the direction consistency inspection, which is used to determine whether the dominant direction of the direction offset rate remains consistent in the nearest time interval; the other is the amplitude evolution inspection, which is used to determine whether the amplitude of the direction offset rate continuously increases, continuously decreases, or fluctuates alternately in the nearest time interval.

[0153] Based on the above inspection results, construct the offset correction vector:

[0154] When the direction is consistent and the amplitude continues to increase, the direction of the offset correction vector is set to the direction from the candidate beam center direction to the actual effective direction estimate. The upper limit of the correction amplitude is constrained by the upper bound of the angular correction. The upper bound of the angular correction is determined by the angular tolerance boundary of the time slice corresponding to the candidate beam direction sequence in the third step, the local perturbation intensity of the angular evolution tensor, and the prediction confidence.

[0155] When the direction alternates and the amplitude fluctuates repeatedly between the upper and lower bounds, the amplitude of the offset correction vector is limited to a moderate amplitude range (40~60%) between the upper and lower bounds of the angular correction, in order to avoid overshooting of the directional fine-tuning.

[0156] When the directional offset rate shows a continuous decreasing trend, only a small directional correction is performed and the correction magnitude is set to the minimum perceptible directional fine-tuning amount in order to maintain the stabilized link.

[0157] If the current timestamp is within a region of directional abrupt change or within its transition duration, the offset correction vector must conform to the directional band constraints of the transition rhythm. The correction direction must not deviate from the allowed directional band, and the correction magnitude must not exceed the allowed step size of that directional band. This ensures that the continuity of angle changes and the continuous coverage of the main lobe in the third step are not disrupted. To ensure consistency, the system writes a correction source label and path segment identifier into the offset correction vector entry to indicate which trend scenario the vector originates from and the corresponding historical residual path segment.

[0158] In the next communication cycle, the offset correction vector is used to compensate and adjust the candidate beam direction set, resulting in a compensated and adjusted candidate beam direction set. Specifically:

[0159] After obtaining the offset correction vector, fine-tuning control is performed on each of the output candidate beam directions;

[0160] For any candidate beam direction Its main lobe center direction is a unit vector This indicates that within the available fine-tuning margin corresponding to that direction, the system proceeds in preset angular steps. Apply small-angle rotations in both the forward and reverse directions to form offset adjustment subsets. ,in, Indicates the first A finely adjusted direction vector This represents the minimum controllable angular step allowed by the array. For each direction in the offset adjustment subset, the system calculates its relationship with the offset correction vector. The included angle between them: ,in, This indicates the angular deviation between the fine-tuning direction and the offset correction direction. This represents the offset correction vector obtained from historical failure feedback, and its direction indicates the main offset trend that the main lobe needs to compensate for. In the calculation... At the same time, it verifies whether the fine-tuning direction is still within the angular tolerance boundary corresponding to the current timestamp. The system prioritizes whether the main lobe maintains continuous coverage for adjacent prediction points, whether the matching priority factor is not lower than a threshold, and whether the stability confidence factor is not lower than a threshold. The smallest fine-tuning direction that simultaneously satisfies the above four constraints is taken as the final fine-tuning beam direction, and a main communication link update instruction is generated and applied at the next available effective time. If there is no direction that meets the conditions within the fine-tuning margin of the candidate direction, the above process is repeated for the next direction in the candidate set until a qualified direction is selected or a backup strategy is triggered.

[0161] To determine whether a transition from fine-tuning to handover is necessary, degradation detection is performed on the communication quality trend over time. Let's assume that in the most recent... Within each observation point, the slope of the linear change of communication quality indicators (such as signal-to-noise ratio, bit error rate, or equivalent quality score) over time is denoted as follows: When the three symbols are identical and appear consecutively, the following condition is met: If so, it is determined that there is a consistent trend of communication quality degradation. This represents the slope of the mass change calculated within adjacent observation windows. This is the sign function. When degradation occurs, the direction deviation vector at that moment is recorded in the beam offset residual path diagram, forming an error sequence ordered by time. ,in: In the formula, This represents the actual beam center direction vector at the time of degradation. This represents the ideal pointing direction vector obtained from the predicted position and the geometric relationship of the array.

[0162] From recent Extracting the dominant offset trend from the failure direction error vectors yields the first principal feature vector. And limit its magnitude to the upper bound of the angular correction. Within this range, to prevent overcompensation; if currently in a region of directional abrupt change, further requirements are needed. It must comply with the transition direction zone constraint, meaning the correction direction cannot exceed the transition zone range and the single-step adjustment range cannot exceed [the limit]. Under this constraint, select the candidate within the fine-tuning margin of the candidate set. The direction with the smallest included angle that still satisfies the matching priority factor and the stability confidence factor not being low is taken as the final fine-tuning direction. The overlap relationship between this direction and its main lobe in the same timestamp is written back to the candidate set, and the continuous segment identifier of the direction is updated. At the same time, the correction result label, correction magnitude, effective time and its relationship with the redundant path are registered in the direction-quality comparison table, and new nodes are added to the beam offset residual path diagram for continuous updating of the direction offset rate and correction strategy in subsequent cycles.

[0163] Construct a redundant path map based on the candidate beam direction set for compensation control:

[0164] The nodes of the redundant path graph are candidate beam direction entries. Each node records the main lobe center direction, available fine-tuning margin, coverage timestamp interval, continuous direction segment identifier, matching priority factor level, and stability confidence factor level. The edges of the redundant path graph represent the main lobe overlap relationship between two nodes. When the main lobe coverage ranges of two candidate beam directions have spatial intersection at the same timestamp and the continuous length of the intersection on the time axis reaches the minimum switching window, a directed edge is established, and the start and end positions of the edge on the time axis are recorded. The minimum switching window is jointly determined by hardware reconfiguration delay calibration and scheduling granularity, requiring at least one complete beam reconfiguration and confirmation feedback cycle to be covered.

[0165] For subsequent selection, the directional complementary structure strength is calculated for each edge. The directional complementary structure strength is defined as strong, medium, and weak in three levels: If the two candidate beam directions maintain main lobe overlap throughout the entire time period of the minimum switching window, and the angle difference between the two directions is within the available fine-tuning margin, and the stability confidence factor of both directions within the window is not lower than the threshold, then the directional complementary structure strength is defined as strong; If the main lobe overlap only covers most of the time period of the minimum switching window, or the angle difference requires the boundary value of the fine-tuning margin to be satisfied, and the stability confidence factor drops to near the threshold for a short period of time, then the directional complementary structure strength is defined as medium; If the main lobe overlap only occurs sporadically within the window, or the angle difference exceeds the fine-tuning margin, or the stability confidence factor does not meet the stability confidence factor threshold within the window, then the directional complementary structure strength is defined as weak.

[0166] After the redundant path graph is constructed, the system uses the candidate beam direction with a high stability confidence factor and a high matching priority factor as the starting node of the main communication path. At the same time, it marks the adjacent nodes that are directly connected to the main communication path and have complementary directions with strong or medium strength on the redundant path graph, forming the initial set of backup path sequences.

[0167] During the operation of the main communication path, the signal-to-noise ratio sequence, bit error rate change rate, and received power fluctuation sequence are continuously collected and aligned, and the communication quality degradation trend is judged based on the execution.

[0168] When, within the minimum observation window, the signal-to-noise ratio sequence shows a continuous decreasing trend and the received power fluctuation sequence shows a continuous decrease, or the bit error rate change rate continuously increases at adjacent observation points, and the change directions of the above three sequences are consistent at most observation points with stable slope signs, the system confirms a communication quality degradation trend and triggers a handover preparation process. The handover preparation process does not immediately change the main communication path, but in the redundant path graph, starting from the node of the current main communication path, it prioritizes exploring edges with strong complementary structural strength, and sequentially collects the adjacent nodes they point to, forming a candidate sequence of backup beam directions arranged in chronological order. If there are no strong edges, then medium-level edges are selected, and at the node level, it is verified again whether the node still meets the angular tolerance boundary at the current timestamp, whether it still has continuous coverage of the main lobe for adjacent prediction point records, whether the matching priority factor is not lower than the threshold, and whether the stability confidence factor is not lower than the stability confidence factor threshold.

[0169] When multiple backup beam directions satisfy the conditions exist, a line-by-line discrimination method that does not rely on summation is adopted:

[0170] First, select the beam with the longer continuous coverage time within the minimum switching window; if there are still beams in a tie, select the one with an earlier expected activation time and a smaller minimum switching distance from the main communication path; if there are still beams in a tie, select the one that better matches the current directional change region marker. The minimum switching distance is defined as the minimum number of angular steps required to rotate from the current main lobe center direction to the main lobe center direction of the backup beam. This number comes from the superposition boundary of the minimum step capability of the array phase control and the available fine-tuning margin, to ensure that the switching action can be completed within a given switching delay window.

[0171] Once the backup beam direction is determined, a beam switching instruction set is generated and execution begins. The beam switching instruction set includes at least the following fields: target direction (i.e., the main lobe center direction of the backup beam and the usage scheme of available fine-tuning margin), estimated effective time (aligned with the scheduling time slot), and modulation adjustment parameters (redundant coding and symbol rate settings used to maintain demodulation robustness during switching). Before issuing the instruction, the spectrum interference avoidance window is simultaneously checked. This window comes from a reserved time period scheduled by the network side, requiring the switching action to avoid critical resources already allocated to neighboring cells or other links in the same cell. If a spectrum interference avoidance window conflict exists within the estimated effective time of the instruction, the estimated effective time of the instruction is postponed to the next available window, while keeping the backup beam direction unchanged. The instruction is dynamically loaded within the switching timing window. After the array completes the direction reconfiguration... Immediately acquire new signal-to-noise ratio sequences, bit error rate change rates, and received power fluctuation sequences at the same time reference, update the direction-quality lookup table, and add new nodes to the beam offset residual path graph for offset correction and trend determination in subsequent cycles. If the stability confidence factor cannot be restored to above the threshold after switching, the system repeats the above selection and execution process according to the next edge on the redundant path graph. If no candidate direction simultaneously satisfies the four constraints of angular tolerance boundary, continuous main lobe coverage, matching priority factor, and stability confidence factor, the system calls the offset correction vector to perform short-range fine-tuning in the current direction to maintain the link, and temporarily expands the main lobe width if necessary, but the expansion must be limited to the energy distribution and interference control range allowed by the state evolution model.

[0172] In summary, with redundant path diagrams, complementary directional structure strength, and switching timing control as the core, the generation and implementation of main communication path initialization, communication quality degradation trend triggering, backup beam direction selection, and beam switching instruction set were completed, effectively improving the feasibility of dynamic beam switching and link stability.

[0173] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A beam steering optimization method based on position prediction in UAV communication, characterized in that, include: Acquire flight trajectory data of the drone during a continuous communication cycle; The trajectory is adaptively segmented based on the flight trajectory data of the UAV during the continuous communication cycle, and a state evolution model is constructed for each segment of the flight trajectory. Short-term and medium-term residual calibration predictions are performed based on the state evolution model of each flight trajectory segment, and prediction correction terms are constructed to generate a prediction point set; The predicted point set is mapped to the directional coordinate system of the ground communication array to construct an angle evolution tensor, and the candidate beam direction set is obtained by filtering based on the angle evolution tensor; The backup beam direction with the maximum path coverage redundancy is selected from the candidate beam direction set to achieve dynamic beam switching and link stability enhancement.

2. The beam steering optimization method based on position prediction in UAV communication according to claim 1, characterized in that, The flight trajectory data of the UAV during the continuous communication cycle is collected after the UAV and the ground communication array establish a synchronized time reference. The flight trajectory data includes position, speed and attitude information. The position includes three-dimensional spatial position and the UAV's own body coordinate orientation. The speed includes a speed vector. The attitude information includes roll angle, pitch angle and heading angle. The three-dimensional spatial position is based on a fixed ground reference coordinate system.

3. The beam steering optimization method based on position prediction in UAV communication according to claim 1, characterized in that, The process of adaptively segmenting the trajectory based on the UAV's flight trajectory data within a continuous communication cycle and constructing a state evolution model for each flight trajectory segment includes: Construct a flight trajectory sequence matrix, wherein the flight trajectory sequence matrix is ​​flight trajectory data under multiple consecutive timestamps; Generate velocity change rate curves, heading change angle sequences, and attitude stability parameters based on the flight trajectory sequence matrix; Traverse the velocity change rate curve, heading change angle sequence, and attitude stability parameters, and compare them with the velocity change rate threshold, heading change angle threshold, and attitude stability threshold, respectively. If at least one of them exceeds the corresponding threshold and the duration meets the minimum duration, segmentation is triggered, and each duration is taken as a flight trajectory segment. For each flight trajectory segment, if the high-value regions of the velocity change rate curve, heading change angle sequence, and attitude stability parameters in the flight trajectory segment overlap in time or any two of them appear consecutively alternately, then the flight trajectory segment is marked as a disturbance segment; otherwise, it is marked as a stable segment. The region in the flight trajectory segment where the heading change angle sequence exceeds the heading angle threshold and the duration is not less than the minimum duration is defined as the heading change region, and the start and end times, peak change angle, and interval between adjacent regions of the heading change region are recorded. A corresponding state evolution model is constructed for each flight trajectory segment based on the disturbance segment, the stable segment, and the region of sudden change in direction.

4. The beam steering optimization method based on position prediction in UAV communication according to claim 1, characterized in that, The process of performing short-term and medium-term residual calibration predictions for each flight trajectory segment, and constructing prediction correction terms to generate a prediction point set, includes: S1. Obtain the velocity change range, heading change range, and attitude stability baseline from the state evolution model corresponding to the flight trajectory segment, and calculate the instantaneous state change trend by combining the flight trajectory data of the current timestamp and the previous timestamp. The instantaneous state change trend includes the rate of change of velocity vector angle, the magnitude of heading angle change, and the real-time evaluation results of attitude stability parameters. S2. Using the velocity vector direction at the current timestamp as the reference direction, determine the direction adjustment direction by referring to the heading angle change range, and determine the single-step displacement range based on the velocity change range. S3. Combine the direction adjustment and single-step displacement amplitude to obtain the short-term predicted position of the current timestamp; S4. Repeat S1~S3 to obtain the short-term predicted position of the flight trajectory segment for all timestamps within the continuous communication period. S5. For a timestamp, the spatial offset vector between the short-term predicted position of the flight trajectory segment and the actual position of the timestamp in the previous communication cycle is used as the offset residual of the timestamp in the current communication cycle. S6. Combine the offset residuals of all timestamps into an offset residual sequence in chronological order, and perform direction consistency inspection and amplitude evolution inspection on the offset residual sequence to obtain the inspection results. S7. Construct a prediction correction term based on the inspection results, and use the prediction correction term to correct the short-term prediction position of the timestamp in the current communication cycle to obtain the medium-term corrected prediction position. S8. Repeat S5~S7 to obtain the mid-term corrected predicted positions of all timestamps. Then, associate the timestamps with their mid-term corrected predicted positions, position tolerance intervals, corresponding trajectory segments, correction source labels, and direction change regions as prediction points. Combine all prediction points to obtain the prediction point set for the flight trajectory segment.

5. The beam steering optimization method based on position prediction in UAV communication according to claim 4, characterized in that, The position tolerance range includes a radial tolerance boundary and an angular tolerance boundary. The radial tolerance boundary is determined by attitude stability parameters, and the angular tolerance boundary is determined by the heading angle variation amplitude, the velocity vector angle variation rate, and the peak change angle and start and end time in the direction change region. The construction of prediction correction terms based on the inspection results includes: The direction of the forecast correction term is set from the short-term forecast position to the actual position; When the direction of the offset residual is consistent and the amplitude shows a continuous diverging trend, the amplitude is limited to the upper limit of the allowable displacement correction. When the direction of the offset residual alternates and the amplitude fluctuates back and forth, the amplitude of the prediction correction term is limited to the 40-60% amplitude range between the upper limit of the allowable displacement correction and the lower limit of the allowable displacement correction. When the magnitude of the offset residual shows a gradual convergence trend, the magnitude of the prediction correction term is limited to no more than 10% of the lower bound of the allowable displacement correction. The upper and lower bounds of the allowable displacement correction are derived based on the allowable velocity variation range and the attitude stability baseline.

6. The beam steering optimization method based on position prediction in UAV communication according to claim 5, characterized in that, The step of mapping the predicted point set to the directional coordinate system of the ground communication array to construct an angle evolution tensor, and filtering the candidate beam direction set based on the angle evolution tensor, includes: A directional coordinate system is constructed with the array phase center of the ground communication array as the origin; The predicted points are mapped one by one to azimuth and pitch angles. The azimuth angle is obtained by comparing the horizontal projection direction of the array phase center to the predicted point with the ground horizontal reference axis as the starting reference direction. The pitch angle is obtained by comparing the angle between the three-dimensional direction of the array phase center to the predicted point and the horizontal plane with the ground horizontal plane as the reference plane. An angle evolution tensor is constructed for each prediction point. Local perturbation intensity index is established in each time slice of the angle evolution tensor, and prediction confidence is calculated based on the corrected source label, the association label of the direction change region, and the attitude stability parameter. An initial direction vector sequence is generated based on the angle evolution tensor, and direction continuity detection is performed. The initial direction vector sequence that meets the detection conditions is retained as the direction sequence. The direction sequence is mapped to the beam codebook of the ground communication array, and the direction sequence closest to the beam center of the codebook is selected as the candidate beam direction sequence. The available fine-tuning margin of the candidate beam direction sequence within the angular tolerance boundary is retained. Calculate the matching priority factor and stability confidence factor for each candidate beam direction sequence; The matching priority factor and stability confidence factor are compared with the matching priority factor threshold and stability confidence factor threshold, respectively. Candidate beam direction sequences with matching priority factors not lower than the matching priority factor threshold and stability confidence factors not lower than the stability confidence factor threshold are selected to form a candidate beam direction set.

7. The beam steering optimization method based on position prediction in UAV communication according to claim 1, characterized in that, The process of selecting the backup beam direction with the maximum path coverage redundancy from the candidate beam direction set includes: Acquire beam configuration and channel quality feedback data of the UAV in the current and historical communication cycles; Based on the beam configuration and channel quality feedback data of the UAV in the current and historical communication cycles, a beam offset residual path map is constructed for the candidate beam direction set. The offset correction vector is extracted from the beam offset residual path map, and the offset correction vector is used to compensate and adjust the candidate beam direction set in the next communication cycle to obtain the candidate beam direction set for compensation and adjustment. A redundant path map is constructed based on the candidate beam direction set for compensation control; A candidate beam direction is selected from the redundant path graph as the starting node of the main communication path, and adjacent nodes are marked on the redundant path graph to form an initial set of backup path sequences. During the operation of the main communication path, beam switching control is triggered based on the communication quality degradation trend to determine the backup beam direction with the maximum path coverage redundancy.

8. The beam steering optimization method based on position prediction in UAV communication according to claim 7, characterized in that, The channel quality feedback data includes a signal-to-noise ratio sequence, a received power fluctuation sequence, a bit error rate change rate, and a delay jitter indication. The method for constructing the beam offset residual path map includes: Based on the azimuth, elevation, effective time period, timestamp, and trajectory segment of the candidate beam direction sequence, communication quality degradation sections and beam failure events are marked using channel quality feedback data, including: Within the minimum observation window, if the signal-to-noise ratio sequence shows a continuous decreasing trend and the received power fluctuation sequence continues to decrease, or the bit error rate change rate continuously increases at adjacent observation points in the candidate beam direction sequence, it is marked as a communication quality degradation segment. If the main lobe coverage time period of a candidate beam direction sequence coincides with the communication quality degradation section, and the candidate beam direction sequence cannot maintain the stability confidence factor above the stability confidence factor threshold within the angular tolerance boundary, then the candidate beam direction sequence is determined to be a beam failure event. For each beam failure event, the actual effective direction is estimated by the direction of the line connecting the center of the ground communication array to the predicted position after mid-term correction of the timestamp corresponding to the beam failure event. The actual effective direction estimate is compared with the beam center direction at the time of failure by angular difference, and a direction offset rate sequence is formed in time order within a fixed-length time window; The beam offset residual path diagram is obtained from the timestamp, beam center direction, communication failure marker, and direction offset rate sequence. The step of extracting the offset correction vector based on the beam offset residual path map and then using the offset correction vector to compensate and adjust the candidate beam direction set in the next communication cycle, resulting in a compensated and adjusted candidate beam direction set, includes: The path segments that match the candidate beam set are retrieved from the beam offset residual path map. The matching conditions are that the trajectory segments are consistent, the direction change region is consistent, and the timestamps are adjacent or the time interval between the timestamps is within the preset time alignment tolerance. The inspection results are obtained by inspecting the path segments for directional consistency and amplitude evolution. Construct an offset correction vector based on the inspection results: If the direction of the path segments is consistent and the amplitude continues to increase, the direction of the offset correction vector is set to the candidate beam center direction pointing to the actual effective direction estimate, and the amplitude is limited to the upper bound of the angular correction. If the direction of the path segment alternates and the amplitude fluctuates repeatedly, the amplitude of the offset correction vector is limited to 40-60% of the amplitude range between the upper bound and the lower bound of the angular correction. If the directional offset rate of a path segment continues to decrease, the direction of the offset correction vector is set to fine-tuning, and the amplitude is limited to a very small amplitude. In the next communication cycle, the offset correction vector is used to compensate and adjust the candidate beam direction set to obtain the compensated and adjusted candidate beam direction set. The upper and lower bounds of the angular correction are derived based on the angular tolerance boundary of the time slice corresponding to the candidate direction sequence, the local perturbation intensity of the angular evolution tensor, and the prediction confidence.

9. The beam steering optimization method based on position prediction in UAV communication according to claim 8, characterized in that, The construction of the redundant path map based on the candidate beam direction set of the compensation control includes: Candidate beam direction entries are used as nodes in the redundant path graph, and the main lobe overlap relationship between two nodes is used as an edge. When the main lobe coverage of two candidate beam directions has a spatial intersection at the same timestamp and the continuous length of the intersection on the time axis reaches the minimum switching window, a directed edge is constructed between the corresponding nodes of the two candidate beam directions. Calculate the complementary structural strength of each side to obtain the redundancy path diagram; The nodes corresponding to the candidate beam directions with high stability confidence factors and high matching priority factors are taken as the starting nodes of the main communication path, and the nodes directly connected to the main communication path with strong or medium complementary strength of the connecting edge are taken as adjacent nodes. During the operation of the main communication path, beam switching control is triggered based on the communication quality degradation trend to determine the backup beam direction with the maximum path coverage redundancy, including: During the operation of the main communication path, the signal-to-noise ratio sequence, bit error rate change rate, and received power fluctuation sequence are continuously acquired and aligned. When, within the minimum observation window, the signal-to-noise ratio (SNR) sequence shows a continuous decreasing trend and the received power fluctuation sequence continuously decreases, or the bit error rate (BER) change rate continuously increases at adjacent observation points, and simultaneously the SNR sequence, BER change rate, and received power fluctuation sequence remain consistent across multiple observation points with stable slope signs, then a communication quality degradation trend is confirmed, and beam switching control is triggered. The beam switching control includes: Taking the starting node of the main communication path in the redundant path graph as the starting point, first search for the edge that is connected to the starting point and has complementary direction and strong structure strength, and then search for the adjacent node pointed to by the edge. The starting point and the adjacent nodes form a candidate sequence of backup beam directions arranged in the order of timestamps. The backup beam direction is selected from the candidate sequence of backup beam directions in order of the following criteria: longest continuous coverage time within the minimum switching window, earliest expected effective time, smallest minimum switching distance with the main communication path, and best match with the current direction change area marker.

10. The beam steering optimization method based on position prediction in UAV communication according to claim 9, characterized in that, The strength of the directional complementary structure is divided into strong, medium, and weak levels, among which, When the main lobes of the two candidate beam directions overlap throughout the entire period of the minimum switching window, and the angle difference between the two candidate beam directions is within the available fine-tuning margin, and the stability confidence factors are not lower than the stability confidence factor threshold, the strength of the complementary direction structure is strong. When the main lobes of the two candidate beam directions overlap for more than 50% of the minimum switching window, or when the angle difference between the two candidate beam directions is at the boundary value of the available fine-tuning margin, and the stability confidence factor drops to the stability confidence factor threshold boundary for a short time, the strength of the complementary structure is at a medium level. When the main lobe overlaps between two candidate beam directions within less than 10% of the minimum switching window, the angle difference between the two candidate beam directions exceeds the available fine-tuning margin, or the stability confidence factor does not meet the threshold, the strength of the directional complementary structure is low.

Images