A beacon source assisted self-calibration and robust beamforming method

Abstract

The present application relates to the technical field of array signal processing and wireless communication, and particularly relates to a beacon source assisted self-calibration and robust beamforming method, comprising: collecting array snapshot data and constructing sample covariance; extracting spatial features based on coarse direction prior of auxiliary signal source and performing reliability determination; estimating geometric drift parameters through per-element phase ratio estimation; constructing parameterized covariance model based on calibrated array manifold and estimating noise power; estimating power coefficients of each signal source by using non-negative constraint fitting estimation; reconstructing INCM and optionally diagonally loading to enhance reversibility; calculating MVDR weight vector based on reconstructed INCM and outputting beamforming result; when reliability determination fails, falling back to nominal geometry or preset geometry to avoid false calibration. The present application has the advantages of geometric self-calibration, low snapshot robustness, false calibration prevention and wide application range.

Description

Technical Field

[0001] This invention relates to the fields of array signal processing and wireless communication / radar / electronic countermeasures, and particularly to a beacon-source-assisted self-calibration and robust beamforming method. It addresses virtual arrays formed by mobile platform clusters (e.g., UAV swarms), achieving adaptive beamforming with interference suppression and robust reception under conditions of quasi-static drift in the relative geometry of the formation and a limited number of snapshots available within the coherent processing time. The virtual array can be a sparse array with extended aperture characteristics, applicable to different one-dimensional formation geometries, and independent of any specific array structure. Background Technology

[0002] Due to their high mobility and flexible deployment, unmanned aerial vehicle (UAV) platforms are widely used for tasks such as aerial communication, remote sensing, and electronic warfare. When multiple UAVs work together to form a swarm, their onboard antennas can work together to form a virtual array and perform coordinated beamforming to obtain a larger equivalent aperture and higher spatial resolution, thereby improving link budget, coverage, and anti-interference capabilities.

[0003] However, in practical applications, the beamforming performance of a cluster of virtual arrays is highly sensitive to the accuracy of the relative geometry of the formation. Due to navigation errors, airframe motion, wind disturbances, and formation control errors, the actual relative position of a UAV often deviates from the nominal formation, exhibiting a quasi-static but non-negligible drift within short coherent processing intervals. If this drift is ignored in the beamforming design and the nominal steering vector is used, steering vector mismatch can easily occur, leading to a significant decrease in output signal-to-interference-plus-noise ratio and other performance characteristics, especially in the presence of strong co-channel interference sources.

[0004] From the perspective of classical array processing, minimum variance distortionless response (MVDR / Capon) beamforming usually achieves interference suppression by estimating and inverting the interference plus noise covariance matrix. However, in high-speed maneuvering scenarios such as UAV swarms or scenarios with limited processing time, the number of snapshots obtainable in each coherent processing interval is often small, and the sample covariance matrix is ​​prone to ill-conditioned behavior, which causes severe performance loss to beamformers based on sample matrix inversion.

[0005] To improve robustness under limited snapshot and mismatch conditions, various robust adaptive beamforming schemes have been proposed in the prior art, such as diagonal loading, robust Capon, and design methods based on uncertainty sets. Another common approach is to regularize or reconstruct the interference plus noise covariance matrix (e.g., based on spatial spectrum or parameterized interference model) to reduce the impact of signal leakage and stabilize matrix inversion.

[0006] However, many of the above methods often use a general uncertain set to characterize array perturbations or assume that the array manifold is already well calibrated. When there is quasi-static geometric drift in the virtual array of a drone swarm, these methods may not be able to utilize strong undesired radiation sources that may exist in the scene for data-driven geometric learning, thus making it difficult to simultaneously ensure geometric consistency and low snapshot stability.

[0007] In addition, array self-calibration technology usually requires joint estimation of sensor position (or amplitude and phase response) and source parameters from snapshot data. However, many self-calibration methods are characterized by strong iterativeness and high data requirements. When snapshots are scarce or the sample covariance is strongly contaminated by co-frequency sources, their reliability may decrease significantly.

[0008] Therefore, to address the robust beamforming requirements of virtual arrays in UAV swarms under quasi-static formation drift and low snapshot conditions, a solution is urgently needed that can reliably operate even in the presence of strong co-channel interference. Without considering strong interference solely as a negative factor, at least one dominant undesired emission source can be used as a "calibration beacon" to assist in learning the array's equivalent geometry. Based on this, a geometrically consistent interference plus noise covariance characterization can be constructed to improve the stability and mismatch resistance of adaptive beamforming such as MVDR under low snapshot conditions. (Invention Content)

[0009] In applications where mobile platform clusters (e.g., drone swarms) collaboratively receive signals to form virtual arrays, adaptive beamforming typically relies on the accuracy of the array steering vector and the interference-plus-noise covariance matrix (INCM). In practical engineering environments, at least two key challenges exist:

[0010] (1) Array geometric uncertainty leads to steering vector mismatch: The cluster formation may experience slow but non-negligible relative geometric drift during coherent processing time, causing the actual steering vector to deviate from the nominal model, resulting in main lobe shift, shallow null trap or interference leakage.

[0011] (2) Low snapshot speed leads to unstable covariance estimation: In short dwell, high maneuver or low latency scenarios, the number of available snapshots is limited, the estimation error of the sample covariance matrix is ​​large or even ill-conditioned, and direct inversion will amplify the error, causing beamforming performance to fluctuate greatly.

[0012] Therefore, this invention proposes a beacon-source-assisted self-calibration and robust beamforming method. Under the conditions of "geometric drift + low snapshot", the array geometry is self-calibrated by using at least one observable auxiliary signal source in the scene (which can be a specially set calibration source or a dominant undesired radiation source). Based on this, a more stable INCM is reconstructed to achieve robust adaptive beamforming.

[0013] Specific technical solutions:

[0014] This invention provides a beacon-source-assisted self-calibration and robust beamforming method, which can be executed by a processor and implemented as a device / system / equipment / storage medium. Its core process includes:

[0015] S1: Collect array snapshot data and construct sample covariance;

[0016] Let the first The receive vector of each snapshot is ,in For the number of collaborative array elements / platforms, , This represents the number of snapshots.

[0017] Let the direction of the desired signal (SOI) be... Its complex base band symbol is ;

[0018] exist The directions of the unwanted signal sources are: , symbol is ;

[0019] The noise vector is .

[0020] Guide vector Indicates direction Below, affected by array disturbance parameters The actual array response after the impact.

[0021] The receiving model can then be expressed as:

[0022] ,

[0023] Will A data matrix is ​​composed of snapshots. The sample covariance matrix is ​​defined as:

[0024] ,

[0025] To enhance numerical stability, Hermitization can be performed (wherein) (Indicates conjugate transpose)

[0026] ,

[0027] Let the nominal geometric position be ,in For the first The nominal position of each array element;

[0028] The actual location is recorded as Using drift amount Represented as .

[0029] Combining drift into a vector .

[0030] To eliminate the indistinguishability of global translation, a fixed reference element can be selected, for example, let .

[0031] In a class of realizable one-dimensional equivalent arrays, the nominal steering vector It can be written as:

[0032] ,

[0033] in A constant related to wavelength and coordinate normalization method (e.g., can be taken as...) or wait).

[0034] The actual steering vector after considering drift is:

[0035] ,

[0036] Note: In addition to positional drift, It can also include more general disturbances such as amplitude error, phase error, and synchronization error. It generalizes to the corresponding set of error parameters.

[0037] S2: Extract spatial features based on coarse directional priors of auxiliary signal sources and determine reliability;

[0038] Choose at least one auxiliary signal source, whose direction is coarsely prior and denoted as . (This prior knowledge may come from detection / tracking / historical estimation, etc.)

[0039] Calculate the nominal steering vector .

[0040] right Perform eigenvalue decomposition:

[0041] ,

[0042] in The eigenvector matrix, For eigenvalue diagonal matrices (which can be sorted by...) (Sort).

[0043] Define the correlation metric ( for -norm):

[0044] ,

[0045] Pick and order , .

[0046] Set threshold parameters :

[0047] like Then stop geometric estimation and fall back to nominal geometry (e.g., take...). );

[0048] like Then proceed to the drift estimation step.

[0049] S3: Estimate geometric drift parameters by the phase ratio of each array element;

[0050] In a class of realizable models, the direction of the auxiliary signal source The actual steering vector can be expressed as the element-by-element phase perturbation of the nominal steering vector:

[0051] ,

[0052] Utilizing spatial features extracted from data With nominal orientation Construct the element-by-element ratio (where Representing vectors The (number of elements)

[0053] ,

[0054] To eliminate common phase factors, reference element normalization can be used, for example, by using... And the drift estimate is obtained from the phase inverse solution ( (For complex phase operators)

[0055] ,

[0056] Thus obtain Furthermore, during the phase inversion process, the phase of the element-by-element ratio is unwound or made continuous, and a reliability criterion is set based on the phase sensitivity of the auxiliary signal source direction and the fitting residual. When the phase sensitivity is lower than the preset threshold, or the phase fitting residual is higher than the preset threshold, the current drift estimation result is stopped, and the system reverts to the nominal geometry or preset geometric parameters.

[0057] Further for any direction Define the calibrated guide vector ;

[0058] Specifically, the calibration guidance for SOI is... , No. The calibration guidance for each interference source is .

[0059] S4: Construct a parameterized covariance model based on the calibrated array manifold and estimate the noise power;

[0060] To improve stability at low snapshot speeds, a power factor is introduced. (SOI power) and (No. (interference power), and noise power .

[0061] Constructing a parameterized model:

[0062]

[0063] in It is an identity matrix.

[0064] Noise power can be estimated using small eigenvalues: Let the eigenvalues ​​of equation (6) be... Take the dimension parameter (For example ),but:

[0065] ,

[0066] S5: Estimate the power coefficients of each signal source using non-negative constraint fitting;

[0067] Define vectorization operators ,structure

[0068] ,

[0069] Let the power vector ,but .

[0070] If the noise power has already been estimated from small eigenvalues, the noise term can be subtracted from the sample covariance matrix first, and then a non-negative fit can be performed on the power coefficients of the SOI and each unwanted signal source; alternatively, the noise term can be treated as an independent dictionary column and used as a non-negative estimate of the power of each signal. One of these two methods should be chosen to avoid the noise power being double-counted.

[0071] To ensure the physical meaning of power and improve robustness, nonnegative constrained least squares is used (transforming the complex problem into an equivalent real number form):

[0072] ,

[0073] S6: Refactor INCM and optionally load diagonally to enhance reversibility;

[0074] Based on the estimation results Reconstruct the interference plus noise covariance matrix (INCM), i.e., without including SOI terms:

[0075] ,

[0076] in for The corresponding number in the middle The components of each interference source.

[0077] When the noise term is subtracted before the signal power is fitted in S5, the estimated noise power term should be added to the interference covariance when reconstructing INCM; when the noise term has already been estimated as a dictionary column in S5, the noise term will not be added again when reconstructing INCM.

[0078] To enhance reversibility, diagonal loading can be optionally applied: let the loading coefficient be... (Usually, the smaller value is taken), then

[0079] ,

[0080] S7: Calculate the MVDR weight vector based on the reconstructed INCM and output the beamforming results;

[0081] based on With SOI calibration guidance Construct MVDR weight vector :

[0082] ,

[0083] For any snapshot, the output is .

[0084] S8: When the reliability determination fails, revert to the nominal geometry (or preset geometry) to avoid miscalibration;

[0085] When the correlation determination fails (i.e.) When ), you can take it directly. (Or take preset geometric parameters), and construct accordingly. and This avoids performance degradation due to miscalibration caused by unreliable auxiliary information.

[0086] The technical effects of the technical solution provided by this invention are as follows:

[0087] (1) Geometric self-calibration: Array perturbation parameters are estimated using the spatial characteristics of the auxiliary signal source. This reduces the mismatch in the guide vector caused by geometric drift.

[0088] (2) Low-speed robustness: Reconstruction through "parametric covariance + non-negative power fitting" To improve the pathological state and the unstable state of seeking reversal.

[0089] (3) Error prevention calibration: through threshold parameters The rollback strategy mitigates the risk of miscalibration and improves engineering availability.

[0090] (4) Wide range of applications: The auxiliary signal source can be a dedicated beacon or a dominant undesired radiation source; It can be expanded to include multiple error parameters other than position drift. Attached Figure Description

[0091] Figure 1 This is a flowchart of the method described in this invention;

[0092] Figure 2 These are the actual drift and estimated drift in the embodiments. Detailed Implementation

[0093] Multi-scenario simulations using a single strong undesired radiation source as a beacon source verify the self-calibration capability, INCM reconstruction stability, and MVDR anti-interference performance improvement of this invention under formation geometry drift and low snapshot conditions. The following are simulation examples; the parameters are for illustrative purposes only and do not constitute a limitation on the scope of this invention.

[0094] Set operating parameters:

[0095] (1) Array / Squadron: Number of drones (number of array elements) The nominal position is normalized to half a wavelength (unit: ). , A sparse linear array (wavelength); the example uses a coprime construction. Nominal half-wavelength index set Remove duplicates and sort in ascending order to get .

[0096] (2) Geometric drift: The first Actual location in each scenario drift vector and fixed reference array elements ; Drift intensity is from Control, example selection (Unit is) ).

[0097] (3) Signal and Interference: Direction of arrival of the desired signal (SOI) There exists at least one undesirable radiation source, the direction of which is... And its interference-to-noise ratio is This undesired radiation source also serves as a "beacon / auxiliary source" (always treated as interference, not as a desired signal).

[0098] (4) Noise: The noise variance is normalized to .

[0099] (5) Quick Shots and Statistics: Number of Quick Shots (Low tempo), Monte Carlo times Input signal-to-noise ratio (Point-by-point frequency scanning).

[0100] (6) Algorithm hyperparameters: correlation threshold Diagonal loading coefficient ; Noise subspace dimension parameter ,in Number of unwanted radiation sources (example: single beacon case) ).

[0101] like Figure 1 The specific process is as follows:

[0102] (1) Construct the nominal array manifold and the receiver model.

[0103] No. The receiving vector of each snapshot is denoted as... In any drifting scenario Down: in In drift Down The guide vector; For SOI symbols; For the first Symbols of undesired radiation sources (including beacon sources); It is noise.

[0104] (2) Generate drift scene and generate snapshot data.

[0105] generate Sub-tests: SOI / beacon / noise snapshots are randomly generated for each test; geometric drift is included in this test. Maintain a near-static state within a snapshot (do not move around) (Changes), but changes between different experiments.

[0106] (3) Calculate the sample covariance and perform Hermitization.

[0107] Will The snapshots are stacked as ,calculate And can be adopted Enhance numerical stability.

[0108] (4) Spatial feature extraction and reliability gating of beacon sources.

[0109] right Perform eigenvalue decomposition to obtain a set of eigenvectors; use the beacon source direction prior. Computational nominal orientation Select from each feature vector that is similar to The vector with the highest normalized correlation is used as the feature of the beacon space; if the highest correlation is less than a threshold... If the current data does not meet the reliable calibration conditions, the calibration is skipped and the system reverts to the nominal geometry (set). ).

[0110] (5) Beacon-assisted self-calibration (if successful, estimate drift).

[0111] When the correlation passes through the gate, the phase difference is extracted using the element-by-element ratio of the "eigenvector / nominal guidance", and the drift estimate is obtained by inverse solving after normalization of the reference elements. ;Depend on Construct a calibration array manifold to obtain the calibration SOI guide. With calibration interference guidance .

[0112] (6) INCM reconstruction (non-negative power fitting + optional diagonal loading).

[0113] A low-dimensional parametric covariance model is constructed using calibration guidance and coarse DOA information. First, the noise power is estimated by the mean of the minimum eigenvalues, and then the power of each interference is estimated by fitting non-negative power, thus reconstructing the interference plus noise covariance matrix. ; Optional Diagonal loading (loading coefficient) This is to improve the stability of inversion under low-speed conditions.

[0114] (7) MVDR weight calculation, output and index statistics.

[0115] Use calibration guide With refactoring INCM Calculate weights in MVDR format , get output .exist The statistical output signal-to-interference-plus-noise ratio in this experiment is: in These represent the components in the output contributed by SOI, interference (including beacon sources), and noise, respectively. Approximated by the Monte Carlo average. Simultaneously, statistical calibration gate pass rate (correlation exceeding...) The proportion of the test) and the drift estimation error (e.g. The RMSE (Reliability and Reliability of Self-calibration) is used to characterize self-calibration reliability.

[0116] Comparison of effects:

[0117] like Figure 2As shown, compared with traditional robust methods, this invention can significantly recover the SINR loss caused by guide mismatch when geometric drift exists. It achieves more stable interference suppression capability under low snapshot conditions through "beacon-assisted geometry learning + geometrically consistent INCM reconstruction". When the gating fails, it automatically reverts to the nominal geometry, thereby avoiding performance degradation caused by miscalibration.

Claims

1. A beacon source-assisted self-calibration and robust beamforming method, characterized in that, Includes the following steps: S1. Collect array snapshot data and construct sample covariance; S2. Extract spatial features based on coarse directional priors of auxiliary signal sources and determine reliability. S3. Estimate the geometric drift parameters by the phase ratio of each array element; S4. Construct a parameterized covariance model based on the calibrated array manifold and estimate the noise power; S5. Use non-negative constraint fitting to estimate the power coefficients of each signal source; S6. Reconstruct INCM and optionally load diagonally to enhance reversibility; S7. Calculate the MVDR weight vector based on the reconstructed INCM and output the beamforming results; S8. When the reliability determination fails, revert to the nominal geometry or the preset geometry to avoid miscalibration.

2. The beacon source-assisted self-calibration and robust beamforming method according to claim 1, characterized in that, The specific method for S1 is as follows: Let the first The receive vector of each snapshot is ,in For the number of collaborative array elements / platforms, , This refers to the number of snapshots; Let the desired signal and direction be... Its complex base band symbol is ; exist The directions of the unwanted signal sources are: , symbol is ; The noise vector is ; Guide vector Indicates direction Below, affected by array disturbance parameters The actual array response after the impact; The receiving model is then represented as: ， Will A data matrix is ​​composed of snapshots. The sample covariance matrix is ​​defined as: ， To enhance numerical stability, Hermitization was performed, in which... Indicates conjugate transpose: ， Let the nominal geometric position be ,in For the first The nominal position of each array element; The actual location is recorded as Using drift amount Represented as ; Combining drift into a vector ; To eliminate the indistinguishability of global translation, a fixed reference element is selected, for example, let... ; In a class of realizable one-dimensional equivalent arrays, the nominal steering vector Written as: ， in These are constants related to wavelength and coordinate normalization method; The actual steering vector after considering drift is: ， In addition to positional drift, It also includes amplitude error, phase error, and synchronization error disturbances. It generalizes to the corresponding set of error parameters.

3. The beacon source-assisted self-calibration and robust beamforming method according to claim 1, characterized in that, The specific method for S2 is as follows: Choose at least one auxiliary signal source, whose direction is coarsely prior and denoted as . ; Calculate the nominal steering vector ; right Perform eigenvalue decomposition: ， in The eigenvector matrix, It is an eigenvalue diagonal matrix; Define a correlation metric. for - Norm: ， Pick and order , ; Set threshold parameters : like If so, then stop geometric estimation and revert to nominal geometry; like Then proceed to the drift estimation step.

4. The beacon source-assisted self-calibration and robust beamforming method according to claim 1, characterized in that, The specific method for S3 is as follows: In a class of realizable models, the direction of the auxiliary signal source The actual steering vector can be expressed as the element-by-element phase perturbation of the nominal steering vector: ， Utilizing spatial features extracted from data With nominal orientation Construct the element-wise ratio, where Representing vectors The One element: ， To eliminate common phase factors, reference element normalization can be used, for example, by using... The drift estimate is obtained from the phase inverse solution. For complex phase operators: ， Thus obtain During the phase inversion process, the phase of the element-by-element ratio is unwound or made continuous, and a reliability criterion is set based on the phase sensitivity and fitting residual of the auxiliary signal source direction. When the phase sensitivity is lower than the preset threshold or the phase fitting residual is higher than the preset threshold, the current drift estimation result is stopped and the algorithm is rolled back to the nominal geometry or preset geometry parameters. For any direction Define the calibrated guide vector ; SOI calibration guidance is , No. The calibration guidance for each interference source is .

5. The beacon source-assisted self-calibration and robust beamforming method according to claim 1, characterized in that, The specific method for S4 is as follows: To improve stability at low snapshot speeds, a power factor is introduced. That is, SOI power, and That is, the first Interference power and noise power ; Constructing a parameterized model: ， in It is the identity matrix; Noise power can be estimated using small eigenvalues: Let the eigenvalues ​​of equation (6) be... Take the dimension parameter ,but: 。 6. The beacon source-assisted self-calibration and robust beamforming method according to claim 1, characterized in that, The specific method for S5 is as follows: Define vectorization operators ,structure: ， Let the power vector ,but ; To ensure the physical meaning of power and improve robustness, nonnegative constrained least squares is used: 。 7. The beacon source-assisted self-calibration and robust beamforming method according to claim 1, characterized in that, The specific method for S6 is as follows: Based on the estimation results Reconstruct the interference plus noise covariance matrix, i.e., one that does not contain SOI terms: ， in for The corresponding number in the middle The components of each interference source; To enhance reversibility, diagonal loading is applied: Let the loading coefficient be... ,but: 。 8. The beacon source-assisted self-calibration and robust beamforming method according to claim 1, characterized in that, The specific method for S7 is as follows: based on With SOI calibration guidance Construct MVDR weight vector : ， For any snapshot, the output is .

9. The beacon source-assisted self-calibration and robust beamforming method according to claim 1, characterized in that, The specific method for S8 is as follows: When the correlation determination fails, that is... At that time, take directly Alternatively, preset geometric parameters can be taken, and construction can be carried out accordingly. and This avoids performance degradation due to miscalibration caused by unreliable auxiliary information.

Images