Antenna beam calibration method based on neural network

Abstract

The invention belongs to the technical field of beam pointing precision correction of phased-array antennas, and particularly relates to an antenna beam calibration method based on a neural network. The method specifically comprises the following steps: selecting a plurality of phased-array antennas with the same specification and the same process to measure beam pointing on a rotary table according to a pitch angle interval of k delta theta, and preprocessing data; performing neural network training by taking the pitch angle as the input of a neural network and the beam pointing error as the output of the neural network; for each phased-array antenna with the same specification and the same process, beam pointing is measured on a rotary table according to the pitch angle interval of m delta theta, and data are preprocessed; performing fine tuning on the trained neural network by using the preprocessing result corresponding to each phased-array antenna to obtain a neural network corresponding to each phased-array antenna; and utilizing the fine-tuned neural network to predict the beam direction of the pitch angle interval delta theta, constructing a lookup table, writing the lookup table into a corresponding phased-array antenna, and utilizing the lookup table to complete antenna beam calibration.

Description

Technical Field

[0001] The invention belongs to the technical field of beam pointing accuracy correction of phased array antennas, and in particular relates to an antenna beam calibration method based on a neural network. Background Art

[0002] In recent years, the advantages of phased array antennas have become increasingly prominent and have been widely used in military and civilian fields. Beam pointing accuracy is one of the important technical indicators of phased array antennas. In order to improve the pointing accuracy, beam pointing correction is required. Traditional correction methods include turntable measurement method and interpolation method. The turntable measurement method measures all beam pointings in turn to establish a lookup table. This method has high accuracy but high cost and time-consuming. The interpolation method uses part of the beam pointing for interpolation to establish a lookup table, which can reduce the number of measurements, but the degree is limited, the efficiency is low and the accuracy is not high.

[0003] Neural networks have a unique network structure and information transmission mechanism. The core feature of a neural network is that it processes data through multiple connected neuron layers, with the output of each layer serving as the input to the next layer. This structure enables neural networks to learn and capture complex patterns and relationships in input data. By adjusting the connection weights between neurons, neural networks can model and predict input data, making them very suitable for processing various types of data. For example, when processing beam pointing errors, neural networks can learn and identify error patterns in input data, thereby more accurately predicting and correcting errors. At present, no researchers have applied neural networks to beam pointing correction. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide an antenna beam calibration method based on a neural network to overcome the shortcomings of the prior art and greatly improve the calibration speed and system real-time performance while ensuring the accuracy of beam pointing correction.

[0005] The technical solution for implementing the present invention is as follows:

[0006] In the first aspect, a neural network-based antenna beam calibration method is used to calibrate the beam pointing of multiple phased arrays of the same specification and process, and the specific process is as follows:

[0007] Select multiple phased array antennas of the same specifications and process to measure beam pointing on a turntable at an interval of kΔθ in pitch angle, and pre-process the data; use the pitch angle as the input of the neural network, and the beam pointing error as the output of the neural network to train the neural network;

[0008] For each phased array antenna of the same specification and process, the beam pointing is measured on the turntable at a pitch angle interval of mΔθ, and the data is preprocessed, where m>k; the preprocessing results corresponding to each phased array antenna are used to fine-tune the trained neural network to obtain a neural network corresponding to each phased array antenna;

[0009] The fine-tuned neural network is used to predict the beam pointing of the pitch angle interval Δθ, a lookup table is constructed and written into the corresponding phased array antenna, and the antenna beam calibration is completed using the lookup table.

[0010] Furthermore, the phased array antenna of the present invention measures beam pointing on a turntable at a pitch angle interval of kΔθ, and the specific process is as follows:

[0011] Set the pitch angle measurement range to -θ M ~θ M , set the azimuth measurement range to -φ M ~φ M ;

[0012] The turntable completes an azimuth scan from left to right, moves downward by kΔθ, and then completes an azimuth scan from right to left, and then moves downward by kΔθ, and so on, until the scan within the measurement range is completed. During the above scanning process, the beam pointing is measured;

[0013] The measured beam pointing is subtracted from the actual beam pointing in the above scanning process to obtain beam pointing error data at all angles.

[0014] Furthermore, the data preprocessing described in the present invention is as follows: the beam pointing error is extracted at equal intervals on the azimuth plane and averaged; the error data is separated according to the left scan and the right scan on the azimuth plane, and used as sample data for training of the neural network respectively. The neural network includes two left / right scan data, which are respectively input into the neural network for training.

[0015] Furthermore, the present invention also includes a verification link for the accuracy of the neural network. There are two verification indicators in total, namely azimuth 3σ and elevation 3σ; among which, the azimuth elevation angle 3σ is calculated by calculating three times the standard deviation 3σ of the difference between the predicted beam pointing angle and the actual measured beam pointing angle.

[0016] Furthermore, the present invention uses the preprocessing results corresponding to each phased array antenna to fine-tune the trained neural network, specifically:

[0017] All hidden layer parameters of the pre-trained network are frozen, and two additional hidden layers are added. The measurement beam is then directed to the input of the pre-trained neural network, and training is performed using a small learning rate to fine-tune the network weight parameters.

[0018] Furthermore, the present invention predicts all scanning data, including: predicting left scanning data and right scanning data respectively according to the pitch angle interval Δθ using a trained neural network; and integrating the predicted left scanning data and right scanning data to obtain all beam pointing.

[0019] Furthermore, after the present invention constructs the lookup table and writes it into the phased array, when the phased array is pointed to a certain beam pointing angle, the phased array sends to the phased array according to the lookup table no longer the theoretical beam pointing, but the corresponding predicted beam pointing, and the predicted beam pointing is used instead of the theoretical beam pointing.

[0020] In the second aspect, a neural network-based antenna beam calibration method is used to calibrate the beam pointing of a phased array. The specific process is as follows:

[0021] Select a phased array antenna to measure the beam pointing at a pitch angle interval of kΔθ on a turntable, and pre-process the data;

[0022] The pitch angle is used as the input of the neural network, and the beam pointing error is used as the output of the neural network to train the neural network;

[0023] The trained neural network is used to predict the beam pointing of the pitch angle interval Δθ, a lookup table is constructed and written into the phased array antenna, and the antenna beam calibration is completed using the lookup table.

[0024] Beneficial effects:

[0025] The phased array antenna beam pointing fast calibration method provided by the present invention can predict the error at all angles when the measured angle does not exceed 1 / k; by introducing transfer learning, the error at all angles can be further predicted when the measured angle does not exceed 1 / m (m>k); the standard deviation of the absolute value of the predicted error and the true error deviation is 3σ<σ std (σ std The prediction accuracy is high. The predicted beam pointing is imported into the lookup table and written into the phased array antenna, so that the pointing accuracy of the phased array antenna can be corrected. The method provided by the present invention effectively improves the beam pointing accuracy and real-time performance of the phased array system, and has low cost and low system complexity. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0027] Figure 1Schematic diagram of beam pointing angle error at similar pitch angles;

[0028] Figure 2 It is a schematic diagram of the probability density distribution of the difference of beam pointing angle errors between similar pitch angles;

[0029] Figure 3 It is a schematic diagram of beam pointing angle error under non-similar pitch angles;

[0030] Figure 4 A schematic diagram of a flow chart of a neural network-based beam pointing rapid calibration method provided by the present invention;

[0031] Figure 5 The measurement trajectory diagram required for the existing calibration method;

[0032] Figure 6 A measurement trajectory diagram provided for an embodiment of the method proposed in the present invention;

[0033] Figure 7 Shown is an explanatory diagram of the result verification indicators in the embodiment of the method proposed in the present invention. DETAILED DESCRIPTION

[0034] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0035] It should be noted that the following embodiments and features in the embodiments may be combined with each other in the absence of conflict; and, based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making any creative work are within the scope of protection of the present disclosure.

[0036] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.

[0037] During the research, the applicant found that when using neural networks for antenna calibration, data is usually collected according to the accuracy requirements. For example, when the accuracy requirement of antenna beam measurement is Δθ, data is collected at intervals of Δθ, and the neural network is trained based on the collected data, which can theoretically meet the corresponding accuracy requirements. However, the above process ignores an important phenomenon: the overall distribution characteristics of beam pointing errors between similar pitch angles are similar and gradual.

[0038] Similarity means that the overall distribution characteristics of beam pointing errors are similar between similar elevation angles, such as Figure 1 As shown, Figure 1 Taking the azimuth error as an example, the beam pointing error at each similar elevation angle is shown in the entire azimuth plane (-θ M ~θ M ) have similar fluctuation characteristics. Figure 2 The probability distribution density of the difference between the beam pointing errors at each elevation angle is shown. In the figure, 98% or more of the difference between the beam pointing errors at each elevation angle is within ±Δθ, that is, the magnitude of the beam pointing errors at each elevation angle is relatively close, indicating the similarity of the distribution characteristics of the beam pointing errors.

[0039] Gradual variation means that the overall distribution characteristics of the beam pointing error gradually change with the change of the pitch angle, as shown in the following example: Figure 6 . Figure 6 The beam errors at intervals of kΔθ (k = 10, pitch angle 1, pitch angle 11, pitch angle 21, pitch angle 31) are plotted for comparison. Figure 1 and Figure 3 It can be seen that the overall distribution of errors at pitch angles 1-7 is similar, while the overall distribution of errors at pitch angles 1, 11, 21, and 31 changes gradually with the change of pitch angle, indicating that the distribution characteristics of beam pointing errors are gradually changing.

[0040] The following is an explanation of the principle of neural network learning error distribution characteristics: The core of the neural network is its hidden layer neuron state h n , which is updated at each sample n and depends on the current input sample x n :

[0041] h n =f(W x ·x n +b)

[0042] Among them, h n is the hidden neuron state of sample n; W x is the input weight matrix; x n is the input of sample n; b is the bias vector; f is the activation function (such as tanh or ReLU).

[0043] In this embodiment, for similar pitch angles, since the angle change is small, the corresponding input sequence x n The change between them will also be smaller. The neural network updates the hidden neuron state h in each sample n When, it mainly depends on the current input x n Since the input changes slightly, the hidden neuron states also change slightly, and the characteristics of the hidden neuron states show similarity. When the characteristics between the input samples gradually change, the hidden neuron states of the neural network will also gradually adjust accordingly, thereby capturing the gradual changes in the input data.

[0044] In the output stage, the neural network is based on the current hidden neuron state h n To predict the output y n , that is, the beam pointing angle:

[0045] y n =g(W y ·h n +c)

[0046] Among them, y n is the output of sample n; W y is the weight matrix of the output layer; c is the bias vector of the output layer; g is the activation function of the output layer.

[0047] By training the neural network, that is, training and optimizing the weight matrix W x and W y , learn the input sequence x n The output beam pointing angle y n The relationship between the two, the neural network can accurately predict the beam pointing angle error based on the new input pitch angle.

[0048] Based on the above analysis of similarity and gradualness, when training the neural network, the input data of the present application no longer needs to be data collected at an interval of Δθ. Instead, a K-fold interval is set on the basis of Δθ of the required accuracy, that is, the pitch angle is collected at an interval of kΔθ to collect sample data, thereby overcoming the problem of too long time required to measure data due to the small measurement interval.

[0049] The present application embodiment provides a neural network-based antenna beam calibration method, which is used to calibrate the beam pointing of a phased array. The specific process is as follows:

[0050] S11: Select a phased array antenna to measure beam pointing on a turntable at a pitch angle interval of kΔθ, and pre-process the data;

[0051] S12: using the pitch angle as the input of the neural network and the beam pointing error as the output of the neural network to perform neural network training;

[0052] S13: Use the trained neural network to predict the beam pointing of the pitch angle interval Δθ, construct a lookup table and write it into the phased array antenna, and use the lookup table to complete the antenna beam calibration.

[0053] Going one step further on this basis, the present invention utilizes the concept of transfer learning to adopt sample data from multiple phased arrays of the same specifications and processes, uses the collected results to pre-train the neural network, and then collects small batches of sample data for each phased array to achieve fine-tuning of the pre-trained neural network, so that the fine-tuned neural network meets the corresponding phased array beam pointing fast calibration requirements.

[0054] like Figure 4 As shown, another embodiment of the present application, an antenna beam calibration method based on a neural network is provided in the embodiment of the present application, and the specific process is as follows:

[0055] S21: Select multiple phased array antennas of the same specification and process to measure beam pointing on a turntable at a pitch angle interval of kΔθ, and pre-process the data; use the pitch angle as the input of the neural network, and the beam pointing error as the output of the neural network to train the neural network;

[0056] S22: for each phased array antenna of the same specification and process, measure the beam pointing on the turntable at a pitch angle interval of mΔθ, and pre-process the data, where m>k; use the pre-processing result corresponding to each phased array antenna to fine-tune the trained neural network to obtain a neural network corresponding to each phased array antenna;

[0057] S23: Use the fine-tuned neural network to predict the beam pointing of the pitch angle interval Δθ, construct a lookup table and write it into the corresponding phased array antenna, and use the lookup table to complete the antenna beam calibration.

[0058] In this embodiment, for phased array radar antennas of the same specifications and process, only one basic neural network training is required, and then for each phased array individual, only a small amount of data is required for fine-tuning, so the purpose of beam calibration of multiple phased array antennas of the same specifications and batches can be quickly achieved. At the same time, in this embodiment, kΔθ is used as the interval during training, so the number of samples required for measurement can be reduced to 1 / k of the original, and the corresponding measurement time is also reduced to 1 / k of the original.

[0059] When determining the appropriate k value, sample data with a pitch angle interval of Δθ can be measured in advance, and on this basis, the pitch angle interval kΔθ of the sample data is sampled, the sampled data is input into the network training and the data with the pitch angle interval Δθ is predicted, and then the prediction result is compared with the result measured in advance. The appropriate k value can be determined when the required accuracy is met.

[0060] In another embodiment of the present application, the phased array antenna measures the beam pointing at a pitch angle interval of kΔθ on a turntable, and the specific process is as follows:

[0061] Set the pitch angle measurement range to -θ M ~θ M , set the azimuth measurement range to -φ M ~φ M ;

[0062] The turntable completes an azimuth scan from left to right, moves downward by kΔθ, and then completes an azimuth scan from right to left, and then moves downward by kΔθ, and so on, until the scan within the measurement range is completed. During the above scanning process, the beam pointing (including the pitch angle and azimuth angle) is measured; the azimuth angle interval in this process is determined by the turntable rotation speed and PRF;

[0063] The measured beam pointing is subtracted from the actual beam pointing in the above scanning process to obtain beam pointing error data at all angles.

[0064] This embodiment is executed according to the above process to measure all beam pointing errors, such as Figure 6 As shown, for the sake of convenience, the pitch angle is only -θ1~θ1, θ1<θ M , traditional beam acquisition is as follows Figure 5 shown.

[0065] In another embodiment of the present application, the data is preprocessed as follows: the beam pointing error is extracted at equal intervals on the azimuth plane and averaged; the error data is separated according to the left scan and the right scan on the azimuth plane, and used as sample data for training the neural network respectively. The neural network includes two, and the left / right scan data is input into the neural network for training respectively. In this embodiment, the measured pitch angle is used as a sample feature, and the corresponding left scan data and right scan data are used as sample labels and input into the neural network for training respectively.

[0066] In another embodiment of the present application, the neural network type is selected as MLP, LSTM or GRU respectively; the neural network structure adopted in this embodiment includes three hidden neuron layers with 512 units, each hidden neuron layer is followed by a ReLU activation layer to enhance the nonlinear feature extraction capability. The input is a univariate sequence of pitch angles, which is processed by these hidden layers and connected to a fully connected layer to output the regression result. The neural network uses the He initialization method, adopts the Adam optimization algorithm, sets an adaptive learning rate adjustment strategy, and is trained on a GPU, which is suitable for processing and predicting sequence data.

[0067] Another embodiment of the present application also includes a verification link for the accuracy of the neural network. There are two verification indicators in total, namely azimuth 3σ and pitch 3σ; wherein, the azimuth pitch angle 3σ is calculated by calculating three times the standard deviation 3σ of the difference between the predicted beam pointing angle and the actual measured beam pointing angle. In this embodiment, the beam pointing error output by the neural network is used to verify the accuracy of the neural network trained in step S2. There are two verification indicators in total, namely azimuth 3σ and pitch 3σ. wherein, the azimuth pitch angle 3σ is calculated by calculating three times the standard deviation (3σ) of the difference between the predicted beam pointing angle and the actual measured beam pointing angle.

[0068] It should be noted that, in this embodiment, the calibration goal is that the theoretical beam pointing of the phased array after calibration and the measured beam pointing satisfy an error of 3σ<σ std , and the calculated index is the 3σ difference between the predicted beam pointing of the phased array and the measured beam pointing. The specific explanation is as follows:

[0069] After the lookup table is constructed and written into the phased array, when the phased array is pointed to a certain beam pointing angle, the phased array sends the corresponding predicted beam pointing to the phased array based on the lookup table instead of the theoretical beam pointing. This process can be regarded as replacing the theoretical beam pointing with the predicted beam pointing, such as Figure 7 Therefore, it is only necessary to ensure that the error between the predicted angle and the measured angle before calibration is 3σ<σ std That's it.

[0070] In another embodiment of the present application, the preprocessing results corresponding to each phased array antenna are used to fine-tune the above-mentioned trained neural network. Specifically, all hidden layer parameters of the pre-trained network are frozen, and two additional hidden layers are added. The measurement beam is then directed to the input of the pre-trained neural network, and training is performed using a small learning rate, thereby achieving fine-tuning of the network weight parameters.

[0071] In another embodiment of the present application, on the premise of distinguishing left-scan and right-scan data, the pitch angles of unmeasured beam pointing (pitch angle interval Δθ) are sequentially input into the trained neural network model network as sample features, and the sample labels output by the network are the predicted beam pointing errors at the corresponding pitch angles.

[0072] In another embodiment of the present application, a trained neural network is used to predict the left scan data and the right scan data respectively according to the pitch angle interval Δθ; the predicted left scan data and right scan data are integrated to obtain all beam pointings, a lookup table is constructed, and the lookup table is written into the phased array antenna to complete the calibration.

[0073] The measured part and the predicted part together construct a lookup table of azimuth and elevation with the predicted minimum angle (Δθ) as the interval.

[0074] Based on the above method, the specific implementation process S11-S13 is simulated and verified on the measured data of 10 different phased arrays (in the simulation experiment, Δθ=0.2°, k=10, σ std =0.25°) The verification results are shown in Figures 1-3.

[0075] Table 1 Results of beam pointing error prediction by MLP network

[0076] Phased Array Number Azimuth 3σ(°) Pitch angle 3σ(°) 1 0.1690 0.1747 2 0.1554 0.1929 3 0.1739 0.1687 4 0.1788 0.2232 5 0.1681 0.1646 6 0.1915 0.1649 7 0.1901 0.2035 8 0.1550 0.1839 9 0.1626 0.1760 10 0.1787 0.1956

[0077] Table 2 Results of LSTM network prediction of beam pointing error

[0078] Phased Array Number Azimuth 3σ(°) Pitch angle 3σ(°) 1 0.1601 0.1565 2 0.1489 0.1791 3 0.1709 0.1623 4 0.1598 0.1897 5 0.1538 0.1597 6 0.1781 0.1491 7 0.1845 0.1751 8 0.1542 0.1768 9 0.1466 0.1664 10 0.1639 0.1646

[0079] Table 3 Results of GRU network prediction beam pointing error

[0080]

[0081]

[0082] In the verification results, the simulation verification of 10 groups of data of 3 networks all reached the standards, which shows the effectiveness of this method.

[0083] Based on the above method, the specific implementation process S21-S23 is used to pre-train the MLP network model using the beam pointing data of 10 phased arrays with the same specifications and processes. The measured beam pointing data of two phased arrays with m=2k and 3k are input into the network for fine-tuning and prediction. The simulation verification results are shown in Figures 4 and Table 5.

[0084] Table 4 Results of predicted beam pointing error of phased array 1

[0085] Sampling ratio Azimuth 3σ(°) Pitch angle 3σ(°) 1 / 30 0.2031 0.2479 1 / 20 0.1806 0.2080

[0086] Table 5 Results of predicted beam pointing error of phased array 2

[0087] Sampling ratio Azimuth 3σ(°) Pitch angle 3σ(°) 1 / 30 0.2158 0.2491 1 / 20 0.1665 0.1892

[0088] In the verification results, the simulation verifications of the two phased arrays both met the standards, which shows the effectiveness of this method.

[0089] Furthermore, in specific practice, the beam pointing accuracy of the phased array antenna can be corrected by finding the predicted beam pointing error corresponding to the designed beam pointing.

[0090] The embodiment provides a beam pointing fast calibration method, which can predict the beam pointing at all angles when only measuring the beam pointing at a few angles. Compared with the prior art, this method has fewer measurements, greatly improved efficiency, high accuracy, low cost, and is easy to implement in engineering.

[0091] It should be noted that, in the description of the present invention, the terms "first", "second" and the like are only used for illustrative purposes and do not indicate or imply relative importance. In addition, in the description of the present application, unless otherwise explicitly stated, the term "plurality" refers to at least two.

[0092] Any process or method described in any flow chart or other manner described in the present invention can be understood as a code module, fragment or part of executable instructions including one or more steps for performing a specific logical function or process. The scope of the preferred embodiment of the present invention also includes other implementations, which may not be performed in the order shown, and will be understood by those skilled in the art to which the present invention relates.

[0093] It is understood that when describing the present invention, terms such as "one embodiment", "some embodiments", "examples", "specific examples" or "some examples" refer to at least one embodiment or example with specific features, structures, materials or characteristics. The use of these terms does not necessarily refer to the same embodiment or example, and the specific features, structures, materials or characteristics in the description can be appropriately combined in one or more embodiments or examples.

[0094] The above specific implementation further describes the purpose, technical methods and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation of the present invention and is not intended to limit the scope of protection of the present invention. Any modification, equivalent replacement or improvement made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A neural network-based antenna beam calibration method for calibrating the beam pointing of multiple phased arrays of the same specification and process, characterized in that: The specific process is: Select multiple phased array antennas of the same specifications and process to measure beam pointing on a turntable at an elevation angle interval of kΔθ, and pre-process the data; use the elevation angle as the input of the neural network and the beam pointing error as the output of the neural network to train the neural network; For each phased array antenna of the same specification and process, the beam pointing is measured on the turntable at a pitch angle interval of mΔθ, and the data is preprocessed, where m>k; the preprocessing results corresponding to each phased array antenna are used to fine-tune the trained neural network to obtain a neural network corresponding to each phased array antenna; The fine-tuned neural network is used to predict the beam pointing of the pitch angle interval Δθ, a lookup table is constructed and written into the corresponding phased array antenna, and the antenna beam calibration is completed using the lookup table.

2. The antenna beam calibration method based on neural network according to claim 1, characterized in that: The phased array antenna measures the beam pointing at a pitch angle interval of kΔθ on a turntable, and the specific process is as follows: Set the pitch angle measurement range to -θ M ~θ M , set the azimuth measurement range to -φ M ~φ M ; The turntable completes an azimuth scan from left to right, moves downward by kΔθ, and then completes an azimuth scan from right to left, and then moves downward by kΔθ, and so on, until the scan within the measurement range is completed. During the above scanning process, the beam pointing is measured; The measured beam pointing is subtracted from the actual beam pointing in the above scanning process to obtain beam pointing error data at all angles.

3. The antenna beam calibration method based on neural network according to claim 1, characterized in that: The data is preprocessed as follows: the beam pointing error is extracted at equal intervals on the azimuth plane and averaged; the error data is separated according to the left scan and the right scan on the azimuth plane, and used as sample data for training the neural network respectively. The neural network includes two left / right scan data, and the left / right scan data are respectively input into the neural network for training.

4. The antenna beam calibration method based on neural network according to claim 1, characterized in that: It also includes a verification phase for the accuracy of the neural network. There are two verification indicators in total, namely azimuth 3σ and elevation 3σ. The azimuth and elevation angles 3σ are calculated by calculating three times the standard deviation 3σ of the difference between the predicted beam pointing angle and the actual measured beam pointing angle.

5. The antenna beam calibration method based on neural network according to claim 1, characterized in that: The preprocessing result corresponding to each phased array antenna is used to fine-tune the trained neural network, specifically: All hidden layer parameters of the pre-trained network are frozen, and two additional hidden layers are added. The measurement beam is then directed to the input of the pre-trained neural network, and training is performed using a small learning rate to fine-tune the network weight parameters.

6. The antenna beam calibration method based on neural network according to claim 1, characterized in that: Predicting all scanning data includes: predicting left scanning data and right scanning data respectively according to the pitch angle interval Δθ using a trained neural network; integrating the predicted left scanning data and right scanning data to obtain all beam pointing.

7. The antenna beam calibration method based on neural network according to claim 1, characterized in that: After the lookup table is constructed and written into the phased array, when the phased array is pointed to a certain beam pointing angle, the phased array no longer sends the theoretical beam pointing to the phased array based on the lookup table, but the corresponding predicted beam pointing. The predicted beam pointing replaces the theoretical beam pointing.

8. A neural network-based antenna beam calibration method for calibrating the beam pointing of a phased array, characterized in that: The specific process is: Select a phased array antenna to measure the beam pointing at a pitch angle interval of kΔθ on a turntable, and pre-process the data; The pitch angle is used as the input of the neural network, and the beam pointing error is used as the output of the neural network to train the neural network; The trained neural network is used to predict the beam pointing of the pitch angle interval Δθ, a lookup table is constructed and written into the phased array antenna, and the antenna beam calibration is completed using the lookup table.

Images