Sparse antenna array synthesis method based on neural network intelligent solution

Abstract

The present invention relates to a sparse antenna array synthesis method based on neural network intelligent solution, comprising: performing full-wave simulation of antenna units according to array synthesis requirements, obtaining unit active directional patterns, and constructing a unit active directional pattern database; constructing an initial sample set according to the unit active directional patterns and an array numerical simulation model; randomly dividing the initial sample set into a training set and a validation set, performing offline training using the training set, and performing array synthesis solution on the validation set data; judging whether the solution result meets the array layout requirements; if not, using a convex optimization algorithm to assist in calculation, updating the data set, and retraining; otherwise, outputting a neural network model; and using the model, performing sparse antenna array synthesis according to unit spacing constraints and active directional patterns, and outputting an array topology. The present invention can realize comprehensive solution of antenna arrays, and has the advantages of high speed, high efficiency, and being more consistent with the actual array working environment.

Description

Technical Field

[0001] The present invention relates to the technical field of phased array antennas, and in particular to a sparse antenna array synthesis method based on neural network intelligent solution. Background Art

[0002] Phased array antennas are widely used in various radar communications, remote sensing, and space-based astronomy systems. As key components for converting electromagnetic waves into guided waves in electronic systems, their performance directly impacts system effectiveness. Array antenna synthesis technology, based on the desired radiation pattern, reversely synthesizes the number of antenna elements, their arrangement, element spacing, excitation amplitude, and phase, and is the starting point and key to antenna design.

[0003] The cost and power consumption of active phased arrays primarily stem from the transceiver components. Especially for high-gain antennas, reducing the number of active channels significantly reduces system complexity, cost, and power consumption. This is where sparse array technology comes in. Compared to traditional uniform arrays, antenna element distribution is no longer limited to rectangular or triangular grids, and element spacing is no longer restricted to half a wavelength. This allows high-gain, high-resolution, narrow-beam radiation to be achieved with less hardware, providing greater flexibility in system design and reducing hardware costs.

[0004] However, sparse or sparse arrays can lead to the generation of scanning grating lobes, which need to be addressed through unit position arrangement and excitation amplitude and phase optimization. The key to design lies in an effective array synthesis algorithm. Array synthesis is often a multi-objective, multi-parameter nonlinear optimization problem. Traditional exhaustive methods, analytical algorithms, gradient iterative optimization algorithms, randomized optimization algorithms, convex optimization algorithms, and their hybrid solutions have demonstrated their respective advantages for different synthesis problems. However, when considering the optimization of sparse arrays, the integration of multi-dimensional variables such as unit mutual coupling and array position makes it difficult to guarantee the convergence and computational efficiency of the algorithm. Summary of the Invention

[0005] In order to solve the technical problems existing in the above-mentioned prior art, the purpose of the present invention is to provide a sparse antenna array synthesis method based on neural network intelligent solution, so as to solve the problem that traditional array synthesis methods converge slowly and are prone to falling into local optimal solutions.

[0006] To achieve the above-mentioned object of the invention, the present invention provides a sparse antenna array synthesis method based on neural network intelligent solution, comprising the following steps:

[0007] Step S1: Perform full-wave simulation of antenna units according to array comprehensive requirements, obtain unit active patterns, and construct a unit active pattern database; the array comprehensive requirements include array layout requirements and expected radiation performance, the array layout requirements include sparsity rate, array dimension, and unit spacing constraints, and the expected radiation performance includes gain, scanning angle, beamwidth, peak sidelobe, and axial ratio;

[0008] Step S2: obtaining the entire array pattern according to the unit active pattern and the array numerical simulation model, and constructing an initial sample set according to the entire array pattern and array parameters;

[0009] Step S3: randomly dividing the initial sample set into a training set and a validation set, and using the training set to perform offline neural network model training;

[0010] Step S4: Based on the trained neural network model, perform array comprehensive solution on the verification set data;

[0011] Step S5: determine whether the solution result of the validation set data meets the array requirements. If not, use a convex optimization algorithm to assist in the calculation, and use the data obtained from the auxiliary calculation to update the data set and retrain the neural network model; otherwise, output the neural network model.

[0012] Step S6: Perform sparse antenna array synthesis based on the input array synthesis requirements and the unit active pattern of the antenna array through the output neural network model, and output the array topology.

[0013] According to a technical solution of the present invention, in step S1, it specifically includes:

[0014] Step S11: According to the unit spacing constraint and antenna unit form, change the array element spacing, perform full-wave simulation, and obtain the unit active radiation pattern where θ and are the elevation and azimuth angles of the antenna coordinate system;

[0015] Step S12: Active directional pattern of the unit according to the operating frequency f Sampling is performed to obtain the unit active pattern corresponding to different operating frequencies Step S13: construct the unit active pattern data set according to the corresponding relationship between the array topology and the unit active pattern, wherein the samples in the unit active pattern data set are represented as

[0016] According to a technical solution of the present invention, the antenna unit form is determined according to the array comprehensive requirements:

[0017] The antenna array aperture is determined according to the gain, the antenna element scale is obtained according to the sparsity rate, and the antenna unit type is determined according to the scanning angle and gain. When the antenna scanning gain loss is small, a wide-beam antenna unit is used for full-wave simulation. When the antenna gain requirement is high, a high-gain antenna unit is used for full-wave simulation.

[0018] According to a technical solution of the present invention, in step S11, the full-wave simulation includes: according to the unit spacing constraint Γ=[d xmin ,d xmax ,d ymin ,d ymax ], within the cell spacing constraint, set the array topology set L = {[x p ,y q ]|p=1,2,…,N x ,q=1,2,…N y}, select the unit and a group of adjacent units for full-wave simulation, change the unit spacing, and obtain the unit active pattern The unit spacing range is expressed as: d x =|x p -x p-1 |∈[d xmin ,d xmax ], d y =|y q -y q-1 |∈[d ymin ,d ymax ], where p and q represent the unit numbers along the x-axis and y-axis respectively, and d x represents the unit spacing along the x-axis, d y represents the unit spacing along the y-axis, d xmin represents the minimum unit spacing along the x-axis, d ymin Indicates the minimum cell spacing along the y-axis.

[0019] According to a technical solution of the present invention, in step S2, it specifically includes:

[0020] Step S21: Based on the array comprehensive requirements and the active directional pattern of the unit, the directional pattern is calculated by the array numerical simulation model to obtain the directional pattern of the entire array, which is expressed as:

[0021]

[0022] in, represents the entire array pattern of an antenna array with an array topology of l and an operating frequency of f; I p,qrepresents the excitation amplitude of the antenna unit; k represents the wave number corresponding to the operating frequency f, k = 2πf / c, c is the speed of light; w = {w p,q |p=1,2,…,N x ,q=1,2,…N y} represents the phase of the antenna array. The phase w of the antenna array is generated according to the array topology and can be expressed as Where k0 represents the wave number k0 = 2πf0 / c when working at frequency f = f0, c is the speed of light, and the scanning frequency is hour, N x 、N y Indicates the number of array elements distributed along the x-axis and y-axis;

[0023] Step S22: construct an initial sample set according to the unit active pattern and the whole array pattern. Each sample in the initial sample set includes an input vector and a sample label. The initial sample set is represented by S={(X m ,Y m )|m=1,2,…M};

[0024] Among them, X m is the input vector, including the unit active pattern, the whole array pattern, the beam width, and the side lobe level, expressed as in is the unit active pattern of the antenna array, and its operating frequency is f, represents the entire array pattern of the antenna array, Indicates scanning to The beam width at Indicates scanning to Sidelobe level at ;

[0025] Y m is the sample label, including the array topology of the antenna array represented by Y m =L.

[0026] According to a technical solution of the present invention, when the desired radiation performance also includes cross-polarization, during full-wave simulation, the cross-polarization unit active pattern component is extracted, and the input vector also includes the entire array cross-polarization level, and is expressed as

[0027]

[0028] Among them, E f,co Indicates the cross-polarization level of the entire array.

[0029] According to a technical solution of the present invention, in step S3, the following steps are specifically included:

[0030] Step S31: dividing the initial sample set into a training set and a validation set in a ratio of 8:2 by a random division method, and normalizing them;

[0031] Step S32: Construct a neural network model, and determine the number of network layers and the number of neurons in each layer based on the input vector dimension and the output vector dimension. The array element size is independent of the input vector dimension but is related to the output vector dimension. The neural network model adopts a double-hidden-layer feedforward network model, and the two hidden layers have the same number of neurons, which is determined by the array dimension.

[0032] Step S33: Set the number of training times, training target, learning rate, transfer function, training function and training error;

[0033] Step S34: Substitute the verification set into the neural network model for verification to determine whether the training error meets the requirements;

[0034] Step S35: When the training error does not meet the requirements, return to step S32 to adjust the network parameters and retrain;

[0035] Step S36: When the output meets the training error requirement, the network model and weight coefficients are saved to obtain the offline trained neural network model.

[0036] According to a technical solution of the present invention, in step S5, the following steps are specifically included:

[0037] Establish a convex optimization model, expressed as

[0038]

[0039] Among them, E ref represents the expected pattern, and ε represents the fitting error;

[0040] Through the convex optimization model, the set of optimized array topologies L'={[x p ',y q ']|p=1,2,…,N x ,q=1,2,…N y};

[0041] The optimized data is added to the training set to form a new training set S'={(X' m ,Y' m )|m=1,2,…M+M a}, retrain the network using the new training set, and at the same time, the set of optimization results can be directly output as the comprehensive results.

[0042] According to one aspect of the present invention, an electronic device is provided, comprising: one or more processors, one or more memories, and one or more computer programs; wherein the processor is connected to the memory, and the one or more computer programs are stored in the memory; when the electronic device is running, the processor executes the one or more computer programs stored in the memory, so that the electronic device performs the above-mentioned sparse antenna array synthesis method based on neural network intelligent solution.

[0043] According to one aspect of the present invention, a computer-readable storage medium is provided for storing computer instructions. When the computer instructions are executed by a processor, the above-mentioned sparse antenna array synthesis method based on neural network intelligent solution is implemented.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention provides a sparse antenna array synthesis method based on neural network intelligent solution, which establishes an artificial neural network with array pattern and pattern performance index as input and unit pattern and array topology as output. This method avoids the tedious manual synthesis and feedback process, improves the efficiency of multi-objective optimization, and has more general practicality. Traditional array synthesis algorithms are often based on the concept of element optimization, optimizing the unit pattern and array topology to the optimal solution respectively, and then directly combining them. The combined result may not be the optimal solution. The present invention, however, realizes intelligent solution of multi-objective optimization by introducing artificial neural network. Based on the concept of system optimization, it introduces the complementary design of unit pattern and array topology to achieve the optimization of system effect.

[0046] The present invention adopts a multi-layer neural network training method for array synthesis, which effectively avoids the problems of slow convergence and easy falling into local optimal solutions in traditional array synthesis methods. By combining array forward analysis with network training, the multi-dimensional parameter constraint relationship of array synthesis can be intelligently obtained to obtain a sparse topology that meets the design constraints.

[0047] The present invention introduces the relationship between array topology changes and the mutual coupling effect of array elements into array synthesis. Through neural network training prediction, the active directional patterns of units under different topological rules are obtained as the element factors of array synthesis, which is more in line with the actual array working environment. In addition, network training based on full-wave simulation data can intelligently fit the nonlinear change relationship, which has advantages over conventional algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be derived from these drawings without inventive effort.

[0049] Figure 1 Schematically showing a flow chart of a sparse antenna array synthesis method based on neural network intelligent solution provided in one embodiment of the present invention;

[0050] Figure 2 Schematically showing a principle diagram of a sparse antenna array synthesis method based on neural network intelligent solution provided in one embodiment of the present invention;

[0051] Figure 3 Schematically illustrates a flow chart of offline neural network training according to one embodiment of the present invention;

[0052] Figure 4 Schematically showing a schematic diagram of the coupling region of the unit modeling in a planar array according to one embodiment of the present invention;

[0053] Figure 5 Schematically showing a low sidelobe array pattern obtained by using a sparse array arrangement according to one embodiment of the present invention;

[0054] Figure 6 The figure schematically shows the topological positions of array elements corresponding to the sparse array low sidelobe pattern according to one embodiment of the present invention. DETAILED DESCRIPTION

[0055] The description of the embodiments in this specification should be combined with the corresponding drawings, which should be considered a complete part of this specification. In the drawings, the shapes and thicknesses of the embodiments may be exaggerated and indicated for simplicity or convenience. Furthermore, the various structural components in the drawings will be described separately. It is worth noting that components not shown in the drawings or not described in words are known to those of ordinary skill in the art.

[0056] The description of the embodiments herein and any references to directions and orientations are for ease of description only and are not to be construed as limiting the scope of the present invention. The following description of the preferred embodiments may involve combinations of features, which may exist independently or in combination. The present invention is not specifically limited to the preferred embodiments. The scope of the present invention is defined by the claims.

[0057] like Figures 1 to 6As shown, a sparse antenna array synthesis method based on neural network intelligent solution of the present invention includes the following steps:

[0058] Step S1: Perform full-wave simulation of antenna elements according to array synthesis requirements, obtain element active patterns, and construct an element active pattern database; the array synthesis requirements include array layout requirements and expected radiation performance, the array layout requirements include sparsity rate, array dimension, and element spacing constraints, and the expected radiation performance includes gain, scan angle, beamwidth, peak sidelobe, axial ratio, and further, cross-polarization level;

[0059] In step S1, it specifically includes:

[0060] Step S11: According to the unit spacing constraint and antenna unit form, change the array element spacing, perform full-wave simulation, and obtain the unit active radiation pattern where θ and are the elevation and azimuth angles of the antenna coordinate system;

[0061] The antenna unit form is determined according to the comprehensive array requirements: the array aperture of the antenna is determined according to the gain, the array element scale of the antenna is obtained according to the sparsity rate, and the unit type of the antenna is determined according to the scanning angle and gain; when the antenna scanning gain loss is small, a wide-beam antenna unit is used for full-wave simulation; when the antenna gain requirement is high, a high-gain antenna unit is used for full-wave simulation.

[0062] Model the wide-angle scanning unit and high-gain antenna unit separately. Select a specific radiation unit model as a representative and optimize the unit model under the radiation boundary conditions. Array modeling is performed based on the unit model. Since antenna mutual coupling is related to the array element spacing and configuration, the adjacent units in the array contribute the most to unit coupling. Therefore, after selecting the unit form, combine the minimum array sparsity requirement to perform full-wave simulation. The steps of full-wave simulation include:

[0063] According to the cell spacing constraint Γ=[d xmin ,d xmax ,d ymin ,d ymax ], within the cell spacing constraint, set the array topology set L = {[x p ,y q ]|p=1,2,…,N x ,q=1,2,…N y}, select the unit and a group of adjacent units for full-wave simulation, change the unit spacing, and obtain the unit active pattern The cell spacing range is expressed as:

[0064] d x =|x p -x p-1|∈[d xmin ,d xmax ], d y =|y q -y q-1 |∈[d ymin ,d ymax ],

[0065] Among them, p and q represent the numbers of the antenna units along the x-axis and y-axis respectively, and d x represents the unit spacing along the x-axis, d y represents the unit spacing along the y-axis, d xmin represents the minimum unit spacing along the x-axis, d ymin Indicates the minimum cell spacing along the y-axis.

[0066] When changing the unit spacing, use step x =0.01λ, step y =0.01λChange the unit spacing, expressed as changing the unit spacing d x =[0.45λ:0.01λ:λ], d y =[0.45λ:0.01λ:λ].

[0067] In order to meet the needs of plane array synthesis, different azimuth planes The active pattern is sampled In order to reduce the amount of data and speed up training, the azimuth sampling data set can be selected as

[0068] Step S12: Active directional pattern of the unit according to the operating frequency f Sampling is performed to obtain the unit active pattern corresponding to different operating frequencies

[0069] Step S13: construct a unit active pattern data set according to the correspondence between the array topology and the unit active pattern. The samples in the unit active pattern data set are represented as

[0070] l represents the active directional pattern of the unit The corresponding array topology is the geometric arrangement of antenna elements in the antenna array. The set of array topologies L contains several different array topologies l, and the element spacing of each array topology l is different.

[0071] Step S2: obtaining the entire array pattern according to the unit active pattern and the array numerical simulation model, and constructing an initial sample set according to the entire array pattern and array parameters;

[0072] In step S2, it specifically includes:

[0073] Step S21: Based on the array comprehensive requirements and the unit active pattern, the pattern calculation is performed through the array numerical simulation model to obtain the whole array pattern, which is expressed as:

[0074]

[0075] in, represents the entire array pattern of an antenna array with an array topology of l and an operating frequency of f; I p,q represents the excitation amplitude of the antenna unit, which can be uniformly weighted to generate the initial distribution; k represents the wave number corresponding to the operating frequency f, k = 2πf / c, c is the speed of light; w represents the phase of the antenna array, which is generated according to the array topology and can be expressed as w = {w p,q |p=1,2,…,N x ,q=1,2,…N y} represents the phase of the antenna array. The phase w is generated according to the array topology and can be expressed as

[0076]

[0077] Where k0 represents the wave number k0 = 2πf0 / c when working at frequency f = f0, c is the speed of light, and the scanning frequency is hour, N x 、N y Indicates the number of array elements distributed along the x-axis and y-axis.

[0078] By setting the excitation amplitude I and phase w of the antenna unit, the whole array pattern with different amplitude weighting and different scanning directions is generated to obtain the corresponding beam width Sidelobe level As the data of the training dataset.

[0079] Considering the amplitude and phase quantization effects of the analog array, the weights are set according to the number of quantization bits and accuracy of the chip components. Generally speaking, for a 6-bit phase shifter, the excitation phase w = [0:5.325°:360°] can be set.

[0080] Step S22: construct an initial sample set according to the unit active pattern and the expected radiation performance. Each sample in the initial sample set includes an input vector and a sample label. The initial sample set is represented by S={(X m ,Y m )||m=1,2,…M};

[0081] Among them, X m is the input vector, including the unit active pattern, the whole array pattern, the beam width, and the side lobe level, expressed as in is the unit active pattern of the antenna array, and its operating frequency is f, represents the entire array pattern of the antenna array, Indicates scanning to The beam width at Indicates scanning to Sidelobe level at ;

[0082] Y m is a vector of sample labels, including a collection of array topologies, denoted as Y m =L.

[0083] When the desired radiation performance also includes cross-polarization, in full-wave simulation, the cross-polarization unit active pattern component is extracted, and the input vector also includes the cross-polarization level of the entire array, which is expressed as

[0084]

[0085] Among them, E f,co Indicates the cross-polarization level of the entire array.

[0086] Step S3: randomly divide the initial sample set into a training set and a validation set, and use the training set to perform offline neural network model training;

[0087] In step S3, it specifically includes:

[0088] Step S31: Divide the initial sample set into a training set and a validation set by a random division method in a ratio of 8:2, and perform normalization;

[0089] Step S32: Construct a neural network model, and determine the number of network layers and the number of neurons in each layer based on the input vector dimension and the output vector dimension. The array element size is independent of the input vector dimension but is related to the output vector dimension. The neural network model adopts a double-hidden-layer feedforward network model, and the two hidden layers have the same number of neurons, which is determined by the array dimension.

[0090] Step S33: Set the number of training times, training target, learning rate, transfer function, training function and training error;

[0091] Optionally, you can set the transfer function to the transfer and purelin functions, and the training function to the trainlm function.

[0092] Step S34: Substitute the validation set into the neural network model for verification to determine whether the training error meets the requirements;

[0093] Step S35: When the training error does not meet the requirements, return to step S32 to adjust the network parameters and retrain;

[0094] Step S36: When the output meets the training error requirement, the network model and weight coefficients are saved to obtain the offline trained neural network model.

[0095] When training a feedforward neural network model, the model convergence is accelerated by reasonably setting the network parameters; at the same time, the corresponding output vector dimension is modified according to the array size, and the number of hidden layer neurons is reasonably set to improve the convergence speed and prediction accuracy.

[0096] Step S4: Based on the trained neural network model, perform array comprehensive solution on the verification set data;

[0097] Model verification is performed based on the trained neural network model. The verification set data is substituted into the array comprehensive solution to obtain the normalized array topology position and amplitude and phase weights.

[0098] Step S5: determine whether the solution of the validation set data meets the layout requirements. If not, use a convex optimization algorithm to assist in the calculation, and use the data obtained from the auxiliary calculation to update the data set and retrain the neural network model; otherwise, output the neural network model.

[0099] When verifying the solution results of the validation set data, we verify whether the solution results meet the sparsity ratio requirements, array dimension requirements, and cell spacing constraint requirements. If they do, we output the neural network model. If not, we use convex optimization to assist in the calculation of the solution results.

[0100] In step S5, the convex optimization algorithm is used to assist in the calculation, specifically including:

[0101] Establish a convex optimization model, expressed as

[0102]

[0103] Among them, min means taking the minimum, st means the constraint condition, E ref represents the expected pattern, and ε represents the fitting error.

[0104] Using the convex optimization algorithm, the array topology set L'={[x p ',y q ']|p=1,2,…,N x ,q=1,2,…N y};

[0105] The optimized data is added to the training set to form a new training set S'={(X' m ,Y' m)|m=1,2,…M+M a}, retrain the network using the new training set, and at the same time, the set of optimization results can be directly output as the comprehensive results.

[0106] Determine whether the array requirements are met. If not, use a convex optimization algorithm to assist in the calculation, update the calculated data into the dataset, and retrain the network model. Furthermore, other types of optimization algorithms can be used to generate new training datasets and update the network model online. The normalized output results are mapped to variables, and the optimized array topology position and channel amplitude and phase results are output.

[0107] Step S6: Using the output neural network model, according to the input unit spacing constraint and active pattern, perform sparse antenna array synthesis and output array topology, excitation amplitude and excitation phase.

[0108] Compared to traditional synthesis algorithms, this method intelligently captures the nonlinear relationship between array multi-dimensional parameters and radiation patterns through forward array analysis and simulation and neural network model training. It also intelligently solves for multiple radiation expectations, facilitating reverse synthesis design under multivariable coupling and multi-objective constraints. Using offline training and online updates, it rapidly generates array topology excitations that meet performance requirements.

[0109] According to one aspect of the present invention, an electronic device includes: one or more processors, one or more memories, and one or more computer programs; wherein the processor is connected to the memory, and the above-mentioned one or more computer programs are stored in the memory. When the electronic device is running, the processor executes the one or more computer programs stored in the memory, so that the electronic device performs the above-mentioned sparse antenna array synthesis method based on neural network intelligent solution.

[0110] According to one aspect of the present invention, a computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the above-mentioned sparse antenna array synthesis method based on neural network intelligent solution is implemented.

[0111] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or terminal device that includes a series of elements includes not only those elements but also other elements not explicitly listed, or also includes elements inherent to such process, method, article, or terminal device. In the absence of further limitations, an element defined by the phrase "comprises a..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the element.

[0112] Finally, it should be noted that the above is a preferred embodiment of the present invention. It should be noted that although the preferred embodiment of the present invention has been described, it is clear that those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles of the present invention. Such improvements and modifications should also be considered as within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the embodiments of the present invention.

Claims

1. A sparse antenna array synthesis method based on neural network intelligent solution, characterized in that: The following steps are involved: Step S1: Perform full-wave simulation of antenna units according to array comprehensive requirements, obtain unit active patterns, and construct a unit active pattern database; the array comprehensive requirements include array layout requirements and expected radiation performance, the array layout requirements include sparsity rate, array dimension, and unit spacing constraints, and the expected radiation performance includes gain, scanning angle, beamwidth, peak sidelobe, and axial ratio; Step S2: obtaining the entire array pattern according to the unit active pattern and the array numerical simulation model, and constructing an initial sample set according to the entire array pattern and array parameters; Step S3: randomly dividing the initial sample set into a training set and a validation set, and using the training set to perform offline neural network model training; Step S4: Based on the trained neural network model, perform array comprehensive solution on the verification set data; Step S5: determine whether the solution result of the validation set data meets the array requirements. If not, use a convex optimization algorithm to assist in the calculation, and use the data obtained from the auxiliary calculation to update the data set and retrain the neural network model; otherwise, output the neural network model. Step S6: Perform sparse antenna array synthesis based on the output neural network model and the input array synthesis requirements and the unit active pattern of the antenna array, and output the array topology; In the step S2, it specifically includes: Step S21: Based on the array comprehensive requirements and the active directional pattern of the unit, the directional pattern is calculated by the array numerical simulation model to obtain the directional pattern of the entire array, which is expressed as: in, represents the entire array pattern of an antenna array with an array topology of l and an operating frequency of f; I p,q represents the excitation amplitude of the antenna unit; k represents the wave number corresponding to the operating frequency f, k = 2πf / c, c is the speed of light; w = {w p,q |p=1,2,…,N x ,q=1,2,…N y } represents the phase of the antenna array. The phase w of the antenna array is generated according to the array topology and can be expressed as Where k0 represents the wave number k0 = 2πf0 / c when working at frequency f = f0, c is the speed of light, and the scanning frequency is hour, N x 、N y Indicates the number of array elements distributed along the x-axis and y-axis; Step S22: construct an initial sample set according to the unit active pattern and the whole array pattern. Each sample in the initial sample set includes an input vector and a sample label. The initial sample set is represented by S={(X m ,Y m )|m=1,2,…M}; Among them, X m is the input vector, including the unit active pattern, the whole array pattern, the beam width, and the side lobe level, expressed as in is the unit active pattern of the antenna array, and its operating frequency is f, represents the entire array pattern of the antenna array, Indicates scanning to The beam width at Indicates scanning to Sidelobe level at ; Y m is the sample label, including the set of array topologies of the antenna array, denoted as Y m =L.

2. The sparse antenna array synthesis method based on neural network intelligent solution according to claim 1 is characterized in that: In the step S1, it specifically includes: Step S11: According to the unit spacing constraint and antenna unit form, change the array element spacing, perform full-wave simulation, and obtain the unit active radiation pattern where θ and are the elevation and azimuth angles of the antenna coordinate system; Step S12: Active directional pattern of the unit according to the operating frequency f Sampling is performed to obtain the unit active pattern corresponding to different operating frequencies Step S13: construct the unit active pattern data set according to the corresponding relationship between the array topology and the unit active pattern, wherein the samples in the unit active pattern data set are represented as 3. The sparse antenna array synthesis method based on neural network intelligent solution according to claim 2 is characterized in that: The antenna unit form is determined according to the comprehensive requirements of the array: The antenna array aperture is determined according to the gain, the antenna element scale is obtained according to the sparsity rate, and the antenna unit type is determined according to the scanning angle and gain. When the antenna scanning gain loss is small, a wide-beam antenna unit is used for full-wave simulation. When the antenna gain requirement is high, a high-gain antenna unit is used for full-wave simulation.

4. The sparse antenna array synthesis method based on neural network intelligent solution according to claim 3 is characterized in that: In step S11, the full-wave simulation includes: According to the cell spacing constraint Γ=[d xmin ,d xmax ,d ymin ,d ymax ], within the cell spacing constraint, set the array topology set L = {[x p ,y q ]|p=1,2,…,N x ,q=1,2,…N y }, select the antenna unit and a group of adjacent antenna units for full-wave simulation, change the unit spacing, and obtain the unit active radiation pattern The unit spacing range is expressed as: d x =|x p -x p-1 |∈[d xmin ,d xmax ], d y =|y q -y q-1 |∈[d ymin ,d ymax ], where p and q represent the numbers of the antenna units along the x-axis and y-axis respectively, and d x represents the unit spacing along the x-axis, d y represents the unit spacing along the y-axis, d xmin represents the minimum unit spacing along the x-axis, d ymin Indicates the minimum cell spacing along the y-axis.

5. The sparse antenna array synthesis method based on neural network intelligent solution according to claim 4 is characterized in that: When the desired radiation performance also includes cross-polarization, in full-wave simulation, the cross-polarization unit active pattern component is extracted, and the input vector also includes the entire array cross-polarization level, which is expressed as Among them, E f，co Indicates the cross-polarization level of the entire array.

6. The sparse antenna array synthesis method based on neural network intelligent solution according to claim 1 is characterized in that: In the step S3, it specifically includes: Step S31: dividing the initial sample set into a training set and a validation set in a ratio of 8:2 by a random division method, and normalizing them; Step S32: Construct a neural network model, and determine the number of network layers and the number of neurons in each layer based on the input vector dimension and the output vector dimension. The array element size is independent of the input vector dimension but is related to the output vector dimension. The neural network model adopts a double-hidden-layer feedforward network model, and the two hidden layers have the same number of neurons, which is determined by the array dimension. Step S33: Set the number of training times, training target, learning rate, transfer function, training function and training error; Step S34: Substitute the verification set into the neural network model for verification to determine whether the training error meets the requirements; Step S35: When the training error does not meet the requirements, return to step S32 to adjust the network parameters and retrain; Step S36: When the output meets the training error requirement, the network model and weight coefficients are saved to obtain the offline trained neural network model.

7. The sparse antenna array synthesis method based on neural network intelligent solution according to claim 1 is characterized in that: In step S5, the convex optimization algorithm-assisted calculation specifically includes: Establish a convex optimization model, expressed as min||w||0 Among them, E ref represents the expected pattern, and ε represents the fitting error; Through the convex optimization model, the set of optimized array topologies L'={[x p ',y q ']|p=1,2,…,N x ,q=1,2,…N y }; The optimized data is added to the training set to form a new training set S'={(X' m ,Y' m )|m=1,2,…M+M a }, retrain the network using the new training set, and at the same time, the set of optimization results can be directly output as the comprehensive results.

8. An electronic device, characterized in that: include: One or more processors, one or more memories, and one or more computer programs; wherein the processor is connected to the memory, and the one or more computer programs are stored in the memory. When the electronic device is running, the processor executes the one or more computer programs stored in the memory to enable the electronic device to perform the sparse antenna array synthesis method based on neural network intelligent solution as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that Used to store computer instructions, which, when executed by a processor, implement the sparse antenna array synthesis method based on neural network intelligent solution as described in any one of claims 1 to 7.

Images