Millimeter wave dual-polarization phased-array antenna beam forming method and device

Abstract

The invention discloses a millimeter wave dual-polarization phased-array antenna beam forming method and device, and relates to the technical field of high-frequency communication, and the method comprises the following steps: obtaining a complex weight signal of each array element in an antenna array under a horizontal polarization channel and a vertical polarization channel, respectively carrying out amplitude response and phase response acquisition and calibration processing on the complex weight signal; performing dynamic extraction and modeling on amplitude errors and phase errors of each array element in different polarization channels; on the basis of error estimation, constructing a machine learning prediction module comprising a time sequence prediction model; performing joint estimation on a real-time error estimation result and a prediction error result; constructing a hardware-software collaborative calibration structure comprising a dynamic gain control module on the basis of the compensation weight; beam focusing of the target space area is realized; according to the invention, the hidden problems of beam offset, sidelobe lifting, gain loss and the like caused by inconsistency of polarization channels are solved.

Description

Technical Field

[0001] This invention relates to the field of high-frequency communication technology, specifically to a method and apparatus for beamforming a millimeter-wave dual-polarized phased array antenna. Background Technology

[0002] As the frequency increases, the propagation loss of millimeter-wave signals becomes greater, which requires more efficient antenna systems to enhance the effective transmission of signals. Dual-polarized antennas can simultaneously transmit and receive horizontally and vertically polarized signals. Through this technology, the antenna system can simultaneously transmit and receive two independent signals in the same frequency band, thereby improving spectrum utilization efficiency. Phased array antennas can electronically adjust the phase of each element in the antenna array to achieve beam pointing control, which is faster and more precise than traditional mechanical antenna adjustment.

[0003] However, in dual-polarized antenna array systems, the horizontal and vertical polarization channels often exhibit different amplitude and phase responses due to hardware differences and inconsistent paths. This inconsistency between channels introduces errors during beamforming, especially during large-angle scanning, leading to implicit performance losses such as reduced array gain, beam shift, and even sidelobe lift. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and apparatus for beamforming a millimeter-wave dual-polarized phased array antenna.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a beamforming method for a millimeter-wave dual-polarized phased array antenna, comprising the following steps: S1: Construct a dual-polarized antenna array structure, obtain the complex weighted signals of each array element in the antenna array under the horizontal polarization channel and the vertical polarization channel, and perform amplitude response and phase response acquisition and calibration processing on the complex weighted signals respectively; S2: Based on the operational status configuration of the dual-polarized antenna array, a real-time error estimation module is configured to dynamically extract and model the amplitude and phase errors of each array element in different polarization channels; S3: Based on error estimation, construct a machine learning prediction module that includes a time series prediction model to model the future error change trend within the target time window and output the prediction error results; S4: Jointly estimate the real-time error estimation results and the prediction error results to calculate the complex form target compensation weight of each array element in the array, and apply it to error compensation; S5: Based on the compensation weight, a hardware-software collaborative calibration structure including a dynamic gain control module is constructed to update the array gain matrix parameters in real time and form an automated beam adjustment mechanism. S6: Based on the compensation weight and calibration structure, the phased array antenna is controlled to form the required beam direction in the target direction, thereby achieving beam focusing on the target spatial region.

[0006] As a preferred embodiment of the present invention, the real-time error estimation module is a closed-loop feedback structure. This structure, based on the reference pilot subcarrier in the received signal, extracts the amplitude error of each array element in the horizontal polarization channel and the vertical polarization channel at the current moment through demodulation. ; With phase error: ; Where t represents the current time and i represents the array element number. This represents the real-time amplitude error estimate of the i-th array element at time t. This represents the estimated real-time phase error of the i-th array element at time t.

[0007] As a preferred embodiment of the present invention, the machine learning prediction module constructs an error prediction model based on a long short-term memory neural network. The model takes the error signal sequence of the past N time points as input and outputs an estimated value of the error after a future time delay of τ. ; and: ; Where N is the length of the input sequence, and τ is the target prediction delay. This represents the magnitude error prediction value of the machine learning model for the i-th array element at a future time t+τ. This represents the phase error prediction value of the i-th array element by the machine learning model at a future time t+τ.

[0008] As a preferred embodiment of the present invention, the complex form target compensation weight The expression is as follows: ; in, φi represents the compensation weight of the i-th array element at time t, φi is the expected phase value of array element i in the target direction, and α and β are the weighting coefficients of prediction error and real-time error, respectively.

[0009] As a preferred embodiment of the present invention, the hardware-software collaborative calibration structure includes a dynamic gain control matrix G(t), a target beam vector Wd(t), and a current output beam vector W0(t). The dynamic gain control matrix G(t) is iteratively updated using the following least squares optimization model: ; Where E(t) represents the beam error at the calibration time.

[0010] As a preferred embodiment of the present invention, the machine learning prediction module and the real-time error estimation module form a double closed-loop structure, including: Feedforward channel: The machine learning module predicts error trends based on historical data and outputs prediction weights for compensation; Feedback channel: The real-time error module samples the current array state and outputs the actual error parameters; During the compensation weight calculation stage, the dual closed-loop structure synchronously introduces prediction and feedback information to form a feedforward and feedback joint modeling strategy, which is used to realize the coupled update of error modeling information in weight calculation.

[0011] As a preferred embodiment of the present invention, the beamforming control is performed independently once in the horizontal polarization channel and the vertical polarization channel. After execution, the phase results of the two channels are input into the polarization coordination control module. The polarization coordination control module performs phase alignment and gain matching processing based on the cross-correlation function between the polarization channels to form a unified phase reference surface. The alignment result is used to eliminate the phase shift caused by the difference in polarization direction.

[0012] As a preferred embodiment of the present invention, in step S4, the error input value required for compensation is calculated using the following sliding weighted fusion model: ; ; Where γ(t) is the error fusion weighting factor, and its domain is [0,1].

[0013] The present invention also provides a millimeter-wave dual-polarized phased array antenna device, comprising: A dual-polarized antenna array structure is used to transmit and receive radio frequency signals in the horizontal polarization channel and the vertical polarization channel, respectively; The real-time error estimation module is configured to extract the amplitude and phase errors of each array element based on the pilot subcarriers in the received signal, forming a closed-loop feedback structure. The machine learning prediction module includes an error trend modeling unit based on a long short-term memory network, which receives historical error signal sequences and outputs error prediction results for future target delays. The weight calculation module is used to calculate the complex form compensation weight of each array element based on the real-time estimation error and the prediction error. The hardware-software co-calibration structure includes a dynamic gain control matrix and a beam control unit, which adjust the transmission state of the antenna array in real time by jointly compensating weights and calibration parameters. The beam control module is used to form a spatial beam in the target direction and perform cooperative alignment and gain matching between polarization channels. The beneficial effects of this invention are: This invention solves the hidden problems of beam offset, sidelobe rise and gain loss caused by polarization channel inconsistency by introducing a multi-module beamforming method that combines real-time feedback estimation, machine learning prediction, hardware-software co-calibration and dual-polarization phase alignment. Especially under large-angle scanning and complex channel conditions, it can effectively improve beam pointing accuracy, array gain consistency and system robustness. Attached Figure Description

[0014] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic flowchart of the beamforming method for millimeter-wave dual-polarized phased array antenna of the present invention. Detailed Implementation

[0015] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0016] Example 1 like Figure 1 As shown, a beamforming method for a millimeter-wave dual-polarized phased array antenna includes the following steps: S1: Construct a dual-polarized antenna array structure, obtain the complex weighted signals of each array element in the antenna array under the horizontal polarization channel and the vertical polarization channel, and perform amplitude response and phase response acquisition and calibration processing on the complex weighted signals respectively; This step begins with designing and deploying an antenna array that supports dual polarization. Each element outputs a corresponding complex weighted signal on both polarization channels, which is a complex signal containing amplitude and phase information. Since various non-ideal factors may exist during the manufacturing, installation, and environmental changes of the antenna array, the amplitude and phase responses of each element must be accurately acquired and compared with the ideal or reference responses for initial calibration. This calibration process provides a benchmark for subsequent error extraction and dynamic compensation.

[0017] S2: Based on the operational status configuration of the dual-polarized antenna array, a real-time error estimation module is configured to dynamically extract and model the amplitude and phase errors of each array element in different polarization channels; During system operation, due to environmental disturbances (such as temperature, humidity, mechanical vibration, etc.), hardware aging, or other random factors, the actual response of each array element often exhibits amplitude attenuation or phase drift. Therefore, this step introduces a real-time error estimation module. This module compares the currently measured array element response with the calibration reference value, extracts the deviation information at the current time point, and establishes a dynamic error model. This model not only records the current error state but also provides data support for subsequent time series prediction.

[0018] S3: Based on error estimation, construct a machine learning prediction module that includes a time series prediction model to model the future error change trend within the target time window and output the prediction error results; Errors are not entirely random, but usually have a certain time correlation and variation pattern. Therefore, this step introduces a time series prediction module based on machine learning, such as LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), or Transformer architecture model, to model and predict the future development trend of errors. The model takes historical error data as input and can output the error change trend within a given future time window after training, including both amplitude and phase. By knowing the potential error development in advance, the system can make pre-compensation before physical errors occur, thereby improving the overall beam control accuracy and system robustness.

[0019] S4: Jointly estimate the real-time error estimation results and the prediction error results to calculate the complex form target compensation weight of each array element in the array, and apply it to error compensation; To achieve high-precision compensation, the system weights and fuses the real-time error estimation results from the second step with the prediction error results from the third step to form a more complete error representation for each array element. Based on this representation, a complex-form weight for compensation is calculated. The design principle of the compensation weight is to adjust the amplitude and phase of the excitation signal in reverse so that the beam direction of the final array output is aligned with the ideal expected value. This weight is usually expressed in complex form, which makes it easy to directly act on the array signal processing link to correct the deviation caused by the error and achieve dynamic error compensation.

[0020] S5: Based on the compensation weight, a hardware-software collaborative calibration structure including a dynamic gain control module is constructed to update the array gain matrix parameters in real time and form an automated beam adjustment mechanism. Compensation calculated solely by software is insufficient to completely offset the impact of physical errors on system performance. Therefore, this step constructs a hardware-software co-calibration structure, which includes a dynamic gain control module. This module adjusts the power amplifier gain and phase settings of each channel in the array in real time based on the compensation weight adjustment results, thereby achieving synchronous updates of the gain matrix parameters of the entire array. The software part is responsible for error calculation and control strategy generation, while the hardware part responds to control signals to complete physical adjustments. The two work together to construct a continuously adaptive closed-loop system, thereby ensuring the long-term stability of beam performance.

[0021] S6: Based on the compensation weight and calibration structure, the phased array antenna is controlled to form the required beam pointing in the target direction, thereby achieving beam focusing on the target spatial region; Finally, the system utilizes compensated and calibrated array excitation weights to achieve precise beamforming and direction control. Specifically, by precisely controlling the amplitude and phase of the excitation signals of each array element, the wavefront generated by the entire array is coherently superimposed in the target direction, thereby enhancing the radiation intensity in the target direction and achieving beam focusing in space. This beam can be dynamically adjusted according to the target's motion state or communication requirements, enabling the system to maintain efficient and stable beam pointing capability even when facing external disturbances or target changes. This process constructs an adaptive beamforming mechanism based on machine learning and error control, providing key technical support for high-frequency communication and sensing systems.

[0022] In summary, this method fully considers the extremely high requirements of phase and amplitude accuracy for systems in the millimeter-wave band. It introduces dynamic error modeling, prediction, and hardware-software co-calibration mechanisms on the basis of traditional phased array beam control, realizing a leap from passive compensation to active prediction and adaptive control. It enhances system capacity through a dual-polarization structure, enhances the foresight of error response through a time-series prediction model, and enhances the accuracy and robustness of beam control through real-time compensation and co-calibration, thus forming an advanced and intelligent beamforming solution.

[0023] Furthermore, the real-time error estimation module is a closed-loop feedback structure. This structure, based on the reference pilot subcarrier in the received signal, extracts the amplitude error of each array element in the horizontal and vertical polarization channels at the current moment through demodulation. Let the actual received amplitude of the i-th array element at time t be... The theoretical expected amplitude is Then the estimated value of the real-time amplitude error is: ; Accordingly, let the actual received phase be... The theoretical phase is Then the phase error is estimated as follows: ; After the aforementioned error is extracted in real time, it is transmitted to the transmitter as a feedback signal through the feedback link inside the system, so as to dynamically adjust and compensate the precoding matrix of beamforming. Where t represents the current time and i represents the array element number. This represents the real-time amplitude error estimate of the i-th array element at time t. This represents the estimated real-time phase error of the i-th array element at time t; The closed-loop feedback structure is based on the control channel in the data frame. Error calibration is performed by decoding the pre-coded information of the pilot signal carried in the frame, forming a closed-loop measurement channel from the receiver to the transmitter. The control channel embeds pilot signals and their precoded information for system calibration, and the specific process is as follows: Receiver processing: The receiving end first decodes the control channel to obtain the expected structure and precoding method of the pilot signal; By comparing with the actual received pilot signal, the amplitude and phase deviation under the current channel condition can be further extracted; Error calibration and feedback: These deviations are encapsulated into feedback control signals and transmitted to the transmitter via the uplink or a specified feedback path. Based on this feedback information, the transmitter dynamically adjusts the beam direction and amplitude distribution to correct errors caused by array mismatch and polarization interference.

[0024] Furthermore, the machine learning prediction module constructs an error prediction model based on a long short-term memory neural network. This model takes the error signal sequence from the past N time points as input, i.e.: ;as well as: ; That is, the historical data of the amplitude and phase errors of the i-th array element at the past N time points are used to output the error estimate after a future time delay of τ: ; and: ; Where N is the length of the input sequence, and τ is the target prediction delay. This represents the magnitude error prediction value of the machine learning model for the i-th array element at a future time t+τ. This represents the phase error prediction value of the i-th array element by the machine learning model at a future time t+τ.

[0025] The key point of this method is: The input is a historical sliding window of length N that captures the temporal trend of the error. The output is the error estimate at a future time point t+τ, for both amplitude and phase. LSTM is used because the error signal has certain time dependence and nonlinear characteristics, which ordinary linear filters cannot capture well. The prediction result can be used for real-time compensation to improve beamforming accuracy.

[0026] For example, in a phased array radar with 16 elements, if the historical amplitude error of the 7th element fluctuates greatly but has a certain pattern (e.g., an offset occurs every 5 seconds), LSTM can learn this pattern, thereby predicting the upcoming error in advance, compensating in advance, and achieving a more stable beam.

[0027] Furthermore, in dynamic scenarios, such as target movement and channel changes, amplitude and phase errors are introduced. Without compensation, these errors will severely affect beam pointing and performance. Therefore, this claim introduces a weight compensation method based on prediction and real-time error estimation to further improve beamforming accuracy. The complex form target compensation weight... The expression is as follows: = ; in, φi represents the compensation weight of the i-th array element at time t, φi is the expected phase value of array element i in the target direction, and α and β are the weight coefficients of prediction error and real-time error, respectively, which are determined by training samples using the minimum mean square error standard.

[0028] in, For the amplitude compensation portion, This is the phase compensation part (expressed in exponential form).

[0029] The core idea of ​​this expression is to merge and compensate for the current error with the error of future predictions, thus ensuring both real-time performance and forward-looking capabilities. Amplitude compensation portion: The current magnitude response error of the array elements is corrected using a weighted linear model (α and β weights); Phase compensation section: The phase error is corrected by embedding it into the phase angle in an exponential function manner, because phase errors in the antenna array can cause beam shift or even sidelobe enhancement, so accurate correction of φi is crucial.

[0030] Furthermore, the hardware-software co-calibration structure includes a dynamic gain control matrix G(t), a target beam vector Wd(t), and a current output beam vector W0(t). The dynamic gain control matrix G(t) is iteratively updated using the following least squares optimization model: ; Where E(t) represents the beam error at the calibration time, which is the difference between the target beam vector Wd(t) and the current output beam vector W0(t) after adjustment by the dynamic gain control matrix G(t). The smaller the error value, the better the beam calibration effect. The specific implementation of G(t) includes a programmable phase shifter and a gain amplification module in the array control circuit. The minimum error configuration is solved by gradient descent or quasi-Newton method. The matrix acts on each channel of the array for joint adjustment of amplitude and phase. It is composed of the compensation weights of each array element calculated in step (4) and represents the beam output of the physical array in the current calibration state. The target beam vector Wd(t) is the target beam vector calculated in step (4). The target beam vector defines the desired beam output of the array, that is, the ideal direction of the signal that the array should emit at a specific time t. It is composed of the array element compensation weights and reflects how each array element contributes to the final beam shape. The current output beam vector W0(t) represents the beam output of the array in the current state. During the calibration process, the difference between the target beam vector and the current output beam vector determines the magnitude of the error. As the optimization process progresses, the current beam vector will gradually approach the target beam vector.

[0031] During calibration, the goal of the system is to minimize beam error, that is, to minimize the difference between the current beam vector W0(t) and the target beam vector Wd(t).

[0032] The update of the gain control matrix G(t) is achieved through optimization algorithms, including gradient descent and quasi-Newton methods. The goal of these methods is to minimize the error function E(t), that is, to gradually reduce the error between the target beam vector Wd(t) and the output beam vector W0(t) by iteratively updating the control matrix. The gain control matrix G(t) is implemented in hardware through a programmable phase shifter and a gain amplification module in the array control circuit.

[0033] Furthermore, the machine learning prediction module and the real-time error estimation module form a dual closed-loop structure, which improves the error modeling and compensation strategy in the beamforming process by combining feedforward and feedback mechanisms, including: Feedforward channel: The machine learning module predicts error trends based on historical data. This historical data can include the system's performance data under different operating conditions, environmental factors (such as interference, noise, etc.), and array status. Based on the analysis of this historical data, the machine learning module will predict the error trend and output prediction weights for compensation to compensate for the errors that are about to occur. Feedback Channel: The real-time error module samples the current array state and outputs actual error parameters, such as the actual gain error of the array, phase offset, etc. This module helps the system continuously correct the current error by sampling in real time and outputting feedback information, thus ensuring the accuracy and stability of the beamforming process. During the compensation weight calculation stage, the dual closed-loop structure synchronously introduces prediction and feedback information to form a feedforward and feedback joint modeling strategy, which is used to realize the coupled update of error modeling information in weight calculation. This strategy combines machine learning-predicted trends with real-time error feedback to dynamically update error modeling, ensuring the system can adaptively optimize during operation. Specifically, the feedforward channel provides predictive adjustments based on historical data, while the feedback channel corrects the current array state using real-time errors. This combination results in more accurate error modeling and maintains efficient beamforming in dynamically changing environments.

[0034] Furthermore, the beamforming control is executed independently once in the horizontal polarization channel and once in the vertical polarization channel. After execution, the phase results of the two channels are input into the polarization coordination control module. The polarization coordination control module performs phase alignment and gain matching processing based on the cross-correlation function between the polarization channels to form a unified phase reference surface. The alignment result is used to eliminate the phase shift caused by the difference in polarization direction.

[0035] In traditional beamforming, horizontal and vertical polarization may be processed together. However, in this invention, the two polarization channels are subjected to independent beamforming control. This means that the horizontal and vertical polarization signals are optimized independently to improve the overall accuracy of beamforming. Beamforming control is performed once in each polarization channel to ensure that the signal of each polarization channel is optimally processed in the corresponding array and to avoid mutual interference between the two channels.

[0036] Once the horizontal and vertical polarization channels have completed their beamforming control, the resulting phase results will be input into a dedicated polarization coordination control module. The main task of this module is to perform phase alignment and gain matching to ensure that the signals from the two polarization channels can be synthesized under the same reference plane.

[0037] To ensure perfect matching of the two polarization channel signals, the cooperative control module uses a cross-correlation function to evaluate and optimize the correlation between the two channel signals. The cross-correlation function helps determine the time or phase difference between the two polarization channel signals. By analyzing the cross-correlation function, the system can optimize the phase alignment of the two polarization channels, so that the synthesis of the two signals can achieve the maximum gain and the optimal beam direction.

[0038] Through the processing of the collaborative control module, the phases of the two polarization channels will be aligned and the gains will be matched. Finally, the synthesis of the two signals will form a unified phase reference surface. In this way, the system can eliminate the phase shift caused by the difference between the horizontal and vertical polarization directions, ensuring that the beam can achieve the optimal shaping effect under different polarization directions. In addition, the unified phase reference surface can effectively eliminate the error caused by the difference in polarization direction, ensuring the overall performance and accuracy of the system.

[0039] Through this independent yet collaborative control approach, the polarization collaborative control module greatly enhances the flexibility and robustness of beamforming methods. In particular, when dealing with complex electromagnetic environments and multipath propagation, it can more effectively optimize beamforming and reduce errors and interference.

[0040] Furthermore, in step S4, the error input value required for compensation is calculated using the following sliding weighted fusion model: ; ; Wherein, γ(t) is the error fusion weight factor, defined in the domain [0,1]. γ(t) is dynamically updated based on the stability parameter and time margin parameter of the current error estimate. When the value of γ(t) is close to 1, the system relies more on maximum likelihood estimation (ML), which usually means that it is used when the error estimate is more stable or the information is more complete. When the value of γ(t) is close to 0, the system relies more on real-time estimation (RT), which is suitable for rapidly changing or rapidly responding situations. and These are the error compensation values ​​for each parameter (such as gain and phase) during beamforming.

[0041] γ(t) is not fixed, but dynamically adjusted based on the current error estimation stability and time delay margin. The calculation of the weighting factor γ(t) in the error fusion process typically involves the following steps: First, the system needs to estimate the current gain error and phase error. These estimates can be obtained through various methods, such as maximum likelihood estimation or real-time feedback. Secondly, based on the stability and time delay margin of the current system, a certain function (such as weighted moving average or exponential weighting method) is used to dynamically adjust the value of γ(t); Finally, based on the updated weighting factors, the system weights and fuses the two different estimates (ML and RT) to obtain the final compensation error value, thereby updating the beamforming parameters.

[0042] This error fusion method is applicable to dynamic beamforming systems, especially in wireless communication, radar, or sonar systems, for real-time adjustment of beam direction, gain, and phase. These systems often face rapid environmental changes, and the impact of incorrect estimations may lead to a degradation in system performance. Therefore, using a sliding weighted fusion model can flexibly address error and delay issues under different conditions, improving the accuracy and stability of the system.

[0043] The present invention also provides a millimeter-wave dual-polarized phased array antenna device, comprising: Dual-polarized antenna array structure is used to transmit and receive radio frequency signals in the horizontal polarization channel and the vertical polarization channel respectively. The array structure can be a planar array (such as 8×8 or 16×16) to meet the beamforming requirements. It usually adopts microstrip patch antenna, slot antenna or dielectric resonant antenna to adapt to the high frequency characteristics of millimeter wave band (above 30GHz). The real-time error estimation module is configured to extract the amplitude and phase errors of each array element based on the pilot subcarriers in the received signal, forming a closed-loop feedback structure. The machine learning prediction module includes an error trend modeling unit based on a long short-term memory network, which receives historical error signal sequences and outputs error prediction results for future target delays. The weight calculation module is used to calculate the complex form compensation weight of each array element based on the real-time estimation error and the prediction error. The hardware-software co-calibration structure includes a dynamic gain control matrix and a beam control unit, which adjust the transmission state of the antenna array in real time by jointly compensating weights and calibration parameters. The beam control module is used to form a spatial beam in the target direction and perform cooperative alignment and gain matching between polarization channels.

[0044] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A beamforming method for a millimeter-wave dual-polarized phased array antenna, characterized in that, The method comprises the following steps: S1: Constructing a dual-polarized antenna array structure, obtaining complex weight signals of each array element in the horizontal polarization channel and the vertical polarization channel, and collecting and calibrating the amplitude response and the phase response of the complex weight signals; S2: Configuring a real-time error estimation module based on the operating state of the dual-polarized antenna array, dynamically extracting and modeling the amplitude error and the phase error of each array element in different polarization channels; S3: On the basis of error estimation, constructing a machine learning prediction module containing a time series prediction model, modeling the future error change trend in the target time window and outputting the predicted error result; S4: Jointly estimating the real-time error estimation result and the predicted error result, calculating the complex form target compensation weight of each array element in the array, and applying it to error compensation; S5: On the basis of the compensation weight, constructing a hardware-software collaborative calibration structure containing a dynamic gain control module, real-time updating the array gain matrix parameters, and forming an automatic beam adjustment mechanism; S6: Based on the compensation weight and the calibration structure, controlling the phased array antenna to form the required beam pointing in the target direction, and realizing beam focusing on the target space region.

2. The method of claim 1, wherein, The real-time error estimation module is a closed-loop feedback structure. Based on the reference pilot subcarrier in the received signal, the amplitude error of each array element in the horizontal polarization channel and the vertical polarization channel at the current time is extracted through demodulation: ； And the phase error is: ； where t represents the current time, i represents the array element number, represents the real-time amplitude error estimation value of the i-th array element at time t, represents the real-time phase error estimation value of the i-th array element at time t.

3. The method of claim 2, wherein, The machine learning prediction module constructs an error prediction model based on a long short-term memory neural network. The model takes the error signal sequence of the past N time points as input and outputs the error estimation value after a future τ time delay: ； And: ； where N is the input sequence length, and τ is the target prediction delay, represents the amplitude error prediction value of the machine learning model for the i-th array element at the future time t+τ, represents the phase error prediction value of the machine learning model for the i-th array element at the future time t+τ.

4. The method of claim 3, wherein, the complex target compensation weight The expression is as follows: ； wherein, represents the compensation weight of the ith array element at time t, φi is the expected phase value of the array element i in the target direction, and α and β are the weight coefficients of the prediction error and the real-time error, respectively.

5. The method of claim 1, wherein, The hardware-software collaborative calibration structure includes a dynamic gain control matrix G(t), a target beam vector Wd(t), and a current output beam vector W0(t). The dynamic gain control matrix G(t) is updated iteratively through the following least squares optimization model: ； Wherein, E(t) represents the beam error at the calibration time.

6. The method of claim 1, wherein, The machine learning prediction module and the real-time error estimation module form a double closed-loop structure, including: Feedforward channel: the machine learning module predicts the error trend based on historical data and outputs the predicted weight for compensation; Feedback channel: the real-time error module samples the current array state and outputs the actual error parameters; In the compensation weight calculation stage, the double closed-loop structure synchronously introduces prediction and feedback information, forms a feedforward and feedback joint modeling strategy, and is used to realize the coupled update of error modeling information in weight calculation.

7. The method of claim 1, wherein, The beam forming control is independently executed once in the horizontal polarization channel and the vertical polarization channel respectively. After execution, the phase results of the two channels are input into the polarization collaborative control module. Based on the cross-correlation function between the polarization channels, the polarization collaborative control module performs phase alignment and gain matching processing, forms a unified phase reference surface, and the alignment result is used to eliminate the phase offset caused by the difference between the polarization directions.

8. The method of claim 1, wherein, In step S4, the error input value required for compensation is calculated through the following sliding weighted fusion model: ； ； Wherein, γ(t) is an error fusion weight factor, and the definition domain is [0, 1].

9. A millimeter-wave dual-polarized phased array antenna device configured to perform the method of any one of claims 1-8. Including: A dual-polarized antenna array structure for transmitting and receiving radio frequency signals in horizontal and vertical polarization channels respectively; A real-time error estimation module configured to extract amplitude and phase errors of each array element based on pilot subcarriers in the received signal, forming a closed-loop feedback structure; A machine learning prediction module including an error trend modeling unit based on a long short-term memory network, which receives a historical error signal sequence and outputs an error prediction result of a future target time delay; A weight calculation module for calculating a complex compensation weight for each array element according to real-time estimated errors and predicted errors; A hardware-software collaborative calibration structure including a dynamic gain control matrix and a beam control unit, which adjusts the transmission state of the antenna array in real time by jointly compensating the weight and the calibration parameter; A beam control module for forming a spatial beam in the target direction and performing collaborative alignment and gain matching between polarization channels.

Images