Joint optimization method and system for network beamforming and target detection

Abstract

This application relates to the field of integrated sensing base station control technology, and discloses a method and system for joint optimization of beamforming and target detection in integrated sensing networks. The invention acquires the device-side operating data, communication feedback data, and sensing observation data of the target base station in the current time slot, constructs local array element cluster-level standardized state observation results, local array element cluster occupancy weights, and initial geometric observation results of candidate targets; performs graph-constrained recursive inversion based on the local array element cluster-level standardized state observation results to obtain the array distortion implicit state vector, and forms the local distortion intensity and the overall array distortion intensity; calculates the communication sensing consistency correction amount based on the array distortion implicit state vector, synchronously corrects the beamforming results and target detection results, obtains the joint correction result, updates the reference baseline and sensitivity mapping based on the high-confidence execution result using constrained recursion, and writes it back to the next time slot processing procedure.

Description

Technical Field

[0001] This application relates to the field of integrated sensing base station control technology, and in particular to a method and system for joint optimization of beamforming and target detection in integrated sensing networks. Background Technology

[0002] As next-generation mobile communication base stations develop towards large-scale antenna arrays, integrated sensing and communication processing, and real-time edge-side collaboration, active antenna arrays (AAUs) have been widely deployed in scenarios such as urban main roads, transportation hubs, park boundaries, port quay cranes, elevated roads, and the vicinity of large venues. In these scenarios, base stations need to continuously complete downlink precoding transmission, uplink signal reception, beam switching, and link maintenance for mobile terminals. On the other hand, they also need to utilize the same array resources to perform target direction estimation, distance sensing, motion state analysis, and abnormal target identification. Existing engineering implementations typically use factory amplitude and phase calibration, power-on self-calibration, periodic online calibration, and beam coarse adjustment based on bit error rate, channel quality indicators, or feedback matrices to maintain array transmission and reception consistency. For example, in high-load macro base stations or indoor high-power units, the system often maintains the availability of the precoding matrix based on the communication link quality and inserts the calibration process during off-peak hours to balance equipment operating efficiency and communication continuity. This type of solution can meet the coverage and throughput requirements in traditional communication services and has therefore become a common technical path in the current manufacturing and operation control of base station equipment.

[0003] However, under the conditions of inductive multiplexing RF front-end and co-array operation, existing technologies still struggle to address the persistent erosion of sensing geometric reliability by array online distortion. Specifically, power amplifier heat accumulation, phase shifter temperature drift, array element mutual coupling changes, power supply fluctuations, uneven local heat dissipation in the cabinet, and long-term occupancy of hot service directions all cause the same set of theoretical precoding weights to correspond to different actual spatial wavefront distributions in different operating phases. Due to the link adaptation and error correction redundancy on the communication side, some array distortion is absorbed, and the system superficially appears to have stable throughput and normal coverage. However, the sensing side relies on array direction consistency and phase geometry, making it more prone to target angle shift, decreased multi-target resolution, and static... Scattering objects are often misidentified as slowly moving targets, and weak dynamic features are sometimes mistaken for real motion due to thermal drift. This is especially true when uplink and downlink loads are unbalanced and the service beam is concentrated in a specific direction for extended periods. In such cases, local array element clusters can experience non-uniform drift. Existing uniform calibration, performed at the whole array or whole channel level, struggles to identify these local anomalies. Consequently, when the sensing results are fed back into the next round of beam control, the false bias is amplified. Therefore, how to finely characterize the array distortion state online and suppress its misleading effect on target detection results in a real base station operating environment, combining equipment characteristic data, communication feedback data, and array operating history, has become a critical technical problem that urgently needs to be solved in integrated sensing base station equipment. Summary of the Invention

[0004] This application proposes a joint optimization method and system for integrated sensing network beamforming and target detection to address the problems mentioned in the background art.

[0005] To achieve the above objectives, this application adopts the following technical solution: a joint optimization method for beamforming and target detection in a sensor-integrated network, comprising the following steps:

[0006] S1. Obtain the device-side operation data, communication feedback data and sensing observation data of the target base station in the current time slot. Divide the local array element clusters according to the array physical adjacency relationship, and construct the local array element cluster-level standardized state observation results, local array element cluster occupancy weights and candidate target initial geometric observation results.

[0007] S2, based on the local array element cluster-level standardized state observation results, combined with the local array element cluster occupancy weight, the physical adjacency relationship between local array element clusters, the reference baseline, the sensitivity mapping and the local distortion state estimation results of the previous time slot, performs graph-constrained recursive inversion to obtain the array distortion hidden state vector, and forms the local distortion intensity and the overall array distortion intensity.

[0008] S3, construct the array distortion compensation result around the array distortion implicit state vector, calculate the communication sensing consistency correction amount by combining the direction consistency loss of the initial direction of the candidate target in the service beam direction and the initial geometric observation result of the candidate target, and simultaneously correct the beamforming result and the target detection result based on the communication sensing consistency correction amount, the local distortion intensity and the overall array distortion intensity to obtain the joint correction result;

[0009] S4 outputs the joint correction results, and updates the reference baseline and sensitivity mapping based on the high-confidence execution results through limited recursion. The updated results are then written back to the local array element cluster-level normalized state observation results construction process and array distortion implicit state vector inversion process in the next time slot.

[0010] Furthermore, the specific operations in S1 for acquiring equipment-side operating data, communication feedback data, and sensing observation data and dividing local array element clusters are as follows: acquiring the average temperature, linear reflection power, power amplifier average current, and phase shifter control word corresponding to each local array element cluster as equipment-side operating data; acquiring the precoding usage frequency within the current window and the uplink and downlink equivalent channel state data after reciprocity calibration and subcarrier alignment as communication feedback data; and acquiring the angular spectrum distribution results and the initial geometric observation results of candidate targets as sensing observation data.

[0011] Based on the physical adjacency of the array, and combined with the distribution of shared power supply branches and shared heat dissipation paths, the local array element clusters are divided.

[0012] Furthermore, the specific operations for constructing the standardized state observation results and local array cluster occupancy weights in S1 are as follows: first, timestamp correction is performed on the equipment-side operation data, communication feedback data, and sensing observation data based on a unified time base; then, short-term missing samples are filled in based on adjacent time slot data, and data samples exceeding the preset operation boundary are removed.

[0013] Subsequently, the corresponding data of each local array element cluster were subjected to dimensional uniform conversion according to the reference temperature baseline, reference reflection power baseline, reference current baseline and reference phase command baseline. Among them, the phase shifter control word was converted into a phase command quantity through the code word phase lookup table and then participated in the standardization process. The uplink and downlink equivalent channel state data were converted into channel residual quantity through residual conversion. The angular spectrum distribution result was converted into angular spectrum perturbation coefficient through adjacent time slot correlation conversion.

[0014] Then, the local array element cluster occupancy weight is formed based on the precoding frequency and the current time slot transmission energy.

[0015] Furthermore, the specific operation of the graph-constrained recursive inversion in S2 is as follows: First, an initial estimate of the local distortion state of the current time slot is formed based on the reference baseline and the estimation result of the local distortion state of the previous time slot. Then, observation fitting constraints are established around the difference between the observation result of the normalized state of the local array element cluster and the sensitivity mapping output. Spatial continuity constraints are established around the difference of the local distortion state between physically adjacent local array element clusters. Time recursion constraints are established around the deviation between the estimation result of the local distortion state of the current time slot and the estimation result of the local distortion state of the previous time slot. Subsequently, the local distortion state of all local array element clusters is iteratively updated by combining the observation fitting constraints, spatial continuity constraints and time recursion constraints until the convergence condition is met, thereby obtaining the array distortion implicit state vector.

[0016] Furthermore, the specific operations for forming the local distortion intensity and the overall array distortion intensity in S2 are as follows: the amplitude drift ratio and normalized phase drift ratio in the local distortion state estimation results are used as local distortion characterization quantities, and the actual phase drift amount is calculated based on the normalized phase drift ratio and the upper bound of the phase drift.

[0017] After the array distortion implicit state vector is determined, the local distortion intensity is first formed based on the magnitude of the local distortion state estimation results. Then, the weighted aggregation of all local distortion intensities is performed in combination with the occupancy weight of each local array element cluster to form the overall array distortion intensity.

[0018] When the local distortion state estimation result exceeds the preset distortion boundary, the current time slot is marked as a strong distortion condition, and conservative correction processing is performed in the subsequent joint correction process.

[0019] Furthermore, the specific operation for calculating the communication-aware consistency correction in S3 is as follows: first, based on the ideal array direction vector and the array distortion compensation result, the directional consistency loss corresponding to the service beam direction and the directional consistency loss corresponding to the initial direction of the candidate target in the initial geometric observation result of the candidate target are formed; then, the communication-aware consistency correction is determined based on the aforementioned two directional consistency losses and the difference between them.

[0020] Furthermore, the specific operation of synchronously correcting beamforming results and target detection results in S3 is as follows: First, the corrected beamforming results are solved based on the communication sensing consistency correction amount, local distortion intensity, and overall array distortion intensity. Then, the basic detection threshold is jointly amplified and corrected based on the directional consistency loss corresponding to the initial direction of the candidate target in the initial geometric observation results of the candidate target and the overall array distortion intensity. Finally, the angle deviation compensation is performed on the target angle estimation results in the initial geometric observation results of the candidate target based on the sensitivity relationship between the array direction vector and the array distortion implicit state vector.

[0021] Furthermore, the specific operation of correcting the target detection result in S3 is as follows: First, construct the target spatial filtering weight based on the corrected target angle estimation result and the array distortion compensation result. Then, use the target spatial filtering weight to perform spatial filtering on the slow time echo sequence. Subsequently, perform Doppler spectrum reconstruction on the filtered slow time echo sequence and determine the corrected target radial velocity estimation result based on the main energy distribution of the Doppler spectrum. Finally, combine the corrected detection threshold, the corrected target angle estimation result, and the corrected target radial velocity estimation result to form the target detection result.

[0022] Furthermore, the specific operations for updating the reference baseline and sensitivity mapping based on the high-confidence execution results in S4 are as follows: First, the reference baseline is updated by a restricted recursive method based on the local array element cluster-level normalized state observation results and the execution results of the current time slot. Then, the sensitivity mapping is updated by a restricted recursive method based on the local distortion state estimation results of the current time slot and the difference between the local array element cluster-level normalized state observation results and the sensitivity mapping output. Finally, the updated reference baseline and sensitivity mapping are written back to the construction process of the local array element cluster-level normalized state observation results and the array distortion implicit state vector inversion process of the next time slot.

[0023] A joint optimization system for integrated sensing network beamforming and target detection includes:

[0024] The state observation construction module is configured to acquire the device-side operation data, communication feedback data and sensing observation data of the target base station in the current time slot, divide the local array element clusters according to the array physical adjacency relationship, and construct the local array element cluster-level standardized state observation results, local array element cluster occupancy weights and candidate target initial geometric observation results;

[0025] The distortion state inversion module is configured to perform graph-constrained recursive inversion based on the local array element cluster-level standardized state observation results, combined with the local array element cluster occupancy weight, the physical adjacency relationship between local array element clusters, the reference baseline, the sensitivity mapping, and the local distortion state estimation results of the previous time slot, to obtain the array distortion implicit state vector and form the local distortion intensity and the overall array distortion intensity.

[0026] The consistency joint correction module is configured to construct array distortion compensation results around the array distortion implicit state vector, calculate the communication sensing consistency correction amount by combining the direction consistency loss of the service beam direction and the candidate target initial direction in the candidate target initial geometric observation results, and simultaneously correct the beamforming results and target detection results based on the communication sensing consistency correction amount, local distortion intensity and array overall distortion intensity to obtain the joint correction result;

[0027] The closed-loop update module is configured to output joint correction results and, based on the high-confidence execution results, perform constrained recursive updates to the reference baseline and sensitivity mapping. The updated results are then written back to the local array element cluster-level normalized state observation results construction process and the array distortion implicit state vector inversion process in the next time slot.

[0028] The beneficial effects of this invention are as follows:

[0029] 1. This invention acquires the device-side operation data, communication feedback data, and sensing observation data of the target base station in the current time slot, and constructs standardized state observation results, local array cluster occupancy weights, and initial geometric observation results of candidate targets at the local array cluster granularity. This unifies the heterogeneous information originally scattered in the device readback link, communication feedback link, and sensing processing link to the same time slot semantics and the same local array cluster granularity, providing a directly solvable observation basis for subsequent array distortion state inversion.

[0030] 2. This invention further performs graph-constrained recursive inversion based on the standardized state observation results at the local array element cluster level, the local array element cluster occupancy weights, the physical adjacency relationships between local array element clusters, the reference baseline, the sensitivity mapping, and the local distortion state estimation results of the previous time slot. This yields the array distortion implicit state vector and forms the local distortion intensity and the overall array distortion intensity. This processing unifies the equipment-side thermal drift, power supply deviation, service load bias, and sensing observation disturbances into the same state representation layer. It can distinguish between real target changes and array geometric offsets at the local array element cluster granularity, thus improving the stability and interpretability of the array distortion state characterization and providing a unified state basis for subsequent synchronous correction.

[0031] 3. This invention also constructs array distortion compensation results based on the implicit state vector of array distortion, calculates the communication-sensing consistency correction amount by combining the directional consistency loss of the initial direction of the candidate target in the service beam direction and the initial geometric observation results of the candidate target, and synchronously corrects the beamforming results and target detection results based on the communication-sensing consistency correction amount, local distortion intensity and overall array distortion intensity. Through this processing, the stability of the service beam direction and the observability of the initial direction of the candidate target are jointly regulated under the same constraint framework. The inconsistency of local array element cluster distortion on the communication direction and sensing direction can be synchronously suppressed, thereby reducing the situation where the service beam direction is available but the target detection result is offset, the number of pseudo-peak triggers increases, and the target angle estimation result and target radial velocity estimation result are distorted.

[0032] 4. Finally, based on the high-reliability execution results, the present invention performs a restricted recursive update on the reference baseline and sensitivity mapping, and writes the update results back to the local array element cluster-level standardized state observation results construction process and the array distortion implicit state vector inversion process in the next time slot. This ensures that the long-term reference layer and state mapping layer are only updated by the execution results that meet the conditions of link error rate, service direction consistency loss and target trajectory continuity. As a result, abnormal execution results will not continue to spread across time slots, and the reference baseline and sensitivity mapping can be updated in a controlled manner as equipment ages, thermal environment changes and load migration occurs. The target base station can maintain the long-term stability and consistency of the sensing geometry results while the communication link remains available. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort:

[0034] Figure 1 This is a flowchart of the method of the present invention;

[0035] Figure 2 This is a flowchart of the S2 process of the present invention;

[0036] Figure 3 This is a system framework diagram of the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Example 1

[0039] like Figure 1 and Figure 2 As shown, this invention discloses a joint optimization method for beamforming and target detection in a sensor-integrated network, including the following specific steps:

[0040] In this embodiment, S1 is used to form a unified input basis required for the subsequent graph-constrained recursive inversion in S2. This step processes the device-side operating data, communication feedback data, and sensing observation data of the target base station in the current time slot, and forms local array cluster-level standardized state observation results, local array cluster occupancy weights, and candidate target initial geometric observation results at the local array cluster granularity. The core purpose of this step is to compress the heterogeneous data that was originally scattered in the device readback link, communication feedback link, and sensing processing link into the same local array cluster granularity and the same time slot semantics, so that the subsequent array distortion implicit state vector inversion has a unified, comparable, and computable observation basis.

[0041] First, acquire the equipment-side operational data, communication feedback data, and sensing observation data of the target base station in the current time slot. The equipment-side operational data includes the average temperature, linear reflection power, average power amplifier current, and phase shifter control word corresponding to each local array element cluster. The average temperature is collected by the array board-level temperature sensor, the power amplifier temperature readback interface, or the internal thermal management monitoring interface; the linear reflection power is collected by the reflection power sampling circuit or the standing wave detection circuit; the average power amplifier current is collected by the power amplifier power supply monitoring interface; and the phase shifter control word is read back from the beam control plane or the radio frequency control bus.

[0042] The communication feedback data includes the precoding usage frequency within the current window and the uplink and downlink equivalent channel state data after reciprocity calibration and subcarrier alignment. This implementation is based on a time-division duplex base station. After reciprocity calibration is completed, the uplink and downlink equivalent channel state data are comparable and are then used to form the channel residual.

[0043] The sensing observation data includes angular spectrum distribution results and initial geometric observation results of candidate targets. The angular spectrum distribution results are obtained by angular domain projection processing of the current sensing time slot echo data. The initial geometric observation results of candidate targets include at least the initial direction of the candidate targets and may include the initial radial velocity of the candidate targets.

[0044] After obtaining the above data, local array element clusters are divided according to the physical adjacency relationship of the array. In this embodiment, the division of local array element clusters is based on the physical adjacency relationship of the array elements on the array surface. The division boundary is further corrected by combining the distribution of shared power supply branches and the distribution of shared heat dissipation paths, so that the array elements in the same local array element cluster have high consistency in heat conduction path, power supply path and service load carrying path. For common two-dimensional planar active arrays, a single local array element cluster preferably contains 4 to 16 array elements, and more preferably 6 to 12 array elements.

[0045] When the number of elements in a single cluster is less than 4, local observations are susceptible to readback noise and instantaneous jitter, resulting in insufficient stability. When the number of elements in a single cluster is greater than 16, local differences are averaged out, and drift characteristics caused by local high load directions are not easily distinguished. The above range matches common AAU board-level grouping, power supply partitioning, and heat dissipation partitioning, and has a practical engineering basis. After completing the local element cluster division, unified time base processing is performed on the equipment-side operating data, communication feedback data, and sensing observation data.

[0046] In practice, the current time slot boundary is used as a unified alignment benchmark. Data collected from different links are timestamped. Time base alignment is performed when the timestamp deviation is no more than 20% of the current time slot duration. Data exceeding this range is recorded as time base mismatch samples and is not included in the construction of the local array element cluster-level standardized state observation results for this time slot. For short-term missing samples, only missing samples of 1 to 3 consecutive time slots are filled in. The filling method is to use linear interpolation of the effective samples before and after or local moving average backfilling. When the length of the consecutive missing samples exceeds 3 time slots, the sample segment is recorded as an invalid segment. For abnormal samples, a sliding window deviation discrimination method is used to remove them. When the sample deviates from the current sliding window center value to the preset deviation threshold, the sample is recorded as an abnormal sample.

[0047] After the unified time base processing is completed, the standardized state observation results at the local array element cluster level are further constructed. This processing is divided into two layers: component conversion and local array element cluster aggregation. In the component conversion layer, the average temperature corresponding to each local array element cluster is first converted into a temperature offset. Specifically, a reference temperature baseline and a temperature scale parameter are used to convert the average temperature into a dimensionless temperature offset. The reference temperature baseline is taken from the stable operation sample after the equipment is installed and debugged, or from the temperature reference baseline updated by the previous round of high-confidence execution results. The temperature scale parameter is preferably taken from 5℃ to 15℃, and more preferably from 8℃ to 12℃. This range covers the local temperature rise fluctuation range that is common for outdoor base stations under normal heat dissipation conditions.

[0048] Subsequently, a reflection power offset conversion is performed on the linear reflection power. Specifically, the linear reflection power is first converted to decibel reflection power using a logarithmic power conversion method, and then converted to a dimensionless reflection power offset based on the reference reflection power baseline and reflection power scale parameters. The reference reflection power baseline is taken from readback samples with normal feeder connection and standing wave within a stable range. Only after this conversion is completed can the reflection power related quantities participate in the subsequent construction of local array element cluster-level normalized state observation results.

[0049] Next, current offset conversion is performed on the average current of the power amplifier. Specifically, a dimensionless current offset is formed by using a reference current baseline and a current scale parameter. The current scale parameter is preferably 10% to 25% of the rated average operating current. This range can reflect the offset of the power amplifier's operating state caused by changes in the service load, while not excessively amplifying the small fluctuations in the normal power adjustment process.

[0050] Subsequently, a phase command offset conversion is performed on the phase shifter control word. Specifically, the existing phase shifter calibration method is first used to convert the phase shifter control word into a phase command quantity through a code word phase lookup table. Then, a phase wrap-around process is performed on the phase command quantity, and a dimensionless phase command offset is formed by combining the reference phase command baseline and phase scale parameters. This conversion chain ensures that the phase command offset quantity that enters the subsequent processing is a phase command offset quantity with physical meaning, rather than a discrete control word difference.

[0051] Next, the precoding usage frequency within the current window is normalized to form a dimensionless quantity corresponding to the precoding usage frequency. The statistical window for the precoding usage frequency is preferably the most recent 20 to 200 time slots, and more preferably the most recent 50 to 100 time slots. This setting can balance the immediate service hotspots and short-term scheduling jitter. Then, residual conversion is performed on the uplink and downlink equivalent channel state data after reciprocity calibration and subcarrier alignment to form the channel residual quantity. This conversion is based on the premise that the time division duplex system and reciprocity calibration are completed. Finally, angular spectrum perturbation conversion is performed on the angular spectrum distribution results. Specifically, the angular spectrum distribution results of the current time slot are first formed, and then the correlation is compared with the angular spectrum distribution results of the same local array element cluster in the previous time slot. The angular spectrum perturbation coefficient is formed based on the comparison results. The angular spectrum perturbation coefficient is used to characterize the stability of angular domain observations between adjacent time slots.

[0052] After completing the above component conversion, the dimensionless quantities belonging to the same local array element cluster are aggregated to form the local array element cluster-level standardized state observation results corresponding to that local array element cluster.

[0053] The preferred aggregation method is to use the average aggregation of the corresponding components of all effective array elements within a local array element cluster. When the number of effective array elements is less than 60% of the original number of array elements after the removal of abnormal samples in a local array element cluster, the local array element cluster-level standardized state observation result of the current time slot of the local array element cluster is recorded as a low-confidence result and is processed as a low-weight observation input in the subsequent S2.

[0054] In this way, temperature offset, reflected power offset, current offset, phase command offset, dimensionless quantity corresponding to precoding frequency, channel residual and angular spectrum perturbation coefficient are uniformly organized into the same local array element cluster granularity, completing the construction of local array element cluster-level standardized state observation results.

[0055] While constructing standardized state observation results at the local array element cluster level, the local array element cluster occupancy weight is further formed. The local array element cluster occupancy weight is determined by two types of information: the first type is the precoding usage frequency within the current window, which reflects the recent continuous occupancy trend of the local array element cluster by service scheduling; the second type is the transmission energy allocated to the local array element cluster in the current time slot, which reflects the real-time carrying capacity of the local array element cluster in the current time slot.

[0056] In actual processing, the transmitted energy is first normalized and then combined with the dimensionless quantity corresponding to the precoding frequency to form the local array element cluster occupancy weight. This weight is limited to the range of 0 to 1. The larger the value, the more concentrated the service load undertaken by the local array element cluster in the current operating state.

[0057] This process has direct engineering significance because array element clusters in the local high load direction are more prone to thermal drift and feed deviation accumulation. In the subsequent S2, the local array element cluster occupancy weight is introduced into the graph constraint recursive inversion, which can improve the pertinence of local distortion state identification.

[0058] In addition, S1 also needs to form the initial geometric observation results of the candidate target. In specific implementation, firstly, the spectral peak is searched based on the angular spectrum distribution results of the current sensing time slot to form the initial direction of the candidate target; then, the initial radial velocity of the candidate target is formed by combining the slow time echo phase change results or the initial Doppler spectrum main peak results; finally, the initial direction and initial radial velocity of the candidate target are combined to form the initial geometric observation results of the candidate target.

[0059] To avoid spurious peaks from directly entering the subsequent joint correction process, it is preferable to perform an initial screening of candidate targets, retaining only those candidate targets that can be observed in two consecutive time slots and whose angular domain position changes are within a continuous range as valid candidate targets. This initial screening step is a supporting processing step, the purpose of which is to ensure that the initial orientation of the candidate targets participating in the orientation consistency loss calculation in the subsequent S3 has basic stability.

[0060] Compared to existing methods that construct calibration inputs based solely on the average state of the entire array, a single communication link indicator, or a single sensing output, this step can simultaneously retain thermal drift, feed deviation, load bias, and corner domain disturbance information at the local element cluster granularity, enabling subsequent steps to identify the differences between local distortion states and actual target changes.

[0061] In this embodiment, S2 takes the local array element cluster-level standardized state observation results, local array element cluster occupancy weights and candidate target initial geometric observation results output by S1, and combines the physical adjacency relationship between local array element clusters, reference baseline, sensitivity mapping and local distortion state estimation results of the previous time slot to perform graph-constrained recursive inversion on the local distortion state of the current time slot to obtain the array distortion implicit state vector, and on this basis form the local distortion intensity and the overall array distortion intensity.

[0062] The technical problem addressed in this step is: when a local array element cluster is subjected to the combined effects of thermal drift, power supply deviation, and service load bias, how to stably invert the state quantity that can represent the current degree of array geometric distortion from the multi-source dimensionless observation results formed by S1, and use this state quantity as the unified input for subsequent communication-sensing joint correction.

[0063] First, regarding the first The local element clusters construct the local distortion state estimation results.

[0064] In this embodiment, the local distortion state estimation result consists of the amplitude drift ratio and the normalized phase drift ratio, both of which are dimensionless quantities. The amplitude drift ratio is used to characterize the degree of deviation of the effective gain of the current local array element cluster relative to the reference gain; the normalized phase drift ratio is used to characterize the degree of normalization of the phase shift of the current local array element cluster relative to the upper bound of the phase drift.

[0065] Furthermore, the normalized phase drift ratio is converted into the true phase drift amount based on the upper bound of the phase drift, and this true phase drift amount is used for subsequent strong distortion condition determination.

[0066] In terms of parameter settings, the actual phase drift is obtained by multiplying the normalized phase drift ratio by the upper bound of the phase drift, and the upper bound of the phase drift is preferably set to [value missing]. to , better to .

[0067] This range corresponds to the common phase shift amplitude of a local array element cluster under normal thermal drift, short-time feed offset, and high load conditions in the hot spot direction: when the actual phase drift is lower than When the actual phase shift is higher than 100%, the effect on array directivity is small; when the actual phase shift is higher than 100%, the effect on array directivity is small. At that time, the local array element cluster had entered a clearly abnormal state.

[0068] The normal inversion range of the amplitude drift ratio is preferably limited to an absolute value of no more than 0.15. When it exceeds this range, it usually corresponds to the power amplifier operating point shift, local power supply abnormality, or local array element response mismatch enhancement.

[0069] After forming the representation of the local distortion state estimation results, a sensitivity mapping is established. The sensitivity mapping is used to describe the mapping relationship between the local distortion state estimation results and the local array element cluster-level standardized state observation results. In specific implementation, the initial sensitivity mapping is first established using the factory calibration data, and then the correction is performed using the online trial calibration data to obtain the sensitivity mapping that is available under the current operating environment. The factory calibration data provides the basic response relationship under the equipment health state, and the online trial calibration data provides the supplementary response relationship under real load, real thermal environment and real power supply conditions.

[0070] It should be clarified that the input to the sensitivity mapping is the current local distortion state estimation result, and the output is the estimated value of the local array element cluster-level standardized state observation result. Both the input and output quantities are dimensionless, so the mapping difference in the observation fitting term is still a dimensionless quantity.

[0071] Subsequently, an adjacency edge set is constructed based on the physical adjacency relationship between local array element clusters. The adjacency edge set is used to limit the scope of the spatial continuity constraint. In this embodiment, an adjacency relationship is established between two local array element clusters only when at least one of the following relationships is true: physical location adjacent, heat diffusion path adjacent, and power supply path adjacent. Furthermore, adjacency weights are assigned to the local array element clusters that have established an adjacency relationship.

[0072] The adjacency weight is preferably between 0.2 and 1.0. When two local array elements simultaneously satisfy physical adjacency, continuous shared heat dissipation path, and continuous shared power supply branch, the adjacency weight is preferably between 0.8 and 1.0. When only physical adjacency is satisfied and the thermal diffusion connection is weak, the adjacency weight is preferably between 0.4 and 0.8. When two local array elements only contact each other in the boundary region, the adjacency weight is preferably between 0.2 and 0.4. This setting allows the spatial continuity constraint to preserve local correlation without completely smoothing out the real drift in the local high load direction.

[0073] After completing the construction of the adjacent edge set, a state recursion matrix is ​​constructed based on the local distortion state estimation result of the previous time slot. The state recursion matrix is ​​used to map the local distortion state estimation result of the previous time slot to the current time slot, forming the initial estimate of the local distortion state of the current time slot.

[0074] In this embodiment, the state recursion matrix preferably adopts a diagonal structure, and its diagonal elements are recursion factors. The recursion factors are preferably between 0.90 and 0.99, and more preferably between 0.94 and 0.98. This range is derived from the short-time continuous variation characteristics of local array element cluster thermal drift, feed offset, and local load bias between adjacent time slots. When the recursion factor is lower than 0.90, the support of the previous time slot state for the current time slot is insufficient; when the recursion factor is higher than 0.99, the effect of the current time slot observation is weakened. Therefore, the above range can take into account both the utilization of time continuity and the correction effect of the current time slot observation.

[0075] After generating the local distortion state estimation results, sensitivity mapping, adjacent edge set, and state recursion matrix, graph-constrained recursive inversion is performed. In this embodiment, the time slot... The objective function is defined as:

[0076] ;

[0077] Among them, the observation fitting term is:

[0078] ;

[0079] The spatial continuity constraint term is:

[0080] ;

[0081] The time recursion constraint is:

[0082] ;

[0083] Minimize the above objective function to obtain the array distortion hidden state vector estimation result for the current time slot:

[0084] ;

[0085] In the above formula, Indicates time slot The objective function, Represents the observation fit term, Represents the spatial continuity constraint term. This represents the time recursion constraint term; This represents the total number of local array element clusters. Indicates the first A local array element cluster in a time slot The weight of local element clusters Indicates the first A local array element cluster in a time slot Local array element cluster-level normalized state observation results, Indicates the first A local array element cluster in a time slot Sensitivity mapping, Indicates the first A local array element cluster in a time slot The results of local distortion state estimation Represents the set of adjacent edges. Indicates the first The local element cluster and the first The adjacency weights between local matrix clusters Indicates the spatial continuity constraint coefficient. Indicates the time recursion constraint coefficient. Indicates the first The state recursion matrix of a local array element cluster Indicates the first The estimation results of the local distortion state of the local element cluster in the previous time slot, This represents the hidden state vector of the current time slot array distortion, formed by concatenating the estimation results of all local distortion states. This indicates the optimal estimation result.

[0086] The effects of the three constraints mentioned above are as follows.

[0087] The observation fitting term is used to constrain the difference between the current local array element cluster-level standardized state observation results and the sensitivity mapping output, so that the solution results fit the current time slot observation.

[0088] The spatial continuity constraint term is used to constrain the differences in local distortion states between physically adjacent local array element clusters, so that the solution results conform to the spatial continuity characteristics of local thermal diffusion and feed coupling.

[0089] The time recursion constraint term is used to constrain the deviation between the local distortion state estimation result of the current time slot and the local distortion state estimation result of the previous time slot, so that the solution result conforms to the short-time continuous change characteristics of adjacent time slots.

[0090] The three constraints working together can achieve a balance between current observation fit, spatial neighborhood consistency, and temporal continuity.

[0091] In terms of the solution method, this implementation method preferably adopts an iterative update method to perform graph constraint recursive inversion.

[0092] In specific processing, an initial estimate of the local distortion state of the current time slot is first formed based on the reference baseline and the local distortion state estimation results of the previous time slot. Then, all local distortion state estimation results are updated cyclically according to the local array element cluster order. After each round of updates, the relative rate of change of the objective function between the current round and the previous round is calculated. When the relative rate of change of the objective function is not higher than... The iteration will stop when the preset limit is reached.

[0093] The upper limit of the number of iterations is preferably 10 to 30 times, and more preferably 15 to 20 times.

[0094] The threshold for the relative rate of change of the objective function and the upper limit for the number of iterations together constitute a dual stopping condition. This setting can ensure the stability of the solution results and meet the computing power constraints of real-time processing on the baseband side under the current time slot. For the overall constraint strength parameters, the spatial continuity constraint coefficient is preferably taken as 0.2 to 0.6, and the time recursion constraint coefficient is preferably taken as 0.5 to 1.0. When the spatial continuity constraint coefficient is lower than 0.2, the spatial continuity constraint is insufficient to suppress the noise of single cluster observations; when the spatial continuity constraint coefficient is higher than 0.6, the actual drift in the local high load direction will be over-smoothed; when the time recursion constraint coefficient is lower than 0.5, the continuity of adjacent time slots is not fully utilized; when the time recursion constraint coefficient is higher than 1.0, the pull of the previous time slot state on the current time slot is too strong, which can easily weaken the correction effect of the current time slot observation. Therefore, the above value range matches the actual change amplitude of the short-term drift of the local array element cluster.

[0095] After completing the array distortion implicit state vector inversion, the local distortion intensity and the overall array distortion intensity are further formed. The local distortion intensity is determined based on the modulus of the local distortion state estimation results of each local array element cluster, and is used to characterize the distortion degree of a single local array element cluster in the current time slot. The overall array distortion intensity is determined by performing weighted aggregation on all local distortion intensities and the corresponding local array element cluster occupancy weights, and is used to characterize the overall distortion degree of the entire array in the current time slot under the combined effect of service load and equipment drift. Weighted aggregation is used here instead of simple averaging because highly occupied local array element clusters have a more direct impact on the current serving beam direction and the initial direction of candidate targets. Simple averaging would weaken the contribution of local drift in hot spot directions to the overall state assessment. Finally, the strong distortion condition is determined based on the local distortion state estimation results.

[0096] In this embodiment, when the absolute value of the amplitude drift ratio of any local element cluster is higher than 0.15, or the absolute value of the true phase drift is higher than... When the current time slot is determined to be in a strong distortion condition, the array distortion implicit state vector of the current time slot is still output after entering the strong distortion condition. However, it is processed conservatively in subsequent S3 and the joint correction strategy of conventional strength is not directly adopted. This can avoid abnormal local array element clusters from exerting too strong pull on the entire communication perception consistency correction process.

[0097] This step can identify local drift formed in the direction of local high load, providing direct state input for the calculation of communication sensing consistency correction in S3 and the synchronous correction of beamforming results and target detection results.

[0098] In this embodiment, S3 receives the array distortion implicit state vector, local distortion intensity, and overall array distortion intensity output by S2. It first forms an array distortion compensation result around the array distortion implicit state vector, and then forms a corresponding directional consistency loss based on the service beam direction and the initial direction of the candidate target in the initial geometric observation results of the candidate target. Based on this, it calculates the communication sensing consistency correction amount. Subsequently, based on the communication sensing consistency correction amount, local distortion intensity, and overall array distortion intensity, it performs synchronous correction on the beamforming result and the target detection result, and finally outputs the joint correction result.

[0099] The technical problem addressed in this step is that local array element cluster distortion can simultaneously change the effective radiation state of the serving beam direction and the effective observation state of the initial direction of the candidate target. If the communication link and the sensing link are corrected separately, the serving beam direction will remain available while the geometric result of the target direction will have shifted. Therefore, it is necessary to perform unified correction on the communication direction and the sensing direction under the same array distortion background.

[0100] The array distortion implicit state vector obtained in S2 is formed by splicing the local distortion state estimation results of all local array element clusters. Each local distortion state estimation result includes the amplitude drift ratio and the normalized phase drift ratio. When entering S3, the normalized phase drift ratio of each local array element cluster is first converted into the true phase drift amount according to the phase drift upper bound determined in S2. Then, the amplitude drift ratio and the true phase drift amount are converted into the complex compensation factor of the corresponding local array element cluster.

[0101] The complex compensation factor has two parts: one part reflects the array response amplitude offset of the current local array element cluster, and the other part reflects the array response phase offset of the current local array element cluster. Subsequently, according to the arrangement order of the local array element clusters on the array plane, the complex compensation factors corresponding to all local array element clusters are written into the array compensation structure to form the array distortion compensation result of the current time slot. This processing chain maps the state characterization result in S2 to the array direction response layer. Only after this mapping is completed can the subsequent consistency comparison between the service beam direction and the initial direction of the candidate target have a unified physical basis.

[0102] In terms of values, the amplitude drift ratio still adopts the normal inversion range in S2, and the absolute value is preferably not higher than 0.15; the absolute value of the true phase drift is preferably not higher than π / 6. When any local array element cluster exceeds this range, the resulting array distortion compensation result is still retained for the expression of the current time slot state, but subsequent corrections are subject to a conservative processing flow.

[0103] After generating the array distortion compensation result, the directional consistency loss corresponding to the serving beam direction and the directional consistency loss corresponding to the initial direction of the candidate target are calculated respectively. Specifically, the reference directional response of the serving beam direction under the condition of a healthy array is first obtained based on the ideal array directional response. Then, the directional response of the serving beam direction under the distortion compensation background is obtained using the current time slot array distortion compensation result. Subsequently, a normalization comparison is performed on the reference directional response and the directional response after distortion compensation to obtain the directional consistency loss corresponding to the serving beam direction. For the initial direction of the candidate target, the processing order is the same, that is, the reference directional response of the initial direction of the candidate target under the condition of a healthy array is first obtained, then the directional response after distortion compensation in the current time slot is obtained, and then the directional consistency loss corresponding to the initial direction of the candidate target is formed by normalization comparison.

[0104] The aforementioned orientation consistency loss is a supporting calculation step. Mature array orientation response comparison methods already exist in this field; this implementation retains only the necessary processing logic.

[0105] First, both the serving beam direction and the initial direction of the candidate target must be input with the same array distortion compensation result; second, the comparison result is normalized to ensure that the dimensions are consistent when the subsequent calculation of the communication sensing consistency correction amount is performed; third, the larger the value of the direction consistency loss, the more obvious the influence of local array element cluster distortion on that direction.

[0106] After obtaining the directional consistency loss corresponding to the service beam direction and the directional consistency loss corresponding to the initial direction of the candidate target, the communication-aware consistency correction amount is calculated.

[0107] The communication-aware consistency correction amount consists of three parts: the first part is the product of the directional consistency loss corresponding to the service beam direction and the service direction distortion weight; the second part is the product of the directional consistency loss corresponding to the initial direction of the candidate target and the target direction distortion weight; and the third part is the product of the absolute value of the difference between the two directional consistency losses and the directional difference weight. Then, these three parts are added together to obtain the communication-aware consistency correction amount.

[0108] Specifically, the directional consistency loss corresponding to the service beam direction reflects the degree of damage to the main direction of the communication side under the current array distortion background; the directional consistency loss corresponding to the initial direction of the candidate target reflects the degree of damage to the target direction of the sensing side under the current array distortion background; the absolute value of the difference between the two reflects the degree of distortion asymmetry between the communication side and the sensing side in the same time slot. Only by weighting and summing the first two items can it be determined whether the communication direction and the sensing direction are damaged respectively, but it cannot be determined whether there is a directional mismatch between the two. After adding the difference term, the communication-sensing consistency correction quantity simultaneously includes three types of information: communication direction damage, sensing direction damage, and communication-sensing direction mismatch. Therefore, this correction quantity can be used as a unified control quantity for the synchronous correction of subsequent beamforming results and target detection results.

[0109] In terms of parameter settings, the service direction distortion weight, target direction distortion weight, and direction difference weight are all taken as non-negative values, and the sum of the three is preferably 1. The service direction distortion weight is preferably taken as 0.30 to 0.45; the target direction distortion weight is preferably taken as 0.30 to 0.45; and the direction difference weight is preferably taken as 0.10 to 0.25. The basis for this range is that both the service beam direction and the initial direction of the candidate target directly participate in the subsequent correction, and their importance is similar, so the dominant weights should be kept close. The direction difference term is used to supplement the description of the degree of distortion inconsistency between the communication direction and the sensing direction, and its proportion is lower than the first two terms. When the direction difference weight is higher than 0.25, the small difference between the service beam direction and the initial direction of the candidate target will be excessively amplified. When the direction difference weight is lower than 0.10, the distortion asymmetry between the communication direction and the sensing direction is difficult to reflect.

[0110] The above scope is consistent with the balance requirements of integrated communication and sensing base stations in terms of ensuring communication services and maintaining target direction.

[0111] After obtaining the communication-aware consistency correction amount, the beamforming result is corrected. Specifically, a distortion suppression term is first constructed based on the local distortion intensity. The higher the local distortion intensity of a local array element cluster, the higher the corresponding distortion suppression intensity. Then, a target orientation preservation term is constructed based on the initial orientation of the candidate target. This term is used to retain the responsiveness to the initial orientation of the candidate target while correcting the service beam orientation. Subsequently, the corrected beamforming result is solved using the communication channel state of the current time slot, the array distortion compensation result, the distortion suppression term, and the target orientation preservation term as common inputs.

[0112] Specifically, first, a quadratic optimization objective is established, which includes three types of constraints:

[0113] The first type is the communication channel adaptation constraint, which is used to ensure that the corrected beamforming result still matches the channel state corresponding to the current time slot serving user; the second type is the distortion suppression constraint, which is used to reduce the pull of local array element clusters with high local distortion intensity on the beamforming result; the third type is the target direction preservation constraint, which is used to maintain the observability of the corrected beamforming result for the initial direction of the candidate target.

[0114] Subsequently, the extreme value of the optimization objective with respect to the beamforming weight is obtained to obtain the corrected beamforming result. Therefore, the corrected beamforming result is jointly determined by the communication channel state, the array distortion compensation result, the local distortion intensity distribution and the initial direction of the candidate target, and is not a fixed scaling of the original beamforming result.

[0115] Regarding parameter settings, the distortion suppression coefficient is preferably between 0.20 and 0.80, more preferably between 0.30 and 0.60; the target orientation preservation coefficient is preferably between 0.10 and 0.60, more preferably between 0.20 and 0.40. When the distortion suppression coefficient is below 0.20, the influence of local array element clusters with high local distortion intensity on the beamforming result is insufficiently suppressed; when it is above 0.80, the adaptability of the corrected beamforming result to the current channel state decreases. When the target orientation preservation coefficient is below 0.10, the preservation effect of the initial direction of the candidate target in the corrected beamforming result is insufficient; when it is above 0.60, the dominance of the serving beam direction decreases. The above ranges are consistent with the engineering balance between the stability of the serving beam direction and the observability of the initial direction of the candidate target.

[0116] After the corrected beamforming results are solved, the target detection results are synchronously corrected. This process includes, in sequence, detection threshold correction, target angle estimation result correction, and target radial velocity estimation result correction.

[0117] First, based on the communication perception consistency correction amount and the overall array distortion intensity, a joint amplification correction is performed on the basic detection threshold. The larger the communication perception consistency correction amount, the higher the overall array distortion intensity, and the greater the increase in the basic detection threshold. This is because an increase in the communication perception consistency correction amount indicates a decrease in the common credibility of the serving beam direction and the initial direction of the candidate target under the current array distortion background. An increase in the overall array distortion intensity indicates that the current array as a whole is more severely affected by local distortion. At this time, increasing the basic detection threshold can suppress pseudo-peaks from entering the target detection results.

[0118] In terms of parameter settings, the magnification factor of the basic detection threshold is preferably 1.05 to 1.50, and more preferably 1.10 to 1.35. When the magnification factor is lower than 1.05, the suppression effect on false peaks is weak; when it is higher than 1.50, weak target echoes are easily suppressed. This range matches the common echo intensity range of vehicle targets, low-altitude targets and park targets.

[0119] Subsequently, the target angle estimation results in the initial geometric observation results of the candidate target are corrected. Specifically, based on the sensitivity relationship between the array direction response and the implicit state vector of array distortion, angle deviation compensation is performed on the target angle estimation results to obtain the corrected target angle estimation results. This compensation is based on the premise that the local small deviation linear approximation holds. Therefore, it is only performed when the deviation between the initial direction of the candidate target and the direction of the service beam center is not higher than a preset angle range. The preset angle range is preferably 10° to 20°, and more preferably 15° to 20°. When the deviation is within this range, the array direction response maintains good linear compensability in the local angular domain. When the deviation is higher than 20°, the local linear approximation error increases, and continuing to perform local angle compensation will reduce the stability of the results. Therefore, when the deviation exceeds this range, the local angle compensation is stopped, and the coarse angular domain search process is started.

[0120] Finally, the radial velocity estimation results of the candidate target in the initial geometric observation results are re-estimated. Instead of directly applying a fixed offset correction to the original radial velocity estimation results, a processing chain of direction correction-spatial filtering-Doppler spectrum reconstruction-principal energy re-estimate is adopted.

[0121] In specific processing, the target spatial filtering weights are first constructed based on the corrected target angle estimation results and array distortion compensation results. Then, spatial filtering is performed on the slow-time echo sequence using these target spatial filtering weights to obtain a slow-time echo sequence focused on the target direction. Subsequently, Doppler spectrum reconstruction is performed on the slow-time echo sequence, and the target radial velocity estimation results are re-determined based on the main energy distribution of the Doppler spectrum. This processing chain conforms to the physical path of velocity estimation because the main impact of local array element cluster distortion on velocity estimation is to change the directional focusing quality and sidelobe leakage, thereby changing the concentration position of the Doppler main energy. By first correcting the direction and then reconstructing the Doppler spectrum, the obtained target radial velocity estimation results are more consistent with the actual echo propagation and reception process.

[0122] In terms of parameter settings, the slow-time observation length is preferably 8 to 64 sampling units, more preferably 16 to 32 sampling units. When it is less than 8 sampling units, the Doppler resolution is insufficient; when it is more than 64 sampling units, the real-time pressure of the current time slot increases, and the impact of the non-stationarity of the target motion is enhanced. Finally, the corrected detection threshold, the corrected target angle estimation result, and the corrected target radial velocity estimation result are combined to form the target detection result. At this time, the beamforming result and the target detection result have been corrected under the same array distortion background, so they together constitute the joint correction result and serve as the input for the high-confidence execution result screening in S4.

[0123] In this embodiment, S4 receives the joint correction result output by S3 and performs two types of processing around the joint correction result. The first type of processing is to implement the corrected control result and detection result into the actual execution link of the current time slot. The second type of processing is to perform a restricted recursive update on the reference baseline and sensitivity mapping based on the credibility of the execution in the current time slot, and write the update result back to the local array element cluster-level normalized state observation result construction process and array distortion implicit state vector inversion process of the next time slot.

[0124] The technical problem addressed by this step is: if the beamforming and target detection results have been corrected in the current time slot, but the next time slot still inherits the execution results of all time slots without difference, then the deviations caused by pseudo-peak triggering, transient disturbances, abnormal load spikes and strong distortion conditions will be written into the reference baseline and sensitivity mapping again, thus forming cross-time slot cumulative offset. S4 controls the source of results entering the long-term memory layer through high-confidence execution result screening and restricted recursive update control, blocking the continuous transmission of abnormal execution results.

[0125] First, the joint correction results output by S3 are implemented in the actual execution link of the current time slot. The joint correction results include at least the corrected beamforming results and the corrected target detection results. The corrected beamforming results are sent to the target base station transmission control link to complete the communication beam control of the current time slot. The corrected target detection results are sent to the current sensing processing link to form the target detection output of the current time slot.

[0126] In this embodiment, the current time slot execution result is defined as: the actual execution state of the corrected beamforming result in the transmission link, and the actual output state of the corrected target detection result in the sensing link. The purpose of this definition is to ensure that the subsequent high-confidence execution result screening uses the execution result after execution, rather than just the intermediate calculation result.

[0127] After the execution in the current time slot is completed, the high-reliability execution result is determined. The high-reliability execution result is jointly determined by three types of quantities: link error rate, service direction consistency loss, and target trajectory continuity.

[0128] For the link block error rate, first count the number of error blocks in the service data block corresponding to the current time slot, then count the total number of blocks within the same statistical window, and divide the number of error blocks by the total number of blocks to obtain the dimensionless link block error rate. The statistical window is preferably the current time slot and the following 5 to 20 symbol cycles, more preferably the current time slot and the following 10 symbol cycles. The basis for this range is that if the statistical window is too short, a single burst error will amplify the degree of link deterioration in the current time slot; if the statistical window is too long, the correction effect of the current time slot will be diluted by subsequent service fluctuations. The link block error rate threshold is preferably 1% to 10%, more preferably 3% to 5%.

[0129] When the percentage is below 1%, a large number of normal short-term disturbances will be excluded; when it is above 10%, the state of obviously damaged links may still be written back.

[0130] For service direction consistency loss, the S3 output result is read directly. The service direction consistency loss threshold is preferably 0.05 to 0.25, and more preferably 0.08 to 0.15. When the threshold is lower than 0.05, small deviations caused by normal thermal drift and local load fluctuations will also be intercepted. When the threshold is higher than 0.25, the service beam direction may be misjudged as a write-back state even though there is obvious distortion.

[0131] For target trajectory continuity, a unified dimensionless continuity score needs to be established first. In specific processing, a sliding window is first established within the most recent 3 to 8 time slots, preferably using the most recent 5 time slots. Within this window, the change in target direction between adjacent time slots is calculated first, and then the change in target radial velocity between adjacent time slots is calculated. At the same time, the duration of target holding within this window is statistically analyzed. Subsequently, the change in target direction is compared with the allowable range of direction change to form a direction continuity evaluation; the change in target radial velocity is compared with the allowable range of velocity change to form a velocity continuity evaluation; and the duration of holding is compared with the window length to form a holding continuity evaluation. Finally, the direction continuity evaluation, velocity continuity evaluation, and holding continuity evaluation are uniformly normalized and fused to obtain a target trajectory continuity score within the range of 0 to 1.

[0132] The allowable range for directional change is preferably no more than 3° to 8° between adjacent time slots, and more preferably 5°; the allowable range for radial velocity change is preferably no more than 1m / s to 5m / s between adjacent time slots, and more preferably 2m / s to 3m / s. These ranges are derived from the continuous motion characteristics of vehicle targets, low-speed low-altitude targets, and mobile targets in the campus in common scenarios of integrated sensing base stations. The target trajectory continuity threshold is preferably 0.60 to 0.90, and more preferably 0.70 to 0.85. When it is below 0.60, the trajectory screening is insufficient in constraining false peaks and mis-segmented trajectories; when it is above 0.90, targets with normal acceleration / deceleration and normal turning are also easily excluded.

[0133] After the link block error rate, service direction consistency loss, and target trajectory continuity have all been calculated, a high-reliability execution result determination is performed. Specifically, when the link block error rate is not higher than the block error rate threshold, the service direction consistency loss is not higher than the consistency loss threshold, and the target trajectory continuity is not lower than the continuity threshold, the current time slot is determined as a high-reliability execution time slot, and the execution result of the current time slot is recorded as a high-reliability execution result; if any of the conditions are not met, the current time slot is recorded as a non-high-reliability execution time slot.

[0134] In non-high-confidence execution slots, the joint correction results are still used for the current slot communication beam control and target detection output, but do not participate in the reference baseline and sensitivity mapping update.

[0135] After determining a high-confidence execution slot, a reference baseline update is performed first. The reference baseline includes at least the temperature reference baseline, the reflection power reference baseline, the current reference baseline, the phase command reference baseline, the precoding usage frequency reference baseline, and the channel residual reference baseline.

[0136] Among them, the temperature reference baseline, the reflection power reference baseline, the current reference baseline, and the phase command reference baseline correspond to the reference values ​​in S1 before the average temperature, linear reflection power, power amplifier average current, and phase command quantity enter the dimensionless conversion, respectively; the precoding usage frequency reference baseline corresponds to the reference value in S1 before the precoding usage frequency enters the normalization; the channel residual reference baseline corresponds to the reference value in S1 before the uplink and downlink equivalent channel state data form the channel residual quantity. Therefore, after the reference baseline is updated, its write-back position is the dimensionless conversion link of the next time slot S1, rather than writing back to the state inversion of S2.

[0137] The reference baseline update adopts a restricted recursive update method. In the specific process, the local array cluster-level standardized state observation results and the execution results of the current time slot are first read. Then, the current observation trend is fused with the existing reference baseline according to the preset update step size to form the updated reference baseline.

[0138] The restrictions here have two layers of meaning: the first layer is that the current observation results are only allowed to participate in the update when the current time slot is determined to be a high-confidence execution time slot; the second layer is that the current observation results will not completely replace the original reference baseline, but will be fused with the original baseline according to the update step size.

[0139] The preferred step size for updating the reference baseline is 0.02 to 0.20, more preferably 0.05 to 0.10. If it is less than 0.02, the reference baseline will respond too slowly to long-term environmental changes and equipment aging; if it is greater than 0.20, short-term disturbances will be written to the reference layer too quickly. This range is consistent with the slow change characteristics of equipment-side operating data and communication feedback data.

[0140] After the reference baseline is updated, the sensitivity map is updated. The sensitivity map describes the mapping relationship between the local distortion state estimation result and the local array element cluster-level normalized state observation result. Its update depends on two inputs: one is the local distortion state estimation result of the current time slot; the other is the difference between the local array element cluster-level normalized state observation result of the current time slot and the sensitivity map output.

[0141] In this embodiment, the latter difference is used as the observation residual. Therefore, the sensitivity mapping update must be performed after the reference baseline update: because after the reference baseline update is completed, the reference layer corresponding to the current time slot observation has been determined. At this time, the sensitivity mapping is updated by combining the local distortion state estimation result and the observation residual.

[0142] The sensitivity mapping update adopts a restricted recursive identification update method. In the specific process, the correction direction of the sensitivity mapping is first determined based on the estimation result of the local distortion state of the current time slot and the current observation residual. Then, the recursive identification is performed according to the sensitivity mapping update step size to obtain the updated sensitivity mapping.

[0143] The restrictions here also include two layers: the first layer is that updates are only allowed in high-confidence execution slots; the second layer is that sensitivity mapping in a single slot only allows small-step corrections and does not allow large jumps.

[0144] The sensitivity mapping update step size is preferably 0.005 to 0.05, more preferably 0.01 to 0.03. When it is lower than 0.005, the sensitivity mapping responds too slowly to long-term equipment drift and environmental changes; when it is higher than 0.05, the single time slot residual will change the mapping relationship too quickly. This range is smaller than the reference baseline update step size range because the sensitivity mapping belongs to the structural relationship from state to observation, and the rate of change should be slower than that of the reference baseline.

[0145] After completing the reference baseline update and sensitivity mapping update, the update results are written back to the next time slot. Specifically, the updated temperature reference baseline, reflection power reference baseline, current reference baseline, phase command reference baseline, precoding usage frequency reference baseline, and channel residual reference baseline are written back to the local array element cluster-level normalized state observation results construction process in the next time slot S1; the updated sensitivity mapping is written back to the array distortion implicit state vector inversion process in the next time slot S2.

[0146] Compared with the existing indiscriminate write-back method, this step can incorporate the execution quality of the communication side, the stability of the service direction, and the continuity of the target trajectory into the write-back gating conditions, so that the long-term reference layer and the state mapping layer are updated only by high-reliability time slots, which meets the actual needs of current base station equipment operation and communication and sensing joint processing.

[0147] Example 2

[0148] like Figure 3 As shown, this invention also discloses a joint optimization system for integrated sensing network beamforming and target detection, comprising:

[0149] The state observation construction module is configured to acquire the device-side operation data, communication feedback data and sensing observation data of the target base station in the current time slot, divide the local array element clusters according to the array physical adjacency relationship, and construct the local array element cluster-level standardized state observation results, local array element cluster occupancy weights and candidate target initial geometric observation results;

[0150] The distortion state inversion module is configured to perform graph-constrained recursive inversion based on the local array element cluster-level standardized state observation results, combined with the local array element cluster occupancy weight, the physical adjacency relationship between local array element clusters, the reference baseline, the sensitivity mapping, and the local distortion state estimation results of the previous time slot, to obtain the array distortion implicit state vector and form the local distortion intensity and the overall array distortion intensity.

[0151] The consistency joint correction module is configured to construct array distortion compensation results around the array distortion implicit state vector, calculate the communication sensing consistency correction amount by combining the direction consistency loss of the service beam direction and the candidate target initial direction in the candidate target initial geometric observation results, and simultaneously correct the beamforming results and target detection results based on the communication sensing consistency correction amount, local distortion intensity and array overall distortion intensity to obtain the joint correction result;

[0152] The closed-loop update module is configured to output joint correction results and, based on the high-confidence execution results, perform constrained recursive updates to the reference baseline and sensitivity mapping. The updated results are then written back to the local array element cluster-level normalized state observation results construction process and the array distortion implicit state vector inversion process in the next time slot.

[0153] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A joint optimization method for beamforming and target detection in a sensor-integrated network, characterized in that, Includes the following steps: S1. Obtain the device-side operation data, communication feedback data and sensing observation data of the target base station in the current time slot. Divide the local array element clusters according to the array physical adjacency relationship, and construct the local array element cluster-level standardized state observation results, local array element cluster occupancy weights and candidate target initial geometric observation results. S2, based on the local array element cluster-level standardized state observation results, combined with the local array element cluster occupancy weight, the physical adjacency relationship between local array element clusters, the reference baseline, the sensitivity mapping and the local distortion state estimation results of the previous time slot, performs graph-constrained recursive inversion to obtain the array distortion hidden state vector, and forms the local distortion intensity and the overall array distortion intensity. S3, construct the array distortion compensation result around the array distortion implicit state vector, calculate the communication sensing consistency correction amount by combining the direction consistency loss of the initial direction of the candidate target in the service beam direction and the initial geometric observation result of the candidate target, and simultaneously correct the beamforming result and the target detection result based on the communication sensing consistency correction amount, the local distortion intensity and the overall array distortion intensity to obtain the joint correction result; S4 outputs the joint correction results, and updates the reference baseline and sensitivity mapping based on the high-confidence execution results through limited recursion. The updated results are then written back to the local array element cluster-level normalized state observation results construction process and array distortion implicit state vector inversion process in the next time slot.

2. The method for joint optimization of beamforming and target detection in a sensory integrated network according to claim 1, characterized in that, The specific operations in S1 for acquiring equipment-side operational data, communication feedback data, and sensing observation data and dividing local array element clusters are as follows: acquiring the average temperature, linear reflection power, power amplifier average current, and phase shifter control word corresponding to each local array element cluster as equipment-side operational data; acquiring the precoding usage frequency and uplink / downlink equivalent channel state data after reciprocity calibration and subcarrier alignment within the current window as communication feedback data; and acquiring the angular spectrum distribution results and candidate target initial geometric observation results as sensing observation data. Based on the physical adjacency of the array, and combined with the distribution of shared power supply branches and shared heat dissipation paths, the local array element clusters are divided.

3. The method for joint optimization of beamforming and target detection in a sensory integrated network according to claim 2, characterized in that, The specific operations for constructing the standardized state observation results and local array cluster occupancy weights in S1 are as follows: First, timestamp correction is performed on the equipment-side operation data, communication feedback data, and sensing observation data based on a unified time base. Then, short-term missing samples are filled in according to adjacent time slot data, and data samples that exceed the preset operation boundary are removed. Subsequently, the corresponding data of each local array element cluster were subjected to dimensional uniform conversion according to the reference temperature baseline, reference reflection power baseline, reference current baseline and reference phase command baseline. Among them, the phase shifter control word was converted into a phase command quantity through the code word phase lookup table and then participated in the standardization process. The uplink and downlink equivalent channel state data were converted into channel residual quantity through residual conversion. The angular spectrum distribution result was converted into angular spectrum perturbation coefficient through adjacent time slot correlation conversion. Then, the local array element cluster occupancy weight is formed based on the precoding frequency and the current time slot transmission energy.

4. The method for joint optimization of beamforming and target detection in a sensory integrated network according to claim 1, characterized in that, The specific operation of the graph-constrained recursive inversion in S2 is as follows: First, an initial estimate of the local distortion state of the current time slot is formed based on the reference baseline and the local distortion state estimation result of the previous time slot. Then, observation fitting constraints are established around the difference between the observation results of the local element cluster-level standardized state and the sensitivity mapping output. Spatial continuity constraints are established around the difference of the local distortion state between physically adjacent local element clusters. Time recursion constraints are established around the deviation between the local distortion state estimation result of the current time slot and the local distortion state estimation result of the previous time slot. Subsequently, the local distortion state of all local element clusters is iteratively updated by combining the observation fitting constraints, spatial continuity constraints and time recursion constraints until the convergence condition is met, thereby obtaining the array distortion implicit state vector.

5. The method for joint optimization of beamforming and target detection in a sensory integrated network according to claim 4, characterized in that, The specific operations for forming the local distortion intensity and the overall array distortion intensity in S2 are as follows: the amplitude drift ratio and normalized phase drift ratio in the local distortion state estimation results are used as local distortion characterization quantities, and the actual phase drift amount is calculated based on the normalized phase drift ratio and the upper bound of the phase drift. After the array distortion implicit state vector is determined, the local distortion intensity is first formed based on the magnitude of the local distortion state estimation results. Then, the weighted aggregation of all local distortion intensities is performed in combination with the occupancy weight of each local array element cluster to form the overall array distortion intensity. When the local distortion state estimation result exceeds the preset distortion boundary, the current time slot is marked as a strong distortion condition, and conservative correction processing is performed in the subsequent joint correction process.

6. The method for joint optimization of beamforming and target detection in a sensory integrated network according to claim 1, characterized in that, The specific operation for calculating the communication sensing consistency correction in S3 is as follows: First, based on the ideal array direction vector and the array distortion compensation result, the directional consistency loss corresponding to the service beam direction and the directional consistency loss corresponding to the initial direction of the candidate target in the initial geometric observation result of the candidate target are formed. Then, the communication sensing consistency correction is determined based on the aforementioned two directional consistency losses and the difference between them.

7. The method for joint optimization of beamforming and target detection in a sensory integrated network according to claim 6, characterized in that, The specific operation of synchronously correcting beamforming results and target detection results in S3 is as follows: First, the corrected beamforming results are solved based on the communication sensing consistency correction amount, local distortion intensity, and overall array distortion intensity. Then, the basic detection threshold is jointly amplified and corrected based on the directional consistency loss corresponding to the initial direction of the candidate target in the initial geometric observation results of the candidate target and the overall array distortion intensity. Finally, the angle deviation compensation is performed on the target angle estimation results in the initial geometric observation results of the candidate target based on the sensitivity relationship between the array direction vector and the array distortion implicit state vector.

8. The method for joint optimization of beamforming and target detection in a sensory integrated network according to claim 7, characterized in that, The specific operation for correcting the target detection result in S3 is as follows: First, construct the target spatial filtering weight based on the corrected target angle estimation result and the array distortion compensation result. Then, use the target spatial filtering weight to perform spatial filtering on the slow time echo sequence. Subsequently, perform Doppler spectrum reconstruction on the filtered slow time echo sequence and determine the corrected target radial velocity estimation result based on the main energy distribution of the Doppler spectrum. Finally, combine the corrected detection threshold, the corrected target angle estimation result, and the corrected target radial velocity estimation result to form the target detection result.

9. The method for joint optimization of beamforming and target detection in a sensory integrated network according to claim 1, characterized in that, The specific operations for updating the reference baseline and sensitivity mapping based on the high-reliability execution results in S4 are as follows: First, the reference baseline is updated by a restricted recursive method based on the local array element cluster-level normalized state observation results and the execution results of the current time slot. Then, the sensitivity mapping is updated by a restricted recursive method based on the local distortion state estimation results of the current time slot and the difference between the local array element cluster-level normalized state observation results and the sensitivity mapping output. Finally, the updated reference baseline and sensitivity mapping are written back to the construction process of the local array element cluster-level normalized state observation results and the array distortion implicit state vector inversion process of the next time slot.

10. A joint optimization system for integrated sensing network beamforming and target detection, employing the joint optimization method for integrated sensing network beamforming and target detection as described in any one of claims 1-9, characterized in that, include: The state observation construction module is configured to acquire the device-side operation data, communication feedback data and sensing observation data of the target base station in the current time slot, divide the local array element clusters according to the array physical adjacency relationship, and construct the local array element cluster-level standardized state observation results, local array element cluster occupancy weights and candidate target initial geometric observation results; The distortion state inversion module is configured to perform graph-constrained recursive inversion based on the local array element cluster-level standardized state observation results, combined with the local array element cluster occupancy weight, the physical adjacency relationship between local array element clusters, the reference baseline, the sensitivity mapping, and the local distortion state estimation results of the previous time slot, to obtain the array distortion implicit state vector and form the local distortion intensity and the overall array distortion intensity. The consistency joint correction module is configured to construct array distortion compensation results around the array distortion implicit state vector, calculate the communication sensing consistency correction amount by combining the direction consistency loss of the service beam direction and the candidate target initial direction in the candidate target initial geometric observation results, and simultaneously correct the beamforming results and target detection results based on the communication sensing consistency correction amount, local distortion intensity and array overall distortion intensity to obtain the joint correction result; The closed-loop update module is configured to output joint correction results and, based on the high-confidence execution results, perform constrained recursive updates to the reference baseline and sensitivity mapping. The updated results are then written back to the local array element cluster-level normalized state observation results construction process and the array distortion implicit state vector inversion process in the next time slot.

Images