Far-field prediction method based on neural network near-field phase reconstruction

Abstract

The invention discloses a far-field prediction method based on neural network near-field phase reconstruction, and the method comprises the following steps: S1, carrying out the near-field simulation of a to-be-tested antenna, obtaining the near-field amplitude and phase data of the to-be-tested antenna, constructing a data set, and dividing the data set into a training set and a verification set; s2, constructing a neural network for predicting a near-field phase according to the near-field amplitude, and training and verifying the neural network through the training set to obtain a trained neural network; s3, acquiring near-field amplitude information of the to-be-tested antenna which is not in the data set, inputting the near-field amplitude information into the trained neural network to obtain a predicted near-field phase, and calculating an equivalent dipole source of the to-be-tested antenna by using a least square method; and S4, calculating the far field of the antenna to be tested according to an equivalent dipole theory. According to the method, the neural network is adopted for near-field phase recovery, then the equivalent dipole source is constructed, far-field prediction of the to-be-tested antenna is finally realized, and the method has relatively high accuracy and timeliness.

Description

Technical Field

[0001] The present invention relates to the field of electromagnetic waves, and in particular to a far-field prediction method based on neural network near-field phase reconstruction. Background Art

[0002] Near-field to far-field conversion is an important part of antenna testing. Its core is to reconstruct the far-field information of the antenna under test (AUT) through near-field measurement data. As electronic systems become larger and more complex, the full-wave method can no longer meet the needs of electromagnetic compatibility modeling due to the huge consumption of computing resources. In the past few decades, the theory of constructing equivalent radiation sources has developed rapidly. Among them, the equivalent dipole theory is widely used in the construction of equivalent sources. Generally, the establishment of an equivalent source requires both the amplitude and phase information of the near field. However, in actual testing, it is difficult or even impossible to accurately obtain the phase information of the near field. On this basis, near-field phase recovery and the method of reconstructing the equivalent source based on the phase-free near field are proposed. There are mainly two ideas: one is to perform redundant measurements, such as measuring the field amplitude information of multiple near-field scanning surfaces; the other is to use optimization algorithms to minimize errors by iterating equivalent dipoles or near-field phases. However, both methods face their own shortcomings. Redundant measurements require more hardware resources and are difficult to ensure under specific measurement conditions; optimization algorithms have high requirements on computer computing power and are difficult to ensure timeliness. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a far-field prediction method based on neural network near-field phase reconstruction, which adopts a neural network to perform near-field phase recovery, then constructs an equivalent dipole source, and finally realizes the far-field prediction of the antenna to be tested, with high accuracy and timeliness.

[0004] The object of the present invention is achieved through the following technical solution: a far field prediction method based on neural network near field phase reconstruction, comprising the following steps:

[0005] S1. Perform near-field simulation on the antenna to be tested, obtain near-field amplitude and phase data of the antenna to be tested, and construct a data set, dividing the data set into a training set and a verification set;

[0006] S2. Construct a neural network for predicting the near-field phase based on the near-field amplitude, train the neural network through the training set, fit the relationship between the near-field amplitude and the near-field phase, verify the trained neural network using the verification set, and obtain the trained neural network after the verification passes;

[0007] S3. For the antenna to be tested that is not in the data set, only the near-field amplitude information is obtained through simulation, and the predicted near-field phase is obtained by inputting it into the trained neural network, thereby obtaining the complete near-field amplitude and phase information, and the equivalent dipole source of the antenna to be tested is calculated using the least squares method;

[0008] S4. Calculate the far field of the antenna under test based on the equivalent dipole theory.

[0009] Preferably, the step S1 comprises:

[0010] S101. Assume that the antenna to be tested is a conical horn antenna, the aperture radius is 1.2λ, λ is the wavelength of the electromagnetic wave, and the near-field sampling surface is a spherical surface with a radius of R and an elevation angle of θ of 1 to 179° and an azimuth angle of θ of 1 to 179°. Set t×t sampling points equally spaced between 0 and 342°, and the number of sampling points is recorded as m=t×t;

[0011] When performing near-field sampling, near-field amplitude data and near-field phase data are collected. The near-field amplitude data includes the radial component |E r |, polar angle component |E θ | and the azimuthal component Three components, the near-field phase data φ also includes the radial component φ r , polar angle component φ θ and the azimuth component Three components, each sampling point has a set of amplitude and phase data;

[0012] S102. Sample equidistantly between 1.5 and 2.4 GHz to obtain p frequencies, and construct an antenna CAD model at each frequency. Under the antenna CAD model at each frequency, repeat step S101 to obtain near-field amplitude data and phase data at each frequency, and add them to the same data set to complete the construction of the data set;

[0013] There are p samples in the dataset, each of which corresponds to an antenna CAD model at a frequency. In each sample, the feature is the near-field amplitude of the horn antenna, with a dimension of m×3, and the label is the near-field phase, also with a dimension of m×3.

[0014] S103. Divide the data set into a training set and a validation set.

[0015] Preferably, in step S2, the neural network CNN convolutional neural network for predicting the near-field phase by the near-field amplitude will train the neural network in turn using each sample during the training process:

[0016] When using any sample to train the neural network, an adaptive momentum optimizer is used. The near-field amplitude of t×t×3 is regarded as a three-channel image of pixel t×t and input into the CNN convolutional neural network. The CNN neural network outputs the predicted phase of t×t×3. The loss function is calculated using the predicted phase of t×t×3 and the near-field phase of t×t×3 in the label. Then, according to the loss function, the gradient is calculated by backpropagation and the parameters of the CNN neural network are updated according to the gradient descent method.

[0017] After reaching the training round, the trained neural network is verified using the verification set, and the trained neural network is obtained after the verification passes.

[0018] Preferably, step S3 comprises:

[0019] S301. Assume that the dipole is placed on a spherical surface with radius R1 at 10° to 170° of θ. t1×t1 points are set equidistantly between 0 and 340°; the number of dipoles is recorded as n=t1×t1, and each dipole P has a polar angle component P θ and the azimuth component Two components;

[0020] S302. For the horn antenna not in the data set, simulate and obtain the near-field amplitude, and calculate the near-field amplitude at r,θ, In three directions, the vectors of the near-field amplitudes of m sampling points are constructed according to the simulation results, denoted as |E r | m×1 、|E θ | m×1 , Therefore, the near-field amplitude is recorded as

[0021] in, Indicates that in r,θ, The vectors of the near-field amplitudes of m sampling points in three directions; |E ri |、|E θi |、 Indicates the amplitude of the i-th near-field sampling point at r,θ, Components in three directions, i = 1, 2…m;

[0022] Will |E| 3m×1 Input the trained neural network to get the predicted near-field phase in The phase of m near-field sampling points is expressed as r,θ, The vector obtained by arranging the three directions; Respectively represent the phase of the i-th near-field sampling point at r,θ, Components in three directions;

[0023] So far, we have obtained the complete information of the near field, which can be recorded as Among them, E r,m×1 、E θ,m×1 , Respectively represent m sampling points at r,θ, Complete near-field information in three directions:

[0024]

[0025] E ri 、E θi , Indicates that the complete information of the i-th near-field sampling point is in r,θ, Components in three directions, i = 1, 2…m;

[0026] S303. Arrange n dipoles according to θ, Arrange in the direction to get a column vector

[0027] in, represents the polar and azimuthal components of n dipoles, P θj , represents the polar component and azimuthal component of the jth dipole, j = 1, 2…n;

[0028] S304. Equivalent dipole theory uses a dipole array to replace the antenna under test to characterize its radiation characteristics;

[0029] Dipole array P 2n×1 The electric field E generated in free space 3m×1 The electric field generated by a single dipole in free space is obtained by coordinate transformation, and its relationship matrix is Make T 3m×2n ×P 2n×1 =E 3m×1 ,in Represents the relationship between the electric field component and the dipole component;

[0030] The dielectric constant in free space is ε0, the magnetic permeability is μ0, and the wave impedance is Wave number The coordinates of a near-field sampling point in the spherical coordinate system are The coordinates in the rectangular coordinate system are (x Ei ,y Ei ,z Ei ); The coordinates of a dipole placement point in the spherical coordinate system are The coordinates in the rectangular coordinate system are (x Pj ,y Pj ,z Pj), the distance between them is recorded as r; when the positions of the near-field sampling point and the dipole are determined, T 3m×2n Each element in

[0031]

[0032]

[0033]

[0034] in,

[0035]

[0036] S305. Calculate the equivalent dipole source P using near-field information 2n×1 , using the least squares method, the objective function is:

[0037] minF(P 2n×1 )=‖E 3m×1 -T 3m×2n P 2n×1 ‖ 2

[0038] Least Squares Solution of Dipole Vector It is obtained through matrix singular value decomposition.

[0039] Preferably, step S4 comprises:

[0040] The far-field sampling surface is a spherical surface with a radius of R2 and an elevation angle of θ between 0 and 180° and an azimuth angle of Set t equally between 0 and 360° ′ 1×t ′ 2 sampling points, the number of sampling points is recorded as m ′ =t ′ 1×t ′ 2;

[0041] The relationship matrix between the dipole array and the far-field sampling points is calculated and recorded as The calculation method is similar to T 3m×2n Similar, except that the number of near-field sampling points m is changed to the number of far-field sampling points m ′ , the sampling point position or (x Ei ,y Ei ,z Ei ), i=1,2…m is changed to or (x ′ Ei ,y ′ Ei ,z ′Ei ),i=1,2…m ′ , the distance r between the corresponding field and the dipole is also changed accordingly;

[0042] Using the matrix equation Complete the calculation of the far field and use the obtained result as the far field prediction result of the antenna to be tested.

[0043] The beneficial effects of the present invention are: the present invention adopts a neural network to perform near-field phase recovery, then constructs an equivalent dipole source, and finally realizes the far-field prediction of the antenna to be tested, with high accuracy and timeliness. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a flow chart of the method of the present invention;

[0045] Figure 2 This is the bottleneck structure diagram with stride 1 and 2;

[0046] Figure 3 It is a schematic diagram of the phase recovery principle;

[0047] Figure 4 is a schematic diagram of an equivalent dipole;

[0048] Figure 5 It is the training loss curve;

[0049] Figure 6 This is the near-field calculation result diagram;

[0050] Figure 7 The far-field actual and predicted images. DETAILED DESCRIPTION

[0051] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings, but the protection scope of the present invention is not limited to the following.

[0052] like Figure 1 As shown, a far-field prediction method based on neural network near-field phase reconstruction includes the following steps:

[0053] S1. Perform near-field simulation on the antenna to be tested, obtain near-field amplitude and phase data of the antenna to be tested, and construct a data set, dividing the data set into a training set and a verification set;

[0054] The AUT uses a conical horn antenna, and FEKO is used to simulate its near field. The geometric dimensions of the conical horn antenna are frequency-dependent, and the aperture radius is 1.2λ. By sampling equidistantly from 1.5 to 2.4 GHz, a total of 800 antenna samples were simulated to obtain the amplitude and phase data of its near field. The near-field sampling surface is a sphere with a radius of 0.6 m, and 20×20 sampling points are set equidistantly between 1 to 179° and 0 to 342°, that is, m=400. The near-field amplitude of the horn antenna in the data set includes |E r |,|E θ |, Three components, phase data φ includes φ r ,φ θ , Three servings.

[0055] The specific steps are as follows:

[0056] In CADFEKO, find the conical horn antenna in the components library, set the operating frequency to 1.5 GHz, and get the model of the conical horn antenna.

[0057] Set the near-field sampling surface with a radius of 0.6 m and a number of 20 × 20, with θ between 1 and 179° and The values ​​are equally spaced between 0 and 342°;

[0058] In order to construct a data set that meets the required frequency band, it is necessary to scan the operating frequency of the horn antenna. The scanning range is 1.5 to 2.4 GHz, and the sampling method is equidistant sampling. In the parameter sweep: create models of the software interface, select the scanning parameter as the operating frequency of the conical horn antenna. Before calculating the near field of the antenna, FEKO will first establish CAD models of 800 horn antennas at different frequencies. During the scanning process, each time the frequency is stepped, the CAD model of the antenna will automatically change accordingly, and then the near field simulation of these 800 antenna samples will be completed in turn. After the near field simulation calculation is completed, the near field amplitude of each horn antenna |E r |,|E θ |, and the near-field phase φ r ,φ θ , It will be saved in the folder where the CAD model is stored, with the file suffix of .out file. The near-field amplitude and phase data of 400 sampling points are extracted from the .out file in the file directory through Python script to complete the construction of the data set. Because the near-field amplitude and phase both contain three components, the structure of the final data set is characterized by the near-field amplitude of the horn antenna, with a dimension of (800,400×3), and the label is the near-field phase, with a dimension of (800,400×3). The 800 samples are then divided into training set and validation set in a ratio of 8:2.

[0059] S2. Construct a neural network for predicting the near-field phase based on the near-field amplitude, train the neural network through the training set, fit the relationship between the near-field amplitude and the near-field phase, verify the trained neural network using the verification set, and obtain the trained neural network after the verification passes;

[0060] Neural Networks is a computational model inspired by the biological nervous system. It processes complex nonlinear relationships by simulating the connections and signal transmission of brain neurons and has been widely used in various fields.

[0061] (1) Structure of Neural Network

[0062] Neural networks usually consist of multiple layers, each containing several neurons (nodes). These layers can be divided into three categories.

[0063] ① Input layer: receives raw data input, each neuron represents an input feature;

[0064] ②Hidden layer: located between the input layer and the output layer, responsible for extracting and converting features. Complex neural networks usually contain multiple hidden layers;

[0065] ③Output layer: Generates the final output result, each neuron corresponds to an output category or value.

[0066] Each neuron is connected to other neurons through weights and biases, and the signal is processed by an activation function. The activation function is generally nonlinear to fit the nonlinear relationship between input and output.

[0067] (2) Working Principle of Neural Network

[0068] The working principle of neural network can be divided into the following steps:

[0069] Forward Propagation: The input data passes through each layer of the network, and is passed layer by layer and weighted and biased, and finally generates the result in the output layer.

[0070] Loss Calculation: By comparing the predicted results with the actual labels, the value of the loss function is calculated to measure the prediction error of the network.

[0071] Backpropagation: According to the gradient of the loss function, adjust the weights and biases of the network to minimize the loss function. This process uses the Gradient Descent or its variant optimization algorithm.

[0072] Iterative Training: Repeat the process of forward propagation and back propagation until the network converges or reaches the preset training rounds.

[0073] For the dataset generated by simulation, this paper uses a neural network to fit the relationship between the near-field amplitude and phase of the AUT. The near-field amplitude of the input feature 400×3 can be regarded as a three-channel image with 20×20 pixels, so this paper intends to use a mature convolutional neural network (CNN) for training and fitting.

[0074] The CNN model used in this article is the MobileNetV2 neural network model, such as Figure 2 The model shown has been successfully applied to near-field plane measurement. The MobileNetV2 neural network model uses the depthwise separable convolution in MobileNetV1, but unlike the latter which uses 1×1 convolution blocks for dimensionality reduction, the former uses 1×1 convolution blocks for dimensionality increase. The MobileNetV2 neural network model introduces linear bottleneck and inverted residuals. Linear bottleneck replaces low-dimensional nonlinear activation transformations with linear transformations to avoid losing too much useful information. Inverted residuals refers to the residual block of the ResNet network. The residual block is a funnel-shaped structure, while inverted residuals is a process of first increasing the dimension and then reducing the dimension. It is a spindle-shaped structure.

[0075] The MobileNetV2 structure used in the present invention is shown in Table 1, and the network is established using Pytorch. The network input is the amplitude data of the near-field simulation of 20×20×3 |E r |,|E θ |, The network output is the predicted three-component near-field phase information of 400 sampling points in the near field. Recorded as There are 1200 values ​​in total. The phase recovery principle diagram is as follows Figure 3 The network is trained using an adaptive momentum optimizer (Adam), with a batch size of 160 and a learning rate of 0.003, and 100 rounds of training using an NVIDIA RTX A5000 GPU. The loss function used is the mean square error (MSE) between the neural network predicted phase and the simulated phase φ.

[0076] Table 1 MobileNetV2 structure diagram

[0077]

[0078] S3. For the antenna to be tested that is not in the data set, only the near-field amplitude information is obtained through simulation, and the predicted near-field phase is obtained by inputting it into the trained neural network, thereby obtaining the complete near-field amplitude and phase information, and the equivalent dipole source of the antenna to be tested is calculated using the least squares method;

[0079] The equivalent dipole theory uses a weighted dipole array instead of the AUT to characterize its radiation characteristics, such as Figure 4 As shown in the figure, the sphere with a larger radius represents the sphere of near-field simulation or measurement, E r ,E θ , represents the three components of the near-field electric field. The sphere with a smaller radius represents the sphere where the dipole is placed. θ , Represents the two components of the dipole. Here we use the electric dipole model. To reduce the computational complexity, the dipole is placed tangentially along the sphere. The electric field of the spherical dipole array and the near-field sphere can be expressed as follows.

[0080] T 3m×2n P 2n×1 =E 3m×1

[0081] Where m is the number of near-field sampling points and n is the number of dipoles. 2n×1 is a dipole array consisting of θ and The column vector of components, E 3m×1 is a near-field electric field consisting of r, θ and A column vector consisting of components, T 3m×2n is the relationship matrix between them, where The six block matrices are only related to the positions of sampling points and dipoles.

[0082]

[0083] For the AUT that is not in the data set, only the near-field amplitude information is obtained through simulation (to address the problem that the near-field phase is difficult to measure accurately in actual testing). According to the trained neural network obtained in the previous step, the near-field amplitude of the AUT |E 3m×1 | Input the neural network to get the predicted near-field phase φ 3m×1 , and thus obtain a complete near-field Calculation of equivalent dipole sources using near-field information It is an electromagnetic inverse problem. The commonly used method is the least squares method, and its objective function is:

[0084] minF(P 2n×1 )=‖E 3m×1 -T 3m×2n P 2n×1 ‖ 2

[0085] The least squares solution of the dipole vector can be obtained by matrix singular value decomposition. In Python, the numpy library can be called to complete the solution.

[0086] The dipole is placed on a spherical surface with a radius of 0.1 m, at 10° to 170° of θ. 18×18 sampling points are set equidistantly between 0 and 340°, that is, n=324. According to the simulated near-field amplitude and the near-field phase predicted by the neural network, the equivalent dipole source of the AUT can be easily calculated by the least squares method.

[0087] S4. Calculate the far field of the antenna under test based on the equivalent dipole theory.

[0088] According to the equivalent dipole theory, to calculate the far field of the AUT, we only need to calculate the relationship matrix between the dipole array and the far field sampling points, which can be expressed as Its form is the same as the relationship matrix T between the dipole array and the near field mentioned above. 3m×2n resemblance. The calculation method is similar to T 3m×2n The sampling point is similar to the previous one, except that the sampling point is set farther away to meet the far field condition, from near field to far field. The far field distance is set to 100m (this distance meets the far field condition), and the sampling points are set to 37×73, that is, at 0~180° and The equivalent dipole source obtained in the previous step is sampled every 5° from 0 to 360°. Using the matrix equation The far field calculation can be completed to predict the far field of the AUT.

[0089] In the embodiment of the present application, the training loss curve of the neural network is as follows: Figure 5 As shown, the dark curve represents the error on the training set, and the light curve represents the error on the validation set. It can be seen that the neural network fitting effect is good, the MSE of the training set and the validation set is below 1000, and there is no serious overfitting phenomenon.

[0090] In order to compare the existing global optimization algorithm with the proposed method, another 100 horn antenna samples that are not in the training set and validation set are selected as the test set. The covariance adaptive evolution strategy (CMA-ES) and the trained neural network are used to construct the equivalent dipole source. Figure 6 The effect of reconstructing the near field of a random horn antenna sample in the test set is shown, where E_real represents the simulated real near field, E_cma represents the near field reconstructed by the CMA-ES optimization algorithm, and E_cnn represents the near field reconstructed by the method proposed in this paper. On two perpendicular near-field sections, the near-field reconstruction effects of the two methods are close and both perform well.

[0091] In order to compare the difference between the two methods in predicting the far field of the AUT, the error is defined as follows.

[0092]

[0093] Where E k represents the electric field component, M s Represents the actual near-field amplitude data of the AUT, M e represents the near-field amplitude data obtained based on the equivalent dipole source. m is the total number of near-field sampling points, i = 1, 2…m and

[0094] Table 2 Results comparison

[0095]

[0096] As can be seen from Table 2, using a neural network to perform phase recovery and construct an equivalent dipole is significantly better than the CMA-ES algorithm, and has great advantages in both timeliness and accuracy. Figure 7 The far-field radiation pattern of the horn antenna predicted by the neural network method is shown, where real represents the actual simulated far field and cnn represents the far field predicted by the neural network. It can be seen that it is highly consistent with the simulation.

[0097] The above is a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, and should not be regarded as excluding other embodiments, but can be used in other combinations, modifications and environments, and can be modified within the scope of the concept described herein through the above teachings or the technology or knowledge of the relevant field. The changes and modifications made by those skilled in the art do not deviate from the spirit and scope of the present invention, and should be within the scope of protection of the claims attached to the present invention.

Claims

1. A far-field prediction method based on near-field phase reconstruction using a neural network, characterized in that: The following steps are involved: S1. Perform near-field simulation on the antenna to be tested, obtain near-field amplitude and phase data of the antenna to be tested, and construct a data set, dividing the data set into a training set and a verification set; S2. Construct a neural network for predicting the near-field phase based on the near-field amplitude, train the neural network through the training set, fit the relationship between the near-field amplitude and the near-field phase, verify the trained neural network using the verification set, and obtain the trained neural network after the verification passes; S3. For the antenna to be tested that is not in the data set, only the near-field amplitude information is obtained through simulation, and the predicted near-field phase is obtained by inputting it into the trained neural network, thereby obtaining the complete near-field amplitude and phase information, and the equivalent dipole source of the antenna to be tested is calculated using the least squares method; S4. Calculate the far field of the antenna under test based on the equivalent dipole theory.

2. The far-field prediction method based on near-field phase reconstruction using a neural network according to claim 1, characterized in that: The step S1 comprises: S101. Assume that the antenna to be tested is a conical horn antenna, the aperture radius is 1.2λ, λ is the wavelength of the electromagnetic wave, and the near-field sampling surface is a spherical surface with a radius of R and the center of the antenna aperture as the origin. T×t sampling points are set equidistantly within the range, and the number of sampling points is recorded as m=t×t; When performing near-field sampling, near-field amplitude data and near-field phase data are collected. The near-field amplitude data includes the radial component |E r |, polar angle component |E θ | and the azimuthal component Three components, the near-field phase data φ also includes the radial component φ r , polar angle component φ θ and the azimuth component Three components, each sampling point has a set of amplitude and phase data; S102. Sample equidistantly within a given frequency range to obtain p frequencies, and construct an antenna CAD model at each frequency. Under the antenna CAD model at each frequency, repeat step S101 to obtain near-field amplitude data and phase data at each frequency, and add them to the same data set to complete the construction of the data set; There are p samples in the dataset, each of which corresponds to an antenna CAD model at a frequency. In each sample, the feature is the near-field amplitude of the horn antenna, with a dimension of m×3, and the label is the near-field phase, also with a dimension of m×3. S103. Divide the data set into a training set and a validation set.

3. The far-field prediction method based on near-field phase reconstruction using a neural network according to claim 1, characterized in that: In step S2, the neural network CNN convolutional neural network for predicting the near-field phase by the near-field amplitude will train the neural network in turn using each sample during the training process: When using any sample to train the neural network, an adaptive momentum optimizer is used. The near-field amplitude of t×t×3 is regarded as a three-channel image of pixel t×t and input into the CNN convolutional neural network. The CNN neural network outputs the predicted phase of t×t×3. The loss function is calculated using the predicted phase of t×t×3 and the near-field phase of t×t×3 in the label. Then, according to the loss function, the gradient is calculated by backpropagation and the parameters of the CNN neural network are updated according to the gradient descent method. After reaching the training round, the trained neural network is verified using the verification set, and the trained neural network is obtained after the verification passes.

4. The far-field prediction method based on neural network near-field phase reconstruction according to claim 1 is characterized in that: The step S3 comprises: S301. Assume that the dipole is placed on a spherical surface with radius R1. Set t1×t1 points equidistantly within the range; the number of dipoles is recorded as n=t1×t1, and each dipole P has a polar angle component P θ and the azimuth component Two components; S302. For the horn antenna not in the data set, simulate and obtain the near-field amplitude, and calculate the near-field amplitude at r,θ, In three directions, the vectors of the near-field amplitudes of m sampling points are constructed according to the simulation results, denoted as |E r | m×1 、|E θ | m×1 , Therefore, the near-field amplitude is recorded as S303. Arrange n dipoles according to θ, Arrange in the direction to get a column vector P θ,n×1 , represents the polar and azimuthal components of n dipoles; S304. Equivalent dipole theory uses a dipole array to replace the antenna under test to characterize its radiation characteristics; Dipole array P 2n×1 The electric field E generated in free space 3m×1 The electric field generated by a single dipole in free space is obtained by coordinate transformation, and its relationship matrix is ​​T 3m×2n , so that T 3m×2n ×P 2n×1 =E 3m×1 ; The dielectric constant in free space is ε0, the magnetic permeability is μ0, and the wave impedance is Wave number The coordinates of a near-field sampling point in the spherical coordinate system are The coordinates in the rectangular coordinate system are (x Ei ,y Ei ,z Ei ); The coordinates of a dipole placement point in the spherical coordinate system are The coordinates in the rectangular coordinate system are (x Pj ,y Pj ,z Pj ), the distance between them is recorded as r; when the positions of the near-field sampling point and the dipole are determined, T 3m×2n Each element in S305. Calculate the equivalent dipole source P using near-field information 2n×1 , using the least squares method, the objective function is: minF(P 2n×1 )=‖E 3m×1 -T 3m×2n P 2n×1 ‖ 2 Least Squares Solution of Dipole Vector It is obtained through matrix singular value decomposition.

5. The far-field prediction method based on near-field phase reconstruction using a neural network according to claim 1, characterized in that: The step S4 comprises: The far-field sampling surface is a spherical surface with a radius of R2 and an elevation angle of θ between 0 and 180° and an azimuth angle of Set t′1×t′2 sampling points equally spaced between 0 and 360°, and the number of sampling points is recorded as m′=t′1×t′2; The relationship matrix between the dipole array and the far-field sampling points is calculated and recorded as The calculation method is similar to T 3m×2n Similar, except that the number of near-field sampling points m is changed to the number of far-field sampling points m ′ , the sampling point position or (x Ei ,y Ei ,z Ei ), i=1,2…m is changed to or (x′ Ei ,y′ Ei ,z′ Ei ), i = 1, 2…m′, the distance r between the corresponding field and the dipole is also changed accordingly; Using the matrix equation Complete the calculation of the far field and use the obtained result as the far field prediction result of the antenna to be tested.

Images