Single-anchor three-dimensional positioning system based on signal angle of arrival fitting

Abstract

The present application relates to the field of radio positioning measurement and data processing, in particular to a single-anchor three-dimensional positioning system based on signal angle of arrival fitting. The system comprises an angle of arrival observation processing module, a direction constraint generation module, a residual field construction module, a residual field partition module, a robust fitting module and a positioning output module. By receiving the azimuth angle, the elevation angle and the angle confidence of the single-anchor, three-dimensional direction constraints are generated. Angle residuals are calculated in the candidate space to form a residual field. According to the spatial continuity of the residual field, the residual transition boundary and the pseudo-convergence characteristics, the effective direction area and the abnormal direction area are divided. The weighted fitting is carried out in combination with the partition results, the time sequence relationship and the angle confidence. The system can suppress the influence of multipath or non-line-of-sight angle observation on the fitting result, reduce the three-dimensional coordinate offset, the height drift and the positioning jump, and improve the stability and reliability of the single-anchor angle of arrival positioning result.

Description

Technical Field

[0001] This invention relates to the field of radio positioning measurement and data processing technology, specifically to a single-anchor-point three-dimensional positioning system based on signal angle of arrival fitting. Background Technology

[0002] 3D positioning based on signal angle of arrival typically uses a known anchor point as the observation reference. The anchor point extracts the azimuth and elevation angles from the received signal, and then converts the angle observations into spatial direction lines in the anchor point coordinate system. The conventional processing flow involves generating candidate positions in a pre-defined 3D candidate space, calculating the angular deviation of the candidate positions relative to the direction lines, and selecting the position with the smaller deviation as the target coordinates. Alternatively, multiple frames of angle observations obtained through continuous sampling can be input into a least-squares or weighted least-squares model, and the 3D coordinates are obtained by minimizing the fitting residuals. To adapt to the application coordinate system, existing schemes also perform coordinate transformations on the anchor point's installation attitude, coordinate origin, and direction axis, and use the transformed angle results as the positioning input. This type of scheme mainly relies on the quality of angle estimation and the geometric fitting process itself, has low requirements for the number of anchor points, and is easy to form a single-anchor-point positioning system. However, its positioning information source is concentrated within a single angle observation link, lacking spatial cross-constraints from other anchor points.

[0003] The existing patent, CN117054965A, discloses a single-anchor-point three-dimensional positioning method based on the angle of arrival. This method collects the azimuth, pitch, and angle confidence of the target signal through a single anchor point, converts the multi-frame angle observations within a continuous sampling window into spatial direction constraints, calculates the angle residuals between each candidate position and the direction constraints in a preset three-dimensional candidate space, sets fitting weights based on the angle confidence and sampling time sequence, and obtains the target's three-dimensional coordinates through weighted least squares iteration.

[0004] The aforementioned existing technical solutions share the following common defects: First, the residuals are only compressed into scalar values ​​in the fitting target. The spatial distribution of residuals in the candidate space and the cross-frame variation relationship are not structurally preserved and analyzed. The residuals only serve the global minimum search and cannot carry the propagation source discrimination information of angle observations. Second, they do not distinguish between local continuous residual regions, abnormal transition regions, and pseudo-convergence regions formed by reflection paths. Low residual regions corresponding to direct paths and reflection paths are treated equally in the model, lacking a mechanism for distinguishing observation sources based on residual spatial characteristics. Third, they lack a mechanism for identifying stable reflection angle observations formed by multi-path or non-line-of-sight propagation. When the intensity of the reflection component is higher than that of the direct component, continuous reflection observations will participate in fitting with lower residuals and pull the positioning results to shift in the wrong direction, causing three-dimensional coordinate shifts, height drift, and positioning jumps. The aforementioned shortcomings form a chain of constraints. Because a structured residual carrier integrating spatial, temporal, and angular dimensions is not constructed, it is impossible to extract the spatial characteristics of continuous or abrupt residual changes to distinguish the source of observation propagation. Furthermore, the inability to accurately identify reflection-type observations allows pseudo-convergence regions generated by reflections to gain an unreasonable low residual weight advantage in iterative solutions, guiding the solution path to continuously deviate from the effective continuous residual region corresponding to the true target. Therefore, this existing technology cannot simultaneously achieve the three core objectives of accurate reflection angle identification, pseudo-convergence region suppression, and stable positioning output, making it difficult to effectively guarantee the accuracy and reliability of single-anchor-point 3D positioning results. Summary of the Invention

[0005] The purpose of this invention is to provide a single-anchor-point three-dimensional positioning system based on signal angle of arrival fitting, which can effectively solve the problems in the above-mentioned background technology where residuals are only used as scalars in calculations, and the inability to distinguish between effective areas and pseudo-converged reflection areas, and where continuous reflection observation is prone to coordinate shift, height drift, and positioning jumps.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A single-anchor-point three-dimensional positioning system based on signal angle of arrival fitting includes: an angle of arrival observation and processing module, a direction constraint generation module, a residual field construction module, a residual field partitioning module, a robust fitting module, and a positioning output module. The feature is that the angle of arrival observation and processing module receives the azimuth angle, pitch angle, and angle confidence degree collected from the single anchor point. The orientation constraint generation module generates three-dimensional orientation constraints based on the azimuth angle and the pitch angle, and the residual field construction module calculates the angular residual between the candidate position and the three-dimensional orientation constraint in the three-dimensional candidate space and forms a residual field. The residual field partitioning module divides the effective direction region and the abnormal direction region according to the spatial continuity and residual mutation characteristics of the residual field. The robust fitting module performs weighted fitting on the angle of arrival observations based on the effective direction region and the abnormal direction region. The positioning output module outputs the system of target three-dimensional coordinates.

[0007] Preferably, the residual field construction module includes a candidate space generation unit, an angle consistency calculation unit, and a residual tensor organization unit. The candidate space generation unit generates a three-dimensional candidate space containing radial candidate points, horizontal candidate points, and height candidate points based on the coordinate system of the single anchor point. The angle consistency calculation unit calculates the azimuth residual and pitch residual of each candidate position relative to the three-dimensional direction constraint of each frame. The residual tensor organization unit organizes the residual data according to the spatial position index, sampling time index, and angle type index, so that the residual changes of the same candidate position in consecutive sampling frames are preserved as a partitionable data structure.

[0008] Preferably, the residual field partitioning module includes a local continuity analysis unit, a residual transition detection unit, and a pseudo-convergence region identification unit. The local continuity analysis unit generates continuous residual patches based on the consistency of residual gradients between adjacent candidate positions. The residual transition detection unit generates abnormal residual patches based on the abrupt change in residual distribution between consecutive sampling frames. The pseudo-convergence region identification unit marks candidate patches where the residual values ​​are locally converged but do not match the temporal direction change as pseudo-convergence regions, and outputs the continuous residual patches, the abnormal residual patches, and the pseudo-convergence regions as the residual field partitioning result.

[0009] Preferably, the robust fitting module includes an observation weight generation unit, a joint residual modeling unit, and an iterative position solving unit. The observation weight generation unit generates fitting weights for each frame's angle of arrival observation based on the residual field partitioning results, the angle confidence level, and the sampling time sequence relationship. The joint residual modeling unit combines the angle residual, the temporal continuous residual, and the spatial feasibility residual into a position fitting target. After each round of solving, the iterative position solving unit updates the fitting weights according to the partition to which the candidate position belongs, and generates the target's three-dimensional coordinates based on the updated fitting weights.

[0010] Preferably, the candidate space generation unit determines the candidate space boundary based on the direction bundle envelope formed by continuous angle of arrival observations, and divides the candidate positions within the direction bundle envelope into a main direction neighborhood, a radial extension neighborhood, and an angle edge neighborhood. The angle consistency calculation unit calculates the normalized angle residuals in the main direction neighborhood, the radial extension neighborhood, and the angle edge neighborhood, respectively. The residual tensor organization unit writes the normalized angle residuals and the corresponding neighborhood identifiers into the residual tensor, so that the residual field construction process includes the spatial attribution information of the candidate positions relative to the direction bundle shape.

[0011] Preferably, the local continuity analysis unit establishes an adjacency graph for candidate positions in the residual field and generates continuous region labels based on the consistency of the residual gradient directions of adjacent candidate positions. The residual transition detection unit marks the cross-region residual break boundary in the adjacency graph and tracks the persistence of the residual break boundary according to the sampling time index. The pseudo-convergence region identification unit marks the set of candidate locations with stable local low residuals and residual break boundaries with adjacent continuous regions as reflection candidate patches, which are part of the abnormal direction region.

[0012] Preferably, the observation weight generation unit includes a partition weight subunit, a temporal weight subunit, and a confidence fusion subunit. The partition weight subunit generates basic weights based on the attribution of the candidate position corresponding to the angle of arrival observation in the effective direction region, the abnormal direction region, and the pseudo-convergence region. The temporal weight subunit generates temporal adjustment weights based on the consistency of direction changes between consecutive sampling frames. The confidence fusion subunit fuses the basic weights, the temporal adjustment weights, and the angle confidence into fitting weights and writes the fitting weights into the joint residual modeling unit.

[0013] Preferably, the residual tensor organization unit further includes a scale consistency marker subunit and a candidate level compression subunit. The scale consistency marker subunit generates a scale consistency marker based on the slope of the residual change of the candidate position in the radial extension direction as the candidate position in the radial extension neighborhood. The candidate level compression subunit performs hierarchical compression of the three-dimensional candidate space based on the neighborhood identifier, the normalized angular residual, and the scale consistency marker, and retains the mapping relationship between the candidate position before compression and the representative position after compression.

[0014] Preferably, the residual transition detection unit further includes a cross-frame boundary association subunit, which generates a boundary association chain based on the positional overlap and boundary direction consistency of the residual break boundaries in different sampling frames; The pseudo-convergence region identification unit filters persistent reflection candidate regions based on the boundary association chain, and marks the residual fracture boundary that appears briefly and does not form a boundary association chain as an instantaneous angular perturbation boundary, so that the residual field partitioning result includes persistent reflection identifiers and instantaneous perturbation identifiers.

[0015] Preferably, the iterative location solving unit includes a weight write-back subunit, a convergence path recording subunit, and a location confidence generation subunit. After each round of fitting, the weight write-back subunit writes back the candidate location assignment partition corresponding to the target 3D coordinates to the observation weight generation unit. The convergence path recording subunit records the migration state of the target 3D coordinates between the effective direction region, the abnormal direction region, and the pseudo-convergence region during continuous iterations. The location confidence generation subunit generates the location confidence corresponding to the target 3D coordinates based on the final fitted weight distribution, the joint residual distribution, and the migration state.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. To address the shortcomings of traditional positioning schemes that compress residuals into a single scalar without preserving the spatial and temporal distribution characteristics of residuals and failing to distinguish between various low-residual regions, this invention constructs a residual tensor that integrates spatial location, sampling time sequence, and angle type three-dimensional indexes, fully preserving continuous residual changes and cross-frame evolution information. Based on residual gradients and residual break boundaries, it divides effective continuous regions, anomalous disturbance regions, and reflection pseudo-convergence regions, achieving differentiated distinction between direct and reflected signal observations. The fitting process no longer assigns equal weights to all angle observations but generates differentiated fitting weights based on the region to which the candidate location belongs. This fundamentally avoids the indiscriminate calculation of low-residual regions formed by reflections and truly effective regions. By relying on the spatial distribution characteristics of residuals to determine the source of observation propagation, it effectively reduces the interference of stable reflection observations on coordinate solving.

[0017] 2. To address the shortcomings of traditional methods, such as the inability to effectively identify continuous reflection observations and the resulting coordinate shifts, height drifts, and positioning jumps, this invention employs a multi-layered collaborative constraint mechanism: It uses the direction bundle envelope to limit the candidate space and divides it into three neighborhoods. Combined with scale consistency labeling and candidate level compression, it distinguishes between the natural radial depth ambiguity region of a single anchor point and the reflection pseudo-convergence region, reducing computational load while avoiding confusion between the two types of low residual regions. It differentiates between instantaneous noise disturbances and long-term stable reflections through cross-frame boundary association chains, only reducing the weight of continuous reflection observations. It constructs a joint fitting objective that integrates angular residuals, temporal constraints, and spatial penalties, adding a weighted write-back loop during iterative solving to dynamically suppress the pull of abnormal regions on the solution path. It records the coordinate iteration migration path throughout the process, combining weight distribution and residual numerical quantification to output positioning reliability, and simultaneously labels the type of abnormal observations. In continuous positioning scenarios, the reliability index can conversely constrain the candidate space range, continuously reducing repeated interference from reflection regions, significantly improving coordinate drift and jump problems in multi-path and non-line-of-sight environments, and enhancing the stability and interpretability of single-anchor-point 3D positioning results. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the overall processing of the single-anchor-point three-dimensional positioning system based on signal angle of arrival fitting according to the present invention. Figure 2 This is a flowchart of the residual field construction and residual tensor organization of the present invention; Figure 3 This is a flowchart of the residual field partitioning and pseudo-convergence region identification process of the present invention; Figure 4 This is a flowchart of the robust fitting, weight write-back, and localization credibility generation process of the present invention. Detailed Implementation

[0019] The flowcharts, tables, and formulas all correspond to the content of the instruction manual. The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please refer to Figure 1This embodiment provides a single-anchor-point 3D positioning system based on signal angle of arrival fitting, including: an angle of arrival observation and processing module, a direction constraint generation module, a residual field construction module, a residual field partitioning module, a robust fitting module, and a positioning output module. The angle of arrival observation and processing module receives the azimuth, elevation, and angle confidence values ​​output by the single anchor point during continuous sampling, and assigns a unified time stamp to angle observations at the same sampling time, enabling subsequent processing to distinguish the angle change relationships between different sampling frames. The direction constraint generation module uses the single anchor point coordinate system as a reference to convert the azimuth and elevation angles into spatial direction constraints. The spatial direction constraint represents a three-dimensional unit direction line starting from the single anchor point position and pointing to the possible target position. The residual field construction module deploys candidate positions in the three-dimensional candidate space and calculates the angle residual between each candidate position and the spatial direction constraint of each sampling frame, thereby expanding the originally single angle fitting error into a residual field with spatial position and time indices. The residual field partitioning module determines the relative stability of the residual changes between adjacent candidate positions in the residual field and the relative stability between different regions. The residual transition boundary and the stability of low residual regions in continuous sampling frames are used to divide the three-dimensional candidate space into effective direction regions, abnormal direction regions, and pseudo-convergence regions. The robust fitting module does not directly input all arrival angle observations into the minimum residual solution in the same way, but generates fitting weights based on the candidate position partitioning results, angle confidence, and sampling time sequence relationship corresponding to different sampling frames. The angle residual, temporal continuous residual, and spatial feasibility residual are combined into a joint fitting target. During the iterative solution process, the fitting weights are written back according to the region to which the current candidate position belongs. After the robust fitting module completes convergence, the positioning output module outputs the target three-dimensional coordinates and can simultaneously output the fitting residual distribution, abnormal observation markers, and positioning confidence. In this embodiment, the core processing logic of the system is to separate the reflection angle observations caused by multipath or non-line-of-sight from ordinary angle noise, so that the stable but spatially unreasonable low residual regions are no longer included in the solution as continuous low residual regions near the direct direction. Therefore, it can reduce the three-dimensional coordinate offset and height drift caused by reflection path traction when a single anchor point lacks multi-anchor point intersection verification.

[0021] Specifically, when the angle of arrival observation processing module receives a frame of angle observation, it records the azimuth angle as... The pitch angle is denoted as The angle confidence level is denoted as ,in Indicates the sampling time index. The value is a normalized confidence value, and the direction constraint generation module calculates the three-dimensional unit direction vector according to the following formula: ; in Indicates the first Spatial orientation constraints corresponding to frame angle of arrival observations and These represent the cosine and sine operations, respectively. This represents the transpose of a vector. and Before being used in calculations, all values ​​are converted to radians. If the azimuth angle of a certain frame is... Pitch angle is ,but , After substituting, we get Approximately Its physical meaning is the normalized pointing components of the target relative to a single anchor point in the horizontal first axis direction, the horizontal second axis direction, and the height direction. The residual field construction module takes candidate positions in the candidate space. The single anchor point position is By using the vector of the candidate position relative to a single anchor point and The angle relationship is calculated to ensure angle consistency, thereby converting the match between each candidate position and each frame direction constraint into a comparable residual value. In this embodiment, the candidate position does not need to preset the target plane, nor does it need to introduce additional hardware ranging data. Instead, it uses the residual distribution formed in three-dimensional space by continuous angle observation to complete multi-path observation identification and three-dimensional coordinate solution. The advantage of this embodiment is that it uses the residual field formed by the arrival angle observation itself as the discrimination basis, so that the single anchor point system has the ability to suppress reflection angles without changing the hardware conditions.

[0022] In one embodiment, reference Figure 2The residual field construction module includes a candidate space generation unit, an angle consistency calculation unit, and a residual tensor organization unit. The candidate space generation unit generates a three-dimensional candidate space based on a single anchor point coordinate system. This three-dimensional candidate space can consist of radial candidate points, horizontal candidate points, and height candidate points. Radial candidate points cover the depth position extending from the single anchor point along the directional constraint; horizontal candidate points cover the lateral position formed by the left and right deviations of the azimuth angle; and height candidate points cover the vertical position formed by the deviation of the pitch angle. The angle consistency calculation unit calculates the azimuth residual and pitch residual for each candidate position and organizes these two types of residuals into a unified angular residual of the candidate position relative to the sampling frame. The residual tensor organization unit organizes the residual data according to the spatial position index, sampling time index, and angle type index, so that the residual changes of the same candidate position in consecutive sampling frames can be preserved. The residual tensor can be represented as an array with three types of indices, where the spatial position index is used to locate the candidate position, the sampling time index... For associating continuous angle of arrival observations, the angle type index is used to distinguish azimuth residuals, pitch residuals, and normalized composite residuals. In specific implementation, the candidate space generation unit can determine the candidate space boundary based on the direction bundle envelope formed by continuous angle of arrival observations, and then generate candidate points inside the direction bundle envelope, instead of exhaustively listing all positions in the entire irrelevant space. This allows the calculation of the residual field to be concentrated in the area that can be constrained by the angle observations. The candidate space generation unit can also retain the relative relationship between each candidate position and the direction bundle centerline, including the attribution identifier located in the main direction neighborhood, radial extension neighborhood, or angle edge neighborhood. The attribution identifier is input together with the residual value into subsequent partitioning processing. In this embodiment, the residual field is not simply a temporary calculation object for finding the minimum value, but serves as a structured data basis for determining the source of the angle of arrival observations. The advantage of this embodiment is that it preserves the relationship between candidate positions, sampling frames, and angle types in tensor form, enabling subsequent analysis of multipath interference from the perspective of spatial continuity and cross-frame variation.

[0023] Preferably, the angle consistency calculation unit calculates the comprehensive angle residual using the angle between the candidate direction vector and the observed direction vector, and the specific formula is as follows: ; in Indicates the first The candidate position is relative to the first The combined angular residuals of frame angle of arrival observations Represents the dot product of vectors. Represents the L2 norm, Represents the inverse cosine operation. This represents the spatial vector pointing from a single anchor point to a candidate position. The denominator is used to normalize the candidate direction. If the single anchor point is located at... Candidate positions are The aforementioned observation direction vector is The magnitude of the candidate direction vector is approximately The inner product is approximately The normalized inner product is approximately ,get Approximately In radians, this indicates a small angular deviation between the candidate position and the directional constraint of the frame. If the residual between another candidate position and the same directional constraint is... If the radius is less than the observation direction, then its consistency with the observation direction is lower. In this embodiment, the azimuth residual and pitch residual can be obtained by comparing the horizontal projection angle and height projection angle of the candidate direction with the observation angle, respectively. The combined angular residual is used as the master index of the residual field, and the azimuth residual and pitch residual are used to subsequently identify the source of altitude drift or horizontal offset. The residual tensor organization unit divides each Write the residual tensor and retain the corresponding data. Regarding the sampling time, the advantage of this embodiment lies in forming the residual field through explicit angular geometric operations, which provides a reproducible computational basis for residual partitioning and robust fitting. The residual tensor index and field meanings are shown in Table 1.

[0024] Table 1 Residual Tensor Index and Field Meanings

[0025] In a preferred embodiment, the candidate space generation unit determines the candidate space boundary based on the direction bundle envelope formed by continuous angle of arrival observations. The direction bundle envelope is jointly determined by multiple three-dimensional direction constraints within a continuous sampling window. The system can define the direction bundle centerline as the normalized weighted average direction of several observation direction vectors, define the direction bundle boundary as the angular extension range of each observation direction relative to the centerline, and then generate a radially extended neighborhood along the centerline direction, a horizontal candidate neighborhood along the azimuth disturbance direction, and an altitude candidate neighborhood along the pitch disturbance direction. The main direction neighborhood is used to retain candidate positions close to most observation directions, the radially extended neighborhood is used to retain candidate positions with depth uncertainty along the same angular ray, and the angular edge neighborhood is used to retain candidate positions near the azimuth or pitch angle boundaries that may be affected by reflection. The angle consistency calculation unit calculates the normalized values ​​in different neighborhoods. The angle residual is processed and the normalized angle residual and the neighborhood identifier are written together into the residual tensor. The normalization process can perform scale transformation based on the residual distribution of candidate positions within the same sampling frame, so that the residual values ​​between different sampling frames are comparable. The residual tensor organization unit further binds the belonging relationship of the candidate position in the direction bundle envelope with the residual value, so that the subsequent residual field partitioning is not only based on the residual size, but also on the positional relationship of the low residual region in the main direction neighborhood, radial extension neighborhood or angle edge neighborhood. In this embodiment, since the pseudo low residual region formed by the reflection path is usually distributed in the angle edge neighborhood or local patches with residual breaks with the main direction neighborhood, the direction bundle envelope and neighborhood identifier can provide spatial prior for pseudo convergence identification. The advantage of the embodiment is that it reduces irrelevant candidate regions from entering the fitting process and retains the spatial structure information carried by the single anchor point angle bundle itself.

[0026] Furthermore, the residual tensor organization unit may include a scale consistency labeling subunit and a candidate hierarchy compression subunit. The scale consistency labeling subunit generates a scale consistency label for candidate positions within the radially extended neighborhood based on the slope of the residual change in the radial extension direction. The slope of the residual change describes whether the residual is smooth when the candidate position near the same direction bundle changes radially in depth. If the residual change is small when moving radially, it indicates that the angular information has limited distance resolution in that direction. If the residual changes abruptly in the radial direction, it indicates that the vicinity of that position may be affected by abnormal angles or boundary effects. The candidate hierarchy compression subunit performs hierarchical compression of the three-dimensional candidate space based on the neighborhood identifier, normalized angular residual, and scale consistency label. Specifically, it can compress candidates with similar residual distributions and scale consistency labels within the same neighborhood. Adjacent candidate positions with the same name are merged into representative positions, and the mapping relationship between candidate positions before compression and representative positions after compression is saved. Subsequent fitting completes the main iterations at the representative position level. When locating the output, the mapping relationship is used to backtrack to the original candidate position set. The compression process does not change the physical coordinate meaning of the candidate positions, nor does it discard the marked abnormal boundary information. Instead, it makes the residual field partitioning and fitting calculations revolve around the representative residual structure. In this embodiment, scale consistency marking can prevent radial depth uncertainty from being misjudged as reflection pseudo-convergence. Candidate level compression can reduce the interference of irrelevant candidate positions on the residual continuity analysis. The advantage of this embodiment is that it forms a clearer candidate level while preserving the residual field structure, so that subsequent partitioning and fitting results can be backtracked to the original candidate space.

[0027] In one embodiment, reference Figure 3The residual field partitioning module includes a local continuity analysis unit, a residual transition detection unit, and a pseudo-convergence region identification unit. The local continuity analysis unit establishes adjacency relationships between candidate positions based on spatial location indices in the residual tensor and generates continuous residual patches based on the consistency of residual gradients between adjacent candidate positions. The residual transition detection unit detects residual break boundaries between adjacent continuous residual patches and determines whether the break boundaries are caused by differences in stable propagation paths based on changes in boundary position and direction in consecutive sampling frames. The pseudo-convergence region identification unit marks candidate patches where the residual values ​​are locally converged but do not match the temporal direction changes as pseudo-convergence regions, and outputs continuous residual patches, abnormal residual patches, and pseudo-convergence regions as the residual field partitioning result. Specifically, the local continuity analysis unit can establish an edge between each candidate position and its spatially adjacent candidate positions. The edge attributes include the difference between adjacent residuals, The residual gradient direction and candidate position neighborhood identification are used. If the residual gradient changes smoothly between multiple candidate positions and the low residual region moves continuously with the sampling frame, the corresponding region can be used as a candidate for effective direction region. If the residual change between adjacent regions suddenly increases and the low residual region is spatially isolated or disconnected from the main direction neighborhood, the corresponding region enters the abnormal direction region candidate. The pseudo-convergence region identification unit further compares whether the position change of the region in the continuous sampling frames is consistent with the change of the observation direction bundle. If the residual of the region remains low but the spatial position is inconsistent with the overall migration of the direction bundle, it is marked as a pseudo-convergence region. In this embodiment, the partitioning module converts multi-path identification from single-frame confidence judgment to spatial residual structure and temporal boundary structure judgment. The advantage of this embodiment is that it can identify stable and high-confidence reflection angle observations, reducing the situation where reflection observations are treated as ordinary direct observations in the solution.

[0028] Preferably, the local continuity analysis unit is configured according to the formula. Calculate the residual gradient magnitude between candidate positions, where Indicates the first Frame 1 The candidate position and the first The residual gradient magnitude between adjacent candidate positions and These represent the combined angular residuals of the two candidate positions relative to the same sampling frame. This represents the spatial distance between two candidate locations. This represents the absolute value operation. , , ,but This indicates that the residual change between two adjacent candidate positions is relatively gradual; if the residual of another adjacent candidate position is... Under the same spatial distance, corresponding This indicates that there is a significant residual transition between two points. The local continuity analysis unit can connect multiple low-gradient edges to form a continuous region and use high-gradient edges as candidate break boundaries between continuous regions. The residual transition detection unit then combines the persistence of the candidate break boundaries in different sampling frames for screening, avoiding directly judging local residual anomalies caused by single-frame angular noise as stable reflection boundaries. In this embodiment, the residual gradient magnitude is not used as a separate fixed threshold judgment basis, but is used together with the neighborhood identifier, cross-frame persistence, and the positional relationship of low residual areas to form a partitioning basis. The advantage of this embodiment is that it can distinguish between continuous residual regions and abrupt change regions through a calculable residual gradient structure, so that the identification of multipath anomalies has a clear data processing path.

[0029] Furthermore, the residual transition detection unit may include a cross-frame boundary association subunit. This subunit generates a boundary association chain based on the overlap of residual break boundaries in different sampling frames and the consistency of their orientations. The pseudo-convergence region identification unit filters persistent reflection candidate regions based on the boundary association chain and marks short-lived residual break boundaries that do not form a boundary association chain as instantaneous angular perturbation boundaries. The boundary association chain can be formed by connecting break boundary segments from multiple sampling frames in chronological order. The connection conditions include the overlapping regions of the boundary segments in the candidate space, the angular relationship between the boundary normal directions, and the migration direction of low-residual regions in adjacent frames. If a break boundary surrounds the same isolated low-residual region in multiple frames... If the isolated region persists and remains disconnected from its main directional neighborhood, the pseudo-convergence region identification unit marks it as a persistent reflection candidate region. If the break boundary appears only in a single frame or a few discontinuous sampling frames, and the residual field at the corresponding position recovers continuity in subsequent sampling frames, it is marked as an instantaneous angular perturbation boundary. Both the persistent reflection identifier and the instantaneous perturbation identifier are written into the residual field partitioning result. In this embodiment, the introduction of the boundary association chain enables the system to distinguish between long-term stable multipath phenomena and short-term noise, avoiding excessive weighting caused by single-frame residual anomalies. The advantage of this embodiment is that it improves the partitioning result's ability to distinguish reflection paths and provides a more detailed basis for subsequent fitting weight allocation. The residual field partitioning types and judgment criteria are shown in Table 2.

[0030] Table 2 Residual Field Zoning Types and Judgment Criteria

[0031] In one embodiment, the pseudo-convergence region identification unit can also use a consistency metric between low residual regions and beam migration for judgment, with the specific formula as follows: ; in Indicates the first Temporal inconsistency of low residual regions This represents the set of sampling times in which the low residual area continuously occurs. This represents the spatial migration vector of the center of the region in adjacent sampling frames. This indicates the direction of change of the observation direction vector between adjacent sampling frames. This represents the number of sampling frames within the time set. If the low residual patch moves with the observation direction, the normalized inner product of the two changing directions is approximately equal to... The value in parentheses is close to If the region remains near the reflection direction and is inconsistent with the direction beam migration, the normalized inner product is smaller, and the value within the parentheses is larger. For example, when the values ​​within the parentheses calculated for a certain region in three adjacent sampling intervals are respectively... , and ,but This indicates that the region is relatively consistent with the direction beam migration relationship, while the corresponding values ​​for another region are respectively , and ,but This indicates that it is more likely to be a pseudo-converged region that is inconsistent with the true direction of the target's movement. In this embodiment, the center of the region in the formula can be obtained by weighting the candidate positions within the region according to the low residual weight, and the direction change vector is obtained by the difference of the continuous observation direction vector. The advantage of this embodiment is that it establishes the pseudo-convergence judgment on the correspondence between spatial region migration and angle bundle change, so that regions with continuous low residuals but abnormal sources can be identified.

[0032] In one embodiment, reference Figure 4The robust fitting module includes an observation weight generation unit, a joint residual modeling unit, and an iterative position solving unit. The observation weight generation unit generates fitting weights for each frame's angle of arrival observation based on the residual field partitioning results, angle confidence, and sampling time sequence relationships. The joint residual modeling unit combines angle residuals, temporal continuous residuals, and spatial feasibility residuals into a position fitting target. The iterative position solving unit updates the fitting weights according to the partition to which the candidate position belongs after each round of solving, and generates the target's 3D coordinates based on the updated fitting weights. Specifically, the observation weight generation unit may include a partition weight subunit, a temporal weight subunit, and a confidence fusion subunit. The partition weight subunit generates basic weights based on the assignment of the candidate position corresponding to the angle of arrival observation in the effective direction region, abnormal direction region, and pseudo-convergence region. The temporal weight subunit generates weights based on the confidence fusion between consecutive sampling frames. The consistency of directional change generates time-series adjustment weights. The confidence fusion subunit fuses the basic weights, time-series adjustment weights, and angle confidence into fitting weights and writes the fitting weights into the joint residual modeling unit. After the iterative position solution unit obtains the target 3D coordinates in one round, the weight write-back subunit feeds back the candidate position assignment partition corresponding to the target 3D coordinates to the observation weight generation unit. If the coordinates fall into the pseudo-convergence region or migrate to the abnormal direction region, the relevant observation weights are reduced in the next round of fitting. If the coordinates are located in the effective direction region and the joint residual distribution remains continuous, the relevant observations retain the main constraints. In this embodiment, the fitting weights are not static quality scores, but constraint quantities that change together with the residual field partitioning, solution path, and time-series consistency. The advantage of this embodiment is that it avoids continuous reflection observations from dominating the position solution for a long time due to high angle confidence.

[0033] Preferably, the observation weight generation unit can generate fitted weights by fusing partition weights, time-series weights, and confidence scores, as specified in the formula: ; in Indicates the first Normalized fitting weights for frame angle of arrival observations. Indicates the first Frame angle confidence, This represents the partition weight obtained from the residual field partitioning results. This represents the temporal weights derived from the consistency of directional changes in consecutive sampling frames. This represents the set of sampled frames within the current fitting window. The denominator represents the normalized fusion weights applied to all sampled frames within the window. If the current window contains 3 observations, the weights of the 3 frames are normalized. They are respectively , and Then the normalized denominator is The corresponding fitting weights are , and This indicates that even if the second frame has usable angular observations, its position fitting is less affected by weaker partitioning or temporal consistency; partitioning weights... Different types of weights, including time-series weights, can be obtained from the effective direction region, abnormal direction region, pseudo-convergence region, or instantaneous perturbation boundary based on the candidate position assignment. The direction vector of this frame can be determined based on whether its direction change is continuous with that of adjacent frames. The formula does not use any single indicator as the sole basis, but rather requires that angle confidence, spatial partitioning, and temporal continuity jointly determine the fitting contribution. In this embodiment, the normalized... The weights are written into the joint residual modeling unit and participate in the position solution. The advantage of this embodiment is that it forms a traceable weight generation path, so that the contributions of the reflection angle and the short-term disturbance angle in the fitting are distinguished and processed.

[0034] Furthermore, the fitting objective constructed by the joint residual modeling unit can be expressed as: ; in Representing candidate 3D coordinates The joint residual objective Indicates the first Fitting weights for frame angle of arrival observations. Indicates the candidate 3D coordinates and the first Angular residuals between frame observation direction constraints The weighting coefficients represent the weighting coefficients of the time-series continuous residuals. This indicates the reference position obtained from the previous sampling state or the reference state of the adjacent window. The weighting coefficients represent the spatial feasibility residuals. This represents the spatial feasibility residual term generated when a candidate 3D coordinate falls into an abnormal orientation region or an unusable candidate region. If the residuals of a candidate coordinate relative to the angle observations of 3 frames are respectively... , , The corresponding weight is , , Then the angular residual is If the squared distance between the candidate coordinate and the reference position is ,and The time-series continuous term is If the candidate coordinates do not fall into the abnormal direction region and The joint residual is In this embodiment, the iterative location solving unit can calculate at the candidate representative location level. Then, local updates are performed along continuous regions with lower residuals. The advantage of this embodiment is that it unifies angular consistency, temporal continuity, and spatial partitioning constraints into a single solution objective, so that the location solution is no longer directly dominated by a single low-residual reflection area. The fitting weight generation fields and writing relationships are shown in Table 3.

[0035] Table 3. Fitting weight generation fields and their writing relationships

[0036] In one embodiment, the iterative location solving unit includes a weight write-back subunit, a convergence path recording subunit, and a location confidence generation subunit. After each round of fitting, the weight write-back subunit writes back the candidate location assignment partition corresponding to the target's 3D coordinates to the observation weight generation unit. The convergence path recording subunit records the migration state of the target's 3D coordinates between the effective direction region, the abnormal direction region, and the pseudo-convergence region during continuous iterations. The location confidence generation subunit generates the location confidence corresponding to the target's 3D coordinates based on the final fitted weight distribution, the joint residual distribution, and the migration state. Specifically, the iterative location solving unit can select initial candidate coordinates from a low-residue continuous region in the residual field. After one round of joint residual solving, updated coordinates are obtained. Then, the assignment of the updated coordinates in the residual field partitioning results is queried. If the updated coordinates move from the effective direction region to the pseudo-convergence region, the weight write-back subunit assigns the relevant... The partition identifier of the sampling frame is updated to a pseudo-convergence traction state. The observation weight generation unit reduces the base weight of this type of observation in the next round of weight generation. If the updated coordinates are always located in the effective direction region and the residual distribution shows continuous convergence, the convergence path recording subunit records it as a stable path. The positioning confidence generation subunit can generate confidence by combining whether the final weights are concentrated in the effective direction observations, whether the joint residuals are dominated by a single abnormal frame, and whether the convergence path crosses the pseudo-convergence region. The confidence does not replace the coordinate output, but together with the coordinates, it describes the solution basis of the positioning result in the residual field. In this embodiment, the recording of the migration state enables the system to discover the process of coordinate solution being pulled by the abnormal region, rather than just judging whether the final residual is small. The advantage of this embodiment is that it makes the positioning output have an interpretable abnormal source mark, which facilitates the subsequent continuous positioning link to filter and constrain low confidence results.

[0037] Preferably, the reliability generation sub-unit can calculate the reliability using residual concentration, regional stability, and convergence migration state, as shown in the following formula: ; in Indicates the credibility of the location. Represents a logical mapping function, This indicates the proportion of the final fitted weights that fall within the valid direction region. This indicates the stability of the coordinates remaining within the effective directional region during continuous iteration. This represents the migration penalty when the convergence path traverses regions of abnormal directions or spurious convergence. , , These represent the fusion coefficients of the three types of indicators, with fixed meanings as the effective area weight coefficient, the area stability coefficient, and the anomaly migration coefficient. If in a certain positioning... , , and take , , Then the logical mapping input is ,get Approximately This indicates that the positioning coordinates are mainly supported by the effective direction area. If another positioning... , , The logical mapping input is ,get Approximately This indicates that the convergence path is significantly affected by abnormal regions. In this embodiment, the inputs for positioning confidence all come from the residual field partition, fitting weights, and convergence path records. No data fields unrelated to positioning are introduced. The advantage of this embodiment is that the coordinate output is accompanied by a computable confidence criterion, reducing the direct transmission of low-confidence 3D coordinates in the continuous positioning link.

[0038] In a preferred embodiment, the residual field construction module, the residual field partitioning module, and the robust fitting module can operate collaboratively within a sliding sampling window. The sliding sampling window includes the current sampling frame and several historical sampling frames. The angle of arrival observation processing module retains the angle confidence, sampling sequence, and abnormal observation markers when updating the window. The direction constraint generation module generates three-dimensional direction constraints for newly added sampling frames within the window and reuses the direction vectors of existing sampling frames. The residual field construction module only updates the residual tensor for candidate positions affected by the newly added direction constraints. The residual field partitioning module reanalyzes local continuous regions and residual fracture boundaries in the updated residual field. The robust fitting module generates new fitting weights based on the updated partitioning results and outputs the current three-dimensional coordinates. Specifically, if a low-residual region in the previous window has been marked as a persistent reflection region, the same boundary condition will still appear in the new window. If the link is established, the partition identifier is retained. If the region disappears in a subsequent window and the residual field becomes continuous again, the persistent reflection identifier is terminated and recorded as a historical anomaly. This process enables the system to adapt to the occurrence, persistence, and fading of multipath states, avoiding the permanent application of historical anomaly identifiers to subsequent normal observations. The sliding window can also use the positioning reliability and convergence migration state of the previous window as a reference for the generation of the candidate space of the current window, so that the current window prioritizes retaining candidate positions adjacent to the reliable positioning area, while still retaining the angle edge neighborhood to prevent changes in the true direction of the target. In this embodiment, the windowing process does not change the basic framework of the single anchor point arrival angle fitting, but rather makes the residual field partitioning and weight writing back maintain temporal consistency in the continuous positioning process. The advantage of this embodiment is that it enables the system to distinguish between stable reflections, short-term disturbances, and normal direction changes in the continuous positioning process.

[0039] Furthermore, to avoid the impact of residual scale differences in different regions on partitioning decisions, the residual field construction module can normalize the comprehensive angle residual within each sampling frame and retain the unnormalized residual for location output verification. The normalized residual can be calculated using the following formula: ; in Indicates the first The candidate position is in the Normalized angle residuals under frame, This represents the composite angle residual before normalization. Indicates the first The low-bit reference value of the synthesized angle residual within the frame candidate space. Indicates the first High-order reference value of the composite angle residual within the frame candidate space This indicates a positive number used to avoid a denominator of zero, if within a certain sampling frame , , The residual at a certain candidate position is ,but The residual of the other candidate position is ,but The normalized result is used for relative comparison of candidate regions within the same frame and for comparison of residual field morphology between different frames. (Before normalization) Still used for calculating physical angle error in joint residual targets, in this embodiment, the low-order reference value and the high-order reference value can be determined by the quantile statistics in the residual distribution to reduce the influence of extreme anomaly candidate positions on scale transformation. The advantage of this embodiment is that the cross-frame residual field partitioning is not affected by the difference in the magnitude of single-frame residuals, while retaining the original angle residuals for the final solution.

[0040] In one embodiment, the spatial feasibility residual term It can be generated based on the region to which the candidate coordinates belong, the candidate level compression mapping, and the boundary association chain status. If the candidate coordinates are located in the valid direction region and the corresponding representative position is fully mapped to the original candidate position set, then... Take the lowest penalty value. If the candidate coordinates are located in an abnormal direction region or a persistent reflection area, then... A higher penalty value is applied based on its relationship with the anomaly boundary. If the candidate coordinates are located near the instantaneous perturbation boundary but no boundary association chain is formed in subsequent sampling frames, then... Temporary penalties are applied and recalculated with window updates. Spatial feasibility residuals are not used to exclude all candidate locations in abnormal regions, but rather to control the competitiveness of these candidate locations in the joint fitting target. This prevents reflective candidate regions from gaining an unreasonable global advantage due to low local angular residuals. In specific implementation, the candidate-level compression sub-unit retains the mapping relationship between the representative position and the original candidate position. The iterative position solving unit calculates the spatial feasibility residual at the representative position level and backtracks to the low residual sub-region in the original candidate position set when outputting coordinates. If the representative position covers multiple original candidate positions and the internal residual distribution is inconsistent, the positioning output module can select the original candidate position with the lowest joint residual in the set that does not cross the continuous reflection boundary as the output coordinate. In this embodiment, spatial feasibility residuals and residual field partitioning jointly limit the solution path, so that local pseudo-low residuals no longer determine the coordinate results independently. The advantage of this embodiment is that it enhances the correspondence between three-dimensional coordinates and continuous regions of the residual field.

[0041] In a preferred embodiment, the residual field partitioning module can establish sub-residual fields for the azimuth residual and the pitch residual respectively, and analyze the sources of horizontal and vertical offsets outside the integrated residual field. Specifically, if a candidate region is continuous in the azimuth residual sub-field and exhibits a boundary transition in the pitch residual sub-field, the system can mark the region as a vertical anomaly candidate region. If a candidate region is continuous in the pitch residual sub-field and exhibits a boundary transition in the azimuth residual sub-field, the system can mark the region as a horizontal anomaly candidate region. If both sub-residual fields show isolated low residuals and form a cross-frame boundary correlation chain, the system classifies it as a persistent reflection candidate region. When fusing weights, the observation weight generation unit can determine the source of the horizontal and pitch residuals based on the anomaly type. The residuals are assigned different contributions, so that altitude drift and horizontal offset are constrained separately in the joint residual target. For example, near the candidate area of ​​vertical anomaly, the pull of pitch residual on coordinate solution can be reduced, while the azimuth direction can still retain a certain constraint, avoiding the discarding of the whole frame observation due to a single angle dimension anomaly. In this embodiment, the azimuth residual subfield, pitch residual subfield and comprehensive residual field all use the same spatial location index and sampling time index. Therefore, the area attribution can be directly compared between different residual fields. The positioning output module can also write the anomaly type corresponding to the output coordinates into the positioning confidence description field. The advantage of the embodiment is that it refines the source of angle anomaly, enabling the system to handle situations where the reflection in the altitude direction and the reflection in the horizontal direction are different.

[0042] In one embodiment, the overall workflow of the system can be realized through the transfer relationship of data objects between modules. The arrival angle observation processing module outputs angle observation records, the direction constraint generation module converts the angle observation records into direction constraint records, the candidate space generation unit generates candidate space records based on the direction constraint records and the previous window's confidence coordinates, the angle consistency calculation unit combines the candidate space records and the direction constraint records into residual records, the residual tensor organization unit organizes the residual records into residual tensors, the residual field partitioning module converts the residual tensors into partition records, the observation weight generation unit converts the partition records, angle confidence, and temporal relationships into fitting weight records, the joint residual modeling unit generates a joint residual objective based on the fitting weight records, the iterative position solving unit outputs three-dimensional coordinate records, convergence path records, and write-back identifiers, and the positioning output module outputs the three-dimensional coordinates... The data objects, including records, location confidence, and anomaly observation markers, are used as the final output. These data objects can be associated using a unified time index and spatial location index, preventing the partitioning results from being unable to be written back due to index misalignment between angle observations, candidate positions, and fitting weights. In engineering implementation, candidate positions can be stored as coordinate vectors, directional constraints as unit direction vectors, residual tensors as arrays arranged continuously by spatial index, partition records as label fields corresponding to the spatial index, and fitting weight records as numerical fields corresponding to the sampling time index. In this embodiment, the modules do not rely on additional hardware structures but instead complete multipath suppression-type 3D positioning through data structures and algorithmic logic. The advantage of this embodiment is that it discloses a complete data stream from the original angle of arrival observation to the 3D coordinate output, enabling the system to implement according to a defined data processing logic.

[0043] In one embodiment, if the angle of arrival observation processing module receives consecutive sampling frames containing observations with low angle confidence or abnormal sampling states, these observations are not directly deleted. Instead, they are entered into the residual field construction module with observation status markers. The angle consistency calculation unit can still calculate the angle residuals between these observations and candidate positions. When associating cross-frame boundaries, the residual field partitioning module reduces the contribution of the frame to the formation of the continuous boundary based on the observation status markers. When fusing weights, the observation weight generation unit reflects the observation status markers in the angle confidence or temporal weights, thus constraining the impact of low-confidence observations on the fitting target. If the frame containing the low-confidence observation still forms a continuous residual patch with the preceding and following high-confidence observations, it can serve as an auxiliary constraint in the fitting process. If a low-confidence observation causes an isolated break without forming a boundary correlation chain, it is marked as a transient disturbance. Specifically, the system processes low-confidence observations through residual field partitioning, rather than eliminating all low-confidence observations at the front end using fixed rules. This is because the confidence of the direct angle during some occlusion transformation processes may drop temporarily, but its spatial residual structure is still continuous with the real direction region. Direct elimination would weaken the temporal constraints. Conversely, some reflection angles may have high confidence, but the residual field structure is inconsistent with the direction beam migration, so they should be downweighted at the back end. In this embodiment, both low-confidence observations and high-confidence reflection observations are processed through the same residual field mechanism. The advantage of this embodiment is that it avoids erroneous deletion or retention caused by relying solely on angle confidence.

[0044] In one embodiment, the system can perform inter-window consistency verification on continuous positioning output. The positioning output module stores the three-dimensional coordinates, positioning confidence, effective direction region identifier, and pseudo-convergence region identifier of the current window as historical positioning records. When the next window runs, the candidate space generation unit reads the coordinates with higher confidence in the historical positioning records as the reference for the candidate space center, but still regenerates the boundary based on the newly added angle of arrival direction bundle. The residual field partitioning module reads the historical pseudo-convergence region identifier and checks whether the corresponding position still forms a low residual isolated area in the new residual field. If the historical pseudo-convergence region is consistent with the continuous reflection area in the new window, the observation weight generation unit reduces the relevant observation contribution in advance in the initial fitting weight of the new window. If the historical pseudo-convergence region is consistent with the continuous reflection area in the new window, the observation weight generation unit reduces the relevant observation contribution in advance in the initial fitting weight of the new window. If the convergence region no longer forms a low residual area in the new window, the relevant identifier will no longer participate in the current weight generation. The convergence path recording sub-unit can also compare whether the output coordinates of adjacent windows continuously cross abnormal direction areas. If the crossing process is accompanied by an increase in joint residual and a decrease in the weight of the effective direction area, the positioning confidence generation sub-unit reduces the output confidence of the corresponding window. In this embodiment, the consistency check between windows does not replace positioning with smoothing, but uses the residual field partition and convergence path of the previous window as auxiliary judgment information for the current window. The main body of the residual field is still determined by the arrival angle observation of the current window. The advantage of the embodiment is that the continuous reflection path is stably identified in multiple windows, while avoiding long-term interference of the abnormal identifier of the old window to the current normal observation.

[0045] In one embodiment, to ensure consistency of residual field partitioning results under different coordinate system transformations, the orientation constraint generation module performs coordinate system unification processing on the azimuth and pitch angles before generating the 3D orientation constraints. This processing includes converting angle observations in the single-anchor-point local coordinate system into direction vectors in the positioning calculation coordinate system and saving the version identifier corresponding to the transformation matrix. The residual field construction module calculates the angular residuals using the transformed direction vectors. The positioning output module saves the transformation version identifier used when outputting the 3D coordinates. If the transformation relationship between the single-anchor-point coordinate system and the positioning calculation coordinate system is updated, the system can regenerate the orientation constraints and reconstruct the residual field instead of directly reusing the residual tensor under the old transformation relationship. Specifically, the orientation vector transformation can be achieved by... It means that, among them This represents the direction vector in the positioning calculation coordinate system. This represents the rotation matrix from the local coordinate system at a single anchor point to the coordinate system used for positioning calculation. This represents the direction vector in the local coordinate system of a single anchor point. If it is an identity matrix, then the direction vectors are the same before and after the transformation. Indicates rotation about the height axis Original direction vector After conversion In this embodiment, all directional constraints within the same residual tensor use the same transformation version. The advantage of this embodiment is that it avoids false fracture boundaries or pseudo-convergence patches in the residual field caused by inconsistencies in the coordinate system.

[0046] In one embodiment, the system can also generate abnormal observation markers during the positioning output stage. These abnormal observation markers include persistent reflection markers, transient disturbance markers, vertical anomaly markers, and horizontal anomaly markers. The abnormal observation markers are jointly determined by the residual field partitioning module and the convergence path recording subunit. If a low residual patch corresponding to a sampling frame is located in a persistent reflection patch and the coordinates are repeatedly pulled towards the pseudo-convergence region during the fitting process, the corresponding sampling frame is marked as a persistent reflection observation. If a sampling frame only forms a residual break in a single frame and does not form a boundary correlation chain, the corresponding sampling frame is marked as a transient disturbance observation. If the sub-residual field shows an abnormal pitch residual but a normal azimuth residual, the system can also generate abnormal observations. If the residuals are continuous, they are marked as vertical anomalies. If the sub-residual field shows azimuth residuals as abnormal but pitch residuals as continuous, they are marked as horizontal anomalies. The positioning output module outputs these marks along with the target's 3D coordinates, positioning confidence, and final fitting weight distribution. Subsequent positioning processing can select whether to use the current coordinates to participate in trajectory continuity constraints based on the marks. In this embodiment, all anomaly observation marks are generated by the angle of arrival residual field and fitting process data, without introducing business scenarios or external sensing data. The advantage of this embodiment is that it enables the positioning results to reflect the type and source of abnormal angle observations, providing a technical basis for coordinate screening in the continuous positioning process.

[0047] In a preferred embodiment, if the target is in a state of small angular change, the direction bundle envelope formed by continuous sampling frames may be narrow, and the residual change in the radially extended neighborhood in the residual field is relatively gentle. The system can identify this type of depth uncertainty through the scale consistency marking subunit and avoid misclassifying radially gentle low residual regions as pseudo-convergence regions. Specifically, the scale consistency marking subunit compares the residual slopes of different radial depth candidate positions in the same principal direction neighborhood. If the low residual region extends continuously in the radial direction and remains connected to the principal direction neighborhood, it is marked as a radial scale uncertainty region. In this region, the joint residual modeling unit mainly relies on temporally continuous residuals and historical references. If a low residual region has a residual break boundary with its main direction neighborhood and is stable across frames, it is still treated as a pseudo-convergence region. The differentiation process enables the system to handle the inherent radial uncertainty of single anchor point angle observations and not to regard all low residual regions existing radially as reflection anomalies. In this embodiment, radial scale uncertainty regions participate in fitting, but their positioning reliability is affected by scale stability. The positioning output module can use them as the region stability input in reliability generation. The advantage of this embodiment is that it takes into account the geometric constraints of single anchor point angle positioning and the requirements of multi-path recognition, and reduces the confusion between radial depth uncertainty and reflection pseudo-convergence.

[0048] In one embodiment, each module of the system can be deployed as functional logic running within the same positioning processing program. The angle of arrival observation processing module receives angle observation records output from a single anchor point via a data interface. The direction constraint generation module takes the angle observation records as input and outputs a direction vector array. The residual field construction module takes the direction vector array and candidate space records as input and outputs a residual tensor. The residual field partitioning module takes the residual tensor as input and outputs a partition label array. The robust fitting module takes the partition label array, residual tensor, and angle confidence array as input and outputs three-dimensional coordinates and fitting status. The positioning output module encapsulates the three-dimensional coordinates and fitting status into a positioning result record. The aforementioned functional logic can be completed by the processor executing program instructions in the storage medium, or by the positioning service process calling modules. Implementation does not require changes to the antenna structure, communication structure, or shell structure of a single anchor point. All innovations are reflected in the angle-of-arrival data processing, residual field partitioning, fitting weight generation, and iterative solution logic. In this embodiment, collaboration between functional logics is achieved through explicit data object transfer. Candidate positions, direction vectors, residuals, partition labels, and fitting weights all have traceable indexes. The advantage of this embodiment is that it ensures the software implementation path is consistent with the engineering deployment method, facilitating the replacement of conventional fitting processes in existing single-anchor-point angle-of-arrival positioning systems.

[0049] In one embodiment, after the system completes a localization, it can save the correspondence between the current residual field partitioning result and the output coordinates as a localization process record. This record includes a sampling window identifier, an orientation constraint array, candidate spatial boundaries, a residual tensor summary, effective orientation region labels, abnormal orientation region labels, pseudo-convergence region labels, a fitted weight sequence, a convergence path, and localization confidence. The localization process record is used for subsequent debugging, anomaly verification, and continuous window weight initialization, but it does not change the calculation result of the current localization coordinates. If repeated pseudo-convergence of the same orientation region occurs in subsequent windows, the system can read the historical boundary association chain through the localization process record and apply it to the current residual field. The system verifies whether the historical identifier still exists. Only when the corresponding structure is formed again in the current residual field will the historical identifier participate in the current weight generation, thus avoiding incorrect weight reduction caused by simply relying on historical labels. The positioning process record can also be used to perform offline calibration of the fusion coefficients in the formula. During offline calibration, the correctness of the residual field partition, the continuity of coordinates, and the consistency of the anomaly markers are used as evaluation objects, rather than the confidence of a single angle. In this embodiment, the positioning process record does not introduce new positioning inputs, but only retains intermediate data in the internal calculation process of the system. The advantage of this embodiment is that it improves the reproducibility of the algorithm operation process and makes the reflection area identification have consistent data basis in the continuous window.

[0050] In one embodiment, if the residual field partitioning module detects multiple low-residual continuous regions simultaneously, the robust fitting module can use a multi-candidate path parallel evaluation method to solve the location problem. Specifically, it generates initial candidate coordinates for each low-residual continuous region and calculates the joint residual target, partition weight distribution, and convergence path migration state. The positioning confidence generation sub-unit generates confidence for each candidate path. The positioning output module selects candidate paths with lower joint residuals and whose convergence paths do not traverse persistent reflection regions to output three-dimensional coordinates. If a candidate path has low angular residuals but its fitting weights mainly come from pseudo-convergence regions or other factors, the module can resolve the issue. If the convergence path remains in the abnormal direction region for a long time, it will not be used as the priority output path. If multiple candidate paths do not meet the support conditions of the effective direction region, the positioning output module will output coordinates with low confidence and retain the abnormal observation mark. In this embodiment, the multi-candidate path evaluation is still based on the same residual field and the same fitting target. It is not a simple repetition of multiple positioning runs, but takes the source difference of different low residual regions in the residual field as the basis for path selection. The advantage of this embodiment is that it reduces the risk of misselection when the global lowest residual is occupied by the reflection area, and makes the final coordinate output more consistent with the continuous structure and temporal structure of the residual field.

[0051] In one embodiment, if a sudden change in direction occurs in the angle of arrival observation within a continuous sampling window, the system uses residual transition detection and temporal weights to jointly determine its impact on the positioning results. The angle of arrival observation processing module retains the angle confidence and time index of the sudden frame, the direction constraint generation module generates the corresponding direction constraints, and the residual field construction module calculates the residual distribution under that frame. If the low residual region caused by the sudden frame is disconnected from the effective direction region of the preceding and following frames in the candidate space and does not form a cross-frame boundary association chain, the residual field partitioning module marks it as an instantaneous perturbation boundary, and the observation weight generation unit reduces the impact of the sudden frame on the joint [position / location] using temporal weights. The influence of residual targets: if the directional beams migrate to the new spatial region for multiple consecutive frames after the abrupt change frame, and the low residual continuous area is consistent with the change of the new directional beam, then the residual field partitioning module will not continue to use the instantaneous disturbance marker, but will include the new continuous area in the candidate effective directional region. In this embodiment, the system will not treat all angle abrupt changes as abnormal reflections, nor will it accept all abrupt changes as the real motion of the target. Instead, it will determine the type of the abrupt change through the cross-frame structure of the residual field and the directional beam migration relationship. The advantage of this embodiment is that it takes into account the real direction change of the target and the non-line-of-sight instantaneous disturbance, and reduces the coordinate jump error in continuous positioning.

[0052] In one embodiment, the system uses a unified naming convention and fixed meaning for the intermediate variables in residual field partitioning, weight generation, and location output. Always indicates the first The spatial direction vector corresponding to the frame angle of arrival observation. Always indicates the first The candidate position is relative to the first The combined angular residuals of frame orientation constraints Always represent the normalized composite residual. Always represents the residual gradient magnitude between adjacent candidate positions. Always indicates the first Normalized fitting weights for frame angle of arrival observations. Always represent the joint residual objective of the candidate 3D coordinates, To consistently represent location reliability, the system maintains the physical meaning of these variables when passing them between modules, avoiding the same symbol representing different objects in different modules. Specifically, in implementation, the residual field construction module outputs... The residual field partitioning module is used for continuity analysis, the joint residual modeling unit is used for angular residual term calculation, and the observation weight generation unit outputs... The joint residual modeling unit is used for weighted fitting, and the positioning confidence generation subunit is used for calculating the weight ratio of the effective direction region. The variable flow enables the data processing chain to have a consistent computational basis. In this embodiment, all formulas correspond to data objects that can be obtained within the system, and the formula results can be traced through candidate positions, sampling frames, and region labels. The advantage of this embodiment is that it reduces symbolic ambiguity and module interface ambiguity in algorithm implementation, enabling the system to reproduce the actual positioning process according to the computational logic in the specification.

Claims

1. A single-anchor-point three-dimensional positioning system based on signal angle of arrival fitting, comprising: The system comprises an angle of arrival observation and processing module, a direction constraint generation module, a residual field construction module, a residual field partitioning module, a robust fitting module, and a positioning output module. The angle of arrival observation and processing module receives azimuth, pitch, and angle confidence values ​​collected from a single anchor point. The orientation constraint generation module generates three-dimensional orientation constraints based on the azimuth angle and the pitch angle, and the residual field construction module calculates the angular residual between the candidate position and the three-dimensional orientation constraint in the three-dimensional candidate space and forms a residual field. The residual field partitioning module divides the effective direction region and the abnormal direction region according to the spatial continuity and abrupt change characteristics of the residual field. The robust fitting module performs weighted fitting on the angle of arrival observations based on the effective direction region and the abnormal direction region. The positioning output module outputs the system of the target's three-dimensional coordinates.

2. The single-anchor-point three-dimensional positioning system based on signal angle of arrival fitting according to claim 1, characterized in that, The residual field construction module includes a candidate space generation unit, an angle consistency calculation unit, and a residual tensor organization unit. The candidate space generation unit generates a three-dimensional candidate space containing radial candidate points, horizontal candidate points, and height candidate points based on the coordinate system of the single anchor point. The angle consistency calculation unit calculates the azimuth residual and pitch residual of each candidate position relative to the three-dimensional directional constraints of each frame. The residual tensor organization unit organizes the residual data according to the spatial position index, sampling time index, and angle type index, so that the residual changes of the same candidate position in consecutive sampling frames are preserved as a partitionable data structure.

3. The single-anchor-point three-dimensional positioning system based on signal angle of arrival fitting according to claim 1, characterized in that, The residual field partitioning module includes a local continuity analysis unit, a residual transition detection unit, and a pseudo-convergence region identification unit. The local continuity analysis unit generates continuous residual patches based on the consistency of residual gradients between adjacent candidate positions. The residual transition detection unit generates abnormal residual patches based on the abrupt change in residual distribution between consecutive sampling frames. The pseudo-convergence region identification unit marks candidate patches where the residual values ​​are locally converged but do not match the temporal direction change as pseudo-convergence regions, and outputs the continuous residual patches, the abnormal residual patches, and the pseudo-convergence regions as the residual field partitioning results.

4. A single-anchor-point three-dimensional positioning system based on signal angle of arrival fitting according to claim 1, characterized in that, The robust fitting module includes an observation weight generation unit, a joint residual modeling unit, and an iterative position solving unit. The observation weight generation unit generates fitting weights for each frame's angle of arrival observation based on the residual field partitioning results, the angle confidence level, and the sampling time sequence relationship. The joint residual modeling unit combines the angle residual, the temporal continuous residual, and the spatial feasibility residual into a position fitting target. After each round of solving, the iterative position solving unit updates the fitting weights according to the partition to which the candidate position belongs, and generates the target's three-dimensional coordinates based on the updated fitting weights.

5. A single-anchor-point three-dimensional positioning system based on signal angle of arrival fitting according to claim 2, characterized in that, The candidate space generation unit determines the candidate space boundary based on the direction bundle envelope formed by continuous angle of arrival observations, and divides the candidate positions within the direction bundle envelope into a main direction neighborhood, a radial extension neighborhood, and an angle edge neighborhood. The angle consistency calculation unit calculates the normalized angle residuals in the main direction neighborhood, the radial extension neighborhood, and the angle edge neighborhood, respectively. The residual tensor organization unit writes the normalized angle residuals and the corresponding neighborhood identifiers into the residual tensor. The residual field construction process includes the spatial attribution information of the candidate positions relative to the direction bundle shape.

6. A single-anchor-point three-dimensional positioning system based on signal angle of arrival fitting according to claim 3, characterized in that, The local continuity analysis unit establishes an adjacency graph for candidate positions in the residual field and generates continuous region labels based on the consistency of the residual gradient directions of adjacent candidate positions. The residual transition detection unit marks the cross-region residual break boundary in the adjacency graph and tracks the persistence of the residual break boundary according to the sampling time index. The pseudo-convergence region identification unit marks the set of candidate locations with stable local low residuals and residual break boundaries between them and adjacent continuous regions as reflection candidate patches, which are part of the abnormal direction region.

7. A single-anchor-point three-dimensional positioning system based on signal angle of arrival fitting according to claim 4, characterized in that, The observation weight generation unit includes a partition weight subunit, a temporal weight subunit, and a confidence fusion subunit. The partition weight subunit generates basic weights based on the assignment of the candidate position corresponding to the angle of arrival observation in the effective direction region, the abnormal direction region, and the pseudo-convergence region. The temporal weight subunit generates temporal adjustment weights based on the consistency of direction changes between consecutive sampling frames. The confidence fusion subunit fuses the basic weights, the temporal adjustment weights, and the angle confidence into fitting weights and writes the fitting weights into the joint residual modeling unit.

8. A single-anchor-point three-dimensional positioning system based on signal angle of arrival fitting according to claim 5, characterized in that, The residual tensor organization unit further includes a scale consistency marker subunit and a candidate level compression subunit. The scale consistency marker subunit generates a scale consistency marker based on the slope of the residual change of the candidate position in the radial extension direction. The candidate level compression subunit performs hierarchical compression on the three-dimensional candidate space based on the neighborhood identifier, the normalized angular residual, and the scale consistency marker, and retains the mapping relationship between the candidate position before compression and the representative position after compression.

9. A single-anchor-point three-dimensional positioning system based on signal angle of arrival fitting according to claim 3, characterized in that, The residual transition detection unit also includes a cross-frame boundary association subunit, which generates a boundary association chain based on the positional overlap and boundary orientation consistency of the residual break boundaries in different sampling frames. The pseudo-convergence region identification unit filters persistent reflection candidate regions based on the boundary association chain, and marks the residual fracture boundary that appears briefly and does not form a boundary association chain as an instantaneous angular perturbation boundary, so that the residual field partitioning result includes persistent reflection identifiers and instantaneous perturbation identifiers.

10. A single-anchor-point three-dimensional positioning system based on signal angle of arrival fitting according to claim 4, characterized in that, The iterative location solving unit includes a weight back-writing subunit, a convergence path recording subunit, and a location confidence generation subunit. After each round of fitting, the weight back-writing subunit writes back the candidate location assignment partition corresponding to the target's three-dimensional coordinates to the observation weight generation unit. The convergence path recording subunit records the migration state of the target's three-dimensional coordinates between the effective direction region, the abnormal direction region, and the pseudo-convergence region during continuous iterations. The location confidence generation subunit generates the location confidence corresponding to the target's three-dimensional coordinates based on the final fitted weight distribution, the joint residual distribution, and the migration state.

Images