Intelligent multi-angle beam recovery method for deformation phased-array antenna

Abstract

The invention discloses an intelligent multi-angle beam recovery method for a deformation phased-array antenna, and mainly solves the problem of poor beam recovery effect in a complex deformation scene in the prior art. Comprising the following steps: 1) arranging a displacement sensor on the back of each unit of the phased-array antenna; 2) obtaining the normal vector of the upper surface of the antenna unit before and after deformation through vector operation and calculating a deflection angle; 3) sampling directional diagrams of different wave beam directions in the scanning surface to obtain directional diagram column vectors; 4) constructing a multi-angle beam recovery network based on a parallel sub-network architecture, wherein a sub-network is used for predicting excitation amplitudes and phases required when the deformation phased-array antenna executes different beam pointing tasks; 5, the deflection angle serves as input, the excitation amplitude and phase prediction result of the antenna unit are output through a network, and multi-angle beam recovery is achieved.According to the method, accurate excitation information can be rapidly provided for a feed system of the deformation phased-array antenna, and the beam scanning requirement is met.

Description

Technical Field

[0001] The present invention belongs to the field of wireless communication technology and further relates to beam recovery technology for deformable phased array antennas. Specifically, it provides an intelligent multi-angle beam recovery method for deformable phased array antennas, which can be used to improve the working performance of antenna systems in fields such as satellite communications, automotive radar, and aerospace. Background Art

[0002] Phased array antennas, with their beamforming and scanning properties, have become core components of modern communications and radar systems, and are widely used in satellite communications, automotive radar, aerospace, and other fields. Under actual operating conditions, factors such as spacecraft launch vibrations, vehicle driving bumps, and aerodynamic loads during high-speed flight can cause the phased array antenna surface to bend and deform. This leads to a series of problems such as beam pointing deviation and gain reduction, which not only seriously impair the antenna's signal transmission and reception performance, but also have a negative impact on the reliability and stability of the entire antenna system. Therefore, the research of beam recovery methods suitable for deformable phased array antennas has become a technical challenge that needs to be solved urgently.

[0003] Traditional beam recovery techniques rely primarily on optimization algorithms, which often have limitations in accuracy and adaptability. Examples include genetic algorithms in RAB Saleem, AA Shah, H. Munsif, AI Najam, S. Khattak and I.Ullah, “Radiation pattern correction of faulty planar phased array using genetic algorithm,” Advanced Electromagnetics, vol. 13, no. 2, pp. 23-31, July 2024, and compressed sensing algorithms in F. Zardi, G. Oliveri, M. Salucci and A. Massa, “Minimum-complexity failure correction in linear arrays via compressive processing,” IEEE Transactions on Antennas and Propagation, vol. 69, no. 8, pp. 4504–4516, Dec. 2020. Among them, genetic algorithms usually require a large number of iterative calculations, have slow calculation speeds, and cannot meet the real-time requirements of practical applications; compressed sensing algorithms encounter difficulties when processing antennas with complex deformations, and the recovery effect is usually affected by factors such as noise; most of these methods use phased array antennas that have not undergone deformation as research objects, and generally have problems such as low accuracy, poor real-time performance, and weak reliability, making it difficult to meet the needs of actual engineering.

[0004] Currently, there is a lack of technologies for beam restoration of deformed phased array antennas. For phased array antennas with known deformation, a deep learning-based beam restoration method is reported in K. Cao, C. Jin, B. Zhang, Q. Lv, and F. Lu, “Beam stabilization of deformed conformal array antenna based on physical-method-driven deep learning,” IEEE Transactions on Antennas and Propagation, vol. 71, no. 5, pp. 4115-4127, May 2023. This method uses the normal vector of the deformed antenna element as the network input and calculates the excitation amplitude and phase of each element, achieving millisecond-level beam restoration. However, this method is only applicable to scenarios where the main beam pointing is fixed and cannot be directly applied to scenarios requiring beam scanning.

[0005] At present, most traditional beam recovery technologies are based on undeformed phased array antennas. There is a lack of effective response measures for deformation scenarios caused by external forces and other factors in actual applications. Existing deep learning-based beam recovery methods cannot meet the common beam scanning requirements in fields such as communications and radar detection. Summary of the Invention

[0006] The purpose of the present invention is to address the deficiencies of the above-mentioned prior art and propose an intelligent multi-angle beam recovery method for a deformable phased array antenna. It is used to solve the problem of poor beam recovery effect of the prior art in complex deformation scenarios. By constructing a multi-angle beam recovery network based on a parallel subnetwork architecture, wherein the subnetwork is used to predict the excitation amplitude and phase required when the deformable phased array antenna performs different beam pointing tasks; each subnetwork uses the deflection angle calculated using the displacement sensor data on the antenna unit as the network input, and the amplitude and phase of the excitation of the phased array antenna unit used to restore the beam pointing as the output, thereby realizing multi-angle beam recovery of the deformable phased array antenna. The present invention can quickly provide accurate excitation information for the feeding system of the deformable phased array antenna, effectively realize multi-angle beam recovery of the deformable phased array antenna, and thus meet the beam scanning requirements.

[0007] To achieve the above object, the technical solution of the present invention includes the following steps:

[0008] (1) Arrange three displacement sensors on the back of each unit floor of the phased array antenna. Indicates the antenna units, where represents the number of antenna elements in the phased array antenna; assuming no deformation occurs The coordinates of the three displacement sensors on each antenna unit are 、 and , after deformation The coordinates of the three displacement sensors on each antenna unit are 、 and ;

[0009] (2) Constructing the original array No. 1 reference vector and reference vector No. 2 , and the deformed array No. 1 reference vector and reference vector No. 2 , and through vector cross multiplication operation, the surface normal vector of the i-th antenna unit when it is not deformed and deformed is obtained and :

[0010] ,

[0011] ,

[0012] in 、 and Respectively represent the unit vectors along the positive directions of the x, y, and z axes in the three-dimensional Cartesian coordinate system;

[0013] (3) Calculate the angle between the normal vectors of the upper surface of the current antenna unit before and after deformation, i.e., the deflection angle of the i-th antenna unit, according to the following formula: :

[0014] ,

[0015] Stipulate that clockwise rotation is negative and counterclockwise rotation is positive, calculate and collect the deflection angle of each antenna unit in the deformable phased array antenna, and construct the deflection angle column vector :

[0016] ,

[0017] (4) Assuming that the phased array antenna does not deform, there are M beam directions when its beam is scanned. Let j = 1, 2, ..., M represent the jth beam direction. Sample the directional pattern of the beam direction in the scanning plane to obtain the directional pattern column vector ;

[0018] (5) Based on the fully connected neural network architecture, a sub-network with the same number of beam pointings as the number of beam pointings is constructed, i.e., M sub-networks for recovering different beam pointings of the phased array antenna, and they are incorporated into the same parallel framework to obtain a multi-angle beam recovery network;

[0019] (6) Using the deflection angle column vector and the direction pattern column vector Construct a dataset corresponding to the jth beam pointing; take j = 1, 2, ..., M as the recovery task for each beam pointing in the multi-angle beam recovery network, and generate a dataset for each task; divide each dataset into a training set and a test set according to a preset ratio for training and testing each sub-network;

[0020] (7) Each sub-network is based on the deflection angle vector As the input of the network, the amplitude and phase of the phased array antenna unit excitation pointing to a specific beam are restored as output; the excitation amplitude and phase vector output by the jth sub-network are It is expressed as follows:

[0021] ,

[0022] in and They represent the excitation amplitude and phase used by the i-th antenna unit when the phased array antenna scans to the j-th beam direction;

[0023] (8) Based on the active pattern AEP, an active pattern module is built to determine the excitation amplitude and phase vector The directional pattern of the synthetic phased array antenna is shown as follows:

[0024] ,

[0025] in, represents the directivity pattern of the j-th beam pointing of the predicted deformable phased array antenna; The i-th antenna unit of the phased array antenna is deflected according to its Direction diagram after rotation; represents the phase difference;

[0026] (9) Connect the active pattern module to the output layer of each sub-network used to recover different beam pointing directions, and use the mean square error (MSE) to calculate the loss function:

[0027] ,

[0028] Set the maximum number of iterations to T, and T > 300; use the training set and the Adam optimization algorithm to train the multi-angle beam recovery network;

[0029] (10) Disconnect the active pattern module from the trained multi-angle beam recovery network, select the corresponding sub-network in the multi-angle beam recovery network according to the desired beam pointing, input the deflection angle column vector in the test set, quickly predict the amplitude and phase of the antenna unit excitation through the multi-angle beam recovery network, and transmit this data to the feeding system of the phased array antenna, re-feed the phased array antenna, and realize the recovery of the multi-beam pointing of the curved deformation phased array antenna.

[0030] Compared with the prior art, the present invention has the following advantages:

[0031] The present invention improves upon the existing beam recovery network architecture and constructs a multi-angle beam recovery network based on parallel subnetworks. Each subnetwork independently corresponds to a different beam pointing task and outputs multiple sets of excitation amplitudes and phases in parallel. By adopting an innovatively constructed parallel subnetwork architecture as the beam recovery network, the deflection angle calculated from the antenna unit displacement sensor data is used as input, directly outputting the excitation amplitude and phase required by the deformable phased array antenna under different beam pointing tasks. This allows the phased array antenna to quickly process beam pointing deviations caused by deformation, making it suitable for scenarios requiring real-time recovery of multiple beam pointings, enabling multi-angle beam recovery, thereby meeting the requirements of phased array beam scanning and improving the adaptability and reliability of phased array antennas in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a flowchart of the overall implementation of the method of the present invention;

[0033] Figure 2 The aircraft wing provided by the embodiment of the present invention Schematic diagram of the phased array antenna and sensor locations;

[0034] Figure 3 Schematic diagram of vectors on the phased array antenna unit in the present invention.

[0035] Figure 4 Schematic diagram of the normal vector and deflection angle of the phased array antenna unit in the present invention; (a) represents the normal vector, and (b) represents the deflection angle;

[0036] Figure 5 This is a schematic diagram of the multi-angle beam recovery network structure constructed in the present invention;

[0037] Figure 6 is a schematic diagram of the active pattern module in the present invention;

[0038] Figure 7 The X-band in the present invention Schematic diagram of phased array antenna model;

[0039] Figure 8The figures are simulation results of pattern recovery using the method of the present invention; (a)-(d) respectively represent the restoration of the pattern when the desired beam pointing is 0°, 4°, 8°, and 12°. DETAILED DESCRIPTION

[0040] The present invention will be further described below with reference to the accompanying drawings.

[0041] Example 1: Refer to the attached Figure 1 The present invention proposes an intelligent multi-angle beam restoration method for a deformable phased array antenna, which specifically includes the following steps:

[0042] Step 1) Place three displacement sensors on the back of each unit floor of the phased array antenna. Indicates the antenna units, where represents the number of antenna elements in the phased array antenna; assuming no deformation occurs The coordinates of the three displacement sensors on each antenna unit are 、 and , after deformation The coordinates of the three displacement sensors on each antenna unit are 、 and , according to the following formula:

[0043] ,

[0044] ,

[0045] ,

[0046] in, 、 and After deformation, The three displacement sensors on each antenna unit output the x, y, and z displacements.

[0047] Step 2) Construct the original array reference vector No. 1 and reference vector No. 2 , and the deformed array No. 1 reference vector and reference vector No. 2 , and through vector cross multiplication operation, the surface normal vector of the i-th antenna unit when it is not deformed and deformed is obtained and :

[0048] ,

[0049] ,

[0050] in 、 and Respectively represent the unit vectors along the positive directions of the x, y, and z axes in the three-dimensional Cartesian coordinate system;

[0051] In this embodiment, the above two normal vectors are obtained specifically according to the following steps. and :

[0052] (2.1) When the phased array antenna is not deformed, the original array reference vector No. 1 is constructed according to the coordinates of the three displacement sensors of the i-th antenna unit. and the original array No. 2 reference vector :

[0053] ,

[0054] ,

[0055] (2.2) Through vector and The cross product operation is performed to obtain the surface normal vector of the i-th antenna unit when no deformation occurs. ;

[0056] (2.3) When the phased array antenna is deformed, the coordinates of the three displacement sensors of the i-th antenna unit after deformation are used to construct the deformation array surface reference vector No. 1 and the deformed array No. 2 reference vector :

[0057] ,

[0058] ,

[0059] (2.4) Through vector and The cross product operation is performed to obtain the surface normal vector of the i-th antenna unit after deformation. .

[0060] Step 3) Calculate the angle between the normal vectors of the upper surface of the current antenna unit before and after deformation, i.e. the deflection angle of the i-th antenna unit, according to the following formula: :

[0061] ,

[0062] Stipulate that clockwise rotation is negative and counterclockwise rotation is positive, calculate and collect the deflection angle of each antenna unit in the deformable phased array antenna, and construct the deflection angle column vector :

[0063] ,

[0064] Step 4) Assume that the phased array antenna is not deformed and there are M beam directions when the beam is scanned. Let j = 1, 2, ..., M represent the jth beam direction. Sample the directional pattern of the beam direction in the scanning plane to obtain the directional pattern column vector ;

[0065] Step 5) Based on the fully connected neural network architecture, build subnetworks with the same number of beam pointings, that is, M subnetworks for recovering different beam pointings of the phased array antenna, and incorporate them into the same parallel framework to obtain a multi-angle beam recovery network. In this embodiment, each subnetwork built in this step has a four-layer structure, where the first three layers are hidden layers and the fourth layer is a linear output layer. The number of neurons in the first three layers is an adjustable parameter, with initial values ​​set to 32, 64, and 128, respectively. The number of neurons in the fourth layer is fixed to 2N.

[0066] Step 6) Use the deflection angle column vector and the direction pattern column vector Construct a dataset corresponding to the jth beam pointing; take j = 1, 2, ..., M as the recovery task for each beam pointing in the multi-angle beam recovery network to generate a dataset; divide each dataset into a training set and a test set according to a preset ratio for training and testing each sub-network; the preset ratio in this step is preferably 4:1. The generated dataset is obtained in the following way: Assume that the phased array antenna undergoes K different degrees of deformation, and the corresponding deflection angle column vector As the input data of the multi-angle beam recovery network, the direction pattern column vector As the label data of the multi-angle beam recovery network; use the input data and label data to form the corresponding dataset of the j-th beam pointing; finally, generate a set of datasets for each beam pointing recovery task in the multi-angle beam recovery network, where the total number of samples in each dataset is K.

[0067] Step 7) Each sub-network is transformed into a deflection angle vector As the input of the network, the amplitude and phase of the phased array antenna unit excitation pointing to a specific beam are restored as output; the excitation amplitude and phase vector output by the jth sub-network are It is expressed as follows:

[0068] ,

[0069] in and They represent the excitation amplitude and phase used by the i-th antenna unit when the phased array antenna scans to the j-th beam direction;

[0070] Step 8) Build an active pattern module based on the active pattern AEP to The directional pattern of the synthetic phased array antenna is shown as follows:

[0071] ,

[0072] in, represents the directivity pattern of the jth beam pointing of the predicted deformable phased array antenna; The i-th antenna unit of the phased array antenna is deflected according to its Direction diagram after rotation; Represents the phase difference, and its expression is as follows:

[0073] ,

[0074] in represents the frequency, is the speed of light, is the relative dielectric constant of the propagation medium, is the path difference between the ith antenna unit and the reference unit, where the reference unit is the first antenna unit. In this embodiment, the path difference is obtained according to the following steps:

[0075] (8.1) The data of all displacement sensors No. 1 when the phased array antenna is deformed are expressed as the following three column vectors:

[0076] ,

[0077] ,

[0078] ,

[0079] in 、 and Respectively represent the displacement data in the x-direction, y-direction, and z-direction collected by all No. 1 displacement sensors;

[0080] (8.2) The position vectors of the i-th antenna element and the reference element are obtained according to the following formula: :

[0081] ,

[0082] in 、 and are the components of the position vector in the x, y, and z directions;

[0083] (8.3) Based on position vector and beam pointing Get the path difference of the i-th antenna unit :

[0084] .

[0085] Step 9) Connect the active pattern module to the output layer of each sub-network used to recover different beam pointing directions, and use the mean square error (MSE) to calculate the loss function:

[0086] ,

[0087] Set the maximum number of iterations to T, and T > 300; use the training set and the Adam optimization algorithm to train the multi-angle beam recovery network;

[0088] Step 10) Disconnect the active pattern module from the trained multi-angle beam recovery network. Select the corresponding subnetwork in the multi-angle beam recovery network based on the desired beam pointing. Input the deflection angle column vector in the test set. Use the multi-angle beam recovery network to quickly predict the amplitude and phase of the antenna unit excitation. This data is transmitted to the feeding system of the phased array antenna, and the phased array antenna is re-fed to achieve the recovery of the multi-beam pointing of the curved deformation phased array antenna.

[0089] Example 2: The overall implementation steps of the beam recovery method proposed in this example are the same as those in Example 1. Figure 2-6 , give specific examples to further describe the implementation process of the present invention in detail:

[0090] Step 1. Figure 2 The assembly shown is mounted on the lower surface of the aircraft wing The phased array antenna is used as the research object, and the generation process of the deflection angle data as the network input is introduced:

[0091] First, three displacement sensors are arranged on the back of the floor of each unit of the phased array antenna. Taking the i-th antenna unit as an example, assuming that no deformation occurs, the coordinates of the displacement sensor No. 1 on the unit are known to be , the coordinates of the second displacement sensor are , the coordinates of displacement sensor No. 3 are After deformation occurs, it is known that the x, y, and z displacements output by the No. 1 displacement sensor on the unit are , the x, y, and z displacements output by displacement sensor No. 2 are , the x, y, and z displacements output by the No. 3 displacement sensor are , then the coordinates of displacement sensor No. 1 on the unit after deformation are , the coordinates of the second displacement sensor , and the coordinates of displacement sensor No. 3 Satisfies the following relationship:

[0092] <1>

[0093] <2>

[0094] <3>

[0095] When the phased array antenna is not deformed, the vector can be constructed based on the coordinates of the first, second and third displacement sensors of the i-th antenna unit. and ,like Figure 3 As shown. The expressions of the two vectors are:

[0096] <4>

[0097] <5>

[0098] By vector and The cross product operation can be obtained as Figure 3 The normal vector of the upper surface of the i-th antenna unit when no deformation occurs is shown in (a) , the specific calculation formula is as follows:

[0099] , <6>

[0100] When the phased array antenna is deformed, the vector can be constructed based on the coordinates of the first, second and third displacement sensors of the i-th antenna unit after deformation. and The expressions of the two vectors are:

[0101] <7>

[0102] <8>

[0103] By vector and The cross product operation can be obtained as Figure 4 The normal vector of the upper surface of the i-th antenna unit after deformation shown in (b) , the specific calculation formula is as follows:

[0104] , <9>

[0105] Then use the normal vector of the i-th antenna unit when it is not deformed and the normal vector of the i-th antenna element after deformation , based on the following formula, calculate the deflection angle of the antenna unit :

[0106] <10>

[0107] The deflection angle is the angle between the normal vectors of the antenna unit surface before and after deformation. It is stipulated that clockwise rotation is negative and counterclockwise rotation is positive, such as Figure 3 The same process is used to calculate and collect deformation. The deflection angle of each antenna unit of the phased array antenna is constructed as a Column vector of :

[0108] <11>

[0109] The deflection angle vector It will then serve as the input to the multi-angle beam recovery network.

[0110] Assuming that the phased array antenna has not deformed, there are M beam directions when its beam is scanned. Consider the jth beam direction, j = 1, 2, ..., M, and sample the directional pattern of the beam direction in the scanning plane with a sampling interval of 0.01°. 36001 data are obtained, which form the following: The direction pattern column vector :

[0111] <12>

[0112] in, Represents the direction pattern column vector The qth directional pattern data in . The directional pattern column vector It will be used as the label data of the sub-network that recovers the j-th beam pointing in the multi-angle beam recovery network.

[0113] Finally, the input data and label data are used to form a data set. The phased array antenna has a total of M beam directions. The multi-angle beam recovery network proposed in this invention corresponds to M sub-networks. Each sub-network only recovers one desired beam direction. Therefore, it is necessary to generate a data set for each sub-network separately, and finally generate M sets of data sets. Considering the j-th beam direction, the phased array antenna undergoes K different degrees of deformation, and the corresponding displacement sensor collects K sets of displacement data. According to the formula <1> to <11> K deflection angle column vectors can be calculated , as the input data of the jth sub-network for recovering the jth beam pointing. Since the jth sub-network remains unchanged when the phased array antenna undergoes different degrees of deformation, it is only necessary to convert the desired pattern column vector After replicating K times, we can obtain K labeled data. Then, we associate the deflection angle column vector with the labeled data to generate a dataset with a total of K samples. Finally, M subnetworks correspond to M datasets with a total of K samples. Each dataset is divided into a training set and a test set in a 4:1 ratio for training and testing each subnetwork.

[0114] Step 2. Figure 5 As shown in the figure, based on the fully connected neural network architecture, M sub-networks are built to recover the different beam pointing directions of the phased array antenna, and these sub-networks are incorporated into the same parallel framework to form a multi-angle beam recovery network. Each sub-network has a structure of three hidden layers and one linear output layer, and the number of neurons in each layer is 32, 64, 128, and 2N respectively. Each sub-network is based on the deflection angle vector As the input of the network, the amplitude and phase of the phased array antenna element excitation used to recover the specific beam pointing are output. The output vector of the jth sub-network is It can be expressed as:

[0115] <13>

[0116] in and They respectively represent the excitation amplitude and phase that should be adopted by the i-th antenna unit when the phased array antenna scans to the j-th beam direction.

[0117] Step 3. Recover the excitation amplitude and phase vector based on the output of the multi-angle beam recovery network , j = 1, 2, …, M, and quickly synthesize the directional pattern of the phased array antenna. This process is implemented by building an active directional pattern module based on the active element pattern (AEP) [DM Pozar, “The active element pattern,” IEEE Transactions on Antennas and Propagation, vol. 42, no. 8, pp. 1176–1178, Aug. 1994.] method, such as Figure 6 Taking the jth subnetwork as an example, this module can quickly predict the directional pattern of the deformable phased array antenna based on the following formula: :

[0118] <14>

[0119] in The active radiation pattern of the i-th antenna unit of the phased array antenna is calculated according to the deflection angle Direction diagram after rotation. Represents the phase difference, and its expression is:

[0120] <15>

[0121] in represents the frequency, is the speed of light, is the relative dielectric constant of the propagation medium, is the path difference between the i-th antenna element and the first antenna element (reference element). The path difference used in this invention is the projection of the actual distance between the first displacement sensor of the i-th antenna element and the first displacement sensor of the reference element in the beam pointing direction. When the phased array antenna deforms, the data from all first displacement sensors can be represented as the following three column vectors:

[0122] <16>

[0123] <17>

[0124] <18>

[0125] in 、 and Respectively represent the displacement data collected by all displacement sensors No. 1 in the x, y, and z directions. Taking the i-th antenna unit of the deformable phased array antenna as an example, its position vector with the reference unit is for:

[0126] <19>

[0127] in 、 and is the component of the position vector in the x, y, and z directions. Based on the position vector and beam pointing , we can get the path difference of the i-th antenna unit The expression is:

[0128] <20>

[0129] Step 4. Connect the active pattern module to the output layer of each sub-network used to recover different beam directions. Taking the j-th sub-network as an example, the output vector of the j-th sub-network in the second step is Input the active pattern module built in the third step, which can quickly predict the pattern of the deformable phased array antenna , then based on this pattern data and the label data generated in the first step to restore the j-th beam pointing Calculate the loss function, which uses the mean square error (MSE) [Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp. 436–444, 2015]:

[0130] <21>

[0131] Where K is the total number of samples. Set the number of training rounds to 300 and use the training set generated in the first step to train the multi-angle beam recovery network using the Adam algorithm [DP Kingma and JL Ba, “Adam: Amethod for stochastic optimization,” International Conference on LearningRepresentations, pp. 1–41, 2015.]. Use the method described in this step to train all M networks.

[0132] Step 5. Disconnect the active pattern module from the trained multi-angle beam recovery network. Select the corresponding multi-angle beam recovery network based on the desired beam pointing direction. Input the deflection angle column vector in the test set. Use the multi-angle beam recovery network to quickly predict the amplitude and phase of the antenna unit excitation. This data is transmitted to the feeding system of the phased array antenna, and the phased array antenna is re-fed, thereby realizing the recovery of the multi-beam pointing direction of the curved deformation phased array antenna.

[0133] The effects of the present invention will be further described below in conjunction with simulation experiments.

[0134] 1. Simulation conditions:

[0135] The simulation experiments of the present invention were conducted in a workstation hardware environment equipped with an Intel(R) Xeon(R) E5-2680 v2 processor, 128 GB of memory, and an NVIDIA RTX 4090D graphics processor, and a software environment based on the PyTorch 2.4.0 framework, Python 3.10 programming language, and Windows 10 operating system.

[0136] 2. Simulation content:

[0137] Use as Figure 7The phased array antenna shown in the dataset generated has a baseplate size of 552 mm × 29.9 mm, a dielectric plate thickness of 0.89 mm, and is made of RT / duroid 5880 laminates. The phased array antenna has a 1×24 configuration, a port excitation of 1 W, and an operating frequency of 10 GHz. The desired beam directions are 0°, 4°, 8°, and 12°. Considering 4000 sets of displacement sensor data, the dataset generation process described in the first implementation step of the present invention was used to generate four datasets with a total of 4000 samples. A multi-angle beam recovery network was then constructed according to the process described in the second implementation step of the present invention. Finally, the active pattern module was constructed according to the processes described in the third and fourth implementation steps of the present invention, and the dataset was used to train the multi-angle beam recovery network.

[0138] 3. Simulation results:

[0139] Figure 8 The figure shows the effect of multi-angle beam restoration of a bent and deformed phased array antenna using the method of the present invention. It can be seen that the restored beam pointing angles of 0 degrees, 4 degrees, 8 degrees, and 12 degrees are consistent with the expected beam pointing angle, and the average error of the main beam pointing angle is less than 10 -3 The accuracy of the present invention is proved. Meanwhile, the average calculation time of the beam recovery of the present invention is 2.75 milliseconds, which proves the real-time performance of the present invention.

[0140] The above simulation analysis proves the correctness and effectiveness of the method proposed in the present invention.

[0141] This invention can accurately and efficiently restore beam pointing deviations caused by bending deformation in phased array antennas used in fields such as communications, radar detection, aerospace, and intelligent transportation. It can be directly applied in scenarios where phased array antennas require beam scanning, and has significant economic and practical value in both military and civilian fields. The effectiveness of the invention was verified by studying multi-angle beam restoration for phased array antennas with varying degrees of bending deformation, and its efficiency was demonstrated through beam pointing error and computation time.

[0142] Parts of the present invention that are not described in detail belong to common knowledge among those skilled in the art.

[0143] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Obviously, for professionals in this field, after understanding the content and principles of the present invention, they may make various modifications and changes in form and details without departing from the principles and structure of the present invention. However, these modifications and changes based on the ideas of the present invention are still within the scope of protection of the claims of the present invention.

Claims

1. A method for intelligent multi-angle beam restoration of a deformable phased array antenna, characterized in that: The steps include: (1) Arrange three displacement sensors on the back of each unit floor of the phased array antenna. Indicates the antenna units, where represents the number of antenna elements in the phased array antenna; assuming no deformation occurs The coordinates of the three displacement sensors on each antenna unit are 、 and , after deformation The coordinates of the three displacement sensors on each antenna unit are 、 and ; (2) Constructing the original array No. 1 reference vector and reference vector No. 2 , and the deformed array No. 1 reference vector and reference vector No. 2 And through the vector cross multiplication operation, the surface normal vector of the i-th antenna unit when it is not deformed and deformed is obtained and : ， ， in 、 and Respectively represent the unit vectors along the positive directions of the x, y, and z axes in the three-dimensional Cartesian coordinate system; (3) Calculate the angle between the normal vectors of the upper surface of the current antenna unit before and after deformation, i.e., the deflection angle of the i-th antenna unit, according to the following formula: : ， Stipulate that clockwise rotation is negative and counterclockwise rotation is positive, calculate and collect the deflection angle of each antenna unit in the deformable phased array antenna, and construct the deflection angle column vector : ， (4) Assuming that the phased array antenna does not deform, there are M beam directions when its beam is scanned. Let j = 1, 2, ..., M represent the jth beam direction. Sample the directional pattern of the beam direction in the scanning plane to obtain the directional pattern column vector ; (5) Based on the fully connected neural network architecture, a sub-network with the same number of beam pointings as the number of beam pointings is constructed, i.e., M sub-networks for recovering different beam pointings of the phased array antenna, and they are incorporated into the same parallel framework to obtain a multi-angle beam recovery network; (6) Using the deflection angle column vector and the direction pattern column vector Construct a dataset corresponding to the jth beam pointing; take j = 1, 2, ..., M as the recovery task for each beam pointing in the multi-angle beam recovery network, and generate a dataset for each task; divide each dataset into a training set and a test set according to a preset ratio for training and testing each sub-network; (7) Each sub-network is based on the deflection angle vector As the input of the network, the amplitude and phase of the phased array antenna unit excitation pointing to a specific beam are restored as output; the excitation amplitude and phase vector output by the jth sub-network are It is expressed as follows: ， in and They represent the excitation amplitude and phase used by the i-th antenna unit when the phased array antenna scans to the j-th beam direction; (8) Based on the active pattern AEP, an active pattern module is built to determine the excitation amplitude and phase vector The directional pattern of the synthetic phased array antenna is shown as follows: ， in, represents the directivity pattern of the jth beam pointing of the predicted deformable phased array antenna; The i-th antenna unit of the phased array antenna is deflected according to its Direction diagram after rotation; represents the phase difference; (9) Connect the active pattern module to the output layer of each sub-network used to recover different beam pointing directions, and use the mean square error (MSE) to calculate the loss function: ， Set the maximum number of iterations to T, and T > 300; use the training set and the Adam optimization algorithm to train the multi-angle beam recovery network; (10) Disconnect the active pattern module from the trained multi-angle beam recovery network, select the corresponding sub-network in the multi-angle beam recovery network according to the desired beam pointing, input the deflection angle column vector in the test set, quickly predict the amplitude and phase of the antenna unit excitation through the multi-angle beam recovery network, and transmit this data to the feeding system of the phased array antenna, re-feed the phased array antenna, and realize the recovery of the multi-beam pointing of the curved deformation phased array antenna.

2. The method according to claim 1, wherein: After the deformation in step (1), The coordinates of the three displacement sensors on each antenna unit are 、 and , according to the following formula: ， ， ， in, 、 and After deformation, The three displacement sensors on each antenna unit output the x, y, and z displacements.

3. The method according to claim 1, wherein: The normal vector of the upper surface of the i-th antenna unit when there is no deformation and deformation in step (2) and , according to the following steps: (2.1) When the phased array antenna is not deformed, the original array reference vector No. 1 is constructed according to the coordinates of the three displacement sensors of the i-th antenna unit. and the original array No. 2 reference vector : ， ， (2.2) Through vector and The cross product operation is performed to obtain the surface normal vector of the i-th antenna unit when no deformation occurs. ; (2.3) When the phased array antenna is deformed, the coordinates of the three displacement sensors of the i-th antenna unit after deformation are used to construct the deformation array surface reference vector No. 1 and deformation array No. 2 reference vector : ， ， (2.4) Through vector and The cross product operation is performed to obtain the surface normal vector of the i-th antenna unit after deformation. .

4. The method according to claim 1, wherein: The sub-networks described in step (5) are all four-layer structures, in which the first three layers are hidden layers and the fourth layer is a linear output layer. The number of neurons in the first three layers is an adjustable parameter, and its initial values ​​are set to 32, 64 and 128 respectively. The number of neurons in the fourth layer is fixed to 2N.

5. The method according to claim 1, wherein: The data set in step (6) is generated in the following way: Assuming that the phased array antenna undergoes K different degrees of deformation, the corresponding deflection angle column vector As the input data of the multi-angle beam recovery network, the direction pattern column vector As the label data of the multi-angle beam recovery network; use the input data and label data to form the corresponding dataset of the j-th beam pointing; finally, generate a set of datasets for each beam pointing recovery task in the multi-angle beam recovery network, where the total number of samples in each dataset is K.

6. The method according to claim 1, wherein: In step (6), each data set is divided into a training set and a test set according to a preset ratio, and the preset ratio is 4:

1.

7. The method according to claim 1, wherein: The phase difference in step (8) , which is expressed as follows: ， in represents the frequency, is the speed of light, is the relative dielectric constant of the propagation medium, is the path difference between the ith antenna element and the reference element.

8. The method according to claim 7, wherein: The path difference between the i-th antenna element and the reference element , according to the following steps: (8.1) The data of all displacement sensors No. 1 when the phased array antenna is deformed are expressed as the following three column vectors: ， ， ， in 、 and Respectively represent the displacement data in the x-direction, y-direction, and z-direction collected by all No. 1 displacement sensors; (8.2) The position vectors of the i-th antenna element and the reference element are obtained according to the following formula: : ， in 、 and are the components of the position vector in the x, y, and z directions; (8.3) Based on position vector and beam pointing Get the path difference of the i-th antenna unit : 。 9. The method according to claim 8, characterized in that: The reference unit is the first antenna unit.

Images