Deep learning-based phased array radar signal adaptive optimization algorithm

Abstract

The invention discloses a phased array radar signal adaptive optimization algorithm based on deep learning, and relates to the technical field of radar signal processing, and the algorithm comprises the following steps: S1, collecting an electromagnetic propagation environment disturbance intensity curve in a detection task, superposing the disturbance intensity curve to an original radar echo point by point, and generating a radar echo set with a disturbance mark; according to the invention, by introducing a polar phase spiral traction structure based on disturbance response characteristics, adaptive return and stable correction of a radar main lobe in a complex propagation environment are realized, and distance errors caused by path jump are significantly reduced; and meanwhile, a dynamic rhythm constraint and traction path guiding mechanism is combined to construct a beam regulation and control framework with interference resistance and continuity, so that stable control from disturbance sensing to direction recovery is realized, and the target recognition precision and the system reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing technology, and more specifically to an adaptive optimization algorithm for phased array radar signals based on deep learning. Background Technology

[0002] The deep learning-based adaptive optimization algorithm for phased array radar signals refers to an intelligent technology approach that tightly integrates artificial intelligence with phased array radar signal processing. Its core idea is to utilize a deep learning model to analyze the raw echo received by the radar in real time. Under complex environments such as noise interference, electromagnetic spurious signals, and strong echo obstruction, it automatically performs noise suppression, clutter removal, and multi-target separation. Combined with the spatiotemporal reference provided by BeiDou positioning, it dynamically adjusts the beam direction, weighting coefficients, and array configuration of the phased array radar under changing environmental parameters such as weather, sea state, atmospheric refraction, and electromagnetic interference, ensuring the radar maintains optimal detection capabilities in complex application scenarios. Through this method, the radar can achieve higher target detection accuracy and tracking stability in dynamic environments such as maritime surveillance, civil aviation guidance, and special detection, reducing manual parameter tuning and improving the system's intelligence and adaptability. This demonstrates the novel technological advantages of AI-driven automated signal optimization and parameter adjustment.

[0003] The existing technology has the following shortcomings: In complex electromagnetic environments, strong spatiotemporal disturbances occur in the atmosphere, causing electromagnetic wave propagation paths to abruptly change within a very short time. When the detection wave of a phased array radar traverses this unstable medium, the originally continuous and stable propagation path exhibits abnormal backtracking, disrupting the spatial hierarchy of the echo signal. When analyzing the disturbed echo data, deep learning models are prone to misinterpreting the reflection characteristics of distant targets as those of near-range targets, resulting in confusion in the model's internal spatial feature representation. Consequently, the radar system may exhibit significant deviations in target identification and distance determination, potentially misidentifying non-threatening targets as nearby threats, thus triggering erroneous warnings and control commands.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide an adaptive optimization algorithm for phased array radar signals based on deep learning to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a deep learning-based adaptive optimization algorithm for phased array radar signals, comprising the following steps: S1, during the detection mission, collect the electromagnetic propagation environment disturbance intensity curve, and superimpose the disturbance intensity curve point by point onto the original radar echo to generate a radar echo set with disturbance markers, which is used to provide input data for subsequent propagation path anomaly analysis; S2, read the change in disturbance intensity from the radar echo set with disturbance markers, segment the continuous radar echo segments according to the fluctuation characteristics of the disturbance intensity in the time dimension, identify the propagation path breakover abrupt change section in each segment, and construct the breakover risk index band to mark the source region of spatial propagation distortion; S3, input the fold risk index band into the deep learning processing link, lock the feature channels corresponding to the intermediate layers of the deep learning network through the mutation segments marked in the index band, and generate a list of long-distance feature misalignment based on the locked feature channels to provide a basis for correcting the beam adjustment strategy; S4, rearrange the beam pointing update order according to the long-range feature misalignment list, and restrict the adjustment rhythm of the short-range pointing during the update process, so that the beam pointing adjustment forms an anti-bend beam adjustment baseline, which is used to establish a stable dynamic control reference. S5, along the anti-bend beam adjustment baseline, a poloidal phase spiral traction ring is deployed. When the intensity of electromagnetic disturbance increases rapidly, the poloidal phase spiral traction ring is used to implement segmented traction, so that the main lobe pointing is gradually moved back to the real target reflection position along the spiral traction path, thereby realizing distance recovery and false alarm suppression under propagation anomalies.

[0007] Preferably, step S1 includes: Deploy environmental sensing receiving equipment with broadband reception capabilities to obtain disturbance characteristics in electromagnetic propagation media; The electromagnetic propagation environment disturbance intensity curve is interpolated, smoothed, and normalized, and then mapped to the radar working coordinate system; The disturbance intensity curve is time-aligned and fused with the original radar echo data point by point to generate a radar echo sequence with disturbance markers. The echo samples are reordered based on the changes in perturbation intensity and perturbation gradient weights are added to construct a highly robust input echo set for propagation path anomaly identification.

[0008] Preferably, step S2 includes: Based on radar echo sets with disturbance intensity markers, a disturbance intensity time series is constructed for continuous echo samples according to the acquisition time sequence, and the disturbance inflection point is identified according to the rate of change of disturbance intensity. The echo sequence is then divided into multiple time segments with independent dynamic characteristics. Propagation path stability was assessed for each time segment, and abrupt transitions with propagation path breakpoints were identified by analyzing the offset distance and offset slope of the main lobe energy distribution. Combining the characteristics of disturbance intensity variation and spatial offset, the start and end frames of the abrupt change segment are timestamped, and the disturbance peak, duration, spatial offset amplitude and energy form variability are extracted to construct the break-through risk unit. Based on the temporal continuity and spatial overlap characteristics of risk units, they are aggregated to form risk clusters, which are then added to the original radar echo sequence to generate a fold risk index band with spatiotemporal pointing characteristics.

[0009] Preferably, the propagation path stability assessment includes constructing a main lobe energy distribution map within each time segment, determining the breakover region by calculating the energy offset distance and offset slope in the azimuth and pitch directions, and identifying the segment as a propagation path breakover abrupt change segment when the energy distribution offset of consecutive echo frames exceeds a preset threshold and continues for several frames, and using it to generate breakover risk units.

[0010] Preferably, step S3 includes: Radar echo samples with timestamps consistent with the jump risk index band are selected as input sequences. Echo samples with jump risk markers are input into the deep learning processing link, and abrupt change segments in the jump risk index band are loaded simultaneously to guide feature response localization. In deep learning processing, the response of the feature channels in each layer of the network is monitored. By analyzing the matching relationship between the temporal response gradient and the perturbation features of the breakpoint, abnormal response channels are identified, and feature channel groups in the intermediate layer that are highly correlated with the propagation path anomalies are determined. Based on the locked feature channels, the spatiotemporal distribution information of the feature response is extracted and the channel feature clustering spectrum is constructed. The long-distance misaligned channels are identified through a multi-channel cross-validation mechanism and a long-distance feature misalignment list is generated. The list of long-distance feature misalignment layers is mapped and associated with the breakpoint risk index band to establish a spatiotemporal correspondence between propagation anomalies and feature responses, and the output parameters are used for beam adjustment strategy correction.

[0011] Preferably, in deep learning processing, spatiotemporal feature peak analysis is performed on the locked feature channels. The far-distance misalignment position is determined by the feature peak frame number and spatial positioning parameters. Near-distance high-energy false reflection structures are identified in the channel feature distribution map. The channel group that maintains the consistency of spatial offset direction and response amplitude is taken as the far-distance misalignment channel set to improve the accuracy and stability of beam adjustment.

[0012] Preferably, step S4 includes: Read the feature channel number, spatial offset position and response frame number in the long-range feature misalignment list, and construct the misalignment intensity distribution map by combining the response intensity of each channel, which is used to identify the directional segment and target distance segment that are significantly affected by disturbances within the radar scanning range; Based on the high-risk areas in the layer intensity distribution map, the beam pointing update order is rearranged, and a time buffer window is set in the low-risk near-distance direction to delay the update rhythm, so that the far-distance direction forms a priority repair and the near-distance direction forms a response inertia update mechanism. Based on the update sequence adjustment, an anti-jump adjustment weight distribution table is introduced. The stability rating of beam pointing is determined according to the response density of the staggered channel and the spatial energy stability. An energy transition gap is set to prevent signal abrupt changes caused by pointing jumps. After completing the design of beam update sequence and rhythm constraints, the dynamic evolution trajectory of the main lobe pointing in three-dimensional space is extracted and fitted as an anti-bend beam adjustment baseline, which is used as a dynamic control reference and deviation limit for real-time beam adjustment.

[0013] Preferably, when constructing the layered intensity distribution map, the beam coverage angle and target distance are normalized according to the two parameters, and the layered risk level is divided by energy density gradient. When introducing the anti-bend adjustment weight distribution table, a multiple confirmation mechanism is set for the beam pointing with the lowest stability rating, and a fixed-duration energy buffer gap is inserted during the pointing switching process to ensure that the main lobe achieves a smooth transition during the update process, thereby improving the continuity and adjustment stability of the beam path.

[0014] Preferably, step S5 includes: Based on the trajectory parameters of the anti-bend beam adjustment baseline in three-dimensional space, a poloidal spiral traction path is constructed, and a poloidal spiral traction structure composed of multiple adjacent spiral rings is formed to guide the main lobe to gradually return to the real target reflection position along a predetermined path. Directional traction command points with phase control capability are set up on the key nodes of the spiral ring in the spiral traction path to form a polar phase spiral traction ring array. The main lobe direction is fine-tuned by controlling the phase difference and interval between the traction points. When the intensity of electromagnetic disturbance increases rapidly, segmented traction is implemented based on the angle difference between the current position of the main lobe and the target spiral node, so that the main lobe advances step by step along the spiral traction path and forms a smooth and continuous return trajectory. After completing the spiral traction path, the trajectory data of the traction node in three-dimensional space is extracted, the lowest offset point of the main lobe trajectory is determined, and the stable target position of the main lobe is constructed, which is used as the initial benchmark and dynamic control reference for the next cycle beam adjustment.

[0015] Preferably, the traction action of the poloidal phase spiral traction ring is dynamically adjusted according to the rate of change of disturbance intensity, and the traction angle and step size are adjusted accordingly. When the rate of increase of disturbance intensity increases, the rotation density of the spiral path is increased synchronously to enhance the main lobe trajectory constraint capability. During the traction process, the smooth transition of the main lobe direction is achieved through continuous phase compensation between adjacent traction rings, so that the main lobe maintains stable angle response and balanced energy distribution during segmented return, thereby further improving the pointing accuracy and return consistency of the main lobe under strong disturbance environment.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention achieves adaptive backtracking and stable correction of the radar main lobe pointing under complex propagation environments by introducing a poloidal phase spiral traction structure based on disturbance response characteristics during beam modulation. This structure utilizes the synergistic effect of the fold risk index band and the long-range feature misalignment list to enable the main lobe to dynamically correct along the anti-fold beam adjustment baseline, thereby effectively eliminating the impact of abrupt propagation path changes caused by electromagnetic disturbances. The step-by-step backtracking of the main lobe pointing along the spiral traction path allows the radar to maintain high-precision spatial positioning and target recognition capabilities even under strong disturbance conditions, significantly reducing range errors caused by echo misalignment and improving the accuracy of long-range reflective target identification.

[0017] This invention applies dynamic rhythm constraints and traction path guidance mechanisms to the beam update sequence, enabling the beam adjustment process to be both disturbance-resistant and continuous, thus constructing a dynamic control framework with physical constraints. This scheme allows the main lobe to maintain a smooth energy distribution and stable angular response even during propagation anomalies, avoiding pointing drift and false alarms caused by sudden jumps. The radar system thus achieves end-to-end stable control from disturbance detection and path compensation to direction recovery, significantly improving beam steady-state adjustment capabilities and target identification reliability in complex propagation environments. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 This is a flowchart of the method for adaptive optimization algorithm of phased array radar signals based on deep learning according to the present invention. Detailed Implementation

[0020] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0021] This invention provides, for example Figure 1 The deep learning-based adaptive optimization algorithm for phased array radar signals, as shown, includes the following steps: S1, during the detection mission, collect the electromagnetic propagation environment disturbance intensity curve, and superimpose the disturbance intensity curve point by point onto the original radar echo to generate a radar echo set with disturbance markers, which is used to provide input data for subsequent propagation path anomaly analysis; When phased array radar performs detection missions, in order to effectively identify and resist signal path abrupt changes caused by electromagnetic propagation environmental disturbances, it is necessary to construct a radar echo data input set with environmental awareness capabilities. The following is a specific implementation method for acquiring and fusing electromagnetic propagation environmental disturbance intensity curves: An environmental sensing receiving device with broadband reception capability is deployed to continuously acquire disturbance characteristics in the electromagnetic propagation medium near the radar's operating frequency band. By frequently sampling the rate of change of charge density and the amplitude of field strength fluctuations in the propagation medium, corresponding disturbance intensity curves are generated in real time. The sampling frequency is set to high-frequency continuous acquisition to ensure that every instantaneous fluctuation point is fully recorded when the propagation disturbance changes rapidly. To improve the spatiotemporal stability of the disturbance intensity curves, multiple sets of environmental disturbance data curves distributed at different azimuth and elevation angles are acquired simultaneously within the radar's operating cycle. Finally, through interpolation, smoothing, and normalization processing, the multi-point disturbance curves are uniformly mapped to the radar's operating coordinate system, forming a complete set of spatiotemporally synchronized disturbance intensity curves.

[0022] After obtaining the disturbance intensity curve covering the entire observation period, the curve is uniformly discretized along the time axis. Each discrete sampling point is used as a disturbance marker, and time-aligned and fused with the original radar echo data point by point. During the fusion process, the timestamp of each frame of radar echo data must strictly match the disturbance value at the corresponding time point in the disturbance curve, ensuring that each echo sample is appended with a disturbance amplitude value. In this way, each frame of echo sample in the radar reception sequence is given synchronized environmental disturbance information, forming an echo sequence with disturbance markers. For example, when the radar frame frequency is a fixed sampling rate, each frame of echo sample corresponds to several disturbance sampling points. The disturbance intensity value of each frame is calculated by moving average and appended to the sample, thereby generating a preliminary labeled radar echo set.

[0023] Based on radar echo sets with disturbance markers, multi-channel spatial signal domain projection and energy distribution analysis are performed on the disturbance intensity values ​​in each echo sample frame. The degree of disturbance impact is calculated by the angular difference between the disturbance intensity value and the beam receiving direction. When the intensity value of a disturbance point is significantly higher than the average value of that point under steady-state conditions, the sample is identified as a high-disturbance point and marked as a first-level disturbance in the echo data. All first-level disturbance-marked samples are prioritized for analysis in subsequent propagation path anomaly identification to identify potential path transition regions. Taking a continuous observation sequence as an example, the high-disturbance samples identified in hundreds of echo samples are usually concentrated in the early stage of the disturbance outbreak period, thus forming a disturbance surge band in the time series.

[0024] To enhance the sensitivity and accuracy of subsequent path anomaly analysis, the labeled radar echo sequences are reordered from low to high perturbation intensity values, placing high-perturbation samples at the forefront of the input sequence to highlight the impact of environmental changes on signal propagation. A perturbation gradient weight is introduced during the reordering process. This weight is calculated based on the difference in perturbation intensity between adjacent samples and describes the rate of perturbation change. If the proportion of high-gradient samples exceeds a threshold within the working period, the sequence is defined as a high-dynamic perturbation sequence and prioritized for analysis in subsequent path anomaly detection. Ultimately, the radar echo set containing perturbation intensity values, perturbation labels, and perturbation gradient weights forms a highly robust input set for propagation path anomaly identification and fold feature extraction. This set possesses complete environmental perturbation and radar response coupling characteristics, providing a reliable input foundation for subsequent depth signal optimization and beam adaptive adjustment.

[0025] S2, read the change in disturbance intensity from the radar echo set with disturbance markers, segment the continuous radar echo segments according to the fluctuation characteristics of the disturbance intensity in the time dimension, identify the propagation path breakover abrupt change section in each segment, and construct the breakover risk index band to mark the source region of spatial propagation distortion; To extract propagation anomaly regions from radar echoes marked with disturbances and to provide a precise spatial positioning basis for subsequent path-jump analysis, temporal structure reshaping and spatial abrupt change identification of the echo sequence are required. The specific implementation method of this process is as follows: Based on the generated radar echo set with disturbance intensity markers, the disturbance intensity values ​​of each frame of echo data are time-series processed, constructing a set of disturbance sequences arranged at equal time intervals according to the acquisition time order of consecutive echo frames. During the construction process, the average disturbance change gradient is calculated by statistically analyzing the disturbance intensity fluctuation range of any consecutive echo samples, and this is used as a dynamic reference value. In this sequence, when the disturbance intensity change rate of consecutive frames exceeds twice the dynamic reference value, it is marked as a disturbance inflection point. The intervals between inflection points constitute disturbance change segments, and each change segment is defined as a time segment with independent dynamic characteristics. In this way, the radar echo samples within a continuous working cycle can be divided into multiple disturbance time segments of unequal length, each segment corresponding to a specific propagation dynamic state.

[0026] Propagation path stability is assessed for each defined time segment. Representative echo samples are selected from each time segment, and a two-dimensional energy distribution map is constructed using spatial interferometry. The offset distance and slope of the main lobe energy in the azimuth and elevation directions are statistically analyzed. When the energy distribution offset exceeds the normal path variance threshold for several consecutive frames within a time segment, a propagation path breakpoint is identified in that segment. The energy distribution maps of these suspected breakpoint segments are further analyzed to extract the main lobe morphology change curve and calculate the degree of deformation. When the average offset angle and offset slope of the main lobe are significantly higher than the average under stable propagation conditions, it is confirmed as a propagation path breakpoint abrupt change segment, reflecting local abnormal reflection characteristics of the propagation path.

[0027] After identifying the abrupt transition sections in the propagation path, each transition section is finely delineated by combining the temporal variation characteristics and spatial offset features of the disturbance intensity, and its start and end frames are marked with timestamps. Further, multi-dimensional indicators such as disturbance peak value, duration, spatial offset amplitude, and energy pattern variability of each transition section are extracted and encapsulated into risk units. By analyzing the temporal continuity and spatial interference pattern similarity among risk units, transition regions with interrelated or shared propagation characteristics are screened and divided into several risk clusters according to their temporal frequency and spatial distribution overlap. Each risk cluster represents an aggregation region of propagation path transition trends, providing prior location information for subsequent in-depth feature analysis and beamforming strategies.

[0028] All identified hop risk clusters are numbered and sorted chronologically, and hop risk labels are added to the corresponding sample frames in the original radar echo sequence to generate a complete hop risk index band. This index band is organized synchronously in both temporal and spatial dimensions, providing temporal boundary information for the disturbance intensity change trajectory and hop events, while also recording the projected contours and deformation characteristics of the spatial hop risk region. The hop risk index band features high temporal resolution and spatial positioning accuracy, and is dynamically updated, expanding and adaptively adjusting in real time during radar operation. Ultimately, this index band forms a set of hop risk characterization sequences with spatiotemporal pointing characteristics, fully describing the path anomaly evolution process caused by propagation disturbances, providing a traceable and quantifiable input basis for subsequent deep signal analysis and beamforming.

[0029] S3, input the fold risk index band into the deep learning processing link, lock the feature channels corresponding to the intermediate layers of the deep learning network through the mutation segments marked in the index band, and generate a list of long-distance feature misalignment based on the locked feature channels to provide a basis for correcting the beam adjustment strategy; To achieve in-depth identification and adaptive correction of spatial misalignment in radar signals, the constructed fold risk index needs to be input into the deep learning processing flow. Combined with the response performance of intermediate layer feature channels, a long-range feature misalignment list is extracted, providing accurate decision-making basis for subsequent beam adjustment strategies. The specific implementation method of this process is as follows: Radar echo samples with timestamps consistent with the bounce risk index band are selected as input sequences for deep learning processing. These echo samples with bounce risk markers are input into a trained radar feature extraction network in the order of acquisition. To ensure that the input sequence covers the dynamic changes before and after the disturbance, the input samples are extended forward and backward along the time axis to capture the complete disturbance response process. Each frame of echo sample contains multi-dimensional signal information such as temporal echo, spatial distribution, polarization state, and spectral components. This information is mapped layer by layer into spatial feature representations in the network. Simultaneously, abrupt change segments in the bounce risk index band are loaded as intervention signals to guide the feature response localization during network processing, ensuring that the model can focus on response changes in spatially anomalous regions within specific time periods.

[0030] In deep learning processing, a set of feature channels highly correlated with the transition periods is extracted by monitoring the feature responses of each layer of the network channel by channel. Specifically, the temporal response gradients of each layer channel are statistically analyzed and matched with the perturbation characteristics of the transition section to identify channels exhibiting anomalous activation, decay, or reversal responses during the perturbation period. When the response value fluctuation of a channel in the transition section significantly exceeds the average level of the stable phase, and its spatial feature map shows significant energy concentration in the far-range reflection region, the channel is identified as a potential spatial misalignment channel. Through comparative analysis of multi-layer features, a group of intermediate layer channels that are anomalously highly correlated with the propagation path can be identified. These channels collectively reflect the structural misalignment characteristics of the signal in the distance and direction dimensions.

[0031] For identified anomalous channels, the spatiotemporal distribution information of channel feature responses is extracted, and features are classified and compared by constructing a channel feature clustering spectrum. Based on the time frame of the feature peak and its corresponding spatial positioning parameters, the location of long-range misalignment is determined. When a channel's response peak shifts significantly forward before and after a fold, and forms a near-range high-energy false reflection structure in the spatial feature distribution map, this channel is identified as a representative of long-range misaligned channels. To ensure the accuracy of the identification results, a multi-channel cross-validation mechanism is used to detect the consistency of feature drift across different channels, retaining only channel groups that maintain coordination in spatial offset direction and response amplitude. After summary analysis, a long-range feature misalignment list is generated. This list includes key parameters such as channel number, feature peak frame number, spatial offset distance, and disturbance correlation strength score, used to quantify the degree of spatial layer misalignment caused by propagation path disturbances.

[0032] Finally, the generated long-range feature misalignment list is mapped and associated with the bounce risk index band to establish a spatiotemporal correspondence between propagation anomalies and feature responses. This mapping relationship indicates the source of feature offsets caused by each bounce segment and their response channel positions in the deep learning network, providing precise input for beam adjustment strategies. Through this list, the radar control process can dynamically adjust the beam update rhythm based on the degree of misalignment and the intensity of disturbance, thereby suppressing erroneous beam pointing adjustments caused by path disturbances. Ultimately, the long-range feature misalignment list achieves a quantifiable description and traceable record of spatial misalignment phenomena, providing reliable data support for subsequent dynamic beam control and propagation stability adjustment.

[0033] S4, rearrange the beam pointing update order according to the long-range feature misalignment list, and restrict the adjustment rhythm of the short-range pointing during the update process, so that the beam pointing adjustment forms an anti-bend beam adjustment baseline, which is used to establish a stable dynamic control reference. To effectively address the spatial misalignment problem of long-range targets caused by propagation path folding disturbances, it is necessary to reorganize the beam pointing update sequence of the phased array radar based on the constructed long-range characteristic misalignment list, and apply rhythm constraints during the update process to establish a stable anti-disturbance beam adjustment baseline. The specific implementation method of this process is as follows: The locked feature channel numbers, spatial offset positions, and corresponding response frame numbers are read from the output long-range feature misalignment list. Combined with the response intensity of each channel during the disturbance, a misalignment intensity distribution map is constructed. This map, with the radar beam coverage angle on the horizontal axis and the target range on the vertical axis, overlays the energy density distribution of the misaligned regions to characterize the directional and range segments most significantly affected by disturbances within the entire radar scan range. To improve spatial resolution accuracy, constant-amplitude energy density contour lines are introduced into the map to divide misaligned regions of different intensity levels into layers. Multiple mission tests show that misalignment hotspots are typically concentrated in specific azimuths and mid-to-long-range regions, exhibiting characteristics of unstable mainlobe response and multiple false backtracking.

[0034] Based on the high-impact areas shown in the layering intensity distribution map, the original beam scanning path is adjusted, prioritizing high-risk areas with long-range layering errors in the update sequence to ensure that long-range pointing corrections are completed before the radar enters a period of strong disturbance. Simultaneously, a time buffer window is set for low-risk near-range areas within the scanning path. Within this window, the update rhythm of the beam direction is delayed to avoid prematurely adjusting the near-range direction before long-range corrections are completed, which could lead to overall spatial positioning structure disorder. In practice, this can be achieved by reordering the original beam scanning angle intervals, prioritizing directional segments with significant long-range errors, and delaying the update sequence of the near-range beam. This creates a priority repair mechanism for long-range directions and a reactive, inertial update mechanism for near-range directions, thereby constructing a pointing adjustment system with temporal stability.

[0035] To ensure structural stability of the beam update sequence and rhythm throughout the scanning cycle, an anti-flip adjustment weight distribution table is introduced based on the aforementioned adjustments. This table is calculated by combining the response density of the staggered channel and spatial energy stability, assigning a stability rating to each beam pointing position. The stability rating is divided into multiple levels, from low to high, corresponding to different beam adjustment priorities and delay parameters. When the stability rating for a certain direction is at the lowest level, multiple pointing confirmations are required during the update cycle, and an energy transition gap is set during the switching process. This allows for short-term smooth energy adjustment of the main lobe between old and new directions, preventing signal abrupt changes caused by pointing jumps. This energy buffering mechanism effectively reduces transient distortion during rapid scanning and improves the continuity and physical stability of the beam path.

[0036] After completing the beam update sequence reconstruction and rhythm constraint design, the beam adjustment trajectory throughout the entire scanning cycle was recorded and fitted to extract the dynamic evolution path of the main lobe pointing in three-dimensional space. A desired beam adjustment trend curve, i.e., the anti-fold beam adjustment baseline, was formed with time as the axis, angle as the latitude, and distance as the longitude. This baseline serves as a dynamic control reference, used to compare the actual pointing deviation during real-time beam adjustment. When the main lobe trajectory deviates from the baseline by more than a set threshold, a pointing limitation mechanism is immediately triggered to restrict rapid beam shift and prevent error accumulation and spatial misalignment expansion. Through the construction of this baseline, the radar achieved stable control of the main lobe pointing path in a complex electromagnetic disturbance environment. Test results show that after adopting this baseline, beam pointing deviation is significantly reduced, false alarm rate is lowered, and response speed is improved. This anti-fold beam adjustment baseline not only forms a stable spatial dynamic control reference but also provides continuous structural support and reliable physical calibration basis for subsequent beam optimization and depth signal correction.

[0037] S5, along the anti-bend beam adjustment baseline, a poloidal phase spiral traction ring is deployed. When the intensity of electromagnetic disturbance increases rapidly, the poloidal phase spiral traction ring is used to implement segmented traction, so that the main lobe pointing is gradually moved back to the real target reflection position along the spiral traction path, thereby realizing distance recovery and false alarm suppression under propagation anomalies.

[0038] To achieve precise return and range recovery control of the radar main lobe under strong electromagnetic disturbances, it is necessary to rely on the anti-bend beam adjustment baseline established in the previous stage, and deploy a poloidal phase spiral traction ring with controllable traction capability along its spatial path. This traction structure is then used to guide the main lobe pointing gradually along a predetermined trajectory to the actual target reflection position. The specific implementation method of this process is as follows: Based on the trajectory parameters of the generated anti-bend beam adjustment baseline in the three-dimensional coordinate system, a spatial topological skeleton for the poloidal spiral traction path is constructed. This path starts at the current pointing position of the radar main lobe and ends at the actual target reflection position marked in the long-range feature misalignment list, planning a spiral traction path that advances step-by-step along the poloidal dimension between the two. The spiral path consists of multiple adjacent spiral loops, each rotating around the polar axis by a certain angle and advancing towards the target with a fixed step size. The spiral loop radius is proportionally set according to the spatial offset of the main lobe, while the angle advancement step size is adjusted according to the disturbance intensity recorded in the previous bend risk index band. The stronger the disturbance, the higher the rotation density of the spiral path, thereby enhancing the trajectory constraint capability of the main lobe pointing. In this way, the pointing correction of the main lobe under strong disturbances can exhibit continuous and smooth poloidal return characteristics.

[0039] Along the spiral path framework, directional traction command points with phase control capabilities are deployed at key nodes of each spiral ring, forming a poloidal phase spiral traction ring array. Each traction ring array uses the main lobe synthesis center of the radar array antenna as its geometric reference, achieving fine-tuning control of the main lobe direction by precisely controlling the phase difference and spacing between the traction points. The number of traction command points in each ring array is set according to the array resolution to ensure that the main lobe can perform high-precision angular response adjustments within a local space. When the system detects that the electromagnetic disturbance intensity rapidly rises to the warning threshold within a short period, it automatically triggers the first stage of traction action in the spiral traction ring array. During traction, the main lobe uses the spiral path as its guide trajectory, achieving graded adjustments in space. Each traction angle remains within a small range, ensuring continuous pointing transitions and smooth responses, thereby avoiding energy dissipation and misjudgment caused by abrupt changes.

[0040] When electromagnetic disturbances continuously intensify or maintain high-level fluctuations, the system employs a segmented traction method to control the main lobe's gradual return along a spiral path. The angular difference between the main lobe's current position and the target spiral node is monitored in real time. When this difference falls below a set threshold, the system switches to the next spiral node, triggering the next traction command. This segmented mechanism limits the magnitude of a single adjustment, ensuring the uniqueness and gradualness of the adjustment direction, allowing the main lobe to form a stable return trajectory along the traction path. In continuous observation missions, the main lobe gradually approaches the true reflection position under multiple tractions, significantly reducing spatial offset errors, and the overall trajectory exhibits a spiral convergence trend. This method not only effectively prevents over-correction of the main lobe under disturbance conditions but also ensures the consistency of echo characteristics in both time and space, thereby maintaining radar ranging and positioning accuracy.

[0041] To further solidify the stability and reliability of the main lobe pointing, after completing the entire spiral traction path, the three-dimensional spatial coordinates of all traction nodes were fitted, and the lowest offset point on the main lobe trajectory was extracted as the final stable target position. This target position serves not only as the main lobe reference pointing for the current observation cycle but also as the initial benchmark for beam adjustment in the next cycle, ensuring that the main lobe can adaptively return along the predetermined trajectory under continuous disturbance conditions. Multiple verification missions revealed that adopting the poloidal phase spiral traction ring structure significantly enhances the dynamic stability of the main lobe pointing, significantly reduces the positioning error of distant targets, and effectively suppresses the system's false alarm rate. The final spiral traction structure achieves both physical guidance and spatial calibration for the main lobe return, constructing a radar pointing adaptive control mechanism that can dynamically respond to disturbance intensity and possesses directional traction and energy mitigation capabilities. This provides highly robust dynamic control support for accurate radar signal identification in complex propagation environments.

[0042] This invention achieves adaptive backtracking and stable correction of the radar main lobe pointing under complex propagation environments by introducing a poloidal phase spiral traction structure based on disturbance response characteristics during beam modulation. This structure utilizes the synergistic effect of the fold risk index band and the long-range feature misalignment list to enable the main lobe to dynamically correct along the anti-fold beam adjustment baseline, thereby effectively eliminating the impact of abrupt propagation path changes caused by electromagnetic disturbances. The step-by-step backtracking of the main lobe pointing along the spiral traction path allows the radar to maintain high-precision spatial positioning and target recognition capabilities even under strong disturbance conditions, significantly reducing range errors caused by echo misalignment and improving the accuracy of long-range reflective target identification.

[0043] This invention applies dynamic rhythm constraints and traction path guidance mechanisms to the beam update sequence, enabling the beam adjustment process to be both disturbance-resistant and continuous, thus constructing a dynamic control framework with physical constraints. This scheme allows the main lobe to maintain a smooth energy distribution and stable angular response even during propagation anomalies, avoiding pointing drift and false alarms caused by sudden jumps. The radar system thus achieves end-to-end stable control from disturbance detection and path compensation to direction recovery, significantly improving beam steady-state adjustment capabilities and target identification reliability in complex propagation environments.

[0044] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A deep learning-based adaptive optimization algorithm for phased array radar signals, characterized in that, Includes the following steps: S1, during the detection mission, collect the electromagnetic propagation environment disturbance intensity curve, and superimpose the disturbance intensity curve point by point onto the original radar echo to generate a radar echo set with disturbance markers. S2: Read the changes in disturbance intensity from the radar echo set with disturbance markers, divide the continuous radar echo segments according to the fluctuation characteristics of the disturbance intensity in the time dimension, identify the propagation path breakover abrupt change section in each segment, and construct the breakover risk index band. S3, input the bounce risk index band into the deep learning processing link, lock the feature channels corresponding to the intermediate layers of the deep learning network through the mutation segments marked in the index band, and generate a list of far-distance feature misalignment based on the locked feature channels; S4, rearrange the beam pointing update order according to the long-range feature misalignment list, and restrict the adjustment rhythm of the short-range pointing during the update process, so that the beam pointing adjustment forms an anti-bend beam adjustment baseline. S5, along the anti-bend beam adjustment baseline, a poloidal phase spiral traction ring is deployed. When the intensity of electromagnetic disturbance increases rapidly, segmented traction is implemented through the poloidal phase spiral traction ring, so that the main lobe pointing is gradually moved back to the actual target reflection position along the spiral traction path.

2. The deep learning-based adaptive optimization algorithm for phased array radar signals according to claim 1, characterized in that, Step S1 includes: Deploy environmental sensing receiving equipment with broadband reception capabilities to obtain disturbance characteristics in electromagnetic propagation media; The electromagnetic propagation environment disturbance intensity curve is interpolated, smoothed, and normalized, and then mapped to the radar working coordinate system; The disturbance intensity curve is time-aligned and fused with the original radar echo data point by point to generate a radar echo sequence with disturbance markers. The echo samples are reordered based on the changes in perturbation intensity and perturbation gradient weights are added to construct a highly robust input echo set for propagation path anomaly identification.

3. The deep learning-based adaptive optimization algorithm for phased array radar signals according to claim 2, characterized in that, Step S2 includes: Based on radar echo sets with disturbance intensity markers, a disturbance intensity time series is constructed for continuous echo samples according to the acquisition time sequence, and the disturbance inflection point is identified according to the rate of change of disturbance intensity. The echo sequence is then divided into multiple time segments with independent dynamic characteristics. Propagation path stability was assessed for each time segment, and abrupt transitions with propagation path breakpoints were identified by analyzing the offset distance and offset slope of the main lobe energy distribution. Combining the characteristics of disturbance intensity variation and spatial offset, the start and end frames of the abrupt change segment are timestamped, and the disturbance peak, duration, spatial offset amplitude and energy form variability are extracted to construct the break-through risk unit. Based on the temporal continuity and spatial overlap characteristics of risk units, they are aggregated to form risk clusters, which are then added to the original radar echo sequence to generate a fold risk index band with spatiotemporal pointing characteristics.

4. The deep learning-based adaptive optimization algorithm for phased array radar signals according to claim 3, characterized in that, The propagation path stability assessment includes constructing a main lobe energy distribution map within each time segment, determining the breakover region by calculating the energy offset distance and offset slope in the azimuth and elevation directions, and identifying the breakover abrupt change segment when the energy distribution offset of consecutive echo frames exceeds a preset threshold and continues for several frames. This segment is then used to generate breakover risk units.

5. The deep learning-based adaptive optimization algorithm for phased array radar signals according to claim 3, characterized in that, Step S3 includes: Radar echo samples with timestamps consistent with the jump risk index band are selected as input sequences. Echo samples with jump risk markers are input into the deep learning processing link, and abrupt change segments in the jump risk index band are loaded simultaneously to guide feature response localization. In deep learning processing, the response of the feature channels in each layer of the network is monitored. By analyzing the matching relationship between the temporal response gradient and the perturbation features of the breakpoint, abnormal response channels are identified, and feature channel groups in the intermediate layer that are highly correlated with the propagation path anomalies are determined. Based on the locked feature channels, the spatiotemporal distribution information of the feature response is extracted and the channel feature clustering spectrum is constructed. The long-distance misaligned channels are identified through a multi-channel cross-validation mechanism and a long-distance feature misalignment list is generated. The list of long-distance feature misalignment layers is mapped and associated with the breakpoint risk index band to establish a spatiotemporal correspondence between propagation anomalies and feature responses, and the output parameters are used for beam adjustment strategy correction.

6. The deep learning-based adaptive optimization algorithm for phased array radar signals according to claim 5, characterized in that, In deep learning processing, spatiotemporal feature peak analysis is performed on the locked feature channels. The location of the far-distance misalignment is determined by the feature peak frame number and spatial positioning parameters. Near-distance high-energy false reflection structures are identified in the channel feature distribution map. The channel group that maintains the consistency of spatial offset direction and response amplitude is used as the far-distance misalignment channel set to improve the accuracy and stability of beam adjustment.

7. The deep learning-based adaptive optimization algorithm for phased array radar signals according to claim 6, characterized in that, Step S4 includes: Read the feature channel number, spatial offset position and response frame number in the long-range feature misalignment list, and construct the misalignment intensity distribution map by combining the response intensity of each channel, which is used to identify the directional segment and target distance segment that are significantly affected by disturbances within the radar scanning range; Based on the high-risk areas in the layer intensity distribution map, the beam pointing update order is rearranged, and a time buffer window is set in the low-risk near-distance direction to delay the update rhythm, so that the far-distance direction forms a priority repair and the near-distance direction forms a response inertia update mechanism. Based on the update sequence adjustment, an anti-jump adjustment weight distribution table is introduced. The stability rating of beam pointing is determined according to the response density of the staggered channel and the spatial energy stability. An energy transition gap is set to prevent signal abrupt changes caused by pointing jumps. After completing the design of beam update sequence and rhythm constraints, the dynamic evolution trajectory of the main lobe pointing in three-dimensional space is extracted and fitted as an anti-bend beam adjustment baseline, which is used as a dynamic control reference and deviation limit for real-time beam adjustment.

8. The deep learning-based adaptive optimization algorithm for phased array radar signals according to claim 7, characterized in that, When constructing the intensity distribution map of the misalignment, the beam coverage angle and target distance are normalized according to the two parameters, and the risk level of the misalignment is divided by the energy density gradient. When introducing the anti-flip adjustment weight distribution table, a multiple confirmation mechanism is set for the beam pointing with the lowest stability rating, and a fixed-duration energy buffer gap is inserted during the pointing switching process to ensure that the main lobe achieves a smooth transition during the update process.

9. The deep learning-based adaptive optimization algorithm for phased array radar signals according to claim 7, characterized in that, Step S5 includes: Based on the trajectory parameters of the anti-bend beam adjustment baseline in three-dimensional space, a poloidal spiral traction path is constructed, and a poloidal spiral traction structure composed of multiple adjacent spiral rings is formed to guide the main lobe to gradually return to the real target reflection position along a predetermined path. Directional traction command points with phase control capability are set up on the key nodes of the spiral ring in the spiral traction path to form a polar phase spiral traction ring array. The main lobe direction is fine-tuned by controlling the phase difference and interval between the traction points. When the intensity of electromagnetic disturbance increases rapidly, segmented traction is implemented based on the angle difference between the current position of the main lobe and the target spiral node, so that the main lobe advances step by step along the spiral traction path and forms a smooth and continuous return trajectory. After completing the spiral traction path, the trajectory data of the traction node in three-dimensional space is extracted, the lowest offset point of the main lobe trajectory is determined, and the stable target position of the main lobe is constructed, which is used as the initial benchmark and dynamic control reference for the next cycle beam adjustment.

10. The deep learning-based adaptive optimization algorithm for phased array radar signals according to claim 9, characterized in that, The traction action of the poloidal phase spiral traction ring dynamically adjusts the traction angle and step size according to the rate of change of disturbance intensity. When the rate of increase of disturbance intensity increases, the rotation density of the spiral path is increased synchronously to enhance the main lobe trajectory constraint capability. During the traction process, the smooth transition of the main lobe direction is achieved through continuous phase compensation between adjacent traction rings, so that the main lobe maintains stable angle response and balanced energy distribution during segmented return, thereby further improving the pointing accuracy and return consistency of the main lobe under strong disturbance environment.

Images